From 0793a61d4df8daeac6492dbf8d2f3e5713caae5e Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Thu, 4 Dec 2008 20:12:29 +0100 Subject: performance counters: core code Implement the core kernel bits of Performance Counters subsystem. The Linux Performance Counter subsystem provides an abstraction of performance counter hardware capabilities. It provides per task and per CPU counters, and it provides event capabilities on top of those. Performance counters are accessed via special file descriptors. There's one file descriptor per virtual counter used. The special file descriptor is opened via the perf_counter_open() system call: int perf_counter_open(u32 hw_event_type, u32 hw_event_period, u32 record_type, pid_t pid, int cpu); The syscall returns the new fd. The fd can be used via the normal VFS system calls: read() can be used to read the counter, fcntl() can be used to set the blocking mode, etc. Multiple counters can be kept open at a time, and the counters can be poll()ed. See more details in Documentation/perf-counters.txt. Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- kernel/Makefile | 1 + kernel/fork.c | 1 + kernel/perf_counter.c | 943 ++++++++++++++++++++++++++++++++++++++++++++++++++ kernel/sched.c | 24 ++ kernel/sys_ni.c | 3 + 5 files changed, 972 insertions(+) create mode 100644 kernel/perf_counter.c (limited to 'kernel') diff --git a/kernel/Makefile b/kernel/Makefile index 19fad003b19d..1f184a1dc406 100644 --- a/kernel/Makefile +++ b/kernel/Makefile @@ -89,6 +89,7 @@ obj-$(CONFIG_HAVE_GENERIC_DMA_COHERENT) += dma-coherent.o obj-$(CONFIG_FUNCTION_TRACER) += trace/ obj-$(CONFIG_TRACING) += trace/ obj-$(CONFIG_SMP) += sched_cpupri.o +obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o ifneq ($(CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER),y) # According to Alan Modra , the -fno-omit-frame-pointer is diff --git a/kernel/fork.c b/kernel/fork.c index 2a372a0e206f..441fadff1fa4 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -975,6 +975,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, goto fork_out; rt_mutex_init_task(p); + perf_counter_init_task(p); #ifdef CONFIG_PROVE_LOCKING DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c new file mode 100644 index 000000000000..20508f053658 --- /dev/null +++ b/kernel/perf_counter.c @@ -0,0 +1,943 @@ +/* + * Performance counter core code + * + * Copyright(C) 2008 Thomas Gleixner + * Copyright(C) 2008 Red Hat, Inc., Ingo Molnar + * + * For licencing details see kernel-base/COPYING + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* + * Each CPU has a list of per CPU counters: + */ +DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); + +int perf_max_counters __read_mostly; +static int perf_reserved_percpu __read_mostly; +static int perf_overcommit __read_mostly = 1; + +/* + * Mutex for (sysadmin-configurable) counter reservations: + */ +static DEFINE_MUTEX(perf_resource_mutex); + +/* + * Architecture provided APIs - weak aliases: + */ + +int __weak hw_perf_counter_init(struct perf_counter *counter, u32 hw_event_type) +{ + return -EINVAL; +} + +void __weak hw_perf_counter_enable(struct perf_counter *counter) { } +void __weak hw_perf_counter_disable(struct perf_counter *counter) { } +void __weak hw_perf_counter_read(struct perf_counter *counter) { } +void __weak hw_perf_disable_all(void) { } +void __weak hw_perf_enable_all(void) { } +void __weak hw_perf_counter_setup(void) { } + +#if BITS_PER_LONG == 64 + +/* + * Read the cached counter in counter safe against cross CPU / NMI + * modifications. 64 bit version - no complications. + */ +static inline u64 perf_read_counter_safe(struct perf_counter *counter) +{ + return (u64) atomic64_read(&counter->count); +} + +#else + +/* + * Read the cached counter in counter safe against cross CPU / NMI + * modifications. 32 bit version. + */ +static u64 perf_read_counter_safe(struct perf_counter *counter) +{ + u32 cntl, cnth; + + local_irq_disable(); + do { + cnth = atomic_read(&counter->count32[1]); + cntl = atomic_read(&counter->count32[0]); + } while (cnth != atomic_read(&counter->count32[1])); + + local_irq_enable(); + + return cntl | ((u64) cnth) << 32; +} + +#endif + +/* + * Cross CPU call to remove a performance counter + * + * We disable the counter on the hardware level first. After that we + * remove it from the context list. + */ +static void __perf_remove_from_context(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_counter *counter = info; + struct perf_counter_context *ctx = counter->ctx; + + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + + spin_lock(&ctx->lock); + + if (counter->active) { + hw_perf_counter_disable(counter); + counter->active = 0; + ctx->nr_active--; + cpuctx->active_oncpu--; + counter->task = NULL; + } + ctx->nr_counters--; + + /* + * Protect the list operation against NMI by disabling the + * counters on a global level. NOP for non NMI based counters. + */ + hw_perf_disable_all(); + list_del_init(&counter->list); + hw_perf_enable_all(); + + if (!ctx->task) { + /* + * Allow more per task counters with respect to the + * reservation: + */ + cpuctx->max_pertask = + min(perf_max_counters - ctx->nr_counters, + perf_max_counters - perf_reserved_percpu); + } + + spin_unlock(&ctx->lock); +} + + +/* + * Remove the counter from a task's (or a CPU's) list of counters. + * + * Must be called with counter->mutex held. + * + * CPU counters are removed with a smp call. For task counters we only + * call when the task is on a CPU. + */ +static void perf_remove_from_context(struct perf_counter *counter) +{ + struct perf_counter_context *ctx = counter->ctx; + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Per cpu counters are removed via an smp call and + * the removal is always sucessful. + */ + smp_call_function_single(counter->cpu, + __perf_remove_from_context, + counter, 1); + return; + } + +retry: + task_oncpu_function_call(task, __perf_remove_from_context, + counter); + + spin_lock_irq(&ctx->lock); + /* + * If the context is active we need to retry the smp call. + */ + if (ctx->nr_active && !list_empty(&counter->list)) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + + /* + * The lock prevents that this context is scheduled in so we + * can remove the counter safely, if it the call above did not + * succeed. + */ + if (!list_empty(&counter->list)) { + ctx->nr_counters--; + list_del_init(&counter->list); + counter->task = NULL; + } + spin_unlock_irq(&ctx->lock); +} + +/* + * Cross CPU call to install and enable a preformance counter + */ +static void __perf_install_in_context(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_counter *counter = info; + struct perf_counter_context *ctx = counter->ctx; + int cpu = smp_processor_id(); + + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + + spin_lock(&ctx->lock); + + /* + * Protect the list operation against NMI by disabling the + * counters on a global level. NOP for non NMI based counters. + */ + hw_perf_disable_all(); + list_add_tail(&counter->list, &ctx->counters); + hw_perf_enable_all(); + + ctx->nr_counters++; + + if (cpuctx->active_oncpu < perf_max_counters) { + hw_perf_counter_enable(counter); + counter->active = 1; + counter->oncpu = cpu; + ctx->nr_active++; + cpuctx->active_oncpu++; + } + + if (!ctx->task && cpuctx->max_pertask) + cpuctx->max_pertask--; + + spin_unlock(&ctx->lock); +} + +/* + * Attach a performance counter to a context + * + * First we add the counter to the list with the hardware enable bit + * in counter->hw_config cleared. + * + * If the counter is attached to a task which is on a CPU we use a smp + * call to enable it in the task context. The task might have been + * scheduled away, but we check this in the smp call again. + */ +static void +perf_install_in_context(struct perf_counter_context *ctx, + struct perf_counter *counter, + int cpu) +{ + struct task_struct *task = ctx->task; + + counter->ctx = ctx; + if (!task) { + /* + * Per cpu counters are installed via an smp call and + * the install is always sucessful. + */ + smp_call_function_single(cpu, __perf_install_in_context, + counter, 1); + return; + } + + counter->task = task; +retry: + task_oncpu_function_call(task, __perf_install_in_context, + counter); + + spin_lock_irq(&ctx->lock); + /* + * If the context is active and the counter has not been added + * we need to retry the smp call. + */ + if (ctx->nr_active && list_empty(&counter->list)) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + + /* + * The lock prevents that this context is scheduled in so we + * can add the counter safely, if it the call above did not + * succeed. + */ + if (list_empty(&counter->list)) { + list_add_tail(&counter->list, &ctx->counters); + ctx->nr_counters++; + } + spin_unlock_irq(&ctx->lock); +} + +/* + * Called from scheduler to remove the counters of the current task, + * with interrupts disabled. + * + * We stop each counter and update the counter value in counter->count. + * + * This does not protect us against NMI, but hw_perf_counter_disable() + * sets the disabled bit in the control field of counter _before_ + * accessing the counter control register. If a NMI hits, then it will + * not restart the counter. + */ +void perf_counter_task_sched_out(struct task_struct *task, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct perf_counter *counter; + + if (likely(!cpuctx->task_ctx)) + return; + + spin_lock(&ctx->lock); + list_for_each_entry(counter, &ctx->counters, list) { + if (!ctx->nr_active) + break; + if (counter->active) { + hw_perf_counter_disable(counter); + counter->active = 0; + counter->oncpu = -1; + ctx->nr_active--; + cpuctx->active_oncpu--; + } + } + spin_unlock(&ctx->lock); + cpuctx->task_ctx = NULL; +} + +/* + * Called from scheduler to add the counters of the current task + * with interrupts disabled. + * + * We restore the counter value and then enable it. + * + * This does not protect us against NMI, but hw_perf_counter_enable() + * sets the enabled bit in the control field of counter _before_ + * accessing the counter control register. If a NMI hits, then it will + * keep the counter running. + */ +void perf_counter_task_sched_in(struct task_struct *task, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct perf_counter *counter; + + if (likely(!ctx->nr_counters)) + return; + + spin_lock(&ctx->lock); + list_for_each_entry(counter, &ctx->counters, list) { + if (ctx->nr_active == cpuctx->max_pertask) + break; + if (counter->cpu != -1 && counter->cpu != cpu) + continue; + + hw_perf_counter_enable(counter); + counter->active = 1; + counter->oncpu = cpu; + ctx->nr_active++; + cpuctx->active_oncpu++; + } + spin_unlock(&ctx->lock); + cpuctx->task_ctx = ctx; +} + +void perf_counter_task_tick(struct task_struct *curr, int cpu) +{ + struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter *counter; + + if (likely(!ctx->nr_counters)) + return; + + perf_counter_task_sched_out(curr, cpu); + + spin_lock(&ctx->lock); + + /* + * Rotate the first entry last: + */ + hw_perf_disable_all(); + list_for_each_entry(counter, &ctx->counters, list) { + list_del(&counter->list); + list_add_tail(&counter->list, &ctx->counters); + break; + } + hw_perf_enable_all(); + + spin_unlock(&ctx->lock); + + perf_counter_task_sched_in(curr, cpu); +} + +/* + * Initialize the perf_counter context in task_struct + */ +void perf_counter_init_task(struct task_struct *task) +{ + struct perf_counter_context *ctx = &task->perf_counter_ctx; + + spin_lock_init(&ctx->lock); + INIT_LIST_HEAD(&ctx->counters); + ctx->nr_counters = 0; + ctx->task = task; +} + +/* + * Cross CPU call to read the hardware counter + */ +static void __hw_perf_counter_read(void *info) +{ + hw_perf_counter_read(info); +} + +static u64 perf_read_counter(struct perf_counter *counter) +{ + /* + * If counter is enabled and currently active on a CPU, update the + * value in the counter structure: + */ + if (counter->active) { + smp_call_function_single(counter->oncpu, + __hw_perf_counter_read, counter, 1); + } + + return perf_read_counter_safe(counter); +} + +/* + * Cross CPU call to switch performance data pointers + */ +static void __perf_switch_irq_data(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_counter *counter = info; + struct perf_counter_context *ctx = counter->ctx; + struct perf_data *oldirqdata = counter->irqdata; + + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. + */ + if (ctx->task) { + if (cpuctx->task_ctx != ctx) + return; + spin_lock(&ctx->lock); + } + + /* Change the pointer NMI safe */ + atomic_long_set((atomic_long_t *)&counter->irqdata, + (unsigned long) counter->usrdata); + counter->usrdata = oldirqdata; + + if (ctx->task) + spin_unlock(&ctx->lock); +} + +static struct perf_data *perf_switch_irq_data(struct perf_counter *counter) +{ + struct perf_counter_context *ctx = counter->ctx; + struct perf_data *oldirqdata = counter->irqdata; + struct task_struct *task = ctx->task; + + if (!task) { + smp_call_function_single(counter->cpu, + __perf_switch_irq_data, + counter, 1); + return counter->usrdata; + } + +retry: + spin_lock_irq(&ctx->lock); + if (!counter->active) { + counter->irqdata = counter->usrdata; + counter->usrdata = oldirqdata; + spin_unlock_irq(&ctx->lock); + return oldirqdata; + } + spin_unlock_irq(&ctx->lock); + task_oncpu_function_call(task, __perf_switch_irq_data, counter); + /* Might have failed, because task was scheduled out */ + if (counter->irqdata == oldirqdata) + goto retry; + + return counter->usrdata; +} + +static void put_context(struct perf_counter_context *ctx) +{ + if (ctx->task) + put_task_struct(ctx->task); +} + +static struct perf_counter_context *find_get_context(pid_t pid, int cpu) +{ + struct perf_cpu_context *cpuctx; + struct perf_counter_context *ctx; + struct task_struct *task; + + /* + * If cpu is not a wildcard then this is a percpu counter: + */ + if (cpu != -1) { + /* Must be root to operate on a CPU counter: */ + if (!capable(CAP_SYS_ADMIN)) + return ERR_PTR(-EACCES); + + if (cpu < 0 || cpu > num_possible_cpus()) + return ERR_PTR(-EINVAL); + + /* + * We could be clever and allow to attach a counter to an + * offline CPU and activate it when the CPU comes up, but + * that's for later. + */ + if (!cpu_isset(cpu, cpu_online_map)) + return ERR_PTR(-ENODEV); + + cpuctx = &per_cpu(perf_cpu_context, cpu); + ctx = &cpuctx->ctx; + + WARN_ON_ONCE(ctx->task); + return ctx; + } + + rcu_read_lock(); + if (!pid) + task = current; + else + task = find_task_by_vpid(pid); + if (task) + get_task_struct(task); + rcu_read_unlock(); + + if (!task) + return ERR_PTR(-ESRCH); + + ctx = &task->perf_counter_ctx; + ctx->task = task; + + /* Reuse ptrace permission checks for now. */ + if (!ptrace_may_access(task, PTRACE_MODE_READ)) { + put_context(ctx); + return ERR_PTR(-EACCES); + } + + return ctx; +} + +/* + * Called when the last reference to the file is gone. + */ +static int perf_release(struct inode *inode, struct file *file) +{ + struct perf_counter *counter = file->private_data; + struct perf_counter_context *ctx = counter->ctx; + + file->private_data = NULL; + + mutex_lock(&counter->mutex); + + perf_remove_from_context(counter); + put_context(ctx); + + mutex_unlock(&counter->mutex); + + kfree(counter); + + return 0; +} + +/* + * Read the performance counter - simple non blocking version for now + */ +static ssize_t +perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) +{ + u64 cntval; + + if (count != sizeof(cntval)) + return -EINVAL; + + mutex_lock(&counter->mutex); + cntval = perf_read_counter(counter); + mutex_unlock(&counter->mutex); + + return put_user(cntval, (u64 __user *) buf) ? -EFAULT : sizeof(cntval); +} + +static ssize_t +perf_copy_usrdata(struct perf_data *usrdata, char __user *buf, size_t count) +{ + if (!usrdata->len) + return 0; + + count = min(count, (size_t)usrdata->len); + if (copy_to_user(buf, usrdata->data + usrdata->rd_idx, count)) + return -EFAULT; + + /* Adjust the counters */ + usrdata->len -= count; + if (!usrdata->len) + usrdata->rd_idx = 0; + else + usrdata->rd_idx += count; + + return count; +} + +static ssize_t +perf_read_irq_data(struct perf_counter *counter, + char __user *buf, + size_t count, + int nonblocking) +{ + struct perf_data *irqdata, *usrdata; + DECLARE_WAITQUEUE(wait, current); + ssize_t res; + + irqdata = counter->irqdata; + usrdata = counter->usrdata; + + if (usrdata->len + irqdata->len >= count) + goto read_pending; + + if (nonblocking) + return -EAGAIN; + + spin_lock_irq(&counter->waitq.lock); + __add_wait_queue(&counter->waitq, &wait); + for (;;) { + set_current_state(TASK_INTERRUPTIBLE); + if (usrdata->len + irqdata->len >= count) + break; + + if (signal_pending(current)) + break; + + spin_unlock_irq(&counter->waitq.lock); + schedule(); + spin_lock_irq(&counter->waitq.lock); + } + __remove_wait_queue(&counter->waitq, &wait); + __set_current_state(TASK_RUNNING); + spin_unlock_irq(&counter->waitq.lock); + + if (usrdata->len + irqdata->len < count) + return -ERESTARTSYS; +read_pending: + mutex_lock(&counter->mutex); + + /* Drain pending data first: */ + res = perf_copy_usrdata(usrdata, buf, count); + if (res < 0 || res == count) + goto out; + + /* Switch irq buffer: */ + usrdata = perf_switch_irq_data(counter); + if (perf_copy_usrdata(usrdata, buf + res, count - res) < 0) { + if (!res) + res = -EFAULT; + } else { + res = count; + } +out: + mutex_unlock(&counter->mutex); + + return res; +} + +static ssize_t +perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) +{ + struct perf_counter *counter = file->private_data; + + switch (counter->record_type) { + case PERF_RECORD_SIMPLE: + return perf_read_hw(counter, buf, count); + + case PERF_RECORD_IRQ: + case PERF_RECORD_GROUP: + return perf_read_irq_data(counter, buf, count, + file->f_flags & O_NONBLOCK); + } + return -EINVAL; +} + +static unsigned int perf_poll(struct file *file, poll_table *wait) +{ + struct perf_counter *counter = file->private_data; + unsigned int events = 0; + unsigned long flags; + + poll_wait(file, &counter->waitq, wait); + + spin_lock_irqsave(&counter->waitq.lock, flags); + if (counter->usrdata->len || counter->irqdata->len) + events |= POLLIN; + spin_unlock_irqrestore(&counter->waitq.lock, flags); + + return events; +} + +static const struct file_operations perf_fops = { + .release = perf_release, + .read = perf_read, + .poll = perf_poll, +}; + +/* + * Allocate and initialize a counter structure + */ +static struct perf_counter * +perf_counter_alloc(u32 hw_event_period, int cpu, u32 record_type) +{ + struct perf_counter *counter = kzalloc(sizeof(*counter), GFP_KERNEL); + + if (!counter) + return NULL; + + mutex_init(&counter->mutex); + INIT_LIST_HEAD(&counter->list); + init_waitqueue_head(&counter->waitq); + + counter->irqdata = &counter->data[0]; + counter->usrdata = &counter->data[1]; + counter->cpu = cpu; + counter->record_type = record_type; + counter->__irq_period = hw_event_period; + counter->wakeup_pending = 0; + + return counter; +} + +/** + * sys_perf_task_open - open a performance counter associate it to a task + * @hw_event_type: event type for monitoring/sampling... + * @pid: target pid + */ +asmlinkage int +sys_perf_counter_open(u32 hw_event_type, + u32 hw_event_period, + u32 record_type, + pid_t pid, + int cpu) +{ + struct perf_counter_context *ctx; + struct perf_counter *counter; + int ret; + + ctx = find_get_context(pid, cpu); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + ret = -ENOMEM; + counter = perf_counter_alloc(hw_event_period, cpu, record_type); + if (!counter) + goto err_put_context; + + ret = hw_perf_counter_init(counter, hw_event_type); + if (ret) + goto err_free_put_context; + + perf_install_in_context(ctx, counter, cpu); + + ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); + if (ret < 0) + goto err_remove_free_put_context; + + return ret; + +err_remove_free_put_context: + mutex_lock(&counter->mutex); + perf_remove_from_context(counter); + mutex_unlock(&counter->mutex); + +err_free_put_context: + kfree(counter); + +err_put_context: + put_context(ctx); + + return ret; +} + +static void __cpuinit perf_init_cpu(int cpu) +{ + struct perf_cpu_context *ctx; + + ctx = &per_cpu(perf_cpu_context, cpu); + spin_lock_init(&ctx->ctx.lock); + INIT_LIST_HEAD(&ctx->ctx.counters); + + mutex_lock(&perf_resource_mutex); + ctx->max_pertask = perf_max_counters - perf_reserved_percpu; + mutex_unlock(&perf_resource_mutex); + hw_perf_counter_setup(); +} + +#ifdef CONFIG_HOTPLUG_CPU +static void __perf_exit_cpu(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_counter_context *ctx = &cpuctx->ctx; + struct perf_counter *counter, *tmp; + + list_for_each_entry_safe(counter, tmp, &ctx->counters, list) + __perf_remove_from_context(counter); + +} +static void perf_exit_cpu(int cpu) +{ + smp_call_function_single(cpu, __perf_exit_cpu, NULL, 1); +} +#else +static inline void perf_exit_cpu(int cpu) { } +#endif + +static int __cpuinit +perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) +{ + unsigned int cpu = (long)hcpu; + + switch (action) { + + case CPU_UP_PREPARE: + case CPU_UP_PREPARE_FROZEN: + perf_init_cpu(cpu); + break; + + case CPU_DOWN_PREPARE: + case CPU_DOWN_PREPARE_FROZEN: + perf_exit_cpu(cpu); + break; + + default: + break; + } + + return NOTIFY_OK; +} + +static struct notifier_block __cpuinitdata perf_cpu_nb = { + .notifier_call = perf_cpu_notify, +}; + +static int __init perf_counter_init(void) +{ + perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, + (void *)(long)smp_processor_id()); + register_cpu_notifier(&perf_cpu_nb); + + return 0; +} +early_initcall(perf_counter_init); + +static ssize_t perf_show_reserve_percpu(struct sysdev_class *class, char *buf) +{ + return sprintf(buf, "%d\n", perf_reserved_percpu); +} + +static ssize_t +perf_set_reserve_percpu(struct sysdev_class *class, + const char *buf, + size_t count) +{ + struct perf_cpu_context *cpuctx; + unsigned long val; + int err, cpu, mpt; + + err = strict_strtoul(buf, 10, &val); + if (err) + return err; + if (val > perf_max_counters) + return -EINVAL; + + mutex_lock(&perf_resource_mutex); + perf_reserved_percpu = val; + for_each_online_cpu(cpu) { + cpuctx = &per_cpu(perf_cpu_context, cpu); + spin_lock_irq(&cpuctx->ctx.lock); + mpt = min(perf_max_counters - cpuctx->ctx.nr_counters, + perf_max_counters - perf_reserved_percpu); + cpuctx->max_pertask = mpt; + spin_unlock_irq(&cpuctx->ctx.lock); + } + mutex_unlock(&perf_resource_mutex); + + return count; +} + +static ssize_t perf_show_overcommit(struct sysdev_class *class, char *buf) +{ + return sprintf(buf, "%d\n", perf_overcommit); +} + +static ssize_t +perf_set_overcommit(struct sysdev_class *class, const char *buf, size_t count) +{ + unsigned long val; + int err; + + err = strict_strtoul(buf, 10, &val); + if (err) + return err; + if (val > 1) + return -EINVAL; + + mutex_lock(&perf_resource_mutex); + perf_overcommit = val; + mutex_unlock(&perf_resource_mutex); + + return count; +} + +static SYSDEV_CLASS_ATTR( + reserve_percpu, + 0644, + perf_show_reserve_percpu, + perf_set_reserve_percpu + ); + +static SYSDEV_CLASS_ATTR( + overcommit, + 0644, + perf_show_overcommit, + perf_set_overcommit + ); + +static struct attribute *perfclass_attrs[] = { + &attr_reserve_percpu.attr, + &attr_overcommit.attr, + NULL +}; + +static struct attribute_group perfclass_attr_group = { + .attrs = perfclass_attrs, + .name = "perf_counters", +}; + +static int __init perf_counter_sysfs_init(void) +{ + return sysfs_create_group(&cpu_sysdev_class.kset.kobj, + &perfclass_attr_group); +} +device_initcall(perf_counter_sysfs_init); + diff --git a/kernel/sched.c b/kernel/sched.c index b7480fb5c3dc..254d56de2548 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2212,6 +2212,27 @@ static int sched_balance_self(int cpu, int flag) #endif /* CONFIG_SMP */ +/** + * task_oncpu_function_call - call a function on the cpu on which a task runs + * @p: the task to evaluate + * @func: the function to be called + * @info: the function call argument + * + * Calls the function @func when the task is currently running. This might + * be on the current CPU, which just calls the function directly + */ +void task_oncpu_function_call(struct task_struct *p, + void (*func) (void *info), void *info) +{ + int cpu; + + preempt_disable(); + cpu = task_cpu(p); + if (task_curr(p)) + smp_call_function_single(cpu, func, info, 1); + preempt_enable(); +} + /*** * try_to_wake_up - wake up a thread * @p: the to-be-woken-up thread @@ -2534,6 +2555,7 @@ prepare_task_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next) { fire_sched_out_preempt_notifiers(prev, next); + perf_counter_task_sched_out(prev, cpu_of(rq)); prepare_lock_switch(rq, next); prepare_arch_switch(next); } @@ -2574,6 +2596,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev) */ prev_state = prev->state; finish_arch_switch(prev); + perf_counter_task_sched_in(current, cpu_of(rq)); finish_lock_switch(rq, prev); #ifdef CONFIG_SMP if (current->sched_class->post_schedule) @@ -4296,6 +4319,7 @@ void scheduler_tick(void) rq->idle_at_tick = idle_cpu(cpu); trigger_load_balance(rq, cpu); #endif + perf_counter_task_tick(curr, cpu); } #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c index e14a23281707..4be8bbc7577c 100644 --- a/kernel/sys_ni.c +++ b/kernel/sys_ni.c @@ -174,3 +174,6 @@ cond_syscall(compat_sys_timerfd_settime); cond_syscall(compat_sys_timerfd_gettime); cond_syscall(sys_eventfd); cond_syscall(sys_eventfd2); + +/* performance counters: */ +cond_syscall(sys_perf_counter_open); -- cgit v1.2.3 From eab656ae04b9d3b83265e3db01c0d2c46b748ef7 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 8 Dec 2008 19:26:59 +0100 Subject: perf counters: clean up 'raw' type API Impact: cleanup Introduce a separate hw_event type. Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 20508f053658..96c333a5b0fc 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -734,26 +734,27 @@ perf_counter_alloc(u32 hw_event_period, int cpu, u32 record_type) * @pid: target pid */ asmlinkage int -sys_perf_counter_open(u32 hw_event_type, - u32 hw_event_period, - u32 record_type, - pid_t pid, - int cpu) +sys_perf_counter_open(struct perf_counter_event __user *uevent, u32 record_type, + pid_t pid, int cpu, int masterfd) { struct perf_counter_context *ctx; + struct perf_counter_event event; struct perf_counter *counter; int ret; + if (copy_from_user(&event, uevent, sizeof(event)) != 0) + return -EFAULT; + ctx = find_get_context(pid, cpu); if (IS_ERR(ctx)) return PTR_ERR(ctx); ret = -ENOMEM; - counter = perf_counter_alloc(hw_event_period, cpu, record_type); + counter = perf_counter_alloc(event.hw_event_period, cpu, record_type); if (!counter) goto err_put_context; - ret = hw_perf_counter_init(counter, hw_event_type); + ret = hw_perf_counter_init(counter, event.hw_event_type); if (ret) goto err_free_put_context; -- cgit v1.2.3 From dfa7c899b401d7dc5d85aca416aee64ac82812f2 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 8 Dec 2008 19:35:37 +0100 Subject: perf counters: expand use of counter->event Impact: change syscall, cleanup Make use of the new perf_counters event type. Signed-off-by: Thomas Gleixner Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 96c333a5b0fc..2557c670a3bb 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -37,7 +37,7 @@ static DEFINE_MUTEX(perf_resource_mutex); * Architecture provided APIs - weak aliases: */ -int __weak hw_perf_counter_init(struct perf_counter *counter, u32 hw_event_type) +int __weak hw_perf_counter_init(struct perf_counter *counter) { return -EINVAL; } @@ -707,7 +707,7 @@ static const struct file_operations perf_fops = { * Allocate and initialize a counter structure */ static struct perf_counter * -perf_counter_alloc(u32 hw_event_period, int cpu, u32 record_type) +perf_counter_alloc(struct perf_counter_event *event, int cpu, u32 record_type) { struct perf_counter *counter = kzalloc(sizeof(*counter), GFP_KERNEL); @@ -722,7 +722,7 @@ perf_counter_alloc(u32 hw_event_period, int cpu, u32 record_type) counter->usrdata = &counter->data[1]; counter->cpu = cpu; counter->record_type = record_type; - counter->__irq_period = hw_event_period; + counter->event = *event; counter->wakeup_pending = 0; return counter; @@ -750,11 +750,11 @@ sys_perf_counter_open(struct perf_counter_event __user *uevent, u32 record_type, return PTR_ERR(ctx); ret = -ENOMEM; - counter = perf_counter_alloc(event.hw_event_period, cpu, record_type); + counter = perf_counter_alloc(&event, cpu, record_type); if (!counter) goto err_put_context; - ret = hw_perf_counter_init(counter, event.hw_event_type); + ret = hw_perf_counter_init(counter); if (ret) goto err_free_put_context; -- cgit v1.2.3 From 9f66a3810fe0d4100972db84290f3ae4a4d77025 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 10 Dec 2008 12:33:23 +0100 Subject: perf counters: restructure the API Impact: clean up new API Thorough cleanup of the new perf counters API, we now get clean separation of the various concepts: - introduce perf_counter_hw_event to separate out the event source details - move special type flags into separate attributes: PERF_COUNT_NMI, PERF_COUNT_RAW - extend the type to u64 and reserve it fully to the architecture in the raw type case. And make use of all these changes in the core and x86 perfcounters code. Also change the syscall signature to: asmlinkage int sys_perf_counter_open( struct perf_counter_hw_event *hw_event_uptr __user, pid_t pid, int cpu, int group_fd); ( Note that group_fd is unused for now - it's reserved for the counter groups abstraction. ) Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 38 ++++++++++++++++++++++---------------- 1 file changed, 22 insertions(+), 16 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 2557c670a3bb..0d323ceda3a4 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -669,7 +669,7 @@ perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) { struct perf_counter *counter = file->private_data; - switch (counter->record_type) { + switch (counter->hw_event.record_type) { case PERF_RECORD_SIMPLE: return perf_read_hw(counter, buf, count); @@ -707,7 +707,7 @@ static const struct file_operations perf_fops = { * Allocate and initialize a counter structure */ static struct perf_counter * -perf_counter_alloc(struct perf_counter_event *event, int cpu, u32 record_type) +perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu) { struct perf_counter *counter = kzalloc(sizeof(*counter), GFP_KERNEL); @@ -718,31 +718,37 @@ perf_counter_alloc(struct perf_counter_event *event, int cpu, u32 record_type) INIT_LIST_HEAD(&counter->list); init_waitqueue_head(&counter->waitq); - counter->irqdata = &counter->data[0]; - counter->usrdata = &counter->data[1]; - counter->cpu = cpu; - counter->record_type = record_type; - counter->event = *event; - counter->wakeup_pending = 0; + counter->irqdata = &counter->data[0]; + counter->usrdata = &counter->data[1]; + counter->cpu = cpu; + counter->hw_event = *hw_event; + counter->wakeup_pending = 0; return counter; } /** - * sys_perf_task_open - open a performance counter associate it to a task - * @hw_event_type: event type for monitoring/sampling... + * sys_perf_task_open - open a performance counter, associate it to a task/cpu + * + * @hw_event_uptr: event type attributes for monitoring/sampling * @pid: target pid + * @cpu: target cpu + * @group_fd: group leader counter fd */ -asmlinkage int -sys_perf_counter_open(struct perf_counter_event __user *uevent, u32 record_type, - pid_t pid, int cpu, int masterfd) +asmlinkage int sys_perf_counter_open( + + struct perf_counter_hw_event *hw_event_uptr __user, + pid_t pid, + int cpu, + int group_fd) + { struct perf_counter_context *ctx; - struct perf_counter_event event; + struct perf_counter_hw_event hw_event; struct perf_counter *counter; int ret; - if (copy_from_user(&event, uevent, sizeof(event)) != 0) + if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0) return -EFAULT; ctx = find_get_context(pid, cpu); @@ -750,7 +756,7 @@ sys_perf_counter_open(struct perf_counter_event __user *uevent, u32 record_type, return PTR_ERR(ctx); ret = -ENOMEM; - counter = perf_counter_alloc(&event, cpu, record_type); + counter = perf_counter_alloc(&hw_event, cpu); if (!counter) goto err_put_context; -- cgit v1.2.3 From 04289bb9891882202d7e961c4c04d2376930e9f9 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 08:38:42 +0100 Subject: perf counters: add support for group counters Impact: add group counters This patch adds the "counter groups" abstraction. Groups of counters behave much like normal 'single' counters, with a few semantic and behavioral extensions on top of that. A counter group is created by creating a new counter with the open() syscall's group-leader group_fd file descriptor parameter pointing to another, already existing counter. Groups of counters are scheduled in and out in one atomic group, and they are also roundrobin-scheduled atomically. Counters that are member of a group can also record events with an (atomic) extended timestamp that extends to all members of the group, if the record type is set to PERF_RECORD_GROUP. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 282 ++++++++++++++++++++++++++++++++++++++------------ 1 file changed, 216 insertions(+), 66 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0d323ceda3a4..fa59fe8c02d5 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -10,6 +10,7 @@ #include #include #include +#include #include #include #include @@ -55,7 +56,7 @@ void __weak hw_perf_counter_setup(void) { } * Read the cached counter in counter safe against cross CPU / NMI * modifications. 64 bit version - no complications. */ -static inline u64 perf_read_counter_safe(struct perf_counter *counter) +static inline u64 perf_counter_read_safe(struct perf_counter *counter) { return (u64) atomic64_read(&counter->count); } @@ -66,7 +67,7 @@ static inline u64 perf_read_counter_safe(struct perf_counter *counter) * Read the cached counter in counter safe against cross CPU / NMI * modifications. 32 bit version. */ -static u64 perf_read_counter_safe(struct perf_counter *counter) +static u64 perf_counter_read_safe(struct perf_counter *counter) { u32 cntl, cnth; @@ -83,13 +84,55 @@ static u64 perf_read_counter_safe(struct perf_counter *counter) #endif +static void +list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) +{ + struct perf_counter *group_leader = counter->group_leader; + + /* + * Depending on whether it is a standalone or sibling counter, + * add it straight to the context's counter list, or to the group + * leader's sibling list: + */ + if (counter->group_leader == counter) + list_add_tail(&counter->list_entry, &ctx->counter_list); + else + list_add_tail(&counter->list_entry, &group_leader->sibling_list); +} + +static void +list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) +{ + struct perf_counter *sibling, *tmp; + + list_del_init(&counter->list_entry); + + if (list_empty(&counter->sibling_list)) + return; + + /* + * If this was a group counter with sibling counters then + * upgrade the siblings to singleton counters by adding them + * to the context list directly: + */ + list_for_each_entry_safe(sibling, tmp, + &counter->sibling_list, list_entry) { + + list_del_init(&sibling->list_entry); + list_add_tail(&sibling->list_entry, &ctx->counter_list); + WARN_ON_ONCE(!sibling->group_leader); + WARN_ON_ONCE(sibling->group_leader == sibling); + sibling->group_leader = sibling; + } +} + /* * Cross CPU call to remove a performance counter * * We disable the counter on the hardware level first. After that we * remove it from the context list. */ -static void __perf_remove_from_context(void *info) +static void __perf_counter_remove_from_context(void *info) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter *counter = info; @@ -119,7 +162,7 @@ static void __perf_remove_from_context(void *info) * counters on a global level. NOP for non NMI based counters. */ hw_perf_disable_all(); - list_del_init(&counter->list); + list_del_counter(counter, ctx); hw_perf_enable_all(); if (!ctx->task) { @@ -144,7 +187,7 @@ static void __perf_remove_from_context(void *info) * CPU counters are removed with a smp call. For task counters we only * call when the task is on a CPU. */ -static void perf_remove_from_context(struct perf_counter *counter) +static void perf_counter_remove_from_context(struct perf_counter *counter) { struct perf_counter_context *ctx = counter->ctx; struct task_struct *task = ctx->task; @@ -155,32 +198,32 @@ static void perf_remove_from_context(struct perf_counter *counter) * the removal is always sucessful. */ smp_call_function_single(counter->cpu, - __perf_remove_from_context, + __perf_counter_remove_from_context, counter, 1); return; } retry: - task_oncpu_function_call(task, __perf_remove_from_context, + task_oncpu_function_call(task, __perf_counter_remove_from_context, counter); spin_lock_irq(&ctx->lock); /* * If the context is active we need to retry the smp call. */ - if (ctx->nr_active && !list_empty(&counter->list)) { + if (ctx->nr_active && !list_empty(&counter->list_entry)) { spin_unlock_irq(&ctx->lock); goto retry; } /* * The lock prevents that this context is scheduled in so we - * can remove the counter safely, if it the call above did not + * can remove the counter safely, if the call above did not * succeed. */ - if (!list_empty(&counter->list)) { + if (!list_empty(&counter->list_entry)) { ctx->nr_counters--; - list_del_init(&counter->list); + list_del_counter(counter, ctx); counter->task = NULL; } spin_unlock_irq(&ctx->lock); @@ -211,7 +254,7 @@ static void __perf_install_in_context(void *info) * counters on a global level. NOP for non NMI based counters. */ hw_perf_disable_all(); - list_add_tail(&counter->list, &ctx->counters); + list_add_counter(counter, ctx); hw_perf_enable_all(); ctx->nr_counters++; @@ -268,7 +311,7 @@ retry: * If the context is active and the counter has not been added * we need to retry the smp call. */ - if (ctx->nr_active && list_empty(&counter->list)) { + if (ctx->nr_active && list_empty(&counter->list_entry)) { spin_unlock_irq(&ctx->lock); goto retry; } @@ -278,13 +321,45 @@ retry: * can add the counter safely, if it the call above did not * succeed. */ - if (list_empty(&counter->list)) { - list_add_tail(&counter->list, &ctx->counters); + if (list_empty(&counter->list_entry)) { + list_add_counter(counter, ctx); ctx->nr_counters++; } spin_unlock_irq(&ctx->lock); } +static void +counter_sched_out(struct perf_counter *counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx) +{ + if (!counter->active) + return; + + hw_perf_counter_disable(counter); + counter->active = 0; + counter->oncpu = -1; + + cpuctx->active_oncpu--; + ctx->nr_active--; +} + +static void +group_sched_out(struct perf_counter *group_counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx) +{ + struct perf_counter *counter; + + counter_sched_out(group_counter, cpuctx, ctx); + + /* + * Schedule out siblings (if any): + */ + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) + counter_sched_out(counter, cpuctx, ctx); +} + /* * Called from scheduler to remove the counters of the current task, * with interrupts disabled. @@ -306,21 +381,48 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) return; spin_lock(&ctx->lock); - list_for_each_entry(counter, &ctx->counters, list) { - if (!ctx->nr_active) - break; - if (counter->active) { - hw_perf_counter_disable(counter); - counter->active = 0; - counter->oncpu = -1; - ctx->nr_active--; - cpuctx->active_oncpu--; - } + if (ctx->nr_active) { + list_for_each_entry(counter, &ctx->counter_list, list_entry) + group_sched_out(counter, cpuctx, ctx); } spin_unlock(&ctx->lock); cpuctx->task_ctx = NULL; } +static void +counter_sched_in(struct perf_counter *counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx, + int cpu) +{ + if (!counter->active) + return; + + hw_perf_counter_enable(counter); + counter->active = 1; + counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ + + cpuctx->active_oncpu++; + ctx->nr_active++; +} + +static void +group_sched_in(struct perf_counter *group_counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx, + int cpu) +{ + struct perf_counter *counter; + + counter_sched_in(group_counter, cpuctx, ctx, cpu); + + /* + * Schedule in siblings as one group (if any): + */ + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) + counter_sched_in(counter, cpuctx, ctx, cpu); +} + /* * Called from scheduler to add the counters of the current task * with interrupts disabled. @@ -342,19 +444,21 @@ void perf_counter_task_sched_in(struct task_struct *task, int cpu) return; spin_lock(&ctx->lock); - list_for_each_entry(counter, &ctx->counters, list) { + list_for_each_entry(counter, &ctx->counter_list, list_entry) { if (ctx->nr_active == cpuctx->max_pertask) break; + + /* + * Listen to the 'cpu' scheduling filter constraint + * of counters: + */ if (counter->cpu != -1 && counter->cpu != cpu) continue; - hw_perf_counter_enable(counter); - counter->active = 1; - counter->oncpu = cpu; - ctx->nr_active++; - cpuctx->active_oncpu++; + group_sched_in(counter, cpuctx, ctx, cpu); } spin_unlock(&ctx->lock); + cpuctx->task_ctx = ctx; } @@ -371,12 +475,12 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) spin_lock(&ctx->lock); /* - * Rotate the first entry last: + * Rotate the first entry last (works just fine for group counters too): */ hw_perf_disable_all(); - list_for_each_entry(counter, &ctx->counters, list) { - list_del(&counter->list); - list_add_tail(&counter->list, &ctx->counters); + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + list_del(&counter->list_entry); + list_add_tail(&counter->list_entry, &ctx->counter_list); break; } hw_perf_enable_all(); @@ -386,17 +490,24 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) perf_counter_task_sched_in(curr, cpu); } +/* + * Initialize the perf_counter context in a task_struct: + */ +static void +__perf_counter_init_context(struct perf_counter_context *ctx, + struct task_struct *task) +{ + spin_lock_init(&ctx->lock); + INIT_LIST_HEAD(&ctx->counter_list); + ctx->nr_counters = 0; + ctx->task = task; +} /* * Initialize the perf_counter context in task_struct */ void perf_counter_init_task(struct task_struct *task) { - struct perf_counter_context *ctx = &task->perf_counter_ctx; - - spin_lock_init(&ctx->lock); - INIT_LIST_HEAD(&ctx->counters); - ctx->nr_counters = 0; - ctx->task = task; + __perf_counter_init_context(&task->perf_counter_ctx, task); } /* @@ -407,7 +518,7 @@ static void __hw_perf_counter_read(void *info) hw_perf_counter_read(info); } -static u64 perf_read_counter(struct perf_counter *counter) +static u64 perf_counter_read(struct perf_counter *counter) { /* * If counter is enabled and currently active on a CPU, update the @@ -418,7 +529,7 @@ static u64 perf_read_counter(struct perf_counter *counter) __hw_perf_counter_read, counter, 1); } - return perf_read_counter_safe(counter); + return perf_counter_read_safe(counter); } /* @@ -555,7 +666,7 @@ static int perf_release(struct inode *inode, struct file *file) mutex_lock(&counter->mutex); - perf_remove_from_context(counter); + perf_counter_remove_from_context(counter); put_context(ctx); mutex_unlock(&counter->mutex); @@ -577,7 +688,7 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) return -EINVAL; mutex_lock(&counter->mutex); - cntval = perf_read_counter(counter); + cntval = perf_counter_read(counter); mutex_unlock(&counter->mutex); return put_user(cntval, (u64 __user *) buf) ? -EFAULT : sizeof(cntval); @@ -707,15 +818,25 @@ static const struct file_operations perf_fops = { * Allocate and initialize a counter structure */ static struct perf_counter * -perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu) +perf_counter_alloc(struct perf_counter_hw_event *hw_event, + int cpu, + struct perf_counter *group_leader) { struct perf_counter *counter = kzalloc(sizeof(*counter), GFP_KERNEL); if (!counter) return NULL; + /* + * Single counters are their own group leaders, with an + * empty sibling list: + */ + if (!group_leader) + group_leader = counter; + mutex_init(&counter->mutex); - INIT_LIST_HEAD(&counter->list); + INIT_LIST_HEAD(&counter->list_entry); + INIT_LIST_HEAD(&counter->sibling_list); init_waitqueue_head(&counter->waitq); counter->irqdata = &counter->data[0]; @@ -723,6 +844,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu) counter->cpu = cpu; counter->hw_event = *hw_event; counter->wakeup_pending = 0; + counter->group_leader = group_leader; return counter; } @@ -743,20 +865,45 @@ asmlinkage int sys_perf_counter_open( int group_fd) { - struct perf_counter_context *ctx; + struct perf_counter *counter, *group_leader; struct perf_counter_hw_event hw_event; - struct perf_counter *counter; + struct perf_counter_context *ctx; + struct file *group_file = NULL; + int fput_needed = 0; int ret; if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0) return -EFAULT; + /* + * Look up the group leader: + */ + group_leader = NULL; + if (group_fd != -1) { + ret = -EINVAL; + group_file = fget_light(group_fd, &fput_needed); + if (!group_file) + goto out_fput; + if (group_file->f_op != &perf_fops) + goto out_fput; + + group_leader = group_file->private_data; + /* + * Do not allow a recursive hierarchy: + */ + if (group_leader->group_leader) + goto out_fput; + } + + /* + * Get the target context (task or percpu): + */ ctx = find_get_context(pid, cpu); if (IS_ERR(ctx)) return PTR_ERR(ctx); ret = -ENOMEM; - counter = perf_counter_alloc(&hw_event, cpu); + counter = perf_counter_alloc(&hw_event, cpu, group_leader); if (!counter) goto err_put_context; @@ -770,11 +917,14 @@ asmlinkage int sys_perf_counter_open( if (ret < 0) goto err_remove_free_put_context; +out_fput: + fput_light(group_file, fput_needed); + return ret; err_remove_free_put_context: mutex_lock(&counter->mutex); - perf_remove_from_context(counter); + perf_counter_remove_from_context(counter); mutex_unlock(&counter->mutex); err_free_put_context: @@ -783,40 +933,40 @@ err_free_put_context: err_put_context: put_context(ctx); - return ret; + goto out_fput; } -static void __cpuinit perf_init_cpu(int cpu) +static void __cpuinit perf_counter_init_cpu(int cpu) { - struct perf_cpu_context *ctx; + struct perf_cpu_context *cpuctx; - ctx = &per_cpu(perf_cpu_context, cpu); - spin_lock_init(&ctx->ctx.lock); - INIT_LIST_HEAD(&ctx->ctx.counters); + cpuctx = &per_cpu(perf_cpu_context, cpu); + __perf_counter_init_context(&cpuctx->ctx, NULL); mutex_lock(&perf_resource_mutex); - ctx->max_pertask = perf_max_counters - perf_reserved_percpu; + cpuctx->max_pertask = perf_max_counters - perf_reserved_percpu; mutex_unlock(&perf_resource_mutex); + hw_perf_counter_setup(); } #ifdef CONFIG_HOTPLUG_CPU -static void __perf_exit_cpu(void *info) +static void __perf_counter_exit_cpu(void *info) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter_context *ctx = &cpuctx->ctx; struct perf_counter *counter, *tmp; - list_for_each_entry_safe(counter, tmp, &ctx->counters, list) - __perf_remove_from_context(counter); + list_for_each_entry_safe(counter, tmp, &ctx->counter_list, list_entry) + __perf_counter_remove_from_context(counter); } -static void perf_exit_cpu(int cpu) +static void perf_counter_exit_cpu(int cpu) { - smp_call_function_single(cpu, __perf_exit_cpu, NULL, 1); + smp_call_function_single(cpu, __perf_counter_exit_cpu, NULL, 1); } #else -static inline void perf_exit_cpu(int cpu) { } +static inline void perf_counter_exit_cpu(int cpu) { } #endif static int __cpuinit @@ -828,12 +978,12 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) case CPU_UP_PREPARE: case CPU_UP_PREPARE_FROZEN: - perf_init_cpu(cpu); + perf_counter_init_cpu(cpu); break; case CPU_DOWN_PREPARE: case CPU_DOWN_PREPARE_FROZEN: - perf_exit_cpu(cpu); + perf_counter_exit_cpu(cpu); break; default: -- cgit v1.2.3 From ccff286d85098ba5438e22aa2ea807fc1e18cf2f Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 11:26:29 +0100 Subject: perf counters: group counter, fixes Impact: bugfix Check that a group does not span outside the context of a CPU or a task. Also, do not allow deep recursive hierarchies. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 39 ++++++++++++++++++++------------------- 1 file changed, 20 insertions(+), 19 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index fa59fe8c02d5..278209c547a8 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -107,9 +107,6 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_del_init(&counter->list_entry); - if (list_empty(&counter->sibling_list)) - return; - /* * If this was a group counter with sibling counters then * upgrade the siblings to singleton counters by adding them @@ -395,9 +392,6 @@ counter_sched_in(struct perf_counter *counter, struct perf_counter_context *ctx, int cpu) { - if (!counter->active) - return; - hw_perf_counter_enable(counter); counter->active = 1; counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ @@ -876,32 +870,39 @@ asmlinkage int sys_perf_counter_open( return -EFAULT; /* - * Look up the group leader: + * Get the target context (task or percpu): + */ + ctx = find_get_context(pid, cpu); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + /* + * Look up the group leader (we will attach this counter to it): */ group_leader = NULL; if (group_fd != -1) { ret = -EINVAL; group_file = fget_light(group_fd, &fput_needed); if (!group_file) - goto out_fput; + goto err_put_context; if (group_file->f_op != &perf_fops) - goto out_fput; + goto err_put_context; group_leader = group_file->private_data; /* - * Do not allow a recursive hierarchy: + * Do not allow a recursive hierarchy (this new sibling + * becoming part of another group-sibling): + */ + if (group_leader->group_leader != group_leader) + goto err_put_context; + /* + * Do not allow to attach to a group in a different + * task or CPU context: */ - if (group_leader->group_leader) - goto out_fput; + if (group_leader->ctx != ctx) + goto err_put_context; } - /* - * Get the target context (task or percpu): - */ - ctx = find_get_context(pid, cpu); - if (IS_ERR(ctx)) - return PTR_ERR(ctx); - ret = -ENOMEM; counter = perf_counter_alloc(&hw_event, cpu, group_leader); if (!counter) -- cgit v1.2.3 From 621a01eac89b5e2f81a4cf576568b31f40a02724 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 12:46:46 +0100 Subject: perf counters: hw driver API Impact: restructure code, introduce hw_ops driver abstraction Introduce this abstraction to handle counter details: struct hw_perf_counter_ops { void (*hw_perf_counter_enable) (struct perf_counter *counter); void (*hw_perf_counter_disable) (struct perf_counter *counter); void (*hw_perf_counter_read) (struct perf_counter *counter); }; This will be useful to support assymetric hw details, and it will also be useful to implement "software counters". (Counters that count kernel managed sw events such as pagefaults, context-switches, wall-clock time or task-local time.) Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 45 ++++++++++++++++++++++++--------------------- 1 file changed, 24 insertions(+), 21 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 278209c547a8..e6e41ca95463 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -37,18 +37,15 @@ static DEFINE_MUTEX(perf_resource_mutex); /* * Architecture provided APIs - weak aliases: */ - -int __weak hw_perf_counter_init(struct perf_counter *counter) +extern __weak struct hw_perf_counter_ops * +hw_perf_counter_init(struct perf_counter *counter) { - return -EINVAL; + return ERR_PTR(-EINVAL); } -void __weak hw_perf_counter_enable(struct perf_counter *counter) { } -void __weak hw_perf_counter_disable(struct perf_counter *counter) { } -void __weak hw_perf_counter_read(struct perf_counter *counter) { } -void __weak hw_perf_disable_all(void) { } -void __weak hw_perf_enable_all(void) { } -void __weak hw_perf_counter_setup(void) { } +void __weak hw_perf_disable_all(void) { } +void __weak hw_perf_enable_all(void) { } +void __weak hw_perf_counter_setup(void) { } #if BITS_PER_LONG == 64 @@ -146,7 +143,7 @@ static void __perf_counter_remove_from_context(void *info) spin_lock(&ctx->lock); if (counter->active) { - hw_perf_counter_disable(counter); + counter->hw_ops->hw_perf_counter_disable(counter); counter->active = 0; ctx->nr_active--; cpuctx->active_oncpu--; @@ -257,7 +254,7 @@ static void __perf_install_in_context(void *info) ctx->nr_counters++; if (cpuctx->active_oncpu < perf_max_counters) { - hw_perf_counter_enable(counter); + counter->hw_ops->hw_perf_counter_enable(counter); counter->active = 1; counter->oncpu = cpu; ctx->nr_active++; @@ -333,7 +330,7 @@ counter_sched_out(struct perf_counter *counter, if (!counter->active) return; - hw_perf_counter_disable(counter); + counter->hw_ops->hw_perf_counter_disable(counter); counter->active = 0; counter->oncpu = -1; @@ -392,7 +389,7 @@ counter_sched_in(struct perf_counter *counter, struct perf_counter_context *ctx, int cpu) { - hw_perf_counter_enable(counter); + counter->hw_ops->hw_perf_counter_enable(counter); counter->active = 1; counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ @@ -509,7 +506,9 @@ void perf_counter_init_task(struct task_struct *task) */ static void __hw_perf_counter_read(void *info) { - hw_perf_counter_read(info); + struct perf_counter *counter = info; + + counter->hw_ops->hw_perf_counter_read(counter); } static u64 perf_counter_read(struct perf_counter *counter) @@ -816,8 +815,10 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu, struct perf_counter *group_leader) { - struct perf_counter *counter = kzalloc(sizeof(*counter), GFP_KERNEL); + struct hw_perf_counter_ops *hw_ops; + struct perf_counter *counter; + counter = kzalloc(sizeof(*counter), GFP_KERNEL); if (!counter) return NULL; @@ -839,6 +840,14 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->hw_event = *hw_event; counter->wakeup_pending = 0; counter->group_leader = group_leader; + counter->hw_ops = NULL; + + hw_ops = hw_perf_counter_init(counter); + if (!hw_ops) { + kfree(counter); + return NULL; + } + counter->hw_ops = hw_ops; return counter; } @@ -908,10 +917,6 @@ asmlinkage int sys_perf_counter_open( if (!counter) goto err_put_context; - ret = hw_perf_counter_init(counter); - if (ret) - goto err_free_put_context; - perf_install_in_context(ctx, counter, cpu); ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); @@ -927,8 +932,6 @@ err_remove_free_put_context: mutex_lock(&counter->mutex); perf_counter_remove_from_context(counter); mutex_unlock(&counter->mutex); - -err_free_put_context: kfree(counter); err_put_context: -- cgit v1.2.3 From 5c92d12411dfe5f0f3d1b1c1e2f756245e6f7249 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 13:21:10 +0100 Subject: perf counters: implement PERF_COUNT_CPU_CLOCK Impact: add new perf-counter type The 'CPU clock' counter counts the amount of CPU clock time that is elapsing, in nanoseconds. (regardless of how much of it the task is spending on a CPU executing) This counter type is a Linux kernel based abstraction, it is available even if the hardware does not support native hardware performance counters. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 95 ++++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 82 insertions(+), 13 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index e6e41ca95463..506286e5ba63 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -37,15 +37,15 @@ static DEFINE_MUTEX(perf_resource_mutex); /* * Architecture provided APIs - weak aliases: */ -extern __weak struct hw_perf_counter_ops * +extern __weak const struct hw_perf_counter_ops * hw_perf_counter_init(struct perf_counter *counter) { return ERR_PTR(-EINVAL); } -void __weak hw_perf_disable_all(void) { } -void __weak hw_perf_enable_all(void) { } -void __weak hw_perf_counter_setup(void) { } +u64 __weak hw_perf_disable_all(void) { return 0; } +void __weak hw_perf_restore_ctrl(u64 ctrl) { } +void __weak hw_perf_counter_setup(void) { } #if BITS_PER_LONG == 64 @@ -58,6 +58,16 @@ static inline u64 perf_counter_read_safe(struct perf_counter *counter) return (u64) atomic64_read(&counter->count); } +void atomic64_counter_set(struct perf_counter *counter, u64 val) +{ + atomic64_set(&counter->count, val); +} + +u64 atomic64_counter_read(struct perf_counter *counter) +{ + return atomic64_read(&counter->count); +} + #else /* @@ -79,6 +89,20 @@ static u64 perf_counter_read_safe(struct perf_counter *counter) return cntl | ((u64) cnth) << 32; } +void atomic64_counter_set(struct perf_counter *counter, u64 val64) +{ + u32 *val32 = (void *)&val64; + + atomic_set(counter->count32 + 0, *(val32 + 0)); + atomic_set(counter->count32 + 1, *(val32 + 1)); +} + +u64 atomic64_counter_read(struct perf_counter *counter) +{ + return atomic_read(counter->count32 + 0) | + (u64) atomic_read(counter->count32 + 1) << 32; +} + #endif static void @@ -131,6 +155,7 @@ static void __perf_counter_remove_from_context(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; + u64 perf_flags; /* * If this is a task context, we need to check whether it is @@ -155,9 +180,9 @@ static void __perf_counter_remove_from_context(void *info) * Protect the list operation against NMI by disabling the * counters on a global level. NOP for non NMI based counters. */ - hw_perf_disable_all(); + perf_flags = hw_perf_disable_all(); list_del_counter(counter, ctx); - hw_perf_enable_all(); + hw_perf_restore_ctrl(perf_flags); if (!ctx->task) { /* @@ -232,6 +257,7 @@ static void __perf_install_in_context(void *info) struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; int cpu = smp_processor_id(); + u64 perf_flags; /* * If this is a task context, we need to check whether it is @@ -247,9 +273,9 @@ static void __perf_install_in_context(void *info) * Protect the list operation against NMI by disabling the * counters on a global level. NOP for non NMI based counters. */ - hw_perf_disable_all(); + perf_flags = hw_perf_disable_all(); list_add_counter(counter, ctx); - hw_perf_enable_all(); + hw_perf_restore_ctrl(perf_flags); ctx->nr_counters++; @@ -457,6 +483,7 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) { struct perf_counter_context *ctx = &curr->perf_counter_ctx; struct perf_counter *counter; + u64 perf_flags; if (likely(!ctx->nr_counters)) return; @@ -468,13 +495,13 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) /* * Rotate the first entry last (works just fine for group counters too): */ - hw_perf_disable_all(); + perf_flags = hw_perf_disable_all(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { list_del(&counter->list_entry); list_add_tail(&counter->list_entry, &ctx->counter_list); break; } - hw_perf_enable_all(); + hw_perf_restore_ctrl(perf_flags); spin_unlock(&ctx->lock); @@ -807,6 +834,42 @@ static const struct file_operations perf_fops = { .poll = perf_poll, }; +static void cpu_clock_perf_counter_enable(struct perf_counter *counter) +{ +} + +static void cpu_clock_perf_counter_disable(struct perf_counter *counter) +{ +} + +static void cpu_clock_perf_counter_read(struct perf_counter *counter) +{ + int cpu = raw_smp_processor_id(); + + atomic64_counter_set(counter, cpu_clock(cpu)); +} + +static const struct hw_perf_counter_ops perf_ops_cpu_clock = { + .hw_perf_counter_enable = cpu_clock_perf_counter_enable, + .hw_perf_counter_disable = cpu_clock_perf_counter_disable, + .hw_perf_counter_read = cpu_clock_perf_counter_read, +}; + +static const struct hw_perf_counter_ops * +sw_perf_counter_init(struct perf_counter *counter) +{ + const struct hw_perf_counter_ops *hw_ops = NULL; + + switch (counter->hw_event.type) { + case PERF_COUNT_CPU_CLOCK: + hw_ops = &perf_ops_cpu_clock; + break; + default: + break; + } + return hw_ops; +} + /* * Allocate and initialize a counter structure */ @@ -815,7 +878,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu, struct perf_counter *group_leader) { - struct hw_perf_counter_ops *hw_ops; + const struct hw_perf_counter_ops *hw_ops; struct perf_counter *counter; counter = kzalloc(sizeof(*counter), GFP_KERNEL); @@ -842,7 +905,13 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->group_leader = group_leader; counter->hw_ops = NULL; - hw_ops = hw_perf_counter_init(counter); + hw_ops = NULL; + if (!hw_event->raw && hw_event->type < 0) + hw_ops = sw_perf_counter_init(counter); + if (!hw_ops) { + hw_ops = hw_perf_counter_init(counter); + } + if (!hw_ops) { kfree(counter); return NULL; @@ -912,7 +981,7 @@ asmlinkage int sys_perf_counter_open( goto err_put_context; } - ret = -ENOMEM; + ret = -EINVAL; counter = perf_counter_alloc(&hw_event, cpu, group_leader); if (!counter) goto err_put_context; -- cgit v1.2.3 From 01b2838c4298c5e0d30b4993c195ac34dd9df61e Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 13:45:51 +0100 Subject: perf counters: consolidate hw_perf save/restore APIs Impact: cleanup Rename them to better match up the usual IRQ disable/enable APIs: hw_perf_disable_all() => hw_perf_save_disable() hw_perf_restore_ctrl() => hw_perf_restore() Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 506286e5ba63..0e93fea17120 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -43,8 +43,8 @@ hw_perf_counter_init(struct perf_counter *counter) return ERR_PTR(-EINVAL); } -u64 __weak hw_perf_disable_all(void) { return 0; } -void __weak hw_perf_restore_ctrl(u64 ctrl) { } +u64 __weak hw_perf_save_disable(void) { return 0; } +void __weak hw_perf_restore(u64 ctrl) { } void __weak hw_perf_counter_setup(void) { } #if BITS_PER_LONG == 64 @@ -180,9 +180,9 @@ static void __perf_counter_remove_from_context(void *info) * Protect the list operation against NMI by disabling the * counters on a global level. NOP for non NMI based counters. */ - perf_flags = hw_perf_disable_all(); + perf_flags = hw_perf_save_disable(); list_del_counter(counter, ctx); - hw_perf_restore_ctrl(perf_flags); + hw_perf_restore(perf_flags); if (!ctx->task) { /* @@ -273,9 +273,9 @@ static void __perf_install_in_context(void *info) * Protect the list operation against NMI by disabling the * counters on a global level. NOP for non NMI based counters. */ - perf_flags = hw_perf_disable_all(); + perf_flags = hw_perf_save_disable(); list_add_counter(counter, ctx); - hw_perf_restore_ctrl(perf_flags); + hw_perf_restore(perf_flags); ctx->nr_counters++; @@ -495,13 +495,13 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) /* * Rotate the first entry last (works just fine for group counters too): */ - perf_flags = hw_perf_disable_all(); + perf_flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { list_del(&counter->list_entry); list_add_tail(&counter->list_entry, &ctx->counter_list); break; } - hw_perf_restore_ctrl(perf_flags); + hw_perf_restore(perf_flags); spin_unlock(&ctx->lock); -- cgit v1.2.3 From bae43c9945ebeef15e7952e317efb02393d3bfc7 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 14:03:20 +0100 Subject: perf counters: implement PERF_COUNT_TASK_CLOCK Impact: add new perf-counter type The 'task clock' counter counts the amount of time a task is executing, in nanoseconds. It stops ticking when a task is scheduled out either due to it blocking, sleeping or it being preempted. This counter type is a Linux kernel based abstraction, it is available even if the hardware does not support native hardware performance counters. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0e93fea17120..a0fe8474ee29 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -855,6 +855,25 @@ static const struct hw_perf_counter_ops perf_ops_cpu_clock = { .hw_perf_counter_read = cpu_clock_perf_counter_read, }; +static void task_clock_perf_counter_enable(struct perf_counter *counter) +{ +} + +static void task_clock_perf_counter_disable(struct perf_counter *counter) +{ +} + +static void task_clock_perf_counter_read(struct perf_counter *counter) +{ + atomic64_counter_set(counter, current->se.sum_exec_runtime); +} + +static const struct hw_perf_counter_ops perf_ops_task_clock = { + .hw_perf_counter_enable = task_clock_perf_counter_enable, + .hw_perf_counter_disable = task_clock_perf_counter_disable, + .hw_perf_counter_read = task_clock_perf_counter_read, +}; + static const struct hw_perf_counter_ops * sw_perf_counter_init(struct perf_counter *counter) { @@ -864,6 +883,9 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_CPU_CLOCK: hw_ops = &perf_ops_cpu_clock; break; + case PERF_COUNT_TASK_CLOCK: + hw_ops = &perf_ops_task_clock; + break; default: break; } -- cgit v1.2.3 From 1d1c7ddbfab358445a542715551301b7fc363e28 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 14:59:31 +0100 Subject: perf counters: add prctl interface to disable/enable counters Add a way for self-monitoring tasks to disable/enable counters summarily, via a prctl: PR_TASK_PERF_COUNTERS_DISABLE 31 PR_TASK_PERF_COUNTERS_ENABLE 32 Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 86 ++++++++++++++++++++++++++++++++++++++++++++++----- kernel/sys.c | 7 +++++ 2 files changed, 86 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index a0fe8474ee29..4e679b91d8bb 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -415,6 +415,9 @@ counter_sched_in(struct perf_counter *counter, struct perf_counter_context *ctx, int cpu) { + if (counter->active == -1) + return; + counter->hw_ops->hw_perf_counter_enable(counter); counter->active = 1; counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ @@ -479,6 +482,79 @@ void perf_counter_task_sched_in(struct task_struct *task, int cpu) cpuctx->task_ctx = ctx; } +int perf_counter_task_disable(void) +{ + struct task_struct *curr = current; + struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter *counter; + u64 perf_flags; + int cpu; + + if (likely(!ctx->nr_counters)) + return 0; + + local_irq_disable(); + cpu = smp_processor_id(); + + perf_counter_task_sched_out(curr, cpu); + + spin_lock(&ctx->lock); + + /* + * Disable all the counters: + */ + perf_flags = hw_perf_save_disable(); + + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + WARN_ON_ONCE(counter->active == 1); + counter->active = -1; + } + hw_perf_restore(perf_flags); + + spin_unlock(&ctx->lock); + + local_irq_enable(); + + return 0; +} + +int perf_counter_task_enable(void) +{ + struct task_struct *curr = current; + struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter *counter; + u64 perf_flags; + int cpu; + + if (likely(!ctx->nr_counters)) + return 0; + + local_irq_disable(); + cpu = smp_processor_id(); + + spin_lock(&ctx->lock); + + /* + * Disable all the counters: + */ + perf_flags = hw_perf_save_disable(); + + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (counter->active != -1) + continue; + counter->active = 0; + } + hw_perf_restore(perf_flags); + + spin_unlock(&ctx->lock); + + perf_counter_task_sched_in(curr, cpu); + + local_irq_enable(); + + return 0; +} + void perf_counter_task_tick(struct task_struct *curr, int cpu) { struct perf_counter_context *ctx = &curr->perf_counter_ctx; @@ -951,13 +1027,9 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, * @cpu: target cpu * @group_fd: group leader counter fd */ -asmlinkage int sys_perf_counter_open( - - struct perf_counter_hw_event *hw_event_uptr __user, - pid_t pid, - int cpu, - int group_fd) - +asmlinkage int +sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, + pid_t pid, int cpu, int group_fd) { struct perf_counter *counter, *group_leader; struct perf_counter_hw_event hw_event; diff --git a/kernel/sys.c b/kernel/sys.c index 31deba8f7d16..0f66633be319 100644 --- a/kernel/sys.c +++ b/kernel/sys.c @@ -14,6 +14,7 @@ #include #include #include +#include #include #include #include @@ -1716,6 +1717,12 @@ asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3, case PR_SET_TSC: error = SET_TSC_CTL(arg2); break; + case PR_TASK_PERF_COUNTERS_DISABLE: + error = perf_counter_task_disable(); + break; + case PR_TASK_PERF_COUNTERS_ENABLE: + error = perf_counter_task_enable(); + break; case PR_GET_TIMERSLACK: error = current->timer_slack_ns; break; -- cgit v1.2.3 From 6a930700c8b655a9e25e42fc4adc0b225ebbcefc Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 11 Dec 2008 15:17:03 +0100 Subject: perf counters: clean up state transitions Impact: cleanup Introduce a proper enum for the 3 states of a counter: PERF_COUNTER_STATE_OFF = -1 PERF_COUNTER_STATE_INACTIVE = 0 PERF_COUNTER_STATE_ACTIVE = 1 and rename counter->active to counter->state and propagate the changes everywhere. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 29 ++++++++++++++--------------- 1 file changed, 14 insertions(+), 15 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4e679b91d8bb..559130b8774d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -167,9 +167,9 @@ static void __perf_counter_remove_from_context(void *info) spin_lock(&ctx->lock); - if (counter->active) { + if (counter->state == PERF_COUNTER_STATE_ACTIVE) { counter->hw_ops->hw_perf_counter_disable(counter); - counter->active = 0; + counter->state = PERF_COUNTER_STATE_INACTIVE; ctx->nr_active--; cpuctx->active_oncpu--; counter->task = NULL; @@ -281,7 +281,7 @@ static void __perf_install_in_context(void *info) if (cpuctx->active_oncpu < perf_max_counters) { counter->hw_ops->hw_perf_counter_enable(counter); - counter->active = 1; + counter->state = PERF_COUNTER_STATE_ACTIVE; counter->oncpu = cpu; ctx->nr_active++; cpuctx->active_oncpu++; @@ -328,7 +328,6 @@ retry: spin_lock_irq(&ctx->lock); /* - * If the context is active and the counter has not been added * we need to retry the smp call. */ if (ctx->nr_active && list_empty(&counter->list_entry)) { @@ -353,12 +352,12 @@ counter_sched_out(struct perf_counter *counter, struct perf_cpu_context *cpuctx, struct perf_counter_context *ctx) { - if (!counter->active) + if (counter->state != PERF_COUNTER_STATE_ACTIVE) return; counter->hw_ops->hw_perf_counter_disable(counter); - counter->active = 0; - counter->oncpu = -1; + counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->oncpu = -1; cpuctx->active_oncpu--; ctx->nr_active--; @@ -415,11 +414,11 @@ counter_sched_in(struct perf_counter *counter, struct perf_counter_context *ctx, int cpu) { - if (counter->active == -1) + if (counter->state == PERF_COUNTER_STATE_OFF) return; counter->hw_ops->hw_perf_counter_enable(counter); - counter->active = 1; + counter->state = PERF_COUNTER_STATE_ACTIVE; counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ cpuctx->active_oncpu++; @@ -506,8 +505,8 @@ int perf_counter_task_disable(void) perf_flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { - WARN_ON_ONCE(counter->active == 1); - counter->active = -1; + WARN_ON_ONCE(counter->state == PERF_COUNTER_STATE_ACTIVE); + counter->state = PERF_COUNTER_STATE_OFF; } hw_perf_restore(perf_flags); @@ -540,9 +539,9 @@ int perf_counter_task_enable(void) perf_flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->active != -1) + if (counter->state != PERF_COUNTER_STATE_OFF) continue; - counter->active = 0; + counter->state = PERF_COUNTER_STATE_INACTIVE; } hw_perf_restore(perf_flags); @@ -620,7 +619,7 @@ static u64 perf_counter_read(struct perf_counter *counter) * If counter is enabled and currently active on a CPU, update the * value in the counter structure: */ - if (counter->active) { + if (counter->state == PERF_COUNTER_STATE_ACTIVE) { smp_call_function_single(counter->oncpu, __hw_perf_counter_read, counter, 1); } @@ -673,7 +672,7 @@ static struct perf_data *perf_switch_irq_data(struct perf_counter *counter) retry: spin_lock_irq(&ctx->lock); - if (!counter->active) { + if (counter->state != PERF_COUNTER_STATE_ACTIVE) { counter->irqdata = counter->usrdata; counter->usrdata = oldirqdata; spin_unlock_irq(&ctx->lock); -- cgit v1.2.3 From ee06094f8279e1312fc0a31591320cc7b6f0ab1e Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sat, 13 Dec 2008 09:00:03 +0100 Subject: perfcounters: restructure x86 counter math Impact: restructure code Change counter math from absolute values to clear delta logic. We try to extract elapsed deltas from the raw hw counter - and put that into the generic counter. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 68 ++++----------------------------------------------- 1 file changed, 5 insertions(+), 63 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 559130b8774d..416861ce8b27 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -44,67 +44,9 @@ hw_perf_counter_init(struct perf_counter *counter) } u64 __weak hw_perf_save_disable(void) { return 0; } -void __weak hw_perf_restore(u64 ctrl) { } +void __weak hw_perf_restore(u64 ctrl) { } void __weak hw_perf_counter_setup(void) { } -#if BITS_PER_LONG == 64 - -/* - * Read the cached counter in counter safe against cross CPU / NMI - * modifications. 64 bit version - no complications. - */ -static inline u64 perf_counter_read_safe(struct perf_counter *counter) -{ - return (u64) atomic64_read(&counter->count); -} - -void atomic64_counter_set(struct perf_counter *counter, u64 val) -{ - atomic64_set(&counter->count, val); -} - -u64 atomic64_counter_read(struct perf_counter *counter) -{ - return atomic64_read(&counter->count); -} - -#else - -/* - * Read the cached counter in counter safe against cross CPU / NMI - * modifications. 32 bit version. - */ -static u64 perf_counter_read_safe(struct perf_counter *counter) -{ - u32 cntl, cnth; - - local_irq_disable(); - do { - cnth = atomic_read(&counter->count32[1]); - cntl = atomic_read(&counter->count32[0]); - } while (cnth != atomic_read(&counter->count32[1])); - - local_irq_enable(); - - return cntl | ((u64) cnth) << 32; -} - -void atomic64_counter_set(struct perf_counter *counter, u64 val64) -{ - u32 *val32 = (void *)&val64; - - atomic_set(counter->count32 + 0, *(val32 + 0)); - atomic_set(counter->count32 + 1, *(val32 + 1)); -} - -u64 atomic64_counter_read(struct perf_counter *counter) -{ - return atomic_read(counter->count32 + 0) | - (u64) atomic_read(counter->count32 + 1) << 32; -} - -#endif - static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { @@ -280,11 +222,11 @@ static void __perf_install_in_context(void *info) ctx->nr_counters++; if (cpuctx->active_oncpu < perf_max_counters) { - counter->hw_ops->hw_perf_counter_enable(counter); counter->state = PERF_COUNTER_STATE_ACTIVE; counter->oncpu = cpu; ctx->nr_active++; cpuctx->active_oncpu++; + counter->hw_ops->hw_perf_counter_enable(counter); } if (!ctx->task && cpuctx->max_pertask) @@ -624,7 +566,7 @@ static u64 perf_counter_read(struct perf_counter *counter) __hw_perf_counter_read, counter, 1); } - return perf_counter_read_safe(counter); + return atomic64_read(&counter->count); } /* @@ -921,7 +863,7 @@ static void cpu_clock_perf_counter_read(struct perf_counter *counter) { int cpu = raw_smp_processor_id(); - atomic64_counter_set(counter, cpu_clock(cpu)); + atomic64_set(&counter->count, cpu_clock(cpu)); } static const struct hw_perf_counter_ops perf_ops_cpu_clock = { @@ -940,7 +882,7 @@ static void task_clock_perf_counter_disable(struct perf_counter *counter) static void task_clock_perf_counter_read(struct perf_counter *counter) { - atomic64_counter_set(counter, current->se.sum_exec_runtime); + atomic64_set(&counter->count, current->se.sum_exec_runtime); } static const struct hw_perf_counter_ops perf_ops_task_clock = { -- cgit v1.2.3 From 9b51f66dcb09ac5eb6bc68fc111d5c7a1e0131d6 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 12 Dec 2008 13:49:45 +0100 Subject: perfcounters: implement "counter inheritance" Impact: implement new performance feature Counter inheritance can be used to run performance counters in a workload, transparently - and pipe back the counter results to the parent counter. Inheritance for performance counters works the following way: when creating a counter it can be marked with the .inherit=1 flag. Such counters are then 'inherited' by all child tasks (be they fork()-ed or clone()-ed). These counters get inherited through exec() boundaries as well (except through setuid boundaries). The counter values get added back to the parent counter(s) when the child task(s) exit - much like stime/utime statistics are gathered. So inherited counters are ideal to gather summary statistics about an application's behavior via shell commands, without having to modify that application. The timec.c command utilizes counter inheritance: http://redhat.com/~mingo/perfcounters/timec.c Sample output: $ ./timec -e 1 -e 3 -e 5 ls -lR /usr/include/ >/dev/null Performance counter stats for 'ls': 163516953 instructions 2295 cache-misses 2855182 branch-misses Signed-off-by: Ingo Molnar --- kernel/exit.c | 7 +- kernel/perf_counter.c | 248 +++++++++++++++++++++++++++++++++++++++++--------- 2 files changed, 210 insertions(+), 45 deletions(-) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index 2d8be7ebb0f7..d336c90a5f13 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -1093,11 +1093,12 @@ NORET_TYPE void do_exit(long code) mpol_put(tsk->mempolicy); tsk->mempolicy = NULL; #endif -#ifdef CONFIG_FUTEX /* - * This must happen late, after the PID is not - * hashed anymore: + * These must happen late, after the PID is not + * hashed anymore, but still at a point that may sleep: */ + perf_counter_exit_task(tsk); +#ifdef CONFIG_FUTEX if (unlikely(!list_empty(&tsk->pi_state_list))) exit_pi_state_list(tsk); if (unlikely(current->pi_state_cache)) diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 416861ce8b27..f5e81dd193d1 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -80,8 +80,6 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_del_init(&sibling->list_entry); list_add_tail(&sibling->list_entry, &ctx->counter_list); - WARN_ON_ONCE(!sibling->group_leader); - WARN_ON_ONCE(sibling->group_leader == sibling); sibling->group_leader = sibling; } } @@ -97,6 +95,7 @@ static void __perf_counter_remove_from_context(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; + unsigned long flags; u64 perf_flags; /* @@ -107,7 +106,7 @@ static void __perf_counter_remove_from_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + spin_lock_irqsave(&ctx->lock, flags); if (counter->state == PERF_COUNTER_STATE_ACTIVE) { counter->hw_ops->hw_perf_counter_disable(counter); @@ -136,7 +135,7 @@ static void __perf_counter_remove_from_context(void *info) perf_max_counters - perf_reserved_percpu); } - spin_unlock(&ctx->lock); + spin_unlock_irqrestore(&ctx->lock, flags); } @@ -199,6 +198,7 @@ static void __perf_install_in_context(void *info) struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; int cpu = smp_processor_id(); + unsigned long flags; u64 perf_flags; /* @@ -209,7 +209,7 @@ static void __perf_install_in_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock(&ctx->lock); + spin_lock_irqsave(&ctx->lock, flags); /* * Protect the list operation against NMI by disabling the @@ -232,7 +232,7 @@ static void __perf_install_in_context(void *info) if (!ctx->task && cpuctx->max_pertask) cpuctx->max_pertask--; - spin_unlock(&ctx->lock); + spin_unlock_irqrestore(&ctx->lock, flags); } /* @@ -446,10 +446,9 @@ int perf_counter_task_disable(void) */ perf_flags = hw_perf_save_disable(); - list_for_each_entry(counter, &ctx->counter_list, list_entry) { - WARN_ON_ONCE(counter->state == PERF_COUNTER_STATE_ACTIVE); + list_for_each_entry(counter, &ctx->counter_list, list_entry) counter->state = PERF_COUNTER_STATE_OFF; - } + hw_perf_restore(perf_flags); spin_unlock(&ctx->lock); @@ -525,26 +524,6 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) perf_counter_task_sched_in(curr, cpu); } -/* - * Initialize the perf_counter context in a task_struct: - */ -static void -__perf_counter_init_context(struct perf_counter_context *ctx, - struct task_struct *task) -{ - spin_lock_init(&ctx->lock); - INIT_LIST_HEAD(&ctx->counter_list); - ctx->nr_counters = 0; - ctx->task = task; -} -/* - * Initialize the perf_counter context in task_struct - */ -void perf_counter_init_task(struct task_struct *task) -{ - __perf_counter_init_context(&task->perf_counter_ctx, task); -} - /* * Cross CPU call to read the hardware counter */ @@ -663,7 +642,6 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) cpuctx = &per_cpu(perf_cpu_context, cpu); ctx = &cpuctx->ctx; - WARN_ON_ONCE(ctx->task); return ctx; } @@ -915,12 +893,13 @@ sw_perf_counter_init(struct perf_counter *counter) static struct perf_counter * perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu, - struct perf_counter *group_leader) + struct perf_counter *group_leader, + gfp_t gfpflags) { const struct hw_perf_counter_ops *hw_ops; struct perf_counter *counter; - counter = kzalloc(sizeof(*counter), GFP_KERNEL); + counter = kzalloc(sizeof(*counter), gfpflags); if (!counter) return NULL; @@ -947,9 +926,8 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, hw_ops = NULL; if (!hw_event->raw && hw_event->type < 0) hw_ops = sw_perf_counter_init(counter); - if (!hw_ops) { + if (!hw_ops) hw_ops = hw_perf_counter_init(counter); - } if (!hw_ops) { kfree(counter); @@ -975,8 +953,10 @@ sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, struct perf_counter *counter, *group_leader; struct perf_counter_hw_event hw_event; struct perf_counter_context *ctx; + struct file *counter_file = NULL; struct file *group_file = NULL; int fput_needed = 0; + int fput_needed2 = 0; int ret; if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0) @@ -1017,25 +997,29 @@ sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, } ret = -EINVAL; - counter = perf_counter_alloc(&hw_event, cpu, group_leader); + counter = perf_counter_alloc(&hw_event, cpu, group_leader, GFP_KERNEL); if (!counter) goto err_put_context; - perf_install_in_context(ctx, counter, cpu); - ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); if (ret < 0) - goto err_remove_free_put_context; + goto err_free_put_context; + + counter_file = fget_light(ret, &fput_needed2); + if (!counter_file) + goto err_free_put_context; + + counter->filp = counter_file; + perf_install_in_context(ctx, counter, cpu); + + fput_light(counter_file, fput_needed2); out_fput: fput_light(group_file, fput_needed); return ret; -err_remove_free_put_context: - mutex_lock(&counter->mutex); - perf_counter_remove_from_context(counter); - mutex_unlock(&counter->mutex); +err_free_put_context: kfree(counter); err_put_context: @@ -1044,6 +1028,186 @@ err_put_context: goto out_fput; } +/* + * Initialize the perf_counter context in a task_struct: + */ +static void +__perf_counter_init_context(struct perf_counter_context *ctx, + struct task_struct *task) +{ + memset(ctx, 0, sizeof(*ctx)); + spin_lock_init(&ctx->lock); + INIT_LIST_HEAD(&ctx->counter_list); + ctx->task = task; +} + +/* + * inherit a counter from parent task to child task: + */ +static int +inherit_counter(struct perf_counter *parent_counter, + struct task_struct *parent, + struct perf_counter_context *parent_ctx, + struct task_struct *child, + struct perf_counter_context *child_ctx) +{ + struct perf_counter *child_counter; + + child_counter = perf_counter_alloc(&parent_counter->hw_event, + parent_counter->cpu, NULL, + GFP_ATOMIC); + if (!child_counter) + return -ENOMEM; + + /* + * Link it up in the child's context: + */ + child_counter->ctx = child_ctx; + child_counter->task = child; + list_add_counter(child_counter, child_ctx); + child_ctx->nr_counters++; + + child_counter->parent = parent_counter; + parent_counter->nr_inherited++; + /* + * inherit into child's child as well: + */ + child_counter->hw_event.inherit = 1; + + /* + * Get a reference to the parent filp - we will fput it + * when the child counter exits. This is safe to do because + * we are in the parent and we know that the filp still + * exists and has a nonzero count: + */ + atomic_long_inc(&parent_counter->filp->f_count); + + return 0; +} + +static void +__perf_counter_exit_task(struct task_struct *child, + struct perf_counter *child_counter, + struct perf_counter_context *child_ctx) +{ + struct perf_counter *parent_counter; + u64 parent_val, child_val; + u64 perf_flags; + + /* + * Disable and unlink this counter. + * + * Be careful about zapping the list - IRQ/NMI context + * could still be processing it: + */ + local_irq_disable(); + perf_flags = hw_perf_save_disable(); + + if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) + child_counter->hw_ops->hw_perf_counter_disable(child_counter); + list_del_init(&child_counter->list_entry); + + hw_perf_restore(perf_flags); + local_irq_enable(); + + parent_counter = child_counter->parent; + /* + * It can happen that parent exits first, and has counters + * that are still around due to the child reference. These + * counters need to be zapped - but otherwise linger. + */ + if (!parent_counter) + return; + + parent_val = atomic64_read(&parent_counter->count); + child_val = atomic64_read(&child_counter->count); + + /* + * Add back the child's count to the parent's count: + */ + atomic64_add(child_val, &parent_counter->count); + + fput(parent_counter->filp); + + kfree(child_counter); +} + +/* + * When a child task exist, feed back counter values to parent counters. + * + * Note: we are running in child context, but the PID is not hashed + * anymore so new counters will not be added. + */ +void perf_counter_exit_task(struct task_struct *child) +{ + struct perf_counter *child_counter, *tmp; + struct perf_counter_context *child_ctx; + + child_ctx = &child->perf_counter_ctx; + + if (likely(!child_ctx->nr_counters)) + return; + + list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, + list_entry) + __perf_counter_exit_task(child, child_counter, child_ctx); +} + +/* + * Initialize the perf_counter context in task_struct + */ +void perf_counter_init_task(struct task_struct *child) +{ + struct perf_counter_context *child_ctx, *parent_ctx; + struct perf_counter *counter, *parent_counter; + struct task_struct *parent = current; + unsigned long flags; + + child_ctx = &child->perf_counter_ctx; + parent_ctx = &parent->perf_counter_ctx; + + __perf_counter_init_context(child_ctx, child); + + /* + * This is executed from the parent task context, so inherit + * counters that have been marked for cloning: + */ + + if (likely(!parent_ctx->nr_counters)) + return; + + /* + * Lock the parent list. No need to lock the child - not PID + * hashed yet and not running, so nobody can access it. + */ + spin_lock_irqsave(&parent_ctx->lock, flags); + + /* + * We dont have to disable NMIs - we are only looking at + * the list, not manipulating it: + */ + list_for_each_entry(counter, &parent_ctx->counter_list, list_entry) { + if (!counter->hw_event.inherit || counter->group_leader != counter) + continue; + + /* + * Instead of creating recursive hierarchies of counters, + * we link inheritd counters back to the original parent, + * which has a filp for sure, which we use as the reference + * count: + */ + parent_counter = counter; + if (counter->parent) + parent_counter = counter->parent; + + if (inherit_counter(parent_counter, parent, + parent_ctx, child, child_ctx)) + break; + } + + spin_unlock_irqrestore(&parent_ctx->lock, flags); +} + static void __cpuinit perf_counter_init_cpu(int cpu) { struct perf_cpu_context *cpuctx; -- cgit v1.2.3 From 8cb391e8786c8072367f0aeb90551903fef074ba Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 12:22:31 +0100 Subject: perfcounters: fix task clock counter Impact: bugfix Update the task clock counter to the new math. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 27 +++++++++++++++++++++++---- 1 file changed, 23 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f5e81dd193d1..1f81cde0dc43 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -850,17 +850,36 @@ static const struct hw_perf_counter_ops perf_ops_cpu_clock = { .hw_perf_counter_read = cpu_clock_perf_counter_read, }; -static void task_clock_perf_counter_enable(struct perf_counter *counter) +static void task_clock_perf_counter_update(struct perf_counter *counter) { + u64 prev, now; + s64 delta; + + prev = atomic64_read(&counter->hw.prev_count); + now = current->se.sum_exec_runtime; + + atomic64_set(&counter->hw.prev_count, now); + + delta = now - prev; + if (WARN_ON_ONCE(delta < 0)) + delta = 0; + + atomic64_add(delta, &counter->count); } -static void task_clock_perf_counter_disable(struct perf_counter *counter) +static void task_clock_perf_counter_read(struct perf_counter *counter) { + task_clock_perf_counter_update(counter); } -static void task_clock_perf_counter_read(struct perf_counter *counter) +static void task_clock_perf_counter_enable(struct perf_counter *counter) +{ + atomic64_set(&counter->hw.prev_count, current->se.sum_exec_runtime); +} + +static void task_clock_perf_counter_disable(struct perf_counter *counter) { - atomic64_set(&counter->count, current->se.sum_exec_runtime); + task_clock_perf_counter_update(counter); } static const struct hw_perf_counter_ops perf_ops_task_clock = { -- cgit v1.2.3 From 5d6a27d8a096868ae313f71f563b06074a7e34fe Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 12:28:33 +0100 Subject: perfcounters: add context switch counter Impact: add new feature, new sw counter Add a counter that counts the number of context-switches a task is doing. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 51 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 1f81cde0dc43..09287091c526 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -888,6 +888,54 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .hw_perf_counter_read = task_clock_perf_counter_read, }; +static u64 get_context_switches(void) +{ + struct task_struct *curr = current; + + return curr->nvcsw + curr->nivcsw; +} + +static void context_switches_perf_counter_update(struct perf_counter *counter) +{ + u64 prev, now; + s64 delta; + + prev = atomic64_read(&counter->hw.prev_count); + now = get_context_switches(); + + atomic64_set(&counter->hw.prev_count, now); + + delta = now - prev; + if (WARN_ON_ONCE(delta < 0)) + delta = 0; + + atomic64_add(delta, &counter->count); +} + +static void context_switches_perf_counter_read(struct perf_counter *counter) +{ + context_switches_perf_counter_update(counter); +} + +static void context_switches_perf_counter_enable(struct perf_counter *counter) +{ + /* + * ->nvcsw + curr->nivcsw is a per-task value already, + * so we dont have to clear it on switch-in. + */ +} + +static void context_switches_perf_counter_disable(struct perf_counter *counter) +{ + context_switches_perf_counter_update(counter); +} + +static const struct hw_perf_counter_ops perf_ops_context_switches = { + .hw_perf_counter_enable = context_switches_perf_counter_enable, + .hw_perf_counter_disable = context_switches_perf_counter_disable, + .hw_perf_counter_read = context_switches_perf_counter_read, +}; + static const struct hw_perf_counter_ops * sw_perf_counter_init(struct perf_counter *counter) { @@ -900,6 +948,9 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_TASK_CLOCK: hw_ops = &perf_ops_task_clock; break; + case PERF_COUNT_CONTEXT_SWITCHES: + hw_ops = &perf_ops_context_switches; + break; default: break; } -- cgit v1.2.3 From 6c594c21fcb02c662f11c97be4d7d2b73060a205 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 12:34:15 +0100 Subject: perfcounters: add task migrations counter Impact: add new feature, new sw counter Add a counter that counts the number of cross-CPU migrations a task is suffering. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++++ kernel/sched.c | 7 +++++-- 2 files changed, 54 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 09287091c526..fb11e351e44e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -936,6 +936,52 @@ static const struct hw_perf_counter_ops perf_ops_context_switches = { .hw_perf_counter_read = context_switches_perf_counter_read, }; +static inline u64 get_cpu_migrations(void) +{ + return current->se.nr_migrations; +} + +static void cpu_migrations_perf_counter_update(struct perf_counter *counter) +{ + u64 prev, now; + s64 delta; + + prev = atomic64_read(&counter->hw.prev_count); + now = get_cpu_migrations(); + + atomic64_set(&counter->hw.prev_count, now); + + delta = now - prev; + if (WARN_ON_ONCE(delta < 0)) + delta = 0; + + atomic64_add(delta, &counter->count); +} + +static void cpu_migrations_perf_counter_read(struct perf_counter *counter) +{ + cpu_migrations_perf_counter_update(counter); +} + +static void cpu_migrations_perf_counter_enable(struct perf_counter *counter) +{ + /* + * se.nr_migrations is a per-task value already, + * so we dont have to clear it on switch-in. + */ +} + +static void cpu_migrations_perf_counter_disable(struct perf_counter *counter) +{ + cpu_migrations_perf_counter_update(counter); +} + +static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { + .hw_perf_counter_enable = cpu_migrations_perf_counter_enable, + .hw_perf_counter_disable = cpu_migrations_perf_counter_disable, + .hw_perf_counter_read = cpu_migrations_perf_counter_read, +}; + static const struct hw_perf_counter_ops * sw_perf_counter_init(struct perf_counter *counter) { @@ -951,6 +997,9 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_CONTEXT_SWITCHES: hw_ops = &perf_ops_context_switches; break; + case PERF_COUNT_CPU_MIGRATIONS: + hw_ops = &perf_ops_cpu_migrations; + break; default: break; } diff --git a/kernel/sched.c b/kernel/sched.c index 5c3f4106314e..382cfdb5e38d 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -1852,12 +1852,14 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) p->se.sleep_start -= clock_offset; if (p->se.block_start) p->se.block_start -= clock_offset; +#endif if (old_cpu != new_cpu) { - schedstat_inc(p, se.nr_migrations); + p->se.nr_migrations++; +#ifdef CONFIG_SCHEDSTATS if (task_hot(p, old_rq->clock, NULL)) schedstat_inc(p, se.nr_forced2_migrations); - } #endif + } p->se.vruntime -= old_cfsrq->min_vruntime - new_cfsrq->min_vruntime; @@ -2375,6 +2377,7 @@ static void __sched_fork(struct task_struct *p) p->se.exec_start = 0; p->se.sum_exec_runtime = 0; p->se.prev_sum_exec_runtime = 0; + p->se.nr_migrations = 0; p->se.last_wakeup = 0; p->se.avg_overlap = 0; -- cgit v1.2.3 From e06c61a879910869aa5bf3f8f634abfee1a7bebc Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 14:44:31 +0100 Subject: perfcounters: add nr-of-faults counter Impact: add new feature, new sw counter Add a counter that counts the number of pagefaults a task is experiencing. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 51 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index fb11e351e44e..59c52f9ee431 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -888,6 +888,54 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .hw_perf_counter_read = task_clock_perf_counter_read, }; +static u64 get_page_faults(void) +{ + struct task_struct *curr = current; + + return curr->maj_flt + curr->min_flt; +} + +static void page_faults_perf_counter_update(struct perf_counter *counter) +{ + u64 prev, now; + s64 delta; + + prev = atomic64_read(&counter->hw.prev_count); + now = get_page_faults(); + + atomic64_set(&counter->hw.prev_count, now); + + delta = now - prev; + if (WARN_ON_ONCE(delta < 0)) + delta = 0; + + atomic64_add(delta, &counter->count); +} + +static void page_faults_perf_counter_read(struct perf_counter *counter) +{ + page_faults_perf_counter_update(counter); +} + +static void page_faults_perf_counter_enable(struct perf_counter *counter) +{ + /* + * page-faults is a per-task value already, + * so we dont have to clear it on switch-in. + */ +} + +static void page_faults_perf_counter_disable(struct perf_counter *counter) +{ + page_faults_perf_counter_update(counter); +} + +static const struct hw_perf_counter_ops perf_ops_page_faults = { + .hw_perf_counter_enable = page_faults_perf_counter_enable, + .hw_perf_counter_disable = page_faults_perf_counter_disable, + .hw_perf_counter_read = page_faults_perf_counter_read, +}; + static u64 get_context_switches(void) { struct task_struct *curr = current; @@ -994,6 +1042,9 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_TASK_CLOCK: hw_ops = &perf_ops_task_clock; break; + case PERF_COUNT_PAGE_FAULTS: + hw_ops = &perf_ops_page_faults; + break; case PERF_COUNT_CONTEXT_SWITCHES: hw_ops = &perf_ops_context_switches; break; -- cgit v1.2.3 From 088e2852c858159d47f71ee8da38e0fb1b21f806 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 20:21:00 +0100 Subject: perfcounters, x86: fix sw counters on non-PMC CPUs Make perf_max_counters default to at least 1 - this allows the sw counters to be used. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 59c52f9ee431..539fa8283a06 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -25,7 +25,7 @@ */ DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); -int perf_max_counters __read_mostly; +int perf_max_counters __read_mostly = 1; static int perf_reserved_percpu __read_mostly; static int perf_overcommit __read_mostly = 1; -- cgit v1.2.3 From 0cc0c027d4e028632933f1be2dc4cd730358183b Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 14 Dec 2008 23:20:36 +0100 Subject: perfcounters: release CPU context when exiting task counters If counters are exiting via do_exit() not via filp close, then the CPU context needs to be released - otherwise future percpu counter creations might fail. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 539fa8283a06..16396e9406fa 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1273,8 +1273,19 @@ __perf_counter_exit_task(struct task_struct *child, local_irq_disable(); perf_flags = hw_perf_save_disable(); - if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) + if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) { + struct perf_cpu_context *cpuctx; + + cpuctx = &__get_cpu_var(perf_cpu_context); + child_counter->hw_ops->hw_perf_counter_disable(child_counter); + child_counter->state = PERF_COUNTER_STATE_INACTIVE; + child_counter->oncpu = -1; + + cpuctx->active_oncpu--; + child_ctx->nr_active--; + } + list_del_init(&child_counter->list_entry); hw_perf_restore(perf_flags); @@ -1539,4 +1550,3 @@ static int __init perf_counter_sysfs_init(void) &perfclass_attr_group); } device_initcall(perf_counter_sysfs_init); - -- cgit v1.2.3 From a86ed50859d65a08beec9474df97b88438a996df Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 17 Dec 2008 00:43:10 +0100 Subject: perfcounters: use hw_event.disable flag Impact: implement default-off counters Make sure that counters that are created with counter.hw_event.disabled=1, get created in disabled state. They can be enabled via: prctl(PR_TASK_PERF_COUNTERS_ENABLE); Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 16396e9406fa..5431e790b5d6 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1093,6 +1093,9 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->group_leader = group_leader; counter->hw_ops = NULL; + if (hw_event->disabled) + counter->state = PERF_COUNTER_STATE_OFF; + hw_ops = NULL; if (!hw_event->raw && hw_event->type < 0) hw_ops = sw_perf_counter_init(counter); -- cgit v1.2.3 From 8fb9331391af95ca1f4e5c0a0da8120b13cbae01 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 23 Dec 2008 12:04:16 +0100 Subject: perfcounters: remove warnings Impact: remove debug checks Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 8 -------- 1 file changed, 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 5431e790b5d6..aab6c123b02c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -861,8 +861,6 @@ static void task_clock_perf_counter_update(struct perf_counter *counter) atomic64_set(&counter->hw.prev_count, now); delta = now - prev; - if (WARN_ON_ONCE(delta < 0)) - delta = 0; atomic64_add(delta, &counter->count); } @@ -906,8 +904,6 @@ static void page_faults_perf_counter_update(struct perf_counter *counter) atomic64_set(&counter->hw.prev_count, now); delta = now - prev; - if (WARN_ON_ONCE(delta < 0)) - delta = 0; atomic64_add(delta, &counter->count); } @@ -954,8 +950,6 @@ static void context_switches_perf_counter_update(struct perf_counter *counter) atomic64_set(&counter->hw.prev_count, now); delta = now - prev; - if (WARN_ON_ONCE(delta < 0)) - delta = 0; atomic64_add(delta, &counter->count); } @@ -1000,8 +994,6 @@ static void cpu_migrations_perf_counter_update(struct perf_counter *counter) atomic64_set(&counter->hw.prev_count, now); delta = now - prev; - if (WARN_ON_ONCE(delta < 0)) - delta = 0; atomic64_add(delta, &counter->count); } -- cgit v1.2.3 From 7995888fcb0246543ee8027bf2835a250ba8c925 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 17 Dec 2008 08:54:56 +0100 Subject: perfcounters: tweak group scheduling Impact: schedule in groups atomically If there are multiple groups in a task, make sure they are scheduled in and out atomically. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index aab6c123b02c..f8a4d9a5d5d3 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -367,21 +367,26 @@ counter_sched_in(struct perf_counter *counter, ctx->nr_active++; } -static void +static int group_sched_in(struct perf_counter *group_counter, struct perf_cpu_context *cpuctx, struct perf_counter_context *ctx, int cpu) { struct perf_counter *counter; + int was_group = 0; counter_sched_in(group_counter, cpuctx, ctx, cpu); /* * Schedule in siblings as one group (if any): */ - list_for_each_entry(counter, &group_counter->sibling_list, list_entry) + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { counter_sched_in(counter, cpuctx, ctx, cpu); + was_group = 1; + } + + return was_group; } /* @@ -416,7 +421,12 @@ void perf_counter_task_sched_in(struct task_struct *task, int cpu) if (counter->cpu != -1 && counter->cpu != cpu) continue; - group_sched_in(counter, cpuctx, ctx, cpu); + /* + * If we scheduled in a group atomically and + * exclusively, break out: + */ + if (group_sched_in(counter, cpuctx, ctx, cpu)) + break; } spin_unlock(&ctx->lock); -- cgit v1.2.3 From 7671581f1666ef4b54a1c1e598c51ac44c060a9b Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 17 Dec 2008 14:20:28 +0100 Subject: perfcounters: hw ops rename Impact: rename field names Shorten them. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 50 +++++++++++++++++++++++++------------------------- 1 file changed, 25 insertions(+), 25 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f8a4d9a5d5d3..961d651aa574 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -109,7 +109,7 @@ static void __perf_counter_remove_from_context(void *info) spin_lock_irqsave(&ctx->lock, flags); if (counter->state == PERF_COUNTER_STATE_ACTIVE) { - counter->hw_ops->hw_perf_counter_disable(counter); + counter->hw_ops->disable(counter); counter->state = PERF_COUNTER_STATE_INACTIVE; ctx->nr_active--; cpuctx->active_oncpu--; @@ -226,7 +226,7 @@ static void __perf_install_in_context(void *info) counter->oncpu = cpu; ctx->nr_active++; cpuctx->active_oncpu++; - counter->hw_ops->hw_perf_counter_enable(counter); + counter->hw_ops->enable(counter); } if (!ctx->task && cpuctx->max_pertask) @@ -297,7 +297,7 @@ counter_sched_out(struct perf_counter *counter, if (counter->state != PERF_COUNTER_STATE_ACTIVE) return; - counter->hw_ops->hw_perf_counter_disable(counter); + counter->hw_ops->disable(counter); counter->state = PERF_COUNTER_STATE_INACTIVE; counter->oncpu = -1; @@ -327,7 +327,7 @@ group_sched_out(struct perf_counter *group_counter, * * We stop each counter and update the counter value in counter->count. * - * This does not protect us against NMI, but hw_perf_counter_disable() + * This does not protect us against NMI, but disable() * sets the disabled bit in the control field of counter _before_ * accessing the counter control register. If a NMI hits, then it will * not restart the counter. @@ -359,7 +359,7 @@ counter_sched_in(struct perf_counter *counter, if (counter->state == PERF_COUNTER_STATE_OFF) return; - counter->hw_ops->hw_perf_counter_enable(counter); + counter->hw_ops->enable(counter); counter->state = PERF_COUNTER_STATE_ACTIVE; counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ @@ -395,7 +395,7 @@ group_sched_in(struct perf_counter *group_counter, * * We restore the counter value and then enable it. * - * This does not protect us against NMI, but hw_perf_counter_enable() + * This does not protect us against NMI, but enable() * sets the enabled bit in the control field of counter _before_ * accessing the counter control register. If a NMI hits, then it will * keep the counter running. @@ -537,11 +537,11 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) /* * Cross CPU call to read the hardware counter */ -static void __hw_perf_counter_read(void *info) +static void __read(void *info) { struct perf_counter *counter = info; - counter->hw_ops->hw_perf_counter_read(counter); + counter->hw_ops->read(counter); } static u64 perf_counter_read(struct perf_counter *counter) @@ -552,7 +552,7 @@ static u64 perf_counter_read(struct perf_counter *counter) */ if (counter->state == PERF_COUNTER_STATE_ACTIVE) { smp_call_function_single(counter->oncpu, - __hw_perf_counter_read, counter, 1); + __read, counter, 1); } return atomic64_read(&counter->count); @@ -855,9 +855,9 @@ static void cpu_clock_perf_counter_read(struct perf_counter *counter) } static const struct hw_perf_counter_ops perf_ops_cpu_clock = { - .hw_perf_counter_enable = cpu_clock_perf_counter_enable, - .hw_perf_counter_disable = cpu_clock_perf_counter_disable, - .hw_perf_counter_read = cpu_clock_perf_counter_read, + .enable = cpu_clock_perf_counter_enable, + .disable = cpu_clock_perf_counter_disable, + .read = cpu_clock_perf_counter_read, }; static void task_clock_perf_counter_update(struct perf_counter *counter) @@ -891,9 +891,9 @@ static void task_clock_perf_counter_disable(struct perf_counter *counter) } static const struct hw_perf_counter_ops perf_ops_task_clock = { - .hw_perf_counter_enable = task_clock_perf_counter_enable, - .hw_perf_counter_disable = task_clock_perf_counter_disable, - .hw_perf_counter_read = task_clock_perf_counter_read, + .enable = task_clock_perf_counter_enable, + .disable = task_clock_perf_counter_disable, + .read = task_clock_perf_counter_read, }; static u64 get_page_faults(void) @@ -937,9 +937,9 @@ static void page_faults_perf_counter_disable(struct perf_counter *counter) } static const struct hw_perf_counter_ops perf_ops_page_faults = { - .hw_perf_counter_enable = page_faults_perf_counter_enable, - .hw_perf_counter_disable = page_faults_perf_counter_disable, - .hw_perf_counter_read = page_faults_perf_counter_read, + .enable = page_faults_perf_counter_enable, + .disable = page_faults_perf_counter_disable, + .read = page_faults_perf_counter_read, }; static u64 get_context_switches(void) @@ -983,9 +983,9 @@ static void context_switches_perf_counter_disable(struct perf_counter *counter) } static const struct hw_perf_counter_ops perf_ops_context_switches = { - .hw_perf_counter_enable = context_switches_perf_counter_enable, - .hw_perf_counter_disable = context_switches_perf_counter_disable, - .hw_perf_counter_read = context_switches_perf_counter_read, + .enable = context_switches_perf_counter_enable, + .disable = context_switches_perf_counter_disable, + .read = context_switches_perf_counter_read, }; static inline u64 get_cpu_migrations(void) @@ -1027,9 +1027,9 @@ static void cpu_migrations_perf_counter_disable(struct perf_counter *counter) } static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { - .hw_perf_counter_enable = cpu_migrations_perf_counter_enable, - .hw_perf_counter_disable = cpu_migrations_perf_counter_disable, - .hw_perf_counter_read = cpu_migrations_perf_counter_read, + .enable = cpu_migrations_perf_counter_enable, + .disable = cpu_migrations_perf_counter_disable, + .read = cpu_migrations_perf_counter_read, }; static const struct hw_perf_counter_ops * @@ -1283,7 +1283,7 @@ __perf_counter_exit_task(struct task_struct *child, cpuctx = &__get_cpu_var(perf_cpu_context); - child_counter->hw_ops->hw_perf_counter_disable(child_counter); + child_counter->hw_ops->disable(child_counter); child_counter->state = PERF_COUNTER_STATE_INACTIVE; child_counter->oncpu = -1; -- cgit v1.2.3 From aa9c4c0f967fdb482ea95e8473ec3d201e6e0781 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 17 Dec 2008 14:10:57 +0100 Subject: perfcounters: fix task clock counter Impact: fix per task clock counter precision Signed-off-by: Ingo Molnar --- kernel/exit.c | 17 +++++++++---- kernel/perf_counter.c | 70 +++++++++++++++++++++++++++++++++++++++------------ kernel/sched.c | 49 +++++++++++++++++++++++++++++++++--- 3 files changed, 112 insertions(+), 24 deletions(-) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index d336c90a5f13..244edfd96865 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -922,6 +922,12 @@ static void exit_notify(struct task_struct *tsk, int group_dead) forget_original_parent(tsk); exit_task_namespaces(tsk); + /* + * Flush inherited counters to the parent - before the parent + * gets woken up by child-exit notifications. + */ + perf_counter_exit_task(tsk); + write_lock_irq(&tasklist_lock); if (group_dead) kill_orphaned_pgrp(tsk->group_leader, NULL); @@ -1093,11 +1099,6 @@ NORET_TYPE void do_exit(long code) mpol_put(tsk->mempolicy); tsk->mempolicy = NULL; #endif - /* - * These must happen late, after the PID is not - * hashed anymore, but still at a point that may sleep: - */ - perf_counter_exit_task(tsk); #ifdef CONFIG_FUTEX if (unlikely(!list_empty(&tsk->pi_state_list))) exit_pi_state_list(tsk); @@ -1121,6 +1122,12 @@ NORET_TYPE void do_exit(long code) if (tsk->splice_pipe) __free_pipe_info(tsk->splice_pipe); + /* + * These must happen late, after the PID is not + * hashed anymore, but still at a point that may sleep: + */ + perf_counter_exit_task(tsk); + preempt_disable(); /* causes final put_task_struct in finish_task_switch(). */ tsk->state = TASK_DEAD; diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 961d651aa574..f1110ac1267b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -18,6 +18,7 @@ #include #include #include +#include #include /* @@ -106,7 +107,8 @@ static void __perf_counter_remove_from_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock_irqsave(&ctx->lock, flags); + curr_rq_lock_irq_save(&flags); + spin_lock(&ctx->lock); if (counter->state == PERF_COUNTER_STATE_ACTIVE) { counter->hw_ops->disable(counter); @@ -135,7 +137,8 @@ static void __perf_counter_remove_from_context(void *info) perf_max_counters - perf_reserved_percpu); } - spin_unlock_irqrestore(&ctx->lock, flags); + spin_unlock(&ctx->lock); + curr_rq_unlock_irq_restore(&flags); } @@ -209,7 +212,8 @@ static void __perf_install_in_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - spin_lock_irqsave(&ctx->lock, flags); + curr_rq_lock_irq_save(&flags); + spin_lock(&ctx->lock); /* * Protect the list operation against NMI by disabling the @@ -232,7 +236,8 @@ static void __perf_install_in_context(void *info) if (!ctx->task && cpuctx->max_pertask) cpuctx->max_pertask--; - spin_unlock_irqrestore(&ctx->lock, flags); + spin_unlock(&ctx->lock); + curr_rq_unlock_irq_restore(&flags); } /* @@ -438,15 +443,19 @@ int perf_counter_task_disable(void) struct task_struct *curr = current; struct perf_counter_context *ctx = &curr->perf_counter_ctx; struct perf_counter *counter; + unsigned long flags; u64 perf_flags; int cpu; if (likely(!ctx->nr_counters)) return 0; - local_irq_disable(); + curr_rq_lock_irq_save(&flags); cpu = smp_processor_id(); + /* force the update of the task clock: */ + __task_delta_exec(curr, 1); + perf_counter_task_sched_out(curr, cpu); spin_lock(&ctx->lock); @@ -463,7 +472,7 @@ int perf_counter_task_disable(void) spin_unlock(&ctx->lock); - local_irq_enable(); + curr_rq_unlock_irq_restore(&flags); return 0; } @@ -473,15 +482,19 @@ int perf_counter_task_enable(void) struct task_struct *curr = current; struct perf_counter_context *ctx = &curr->perf_counter_ctx; struct perf_counter *counter; + unsigned long flags; u64 perf_flags; int cpu; if (likely(!ctx->nr_counters)) return 0; - local_irq_disable(); + curr_rq_lock_irq_save(&flags); cpu = smp_processor_id(); + /* force the update of the task clock: */ + __task_delta_exec(curr, 1); + spin_lock(&ctx->lock); /* @@ -493,6 +506,7 @@ int perf_counter_task_enable(void) if (counter->state != PERF_COUNTER_STATE_OFF) continue; counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->hw_event.disabled = 0; } hw_perf_restore(perf_flags); @@ -500,7 +514,7 @@ int perf_counter_task_enable(void) perf_counter_task_sched_in(curr, cpu); - local_irq_enable(); + curr_rq_unlock_irq_restore(&flags); return 0; } @@ -540,8 +554,11 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) static void __read(void *info) { struct perf_counter *counter = info; + unsigned long flags; + curr_rq_lock_irq_save(&flags); counter->hw_ops->read(counter); + curr_rq_unlock_irq_restore(&flags); } static u64 perf_counter_read(struct perf_counter *counter) @@ -860,13 +877,27 @@ static const struct hw_perf_counter_ops perf_ops_cpu_clock = { .read = cpu_clock_perf_counter_read, }; -static void task_clock_perf_counter_update(struct perf_counter *counter) +/* + * Called from within the scheduler: + */ +static u64 task_clock_perf_counter_val(struct perf_counter *counter, int update) { - u64 prev, now; + struct task_struct *curr = counter->task; + u64 delta; + + WARN_ON_ONCE(counter->task != current); + + delta = __task_delta_exec(curr, update); + + return curr->se.sum_exec_runtime + delta; +} + +static void task_clock_perf_counter_update(struct perf_counter *counter, u64 now) +{ + u64 prev; s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = current->se.sum_exec_runtime; atomic64_set(&counter->hw.prev_count, now); @@ -877,17 +908,23 @@ static void task_clock_perf_counter_update(struct perf_counter *counter) static void task_clock_perf_counter_read(struct perf_counter *counter) { - task_clock_perf_counter_update(counter); + u64 now = task_clock_perf_counter_val(counter, 1); + + task_clock_perf_counter_update(counter, now); } static void task_clock_perf_counter_enable(struct perf_counter *counter) { - atomic64_set(&counter->hw.prev_count, current->se.sum_exec_runtime); + u64 now = task_clock_perf_counter_val(counter, 0); + + atomic64_set(&counter->hw.prev_count, now); } static void task_clock_perf_counter_disable(struct perf_counter *counter) { - task_clock_perf_counter_update(counter); + u64 now = task_clock_perf_counter_val(counter, 0); + + task_clock_perf_counter_update(counter, now); } static const struct hw_perf_counter_ops perf_ops_task_clock = { @@ -1267,6 +1304,7 @@ __perf_counter_exit_task(struct task_struct *child, { struct perf_counter *parent_counter; u64 parent_val, child_val; + unsigned long flags; u64 perf_flags; /* @@ -1275,7 +1313,7 @@ __perf_counter_exit_task(struct task_struct *child, * Be careful about zapping the list - IRQ/NMI context * could still be processing it: */ - local_irq_disable(); + curr_rq_lock_irq_save(&flags); perf_flags = hw_perf_save_disable(); if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) { @@ -1294,7 +1332,7 @@ __perf_counter_exit_task(struct task_struct *child, list_del_init(&child_counter->list_entry); hw_perf_restore(perf_flags); - local_irq_enable(); + curr_rq_unlock_irq_restore(&flags); parent_counter = child_counter->parent; /* diff --git a/kernel/sched.c b/kernel/sched.c index 382cfdb5e38d..4d84ff4c8774 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -638,7 +638,7 @@ static inline int cpu_of(struct rq *rq) #define task_rq(p) cpu_rq(task_cpu(p)) #define cpu_curr(cpu) (cpu_rq(cpu)->curr) -static inline void update_rq_clock(struct rq *rq) +inline void update_rq_clock(struct rq *rq) { rq->clock = sched_clock_cpu(cpu_of(rq)); } @@ -969,6 +969,26 @@ static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) } } +void curr_rq_lock_irq_save(unsigned long *flags) + __acquires(rq->lock) +{ + struct rq *rq; + + local_irq_save(*flags); + rq = cpu_rq(smp_processor_id()); + spin_lock(&rq->lock); +} + +void curr_rq_unlock_irq_restore(unsigned long *flags) + __releases(rq->lock) +{ + struct rq *rq; + + rq = cpu_rq(smp_processor_id()); + spin_unlock(&rq->lock); + local_irq_restore(*flags); +} + void task_rq_unlock_wait(struct task_struct *p) { struct rq *rq = task_rq(p); @@ -2558,7 +2578,6 @@ prepare_task_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next) { fire_sched_out_preempt_notifiers(prev, next); - perf_counter_task_sched_out(prev, cpu_of(rq)); prepare_lock_switch(rq, next); prepare_arch_switch(next); } @@ -4089,6 +4108,29 @@ DEFINE_PER_CPU(struct kernel_stat, kstat); EXPORT_PER_CPU_SYMBOL(kstat); +/* + * Return any ns on the sched_clock that have not yet been banked in + * @p in case that task is currently running. + */ +unsigned long long __task_delta_exec(struct task_struct *p, int update) +{ + s64 delta_exec; + struct rq *rq; + + rq = task_rq(p); + WARN_ON_ONCE(!runqueue_is_locked()); + WARN_ON_ONCE(!task_current(rq, p)); + + if (update) + update_rq_clock(rq); + + delta_exec = rq->clock - p->se.exec_start; + + WARN_ON_ONCE(delta_exec < 0); + + return delta_exec; +} + /* * Return any ns on the sched_clock that have not yet been banked in * @p in case that task is currently running. @@ -4316,13 +4358,13 @@ void scheduler_tick(void) update_rq_clock(rq); update_cpu_load(rq); curr->sched_class->task_tick(rq, curr, 0); + perf_counter_task_tick(curr, cpu); spin_unlock(&rq->lock); #ifdef CONFIG_SMP rq->idle_at_tick = idle_cpu(cpu); trigger_load_balance(rq, cpu); #endif - perf_counter_task_tick(curr, cpu); } #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ @@ -4512,6 +4554,7 @@ need_resched_nonpreemptible: if (likely(prev != next)) { sched_info_switch(prev, next); + perf_counter_task_sched_out(prev, cpu); rq->nr_switches++; rq->curr = next; -- cgit v1.2.3 From eef6cbf5844c620d9db9be99e4908cdf92492fb9 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 19 Dec 2008 10:20:42 +0100 Subject: perfcounters: pull inherited counters Change counter inheritance from a 'push' to a 'pull' model: instead of child tasks pushing their final counts to the parent, reuse the wait4 infrastructure to pull counters as child tasks are exit-processed, much like how cutime/cstime is collected. Signed-off-by: Ingo Molnar --- kernel/exit.c | 21 +++++++++------------ 1 file changed, 9 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index 244edfd96865..101b7eeff44c 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -153,6 +153,9 @@ static void delayed_put_task_struct(struct rcu_head *rhp) { struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); +#ifdef CONFIG_PERF_COUNTERS + WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list)); +#endif trace_sched_process_free(tsk); put_task_struct(tsk); } @@ -922,12 +925,6 @@ static void exit_notify(struct task_struct *tsk, int group_dead) forget_original_parent(tsk); exit_task_namespaces(tsk); - /* - * Flush inherited counters to the parent - before the parent - * gets woken up by child-exit notifications. - */ - perf_counter_exit_task(tsk); - write_lock_irq(&tasklist_lock); if (group_dead) kill_orphaned_pgrp(tsk->group_leader, NULL); @@ -1122,12 +1119,6 @@ NORET_TYPE void do_exit(long code) if (tsk->splice_pipe) __free_pipe_info(tsk->splice_pipe); - /* - * These must happen late, after the PID is not - * hashed anymore, but still at a point that may sleep: - */ - perf_counter_exit_task(tsk); - preempt_disable(); /* causes final put_task_struct in finish_task_switch(). */ tsk->state = TASK_DEAD; @@ -1371,6 +1362,12 @@ static int wait_task_zombie(struct task_struct *p, int options, */ read_unlock(&tasklist_lock); + /* + * Flush inherited counters to the parent - before the parent + * gets woken up by child-exit notifications. + */ + perf_counter_exit_task(p); + retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0; status = (p->signal->flags & SIGNAL_GROUP_EXIT) ? p->signal->group_exit_code : p->exit_code; -- cgit v1.2.3 From 95cdd2e7851cce79ab839cb0b3cbe68d7911d0f1 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 21 Dec 2008 13:50:42 +0100 Subject: perfcounters: enable lowlevel pmc code to schedule counters Allow lowlevel ->enable() op to return an error if a counter can not be added. This can be used to handle counter constraints. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 62 +++++++++++++++++++++++++++++++++++++-------------- 1 file changed, 45 insertions(+), 17 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f1110ac1267b..2e73929a6959 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -355,21 +355,25 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) cpuctx->task_ctx = NULL; } -static void +static int counter_sched_in(struct perf_counter *counter, struct perf_cpu_context *cpuctx, struct perf_counter_context *ctx, int cpu) { if (counter->state == PERF_COUNTER_STATE_OFF) - return; + return 0; + + if (counter->hw_ops->enable(counter)) + return -EAGAIN; - counter->hw_ops->enable(counter); counter->state = PERF_COUNTER_STATE_ACTIVE; counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ cpuctx->active_oncpu++; ctx->nr_active++; + + return 0; } static int @@ -378,20 +382,38 @@ group_sched_in(struct perf_counter *group_counter, struct perf_counter_context *ctx, int cpu) { - struct perf_counter *counter; - int was_group = 0; + struct perf_counter *counter, *partial_group; + int ret = 0; - counter_sched_in(group_counter, cpuctx, ctx, cpu); + if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) + return -EAGAIN; /* * Schedule in siblings as one group (if any): */ list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { - counter_sched_in(counter, cpuctx, ctx, cpu); - was_group = 1; + if (counter_sched_in(counter, cpuctx, ctx, cpu)) { + partial_group = counter; + goto group_error; + } + ret = -EAGAIN; } - return was_group; + return ret; + +group_error: + /* + * Groups can be scheduled in as one unit only, so undo any + * partial group before returning: + */ + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { + if (counter == partial_group) + break; + counter_sched_out(counter, cpuctx, ctx); + } + counter_sched_out(group_counter, cpuctx, ctx); + + return -EAGAIN; } /* @@ -416,9 +438,6 @@ void perf_counter_task_sched_in(struct task_struct *task, int cpu) spin_lock(&ctx->lock); list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (ctx->nr_active == cpuctx->max_pertask) - break; - /* * Listen to the 'cpu' scheduling filter constraint * of counters: @@ -856,8 +875,9 @@ static const struct file_operations perf_fops = { .poll = perf_poll, }; -static void cpu_clock_perf_counter_enable(struct perf_counter *counter) +static int cpu_clock_perf_counter_enable(struct perf_counter *counter) { + return 0; } static void cpu_clock_perf_counter_disable(struct perf_counter *counter) @@ -913,11 +933,13 @@ static void task_clock_perf_counter_read(struct perf_counter *counter) task_clock_perf_counter_update(counter, now); } -static void task_clock_perf_counter_enable(struct perf_counter *counter) +static int task_clock_perf_counter_enable(struct perf_counter *counter) { u64 now = task_clock_perf_counter_val(counter, 0); atomic64_set(&counter->hw.prev_count, now); + + return 0; } static void task_clock_perf_counter_disable(struct perf_counter *counter) @@ -960,12 +982,14 @@ static void page_faults_perf_counter_read(struct perf_counter *counter) page_faults_perf_counter_update(counter); } -static void page_faults_perf_counter_enable(struct perf_counter *counter) +static int page_faults_perf_counter_enable(struct perf_counter *counter) { /* * page-faults is a per-task value already, * so we dont have to clear it on switch-in. */ + + return 0; } static void page_faults_perf_counter_disable(struct perf_counter *counter) @@ -1006,12 +1030,14 @@ static void context_switches_perf_counter_read(struct perf_counter *counter) context_switches_perf_counter_update(counter); } -static void context_switches_perf_counter_enable(struct perf_counter *counter) +static int context_switches_perf_counter_enable(struct perf_counter *counter) { /* * ->nvcsw + curr->nivcsw is a per-task value already, * so we dont have to clear it on switch-in. */ + + return 0; } static void context_switches_perf_counter_disable(struct perf_counter *counter) @@ -1050,12 +1076,14 @@ static void cpu_migrations_perf_counter_read(struct perf_counter *counter) cpu_migrations_perf_counter_update(counter); } -static void cpu_migrations_perf_counter_enable(struct perf_counter *counter) +static int cpu_migrations_perf_counter_enable(struct perf_counter *counter) { /* * se.nr_migrations is a per-task value already, * so we dont have to clear it on switch-in. */ + + return 0; } static void cpu_migrations_perf_counter_disable(struct perf_counter *counter) -- cgit v1.2.3 From 8fe91e61cdc407c7556d3cd71cf20141a25bbcea Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 23 Dec 2008 12:29:25 +0100 Subject: perfcounters: remove ->nr_inherited Impact: remove dead code nr_inherited was not maintained correctly (not decremented) - and also not used - remove it. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 1 - 1 file changed, 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 2e73929a6959..48e1dbcdc1cd 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1308,7 +1308,6 @@ inherit_counter(struct perf_counter *parent_counter, child_ctx->nr_counters++; child_counter->parent = parent_counter; - parent_counter->nr_inherited++; /* * inherit into child's child as well: */ -- cgit v1.2.3 From 235c7fc7c500e4fd1700c4ad01b5612bcdc1b449 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 21 Dec 2008 14:43:25 +0100 Subject: perfcounters: generalize the counter scheduler Impact: clean up and refactor code refactor the counter scheduler: separate out in/out functions and introduce a counter-rotation function as well. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 220 ++++++++++++++++++++++++++++++++------------------ 1 file changed, 142 insertions(+), 78 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 48e1dbcdc1cd..d7a79f321b1c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -111,11 +111,12 @@ static void __perf_counter_remove_from_context(void *info) spin_lock(&ctx->lock); if (counter->state == PERF_COUNTER_STATE_ACTIVE) { - counter->hw_ops->disable(counter); counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->hw_ops->disable(counter); ctx->nr_active--; cpuctx->active_oncpu--; counter->task = NULL; + counter->oncpu = -1; } ctx->nr_counters--; @@ -192,8 +193,36 @@ retry: spin_unlock_irq(&ctx->lock); } +static int +counter_sched_in(struct perf_counter *counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx, + int cpu) +{ + if (counter->state == PERF_COUNTER_STATE_OFF) + return 0; + + counter->state = PERF_COUNTER_STATE_ACTIVE; + counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ + /* + * The new state must be visible before we turn it on in the hardware: + */ + smp_wmb(); + + if (counter->hw_ops->enable(counter)) { + counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->oncpu = -1; + return -EAGAIN; + } + + cpuctx->active_oncpu++; + ctx->nr_active++; + + return 0; +} + /* - * Cross CPU call to install and enable a preformance counter + * Cross CPU call to install and enable a performance counter */ static void __perf_install_in_context(void *info) { @@ -220,22 +249,17 @@ static void __perf_install_in_context(void *info) * counters on a global level. NOP for non NMI based counters. */ perf_flags = hw_perf_save_disable(); - list_add_counter(counter, ctx); - hw_perf_restore(perf_flags); + list_add_counter(counter, ctx); ctx->nr_counters++; - if (cpuctx->active_oncpu < perf_max_counters) { - counter->state = PERF_COUNTER_STATE_ACTIVE; - counter->oncpu = cpu; - ctx->nr_active++; - cpuctx->active_oncpu++; - counter->hw_ops->enable(counter); - } + counter_sched_in(counter, cpuctx, ctx, cpu); if (!ctx->task && cpuctx->max_pertask) cpuctx->max_pertask--; + hw_perf_restore(perf_flags); + spin_unlock(&ctx->lock); curr_rq_unlock_irq_restore(&flags); } @@ -302,8 +326,8 @@ counter_sched_out(struct perf_counter *counter, if (counter->state != PERF_COUNTER_STATE_ACTIVE) return; - counter->hw_ops->disable(counter); counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->hw_ops->disable(counter); counter->oncpu = -1; cpuctx->active_oncpu--; @@ -326,6 +350,22 @@ group_sched_out(struct perf_counter *group_counter, counter_sched_out(counter, cpuctx, ctx); } +void __perf_counter_sched_out(struct perf_counter_context *ctx, + struct perf_cpu_context *cpuctx) +{ + struct perf_counter *counter; + + if (likely(!ctx->nr_counters)) + return; + + spin_lock(&ctx->lock); + if (ctx->nr_active) { + list_for_each_entry(counter, &ctx->counter_list, list_entry) + group_sched_out(counter, cpuctx, ctx); + } + spin_unlock(&ctx->lock); +} + /* * Called from scheduler to remove the counters of the current task, * with interrupts disabled. @@ -341,39 +381,18 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); struct perf_counter_context *ctx = &task->perf_counter_ctx; - struct perf_counter *counter; if (likely(!cpuctx->task_ctx)) return; - spin_lock(&ctx->lock); - if (ctx->nr_active) { - list_for_each_entry(counter, &ctx->counter_list, list_entry) - group_sched_out(counter, cpuctx, ctx); - } - spin_unlock(&ctx->lock); + __perf_counter_sched_out(ctx, cpuctx); + cpuctx->task_ctx = NULL; } -static int -counter_sched_in(struct perf_counter *counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, - int cpu) +static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx) { - if (counter->state == PERF_COUNTER_STATE_OFF) - return 0; - - if (counter->hw_ops->enable(counter)) - return -EAGAIN; - - counter->state = PERF_COUNTER_STATE_ACTIVE; - counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ - - cpuctx->active_oncpu++; - ctx->nr_active++; - - return 0; + __perf_counter_sched_out(&cpuctx->ctx, cpuctx); } static int @@ -416,21 +435,10 @@ group_error: return -EAGAIN; } -/* - * Called from scheduler to add the counters of the current task - * with interrupts disabled. - * - * We restore the counter value and then enable it. - * - * This does not protect us against NMI, but enable() - * sets the enabled bit in the control field of counter _before_ - * accessing the counter control register. If a NMI hits, then it will - * keep the counter running. - */ -void perf_counter_task_sched_in(struct task_struct *task, int cpu) +static void +__perf_counter_sched_in(struct perf_counter_context *ctx, + struct perf_cpu_context *cpuctx, int cpu) { - struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &task->perf_counter_ctx; struct perf_counter *counter; if (likely(!ctx->nr_counters)) @@ -453,10 +461,35 @@ void perf_counter_task_sched_in(struct task_struct *task, int cpu) break; } spin_unlock(&ctx->lock); +} +/* + * Called from scheduler to add the counters of the current task + * with interrupts disabled. + * + * We restore the counter value and then enable it. + * + * This does not protect us against NMI, but enable() + * sets the enabled bit in the control field of counter _before_ + * accessing the counter control register. If a NMI hits, then it will + * keep the counter running. + */ +void perf_counter_task_sched_in(struct task_struct *task, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_counter_context *ctx = &task->perf_counter_ctx; + + __perf_counter_sched_in(ctx, cpuctx, cpu); cpuctx->task_ctx = ctx; } +static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) +{ + struct perf_counter_context *ctx = &cpuctx->ctx; + + __perf_counter_sched_in(ctx, cpuctx, cpu); +} + int perf_counter_task_disable(void) { struct task_struct *curr = current; @@ -514,6 +547,8 @@ int perf_counter_task_enable(void) /* force the update of the task clock: */ __task_delta_exec(curr, 1); + perf_counter_task_sched_out(curr, cpu); + spin_lock(&ctx->lock); /* @@ -538,19 +573,18 @@ int perf_counter_task_enable(void) return 0; } -void perf_counter_task_tick(struct task_struct *curr, int cpu) +/* + * Round-robin a context's counters: + */ +static void rotate_ctx(struct perf_counter_context *ctx) { - struct perf_counter_context *ctx = &curr->perf_counter_ctx; struct perf_counter *counter; u64 perf_flags; - if (likely(!ctx->nr_counters)) + if (!ctx->nr_counters) return; - perf_counter_task_sched_out(curr, cpu); - spin_lock(&ctx->lock); - /* * Rotate the first entry last (works just fine for group counters too): */ @@ -563,7 +597,24 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) hw_perf_restore(perf_flags); spin_unlock(&ctx->lock); +} + +void perf_counter_task_tick(struct task_struct *curr, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_counter_context *ctx = &curr->perf_counter_ctx; + const int rotate_percpu = 0; + + if (rotate_percpu) + perf_counter_cpu_sched_out(cpuctx); + perf_counter_task_sched_out(curr, cpu); + if (rotate_percpu) + rotate_ctx(&cpuctx->ctx); + rotate_ctx(ctx); + + if (rotate_percpu) + perf_counter_cpu_sched_in(cpuctx, cpu); perf_counter_task_sched_in(curr, cpu); } @@ -905,8 +956,6 @@ static u64 task_clock_perf_counter_val(struct perf_counter *counter, int update) struct task_struct *curr = counter->task; u64 delta; - WARN_ON_ONCE(counter->task != current); - delta = __task_delta_exec(curr, update); return curr->se.sum_exec_runtime + delta; @@ -1160,6 +1209,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->group_leader = group_leader; counter->hw_ops = NULL; + counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) counter->state = PERF_COUNTER_STATE_OFF; @@ -1331,35 +1381,49 @@ __perf_counter_exit_task(struct task_struct *child, { struct perf_counter *parent_counter; u64 parent_val, child_val; - unsigned long flags; - u64 perf_flags; /* - * Disable and unlink this counter. - * - * Be careful about zapping the list - IRQ/NMI context - * could still be processing it: + * If we do not self-reap then we have to wait for the + * child task to unschedule (it will happen for sure), + * so that its counter is at its final count. (This + * condition triggers rarely - child tasks usually get + * off their CPU before the parent has a chance to + * get this far into the reaping action) */ - curr_rq_lock_irq_save(&flags); - perf_flags = hw_perf_save_disable(); - - if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) { + if (child != current) { + wait_task_inactive(child, 0); + list_del_init(&child_counter->list_entry); + } else { struct perf_cpu_context *cpuctx; + unsigned long flags; + u64 perf_flags; + + /* + * Disable and unlink this counter. + * + * Be careful about zapping the list - IRQ/NMI context + * could still be processing it: + */ + curr_rq_lock_irq_save(&flags); + perf_flags = hw_perf_save_disable(); cpuctx = &__get_cpu_var(perf_cpu_context); - child_counter->hw_ops->disable(child_counter); - child_counter->state = PERF_COUNTER_STATE_INACTIVE; - child_counter->oncpu = -1; + if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) { + child_counter->state = PERF_COUNTER_STATE_INACTIVE; + child_counter->hw_ops->disable(child_counter); + cpuctx->active_oncpu--; + child_ctx->nr_active--; + child_counter->oncpu = -1; + } - cpuctx->active_oncpu--; - child_ctx->nr_active--; - } + list_del_init(&child_counter->list_entry); - list_del_init(&child_counter->list_entry); + child_ctx->nr_counters--; - hw_perf_restore(perf_flags); - curr_rq_unlock_irq_restore(&flags); + hw_perf_restore(perf_flags); + curr_rq_unlock_irq_restore(&flags); + } parent_counter = child_counter->parent; /* -- cgit v1.2.3 From 01ea1ccaa24dea3552e103be13b7897211607a8b Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Fri, 26 Dec 2008 21:05:06 -0800 Subject: perf_counter: more barrier in blank weak function Impact: fix panic possible panic Some versions of GCC inline the weak global function if it's empty. Add a barrier() to work it around. Signed-off-by: Yinghai Lu Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d7a79f321b1c..37f771691f93 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -45,8 +45,8 @@ hw_perf_counter_init(struct perf_counter *counter) } u64 __weak hw_perf_save_disable(void) { return 0; } -void __weak hw_perf_restore(u64 ctrl) { } -void __weak hw_perf_counter_setup(void) { } +void __weak hw_perf_restore(u64 ctrl) { barrier(); } +void __weak hw_perf_counter_setup(void) { barrier(); } static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) -- cgit v1.2.3 From ff6f05416ece2caec1a7a1f8180d6598e0ab9272 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 9 Jan 2009 16:19:25 +1100 Subject: perf_counter: Fix return value from dummy hw_perf_counter_init Impact: fix oops-causing bug Currently, if you try to use perf_counters on an architecture that has no hardware support, and you select an event that doesn't map to any of the defined software counters, you get an oops rather than an error. This is because the dummy hw_perf_counter_init returns ERR_PTR(-EINVAL) but the caller (perf_counter_alloc) only tests for NULL. This makes the dummy hw_perf_counter_init return NULL instead. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 37f771691f93..4be1a8d872b4 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -41,7 +41,7 @@ static DEFINE_MUTEX(perf_resource_mutex); extern __weak const struct hw_perf_counter_ops * hw_perf_counter_init(struct perf_counter *counter) { - return ERR_PTR(-EINVAL); + return NULL; } u64 __weak hw_perf_save_disable(void) { return 0; } -- cgit v1.2.3 From 9abf8a08bc8f18a3b125f834f00e2e71b49c15d2 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 9 Jan 2009 16:26:43 +1100 Subject: perf_counter: Fix the cpu_clock software counter Impact: bug fix Currently if you do (e.g.) timec -e -1 ls, it will report 0 for the value of the cpu_clock counter. The reason is that the core assumes that a counter's count field is up-to-date when the counter is inactive, and doesn't call the counter's read function. However, the cpu_clock counter code only updates the count in the read function. This fixes it by making both the read and disable functions update the count. It also makes the counter ignore time passing while the counter is disabled, by making the enable function update the hw.prev_count field. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4be1a8d872b4..b7a027a2ef02 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -928,18 +928,32 @@ static const struct file_operations perf_fops = { static int cpu_clock_perf_counter_enable(struct perf_counter *counter) { + int cpu = raw_smp_processor_id(); + + atomic64_set(&counter->hw.prev_count, cpu_clock(cpu)); return 0; } +static void cpu_clock_perf_counter_update(struct perf_counter *counter) +{ + int cpu = raw_smp_processor_id(); + s64 prev; + u64 now; + + now = cpu_clock(cpu); + prev = atomic64_read(&counter->hw.prev_count); + atomic64_set(&counter->hw.prev_count, now); + atomic64_add(now - prev, &counter->count); +} + static void cpu_clock_perf_counter_disable(struct perf_counter *counter) { + cpu_clock_perf_counter_update(counter); } static void cpu_clock_perf_counter_read(struct perf_counter *counter) { - int cpu = raw_smp_processor_id(); - - atomic64_set(&counter->count, cpu_clock(cpu)); + cpu_clock_perf_counter_update(counter); } static const struct hw_perf_counter_ops perf_ops_cpu_clock = { -- cgit v1.2.3 From 3cbed429a9ccdb7a243f733b1056fe5c39e9004c Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 9 Jan 2009 16:43:42 +1100 Subject: perf_counter: Add optional hw_perf_group_sched_in arch function Impact: extend perf_counter infrastructure This adds an optional hw_perf_group_sched_in() arch function that enables a whole group of counters in one go. It returns 1 if it added the group successfully, 0 if it did nothing (and therefore the core needs to add the counters individually), or a negative number if an error occurred. It should add all the counters and enable any software counters in the group, or else add none of them and return an error. There are a couple of related changes/improvements in the group handling here: * As an optimization, group_sched_out() and group_sched_in() now check the state of the group leader, and do nothing if the leader is not active or disabled. * We now call hw_perf_save_disable/hw_perf_restore around the complete set of counter enable/disable calls in __perf_counter_sched_in/out, to give the arch code the opportunity to defer updating the hardware state until the hw_perf_restore call if it wants. * We no longer stop adding groups after we get to a group that has more than one counter. We will ultimately add an option for a group to be exclusive. The current code doesn't really implement exclusive groups anyway, since a group could end up going on with other counters that get added before it. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 31 ++++++++++++++++++++++++++----- 1 file changed, 26 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index b7a027a2ef02..9ad11e44d9ab 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -47,6 +47,12 @@ hw_perf_counter_init(struct perf_counter *counter) u64 __weak hw_perf_save_disable(void) { return 0; } void __weak hw_perf_restore(u64 ctrl) { barrier(); } void __weak hw_perf_counter_setup(void) { barrier(); } +int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx, int cpu) +{ + return 0; +} static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) @@ -341,6 +347,9 @@ group_sched_out(struct perf_counter *group_counter, { struct perf_counter *counter; + if (group_counter->state != PERF_COUNTER_STATE_ACTIVE) + return; + counter_sched_out(group_counter, cpuctx, ctx); /* @@ -354,15 +363,18 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx) { struct perf_counter *counter; + u64 flags; if (likely(!ctx->nr_counters)) return; spin_lock(&ctx->lock); + flags = hw_perf_save_disable(); if (ctx->nr_active) { list_for_each_entry(counter, &ctx->counter_list, list_entry) group_sched_out(counter, cpuctx, ctx); } + hw_perf_restore(flags); spin_unlock(&ctx->lock); } @@ -402,7 +414,14 @@ group_sched_in(struct perf_counter *group_counter, int cpu) { struct perf_counter *counter, *partial_group; - int ret = 0; + int ret; + + if (group_counter->state == PERF_COUNTER_STATE_OFF) + return 0; + + ret = hw_perf_group_sched_in(group_counter, cpuctx, ctx, cpu); + if (ret) + return ret < 0 ? ret : 0; if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) return -EAGAIN; @@ -415,10 +434,9 @@ group_sched_in(struct perf_counter *group_counter, partial_group = counter; goto group_error; } - ret = -EAGAIN; } - return ret; + return 0; group_error: /* @@ -440,11 +458,13 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx, int cpu) { struct perf_counter *counter; + u64 flags; if (likely(!ctx->nr_counters)) return; spin_lock(&ctx->lock); + flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { /* * Listen to the 'cpu' scheduling filter constraint @@ -454,12 +474,13 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, continue; /* - * If we scheduled in a group atomically and - * exclusively, break out: + * If we scheduled in a group atomically and exclusively, + * or if this group can't go on, break out: */ if (group_sched_in(counter, cpuctx, ctx, cpu)) break; } + hw_perf_restore(flags); spin_unlock(&ctx->lock); } -- cgit v1.2.3 From 4eb96fcfe07b7f2a05577e57533840f8e26bea53 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 9 Jan 2009 17:24:34 +1100 Subject: perf_counter: Add dummy perf_counter_print_debug function Impact: minimize requirements on architectures Currently, an architecture just enabling CONFIG_PERF_COUNTERS but not providing any extra functions will fail to build with perf_counter_print_debug being undefined, since we don't provide an empty dummy definition like we do with the hw_perf_* functions. This provides an empty dummy perf_counter_print_debug() to make it easier for architectures to turn on CONFIG_PERF_COUNTERS. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 9ad11e44d9ab..4c0dccb756ad 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -54,6 +54,8 @@ int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, return 0; } +void __weak perf_counter_print_debug(void) { } + static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { -- cgit v1.2.3 From dd0e6ba22ea21bcc2c420b385a170593c58f4c08 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 12 Jan 2009 15:11:00 +1100 Subject: perf_counter: Always schedule all software counters in Software counters aren't subject to the limitations imposed by the fixed number of hardware counter registers, so there is no reason not to enable them all in __perf_counter_sched_in. Previously we used to break out of the loop when we got to a group that wouldn't fit on the PMU; with this we continue through the list but only schedule in software counters (or groups containing only software counters) from there on. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 33 ++++++++++++++++++++++++++++++--- 1 file changed, 30 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4c0dccb756ad..3aef3062ff78 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -455,12 +455,37 @@ group_error: return -EAGAIN; } +/* + * Return 1 for a software counter, 0 for a hardware counter + */ +static inline int is_software_counter(struct perf_counter *counter) +{ + return !counter->hw_event.raw && counter->hw_event.type < 0; +} + +/* + * Return 1 for a group consisting entirely of software counters, + * 0 if the group contains any hardware counters. + */ +static int is_software_only_group(struct perf_counter *leader) +{ + struct perf_counter *counter; + + if (!is_software_counter(leader)) + return 0; + list_for_each_entry(counter, &leader->sibling_list, list_entry) + if (!is_software_counter(counter)) + return 0; + return 1; +} + static void __perf_counter_sched_in(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx, int cpu) { struct perf_counter *counter; u64 flags; + int can_add_hw = 1; if (likely(!ctx->nr_counters)) return; @@ -477,10 +502,12 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, /* * If we scheduled in a group atomically and exclusively, - * or if this group can't go on, break out: + * or if this group can't go on, don't add any more + * hardware counters. */ - if (group_sched_in(counter, cpuctx, ctx, cpu)) - break; + if (can_add_hw || is_software_only_group(counter)) + if (group_sched_in(counter, cpuctx, ctx, cpu)) + can_add_hw = 0; } hw_perf_restore(flags); spin_unlock(&ctx->lock); -- cgit v1.2.3 From 01d0287f068de2934109ba9b989d8807526cccc2 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Wed, 14 Jan 2009 13:44:19 +1100 Subject: powerpc/perf_counter: Make sure PMU gets enabled properly This makes sure that we call the platform-specific ppc_md.enable_pmcs function on each CPU before we try to use the PMU on that CPU. If the CPU goes off-line and then on-line, we need to do the enable_pmcs call again, so we use the hw_perf_counter_setup hook to ensure that. It gets called as each CPU comes online, but it isn't called on the CPU that is coming up, so this adds the CPU number as an argument to it (there were no non-empty instances of hw_perf_counter_setup before). This also arranges to set the pmcregs_in_use field of the lppaca (data structure shared with the hypervisor) on each CPU when we are using the PMU and clear it when we are not. This allows the hypervisor to optimize partition switches by not saving/restoring the PMU registers when we aren't using the PMU. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 3aef3062ff78..52f2f526248e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -46,7 +46,7 @@ hw_perf_counter_init(struct perf_counter *counter) u64 __weak hw_perf_save_disable(void) { return 0; } void __weak hw_perf_restore(u64 ctrl) { barrier(); } -void __weak hw_perf_counter_setup(void) { barrier(); } +void __weak hw_perf_counter_setup(int cpu) { barrier(); } int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, struct perf_cpu_context *cpuctx, struct perf_counter_context *ctx, int cpu) @@ -1598,7 +1598,7 @@ static void __cpuinit perf_counter_init_cpu(int cpu) cpuctx->max_pertask = perf_max_counters - perf_reserved_percpu; mutex_unlock(&perf_resource_mutex); - hw_perf_counter_setup(); + hw_perf_counter_setup(cpu); } #ifdef CONFIG_HOTPLUG_CPU -- cgit v1.2.3 From 3b6f9e5cb21964b7ce12bf81076f830885563ec8 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Wed, 14 Jan 2009 21:00:30 +1100 Subject: perf_counter: Add support for pinned and exclusive counter groups Impact: New perf_counter features A pinned counter group is one that the user wants to have on the CPU whenever possible, i.e. whenever the associated task is running, for a per-task group, or always for a per-cpu group. If the system cannot satisfy that, it puts the group into an error state where it is not scheduled any more and reads from it return EOF (i.e. 0 bytes read). The group can be released from error state and made readable again using prctl(PR_TASK_PERF_COUNTERS_ENABLE). When we have finer-grained enable/disable controls on counters we'll be able to reset the error state on individual groups. An exclusive group is one that the user wants to be the only group using the CPU performance monitor hardware whenever it is on. The counter group scheduler will not schedule an exclusive group if there are already other groups on the CPU and will not schedule other groups onto the CPU if there is an exclusive group scheduled (that statement does not apply to groups containing only software counters, which can always go on and which do not prevent an exclusive group from going on). With an exclusive group, we will be able to let users program PMU registers at a low level without the concern that those settings will perturb other measurements. Along the way this reorganizes things a little: - is_software_counter() is moved to perf_counter.h. - cpuctx->active_oncpu now records the number of hardware counters on the CPU, i.e. it now excludes software counters. Nothing was reading cpuctx->active_oncpu before, so this change is harmless. - A new cpuctx->exclusive field records whether we currently have an exclusive group on the CPU. - counter_sched_out moves higher up in perf_counter.c and gets called from __perf_counter_remove_from_context and __perf_counter_exit_task, where we used to have essentially the same code. - __perf_counter_sched_in now goes through the counter list twice, doing the pinned counters in the first loop and the non-pinned counters in the second loop, in order to give the pinned counters the best chance to be scheduled in. Note that only a group leader can be exclusive or pinned, and that attribute applies to the whole group. This avoids some awkwardness in some corner cases (e.g. where a group leader is closed and the other group members get added to the context list). If we want to relax that restriction later, we can, and it is easier to relax a restriction than to apply a new one. This doesn't yet handle the case where a pinned counter is inherited and goes into error state in the child - the error state is not propagated up to the parent when the child exits, and arguably it should. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 226 ++++++++++++++++++++++++++++++++++---------------- 1 file changed, 154 insertions(+), 72 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 52f2f526248e..faf671b29566 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -93,6 +93,25 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) } } +static void +counter_sched_out(struct perf_counter *counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx) +{ + if (counter->state != PERF_COUNTER_STATE_ACTIVE) + return; + + counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->hw_ops->disable(counter); + counter->oncpu = -1; + + if (!is_software_counter(counter)) + cpuctx->active_oncpu--; + ctx->nr_active--; + if (counter->hw_event.exclusive || !cpuctx->active_oncpu) + cpuctx->exclusive = 0; +} + /* * Cross CPU call to remove a performance counter * @@ -118,14 +137,9 @@ static void __perf_counter_remove_from_context(void *info) curr_rq_lock_irq_save(&flags); spin_lock(&ctx->lock); - if (counter->state == PERF_COUNTER_STATE_ACTIVE) { - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->hw_ops->disable(counter); - ctx->nr_active--; - cpuctx->active_oncpu--; - counter->task = NULL; - counter->oncpu = -1; - } + counter_sched_out(counter, cpuctx, ctx); + + counter->task = NULL; ctx->nr_counters--; /* @@ -207,7 +221,7 @@ counter_sched_in(struct perf_counter *counter, struct perf_counter_context *ctx, int cpu) { - if (counter->state == PERF_COUNTER_STATE_OFF) + if (counter->state <= PERF_COUNTER_STATE_OFF) return 0; counter->state = PERF_COUNTER_STATE_ACTIVE; @@ -223,12 +237,63 @@ counter_sched_in(struct perf_counter *counter, return -EAGAIN; } - cpuctx->active_oncpu++; + if (!is_software_counter(counter)) + cpuctx->active_oncpu++; ctx->nr_active++; + if (counter->hw_event.exclusive) + cpuctx->exclusive = 1; + return 0; } +/* + * Return 1 for a group consisting entirely of software counters, + * 0 if the group contains any hardware counters. + */ +static int is_software_only_group(struct perf_counter *leader) +{ + struct perf_counter *counter; + + if (!is_software_counter(leader)) + return 0; + list_for_each_entry(counter, &leader->sibling_list, list_entry) + if (!is_software_counter(counter)) + return 0; + return 1; +} + +/* + * Work out whether we can put this counter group on the CPU now. + */ +static int group_can_go_on(struct perf_counter *counter, + struct perf_cpu_context *cpuctx, + int can_add_hw) +{ + /* + * Groups consisting entirely of software counters can always go on. + */ + if (is_software_only_group(counter)) + return 1; + /* + * If an exclusive group is already on, no other hardware + * counters can go on. + */ + if (cpuctx->exclusive) + return 0; + /* + * If this group is exclusive and there are already + * counters on the CPU, it can't go on. + */ + if (counter->hw_event.exclusive && cpuctx->active_oncpu) + return 0; + /* + * Otherwise, try to add it if all previous groups were able + * to go on. + */ + return can_add_hw; +} + /* * Cross CPU call to install and enable a performance counter */ @@ -240,6 +305,7 @@ static void __perf_install_in_context(void *info) int cpu = smp_processor_id(); unsigned long flags; u64 perf_flags; + int err; /* * If this is a task context, we need to check whether it is @@ -261,9 +327,21 @@ static void __perf_install_in_context(void *info) list_add_counter(counter, ctx); ctx->nr_counters++; - counter_sched_in(counter, cpuctx, ctx, cpu); + /* + * An exclusive counter can't go on if there are already active + * hardware counters, and no hardware counter can go on if there + * is already an exclusive counter on. + */ + if (counter->state == PERF_COUNTER_STATE_INACTIVE && + !group_can_go_on(counter, cpuctx, 1)) + err = -EEXIST; + else + err = counter_sched_in(counter, cpuctx, ctx, cpu); + + if (err && counter->hw_event.pinned) + counter->state = PERF_COUNTER_STATE_ERROR; - if (!ctx->task && cpuctx->max_pertask) + if (!err && !ctx->task && cpuctx->max_pertask) cpuctx->max_pertask--; hw_perf_restore(perf_flags); @@ -326,22 +404,6 @@ retry: spin_unlock_irq(&ctx->lock); } -static void -counter_sched_out(struct perf_counter *counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx) -{ - if (counter->state != PERF_COUNTER_STATE_ACTIVE) - return; - - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->hw_ops->disable(counter); - counter->oncpu = -1; - - cpuctx->active_oncpu--; - ctx->nr_active--; -} - static void group_sched_out(struct perf_counter *group_counter, struct perf_cpu_context *cpuctx, @@ -359,6 +421,9 @@ group_sched_out(struct perf_counter *group_counter, */ list_for_each_entry(counter, &group_counter->sibling_list, list_entry) counter_sched_out(counter, cpuctx, ctx); + + if (group_counter->hw_event.exclusive) + cpuctx->exclusive = 0; } void __perf_counter_sched_out(struct perf_counter_context *ctx, @@ -455,30 +520,6 @@ group_error: return -EAGAIN; } -/* - * Return 1 for a software counter, 0 for a hardware counter - */ -static inline int is_software_counter(struct perf_counter *counter) -{ - return !counter->hw_event.raw && counter->hw_event.type < 0; -} - -/* - * Return 1 for a group consisting entirely of software counters, - * 0 if the group contains any hardware counters. - */ -static int is_software_only_group(struct perf_counter *leader) -{ - struct perf_counter *counter; - - if (!is_software_counter(leader)) - return 0; - list_for_each_entry(counter, &leader->sibling_list, list_entry) - if (!is_software_counter(counter)) - return 0; - return 1; -} - static void __perf_counter_sched_in(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx, int cpu) @@ -492,7 +533,38 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, spin_lock(&ctx->lock); flags = hw_perf_save_disable(); + + /* + * First go through the list and put on any pinned groups + * in order to give them the best chance of going on. + */ + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (counter->state <= PERF_COUNTER_STATE_OFF || + !counter->hw_event.pinned) + continue; + if (counter->cpu != -1 && counter->cpu != cpu) + continue; + + if (group_can_go_on(counter, cpuctx, 1)) + group_sched_in(counter, cpuctx, ctx, cpu); + + /* + * If this pinned group hasn't been scheduled, + * put it in error state. + */ + if (counter->state == PERF_COUNTER_STATE_INACTIVE) + counter->state = PERF_COUNTER_STATE_ERROR; + } + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + /* + * Ignore counters in OFF or ERROR state, and + * ignore pinned counters since we did them already. + */ + if (counter->state <= PERF_COUNTER_STATE_OFF || + counter->hw_event.pinned) + continue; + /* * Listen to the 'cpu' scheduling filter constraint * of counters: @@ -500,14 +572,10 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, if (counter->cpu != -1 && counter->cpu != cpu) continue; - /* - * If we scheduled in a group atomically and exclusively, - * or if this group can't go on, don't add any more - * hardware counters. - */ - if (can_add_hw || is_software_only_group(counter)) + if (group_can_go_on(counter, cpuctx, can_add_hw)) { if (group_sched_in(counter, cpuctx, ctx, cpu)) can_add_hw = 0; + } } hw_perf_restore(flags); spin_unlock(&ctx->lock); @@ -567,8 +635,10 @@ int perf_counter_task_disable(void) */ perf_flags = hw_perf_save_disable(); - list_for_each_entry(counter, &ctx->counter_list, list_entry) - counter->state = PERF_COUNTER_STATE_OFF; + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (counter->state != PERF_COUNTER_STATE_ERROR) + counter->state = PERF_COUNTER_STATE_OFF; + } hw_perf_restore(perf_flags); @@ -607,7 +677,7 @@ int perf_counter_task_enable(void) perf_flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->state != PERF_COUNTER_STATE_OFF) + if (counter->state > PERF_COUNTER_STATE_OFF) continue; counter->state = PERF_COUNTER_STATE_INACTIVE; counter->hw_event.disabled = 0; @@ -849,6 +919,14 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (count != sizeof(cntval)) return -EINVAL; + /* + * Return end-of-file for a read on a counter that is in + * error state (i.e. because it was pinned but it couldn't be + * scheduled on to the CPU at some point). + */ + if (counter->state == PERF_COUNTER_STATE_ERROR) + return 0; + mutex_lock(&counter->mutex); cntval = perf_counter_read(counter); mutex_unlock(&counter->mutex); @@ -884,7 +962,7 @@ perf_read_irq_data(struct perf_counter *counter, { struct perf_data *irqdata, *usrdata; DECLARE_WAITQUEUE(wait, current); - ssize_t res; + ssize_t res, res2; irqdata = counter->irqdata; usrdata = counter->usrdata; @@ -905,6 +983,9 @@ perf_read_irq_data(struct perf_counter *counter, if (signal_pending(current)) break; + if (counter->state == PERF_COUNTER_STATE_ERROR) + break; + spin_unlock_irq(&counter->waitq.lock); schedule(); spin_lock_irq(&counter->waitq.lock); @@ -913,7 +994,8 @@ perf_read_irq_data(struct perf_counter *counter, __set_current_state(TASK_RUNNING); spin_unlock_irq(&counter->waitq.lock); - if (usrdata->len + irqdata->len < count) + if (usrdata->len + irqdata->len < count && + counter->state != PERF_COUNTER_STATE_ERROR) return -ERESTARTSYS; read_pending: mutex_lock(&counter->mutex); @@ -925,11 +1007,12 @@ read_pending: /* Switch irq buffer: */ usrdata = perf_switch_irq_data(counter); - if (perf_copy_usrdata(usrdata, buf + res, count - res) < 0) { + res2 = perf_copy_usrdata(usrdata, buf + res, count - res); + if (res2 < 0) { if (!res) res = -EFAULT; } else { - res = count; + res += res2; } out: mutex_unlock(&counter->mutex); @@ -1348,6 +1431,11 @@ sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, */ if (group_leader->ctx != ctx) goto err_put_context; + /* + * Only a group leader can be exclusive or pinned + */ + if (hw_event.exclusive || hw_event.pinned) + goto err_put_context; } ret = -EINVAL; @@ -1473,13 +1561,7 @@ __perf_counter_exit_task(struct task_struct *child, cpuctx = &__get_cpu_var(perf_cpu_context); - if (child_counter->state == PERF_COUNTER_STATE_ACTIVE) { - child_counter->state = PERF_COUNTER_STATE_INACTIVE; - child_counter->hw_ops->disable(child_counter); - cpuctx->active_oncpu--; - child_ctx->nr_active--; - child_counter->oncpu = -1; - } + counter_sched_out(child_counter, cpuctx, child_ctx); list_del_init(&child_counter->list_entry); -- cgit v1.2.3 From d859e29fe34cb833071b20aef860ee94fbad9bb2 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Sat, 17 Jan 2009 18:10:22 +1100 Subject: perf_counter: Add counter enable/disable ioctls Impact: New perf_counter features This primarily adds a way for perf_counter users to enable and disable counters and groups. Enabling or disabling a counter or group also enables or disables all of the child counters that have been cloned from it to monitor children of the task monitored by the top-level counter. The userspace interface to enable/disable counters is via ioctl on the counter file descriptor. Along the way this extends the code that handles child counters to handle child counter groups properly. A group with multiple counters will be cloned to child tasks if and only if the group leader has the hw_event.inherit bit set - if it is set the whole group is cloned as a group in the child task. In order to be able to enable or disable all child counters of a given top-level counter, we need a way to find them all. Hence I have added a child_list field to struct perf_counter, which is the head of the list of children for a top-level counter, or the link in that list for a child counter. That list is protected by the perf_counter.mutex field. This also adds a mutex to the perf_counter_context struct. Previously the list of counters was protected just by the lock field in the context, which meant that perf_counter_init_task had to take that lock and then take whatever lock/mutex protects the top-level counter's child_list. But the counter enable/disable functions need to take that lock in order to traverse the list, then for each counter take the lock in that counter's context in order to change the counter's state safely, which will lead to a deadlock. To solve this, we now have both a mutex and a spinlock in the context, and taking either is sufficient to ensure the list of counters can't change - you have to take both before changing the list. Now perf_counter_init_task takes the mutex instead of the lock (which incidentally means that inherit_counter can use GFP_KERNEL instead of GFP_ATOMIC) and thus avoids the possible deadlock. Similarly the new enable/disable functions can take the mutex while traversing the list of child counters without incurring a possible deadlock when the counter manipulation code locks the context for a child counter. We also had an misfeature that the first counter added to a context would possibly not go on until the next sched-in, because we were using ctx->nr_active to detect if the context was running on a CPU. But nr_active is the number of active counters, and if that was zero (because the context didn't have any counters yet) it would look like the context wasn't running on a cpu and so the retry code in __perf_install_in_context wouldn't retry. So this adds an 'is_active' field that is set when the context is on a CPU, even if it has no counters. The is_active field is only used for task contexts, not for per-cpu contexts. If we enable a subsidiary counter in a group that is active on a CPU, and the arch code can't enable the counter, then we have to pull the whole group off the CPU. We do this with group_sched_out, which gets moved up in the file so it comes before all its callers. This also adds similar logic to __perf_install_in_context so that the "all on, or none" invariant of groups is preserved when adding a new counter to a group. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 455 +++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 396 insertions(+), 59 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index faf671b29566..1ac18daa424f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -112,6 +112,28 @@ counter_sched_out(struct perf_counter *counter, cpuctx->exclusive = 0; } +static void +group_sched_out(struct perf_counter *group_counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx) +{ + struct perf_counter *counter; + + if (group_counter->state != PERF_COUNTER_STATE_ACTIVE) + return; + + counter_sched_out(group_counter, cpuctx, ctx); + + /* + * Schedule out siblings (if any): + */ + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) + counter_sched_out(counter, cpuctx, ctx); + + if (group_counter->hw_event.exclusive) + cpuctx->exclusive = 0; +} + /* * Cross CPU call to remove a performance counter * @@ -168,7 +190,7 @@ static void __perf_counter_remove_from_context(void *info) /* * Remove the counter from a task's (or a CPU's) list of counters. * - * Must be called with counter->mutex held. + * Must be called with counter->mutex and ctx->mutex held. * * CPU counters are removed with a smp call. For task counters we only * call when the task is on a CPU. @@ -215,6 +237,99 @@ retry: spin_unlock_irq(&ctx->lock); } +/* + * Cross CPU call to disable a performance counter + */ +static void __perf_counter_disable(void *info) +{ + struct perf_counter *counter = info; + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_counter_context *ctx = counter->ctx; + unsigned long flags; + + /* + * If this is a per-task counter, need to check whether this + * counter's task is the current task on this cpu. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + + curr_rq_lock_irq_save(&flags); + spin_lock(&ctx->lock); + + /* + * If the counter is on, turn it off. + * If it is in error state, leave it in error state. + */ + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { + if (counter == counter->group_leader) + group_sched_out(counter, cpuctx, ctx); + else + counter_sched_out(counter, cpuctx, ctx); + counter->state = PERF_COUNTER_STATE_OFF; + } + + spin_unlock(&ctx->lock); + curr_rq_unlock_irq_restore(&flags); +} + +/* + * Disable a counter. + */ +static void perf_counter_disable(struct perf_counter *counter) +{ + struct perf_counter_context *ctx = counter->ctx; + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Disable the counter on the cpu that it's on + */ + smp_call_function_single(counter->cpu, __perf_counter_disable, + counter, 1); + return; + } + + retry: + task_oncpu_function_call(task, __perf_counter_disable, counter); + + spin_lock_irq(&ctx->lock); + /* + * If the counter is still active, we need to retry the cross-call. + */ + if (counter->state == PERF_COUNTER_STATE_ACTIVE) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + + /* + * Since we have the lock this context can't be scheduled + * in, so we can change the state safely. + */ + if (counter->state == PERF_COUNTER_STATE_INACTIVE) + counter->state = PERF_COUNTER_STATE_OFF; + + spin_unlock_irq(&ctx->lock); +} + +/* + * Disable a counter and all its children. + */ +static void perf_counter_disable_family(struct perf_counter *counter) +{ + struct perf_counter *child; + + perf_counter_disable(counter); + + /* + * Lock the mutex to protect the list of children + */ + mutex_lock(&counter->mutex); + list_for_each_entry(child, &counter->child_list, child_list) + perf_counter_disable(child); + mutex_unlock(&counter->mutex); +} + static int counter_sched_in(struct perf_counter *counter, struct perf_cpu_context *cpuctx, @@ -302,6 +417,7 @@ static void __perf_install_in_context(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; + struct perf_counter *leader = counter->group_leader; int cpu = smp_processor_id(); unsigned long flags; u64 perf_flags; @@ -327,23 +443,40 @@ static void __perf_install_in_context(void *info) list_add_counter(counter, ctx); ctx->nr_counters++; + /* + * Don't put the counter on if it is disabled or if + * it is in a group and the group isn't on. + */ + if (counter->state != PERF_COUNTER_STATE_INACTIVE || + (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE)) + goto unlock; + /* * An exclusive counter can't go on if there are already active * hardware counters, and no hardware counter can go on if there * is already an exclusive counter on. */ - if (counter->state == PERF_COUNTER_STATE_INACTIVE && - !group_can_go_on(counter, cpuctx, 1)) + if (!group_can_go_on(counter, cpuctx, 1)) err = -EEXIST; else err = counter_sched_in(counter, cpuctx, ctx, cpu); - if (err && counter->hw_event.pinned) - counter->state = PERF_COUNTER_STATE_ERROR; + if (err) { + /* + * This counter couldn't go on. If it is in a group + * then we have to pull the whole group off. + * If the counter group is pinned then put it in error state. + */ + if (leader != counter) + group_sched_out(leader, cpuctx, ctx); + if (leader->hw_event.pinned) + leader->state = PERF_COUNTER_STATE_ERROR; + } if (!err && !ctx->task && cpuctx->max_pertask) cpuctx->max_pertask--; + unlock: hw_perf_restore(perf_flags); spin_unlock(&ctx->lock); @@ -359,6 +492,8 @@ static void __perf_install_in_context(void *info) * If the counter is attached to a task which is on a CPU we use a smp * call to enable it in the task context. The task might have been * scheduled away, but we check this in the smp call again. + * + * Must be called with ctx->mutex held. */ static void perf_install_in_context(struct perf_counter_context *ctx, @@ -387,7 +522,7 @@ retry: /* * we need to retry the smp call. */ - if (ctx->nr_active && list_empty(&counter->list_entry)) { + if (ctx->is_active && list_empty(&counter->list_entry)) { spin_unlock_irq(&ctx->lock); goto retry; } @@ -404,26 +539,131 @@ retry: spin_unlock_irq(&ctx->lock); } -static void -group_sched_out(struct perf_counter *group_counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx) +/* + * Cross CPU call to enable a performance counter + */ +static void __perf_counter_enable(void *info) { - struct perf_counter *counter; + struct perf_counter *counter = info; + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_counter_context *ctx = counter->ctx; + struct perf_counter *leader = counter->group_leader; + unsigned long flags; + int err; - if (group_counter->state != PERF_COUNTER_STATE_ACTIVE) + /* + * If this is a per-task counter, need to check whether this + * counter's task is the current task on this cpu. + */ + if (ctx->task && cpuctx->task_ctx != ctx) return; - counter_sched_out(group_counter, cpuctx, ctx); + curr_rq_lock_irq_save(&flags); + spin_lock(&ctx->lock); + + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) + goto unlock; + counter->state = PERF_COUNTER_STATE_INACTIVE; /* - * Schedule out siblings (if any): + * If the counter is in a group and isn't the group leader, + * then don't put it on unless the group is on. */ - list_for_each_entry(counter, &group_counter->sibling_list, list_entry) - counter_sched_out(counter, cpuctx, ctx); + if (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE) + goto unlock; - if (group_counter->hw_event.exclusive) - cpuctx->exclusive = 0; + if (!group_can_go_on(counter, cpuctx, 1)) + err = -EEXIST; + else + err = counter_sched_in(counter, cpuctx, ctx, + smp_processor_id()); + + if (err) { + /* + * If this counter can't go on and it's part of a + * group, then the whole group has to come off. + */ + if (leader != counter) + group_sched_out(leader, cpuctx, ctx); + if (leader->hw_event.pinned) + leader->state = PERF_COUNTER_STATE_ERROR; + } + + unlock: + spin_unlock(&ctx->lock); + curr_rq_unlock_irq_restore(&flags); +} + +/* + * Enable a counter. + */ +static void perf_counter_enable(struct perf_counter *counter) +{ + struct perf_counter_context *ctx = counter->ctx; + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Enable the counter on the cpu that it's on + */ + smp_call_function_single(counter->cpu, __perf_counter_enable, + counter, 1); + return; + } + + spin_lock_irq(&ctx->lock); + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) + goto out; + + /* + * If the counter is in error state, clear that first. + * That way, if we see the counter in error state below, we + * know that it has gone back into error state, as distinct + * from the task having been scheduled away before the + * cross-call arrived. + */ + if (counter->state == PERF_COUNTER_STATE_ERROR) + counter->state = PERF_COUNTER_STATE_OFF; + + retry: + spin_unlock_irq(&ctx->lock); + task_oncpu_function_call(task, __perf_counter_enable, counter); + + spin_lock_irq(&ctx->lock); + + /* + * If the context is active and the counter is still off, + * we need to retry the cross-call. + */ + if (ctx->is_active && counter->state == PERF_COUNTER_STATE_OFF) + goto retry; + + /* + * Since we have the lock this context can't be scheduled + * in, so we can change the state safely. + */ + if (counter->state == PERF_COUNTER_STATE_OFF) + counter->state = PERF_COUNTER_STATE_INACTIVE; + out: + spin_unlock_irq(&ctx->lock); +} + +/* + * Enable a counter and all its children. + */ +static void perf_counter_enable_family(struct perf_counter *counter) +{ + struct perf_counter *child; + + perf_counter_enable(counter); + + /* + * Lock the mutex to protect the list of children + */ + mutex_lock(&counter->mutex); + list_for_each_entry(child, &counter->child_list, child_list) + perf_counter_enable(child); + mutex_unlock(&counter->mutex); } void __perf_counter_sched_out(struct perf_counter_context *ctx, @@ -432,16 +672,18 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, struct perf_counter *counter; u64 flags; + spin_lock(&ctx->lock); + ctx->is_active = 0; if (likely(!ctx->nr_counters)) - return; + goto out; - spin_lock(&ctx->lock); flags = hw_perf_save_disable(); if (ctx->nr_active) { list_for_each_entry(counter, &ctx->counter_list, list_entry) group_sched_out(counter, cpuctx, ctx); } hw_perf_restore(flags); + out: spin_unlock(&ctx->lock); } @@ -528,10 +770,11 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, u64 flags; int can_add_hw = 1; + spin_lock(&ctx->lock); + ctx->is_active = 1; if (likely(!ctx->nr_counters)) - return; + goto out; - spin_lock(&ctx->lock); flags = hw_perf_save_disable(); /* @@ -578,6 +821,7 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, } } hw_perf_restore(flags); + out: spin_unlock(&ctx->lock); } @@ -896,12 +1140,14 @@ static int perf_release(struct inode *inode, struct file *file) file->private_data = NULL; + mutex_lock(&ctx->mutex); mutex_lock(&counter->mutex); perf_counter_remove_from_context(counter); put_context(ctx); mutex_unlock(&counter->mutex); + mutex_unlock(&ctx->mutex); kfree(counter); @@ -1053,10 +1299,30 @@ static unsigned int perf_poll(struct file *file, poll_table *wait) return events; } +static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + struct perf_counter *counter = file->private_data; + int err = 0; + + switch (cmd) { + case PERF_COUNTER_IOC_ENABLE: + perf_counter_enable_family(counter); + break; + case PERF_COUNTER_IOC_DISABLE: + perf_counter_disable_family(counter); + break; + default: + err = -ENOTTY; + } + return err; +} + static const struct file_operations perf_fops = { .release = perf_release, .read = perf_read, .poll = perf_poll, + .unlocked_ioctl = perf_ioctl, + .compat_ioctl = perf_ioctl, }; static int cpu_clock_perf_counter_enable(struct perf_counter *counter) @@ -1348,6 +1614,8 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, INIT_LIST_HEAD(&counter->sibling_list); init_waitqueue_head(&counter->waitq); + INIT_LIST_HEAD(&counter->child_list); + counter->irqdata = &counter->data[0]; counter->usrdata = &counter->data[1]; counter->cpu = cpu; @@ -1452,7 +1720,9 @@ sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, goto err_free_put_context; counter->filp = counter_file; + mutex_lock(&ctx->mutex); perf_install_in_context(ctx, counter, cpu); + mutex_unlock(&ctx->mutex); fput_light(counter_file, fput_needed2); @@ -1479,6 +1749,7 @@ __perf_counter_init_context(struct perf_counter_context *ctx, { memset(ctx, 0, sizeof(*ctx)); spin_lock_init(&ctx->lock); + mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->counter_list); ctx->task = task; } @@ -1486,20 +1757,30 @@ __perf_counter_init_context(struct perf_counter_context *ctx, /* * inherit a counter from parent task to child task: */ -static int +static struct perf_counter * inherit_counter(struct perf_counter *parent_counter, struct task_struct *parent, struct perf_counter_context *parent_ctx, struct task_struct *child, + struct perf_counter *group_leader, struct perf_counter_context *child_ctx) { struct perf_counter *child_counter; + /* + * Instead of creating recursive hierarchies of counters, + * we link inherited counters back to the original parent, + * which has a filp for sure, which we use as the reference + * count: + */ + if (parent_counter->parent) + parent_counter = parent_counter->parent; + child_counter = perf_counter_alloc(&parent_counter->hw_event, - parent_counter->cpu, NULL, - GFP_ATOMIC); + parent_counter->cpu, group_leader, + GFP_KERNEL); if (!child_counter) - return -ENOMEM; + return NULL; /* * Link it up in the child's context: @@ -1523,16 +1804,82 @@ inherit_counter(struct perf_counter *parent_counter, */ atomic_long_inc(&parent_counter->filp->f_count); + /* + * Link this into the parent counter's child list + */ + mutex_lock(&parent_counter->mutex); + list_add_tail(&child_counter->child_list, &parent_counter->child_list); + + /* + * Make the child state follow the state of the parent counter, + * not its hw_event.disabled bit. We hold the parent's mutex, + * so we won't race with perf_counter_{en,dis}able_family. + */ + if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE) + child_counter->state = PERF_COUNTER_STATE_INACTIVE; + else + child_counter->state = PERF_COUNTER_STATE_OFF; + + mutex_unlock(&parent_counter->mutex); + + return child_counter; +} + +static int inherit_group(struct perf_counter *parent_counter, + struct task_struct *parent, + struct perf_counter_context *parent_ctx, + struct task_struct *child, + struct perf_counter_context *child_ctx) +{ + struct perf_counter *leader; + struct perf_counter *sub; + + leader = inherit_counter(parent_counter, parent, parent_ctx, + child, NULL, child_ctx); + if (!leader) + return -ENOMEM; + list_for_each_entry(sub, &parent_counter->sibling_list, list_entry) { + if (!inherit_counter(sub, parent, parent_ctx, + child, leader, child_ctx)) + return -ENOMEM; + } return 0; } +static void sync_child_counter(struct perf_counter *child_counter, + struct perf_counter *parent_counter) +{ + u64 parent_val, child_val; + + parent_val = atomic64_read(&parent_counter->count); + child_val = atomic64_read(&child_counter->count); + + /* + * Add back the child's count to the parent's count: + */ + atomic64_add(child_val, &parent_counter->count); + + /* + * Remove this counter from the parent's list + */ + mutex_lock(&parent_counter->mutex); + list_del_init(&child_counter->child_list); + mutex_unlock(&parent_counter->mutex); + + /* + * Release the parent counter, if this was the last + * reference to it. + */ + fput(parent_counter->filp); +} + static void __perf_counter_exit_task(struct task_struct *child, struct perf_counter *child_counter, struct perf_counter_context *child_ctx) { struct perf_counter *parent_counter; - u64 parent_val, child_val; + struct perf_counter *sub, *tmp; /* * If we do not self-reap then we have to wait for the @@ -1561,7 +1908,7 @@ __perf_counter_exit_task(struct task_struct *child, cpuctx = &__get_cpu_var(perf_cpu_context); - counter_sched_out(child_counter, cpuctx, child_ctx); + group_sched_out(child_counter, cpuctx, child_ctx); list_del_init(&child_counter->list_entry); @@ -1577,26 +1924,23 @@ __perf_counter_exit_task(struct task_struct *child, * that are still around due to the child reference. These * counters need to be zapped - but otherwise linger. */ - if (!parent_counter) - return; - - parent_val = atomic64_read(&parent_counter->count); - child_val = atomic64_read(&child_counter->count); - - /* - * Add back the child's count to the parent's count: - */ - atomic64_add(child_val, &parent_counter->count); - - fput(parent_counter->filp); + if (parent_counter) { + sync_child_counter(child_counter, parent_counter); + list_for_each_entry_safe(sub, tmp, &child_counter->sibling_list, + list_entry) { + if (sub->parent) + sync_child_counter(sub, sub->parent); + kfree(sub); + } + } kfree(child_counter); } /* - * When a child task exist, feed back counter values to parent counters. + * When a child task exits, feed back counter values to parent counters. * - * Note: we are running in child context, but the PID is not hashed + * Note: we may be running in child context, but the PID is not hashed * anymore so new counters will not be added. */ void perf_counter_exit_task(struct task_struct *child) @@ -1620,9 +1964,8 @@ void perf_counter_exit_task(struct task_struct *child) void perf_counter_init_task(struct task_struct *child) { struct perf_counter_context *child_ctx, *parent_ctx; - struct perf_counter *counter, *parent_counter; + struct perf_counter *counter; struct task_struct *parent = current; - unsigned long flags; child_ctx = &child->perf_counter_ctx; parent_ctx = &parent->perf_counter_ctx; @@ -1641,32 +1984,22 @@ void perf_counter_init_task(struct task_struct *child) * Lock the parent list. No need to lock the child - not PID * hashed yet and not running, so nobody can access it. */ - spin_lock_irqsave(&parent_ctx->lock, flags); + mutex_lock(&parent_ctx->mutex); /* * We dont have to disable NMIs - we are only looking at * the list, not manipulating it: */ list_for_each_entry(counter, &parent_ctx->counter_list, list_entry) { - if (!counter->hw_event.inherit || counter->group_leader != counter) + if (!counter->hw_event.inherit) continue; - /* - * Instead of creating recursive hierarchies of counters, - * we link inheritd counters back to the original parent, - * which has a filp for sure, which we use as the reference - * count: - */ - parent_counter = counter; - if (counter->parent) - parent_counter = counter->parent; - - if (inherit_counter(parent_counter, parent, + if (inherit_group(counter, parent, parent_ctx, child, child_ctx)) break; } - spin_unlock_irqrestore(&parent_ctx->lock, flags); + mutex_unlock(&parent_ctx->mutex); } static void __cpuinit perf_counter_init_cpu(int cpu) @@ -1692,11 +2025,15 @@ static void __perf_counter_exit_cpu(void *info) list_for_each_entry_safe(counter, tmp, &ctx->counter_list, list_entry) __perf_counter_remove_from_context(counter); - } static void perf_counter_exit_cpu(int cpu) { + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_counter_context *ctx = &cpuctx->ctx; + + mutex_lock(&ctx->mutex); smp_call_function_single(cpu, __perf_counter_exit_cpu, NULL, 1); + mutex_unlock(&ctx->mutex); } #else static inline void perf_counter_exit_cpu(int cpu) { } -- cgit v1.2.3 From 65d370862f64973611a271ced61864b5f9bb6fc0 Mon Sep 17 00:00:00 2001 From: Mike Galbraith Date: Thu, 29 Jan 2009 14:06:52 +0100 Subject: perfcounters: fix refcounting bug don't kfree in use counters. Running... while true; do perfstat -e 1 -c true; done ...on all cores for a while doesn't seem to be eating ram, and my oops is gone. Signed-off-by: Mike Galbraith Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 1ac18daa424f..f27a7e9f3c41 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1934,7 +1934,8 @@ __perf_counter_exit_task(struct task_struct *child, } } - kfree(child_counter); + if (!child_counter->filp || !atomic_long_read(&child_counter->filp->f_count)) + kfree(child_counter); } /* -- cgit v1.2.3 From 23a185ca8abbeef64b6ffc33059b1d630e43ec10 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 9 Feb 2009 22:42:47 +1100 Subject: perf_counters: make software counters work as per-cpu counters Impact: kernel crash fix Yanmin Zhang reported that using a PERF_COUNT_TASK_CLOCK software counter as a per-cpu counter would reliably crash the system, because it calls __task_delta_exec with a null pointer. The page fault, context switch and cpu migration counters also won't function correctly as per-cpu counters since they reference the current task. This fixes the problem by redirecting the task_clock counter to the cpu_clock counter when used as a per-cpu counter, and by implementing per-cpu page fault, context switch and cpu migration counters. Along the way, this: - Initializes counter->ctx earlier, in perf_counter_alloc, so that sw_perf_counter_init can use it - Adds code to kernel/sched.c to count task migrations into each cpu, in rq->nr_migrations_in - Exports the per-cpu context switch and task migration counts via new functions added to kernel/sched.c - Makes sure that if sw_perf_counter_init fails, we don't try to initialize the counter as a hardware counter. Since the user has passed a negative, non-raw event type, they clearly don't intend for it to be interpreted as a hardware event. Reported-by: "Zhang Yanmin" Signed-off-by: Paul Mackerras Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 78 +++++++++++++++++++++++++++++---------------------- kernel/sched.c | 17 +++++++++++ 2 files changed, 62 insertions(+), 33 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f27a7e9f3c41..544193cbc478 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -20,6 +20,8 @@ #include #include #include +#include +#include /* * Each CPU has a list of per CPU counters: @@ -502,7 +504,6 @@ perf_install_in_context(struct perf_counter_context *ctx, { struct task_struct *task = ctx->task; - counter->ctx = ctx; if (!task) { /* * Per cpu counters are installed via an smp call and @@ -1417,11 +1418,19 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; -static u64 get_page_faults(void) +#ifdef CONFIG_VM_EVENT_COUNTERS +#define cpu_page_faults() __get_cpu_var(vm_event_states).event[PGFAULT] +#else +#define cpu_page_faults() 0 +#endif + +static u64 get_page_faults(struct perf_counter *counter) { - struct task_struct *curr = current; + struct task_struct *curr = counter->ctx->task; - return curr->maj_flt + curr->min_flt; + if (curr) + return curr->maj_flt + curr->min_flt; + return cpu_page_faults(); } static void page_faults_perf_counter_update(struct perf_counter *counter) @@ -1430,7 +1439,7 @@ static void page_faults_perf_counter_update(struct perf_counter *counter) s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = get_page_faults(); + now = get_page_faults(counter); atomic64_set(&counter->hw.prev_count, now); @@ -1446,11 +1455,7 @@ static void page_faults_perf_counter_read(struct perf_counter *counter) static int page_faults_perf_counter_enable(struct perf_counter *counter) { - /* - * page-faults is a per-task value already, - * so we dont have to clear it on switch-in. - */ - + atomic64_set(&counter->hw.prev_count, get_page_faults(counter)); return 0; } @@ -1465,11 +1470,13 @@ static const struct hw_perf_counter_ops perf_ops_page_faults = { .read = page_faults_perf_counter_read, }; -static u64 get_context_switches(void) +static u64 get_context_switches(struct perf_counter *counter) { - struct task_struct *curr = current; + struct task_struct *curr = counter->ctx->task; - return curr->nvcsw + curr->nivcsw; + if (curr) + return curr->nvcsw + curr->nivcsw; + return cpu_nr_switches(smp_processor_id()); } static void context_switches_perf_counter_update(struct perf_counter *counter) @@ -1478,7 +1485,7 @@ static void context_switches_perf_counter_update(struct perf_counter *counter) s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = get_context_switches(); + now = get_context_switches(counter); atomic64_set(&counter->hw.prev_count, now); @@ -1494,11 +1501,7 @@ static void context_switches_perf_counter_read(struct perf_counter *counter) static int context_switches_perf_counter_enable(struct perf_counter *counter) { - /* - * ->nvcsw + curr->nivcsw is a per-task value already, - * so we dont have to clear it on switch-in. - */ - + atomic64_set(&counter->hw.prev_count, get_context_switches(counter)); return 0; } @@ -1513,9 +1516,13 @@ static const struct hw_perf_counter_ops perf_ops_context_switches = { .read = context_switches_perf_counter_read, }; -static inline u64 get_cpu_migrations(void) +static inline u64 get_cpu_migrations(struct perf_counter *counter) { - return current->se.nr_migrations; + struct task_struct *curr = counter->ctx->task; + + if (curr) + return curr->se.nr_migrations; + return cpu_nr_migrations(smp_processor_id()); } static void cpu_migrations_perf_counter_update(struct perf_counter *counter) @@ -1524,7 +1531,7 @@ static void cpu_migrations_perf_counter_update(struct perf_counter *counter) s64 delta; prev = atomic64_read(&counter->hw.prev_count); - now = get_cpu_migrations(); + now = get_cpu_migrations(counter); atomic64_set(&counter->hw.prev_count, now); @@ -1540,11 +1547,7 @@ static void cpu_migrations_perf_counter_read(struct perf_counter *counter) static int cpu_migrations_perf_counter_enable(struct perf_counter *counter) { - /* - * se.nr_migrations is a per-task value already, - * so we dont have to clear it on switch-in. - */ - + atomic64_set(&counter->hw.prev_count, get_cpu_migrations(counter)); return 0; } @@ -1569,7 +1572,14 @@ sw_perf_counter_init(struct perf_counter *counter) hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_TASK_CLOCK: - hw_ops = &perf_ops_task_clock; + /* + * If the user instantiates this as a per-cpu counter, + * use the cpu_clock counter instead. + */ + if (counter->ctx->task) + hw_ops = &perf_ops_task_clock; + else + hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_PAGE_FAULTS: hw_ops = &perf_ops_page_faults; @@ -1592,6 +1602,7 @@ sw_perf_counter_init(struct perf_counter *counter) static struct perf_counter * perf_counter_alloc(struct perf_counter_hw_event *hw_event, int cpu, + struct perf_counter_context *ctx, struct perf_counter *group_leader, gfp_t gfpflags) { @@ -1623,6 +1634,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->wakeup_pending = 0; counter->group_leader = group_leader; counter->hw_ops = NULL; + counter->ctx = ctx; counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) @@ -1631,7 +1643,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, hw_ops = NULL; if (!hw_event->raw && hw_event->type < 0) hw_ops = sw_perf_counter_init(counter); - if (!hw_ops) + else hw_ops = hw_perf_counter_init(counter); if (!hw_ops) { @@ -1707,7 +1719,8 @@ sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, } ret = -EINVAL; - counter = perf_counter_alloc(&hw_event, cpu, group_leader, GFP_KERNEL); + counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader, + GFP_KERNEL); if (!counter) goto err_put_context; @@ -1777,15 +1790,14 @@ inherit_counter(struct perf_counter *parent_counter, parent_counter = parent_counter->parent; child_counter = perf_counter_alloc(&parent_counter->hw_event, - parent_counter->cpu, group_leader, - GFP_KERNEL); + parent_counter->cpu, child_ctx, + group_leader, GFP_KERNEL); if (!child_counter) return NULL; /* * Link it up in the child's context: */ - child_counter->ctx = child_ctx; child_counter->task = child; list_add_counter(child_counter, child_ctx); child_ctx->nr_counters++; diff --git a/kernel/sched.c b/kernel/sched.c index 8db1a4cf2082..173768f142ad 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -558,6 +558,7 @@ struct rq { struct load_weight load; unsigned long nr_load_updates; u64 nr_switches; + u64 nr_migrations_in; struct cfs_rq cfs; struct rt_rq rt; @@ -1908,6 +1909,7 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) #endif if (old_cpu != new_cpu) { p->se.nr_migrations++; + new_rq->nr_migrations_in++; #ifdef CONFIG_SCHEDSTATS if (task_hot(p, old_rq->clock, NULL)) schedstat_inc(p, se.nr_forced2_migrations); @@ -2810,6 +2812,21 @@ unsigned long nr_active(void) return running + uninterruptible; } +/* + * Externally visible per-cpu scheduler statistics: + * cpu_nr_switches(cpu) - number of context switches on that cpu + * cpu_nr_migrations(cpu) - number of migrations into that cpu + */ +u64 cpu_nr_switches(int cpu) +{ + return cpu_rq(cpu)->nr_switches; +} + +u64 cpu_nr_migrations(int cpu) +{ + return cpu_rq(cpu)->nr_migrations_in; +} + /* * Update rq->cpu_load[] statistics. This function is usually called every * scheduler tick (TICK_NSEC). -- cgit v1.2.3 From 0475f9ea8e2cc030298908949e0d5da9f2fc2cfe Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Wed, 11 Feb 2009 14:35:35 +1100 Subject: perf_counters: allow users to count user, kernel and/or hypervisor events Impact: new perf_counter feature This extends the perf_counter_hw_event struct with bits that specify that events in user, kernel and/or hypervisor mode should not be counted (i.e. should be excluded), and adds code to program the PMU mode selection bits accordingly on x86 and powerpc. For software counters, we don't currently have the infrastructure to distinguish which mode an event occurs in, so we currently fail the counter initialization if the setting of the hw_event.exclude_* bits would require us to distinguish. Context switches and CPU migrations are currently considered to occur in kernel mode. On x86, this changes the previous policy that only root can count kernel events. Now non-root users can count kernel events or exclude them. Non-root users still can't use NMI events, though. On x86 we don't appear to have any way to control whether hypervisor events are counted or not, so hw_event.exclude_hv is ignored. On powerpc, the selection of whether to count events in user, kernel and/or hypervisor mode is PMU-wide, not per-counter, so this adds a check that the hw_event.exclude_* settings are the same as other events on the PMU. Counters being added to a group have to have the same settings as the other hardware counters in the group. Counters and groups can only be enabled in hw_perf_group_sched_in or power_perf_enable if they have the same settings as any other counters already on the PMU. If we are not running on a hypervisor, the exclude_hv setting is ignored (by forcing it to 0) since we can't ever get any hypervisor events. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 544193cbc478..89d5e3fe9700 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1567,11 +1567,25 @@ sw_perf_counter_init(struct perf_counter *counter) { const struct hw_perf_counter_ops *hw_ops = NULL; + /* + * Software counters (currently) can't in general distinguish + * between user, kernel and hypervisor events. + * However, context switches and cpu migrations are considered + * to be kernel events, and page faults are never hypervisor + * events. + */ switch (counter->hw_event.type) { case PERF_COUNT_CPU_CLOCK: - hw_ops = &perf_ops_cpu_clock; + if (!(counter->hw_event.exclude_user || + counter->hw_event.exclude_kernel || + counter->hw_event.exclude_hv)) + hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_TASK_CLOCK: + if (counter->hw_event.exclude_user || + counter->hw_event.exclude_kernel || + counter->hw_event.exclude_hv) + break; /* * If the user instantiates this as a per-cpu counter, * use the cpu_clock counter instead. @@ -1582,13 +1596,17 @@ sw_perf_counter_init(struct perf_counter *counter) hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_PAGE_FAULTS: - hw_ops = &perf_ops_page_faults; + if (!(counter->hw_event.exclude_user || + counter->hw_event.exclude_kernel)) + hw_ops = &perf_ops_page_faults; break; case PERF_COUNT_CONTEXT_SWITCHES: - hw_ops = &perf_ops_context_switches; + if (!counter->hw_event.exclude_kernel) + hw_ops = &perf_ops_context_switches; break; case PERF_COUNT_CPU_MIGRATIONS: - hw_ops = &perf_ops_cpu_migrations; + if (!counter->hw_event.exclude_kernel) + hw_ops = &perf_ops_cpu_migrations; break; default: break; -- cgit v1.2.3 From 5af759176cc767e7426f89764bde4996ebaaf419 Mon Sep 17 00:00:00 2001 From: Mike Galbraith Date: Wed, 11 Feb 2009 10:53:37 +0100 Subject: perfcounters: fix use after free in perf_release() running... while true; do foo -d 1 -f 1 -c 100000 & sleep 1 kerneltop -d 1 -f 1 -e 1 -c 25000 -p `pidof foo` done while true; do killall foo; killall kerneltop; sleep 2 done ...in two shells with SLUB_DEBUG enabled produces flood of: BUG task_struct: Poison overwritten. Fix the use-after-free bug in perf_release(). Signed-off-by: Mike Galbraith Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 89d5e3fe9700..e0576c3fdb50 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1145,12 +1145,12 @@ static int perf_release(struct inode *inode, struct file *file) mutex_lock(&counter->mutex); perf_counter_remove_from_context(counter); - put_context(ctx); mutex_unlock(&counter->mutex); mutex_unlock(&ctx->mutex); kfree(counter); + put_context(ctx); return 0; } -- cgit v1.2.3 From 4bcf349a0f90d1e69eb35c6df0fa285c886c1cd6 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Wed, 11 Feb 2009 13:53:19 +0100 Subject: perfcounters: fix refcounting bug, take 2 Only free child_counter if it has a parent; if it doesn't, then it has a file pointing to it and we'll free it in perf_release. Signed-off-by: Mike Galbraith Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index e0576c3fdb50..fcefb0a726f3 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1958,14 +1958,13 @@ __perf_counter_exit_task(struct task_struct *child, sync_child_counter(child_counter, parent_counter); list_for_each_entry_safe(sub, tmp, &child_counter->sibling_list, list_entry) { - if (sub->parent) + if (sub->parent) { sync_child_counter(sub, sub->parent); - kfree(sub); + kfree(sub); + } } - } - - if (!child_counter->filp || !atomic_long_read(&child_counter->filp->f_count)) kfree(child_counter); + } } /* -- cgit v1.2.3 From c07c99b67233ccaad38a961c17405dc1e1542aa4 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 13 Feb 2009 22:10:34 +1100 Subject: perfcounters: make context switch and migration software counters work again Jaswinder Singh Rajput reported that commit 23a185ca8abbeef caused the context switch and migration software counters to report zero always. With that commit, the software counters only count events that occur between sched-in and sched-out for a task. This is necessary for the counter enable/disable prctls and ioctls to work. However, the context switch and migration counts are incremented after sched-out for one task and before sched-in for the next. Since the increment doesn't occur while a task is scheduled in (as far as the software counters are concerned) it doesn't count towards any counter. Thus the context switch and migration counters need to count events that occur at any time, provided the counter is enabled, not just those that occur while the task is scheduled in (from the perf_counter subsystem's point of view). The problem though is that the software counter code can't tell the difference between being enabled and being scheduled in, and between being disabled and being scheduled out, since we use the one pair of enable/disable entry points for both. That is, the high-level disable operation simply arranges for the counter to not be scheduled in any more, and the high-level enable operation arranges for it to be scheduled in again. One way to solve this would be to have sched_in/out operations in the hw_perf_counter_ops struct as well as enable/disable. However, this takes a simpler approach: it adds a 'prev_state' field to the perf_counter struct that allows a counter's enable method to know whether the counter was previously disabled or just inactive (scheduled out), and therefore whether the enable method is being called as a result of a high-level enable or a schedule-in operation. This then allows the context switch, migration and page fault counters to reset their hw.prev_count value in their enable functions only if they are called as a result of a high-level enable operation. Although page faults would normally only occur while the counter is scheduled in, this changes the page fault counter code too in case there are ever circumstances where page faults get counted against a task while its counters are not scheduled in. Reported-by: Jaswinder Singh Rajput Signed-off-by: Paul Mackerras Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index fcefb0a726f3..ad62965828d3 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -444,6 +444,7 @@ static void __perf_install_in_context(void *info) list_add_counter(counter, ctx); ctx->nr_counters++; + counter->prev_state = PERF_COUNTER_STATE_OFF; /* * Don't put the counter on if it is disabled or if @@ -562,6 +563,7 @@ static void __perf_counter_enable(void *info) curr_rq_lock_irq_save(&flags); spin_lock(&ctx->lock); + counter->prev_state = counter->state; if (counter->state >= PERF_COUNTER_STATE_INACTIVE) goto unlock; counter->state = PERF_COUNTER_STATE_INACTIVE; @@ -733,6 +735,7 @@ group_sched_in(struct perf_counter *group_counter, if (ret) return ret < 0 ? ret : 0; + group_counter->prev_state = group_counter->state; if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) return -EAGAIN; @@ -740,6 +743,7 @@ group_sched_in(struct perf_counter *group_counter, * Schedule in siblings as one group (if any): */ list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { + counter->prev_state = counter->state; if (counter_sched_in(counter, cpuctx, ctx, cpu)) { partial_group = counter; goto group_error; @@ -1398,9 +1402,9 @@ static void task_clock_perf_counter_read(struct perf_counter *counter) static int task_clock_perf_counter_enable(struct perf_counter *counter) { - u64 now = task_clock_perf_counter_val(counter, 0); - - atomic64_set(&counter->hw.prev_count, now); + if (counter->prev_state <= PERF_COUNTER_STATE_OFF) + atomic64_set(&counter->hw.prev_count, + task_clock_perf_counter_val(counter, 0)); return 0; } @@ -1455,7 +1459,8 @@ static void page_faults_perf_counter_read(struct perf_counter *counter) static int page_faults_perf_counter_enable(struct perf_counter *counter) { - atomic64_set(&counter->hw.prev_count, get_page_faults(counter)); + if (counter->prev_state <= PERF_COUNTER_STATE_OFF) + atomic64_set(&counter->hw.prev_count, get_page_faults(counter)); return 0; } @@ -1501,7 +1506,9 @@ static void context_switches_perf_counter_read(struct perf_counter *counter) static int context_switches_perf_counter_enable(struct perf_counter *counter) { - atomic64_set(&counter->hw.prev_count, get_context_switches(counter)); + if (counter->prev_state <= PERF_COUNTER_STATE_OFF) + atomic64_set(&counter->hw.prev_count, + get_context_switches(counter)); return 0; } @@ -1547,7 +1554,9 @@ static void cpu_migrations_perf_counter_read(struct perf_counter *counter) static int cpu_migrations_perf_counter_enable(struct perf_counter *counter) { - atomic64_set(&counter->hw.prev_count, get_cpu_migrations(counter)); + if (counter->prev_state <= PERF_COUNTER_STATE_OFF) + atomic64_set(&counter->hw.prev_count, + get_cpu_migrations(counter)); return 0; } -- cgit v1.2.3 From f3dfd2656deb81a0addee4f4ceff66b50a387388 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Thu, 26 Feb 2009 22:43:46 +1100 Subject: perfcounters: fix a few minor cleanliness issues This fixes three issues noticed by Arnd Bergmann: - Add #ifdef __KERNEL__ and move some things around in perf_counter.h to make sure only the bits that userspace needs are exported to userspace. - Use __u64, __s64, __u32 types in the structs exported to userspace rather than u64, s64, u32. - Make the sys_perf_counter_open syscall available to the SPUs on Cell platforms. And one issue that I noticed in looking at the code again: - Wrap the perf_counter_open syscall with SYSCALL_DEFINE4 so we get the proper handling of int arguments on ppc64 (and some other 64-bit architectures). Reported-by: Arnd Bergmann Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ad62965828d3..16b14ba99d34 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1690,9 +1690,9 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, * @cpu: target cpu * @group_fd: group leader counter fd */ -asmlinkage int -sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr __user, - pid_t pid, int cpu, int group_fd) +SYSCALL_DEFINE4(perf_counter_open, + const struct perf_counter_hw_event __user *, hw_event_uptr, + pid_t, pid, int, cpu, int, group_fd) { struct perf_counter *counter, *group_leader; struct perf_counter_hw_event hw_event; -- cgit v1.2.3 From 2743a5b0fa6f309da904f2190a9cc25deee34dbd Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Wed, 4 Mar 2009 20:36:51 +1100 Subject: perfcounters: provide expansion room in the ABI Impact: ABI change This expands several fields in the perf_counter_hw_event struct and adds a "flags" argument to the perf_counter_open system call, in order that features can be added in future without ABI changes. In particular the record_type field is expanded to 64 bits, and the space for flag bits has been expanded from 32 to 64 bits. This also adds some new fields: * read_format (64 bits) is intended to provide a way to specify what userspace wants to get back when it does a read() on a simple (non-interrupting) counter; * exclude_idle (1 bit) provides a way for userspace to ask that events that occur when the cpu is idle be excluded; * extra_config_len will provide a way for userspace to supply an arbitrary amount of extra machine-specific PMU configuration data immediately following the perf_counter_hw_event struct, to allow sophisticated users to program things such as instruction matching CAMs and address range registers; * __reserved_3 and __reserved_4 provide space for future expansion. Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 16b14ba99d34..b2e838959f3e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1683,16 +1683,16 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, } /** - * sys_perf_task_open - open a performance counter, associate it to a task/cpu + * sys_perf_counter_open - open a performance counter, associate it to a task/cpu * * @hw_event_uptr: event type attributes for monitoring/sampling * @pid: target pid * @cpu: target cpu * @group_fd: group leader counter fd */ -SYSCALL_DEFINE4(perf_counter_open, +SYSCALL_DEFINE5(perf_counter_open, const struct perf_counter_hw_event __user *, hw_event_uptr, - pid_t, pid, int, cpu, int, group_fd) + pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) { struct perf_counter *counter, *group_leader; struct perf_counter_hw_event hw_event; @@ -1703,6 +1703,10 @@ SYSCALL_DEFINE4(perf_counter_open, int fput_needed2 = 0; int ret; + /* for future expandability... */ + if (flags) + return -EINVAL; + if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0) return -EFAULT; -- cgit v1.2.3 From 7fd7d83d49914f03aefffba6aee09032fcd54cce Mon Sep 17 00:00:00 2001 From: Jeremy Fitzhardinge Date: Tue, 17 Feb 2009 23:24:03 -0800 Subject: x86/pvops: replace arch_enter_lazy_cpu_mode with arch_start_context_switch Impact: simplification, prepare for later changes Make lazy cpu mode more specific to context switching, so that it makes sense to do more context-switch specific things in the callbacks. Signed-off-by: Jeremy Fitzhardinge Acked-by: Peter Zijlstra --- kernel/sched.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 5757e03cfac0..7530fdd7c982 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2746,7 +2746,7 @@ context_switch(struct rq *rq, struct task_struct *prev, * combine the page table reload and the switch backend into * one hypercall. */ - arch_enter_lazy_cpu_mode(); + arch_start_context_switch(); if (unlikely(!mm)) { next->active_mm = oldmm; -- cgit v1.2.3 From 224101ed69d3fbb486868e0f6e0f9fa37302efb4 Mon Sep 17 00:00:00 2001 From: Jeremy Fitzhardinge Date: Wed, 18 Feb 2009 11:18:57 -0800 Subject: x86/paravirt: finish change from lazy cpu to context switch start/end Impact: fix lazy context switch API Pass the previous and next tasks into the context switch start end calls, so that the called functions can properly access the task state (esp in end_context_switch, in which the next task is not yet completely current). Signed-off-by: Jeremy Fitzhardinge Acked-by: Peter Zijlstra --- kernel/sched.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 7530fdd7c982..133762aece50 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2746,7 +2746,7 @@ context_switch(struct rq *rq, struct task_struct *prev, * combine the page table reload and the switch backend into * one hypercall. */ - arch_start_context_switch(); + arch_start_context_switch(prev); if (unlikely(!mm)) { next->active_mm = oldmm; -- cgit v1.2.3 From 3d43321b7015387cfebbe26436d0e9d299162ea1 Mon Sep 17 00:00:00 2001 From: Kees Cook Date: Thu, 2 Apr 2009 15:49:29 -0700 Subject: modules: sysctl to block module loading Implement a sysctl file that disables module-loading system-wide since there is no longer a viable way to remove CAP_SYS_MODULE after the system bounding capability set was removed in 2.6.25. Value can only be set to "1", and is tested only if standard capability checks allow CAP_SYS_MODULE. Given existing /dev/mem protections, this should allow administrators a one-way method to block module loading after initial boot-time module loading has finished. Signed-off-by: Kees Cook Acked-by: Serge Hallyn Signed-off-by: James Morris --- kernel/module.c | 7 +++++-- kernel/sysctl.c | 12 ++++++++++++ 2 files changed, 17 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/module.c b/kernel/module.c index f77ac320d0b5..eeb3f7b1383c 100644 --- a/kernel/module.c +++ b/kernel/module.c @@ -778,6 +778,9 @@ static void wait_for_zero_refcount(struct module *mod) mutex_lock(&module_mutex); } +/* Block module loading/unloading? */ +int modules_disabled = 0; + SYSCALL_DEFINE2(delete_module, const char __user *, name_user, unsigned int, flags) { @@ -785,7 +788,7 @@ SYSCALL_DEFINE2(delete_module, const char __user *, name_user, char name[MODULE_NAME_LEN]; int ret, forced = 0; - if (!capable(CAP_SYS_MODULE)) + if (!capable(CAP_SYS_MODULE) || modules_disabled) return -EPERM; if (strncpy_from_user(name, name_user, MODULE_NAME_LEN-1) < 0) @@ -2349,7 +2352,7 @@ SYSCALL_DEFINE3(init_module, void __user *, umod, int ret = 0; /* Must have permission */ - if (!capable(CAP_SYS_MODULE)) + if (!capable(CAP_SYS_MODULE) || modules_disabled) return -EPERM; /* Only one module load at a time, please */ diff --git a/kernel/sysctl.c b/kernel/sysctl.c index c5ef44ff850f..2fb4246d27de 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -113,6 +113,7 @@ static int ngroups_max = NGROUPS_MAX; #ifdef CONFIG_MODULES extern char modprobe_path[]; +extern int modules_disabled; #endif #ifdef CONFIG_CHR_DEV_SG extern int sg_big_buff; @@ -533,6 +534,17 @@ static struct ctl_table kern_table[] = { .proc_handler = &proc_dostring, .strategy = &sysctl_string, }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "modules_disabled", + .data = &modules_disabled, + .maxlen = sizeof(int), + .mode = 0644, + /* only handle a transition from default "0" to "1" */ + .proc_handler = &proc_dointvec_minmax, + .extra1 = &one, + .extra2 = &one, + }, #endif #if defined(CONFIG_HOTPLUG) && defined(CONFIG_NET) { -- cgit v1.2.3 From 755642322aa66fbc5421a35fd3e1733f73e20083 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:29 +0100 Subject: perf_counter: use list_move_tail() Instead of del/add use a move list-op. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index b2e838959f3e..0fe22c916e29 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -89,8 +89,7 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_for_each_entry_safe(sibling, tmp, &counter->sibling_list, list_entry) { - list_del_init(&sibling->list_entry); - list_add_tail(&sibling->list_entry, &ctx->counter_list); + list_move_tail(&sibling->list_entry, &ctx->counter_list); sibling->group_leader = sibling; } } @@ -959,8 +958,7 @@ static void rotate_ctx(struct perf_counter_context *ctx) */ perf_flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { - list_del(&counter->list_entry); - list_add_tail(&counter->list_entry, &ctx->counter_list); + list_move_tail(&counter->list_entry, &ctx->counter_list); break; } hw_perf_restore(perf_flags); -- cgit v1.2.3 From 15dbf27cc18559a14e99609f78678aa86b9c6ff1 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:32 +0100 Subject: perf_counter: software counter event infrastructure Provide generic software counter infrastructure that supports software events. This will be used to allow sample based profiling based on software events such as pagefaults. The current infrastructure can only provide a count of such events, no place information. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 201 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 201 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0fe22c916e29..eeb1b46cf707 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1328,6 +1328,185 @@ static const struct file_operations perf_fops = { .compat_ioctl = perf_ioctl, }; +/* + * Generic software counter infrastructure + */ + +static void perf_swcounter_update(struct perf_counter *counter) +{ + struct hw_perf_counter *hwc = &counter->hw; + u64 prev, now; + s64 delta; + +again: + prev = atomic64_read(&hwc->prev_count); + now = atomic64_read(&hwc->count); + if (atomic64_cmpxchg(&hwc->prev_count, prev, now) != prev) + goto again; + + delta = now - prev; + + atomic64_add(delta, &counter->count); + atomic64_sub(delta, &hwc->period_left); +} + +static void perf_swcounter_set_period(struct perf_counter *counter) +{ + struct hw_perf_counter *hwc = &counter->hw; + s64 left = atomic64_read(&hwc->period_left); + s64 period = hwc->irq_period; + + if (unlikely(left <= -period)) { + left = period; + atomic64_set(&hwc->period_left, left); + } + + if (unlikely(left <= 0)) { + left += period; + atomic64_add(period, &hwc->period_left); + } + + atomic64_set(&hwc->prev_count, -left); + atomic64_set(&hwc->count, -left); +} + +static void perf_swcounter_save_and_restart(struct perf_counter *counter) +{ + perf_swcounter_update(counter); + perf_swcounter_set_period(counter); +} + +static void perf_swcounter_store_irq(struct perf_counter *counter, u64 data) +{ + struct perf_data *irqdata = counter->irqdata; + + if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) { + irqdata->overrun++; + } else { + u64 *p = (u64 *) &irqdata->data[irqdata->len]; + + *p = data; + irqdata->len += sizeof(u64); + } +} + +static void perf_swcounter_handle_group(struct perf_counter *sibling) +{ + struct perf_counter *counter, *group_leader = sibling->group_leader; + + list_for_each_entry(counter, &group_leader->sibling_list, list_entry) { + perf_swcounter_update(counter); + perf_swcounter_store_irq(sibling, counter->hw_event.type); + perf_swcounter_store_irq(sibling, atomic64_read(&counter->count)); + } +} + +static void perf_swcounter_interrupt(struct perf_counter *counter, + int nmi, struct pt_regs *regs) +{ + perf_swcounter_save_and_restart(counter); + + switch (counter->hw_event.record_type) { + case PERF_RECORD_SIMPLE: + break; + + case PERF_RECORD_IRQ: + perf_swcounter_store_irq(counter, instruction_pointer(regs)); + break; + + case PERF_RECORD_GROUP: + perf_swcounter_handle_group(counter); + break; + } + + if (nmi) { + counter->wakeup_pending = 1; + set_tsk_thread_flag(current, TIF_PERF_COUNTERS); + } else + wake_up(&counter->waitq); +} + +static int perf_swcounter_match(struct perf_counter *counter, + enum hw_event_types event, + struct pt_regs *regs) +{ + if (counter->state != PERF_COUNTER_STATE_ACTIVE) + return 0; + + if (counter->hw_event.raw) + return 0; + + if (counter->hw_event.type != event) + return 0; + + if (counter->hw_event.exclude_user && user_mode(regs)) + return 0; + + if (counter->hw_event.exclude_kernel && !user_mode(regs)) + return 0; + + return 1; +} + +static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, + enum hw_event_types event, u64 nr, + int nmi, struct pt_regs *regs) +{ + struct perf_counter *counter; + unsigned long flags; + int neg; + + if (list_empty(&ctx->counter_list)) + return; + + spin_lock_irqsave(&ctx->lock, flags); + + /* + * XXX: make counter_list RCU safe + */ + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (perf_swcounter_match(counter, event, regs)) { + neg = atomic64_add_negative(nr, &counter->hw.count); + if (counter->hw.irq_period && !neg) + perf_swcounter_interrupt(counter, nmi, regs); + } + } + + spin_unlock_irqrestore(&ctx->lock, flags); +} + +void perf_swcounter_event(enum hw_event_types event, u64 nr, + int nmi, struct pt_regs *regs) +{ + struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); + + perf_swcounter_ctx_event(&cpuctx->ctx, event, nr, nmi, regs); + if (cpuctx->task_ctx) + perf_swcounter_ctx_event(cpuctx->task_ctx, event, nr, nmi, regs); + + put_cpu_var(perf_cpu_context); +} + +static void perf_swcounter_read(struct perf_counter *counter) +{ + perf_swcounter_update(counter); +} + +static int perf_swcounter_enable(struct perf_counter *counter) +{ + perf_swcounter_set_period(counter); + return 0; +} + +static void perf_swcounter_disable(struct perf_counter *counter) +{ + perf_swcounter_update(counter); +} + +/* + * Software counter: cpu wall time clock + */ + static int cpu_clock_perf_counter_enable(struct perf_counter *counter) { int cpu = raw_smp_processor_id(); @@ -1364,6 +1543,10 @@ static const struct hw_perf_counter_ops perf_ops_cpu_clock = { .read = cpu_clock_perf_counter_read, }; +/* + * Software counter: task time clock + */ + /* * Called from within the scheduler: */ @@ -1420,6 +1603,10 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; +/* + * Software counter: page faults + */ + #ifdef CONFIG_VM_EVENT_COUNTERS #define cpu_page_faults() __get_cpu_var(vm_event_states).event[PGFAULT] #else @@ -1473,6 +1660,10 @@ static const struct hw_perf_counter_ops perf_ops_page_faults = { .read = page_faults_perf_counter_read, }; +/* + * Software counter: context switches + */ + static u64 get_context_switches(struct perf_counter *counter) { struct task_struct *curr = counter->ctx->task; @@ -1521,6 +1712,10 @@ static const struct hw_perf_counter_ops perf_ops_context_switches = { .read = context_switches_perf_counter_read, }; +/* + * Software counter: cpu migrations + */ + static inline u64 get_cpu_migrations(struct perf_counter *counter) { struct task_struct *curr = counter->ctx->task; @@ -1572,7 +1767,9 @@ static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { static const struct hw_perf_counter_ops * sw_perf_counter_init(struct perf_counter *counter) { + struct perf_counter_hw_event *hw_event = &counter->hw_event; const struct hw_perf_counter_ops *hw_ops = NULL; + struct hw_perf_counter *hwc = &counter->hw; /* * Software counters (currently) can't in general distinguish @@ -1618,6 +1815,10 @@ sw_perf_counter_init(struct perf_counter *counter) default: break; } + + if (hw_ops) + hwc->irq_period = hw_event->irq_period; + return hw_ops; } -- cgit v1.2.3 From 7dd1fcc258b65da718f01e4684a7b9244501a9fb Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:33 +0100 Subject: perf_counter: provide pagefault software events We use the generic software counter infrastructure to provide page fault events. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 53 +++------------------------------------------------ 1 file changed, 3 insertions(+), 50 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index eeb1b46cf707..1773c5d7427d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1607,57 +1607,10 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { * Software counter: page faults */ -#ifdef CONFIG_VM_EVENT_COUNTERS -#define cpu_page_faults() __get_cpu_var(vm_event_states).event[PGFAULT] -#else -#define cpu_page_faults() 0 -#endif - -static u64 get_page_faults(struct perf_counter *counter) -{ - struct task_struct *curr = counter->ctx->task; - - if (curr) - return curr->maj_flt + curr->min_flt; - return cpu_page_faults(); -} - -static void page_faults_perf_counter_update(struct perf_counter *counter) -{ - u64 prev, now; - s64 delta; - - prev = atomic64_read(&counter->hw.prev_count); - now = get_page_faults(counter); - - atomic64_set(&counter->hw.prev_count, now); - - delta = now - prev; - - atomic64_add(delta, &counter->count); -} - -static void page_faults_perf_counter_read(struct perf_counter *counter) -{ - page_faults_perf_counter_update(counter); -} - -static int page_faults_perf_counter_enable(struct perf_counter *counter) -{ - if (counter->prev_state <= PERF_COUNTER_STATE_OFF) - atomic64_set(&counter->hw.prev_count, get_page_faults(counter)); - return 0; -} - -static void page_faults_perf_counter_disable(struct perf_counter *counter) -{ - page_faults_perf_counter_update(counter); -} - static const struct hw_perf_counter_ops perf_ops_page_faults = { - .enable = page_faults_perf_counter_enable, - .disable = page_faults_perf_counter_disable, - .read = page_faults_perf_counter_read, + .enable = perf_swcounter_enable, + .disable = perf_swcounter_disable, + .read = perf_swcounter_read, }; /* -- cgit v1.2.3 From ac17dc8e58f3069ea895cfff963adf98ff3cf6b2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:34 +0100 Subject: perf_counter: provide major/minor page fault software events Provide separate sw counters for major and minor page faults. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 22 +++++++++------------- 1 file changed, 9 insertions(+), 13 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 1773c5d7427d..68950a3a52bf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1503,6 +1503,12 @@ static void perf_swcounter_disable(struct perf_counter *counter) perf_swcounter_update(counter); } +static const struct hw_perf_counter_ops perf_ops_generic = { + .enable = perf_swcounter_enable, + .disable = perf_swcounter_disable, + .read = perf_swcounter_read, +}; + /* * Software counter: cpu wall time clock */ @@ -1603,16 +1609,6 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; -/* - * Software counter: page faults - */ - -static const struct hw_perf_counter_ops perf_ops_page_faults = { - .enable = perf_swcounter_enable, - .disable = perf_swcounter_disable, - .read = perf_swcounter_read, -}; - /* * Software counter: context switches */ @@ -1753,9 +1749,9 @@ sw_perf_counter_init(struct perf_counter *counter) hw_ops = &perf_ops_cpu_clock; break; case PERF_COUNT_PAGE_FAULTS: - if (!(counter->hw_event.exclude_user || - counter->hw_event.exclude_kernel)) - hw_ops = &perf_ops_page_faults; + case PERF_COUNT_PAGE_FAULTS_MIN: + case PERF_COUNT_PAGE_FAULTS_MAJ: + hw_ops = &perf_ops_generic; break; case PERF_COUNT_CONTEXT_SWITCHES: if (!counter->hw_event.exclude_kernel) -- cgit v1.2.3 From d6d020e9957745c61285ef3da9f294c5e6801f0f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:35 +0100 Subject: perf_counter: hrtimer based sampling for software time events Use hrtimers to profile timer based sampling for the software time counters. This allows platforms without hardware counter support to still perform sample based profiling. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 123 +++++++++++++++++++++++++++++++++++--------------- 1 file changed, 86 insertions(+), 37 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 68950a3a52bf..f9330d5827cf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1395,7 +1395,7 @@ static void perf_swcounter_handle_group(struct perf_counter *sibling) struct perf_counter *counter, *group_leader = sibling->group_leader; list_for_each_entry(counter, &group_leader->sibling_list, list_entry) { - perf_swcounter_update(counter); + counter->hw_ops->read(counter); perf_swcounter_store_irq(sibling, counter->hw_event.type); perf_swcounter_store_irq(sibling, atomic64_read(&counter->count)); } @@ -1404,8 +1404,6 @@ static void perf_swcounter_handle_group(struct perf_counter *sibling) static void perf_swcounter_interrupt(struct perf_counter *counter, int nmi, struct pt_regs *regs) { - perf_swcounter_save_and_restart(counter); - switch (counter->hw_event.record_type) { case PERF_RECORD_SIMPLE: break; @@ -1426,6 +1424,38 @@ static void perf_swcounter_interrupt(struct perf_counter *counter, wake_up(&counter->waitq); } +static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) +{ + struct perf_counter *counter; + struct pt_regs *regs; + + counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); + counter->hw_ops->read(counter); + + regs = get_irq_regs(); + /* + * In case we exclude kernel IPs or are somehow not in interrupt + * context, provide the next best thing, the user IP. + */ + if ((counter->hw_event.exclude_kernel || !regs) && + !counter->hw_event.exclude_user) + regs = task_pt_regs(current); + + if (regs) + perf_swcounter_interrupt(counter, 0, regs); + + hrtimer_forward_now(hrtimer, ns_to_ktime(counter->hw.irq_period)); + + return HRTIMER_RESTART; +} + +static void perf_swcounter_overflow(struct perf_counter *counter, + int nmi, struct pt_regs *regs) +{ + perf_swcounter_save_and_restart(counter); + perf_swcounter_interrupt(counter, nmi, regs); +} + static int perf_swcounter_match(struct perf_counter *counter, enum hw_event_types event, struct pt_regs *regs) @@ -1448,13 +1478,20 @@ static int perf_swcounter_match(struct perf_counter *counter, return 1; } +static void perf_swcounter_add(struct perf_counter *counter, u64 nr, + int nmi, struct pt_regs *regs) +{ + int neg = atomic64_add_negative(nr, &counter->hw.count); + if (counter->hw.irq_period && !neg) + perf_swcounter_overflow(counter, nmi, regs); +} + static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, enum hw_event_types event, u64 nr, int nmi, struct pt_regs *regs) { struct perf_counter *counter; unsigned long flags; - int neg; if (list_empty(&ctx->counter_list)) return; @@ -1465,11 +1502,8 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, * XXX: make counter_list RCU safe */ list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (perf_swcounter_match(counter, event, regs)) { - neg = atomic64_add_negative(nr, &counter->hw.count); - if (counter->hw.irq_period && !neg) - perf_swcounter_interrupt(counter, nmi, regs); - } + if (perf_swcounter_match(counter, event, regs)) + perf_swcounter_add(counter, nr, nmi, regs); } spin_unlock_irqrestore(&ctx->lock, flags); @@ -1513,14 +1547,6 @@ static const struct hw_perf_counter_ops perf_ops_generic = { * Software counter: cpu wall time clock */ -static int cpu_clock_perf_counter_enable(struct perf_counter *counter) -{ - int cpu = raw_smp_processor_id(); - - atomic64_set(&counter->hw.prev_count, cpu_clock(cpu)); - return 0; -} - static void cpu_clock_perf_counter_update(struct perf_counter *counter) { int cpu = raw_smp_processor_id(); @@ -1533,8 +1559,26 @@ static void cpu_clock_perf_counter_update(struct perf_counter *counter) atomic64_add(now - prev, &counter->count); } +static int cpu_clock_perf_counter_enable(struct perf_counter *counter) +{ + struct hw_perf_counter *hwc = &counter->hw; + int cpu = raw_smp_processor_id(); + + atomic64_set(&hwc->prev_count, cpu_clock(cpu)); + if (hwc->irq_period) { + hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hwc->hrtimer.function = perf_swcounter_hrtimer; + __hrtimer_start_range_ns(&hwc->hrtimer, + ns_to_ktime(hwc->irq_period), 0, + HRTIMER_MODE_REL, 0); + } + + return 0; +} + static void cpu_clock_perf_counter_disable(struct perf_counter *counter) { + hrtimer_cancel(&counter->hw.hrtimer); cpu_clock_perf_counter_update(counter); } @@ -1580,27 +1624,33 @@ static void task_clock_perf_counter_update(struct perf_counter *counter, u64 now atomic64_add(delta, &counter->count); } -static void task_clock_perf_counter_read(struct perf_counter *counter) -{ - u64 now = task_clock_perf_counter_val(counter, 1); - - task_clock_perf_counter_update(counter, now); -} - static int task_clock_perf_counter_enable(struct perf_counter *counter) { - if (counter->prev_state <= PERF_COUNTER_STATE_OFF) - atomic64_set(&counter->hw.prev_count, - task_clock_perf_counter_val(counter, 0)); + struct hw_perf_counter *hwc = &counter->hw; + + atomic64_set(&hwc->prev_count, task_clock_perf_counter_val(counter, 0)); + if (hwc->irq_period) { + hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hwc->hrtimer.function = perf_swcounter_hrtimer; + __hrtimer_start_range_ns(&hwc->hrtimer, + ns_to_ktime(hwc->irq_period), 0, + HRTIMER_MODE_REL, 0); + } return 0; } static void task_clock_perf_counter_disable(struct perf_counter *counter) { - u64 now = task_clock_perf_counter_val(counter, 0); + hrtimer_cancel(&counter->hw.hrtimer); + task_clock_perf_counter_update(counter, + task_clock_perf_counter_val(counter, 0)); +} - task_clock_perf_counter_update(counter, now); +static void task_clock_perf_counter_read(struct perf_counter *counter) +{ + task_clock_perf_counter_update(counter, + task_clock_perf_counter_val(counter, 1)); } static const struct hw_perf_counter_ops perf_ops_task_clock = { @@ -1729,16 +1779,12 @@ sw_perf_counter_init(struct perf_counter *counter) */ switch (counter->hw_event.type) { case PERF_COUNT_CPU_CLOCK: - if (!(counter->hw_event.exclude_user || - counter->hw_event.exclude_kernel || - counter->hw_event.exclude_hv)) - hw_ops = &perf_ops_cpu_clock; + hw_ops = &perf_ops_cpu_clock; + + if (hw_event->irq_period && hw_event->irq_period < 10000) + hw_event->irq_period = 10000; break; case PERF_COUNT_TASK_CLOCK: - if (counter->hw_event.exclude_user || - counter->hw_event.exclude_kernel || - counter->hw_event.exclude_hv) - break; /* * If the user instantiates this as a per-cpu counter, * use the cpu_clock counter instead. @@ -1747,6 +1793,9 @@ sw_perf_counter_init(struct perf_counter *counter) hw_ops = &perf_ops_task_clock; else hw_ops = &perf_ops_cpu_clock; + + if (hw_event->irq_period && hw_event->irq_period < 10000) + hw_event->irq_period = 10000; break; case PERF_COUNT_PAGE_FAULTS: case PERF_COUNT_PAGE_FAULTS_MIN: -- cgit v1.2.3 From 592903cdcbf606a838056bae6d03fc557806c914 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:36 +0100 Subject: perf_counter: add an event_list I noticed that the counter_list only includes top-level counters, thus perf_swcounter_event() will miss sw-counters in groups. Since perf_swcounter_event() also wants an RCU safe list, create a new event_list that includes all counters and uses RCU list ops and use call_rcu to free the counter structure. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 30 +++++++++++++++++++----------- 1 file changed, 19 insertions(+), 11 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f9330d5827cf..8d6ecfa64c0e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -22,6 +22,7 @@ #include #include #include +#include /* * Each CPU has a list of per CPU counters: @@ -72,6 +73,8 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_add_tail(&counter->list_entry, &ctx->counter_list); else list_add_tail(&counter->list_entry, &group_leader->sibling_list); + + list_add_rcu(&counter->event_entry, &ctx->event_list); } static void @@ -80,6 +83,7 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) struct perf_counter *sibling, *tmp; list_del_init(&counter->list_entry); + list_del_rcu(&counter->event_entry); /* * If this was a group counter with sibling counters then @@ -1133,6 +1137,14 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) return ctx; } +static void free_counter_rcu(struct rcu_head *head) +{ + struct perf_counter *counter; + + counter = container_of(head, struct perf_counter, rcu_head); + kfree(counter); +} + /* * Called when the last reference to the file is gone. */ @@ -1151,7 +1163,7 @@ static int perf_release(struct inode *inode, struct file *file) mutex_unlock(&counter->mutex); mutex_unlock(&ctx->mutex); - kfree(counter); + call_rcu(&counter->rcu_head, free_counter_rcu); put_context(ctx); return 0; @@ -1491,22 +1503,16 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, int nmi, struct pt_regs *regs) { struct perf_counter *counter; - unsigned long flags; - if (list_empty(&ctx->counter_list)) + if (list_empty(&ctx->event_list)) return; - spin_lock_irqsave(&ctx->lock, flags); - - /* - * XXX: make counter_list RCU safe - */ - list_for_each_entry(counter, &ctx->counter_list, list_entry) { + rcu_read_lock(); + list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { if (perf_swcounter_match(counter, event, regs)) perf_swcounter_add(counter, nr, nmi, regs); } - - spin_unlock_irqrestore(&ctx->lock, flags); + rcu_read_unlock(); } void perf_swcounter_event(enum hw_event_types event, u64 nr, @@ -1846,6 +1852,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, mutex_init(&counter->mutex); INIT_LIST_HEAD(&counter->list_entry); + INIT_LIST_HEAD(&counter->event_entry); INIT_LIST_HEAD(&counter->sibling_list); init_waitqueue_head(&counter->waitq); @@ -1992,6 +1999,7 @@ __perf_counter_init_context(struct perf_counter_context *ctx, spin_lock_init(&ctx->lock); mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->counter_list); + INIT_LIST_HEAD(&ctx->event_list); ctx->task = task; } -- cgit v1.2.3 From 039fc91e064b81c2820ff16c304be5aba35fd126 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 16:43:47 +0100 Subject: perf_counter: fix hrtimer sampling Impact: fix deadlock with perfstat Fix for the perfstat fubar.. We cannot unconditionally call hrtimer_cancel() without ever having done hrtimer_init() on the thing. Signed-off-by: Peter Zijlstra Orig-LKML-Reference: <1236959027.22447.149.camel@twins> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 8d6ecfa64c0e..d6cc22271ef4 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1571,9 +1571,9 @@ static int cpu_clock_perf_counter_enable(struct perf_counter *counter) int cpu = raw_smp_processor_id(); atomic64_set(&hwc->prev_count, cpu_clock(cpu)); + hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hwc->hrtimer.function = perf_swcounter_hrtimer; if (hwc->irq_period) { - hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); - hwc->hrtimer.function = perf_swcounter_hrtimer; __hrtimer_start_range_ns(&hwc->hrtimer, ns_to_ktime(hwc->irq_period), 0, HRTIMER_MODE_REL, 0); @@ -1635,9 +1635,9 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) struct hw_perf_counter *hwc = &counter->hw; atomic64_set(&hwc->prev_count, task_clock_perf_counter_val(counter, 0)); + hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hwc->hrtimer.function = perf_swcounter_hrtimer; if (hwc->irq_period) { - hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); - hwc->hrtimer.function = perf_swcounter_hrtimer; __hrtimer_start_range_ns(&hwc->hrtimer, ns_to_ktime(hwc->irq_period), 0, HRTIMER_MODE_REL, 0); -- cgit v1.2.3 From 4e193bd4dfdc983d12969b51439b4a1fbaf2daad Mon Sep 17 00:00:00 2001 From: Tim Blechmann Date: Sat, 14 Mar 2009 14:29:25 +0100 Subject: perf_counter: include missing header Impact: build fix In order to compile a kernel with performance counter patches, has to be included to provide the declaration of struct pt_regs *get_irq_regs(void); [ This bug was masked by unrelated x86 header file changes in the x86 tree, but occurs in the tip:perfcounters/core standalone tree. ] Signed-off-by: Tim Blechmann Orig-LKML-Reference: <20090314142925.49c29c17@thinkpad> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d6cc22271ef4..0018c5e81249 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -24,6 +24,8 @@ #include #include +#include + /* * Each CPU has a list of per CPU counters: */ -- cgit v1.2.3 From b6c5a71da1477d261bc36254fe1f20d32b57598d Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 16 Mar 2009 21:00:00 +1100 Subject: perf_counter: abstract wakeup flag setting in core to fix powerpc build Impact: build fix for powerpc Commit bd753921015e7905 ("perf_counter: software counter event infrastructure") introduced a use of TIF_PERF_COUNTERS into the core perfcounter code. This breaks the build on powerpc because we use a flag in a per-cpu area to signal wakeups on powerpc rather than a thread_info flag, because the thread_info flags have to be manipulated with atomic operations and are thus slower than per-cpu flags. This fixes the by changing the core to use an abstracted set_perf_counter_pending() function, which is defined on x86 to set the TIF_PERF_COUNTERS flag and on powerpc to set the per-cpu flag (paca->perf_counter_pending). It changes the previous powerpc definition of set_perf_counter_pending to not take an argument and adds a clear_perf_counter_pending, so as to simplify the definition on x86. On x86, set_perf_counter_pending() is defined as a macro. Defining it as a static inline in arch/x86/include/asm/perf_counters.h causes compile failures because gets included early in , and the definitions of set_tsk_thread_flag etc. are therefore not available in . (On powerpc this problem is avoided by defining set_perf_counter_pending etc. in .) Signed-off-by: Paul Mackerras --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0018c5e81249..b39456ad74a1 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1433,7 +1433,7 @@ static void perf_swcounter_interrupt(struct perf_counter *counter, if (nmi) { counter->wakeup_pending = 1; - set_tsk_thread_flag(current, TIF_PERF_COUNTERS); + set_perf_counter_pending(); } else wake_up(&counter->waitq); } -- cgit v1.2.3 From 01ef09d9ffb5ce9f8d62d1e5206da3d5ca612acc Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:11 +0100 Subject: perf_counter: fix uninitialized usage of event_list Impact: fix boot crash When doing the generic context switch event I ran into some early boot hangs, which were caused by inf func recursion (event, fault, event, fault). I eventually tracked it down to event_list not being initialized at the time of the first event. Fix this. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.195392657@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index b39456ad74a1..4c4e9eb37ab0 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1506,7 +1506,7 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, { struct perf_counter *counter; - if (list_empty(&ctx->event_list)) + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) return; rcu_read_lock(); -- cgit v1.2.3 From 4a0deca657f3dbb8a707b5dc8f173beec01e7ed2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:12 +0100 Subject: perf_counter: generic context switch event Impact: cleanup Use the generic software events for context switches. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.283522645@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 60 ++++----------------------------------------------- kernel/sched.c | 6 ------ 2 files changed, 4 insertions(+), 62 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4c4e9eb37ab0..99d5930f0a52 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -710,10 +710,13 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct pt_regs *regs; if (likely(!cpuctx->task_ctx)) return; + regs = task_pt_regs(task); + perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs); __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; @@ -1667,58 +1670,6 @@ static const struct hw_perf_counter_ops perf_ops_task_clock = { .read = task_clock_perf_counter_read, }; -/* - * Software counter: context switches - */ - -static u64 get_context_switches(struct perf_counter *counter) -{ - struct task_struct *curr = counter->ctx->task; - - if (curr) - return curr->nvcsw + curr->nivcsw; - return cpu_nr_switches(smp_processor_id()); -} - -static void context_switches_perf_counter_update(struct perf_counter *counter) -{ - u64 prev, now; - s64 delta; - - prev = atomic64_read(&counter->hw.prev_count); - now = get_context_switches(counter); - - atomic64_set(&counter->hw.prev_count, now); - - delta = now - prev; - - atomic64_add(delta, &counter->count); -} - -static void context_switches_perf_counter_read(struct perf_counter *counter) -{ - context_switches_perf_counter_update(counter); -} - -static int context_switches_perf_counter_enable(struct perf_counter *counter) -{ - if (counter->prev_state <= PERF_COUNTER_STATE_OFF) - atomic64_set(&counter->hw.prev_count, - get_context_switches(counter)); - return 0; -} - -static void context_switches_perf_counter_disable(struct perf_counter *counter) -{ - context_switches_perf_counter_update(counter); -} - -static const struct hw_perf_counter_ops perf_ops_context_switches = { - .enable = context_switches_perf_counter_enable, - .disable = context_switches_perf_counter_disable, - .read = context_switches_perf_counter_read, -}; - /* * Software counter: cpu migrations */ @@ -1808,11 +1759,8 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_PAGE_FAULTS: case PERF_COUNT_PAGE_FAULTS_MIN: case PERF_COUNT_PAGE_FAULTS_MAJ: - hw_ops = &perf_ops_generic; - break; case PERF_COUNT_CONTEXT_SWITCHES: - if (!counter->hw_event.exclude_kernel) - hw_ops = &perf_ops_context_switches; + hw_ops = &perf_ops_generic; break; case PERF_COUNT_CPU_MIGRATIONS: if (!counter->hw_event.exclude_kernel) diff --git a/kernel/sched.c b/kernel/sched.c index 39e708602169..f76e3c0188a2 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2900,14 +2900,8 @@ unsigned long nr_active(void) /* * Externally visible per-cpu scheduler statistics: - * cpu_nr_switches(cpu) - number of context switches on that cpu * cpu_nr_migrations(cpu) - number of migrations into that cpu */ -u64 cpu_nr_switches(int cpu) -{ - return cpu_rq(cpu)->nr_switches; -} - u64 cpu_nr_migrations(int cpu) { return cpu_rq(cpu)->nr_migrations_in; -- cgit v1.2.3 From f16009527595ee562308653bc3d0039166d2ab15 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:16 +0100 Subject: perf_counter: fix up counter free paths Impact: fix crash during perfcounters use I found another counter free path, create a free_counter() call to accomodate generic tear-down. Fixes an RCU bug. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.652078652@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 99d5930f0a52..97f891ffeb40 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1150,6 +1150,11 @@ static void free_counter_rcu(struct rcu_head *head) kfree(counter); } +static void free_counter(struct perf_counter *counter) +{ + call_rcu(&counter->rcu_head, free_counter_rcu); +} + /* * Called when the last reference to the file is gone. */ @@ -1168,7 +1173,7 @@ static int perf_release(struct inode *inode, struct file *file) mutex_unlock(&counter->mutex); mutex_unlock(&ctx->mutex); - call_rcu(&counter->rcu_head, free_counter_rcu); + free_counter(counter); put_context(ctx); return 0; @@ -2128,10 +2133,10 @@ __perf_counter_exit_task(struct task_struct *child, list_entry) { if (sub->parent) { sync_child_counter(sub, sub->parent); - kfree(sub); + free_counter(sub); } } - kfree(child_counter); + free_counter(child_counter); } } -- cgit v1.2.3 From e077df4f439681e43f0db8255b2d215b342ebdc6 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:17 +0100 Subject: perf_counter: hook up the tracepoint events Impact: new perfcounters feature Enable usage of tracepoints as perf counter events. tracepoint event ids can be found in /debug/tracing/event/*/*/id and (for now) are represented as -65536+id in the type field. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.744044174@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 43 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 43 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 97f891ffeb40..0bbe3e45ba0d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1152,6 +1152,9 @@ static void free_counter_rcu(struct rcu_head *head) static void free_counter(struct perf_counter *counter) { + if (counter->destroy) + counter->destroy(counter); + call_rcu(&counter->rcu_head, free_counter_rcu); } @@ -1727,6 +1730,45 @@ static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { .read = cpu_migrations_perf_counter_read, }; +#ifdef CONFIG_EVENT_PROFILE +void perf_tpcounter_event(int event_id) +{ + perf_swcounter_event(PERF_TP_EVENTS_MIN + event_id, 1, 1, + task_pt_regs(current)); +} + +extern int ftrace_profile_enable(int); +extern void ftrace_profile_disable(int); + +static void tp_perf_counter_destroy(struct perf_counter *counter) +{ + int event_id = counter->hw_event.type - PERF_TP_EVENTS_MIN; + + ftrace_profile_disable(event_id); +} + +static const struct hw_perf_counter_ops * +tp_perf_counter_init(struct perf_counter *counter) +{ + int event_id = counter->hw_event.type - PERF_TP_EVENTS_MIN; + int ret; + + ret = ftrace_profile_enable(event_id); + if (ret) + return NULL; + + counter->destroy = tp_perf_counter_destroy; + + return &perf_ops_generic; +} +#else +static const struct hw_perf_counter_ops * +tp_perf_counter_init(struct perf_counter *counter) +{ + return NULL; +} +#endif + static const struct hw_perf_counter_ops * sw_perf_counter_init(struct perf_counter *counter) { @@ -1772,6 +1814,7 @@ sw_perf_counter_init(struct perf_counter *counter) hw_ops = &perf_ops_cpu_migrations; break; default: + hw_ops = tp_perf_counter_init(counter); break; } -- cgit v1.2.3 From b8e83514b64577b48bfb794fe85fcde40a9343ca Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:18 +0100 Subject: perf_counter: revamp syscall input ABI Impact: modify ABI The hardware/software classification in hw_event->type became a little strained due to the addition of tracepoint tracing. Instead split up the field and provide a type field to explicitly specify the counter type, while using the event_id field to specify which event to use. Raw counters still work as before, only the raw config now goes into raw_event. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.836807573@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 83 +++++++++++++++++++++++++++++++-------------------- 1 file changed, 50 insertions(+), 33 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0bbe3e45ba0d..68a56a68bc74 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1395,12 +1395,6 @@ static void perf_swcounter_set_period(struct perf_counter *counter) atomic64_set(&hwc->count, -left); } -static void perf_swcounter_save_and_restart(struct perf_counter *counter) -{ - perf_swcounter_update(counter); - perf_swcounter_set_period(counter); -} - static void perf_swcounter_store_irq(struct perf_counter *counter, u64 data) { struct perf_data *irqdata = counter->irqdata; @@ -1421,7 +1415,7 @@ static void perf_swcounter_handle_group(struct perf_counter *sibling) list_for_each_entry(counter, &group_leader->sibling_list, list_entry) { counter->hw_ops->read(counter); - perf_swcounter_store_irq(sibling, counter->hw_event.type); + perf_swcounter_store_irq(sibling, counter->hw_event.event_config); perf_swcounter_store_irq(sibling, atomic64_read(&counter->count)); } } @@ -1477,21 +1471,25 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) static void perf_swcounter_overflow(struct perf_counter *counter, int nmi, struct pt_regs *regs) { - perf_swcounter_save_and_restart(counter); + perf_swcounter_update(counter); + perf_swcounter_set_period(counter); perf_swcounter_interrupt(counter, nmi, regs); } static int perf_swcounter_match(struct perf_counter *counter, - enum hw_event_types event, - struct pt_regs *regs) + enum perf_event_types type, + u32 event, struct pt_regs *regs) { if (counter->state != PERF_COUNTER_STATE_ACTIVE) return 0; - if (counter->hw_event.raw) + if (counter->hw_event.raw_type) + return 0; + + if (counter->hw_event.type != type) return 0; - if (counter->hw_event.type != event) + if (counter->hw_event.event_id != event) return 0; if (counter->hw_event.exclude_user && user_mode(regs)) @@ -1512,8 +1510,8 @@ static void perf_swcounter_add(struct perf_counter *counter, u64 nr, } static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, - enum hw_event_types event, u64 nr, - int nmi, struct pt_regs *regs) + enum perf_event_types type, u32 event, + u64 nr, int nmi, struct pt_regs *regs) { struct perf_counter *counter; @@ -1522,24 +1520,31 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_swcounter_match(counter, event, regs)) + if (perf_swcounter_match(counter, type, event, regs)) perf_swcounter_add(counter, nr, nmi, regs); } rcu_read_unlock(); } -void perf_swcounter_event(enum hw_event_types event, u64 nr, - int nmi, struct pt_regs *regs) +static void __perf_swcounter_event(enum perf_event_types type, u32 event, + u64 nr, int nmi, struct pt_regs *regs) { struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); - perf_swcounter_ctx_event(&cpuctx->ctx, event, nr, nmi, regs); - if (cpuctx->task_ctx) - perf_swcounter_ctx_event(cpuctx->task_ctx, event, nr, nmi, regs); + perf_swcounter_ctx_event(&cpuctx->ctx, type, event, nr, nmi, regs); + if (cpuctx->task_ctx) { + perf_swcounter_ctx_event(cpuctx->task_ctx, type, event, + nr, nmi, regs); + } put_cpu_var(perf_cpu_context); } +void perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs) +{ + __perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, regs); +} + static void perf_swcounter_read(struct perf_counter *counter) { perf_swcounter_update(counter); @@ -1733,8 +1738,12 @@ static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { #ifdef CONFIG_EVENT_PROFILE void perf_tpcounter_event(int event_id) { - perf_swcounter_event(PERF_TP_EVENTS_MIN + event_id, 1, 1, - task_pt_regs(current)); + struct pt_regs *regs = get_irq_regs(); + + if (!regs) + regs = task_pt_regs(current); + + __perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs); } extern int ftrace_profile_enable(int); @@ -1742,15 +1751,13 @@ extern void ftrace_profile_disable(int); static void tp_perf_counter_destroy(struct perf_counter *counter) { - int event_id = counter->hw_event.type - PERF_TP_EVENTS_MIN; - - ftrace_profile_disable(event_id); + ftrace_profile_disable(counter->hw_event.event_id); } static const struct hw_perf_counter_ops * tp_perf_counter_init(struct perf_counter *counter) { - int event_id = counter->hw_event.type - PERF_TP_EVENTS_MIN; + int event_id = counter->hw_event.event_id; int ret; ret = ftrace_profile_enable(event_id); @@ -1758,6 +1765,7 @@ tp_perf_counter_init(struct perf_counter *counter) return NULL; counter->destroy = tp_perf_counter_destroy; + counter->hw.irq_period = counter->hw_event.irq_period; return &perf_ops_generic; } @@ -1783,7 +1791,7 @@ sw_perf_counter_init(struct perf_counter *counter) * to be kernel events, and page faults are never hypervisor * events. */ - switch (counter->hw_event.type) { + switch (counter->hw_event.event_id) { case PERF_COUNT_CPU_CLOCK: hw_ops = &perf_ops_cpu_clock; @@ -1813,9 +1821,6 @@ sw_perf_counter_init(struct perf_counter *counter) if (!counter->hw_event.exclude_kernel) hw_ops = &perf_ops_cpu_migrations; break; - default: - hw_ops = tp_perf_counter_init(counter); - break; } if (hw_ops) @@ -1870,10 +1875,22 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->state = PERF_COUNTER_STATE_OFF; hw_ops = NULL; - if (!hw_event->raw && hw_event->type < 0) - hw_ops = sw_perf_counter_init(counter); - else + + if (hw_event->raw_type) + hw_ops = hw_perf_counter_init(counter); + else switch (hw_event->type) { + case PERF_TYPE_HARDWARE: hw_ops = hw_perf_counter_init(counter); + break; + + case PERF_TYPE_SOFTWARE: + hw_ops = sw_perf_counter_init(counter); + break; + + case PERF_TYPE_TRACEPOINT: + hw_ops = tp_perf_counter_init(counter); + break; + } if (!hw_ops) { kfree(counter); -- cgit v1.2.3 From 0322cd6ec504b0bf08ca7b2c3d7f43bda37d79c9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 19 Mar 2009 20:26:19 +0100 Subject: perf_counter: unify irq output code Impact: cleanup Having 3 slightly different copies of the same code around does nobody any good. First step in revamping the output format. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Orig-LKML-Reference: <20090319194233.929962222@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 106 ++++++++++++++++++++++++++------------------------ 1 file changed, 56 insertions(+), 50 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 68a56a68bc74..f054b8c9bf96 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1353,6 +1353,60 @@ static const struct file_operations perf_fops = { .compat_ioctl = perf_ioctl, }; +/* + * Output + */ + +static void perf_counter_store_irq(struct perf_counter *counter, u64 data) +{ + struct perf_data *irqdata = counter->irqdata; + + if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) { + irqdata->overrun++; + } else { + u64 *p = (u64 *) &irqdata->data[irqdata->len]; + + *p = data; + irqdata->len += sizeof(u64); + } +} + +static void perf_counter_handle_group(struct perf_counter *counter) +{ + struct perf_counter *leader, *sub; + + leader = counter->group_leader; + list_for_each_entry(sub, &leader->sibling_list, list_entry) { + if (sub != counter) + sub->hw_ops->read(sub); + perf_counter_store_irq(counter, sub->hw_event.event_config); + perf_counter_store_irq(counter, atomic64_read(&sub->count)); + } +} + +void perf_counter_output(struct perf_counter *counter, + int nmi, struct pt_regs *regs) +{ + switch (counter->hw_event.record_type) { + case PERF_RECORD_SIMPLE: + return; + + case PERF_RECORD_IRQ: + perf_counter_store_irq(counter, instruction_pointer(regs)); + break; + + case PERF_RECORD_GROUP: + perf_counter_handle_group(counter); + break; + } + + if (nmi) { + counter->wakeup_pending = 1; + set_perf_counter_pending(); + } else + wake_up(&counter->waitq); +} + /* * Generic software counter infrastructure */ @@ -1395,54 +1449,6 @@ static void perf_swcounter_set_period(struct perf_counter *counter) atomic64_set(&hwc->count, -left); } -static void perf_swcounter_store_irq(struct perf_counter *counter, u64 data) -{ - struct perf_data *irqdata = counter->irqdata; - - if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) { - irqdata->overrun++; - } else { - u64 *p = (u64 *) &irqdata->data[irqdata->len]; - - *p = data; - irqdata->len += sizeof(u64); - } -} - -static void perf_swcounter_handle_group(struct perf_counter *sibling) -{ - struct perf_counter *counter, *group_leader = sibling->group_leader; - - list_for_each_entry(counter, &group_leader->sibling_list, list_entry) { - counter->hw_ops->read(counter); - perf_swcounter_store_irq(sibling, counter->hw_event.event_config); - perf_swcounter_store_irq(sibling, atomic64_read(&counter->count)); - } -} - -static void perf_swcounter_interrupt(struct perf_counter *counter, - int nmi, struct pt_regs *regs) -{ - switch (counter->hw_event.record_type) { - case PERF_RECORD_SIMPLE: - break; - - case PERF_RECORD_IRQ: - perf_swcounter_store_irq(counter, instruction_pointer(regs)); - break; - - case PERF_RECORD_GROUP: - perf_swcounter_handle_group(counter); - break; - } - - if (nmi) { - counter->wakeup_pending = 1; - set_perf_counter_pending(); - } else - wake_up(&counter->waitq); -} - static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) { struct perf_counter *counter; @@ -1461,7 +1467,7 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) regs = task_pt_regs(current); if (regs) - perf_swcounter_interrupt(counter, 0, regs); + perf_counter_output(counter, 0, regs); hrtimer_forward_now(hrtimer, ns_to_ktime(counter->hw.irq_period)); @@ -1473,7 +1479,7 @@ static void perf_swcounter_overflow(struct perf_counter *counter, { perf_swcounter_update(counter); perf_swcounter_set_period(counter); - perf_swcounter_interrupt(counter, nmi, regs); + perf_counter_output(counter, nmi, regs); } static int perf_swcounter_match(struct perf_counter *counter, -- cgit v1.2.3 From f4a2deb4860497f4332cf6a1acddab3dd628ddf0 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Mar 2009 18:22:06 +0100 Subject: perf_counter: remove the event config bitfields Since the bitfields turned into a bit of a mess, remove them and rely on good old masks. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Orig-LKML-Reference: <20090323172417.059499915@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 22 +++++++++++++--------- 1 file changed, 13 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f054b8c9bf96..ca14fc41ccdf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1379,7 +1379,7 @@ static void perf_counter_handle_group(struct perf_counter *counter) list_for_each_entry(sub, &leader->sibling_list, list_entry) { if (sub != counter) sub->hw_ops->read(sub); - perf_counter_store_irq(counter, sub->hw_event.event_config); + perf_counter_store_irq(counter, sub->hw_event.config); perf_counter_store_irq(counter, atomic64_read(&sub->count)); } } @@ -1489,13 +1489,13 @@ static int perf_swcounter_match(struct perf_counter *counter, if (counter->state != PERF_COUNTER_STATE_ACTIVE) return 0; - if (counter->hw_event.raw_type) + if (perf_event_raw(&counter->hw_event)) return 0; - if (counter->hw_event.type != type) + if (perf_event_type(&counter->hw_event) != type) return 0; - if (counter->hw_event.event_id != event) + if (perf_event_id(&counter->hw_event) != event) return 0; if (counter->hw_event.exclude_user && user_mode(regs)) @@ -1757,13 +1757,13 @@ extern void ftrace_profile_disable(int); static void tp_perf_counter_destroy(struct perf_counter *counter) { - ftrace_profile_disable(counter->hw_event.event_id); + ftrace_profile_disable(perf_event_id(&counter->hw_event)); } static const struct hw_perf_counter_ops * tp_perf_counter_init(struct perf_counter *counter) { - int event_id = counter->hw_event.event_id; + int event_id = perf_event_id(&counter->hw_event); int ret; ret = ftrace_profile_enable(event_id); @@ -1797,7 +1797,7 @@ sw_perf_counter_init(struct perf_counter *counter) * to be kernel events, and page faults are never hypervisor * events. */ - switch (counter->hw_event.event_id) { + switch (perf_event_id(&counter->hw_event)) { case PERF_COUNT_CPU_CLOCK: hw_ops = &perf_ops_cpu_clock; @@ -1882,9 +1882,12 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, hw_ops = NULL; - if (hw_event->raw_type) + if (perf_event_raw(hw_event)) { hw_ops = hw_perf_counter_init(counter); - else switch (hw_event->type) { + goto done; + } + + switch (perf_event_type(hw_event)) { case PERF_TYPE_HARDWARE: hw_ops = hw_perf_counter_init(counter); break; @@ -1902,6 +1905,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, kfree(counter); return NULL; } +done: counter->hw_ops = hw_ops; return counter; -- cgit v1.2.3 From 96f6d4444302bb2ea2cf409529eef816462f6ce0 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Mar 2009 18:22:07 +0100 Subject: perf_counter: avoid recursion Tracepoint events like lock_acquire and software counters like pagefaults can recurse into the perf counter code again, avoid that. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Orig-LKML-Reference: <20090323172417.152096433@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ca14fc41ccdf..ce34bff07bda 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -23,6 +23,7 @@ #include #include #include +#include #include @@ -1532,10 +1533,31 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, rcu_read_unlock(); } +static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx) +{ + if (in_nmi()) + return &cpuctx->recursion[3]; + + if (in_irq()) + return &cpuctx->recursion[2]; + + if (in_softirq()) + return &cpuctx->recursion[1]; + + return &cpuctx->recursion[0]; +} + static void __perf_swcounter_event(enum perf_event_types type, u32 event, u64 nr, int nmi, struct pt_regs *regs) { struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); + int *recursion = perf_swcounter_recursion_context(cpuctx); + + if (*recursion) + goto out; + + (*recursion)++; + barrier(); perf_swcounter_ctx_event(&cpuctx->ctx, type, event, nr, nmi, regs); if (cpuctx->task_ctx) { @@ -1543,6 +1565,10 @@ static void __perf_swcounter_event(enum perf_event_types type, u32 event, nr, nmi, regs); } + barrier(); + (*recursion)--; + +out: put_cpu_var(perf_cpu_context); } -- cgit v1.2.3 From 37d81828385f8ff823caaaf1a83e72d065b6cfa1 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 23 Mar 2009 18:22:08 +0100 Subject: perf_counter: add an mmap method to allow userspace to read hardware counters Impact: new feature giving performance improvement This adds the ability for userspace to do an mmap on a hardware counter fd and get access to a read-only page that contains the information needed to translate a hardware counter value to the full 64-bit counter value that would be returned by a read on the fd. This is useful on architectures that allow user programs to read the hardware counters, such as PowerPC. The mmap will only succeed if the counter is a hardware counter monitoring the current process. On my quad 2.5GHz PowerPC 970MP machine, userspace can read a counter and translate it to the full 64-bit value in about 30ns using the mmapped page, compared to about 830ns for the read syscall on the counter, so this does give a significant performance improvement. Signed-off-by: Paul Mackerras Signed-off-by: Peter Zijlstra Orig-LKML-Reference: <20090323172417.297057964@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 76 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 76 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ce34bff07bda..d9cfd902140e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1177,6 +1177,7 @@ static int perf_release(struct inode *inode, struct file *file) mutex_unlock(&counter->mutex); mutex_unlock(&ctx->mutex); + free_page(counter->user_page); free_counter(counter); put_context(ctx); @@ -1346,12 +1347,87 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) return err; } +void perf_counter_update_userpage(struct perf_counter *counter) +{ + struct perf_counter_mmap_page *userpg; + + if (!counter->user_page) + return; + userpg = (struct perf_counter_mmap_page *) counter->user_page; + + ++userpg->lock; + smp_wmb(); + userpg->index = counter->hw.idx; + userpg->offset = atomic64_read(&counter->count); + if (counter->state == PERF_COUNTER_STATE_ACTIVE) + userpg->offset -= atomic64_read(&counter->hw.prev_count); + smp_wmb(); + ++userpg->lock; +} + +static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +{ + struct perf_counter *counter = vma->vm_file->private_data; + + if (!counter->user_page) + return VM_FAULT_SIGBUS; + + vmf->page = virt_to_page(counter->user_page); + get_page(vmf->page); + return 0; +} + +static struct vm_operations_struct perf_mmap_vmops = { + .fault = perf_mmap_fault, +}; + +static int perf_mmap(struct file *file, struct vm_area_struct *vma) +{ + struct perf_counter *counter = file->private_data; + unsigned long userpg; + + if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE)) + return -EINVAL; + if (vma->vm_end - vma->vm_start != PAGE_SIZE) + return -EINVAL; + + /* + * For now, restrict to the case of a hardware counter + * on the current task. + */ + if (is_software_counter(counter) || counter->task != current) + return -EINVAL; + + userpg = counter->user_page; + if (!userpg) { + userpg = get_zeroed_page(GFP_KERNEL); + mutex_lock(&counter->mutex); + if (counter->user_page) { + free_page(userpg); + userpg = counter->user_page; + } else { + counter->user_page = userpg; + } + mutex_unlock(&counter->mutex); + if (!userpg) + return -ENOMEM; + } + + perf_counter_update_userpage(counter); + + vma->vm_flags &= ~VM_MAYWRITE; + vma->vm_flags |= VM_RESERVED; + vma->vm_ops = &perf_mmap_vmops; + return 0; +} + static const struct file_operations perf_fops = { .release = perf_release, .read = perf_read, .poll = perf_poll, .unlocked_ioctl = perf_ioctl, .compat_ioctl = perf_ioctl, + .mmap = perf_mmap, }; /* -- cgit v1.2.3 From b09d2501ed3d294619cbfbcf828ad39324d0e548 Mon Sep 17 00:00:00 2001 From: "H. Peter Anvin" Date: Wed, 1 Apr 2009 17:21:56 -0700 Subject: mutex: drop "inline" from mutex_lock() inside kernel/mutex.c Impact: build fix mutex_lock() is was defined inline in kernel/mutex.c, but wasn't declared so not in . This didn't cause a problem until checkin 3a2d367d9aabac486ac4444c6c7ec7a1dab16267 added the atomic_dec_and_mutex_lock() inline in between declaration and definion. This broke building with CONFIG_ALLOW_WARNINGS=n, e.g. make allnoconfig. Either from the source code nor the allnoconfig binary output I cannot find any internal references to mutex_lock() in kernel/mutex.c, so presumably this "inline" is now-useless legacy. Cc: Eric Paris Cc: Peter Zijlstra Cc: Paul Mackerras Orig-LKML-Reference: Signed-off-by: H. Peter Anvin --- kernel/mutex.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/mutex.c b/kernel/mutex.c index 5d79781394a3..fd95eaa672e6 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -89,7 +89,7 @@ __mutex_lock_slowpath(atomic_t *lock_count); * * This function is similar to (but not equivalent to) down(). */ -void inline __sched mutex_lock(struct mutex *lock) +void __sched mutex_lock(struct mutex *lock) { might_sleep(); /* -- cgit v1.2.3 From 7b732a75047738e4f85438ed2f9cd34bf5f2a19a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 23 Mar 2009 18:22:10 +0100 Subject: perf_counter: new output ABI - part 1 Impact: Rework the perfcounter output ABI use sys_read() only for instant data and provide mmap() output for all async overflow data. The first mmap() determines the size of the output buffer. The mmap() size must be a PAGE_SIZE multiple of 1+pages, where pages must be a power of 2 or 0. Further mmap()s of the same fd must have the same size. Once all maps are gone, you can again mmap() with a new size. In case of 0 extra pages there is no data output and the first page only contains meta data. When there are data pages, a poll() event will be generated for each full page of data. Furthermore, the output is circular. This means that although 1 page is a valid configuration, its useless, since we'll start overwriting it the instant we report a full page. Future work will focus on the output format (currently maintained) where we'll likey want each entry denoted by a header which includes a type and length. Further future work will allow to splice() the fd, also containing the async overflow data -- splice() would be mutually exclusive with mmap() of the data. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Orig-LKML-Reference: <20090323172417.470536358@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 464 ++++++++++++++++++++++++++------------------------ 1 file changed, 245 insertions(+), 219 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d9cfd902140e..0dfe91094fd1 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -4,7 +4,8 @@ * Copyright(C) 2008 Thomas Gleixner * Copyright(C) 2008 Red Hat, Inc., Ingo Molnar * - * For licencing details see kernel-base/COPYING + * + * For licensing details see kernel-base/COPYING */ #include @@ -1022,66 +1023,6 @@ static u64 perf_counter_read(struct perf_counter *counter) return atomic64_read(&counter->count); } -/* - * Cross CPU call to switch performance data pointers - */ -static void __perf_switch_irq_data(void *info) -{ - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter *counter = info; - struct perf_counter_context *ctx = counter->ctx; - struct perf_data *oldirqdata = counter->irqdata; - - /* - * If this is a task context, we need to check whether it is - * the current task context of this cpu. If not it has been - * scheduled out before the smp call arrived. - */ - if (ctx->task) { - if (cpuctx->task_ctx != ctx) - return; - spin_lock(&ctx->lock); - } - - /* Change the pointer NMI safe */ - atomic_long_set((atomic_long_t *)&counter->irqdata, - (unsigned long) counter->usrdata); - counter->usrdata = oldirqdata; - - if (ctx->task) - spin_unlock(&ctx->lock); -} - -static struct perf_data *perf_switch_irq_data(struct perf_counter *counter) -{ - struct perf_counter_context *ctx = counter->ctx; - struct perf_data *oldirqdata = counter->irqdata; - struct task_struct *task = ctx->task; - - if (!task) { - smp_call_function_single(counter->cpu, - __perf_switch_irq_data, - counter, 1); - return counter->usrdata; - } - -retry: - spin_lock_irq(&ctx->lock); - if (counter->state != PERF_COUNTER_STATE_ACTIVE) { - counter->irqdata = counter->usrdata; - counter->usrdata = oldirqdata; - spin_unlock_irq(&ctx->lock); - return oldirqdata; - } - spin_unlock_irq(&ctx->lock); - task_oncpu_function_call(task, __perf_switch_irq_data, counter); - /* Might have failed, because task was scheduled out */ - if (counter->irqdata == oldirqdata) - goto retry; - - return counter->usrdata; -} - static void put_context(struct perf_counter_context *ctx) { if (ctx->task) @@ -1177,7 +1118,6 @@ static int perf_release(struct inode *inode, struct file *file) mutex_unlock(&counter->mutex); mutex_unlock(&ctx->mutex); - free_page(counter->user_page); free_counter(counter); put_context(ctx); @@ -1192,7 +1132,7 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) { u64 cntval; - if (count != sizeof(cntval)) + if (count < sizeof(cntval)) return -EINVAL; /* @@ -1210,122 +1150,21 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) return put_user(cntval, (u64 __user *) buf) ? -EFAULT : sizeof(cntval); } -static ssize_t -perf_copy_usrdata(struct perf_data *usrdata, char __user *buf, size_t count) -{ - if (!usrdata->len) - return 0; - - count = min(count, (size_t)usrdata->len); - if (copy_to_user(buf, usrdata->data + usrdata->rd_idx, count)) - return -EFAULT; - - /* Adjust the counters */ - usrdata->len -= count; - if (!usrdata->len) - usrdata->rd_idx = 0; - else - usrdata->rd_idx += count; - - return count; -} - -static ssize_t -perf_read_irq_data(struct perf_counter *counter, - char __user *buf, - size_t count, - int nonblocking) -{ - struct perf_data *irqdata, *usrdata; - DECLARE_WAITQUEUE(wait, current); - ssize_t res, res2; - - irqdata = counter->irqdata; - usrdata = counter->usrdata; - - if (usrdata->len + irqdata->len >= count) - goto read_pending; - - if (nonblocking) - return -EAGAIN; - - spin_lock_irq(&counter->waitq.lock); - __add_wait_queue(&counter->waitq, &wait); - for (;;) { - set_current_state(TASK_INTERRUPTIBLE); - if (usrdata->len + irqdata->len >= count) - break; - - if (signal_pending(current)) - break; - - if (counter->state == PERF_COUNTER_STATE_ERROR) - break; - - spin_unlock_irq(&counter->waitq.lock); - schedule(); - spin_lock_irq(&counter->waitq.lock); - } - __remove_wait_queue(&counter->waitq, &wait); - __set_current_state(TASK_RUNNING); - spin_unlock_irq(&counter->waitq.lock); - - if (usrdata->len + irqdata->len < count && - counter->state != PERF_COUNTER_STATE_ERROR) - return -ERESTARTSYS; -read_pending: - mutex_lock(&counter->mutex); - - /* Drain pending data first: */ - res = perf_copy_usrdata(usrdata, buf, count); - if (res < 0 || res == count) - goto out; - - /* Switch irq buffer: */ - usrdata = perf_switch_irq_data(counter); - res2 = perf_copy_usrdata(usrdata, buf + res, count - res); - if (res2 < 0) { - if (!res) - res = -EFAULT; - } else { - res += res2; - } -out: - mutex_unlock(&counter->mutex); - - return res; -} - static ssize_t perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) { struct perf_counter *counter = file->private_data; - switch (counter->hw_event.record_type) { - case PERF_RECORD_SIMPLE: - return perf_read_hw(counter, buf, count); - - case PERF_RECORD_IRQ: - case PERF_RECORD_GROUP: - return perf_read_irq_data(counter, buf, count, - file->f_flags & O_NONBLOCK); - } - return -EINVAL; + return perf_read_hw(counter, buf, count); } static unsigned int perf_poll(struct file *file, poll_table *wait) { struct perf_counter *counter = file->private_data; - unsigned int events = 0; - unsigned long flags; + unsigned int events = POLLIN; poll_wait(file, &counter->waitq, wait); - spin_lock_irqsave(&counter->waitq.lock, flags); - if (counter->usrdata->len || counter->irqdata->len) - events |= POLLIN; - spin_unlock_irqrestore(&counter->waitq.lock, flags); - return events; } @@ -1347,78 +1186,207 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) return err; } -void perf_counter_update_userpage(struct perf_counter *counter) +static void __perf_counter_update_userpage(struct perf_counter *counter, + struct perf_mmap_data *data) { - struct perf_counter_mmap_page *userpg; - - if (!counter->user_page) - return; - userpg = (struct perf_counter_mmap_page *) counter->user_page; + struct perf_counter_mmap_page *userpg = data->user_page; + /* + * Disable preemption so as to not let the corresponding user-space + * spin too long if we get preempted. + */ + preempt_disable(); ++userpg->lock; smp_wmb(); userpg->index = counter->hw.idx; userpg->offset = atomic64_read(&counter->count); if (counter->state == PERF_COUNTER_STATE_ACTIVE) userpg->offset -= atomic64_read(&counter->hw.prev_count); + + userpg->data_head = atomic_read(&data->head); smp_wmb(); ++userpg->lock; + preempt_enable(); +} + +void perf_counter_update_userpage(struct perf_counter *counter) +{ + struct perf_mmap_data *data; + + rcu_read_lock(); + data = rcu_dereference(counter->data); + if (data) + __perf_counter_update_userpage(counter, data); + rcu_read_unlock(); } static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) { struct perf_counter *counter = vma->vm_file->private_data; + struct perf_mmap_data *data; + int ret = VM_FAULT_SIGBUS; - if (!counter->user_page) - return VM_FAULT_SIGBUS; + rcu_read_lock(); + data = rcu_dereference(counter->data); + if (!data) + goto unlock; + + if (vmf->pgoff == 0) { + vmf->page = virt_to_page(data->user_page); + } else { + int nr = vmf->pgoff - 1; - vmf->page = virt_to_page(counter->user_page); + if ((unsigned)nr > data->nr_pages) + goto unlock; + + vmf->page = virt_to_page(data->data_pages[nr]); + } get_page(vmf->page); + ret = 0; +unlock: + rcu_read_unlock(); + + return ret; +} + +static int perf_mmap_data_alloc(struct perf_counter *counter, int nr_pages) +{ + struct perf_mmap_data *data; + unsigned long size; + int i; + + WARN_ON(atomic_read(&counter->mmap_count)); + + size = sizeof(struct perf_mmap_data); + size += nr_pages * sizeof(void *); + + data = kzalloc(size, GFP_KERNEL); + if (!data) + goto fail; + + data->user_page = (void *)get_zeroed_page(GFP_KERNEL); + if (!data->user_page) + goto fail_user_page; + + for (i = 0; i < nr_pages; i++) { + data->data_pages[i] = (void *)get_zeroed_page(GFP_KERNEL); + if (!data->data_pages[i]) + goto fail_data_pages; + } + + data->nr_pages = nr_pages; + + rcu_assign_pointer(counter->data, data); + return 0; + +fail_data_pages: + for (i--; i >= 0; i--) + free_page((unsigned long)data->data_pages[i]); + + free_page((unsigned long)data->user_page); + +fail_user_page: + kfree(data); + +fail: + return -ENOMEM; +} + +static void __perf_mmap_data_free(struct rcu_head *rcu_head) +{ + struct perf_mmap_data *data = container_of(rcu_head, + struct perf_mmap_data, rcu_head); + int i; + + free_page((unsigned long)data->user_page); + for (i = 0; i < data->nr_pages; i++) + free_page((unsigned long)data->data_pages[i]); + kfree(data); +} + +static void perf_mmap_data_free(struct perf_counter *counter) +{ + struct perf_mmap_data *data = counter->data; + + WARN_ON(atomic_read(&counter->mmap_count)); + + rcu_assign_pointer(counter->data, NULL); + call_rcu(&data->rcu_head, __perf_mmap_data_free); +} + +static void perf_mmap_open(struct vm_area_struct *vma) +{ + struct perf_counter *counter = vma->vm_file->private_data; + + atomic_inc(&counter->mmap_count); +} + +static void perf_mmap_close(struct vm_area_struct *vma) +{ + struct perf_counter *counter = vma->vm_file->private_data; + + if (atomic_dec_and_mutex_lock(&counter->mmap_count, + &counter->mmap_mutex)) { + perf_mmap_data_free(counter); + mutex_unlock(&counter->mmap_mutex); + } } static struct vm_operations_struct perf_mmap_vmops = { + .open = perf_mmap_open, + .close = perf_mmap_close, .fault = perf_mmap_fault, }; static int perf_mmap(struct file *file, struct vm_area_struct *vma) { struct perf_counter *counter = file->private_data; - unsigned long userpg; + unsigned long vma_size; + unsigned long nr_pages; + unsigned long locked, lock_limit; + int ret = 0; if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE)) return -EINVAL; - if (vma->vm_end - vma->vm_start != PAGE_SIZE) + + vma_size = vma->vm_end - vma->vm_start; + nr_pages = (vma_size / PAGE_SIZE) - 1; + + if (nr_pages == 0 || !is_power_of_2(nr_pages)) return -EINVAL; - /* - * For now, restrict to the case of a hardware counter - * on the current task. - */ - if (is_software_counter(counter) || counter->task != current) + if (vma_size != PAGE_SIZE * (1 + nr_pages)) return -EINVAL; - userpg = counter->user_page; - if (!userpg) { - userpg = get_zeroed_page(GFP_KERNEL); - mutex_lock(&counter->mutex); - if (counter->user_page) { - free_page(userpg); - userpg = counter->user_page; - } else { - counter->user_page = userpg; - } - mutex_unlock(&counter->mutex); - if (!userpg) - return -ENOMEM; - } + if (vma->vm_pgoff != 0) + return -EINVAL; + + locked = vma_size >> PAGE_SHIFT; + locked += vma->vm_mm->locked_vm; - perf_counter_update_userpage(counter); + lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; + lock_limit >>= PAGE_SHIFT; + + if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) + return -EPERM; + + mutex_lock(&counter->mmap_mutex); + if (atomic_inc_not_zero(&counter->mmap_count)) + goto out; + + WARN_ON(counter->data); + ret = perf_mmap_data_alloc(counter, nr_pages); + if (!ret) + atomic_set(&counter->mmap_count, 1); +out: + mutex_unlock(&counter->mmap_mutex); vma->vm_flags &= ~VM_MAYWRITE; vma->vm_flags |= VM_RESERVED; vma->vm_ops = &perf_mmap_vmops; - return 0; + + return ret; } static const struct file_operations perf_fops = { @@ -1434,30 +1402,94 @@ static const struct file_operations perf_fops = { * Output */ -static void perf_counter_store_irq(struct perf_counter *counter, u64 data) +static int perf_output_write(struct perf_counter *counter, int nmi, + void *buf, ssize_t size) { - struct perf_data *irqdata = counter->irqdata; + struct perf_mmap_data *data; + unsigned int offset, head, nr; + unsigned int len; + int ret, wakeup; - if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) { - irqdata->overrun++; - } else { - u64 *p = (u64 *) &irqdata->data[irqdata->len]; + rcu_read_lock(); + ret = -ENOSPC; + data = rcu_dereference(counter->data); + if (!data) + goto out; + + if (!data->nr_pages) + goto out; + + ret = -EINVAL; + if (size > PAGE_SIZE) + goto out; + + do { + offset = head = atomic_read(&data->head); + head += sizeof(u64); + } while (atomic_cmpxchg(&data->head, offset, head) != offset); + + wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); - *p = data; - irqdata->len += sizeof(u64); + nr = (offset >> PAGE_SHIFT) & (data->nr_pages - 1); + offset &= PAGE_SIZE - 1; + + len = min_t(unsigned int, PAGE_SIZE - offset, size); + memcpy(data->data_pages[nr] + offset, buf, len); + size -= len; + + if (size) { + nr = (nr + 1) & (data->nr_pages - 1); + memcpy(data->data_pages[nr], buf + len, size); + } + + /* + * generate a poll() wakeup for every page boundary crossed + */ + if (wakeup) { + __perf_counter_update_userpage(counter, data); + if (nmi) { + counter->wakeup_pending = 1; + set_perf_counter_pending(); + } else + wake_up(&counter->waitq); } + ret = 0; +out: + rcu_read_unlock(); + + return ret; } -static void perf_counter_handle_group(struct perf_counter *counter) +static void perf_output_simple(struct perf_counter *counter, + int nmi, struct pt_regs *regs) +{ + u64 entry; + + entry = instruction_pointer(regs); + + perf_output_write(counter, nmi, &entry, sizeof(entry)); +} + +struct group_entry { + u64 event; + u64 counter; +}; + +static void perf_output_group(struct perf_counter *counter, int nmi) { struct perf_counter *leader, *sub; leader = counter->group_leader; list_for_each_entry(sub, &leader->sibling_list, list_entry) { + struct group_entry entry; + if (sub != counter) sub->hw_ops->read(sub); - perf_counter_store_irq(counter, sub->hw_event.config); - perf_counter_store_irq(counter, atomic64_read(&sub->count)); + + entry.event = sub->hw_event.config; + entry.counter = atomic64_read(&sub->count); + + perf_output_write(counter, nmi, &entry, sizeof(entry)); } } @@ -1469,19 +1501,13 @@ void perf_counter_output(struct perf_counter *counter, return; case PERF_RECORD_IRQ: - perf_counter_store_irq(counter, instruction_pointer(regs)); + perf_output_simple(counter, nmi, regs); break; case PERF_RECORD_GROUP: - perf_counter_handle_group(counter); + perf_output_group(counter, nmi); break; } - - if (nmi) { - counter->wakeup_pending = 1; - set_perf_counter_pending(); - } else - wake_up(&counter->waitq); } /* @@ -1967,10 +1993,10 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, INIT_LIST_HEAD(&counter->sibling_list); init_waitqueue_head(&counter->waitq); + mutex_init(&counter->mmap_mutex); + INIT_LIST_HEAD(&counter->child_list); - counter->irqdata = &counter->data[0]; - counter->usrdata = &counter->data[1]; counter->cpu = cpu; counter->hw_event = *hw_event; counter->wakeup_pending = 0; -- cgit v1.2.3 From c7138f37f905bb7987b1f9f5a8ee73667db39f25 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 24 Mar 2009 13:18:16 +0100 Subject: perf_counter: fix perf_poll() Impact: fix kerneltop 100% CPU usage Only return a poll event when there's actually been one, poll_wait() doesn't actually wait for the waitq you pass it, it only enqueues you on it. Only once all FDs have been iterated and none of thm returned a poll-event will it schedule(). Also make it return POLL_HUP when there's not mmap() area to read from. Further, fix a silly bug in the write code. Reported-by: Mike Galbraith Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Arjan van de Ven Orig-LKML-Reference: <1237897096.24918.181.camel@twins> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0dfe91094fd1..affe227d56a0 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1161,7 +1161,16 @@ perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) static unsigned int perf_poll(struct file *file, poll_table *wait) { struct perf_counter *counter = file->private_data; - unsigned int events = POLLIN; + struct perf_mmap_data *data; + unsigned int events; + + rcu_read_lock(); + data = rcu_dereference(counter->data); + if (data) + events = atomic_xchg(&data->wakeup, 0); + else + events = POLL_HUP; + rcu_read_unlock(); poll_wait(file, &counter->waitq, wait); @@ -1425,7 +1434,7 @@ static int perf_output_write(struct perf_counter *counter, int nmi, do { offset = head = atomic_read(&data->head); - head += sizeof(u64); + head += size; } while (atomic_cmpxchg(&data->head, offset, head) != offset); wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); @@ -1446,6 +1455,7 @@ static int perf_output_write(struct perf_counter *counter, int nmi, * generate a poll() wakeup for every page boundary crossed */ if (wakeup) { + atomic_xchg(&data->wakeup, POLL_IN); __perf_counter_update_userpage(counter, data); if (nmi) { counter->wakeup_pending = 1; -- cgit v1.2.3 From b9cacc7bf193df16532bfa7d7ca77fe50fc3c2e6 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 25 Mar 2009 12:30:22 +0100 Subject: perf_counter: more elaborate write API Provide a begin, copy, end interface to the output buffer. begin() reserves the space, copy() copies the data over, considering page boundaries, end() finalizes the event and does the wakeup. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Mike Galbraith Cc: Arjan van de Ven Cc: Wu Fengguang Orig-LKML-Reference: <20090325113316.740550870@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 109 ++++++++++++++++++++++++++++++++++---------------- 1 file changed, 75 insertions(+), 34 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index affe227d56a0..0422fd9bf627 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -9,6 +9,7 @@ */ #include +#include #include #include #include @@ -16,15 +17,14 @@ #include #include #include +#include +#include +#include #include #include #include #include #include -#include -#include -#include -#include #include @@ -1411,16 +1411,20 @@ static const struct file_operations perf_fops = { * Output */ -static int perf_output_write(struct perf_counter *counter, int nmi, - void *buf, ssize_t size) +struct perf_output_handle { + struct perf_counter *counter; + struct perf_mmap_data *data; + unsigned int offset; + int wakeup; +}; + +static int perf_output_begin(struct perf_output_handle *handle, + struct perf_counter *counter, unsigned int size) { struct perf_mmap_data *data; - unsigned int offset, head, nr; - unsigned int len; - int ret, wakeup; + unsigned int offset, head; rcu_read_lock(); - ret = -ENOSPC; data = rcu_dereference(counter->data); if (!data) goto out; @@ -1428,45 +1432,82 @@ static int perf_output_write(struct perf_counter *counter, int nmi, if (!data->nr_pages) goto out; - ret = -EINVAL; - if (size > PAGE_SIZE) - goto out; - do { offset = head = atomic_read(&data->head); head += size; } while (atomic_cmpxchg(&data->head, offset, head) != offset); - wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); + handle->counter = counter; + handle->data = data; + handle->offset = offset; + handle->wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); - nr = (offset >> PAGE_SHIFT) & (data->nr_pages - 1); - offset &= PAGE_SIZE - 1; + return 0; - len = min_t(unsigned int, PAGE_SIZE - offset, size); - memcpy(data->data_pages[nr] + offset, buf, len); - size -= len; +out: + rcu_read_unlock(); - if (size) { - nr = (nr + 1) & (data->nr_pages - 1); - memcpy(data->data_pages[nr], buf + len, size); - } + return -ENOSPC; +} - /* - * generate a poll() wakeup for every page boundary crossed - */ - if (wakeup) { - atomic_xchg(&data->wakeup, POLL_IN); - __perf_counter_update_userpage(counter, data); +static void perf_output_copy(struct perf_output_handle *handle, + void *buf, unsigned int len) +{ + unsigned int pages_mask; + unsigned int offset; + unsigned int size; + void **pages; + + offset = handle->offset; + pages_mask = handle->data->nr_pages - 1; + pages = handle->data->data_pages; + + do { + unsigned int page_offset; + int nr; + + nr = (offset >> PAGE_SHIFT) & pages_mask; + page_offset = offset & (PAGE_SIZE - 1); + size = min_t(unsigned int, PAGE_SIZE - page_offset, len); + + memcpy(pages[nr] + page_offset, buf, size); + + len -= size; + buf += size; + offset += size; + } while (len); + + handle->offset = offset; +} + +static void perf_output_end(struct perf_output_handle *handle, int nmi) +{ + if (handle->wakeup) { + (void)atomic_xchg(&handle->data->wakeup, POLL_IN); + __perf_counter_update_userpage(handle->counter, handle->data); if (nmi) { - counter->wakeup_pending = 1; + handle->counter->wakeup_pending = 1; set_perf_counter_pending(); } else - wake_up(&counter->waitq); + wake_up(&handle->counter->waitq); } - ret = 0; -out: rcu_read_unlock(); +} + +static int perf_output_write(struct perf_counter *counter, int nmi, + void *buf, ssize_t size) +{ + struct perf_output_handle handle; + int ret; + ret = perf_output_begin(&handle, counter, size); + if (ret) + goto out; + + perf_output_copy(&handle, buf, size); + perf_output_end(&handle, nmi); + +out: return ret; } -- cgit v1.2.3 From 5c1481943250ab65fa5130e05ec479c93216e9f7 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 25 Mar 2009 12:30:23 +0100 Subject: perf_counter: output objects Provide a {type,size} header for each output entry. This should provide extensible output, and the ability to mix multiple streams. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Mike Galbraith Cc: Arjan van de Ven Cc: Wu Fengguang Orig-LKML-Reference: <20090325113316.831607932@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 53 +++++++++++++++++++++++++++++++++++++++------------ 1 file changed, 41 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0422fd9bf627..d76e3112d386 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -75,8 +75,10 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) */ if (counter->group_leader == counter) list_add_tail(&counter->list_entry, &ctx->counter_list); - else + else { list_add_tail(&counter->list_entry, &group_leader->sibling_list); + group_leader->nr_siblings++; + } list_add_rcu(&counter->event_entry, &ctx->event_list); } @@ -89,6 +91,9 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_del_init(&counter->list_entry); list_del_rcu(&counter->event_entry); + if (counter->group_leader != counter) + counter->group_leader->nr_siblings--; + /* * If this was a group counter with sibling counters then * upgrade the siblings to singleton counters by adding them @@ -381,9 +386,11 @@ static int is_software_only_group(struct perf_counter *leader) if (!is_software_counter(leader)) return 0; + list_for_each_entry(counter, &leader->sibling_list, list_entry) if (!is_software_counter(counter)) return 0; + return 1; } @@ -1480,6 +1487,9 @@ static void perf_output_copy(struct perf_output_handle *handle, handle->offset = offset; } +#define perf_output_put(handle, x) \ + perf_output_copy((handle), &(x), sizeof(x)) + static void perf_output_end(struct perf_output_handle *handle, int nmi) { if (handle->wakeup) { @@ -1514,34 +1524,53 @@ out: static void perf_output_simple(struct perf_counter *counter, int nmi, struct pt_regs *regs) { - u64 entry; + struct { + struct perf_event_header header; + u64 ip; + } event; - entry = instruction_pointer(regs); + event.header.type = PERF_EVENT_IP; + event.header.size = sizeof(event); + event.ip = instruction_pointer(regs); - perf_output_write(counter, nmi, &entry, sizeof(entry)); + perf_output_write(counter, nmi, &event, sizeof(event)); } -struct group_entry { - u64 event; - u64 counter; -}; - static void perf_output_group(struct perf_counter *counter, int nmi) { + struct perf_output_handle handle; + struct perf_event_header header; struct perf_counter *leader, *sub; + unsigned int size; + struct { + u64 event; + u64 counter; + } entry; + int ret; + + size = sizeof(header) + counter->nr_siblings * sizeof(entry); + + ret = perf_output_begin(&handle, counter, size); + if (ret) + return; + + header.type = PERF_EVENT_GROUP; + header.size = size; + + perf_output_put(&handle, header); leader = counter->group_leader; list_for_each_entry(sub, &leader->sibling_list, list_entry) { - struct group_entry entry; - if (sub != counter) sub->hw_ops->read(sub); entry.event = sub->hw_event.config; entry.counter = atomic64_read(&sub->count); - perf_output_write(counter, nmi, &entry, sizeof(entry)); + perf_output_put(&handle, entry); } + + perf_output_end(&handle, nmi); } void perf_counter_output(struct perf_counter *counter, -- cgit v1.2.3 From 63e35b25d6b5c3136d22ef249dbbf96716aa08bf Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 25 Mar 2009 12:30:24 +0100 Subject: perf_counter: sanity check on the output API Ensure we never write more than we said we would. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Mike Galbraith Cc: Arjan van de Ven Cc: Wu Fengguang Orig-LKML-Reference: <20090325113316.921433024@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d76e3112d386..7669afe82cc7 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1422,6 +1422,7 @@ struct perf_output_handle { struct perf_counter *counter; struct perf_mmap_data *data; unsigned int offset; + unsigned int head; int wakeup; }; @@ -1447,6 +1448,7 @@ static int perf_output_begin(struct perf_output_handle *handle, handle->counter = counter; handle->data = data; handle->offset = offset; + handle->head = head; handle->wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); return 0; @@ -1485,6 +1487,8 @@ static void perf_output_copy(struct perf_output_handle *handle, } while (len); handle->offset = offset; + + WARN_ON_ONCE(handle->offset > handle->head); } #define perf_output_put(handle, x) \ -- cgit v1.2.3 From ea5d20cf99db5d26d43b6d322d3ace17e08a6552 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 25 Mar 2009 12:30:25 +0100 Subject: perf_counter: optionally provide the pid/tid of the sampled task Allow cpu wide counters to profile userspace by providing what process the sample belongs to. This raises the first issue with the output type, lots of these options: group, tid, callchain, etc.. are non-exclusive and could be combined, suggesting a bitfield. However, things like the mmap() data stream doesn't fit in that. How to split the type field... Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Mike Galbraith Cc: Arjan van de Ven Cc: Wu Fengguang Orig-LKML-Reference: <20090325113317.013775235@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7669afe82cc7..f3e1b27bc1b8 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1528,16 +1528,30 @@ out: static void perf_output_simple(struct perf_counter *counter, int nmi, struct pt_regs *regs) { + unsigned int size; struct { struct perf_event_header header; u64 ip; + u32 pid, tid; } event; event.header.type = PERF_EVENT_IP; - event.header.size = sizeof(event); event.ip = instruction_pointer(regs); - perf_output_write(counter, nmi, &event, sizeof(event)); + size = sizeof(event); + + if (counter->hw_event.include_tid) { + /* namespace issues */ + event.pid = current->group_leader->pid; + event.tid = current->pid; + + event.header.type |= __PERF_EVENT_TID; + } else + size -= sizeof(u64); + + event.header.size = size; + + perf_output_write(counter, nmi, &event, size); } static void perf_output_group(struct perf_counter *counter, int nmi) -- cgit v1.2.3 From 7730d8655880f41f2ea519aca2ca6a1413dfd2c9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 25 Mar 2009 12:48:31 +0100 Subject: perf_counter: allow and require one-page mmap on counting counters A brainfart stopped single page mmap()s working. The rest of the code should be perfectly fine with not having any data pages. Reported-by: Paul Mackerras Signed-off-by: Peter Zijlstra Orig-LKML-Reference: <1237981712.7972.812.camel@twins> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f3e1b27bc1b8..95e02575546b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1369,7 +1369,11 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) vma_size = vma->vm_end - vma->vm_start; nr_pages = (vma_size / PAGE_SIZE) - 1; - if (nr_pages == 0 || !is_power_of_2(nr_pages)) + /* + * If we have data pages ensure they're a power-of-two number, so we + * can do bitmasks instead of modulo. + */ + if (nr_pages != 0 && !is_power_of_2(nr_pages)) return -EINVAL; if (vma_size != PAGE_SIZE * (1 + nr_pages)) -- cgit v1.2.3 From 53cfbf593758916aac41db728f029986a62f1254 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Wed, 25 Mar 2009 22:46:58 +1100 Subject: perf_counter: record time running and time enabled for each counter Impact: new functionality Currently, if there are more counters enabled than can fit on the CPU, the kernel will multiplex the counters on to the hardware using round-robin scheduling. That isn't too bad for sampling counters, but for counting counters it means that the value read from a counter represents some unknown fraction of the true count of events that occurred while the counter was enabled. This remedies the situation by keeping track of how long each counter is enabled for, and how long it is actually on the cpu and counting events. These times are recorded in nanoseconds using the task clock for per-task counters and the cpu clock for per-cpu counters. These values can be supplied to userspace on a read from the counter. Userspace requests that they be supplied after the counter value by setting the PERF_FORMAT_TOTAL_TIME_ENABLED and/or PERF_FORMAT_TOTAL_TIME_RUNNING bits in the hw_event.read_format field when creating the counter. (There is no way to change the read format after the counter is created, though it would be possible to add some way to do that.) Using this information it is possible for userspace to scale the count it reads from the counter to get an estimate of the true count: true_count_estimate = count * total_time_enabled / total_time_running This also lets userspace detect the situation where the counter never got to go on the cpu: total_time_running == 0. This functionality has been requested by the PAPI developers, and will be generally needed for interpreting the count values from counting counters correctly. In the implementation, this keeps 5 time values (in nanoseconds) for each counter: total_time_enabled and total_time_running are used when the counter is in state OFF or ERROR and for reporting back to userspace. When the counter is in state INACTIVE or ACTIVE, it is the tstamp_enabled, tstamp_running and tstamp_stopped values that are relevant, and total_time_enabled and total_time_running are determined from them. (tstamp_stopped is only used in INACTIVE state.) The reason for doing it like this is that it means that only counters being enabled or disabled at sched-in and sched-out time need to be updated. There are no new loops that iterate over all counters to update total_time_enabled or total_time_running. This also keeps separate child_total_time_running and child_total_time_enabled fields that get added in when reporting the totals to userspace. They are separate fields so that they can be atomic. We don't want to use atomics for total_time_running, total_time_enabled etc., because then we would have to use atomic sequences to update them, which are slower than regular arithmetic and memory accesses. It is possible to measure total_time_running by adding a task_clock counter to each group of counters, and total_time_enabled can be measured approximately with a top-level task_clock counter (though inaccuracies will creep in if you need to disable and enable groups since it is not possible in general to disable/enable the top-level task_clock counter simultaneously with another group). However, that adds extra overhead - I measured around 15% increase in the context switch latency reported by lat_ctx (from lmbench) when a task_clock counter was added to each of 2 groups, and around 25% increase when a task_clock counter was added to each of 4 groups. (In both cases a top-level task-clock counter was also added.) In contrast, the code added in this commit gives better information with no overhead that I could measure (in fact in some cases I measured lower times with this code, but the differences were all less than one standard deviation). [ v2: address review comments by Andrew Morton. ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Andrew Morton Orig-LKML-Reference: <18890.6578.728637.139402@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 157 +++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 136 insertions(+), 21 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 95e02575546b..3b862a7988cd 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -116,6 +116,7 @@ counter_sched_out(struct perf_counter *counter, return; counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->tstamp_stopped = ctx->time_now; counter->hw_ops->disable(counter); counter->oncpu = -1; @@ -251,6 +252,60 @@ retry: spin_unlock_irq(&ctx->lock); } +/* + * Get the current time for this context. + * If this is a task context, we use the task's task clock, + * or for a per-cpu context, we use the cpu clock. + */ +static u64 get_context_time(struct perf_counter_context *ctx, int update) +{ + struct task_struct *curr = ctx->task; + + if (!curr) + return cpu_clock(smp_processor_id()); + + return __task_delta_exec(curr, update) + curr->se.sum_exec_runtime; +} + +/* + * Update the record of the current time in a context. + */ +static void update_context_time(struct perf_counter_context *ctx, int update) +{ + ctx->time_now = get_context_time(ctx, update) - ctx->time_lost; +} + +/* + * Update the total_time_enabled and total_time_running fields for a counter. + */ +static void update_counter_times(struct perf_counter *counter) +{ + struct perf_counter_context *ctx = counter->ctx; + u64 run_end; + + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { + counter->total_time_enabled = ctx->time_now - + counter->tstamp_enabled; + if (counter->state == PERF_COUNTER_STATE_INACTIVE) + run_end = counter->tstamp_stopped; + else + run_end = ctx->time_now; + counter->total_time_running = run_end - counter->tstamp_running; + } +} + +/* + * Update total_time_enabled and total_time_running for all counters in a group. + */ +static void update_group_times(struct perf_counter *leader) +{ + struct perf_counter *counter; + + update_counter_times(leader); + list_for_each_entry(counter, &leader->sibling_list, list_entry) + update_counter_times(counter); +} + /* * Cross CPU call to disable a performance counter */ @@ -276,6 +331,8 @@ static void __perf_counter_disable(void *info) * If it is in error state, leave it in error state. */ if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { + update_context_time(ctx, 1); + update_counter_times(counter); if (counter == counter->group_leader) group_sched_out(counter, cpuctx, ctx); else @@ -320,8 +377,10 @@ static void perf_counter_disable(struct perf_counter *counter) * Since we have the lock this context can't be scheduled * in, so we can change the state safely. */ - if (counter->state == PERF_COUNTER_STATE_INACTIVE) + if (counter->state == PERF_COUNTER_STATE_INACTIVE) { + update_counter_times(counter); counter->state = PERF_COUNTER_STATE_OFF; + } spin_unlock_irq(&ctx->lock); } @@ -366,6 +425,8 @@ counter_sched_in(struct perf_counter *counter, return -EAGAIN; } + counter->tstamp_running += ctx->time_now - counter->tstamp_stopped; + if (!is_software_counter(counter)) cpuctx->active_oncpu++; ctx->nr_active++; @@ -425,6 +486,17 @@ static int group_can_go_on(struct perf_counter *counter, return can_add_hw; } +static void add_counter_to_ctx(struct perf_counter *counter, + struct perf_counter_context *ctx) +{ + list_add_counter(counter, ctx); + ctx->nr_counters++; + counter->prev_state = PERF_COUNTER_STATE_OFF; + counter->tstamp_enabled = ctx->time_now; + counter->tstamp_running = ctx->time_now; + counter->tstamp_stopped = ctx->time_now; +} + /* * Cross CPU call to install and enable a performance counter */ @@ -449,6 +521,7 @@ static void __perf_install_in_context(void *info) curr_rq_lock_irq_save(&flags); spin_lock(&ctx->lock); + update_context_time(ctx, 1); /* * Protect the list operation against NMI by disabling the @@ -456,9 +529,7 @@ static void __perf_install_in_context(void *info) */ perf_flags = hw_perf_save_disable(); - list_add_counter(counter, ctx); - ctx->nr_counters++; - counter->prev_state = PERF_COUNTER_STATE_OFF; + add_counter_to_ctx(counter, ctx); /* * Don't put the counter on if it is disabled or if @@ -486,8 +557,10 @@ static void __perf_install_in_context(void *info) */ if (leader != counter) group_sched_out(leader, cpuctx, ctx); - if (leader->hw_event.pinned) + if (leader->hw_event.pinned) { + update_group_times(leader); leader->state = PERF_COUNTER_STATE_ERROR; + } } if (!err && !ctx->task && cpuctx->max_pertask) @@ -548,10 +621,8 @@ retry: * can add the counter safely, if it the call above did not * succeed. */ - if (list_empty(&counter->list_entry)) { - list_add_counter(counter, ctx); - ctx->nr_counters++; - } + if (list_empty(&counter->list_entry)) + add_counter_to_ctx(counter, ctx); spin_unlock_irq(&ctx->lock); } @@ -576,11 +647,13 @@ static void __perf_counter_enable(void *info) curr_rq_lock_irq_save(&flags); spin_lock(&ctx->lock); + update_context_time(ctx, 1); counter->prev_state = counter->state; if (counter->state >= PERF_COUNTER_STATE_INACTIVE) goto unlock; counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->tstamp_enabled = ctx->time_now - counter->total_time_enabled; /* * If the counter is in a group and isn't the group leader, @@ -602,8 +675,10 @@ static void __perf_counter_enable(void *info) */ if (leader != counter) group_sched_out(leader, cpuctx, ctx); - if (leader->hw_event.pinned) + if (leader->hw_event.pinned) { + update_group_times(leader); leader->state = PERF_COUNTER_STATE_ERROR; + } } unlock: @@ -659,8 +734,11 @@ static void perf_counter_enable(struct perf_counter *counter) * Since we have the lock this context can't be scheduled * in, so we can change the state safely. */ - if (counter->state == PERF_COUNTER_STATE_OFF) + if (counter->state == PERF_COUNTER_STATE_OFF) { counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->tstamp_enabled = ctx->time_now - + counter->total_time_enabled; + } out: spin_unlock_irq(&ctx->lock); } @@ -693,6 +771,7 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, ctx->is_active = 0; if (likely(!ctx->nr_counters)) goto out; + update_context_time(ctx, 0); flags = hw_perf_save_disable(); if (ctx->nr_active) { @@ -797,6 +876,13 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, if (likely(!ctx->nr_counters)) goto out; + /* + * Add any time since the last sched_out to the lost time + * so it doesn't get included in the total_time_enabled and + * total_time_running measures for counters in the context. + */ + ctx->time_lost = get_context_time(ctx, 0) - ctx->time_now; + flags = hw_perf_save_disable(); /* @@ -817,8 +903,10 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, * If this pinned group hasn't been scheduled, * put it in error state. */ - if (counter->state == PERF_COUNTER_STATE_INACTIVE) + if (counter->state == PERF_COUNTER_STATE_INACTIVE) { + update_group_times(counter); counter->state = PERF_COUNTER_STATE_ERROR; + } } list_for_each_entry(counter, &ctx->counter_list, list_entry) { @@ -902,8 +990,10 @@ int perf_counter_task_disable(void) perf_flags = hw_perf_save_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->state != PERF_COUNTER_STATE_ERROR) + if (counter->state != PERF_COUNTER_STATE_ERROR) { + update_group_times(counter); counter->state = PERF_COUNTER_STATE_OFF; + } } hw_perf_restore(perf_flags); @@ -946,6 +1036,8 @@ int perf_counter_task_enable(void) if (counter->state > PERF_COUNTER_STATE_OFF) continue; counter->state = PERF_COUNTER_STATE_INACTIVE; + counter->tstamp_enabled = ctx->time_now - + counter->total_time_enabled; counter->hw_event.disabled = 0; } hw_perf_restore(perf_flags); @@ -1009,10 +1101,14 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) static void __read(void *info) { struct perf_counter *counter = info; + struct perf_counter_context *ctx = counter->ctx; unsigned long flags; curr_rq_lock_irq_save(&flags); + if (ctx->is_active) + update_context_time(ctx, 1); counter->hw_ops->read(counter); + update_counter_times(counter); curr_rq_unlock_irq_restore(&flags); } @@ -1025,6 +1121,8 @@ static u64 perf_counter_read(struct perf_counter *counter) if (counter->state == PERF_COUNTER_STATE_ACTIVE) { smp_call_function_single(counter->oncpu, __read, counter, 1); + } else if (counter->state == PERF_COUNTER_STATE_INACTIVE) { + update_counter_times(counter); } return atomic64_read(&counter->count); @@ -1137,10 +1235,8 @@ static int perf_release(struct inode *inode, struct file *file) static ssize_t perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) { - u64 cntval; - - if (count < sizeof(cntval)) - return -EINVAL; + u64 values[3]; + int n; /* * Return end-of-file for a read on a counter that is in @@ -1151,10 +1247,24 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) return 0; mutex_lock(&counter->mutex); - cntval = perf_counter_read(counter); + values[0] = perf_counter_read(counter); + n = 1; + if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = counter->total_time_enabled + + atomic64_read(&counter->child_total_time_enabled); + if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = counter->total_time_running + + atomic64_read(&counter->child_total_time_running); mutex_unlock(&counter->mutex); - return put_user(cntval, (u64 __user *) buf) ? -EFAULT : sizeof(cntval); + if (count < n * sizeof(u64)) + return -EINVAL; + count = n * sizeof(u64); + + if (copy_to_user(buf, values, count)) + return -EFAULT; + + return count; } static ssize_t @@ -2290,8 +2400,7 @@ inherit_counter(struct perf_counter *parent_counter, * Link it up in the child's context: */ child_counter->task = child; - list_add_counter(child_counter, child_ctx); - child_ctx->nr_counters++; + add_counter_to_ctx(child_counter, child_ctx); child_counter->parent = parent_counter; /* @@ -2361,6 +2470,10 @@ static void sync_child_counter(struct perf_counter *child_counter, * Add back the child's count to the parent's count: */ atomic64_add(child_val, &parent_counter->count); + atomic64_add(child_counter->total_time_enabled, + &parent_counter->child_total_time_enabled); + atomic64_add(child_counter->total_time_running, + &parent_counter->child_total_time_running); /* * Remove this counter from the parent's list @@ -2395,6 +2508,7 @@ __perf_counter_exit_task(struct task_struct *child, if (child != current) { wait_task_inactive(child, 0); list_del_init(&child_counter->list_entry); + update_counter_times(child_counter); } else { struct perf_cpu_context *cpuctx; unsigned long flags; @@ -2412,6 +2526,7 @@ __perf_counter_exit_task(struct task_struct *child, cpuctx = &__get_cpu_var(perf_cpu_context); group_sched_out(child_counter, cpuctx, child_ctx); + update_counter_times(child_counter); list_del_init(&child_counter->list_entry); -- cgit v1.2.3 From 925d519ab82b6dd7aca9420d809ee83819c08db2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 30 Mar 2009 19:07:02 +0200 Subject: perf_counter: unify and fix delayed counter wakeup While going over the wakeup code I noticed delayed wakeups only work for hardware counters but basically all software counters rely on them. This patch unifies and generalizes the delayed wakeup to fix this issue. Since we're dealing with NMI context bits here, use a cmpxchg() based single link list implementation to track counters that have pending wakeups. [ This should really be generic code for delayed wakeups, but since we cannot use cmpxchg()/xchg() in generic code, I've let it live in the perf_counter code. -- Eric Dumazet could use it to aggregate the network wakeups. ] Furthermore, the x86 method of using TIF flags was flawed in that its quite possible to end up setting the bit on the idle task, loosing the wakeup. The powerpc method uses per-cpu storage and does appear to be sufficient. Signed-off-by: Peter Zijlstra Acked-by: Paul Mackerras Orig-LKML-Reference: <20090330171023.153932974@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 128 ++++++++++++++++++++++++++++++++++++++++++++++---- kernel/timer.c | 3 ++ 2 files changed, 123 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 3b862a7988cd..f70ff80e79d7 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1197,8 +1197,12 @@ static void free_counter_rcu(struct rcu_head *head) kfree(counter); } +static void perf_pending_sync(struct perf_counter *counter); + static void free_counter(struct perf_counter *counter) { + perf_pending_sync(counter); + if (counter->destroy) counter->destroy(counter); @@ -1528,6 +1532,118 @@ static const struct file_operations perf_fops = { .mmap = perf_mmap, }; +/* + * Perf counter wakeup + * + * If there's data, ensure we set the poll() state and publish everything + * to user-space before waking everybody up. + */ + +void perf_counter_wakeup(struct perf_counter *counter) +{ + struct perf_mmap_data *data; + + rcu_read_lock(); + data = rcu_dereference(counter->data); + if (data) { + (void)atomic_xchg(&data->wakeup, POLL_IN); + __perf_counter_update_userpage(counter, data); + } + rcu_read_unlock(); + + wake_up_all(&counter->waitq); +} + +/* + * Pending wakeups + * + * Handle the case where we need to wakeup up from NMI (or rq->lock) context. + * + * The NMI bit means we cannot possibly take locks. Therefore, maintain a + * single linked list and use cmpxchg() to add entries lockless. + */ + +#define PENDING_TAIL ((struct perf_wakeup_entry *)-1UL) + +static DEFINE_PER_CPU(struct perf_wakeup_entry *, perf_wakeup_head) = { + PENDING_TAIL, +}; + +static void perf_pending_queue(struct perf_counter *counter) +{ + struct perf_wakeup_entry **head; + struct perf_wakeup_entry *prev, *next; + + if (cmpxchg(&counter->wakeup.next, NULL, PENDING_TAIL) != NULL) + return; + + head = &get_cpu_var(perf_wakeup_head); + + do { + prev = counter->wakeup.next = *head; + next = &counter->wakeup; + } while (cmpxchg(head, prev, next) != prev); + + set_perf_counter_pending(); + + put_cpu_var(perf_wakeup_head); +} + +static int __perf_pending_run(void) +{ + struct perf_wakeup_entry *list; + int nr = 0; + + list = xchg(&__get_cpu_var(perf_wakeup_head), PENDING_TAIL); + while (list != PENDING_TAIL) { + struct perf_counter *counter = container_of(list, + struct perf_counter, wakeup); + + list = list->next; + + counter->wakeup.next = NULL; + /* + * Ensure we observe the unqueue before we issue the wakeup, + * so that we won't be waiting forever. + * -- see perf_not_pending(). + */ + smp_wmb(); + + perf_counter_wakeup(counter); + nr++; + } + + return nr; +} + +static inline int perf_not_pending(struct perf_counter *counter) +{ + /* + * If we flush on whatever cpu we run, there is a chance we don't + * need to wait. + */ + get_cpu(); + __perf_pending_run(); + put_cpu(); + + /* + * Ensure we see the proper queue state before going to sleep + * so that we do not miss the wakeup. -- see perf_pending_handle() + */ + smp_rmb(); + return counter->wakeup.next == NULL; +} + +static void perf_pending_sync(struct perf_counter *counter) +{ + wait_event(counter->waitq, perf_not_pending(counter)); +} + +void perf_counter_do_pending(void) +{ + __perf_pending_run(); +} + /* * Output */ @@ -1611,13 +1727,10 @@ static void perf_output_copy(struct perf_output_handle *handle, static void perf_output_end(struct perf_output_handle *handle, int nmi) { if (handle->wakeup) { - (void)atomic_xchg(&handle->data->wakeup, POLL_IN); - __perf_counter_update_userpage(handle->counter, handle->data); - if (nmi) { - handle->counter->wakeup_pending = 1; - set_perf_counter_pending(); - } else - wake_up(&handle->counter->waitq); + if (nmi) + perf_pending_queue(handle->counter); + else + perf_counter_wakeup(handle->counter); } rcu_read_unlock(); } @@ -2211,7 +2324,6 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->cpu = cpu; counter->hw_event = *hw_event; - counter->wakeup_pending = 0; counter->group_leader = group_leader; counter->hw_ops = NULL; counter->ctx = ctx; diff --git a/kernel/timer.c b/kernel/timer.c index b4555568b4e4..672ca25fbc43 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -37,6 +37,7 @@ #include #include #include +#include #include #include @@ -1167,6 +1168,8 @@ static void run_timer_softirq(struct softirq_action *h) { struct tvec_base *base = __get_cpu_var(tvec_bases); + perf_counter_do_pending(); + hrtimer_run_pending(); if (time_after_eq(jiffies, base->timer_jiffies)) -- cgit v1.2.3 From 38ff667b321b00f5e6830e93fb4ab11a653a2920 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 30 Mar 2009 19:07:03 +0200 Subject: perf_counter: fix update_userpage() It just occured to me it is possible to have multiple contending updates of the userpage (mmap information vs overflow vs counter). This would break the seqlock logic. It appear the arch code uses this from NMI context, so we cannot possibly serialize its use, therefore separate the data_head update from it and let it return to its original use. The arch code needs to make sure there are no contending callers by disabling the counter before using it -- powerpc appears to do this nicely. Signed-off-by: Peter Zijlstra Acked-by: Paul Mackerras Orig-LKML-Reference: <20090330171023.241410660@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 38 +++++++++++++++++++++++--------------- 1 file changed, 23 insertions(+), 15 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f70ff80e79d7..c95e92329b97 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1316,10 +1316,22 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) return err; } -static void __perf_counter_update_userpage(struct perf_counter *counter, - struct perf_mmap_data *data) +/* + * Callers need to ensure there can be no nesting of this function, otherwise + * the seqlock logic goes bad. We can not serialize this because the arch + * code calls this from NMI context. + */ +void perf_counter_update_userpage(struct perf_counter *counter) { - struct perf_counter_mmap_page *userpg = data->user_page; + struct perf_mmap_data *data; + struct perf_counter_mmap_page *userpg; + + rcu_read_lock(); + data = rcu_dereference(counter->data); + if (!data) + goto unlock; + + userpg = data->user_page; /* * Disable preemption so as to not let the corresponding user-space @@ -1333,20 +1345,10 @@ static void __perf_counter_update_userpage(struct perf_counter *counter, if (counter->state == PERF_COUNTER_STATE_ACTIVE) userpg->offset -= atomic64_read(&counter->hw.prev_count); - userpg->data_head = atomic_read(&data->head); smp_wmb(); ++userpg->lock; preempt_enable(); -} - -void perf_counter_update_userpage(struct perf_counter *counter) -{ - struct perf_mmap_data *data; - - rcu_read_lock(); - data = rcu_dereference(counter->data); - if (data) - __perf_counter_update_userpage(counter, data); +unlock: rcu_read_unlock(); } @@ -1547,7 +1549,13 @@ void perf_counter_wakeup(struct perf_counter *counter) data = rcu_dereference(counter->data); if (data) { (void)atomic_xchg(&data->wakeup, POLL_IN); - __perf_counter_update_userpage(counter, data); + /* + * Ensure all data writes are issued before updating the + * user-space data head information. The matching rmb() + * will be in userspace after reading this value. + */ + smp_wmb(); + data->user_page->data_head = atomic_read(&data->head); } rcu_read_unlock(); -- cgit v1.2.3 From 0a4a93919bdc5cee48fe4367591e8e0449c1086c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 30 Mar 2009 19:07:05 +0200 Subject: perf_counter: executable mmap() information Currently the profiling information returns userspace IPs but no way to correlate them to userspace code. Userspace could look into /proc/$pid/maps but that might not be current or even present anymore at the time of analyzing the IPs. Therefore provide means to track the mmap information and provide it in the output stream. XXX: only covers mmap()/munmap(), mremap() and mprotect() are missing. Signed-off-by: Peter Zijlstra Acked-by: Paul Mackerras Cc: Andrew Morton Orig-LKML-Reference: <20090330171023.417259499@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 145 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 145 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c95e92329b97..f35e89e3d6a4 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -25,6 +25,7 @@ #include #include #include +#include #include @@ -1843,6 +1844,150 @@ void perf_counter_output(struct perf_counter *counter, } } +/* + * mmap tracking + */ + +struct perf_mmap_event { + struct file *file; + char *file_name; + int file_size; + + struct { + struct perf_event_header header; + + u32 pid; + u32 tid; + u64 start; + u64 len; + u64 pgoff; + } event; +}; + +static void perf_counter_mmap_output(struct perf_counter *counter, + struct perf_mmap_event *mmap_event) +{ + struct perf_output_handle handle; + int size = mmap_event->event.header.size; + int ret = perf_output_begin(&handle, counter, size); + + if (ret) + return; + + perf_output_put(&handle, mmap_event->event); + perf_output_copy(&handle, mmap_event->file_name, + mmap_event->file_size); + perf_output_end(&handle, 0); +} + +static int perf_counter_mmap_match(struct perf_counter *counter, + struct perf_mmap_event *mmap_event) +{ + if (counter->hw_event.mmap && + mmap_event->event.header.type == PERF_EVENT_MMAP) + return 1; + + if (counter->hw_event.munmap && + mmap_event->event.header.type == PERF_EVENT_MUNMAP) + return 1; + + return 0; +} + +static void perf_counter_mmap_ctx(struct perf_counter_context *ctx, + struct perf_mmap_event *mmap_event) +{ + struct perf_counter *counter; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { + if (perf_counter_mmap_match(counter, mmap_event)) + perf_counter_mmap_output(counter, mmap_event); + } + rcu_read_unlock(); +} + +static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) +{ + struct perf_cpu_context *cpuctx; + struct file *file = mmap_event->file; + unsigned int size; + char tmp[16]; + char *buf = NULL; + char *name; + + if (file) { + buf = kzalloc(PATH_MAX, GFP_KERNEL); + if (!buf) { + name = strncpy(tmp, "//enomem", sizeof(tmp)); + goto got_name; + } + name = dentry_path(file->f_dentry, buf, PATH_MAX); + if (IS_ERR(name)) { + name = strncpy(tmp, "//toolong", sizeof(tmp)); + goto got_name; + } + } else { + name = strncpy(tmp, "//anon", sizeof(tmp)); + goto got_name; + } + +got_name: + size = ALIGN(strlen(name), sizeof(u64)); + + mmap_event->file_name = name; + mmap_event->file_size = size; + + mmap_event->event.header.size = sizeof(mmap_event->event) + size; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event); + put_cpu_var(perf_cpu_context); + + perf_counter_mmap_ctx(¤t->perf_counter_ctx, mmap_event); + + kfree(buf); +} + +void perf_counter_mmap(unsigned long addr, unsigned long len, + unsigned long pgoff, struct file *file) +{ + struct perf_mmap_event mmap_event = { + .file = file, + .event = { + .header = { .type = PERF_EVENT_MMAP, }, + .pid = current->group_leader->pid, + .tid = current->pid, + .start = addr, + .len = len, + .pgoff = pgoff, + }, + }; + + perf_counter_mmap_event(&mmap_event); +} + +void perf_counter_munmap(unsigned long addr, unsigned long len, + unsigned long pgoff, struct file *file) +{ + struct perf_mmap_event mmap_event = { + .file = file, + .event = { + .header = { .type = PERF_EVENT_MUNMAP, }, + .pid = current->group_leader->pid, + .tid = current->pid, + .start = addr, + .len = len, + .pgoff = pgoff, + }, + }; + + perf_counter_mmap_event(&mmap_event); +} + /* * Generic software counter infrastructure */ -- cgit v1.2.3 From d5d2bc0dd0379deddb9ede66fec90a3083eaec57 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 30 Mar 2009 19:07:08 +0200 Subject: perf_counter: make it possible for hw_perf_counter_init to return error codes Impact: better error reporting At present, if hw_perf_counter_init encounters an error, all it can do is return NULL, which causes sys_perf_counter_open to return an EINVAL error to userspace. This isn't very informative for userspace; it means that userspace can't tell the difference between "sorry, oprofile is already using the PMU" and "we don't support this CPU" and "this CPU doesn't support the requested generic hardware event". This commit uses the PTR_ERR/ERR_PTR/IS_ERR set of macros to let hw_perf_counter_init return an error code on error rather than just NULL if it wishes. If it does so, that error code will be returned from sys_perf_counter_open to userspace. If it returns NULL, an EINVAL error will be returned to userspace, as before. This also adapts the powerpc hw_perf_counter_init to make use of this to return ENXIO, EINVAL, EBUSY, or EOPNOTSUPP as appropriate. It would be good to add extra error numbers in future to allow userspace to distinguish the various errors that are currently reported as EINVAL, i.e. irq_period < 0, too many events in a group, conflict between exclude_* settings in a group, and PMU resource conflict in a group. [ v2: fix a bug pointed out by Corey Ashford where error returns from hw_perf_counter_init were not handled correctly in the case of raw hardware events.] Signed-off-by: Paul Mackerras Signed-off-by: Peter Zijlstra Orig-LKML-Reference: <20090330171023.682428180@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 35 ++++++++++++++++++++++------------- 1 file changed, 22 insertions(+), 13 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f35e89e3d6a4..d07b45278b4f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2453,10 +2453,11 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, { const struct hw_perf_counter_ops *hw_ops; struct perf_counter *counter; + long err; counter = kzalloc(sizeof(*counter), gfpflags); if (!counter) - return NULL; + return ERR_PTR(-ENOMEM); /* * Single counters are their own group leaders, with an @@ -2505,12 +2506,18 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, hw_ops = tp_perf_counter_init(counter); break; } +done: + err = 0; + if (!hw_ops) + err = -EINVAL; + else if (IS_ERR(hw_ops)) + err = PTR_ERR(hw_ops); - if (!hw_ops) { + if (err) { kfree(counter); - return NULL; + return ERR_PTR(err); } -done: + counter->hw_ops = hw_ops; return counter; @@ -2583,10 +2590,10 @@ SYSCALL_DEFINE5(perf_counter_open, goto err_put_context; } - ret = -EINVAL; counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader, GFP_KERNEL); - if (!counter) + ret = PTR_ERR(counter); + if (IS_ERR(counter)) goto err_put_context; ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); @@ -2658,8 +2665,8 @@ inherit_counter(struct perf_counter *parent_counter, child_counter = perf_counter_alloc(&parent_counter->hw_event, parent_counter->cpu, child_ctx, group_leader, GFP_KERNEL); - if (!child_counter) - return NULL; + if (IS_ERR(child_counter)) + return child_counter; /* * Link it up in the child's context: @@ -2710,15 +2717,17 @@ static int inherit_group(struct perf_counter *parent_counter, { struct perf_counter *leader; struct perf_counter *sub; + struct perf_counter *child_ctr; leader = inherit_counter(parent_counter, parent, parent_ctx, child, NULL, child_ctx); - if (!leader) - return -ENOMEM; + if (IS_ERR(leader)) + return PTR_ERR(leader); list_for_each_entry(sub, &parent_counter->sibling_list, list_entry) { - if (!inherit_counter(sub, parent, parent_ctx, - child, leader, child_ctx)) - return -ENOMEM; + child_ctr = inherit_counter(sub, parent, parent_ctx, + child, leader, child_ctx); + if (IS_ERR(child_ctr)) + return PTR_ERR(child_ctr); } return 0; } -- cgit v1.2.3 From 78d613eb129fc4edf0e2cabfcc6a4c5285482d21 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 30 Mar 2009 19:07:11 +0200 Subject: perf_counter: small cleanup of the output routines Move the nmi argument to the _begin() function, so that _end() only needs the handle. This allows the _begin() function to generate a wakeup on event loss. Signed-off-by: Peter Zijlstra Acked-by: Paul Mackerras Orig-LKML-Reference: <20090330171023.959404268@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 42 ++++++++++++++++++++++++++---------------- 1 file changed, 26 insertions(+), 16 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d07b45278b4f..4471e7e2c109 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1663,10 +1663,20 @@ struct perf_output_handle { unsigned int offset; unsigned int head; int wakeup; + int nmi; }; +static inline void __perf_output_wakeup(struct perf_output_handle *handle) +{ + if (handle->nmi) + perf_pending_queue(handle->counter); + else + perf_counter_wakeup(handle->counter); +} + static int perf_output_begin(struct perf_output_handle *handle, - struct perf_counter *counter, unsigned int size) + struct perf_counter *counter, unsigned int size, + int nmi) { struct perf_mmap_data *data; unsigned int offset, head; @@ -1676,15 +1686,17 @@ static int perf_output_begin(struct perf_output_handle *handle, if (!data) goto out; + handle->counter = counter; + handle->nmi = nmi; + if (!data->nr_pages) - goto out; + goto fail; do { offset = head = atomic_read(&data->head); head += size; } while (atomic_cmpxchg(&data->head, offset, head) != offset); - handle->counter = counter; handle->data = data; handle->offset = offset; handle->head = head; @@ -1692,6 +1704,8 @@ static int perf_output_begin(struct perf_output_handle *handle, return 0; +fail: + __perf_output_wakeup(handle); out: rcu_read_unlock(); @@ -1733,14 +1747,10 @@ static void perf_output_copy(struct perf_output_handle *handle, #define perf_output_put(handle, x) \ perf_output_copy((handle), &(x), sizeof(x)) -static void perf_output_end(struct perf_output_handle *handle, int nmi) +static void perf_output_end(struct perf_output_handle *handle) { - if (handle->wakeup) { - if (nmi) - perf_pending_queue(handle->counter); - else - perf_counter_wakeup(handle->counter); - } + if (handle->wakeup) + __perf_output_wakeup(handle); rcu_read_unlock(); } @@ -1750,12 +1760,12 @@ static int perf_output_write(struct perf_counter *counter, int nmi, struct perf_output_handle handle; int ret; - ret = perf_output_begin(&handle, counter, size); + ret = perf_output_begin(&handle, counter, size, nmi); if (ret) goto out; perf_output_copy(&handle, buf, size); - perf_output_end(&handle, nmi); + perf_output_end(&handle); out: return ret; @@ -1804,7 +1814,7 @@ static void perf_output_group(struct perf_counter *counter, int nmi) size = sizeof(header) + counter->nr_siblings * sizeof(entry); - ret = perf_output_begin(&handle, counter, size); + ret = perf_output_begin(&handle, counter, size, nmi); if (ret) return; @@ -1824,7 +1834,7 @@ static void perf_output_group(struct perf_counter *counter, int nmi) perf_output_put(&handle, entry); } - perf_output_end(&handle, nmi); + perf_output_end(&handle); } void perf_counter_output(struct perf_counter *counter, @@ -1869,7 +1879,7 @@ static void perf_counter_mmap_output(struct perf_counter *counter, { struct perf_output_handle handle; int size = mmap_event->event.header.size; - int ret = perf_output_begin(&handle, counter, size); + int ret = perf_output_begin(&handle, counter, size, 0); if (ret) return; @@ -1877,7 +1887,7 @@ static void perf_counter_mmap_output(struct perf_counter *counter, perf_output_put(&handle, mmap_event->event); perf_output_copy(&handle, mmap_event->file_name, mmap_event->file_size); - perf_output_end(&handle, 0); + perf_output_end(&handle); } static int perf_counter_mmap_match(struct perf_counter *counter, -- cgit v1.2.3 From 5ed00415e304203a0a9dcaef226d6d3f1106070e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 30 Mar 2009 19:07:12 +0200 Subject: perf_counter: re-arrange the perf_event_type Breaks ABI yet again :-) Change the event type so that [0, 2^31-1] are regular event types, but [2^31, 2^32-1] forms a bitmask for overflow events. Signed-off-by: Peter Zijlstra Acked-by: Paul Mackerras Orig-LKML-Reference: <20090330171024.047961770@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 56 +++++++++++++++++++++++---------------------------- 1 file changed, 25 insertions(+), 31 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4471e7e2c109..d93e9ddf7848 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1754,50 +1754,44 @@ static void perf_output_end(struct perf_output_handle *handle) rcu_read_unlock(); } -static int perf_output_write(struct perf_counter *counter, int nmi, - void *buf, ssize_t size) -{ - struct perf_output_handle handle; - int ret; - - ret = perf_output_begin(&handle, counter, size, nmi); - if (ret) - goto out; - - perf_output_copy(&handle, buf, size); - perf_output_end(&handle); - -out: - return ret; -} - static void perf_output_simple(struct perf_counter *counter, int nmi, struct pt_regs *regs) { - unsigned int size; + int ret; + struct perf_output_handle handle; + struct perf_event_header header; + u64 ip; struct { - struct perf_event_header header; - u64 ip; u32 pid, tid; - } event; + } tid_entry; - event.header.type = PERF_EVENT_IP; - event.ip = instruction_pointer(regs); + header.type = PERF_EVENT_OVERFLOW; + header.size = sizeof(header); - size = sizeof(event); + ip = instruction_pointer(regs); + header.type |= __PERF_EVENT_IP; + header.size += sizeof(ip); if (counter->hw_event.include_tid) { /* namespace issues */ - event.pid = current->group_leader->pid; - event.tid = current->pid; + tid_entry.pid = current->group_leader->pid; + tid_entry.tid = current->pid; - event.header.type |= __PERF_EVENT_TID; - } else - size -= sizeof(u64); + header.type |= __PERF_EVENT_TID; + header.size += sizeof(tid_entry); + } - event.header.size = size; + ret = perf_output_begin(&handle, counter, header.size, nmi); + if (ret) + return; + + perf_output_put(&handle, header); + perf_output_put(&handle, ip); + + if (counter->hw_event.include_tid) + perf_output_put(&handle, tid_entry); - perf_output_write(counter, nmi, &event, size); + perf_output_end(&handle); } static void perf_output_group(struct perf_counter *counter, int nmi) -- cgit v1.2.3 From 394ee07623cf556c8daae2b3c00cf5fea47f0811 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 30 Mar 2009 19:07:14 +0200 Subject: perf_counter: provide generic callchain bits Provide the generic callchain support bits. If hw_event->callchain is set the arch specific perf_callchain() function is called upon to provide a perf_callchain_entry structure filled with the current callchain. If it does so, it is added to the overflow output event. Signed-off-by: Peter Zijlstra Acked-by: Paul Mackerras Orig-LKML-Reference: <20090330171024.254266860@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d93e9ddf7848..860cdc26bd7a 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1653,6 +1653,17 @@ void perf_counter_do_pending(void) __perf_pending_run(); } +/* + * Callchain support -- arch specific + */ + +struct perf_callchain_entry * +__attribute__((weak)) +perf_callchain(struct pt_regs *regs) +{ + return NULL; +} + /* * Output */ @@ -1764,6 +1775,8 @@ static void perf_output_simple(struct perf_counter *counter, struct { u32 pid, tid; } tid_entry; + struct perf_callchain_entry *callchain = NULL; + int callchain_size = 0; header.type = PERF_EVENT_OVERFLOW; header.size = sizeof(header); @@ -1781,6 +1794,17 @@ static void perf_output_simple(struct perf_counter *counter, header.size += sizeof(tid_entry); } + if (counter->hw_event.callchain) { + callchain = perf_callchain(regs); + + if (callchain) { + callchain_size = (1 + callchain->nr) * sizeof(u64); + + header.type |= __PERF_EVENT_CALLCHAIN; + header.size += callchain_size; + } + } + ret = perf_output_begin(&handle, counter, header.size, nmi); if (ret) return; @@ -1791,6 +1815,9 @@ static void perf_output_simple(struct perf_counter *counter, if (counter->hw_event.include_tid) perf_output_put(&handle, tid_entry); + if (callchain) + perf_output_copy(&handle, callchain, callchain_size); + perf_output_end(&handle); } -- cgit v1.2.3 From 8a057d84912f36e53f970c4d177cb4bb6b2f9e08 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 2 Apr 2009 11:11:59 +0200 Subject: perf_counter: move the event overflow output bits to record_type Per suggestion from Paul, move the event overflow bits to record_type and sanitize the enums a bit. Breaks the ABI -- again ;-) Suggested-by: Paul Mackerras Signed-off-by: Peter Zijlstra Cc: Corey Ashford Orig-LKML-Reference: <20090402091319.151921176@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 101 ++++++++++++++++++++------------------------------ 1 file changed, 40 insertions(+), 61 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 860cdc26bd7a..995063df910f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1765,27 +1765,34 @@ static void perf_output_end(struct perf_output_handle *handle) rcu_read_unlock(); } -static void perf_output_simple(struct perf_counter *counter, - int nmi, struct pt_regs *regs) +void perf_counter_output(struct perf_counter *counter, + int nmi, struct pt_regs *regs) { int ret; + u64 record_type = counter->hw_event.record_type; struct perf_output_handle handle; struct perf_event_header header; u64 ip; struct { u32 pid, tid; } tid_entry; + struct { + u64 event; + u64 counter; + } group_entry; struct perf_callchain_entry *callchain = NULL; int callchain_size = 0; - header.type = PERF_EVENT_OVERFLOW; + header.type = PERF_EVENT_COUNTER_OVERFLOW; header.size = sizeof(header); - ip = instruction_pointer(regs); - header.type |= __PERF_EVENT_IP; - header.size += sizeof(ip); + if (record_type & PERF_RECORD_IP) { + ip = instruction_pointer(regs); + header.type |= __PERF_EVENT_IP; + header.size += sizeof(ip); + } - if (counter->hw_event.include_tid) { + if (record_type & PERF_RECORD_TID) { /* namespace issues */ tid_entry.pid = current->group_leader->pid; tid_entry.tid = current->pid; @@ -1794,7 +1801,13 @@ static void perf_output_simple(struct perf_counter *counter, header.size += sizeof(tid_entry); } - if (counter->hw_event.callchain) { + if (record_type & PERF_RECORD_GROUP) { + header.type |= __PERF_EVENT_GROUP; + header.size += sizeof(u64) + + counter->nr_siblings * sizeof(group_entry); + } + + if (record_type & PERF_RECORD_CALLCHAIN) { callchain = perf_callchain(regs); if (callchain) { @@ -1810,69 +1823,35 @@ static void perf_output_simple(struct perf_counter *counter, return; perf_output_put(&handle, header); - perf_output_put(&handle, ip); - if (counter->hw_event.include_tid) - perf_output_put(&handle, tid_entry); + if (record_type & PERF_RECORD_IP) + perf_output_put(&handle, ip); - if (callchain) - perf_output_copy(&handle, callchain, callchain_size); - - perf_output_end(&handle); -} - -static void perf_output_group(struct perf_counter *counter, int nmi) -{ - struct perf_output_handle handle; - struct perf_event_header header; - struct perf_counter *leader, *sub; - unsigned int size; - struct { - u64 event; - u64 counter; - } entry; - int ret; - - size = sizeof(header) + counter->nr_siblings * sizeof(entry); + if (record_type & PERF_RECORD_TID) + perf_output_put(&handle, tid_entry); - ret = perf_output_begin(&handle, counter, size, nmi); - if (ret) - return; + if (record_type & PERF_RECORD_GROUP) { + struct perf_counter *leader, *sub; + u64 nr = counter->nr_siblings; - header.type = PERF_EVENT_GROUP; - header.size = size; + perf_output_put(&handle, nr); - perf_output_put(&handle, header); + leader = counter->group_leader; + list_for_each_entry(sub, &leader->sibling_list, list_entry) { + if (sub != counter) + sub->hw_ops->read(sub); - leader = counter->group_leader; - list_for_each_entry(sub, &leader->sibling_list, list_entry) { - if (sub != counter) - sub->hw_ops->read(sub); + group_entry.event = sub->hw_event.config; + group_entry.counter = atomic64_read(&sub->count); - entry.event = sub->hw_event.config; - entry.counter = atomic64_read(&sub->count); - - perf_output_put(&handle, entry); + perf_output_put(&handle, group_entry); + } } - perf_output_end(&handle); -} - -void perf_counter_output(struct perf_counter *counter, - int nmi, struct pt_regs *regs) -{ - switch (counter->hw_event.record_type) { - case PERF_RECORD_SIMPLE: - return; - - case PERF_RECORD_IRQ: - perf_output_simple(counter, nmi, regs); - break; + if (callchain) + perf_output_copy(&handle, callchain, callchain_size); - case PERF_RECORD_GROUP: - perf_output_group(counter, nmi); - break; - } + perf_output_end(&handle); } /* -- cgit v1.2.3 From c457810ab4a825161aec6ef71b581e1bc8febd1a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 2 Apr 2009 11:12:01 +0200 Subject: perf_counter: per event wakeups By request, provide a way to request a wakeup every 'n' events instead of every page of output. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Orig-LKML-Reference: <20090402091319.323309784@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 995063df910f..9bcab10e735d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1760,7 +1760,15 @@ static void perf_output_copy(struct perf_output_handle *handle, static void perf_output_end(struct perf_output_handle *handle) { - if (handle->wakeup) + int wakeup_events = handle->counter->hw_event.wakeup_events; + + if (wakeup_events) { + int events = atomic_inc_return(&handle->data->events); + if (events >= wakeup_events) { + atomic_sub(wakeup_events, &handle->data->events); + __perf_output_wakeup(handle); + } + } else if (handle->wakeup) __perf_output_wakeup(handle); rcu_read_unlock(); } -- cgit v1.2.3 From 5872bdb88a35fae7d224bd6b21e5f377e854ccfc Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 2 Apr 2009 11:12:03 +0200 Subject: perf_counter: add more context information Put in counts to tell which ips belong to what context. ----- | | hv | -- nr | | kernel | -- | | user ----- Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Orig-LKML-Reference: <20090402091319.493101305@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 9bcab10e735d..f105a6e696c2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1819,7 +1819,7 @@ void perf_counter_output(struct perf_counter *counter, callchain = perf_callchain(regs); if (callchain) { - callchain_size = (1 + callchain->nr) * sizeof(u64); + callchain_size = (2 + callchain->nr) * sizeof(u64); header.type |= __PERF_EVENT_CALLCHAIN; header.size += callchain_size; -- cgit v1.2.3 From 92f22a3865abe87eea2609a6f8e5be5123f7ce4f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 2 Apr 2009 11:12:04 +0200 Subject: perf_counter: update mmap() counter read Paul noted that we don't need SMP barriers for the mmap() counter read because its always on the same cpu (otherwise you can't access the hw counter anyway). So remove the SMP barriers and replace them with regular compiler barriers. Further, update the comment to include a race free method of reading said hardware counter. The primary change is putting the pmc_read inside the seq-loop, otherwise we can still race and read rubbish. Noticed-by: Paul Mackerras Signed-off-by: Peter Zijlstra Cc: Corey Ashford Orig-LKML-Reference: <20090402091319.577951445@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f105a6e696c2..2a5d4f525567 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1340,13 +1340,13 @@ void perf_counter_update_userpage(struct perf_counter *counter) */ preempt_disable(); ++userpg->lock; - smp_wmb(); + barrier(); userpg->index = counter->hw.idx; userpg->offset = atomic64_read(&counter->count); if (counter->state == PERF_COUNTER_STATE_ACTIVE) userpg->offset -= atomic64_read(&counter->hw.prev_count); - smp_wmb(); + barrier(); ++userpg->lock; preempt_enable(); unlock: -- cgit v1.2.3 From ca5f9524d61f54b1f618293ab92fc6b49cac864d Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:39:33 -0700 Subject: futex: separate futex_wait_queue_me() logic from futex_wait() Refactor futex_wait() in preparation for futex_wait_requeue_pi(). In order to reuse a good chunk of the futex_wait() code for the upcoming futex_wait_requeue_pi() function, this patch breaks out the queue-to-wakeup section of futex_wait() into futex_wait_queue_me(). Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 138 +++++++++++++++++++++++++++++++-------------------------- 1 file changed, 76 insertions(+), 62 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 6b50a024bca2..ebb48d6d1a87 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1115,24 +1115,87 @@ handle_fault: static long futex_wait_restart(struct restart_block *restart); +/** + * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal + * @hb: the futex hash bucket, must be locked by the caller + * @q: the futex_q to queue up on + * @timeout: the prepared hrtimer_sleeper, or null for no timeout + * @wait: the wait_queue to add to the futex_q after queueing in the hb + */ +static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, + struct hrtimer_sleeper *timeout, + wait_queue_t *wait) +{ + queue_me(q, hb); + + /* + * There might have been scheduling since the queue_me(), as we + * cannot hold a spinlock across the get_user() in case it + * faults, and we cannot just set TASK_INTERRUPTIBLE state when + * queueing ourselves into the futex hash. This code thus has to + * rely on the futex_wake() code removing us from hash when it + * wakes us up. + */ + + /* add_wait_queue is the barrier after __set_current_state. */ + __set_current_state(TASK_INTERRUPTIBLE); + + /* + * Add current as the futex_q waiter. We don't remove ourselves from + * the wait_queue because we are the only user of it. + */ + add_wait_queue(&q->waiter, wait); + + /* Arm the timer */ + if (timeout) { + hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); + if (!hrtimer_active(&timeout->timer)) + timeout->task = NULL; + } + + /* + * !plist_node_empty() is safe here without any lock. + * q.lock_ptr != 0 is not safe, because of ordering against wakeup. + */ + if (likely(!plist_node_empty(&q->list))) { + /* + * If the timer has already expired, current will already be + * flagged for rescheduling. Only call schedule if there + * is no timeout, or if it has yet to expire. + */ + if (!timeout || timeout->task) + schedule(); + } + __set_current_state(TASK_RUNNING); +} + static int futex_wait(u32 __user *uaddr, int fshared, u32 val, ktime_t *abs_time, u32 bitset, int clockrt) { - struct task_struct *curr = current; + struct hrtimer_sleeper timeout, *to = NULL; + DECLARE_WAITQUEUE(wait, current); struct restart_block *restart; - DECLARE_WAITQUEUE(wait, curr); struct futex_hash_bucket *hb; struct futex_q q; u32 uval; int ret; - struct hrtimer_sleeper t; - int rem = 0; if (!bitset) return -EINVAL; q.pi_state = NULL; q.bitset = bitset; + + if (abs_time) { + to = &timeout; + + hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : + CLOCK_MONOTONIC, HRTIMER_MODE_ABS); + hrtimer_init_sleeper(to, current); + hrtimer_set_expires_range_ns(&to->timer, *abs_time, + current->timer_slack_ns); + } + retry: q.key = FUTEX_KEY_INIT; ret = get_futex_key(uaddr, fshared, &q.key); @@ -1178,75 +1241,22 @@ retry_private: goto retry; } ret = -EWOULDBLOCK; + + /* Only actually queue if *uaddr contained val. */ if (unlikely(uval != val)) { queue_unlock(&q, hb); goto out_put_key; } - /* Only actually queue if *uaddr contained val. */ - queue_me(&q, hb); - - /* - * There might have been scheduling since the queue_me(), as we - * cannot hold a spinlock across the get_user() in case it - * faults, and we cannot just set TASK_INTERRUPTIBLE state when - * queueing ourselves into the futex hash. This code thus has to - * rely on the futex_wake() code removing us from hash when it - * wakes us up. - */ - - /* add_wait_queue is the barrier after __set_current_state. */ - __set_current_state(TASK_INTERRUPTIBLE); - add_wait_queue(&q.waiter, &wait); - /* - * !plist_node_empty() is safe here without any lock. - * q.lock_ptr != 0 is not safe, because of ordering against wakeup. - */ - if (likely(!plist_node_empty(&q.list))) { - if (!abs_time) - schedule(); - else { - hrtimer_init_on_stack(&t.timer, - clockrt ? CLOCK_REALTIME : - CLOCK_MONOTONIC, - HRTIMER_MODE_ABS); - hrtimer_init_sleeper(&t, current); - hrtimer_set_expires_range_ns(&t.timer, *abs_time, - current->timer_slack_ns); - - hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS); - if (!hrtimer_active(&t.timer)) - t.task = NULL; - - /* - * the timer could have already expired, in which - * case current would be flagged for rescheduling. - * Don't bother calling schedule. - */ - if (likely(t.task)) - schedule(); - - hrtimer_cancel(&t.timer); - - /* Flag if a timeout occured */ - rem = (t.task == NULL); - - destroy_hrtimer_on_stack(&t.timer); - } - } - __set_current_state(TASK_RUNNING); - - /* - * NOTE: we don't remove ourselves from the waitqueue because - * we are the only user of it. - */ + /* queue_me and wait for wakeup, timeout, or a signal. */ + futex_wait_queue_me(hb, &q, to, &wait); /* If we were woken (and unqueued), we succeeded, whatever. */ ret = 0; if (!unqueue_me(&q)) goto out_put_key; ret = -ETIMEDOUT; - if (rem) + if (to && !to->task) goto out_put_key; /* @@ -1275,6 +1285,10 @@ retry_private: out_put_key: put_futex_key(fshared, &q.key); out: + if (to) { + hrtimer_cancel(&to->timer); + destroy_hrtimer_on_stack(&to->timer); + } return ret; } -- cgit v1.2.3 From 4b1c486b3587d2abf50bee4a05eb488cd4045f2c Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:39:42 -0700 Subject: futex: add helper to find the top prio waiter of a futex Improve legibility by wrapping finding the top waiter in a function. This will be used by the follow-on patches for enabling requeue pi. Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index ebb48d6d1a87..421fb5e42a10 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -276,6 +276,25 @@ void put_futex_key(int fshared, union futex_key *key) drop_futex_key_refs(key); } +/** + * futex_top_waiter() - Return the highest priority waiter on a futex + * @hb: the hash bucket the futex_q's reside in + * @key: the futex key (to distinguish it from other futex futex_q's) + * + * Must be called with the hb lock held. + */ +static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, + union futex_key *key) +{ + struct futex_q *this; + + plist_for_each_entry(this, &hb->chain, list) { + if (match_futex(&this->key, key)) + return this; + } + return NULL; +} + static u32 cmpxchg_futex_value_locked(u32 __user *uaddr, u32 uval, u32 newval) { u32 curval; -- cgit v1.2.3 From 1a52084d0919c2799258737c21fb328a9de159b5 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:39:52 -0700 Subject: futex: split out atomic logic from futex_lock_pi() Refactor the atomic portion of futex_lock_pi() into futex_lock_pi_atomic(). This logic will be needed by requeue_pi, so modularize it to reduce code duplication. The only significant change is passing of the task to try and take the lock for. This simplifies the -EDEADLK test as if the lock is owned by task t, it's a deadlock, regardless of if we are doing requeue pi or not. This patch updates the corresponding comment accordingly. Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 224 +++++++++++++++++++++++++++++++++------------------------ 1 file changed, 130 insertions(+), 94 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 421fb5e42a10..986b16e44534 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -556,6 +556,127 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, return 0; } +/** + * futex_lock_pi_atomic() - atomic work required to acquire a pi aware futex + * @uaddr: the pi futex user address + * @hb: the pi futex hash bucket + * @key: the futex key associated with uaddr and hb + * @ps: the pi_state pointer where we store the result of the lookup + * @task: the task to perform the atomic lock work for. This will be + * "current" except in the case of requeue pi. + * + * Returns: + * 0 - ready to wait + * 1 - acquired the lock + * <0 - error + * + * The hb->lock and futex_key refs shall be held by the caller. + */ +static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, + union futex_key *key, + struct futex_pi_state **ps, + struct task_struct *task) +{ + int lock_taken, ret, ownerdied = 0; + u32 uval, newval, curval; + +retry: + ret = lock_taken = 0; + + /* + * To avoid races, we attempt to take the lock here again + * (by doing a 0 -> TID atomic cmpxchg), while holding all + * the locks. It will most likely not succeed. + */ + newval = task_pid_vnr(task); + + curval = cmpxchg_futex_value_locked(uaddr, 0, newval); + + if (unlikely(curval == -EFAULT)) + return -EFAULT; + + /* + * Detect deadlocks. + */ + if ((unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(task)))) + return -EDEADLK; + + /* + * Surprise - we got the lock. Just return to userspace: + */ + if (unlikely(!curval)) + return 1; + + uval = curval; + + /* + * Set the FUTEX_WAITERS flag, so the owner will know it has someone + * to wake at the next unlock. + */ + newval = curval | FUTEX_WAITERS; + + /* + * There are two cases, where a futex might have no owner (the + * owner TID is 0): OWNER_DIED. We take over the futex in this + * case. We also do an unconditional take over, when the owner + * of the futex died. + * + * This is safe as we are protected by the hash bucket lock ! + */ + if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { + /* Keep the OWNER_DIED bit */ + newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(task); + ownerdied = 0; + lock_taken = 1; + } + + curval = cmpxchg_futex_value_locked(uaddr, uval, newval); + + if (unlikely(curval == -EFAULT)) + return -EFAULT; + if (unlikely(curval != uval)) + goto retry; + + /* + * We took the lock due to owner died take over. + */ + if (unlikely(lock_taken)) + return 1; + + /* + * We dont have the lock. Look up the PI state (or create it if + * we are the first waiter): + */ + ret = lookup_pi_state(uval, hb, key, ps); + + if (unlikely(ret)) { + switch (ret) { + case -ESRCH: + /* + * No owner found for this futex. Check if the + * OWNER_DIED bit is set to figure out whether + * this is a robust futex or not. + */ + if (get_futex_value_locked(&curval, uaddr)) + return -EFAULT; + + /* + * We simply start over in case of a robust + * futex. The code above will take the futex + * and return happy. + */ + if (curval & FUTEX_OWNER_DIED) { + ownerdied = 1; + goto retry; + } + default: + break; + } + } + + return ret; +} + /* * The hash bucket lock must be held when this is called. * Afterwards, the futex_q must not be accessed. @@ -1340,9 +1461,9 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, struct hrtimer_sleeper timeout, *to = NULL; struct task_struct *curr = current; struct futex_hash_bucket *hb; - u32 uval, newval, curval; + u32 uval; struct futex_q q; - int ret, lock_taken, ownerdied = 0; + int ret; if (refill_pi_state_cache()) return -ENOMEM; @@ -1365,81 +1486,15 @@ retry: retry_private: hb = queue_lock(&q); -retry_locked: - ret = lock_taken = 0; - - /* - * To avoid races, we attempt to take the lock here again - * (by doing a 0 -> TID atomic cmpxchg), while holding all - * the locks. It will most likely not succeed. - */ - newval = task_pid_vnr(current); - - curval = cmpxchg_futex_value_locked(uaddr, 0, newval); - - if (unlikely(curval == -EFAULT)) - goto uaddr_faulted; - - /* - * Detect deadlocks. In case of REQUEUE_PI this is a valid - * situation and we return success to user space. - */ - if (unlikely((curval & FUTEX_TID_MASK) == task_pid_vnr(current))) { - ret = -EDEADLK; - goto out_unlock_put_key; - } - - /* - * Surprise - we got the lock. Just return to userspace: - */ - if (unlikely(!curval)) - goto out_unlock_put_key; - - uval = curval; - - /* - * Set the WAITERS flag, so the owner will know it has someone - * to wake at next unlock - */ - newval = curval | FUTEX_WAITERS; - - /* - * There are two cases, where a futex might have no owner (the - * owner TID is 0): OWNER_DIED. We take over the futex in this - * case. We also do an unconditional take over, when the owner - * of the futex died. - * - * This is safe as we are protected by the hash bucket lock ! - */ - if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { - /* Keep the OWNER_DIED bit */ - newval = (curval & ~FUTEX_TID_MASK) | task_pid_vnr(current); - ownerdied = 0; - lock_taken = 1; - } - - curval = cmpxchg_futex_value_locked(uaddr, uval, newval); - - if (unlikely(curval == -EFAULT)) - goto uaddr_faulted; - if (unlikely(curval != uval)) - goto retry_locked; - - /* - * We took the lock due to owner died take over. - */ - if (unlikely(lock_taken)) - goto out_unlock_put_key; - - /* - * We dont have the lock. Look up the PI state (or create it if - * we are the first waiter): - */ - ret = lookup_pi_state(uval, hb, &q.key, &q.pi_state); - + ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current); if (unlikely(ret)) { switch (ret) { - + case 1: + /* We got the lock. */ + ret = 0; + goto out_unlock_put_key; + case -EFAULT: + goto uaddr_faulted; case -EAGAIN: /* * Task is exiting and we just wait for the @@ -1449,25 +1504,6 @@ retry_locked: put_futex_key(fshared, &q.key); cond_resched(); goto retry; - - case -ESRCH: - /* - * No owner found for this futex. Check if the - * OWNER_DIED bit is set to figure out whether - * this is a robust futex or not. - */ - if (get_futex_value_locked(&curval, uaddr)) - goto uaddr_faulted; - - /* - * We simply start over in case of a robust - * futex. The code above will take the futex - * and return happy. - */ - if (curval & FUTEX_OWNER_DIED) { - ownerdied = 1; - goto retry_locked; - } default: goto out_unlock_put_key; } -- cgit v1.2.3 From dd9739980b50c8cde33e1f8eb08b7e0140bcd61e Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:40:02 -0700 Subject: futex: split out fixup owner logic from futex_lock_pi() Refactor the post lock acquisition logic from futex_lock_pi(). This code will be reused in futex_wait_requeue_pi(). Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 158 ++++++++++++++++++++++++++++++++------------------------- 1 file changed, 89 insertions(+), 69 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 986b16e44534..af831fbb7fb4 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1255,6 +1255,79 @@ handle_fault: static long futex_wait_restart(struct restart_block *restart); +/** + * fixup_owner() - Post lock pi_state and corner case management + * @uaddr: user address of the futex + * @fshared: whether the futex is shared (1) or not (0) + * @q: futex_q (contains pi_state and access to the rt_mutex) + * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) + * + * After attempting to lock an rt_mutex, this function is called to cleanup + * the pi_state owner as well as handle race conditions that may allow us to + * acquire the lock. Must be called with the hb lock held. + * + * Returns: + * 1 - success, lock taken + * 0 - success, lock not taken + * <0 - on error (-EFAULT) + */ +static int fixup_owner(u32 __user *uaddr, int fshared, struct futex_q *q, + int locked) +{ + struct task_struct *owner; + int ret = 0; + + if (locked) { + /* + * Got the lock. We might not be the anticipated owner if we + * did a lock-steal - fix up the PI-state in that case: + */ + if (q->pi_state->owner != current) + ret = fixup_pi_state_owner(uaddr, q, current, fshared); + goto out; + } + + /* + * Catch the rare case, where the lock was released when we were on the + * way back before we locked the hash bucket. + */ + if (q->pi_state->owner == current) { + /* + * Try to get the rt_mutex now. This might fail as some other + * task acquired the rt_mutex after we removed ourself from the + * rt_mutex waiters list. + */ + if (rt_mutex_trylock(&q->pi_state->pi_mutex)) { + locked = 1; + goto out; + } + + /* + * pi_state is incorrect, some other task did a lock steal and + * we returned due to timeout or signal without taking the + * rt_mutex. Too late. We can access the rt_mutex_owner without + * locking, as the other task is now blocked on the hash bucket + * lock. Fix the state up. + */ + owner = rt_mutex_owner(&q->pi_state->pi_mutex); + ret = fixup_pi_state_owner(uaddr, q, owner, fshared); + goto out; + } + + /* + * Paranoia check. If we did not take the lock, then we should not be + * the owner, nor the pending owner, of the rt_mutex. + */ + if (rt_mutex_owner(&q->pi_state->pi_mutex) == current) + printk(KERN_ERR "fixup_owner: ret = %d pi-mutex: %p " + "pi-state %p\n", ret, + q->pi_state->pi_mutex.owner, + q->pi_state->owner); + +out: + return ret ? ret : locked; +} + /** * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal * @hb: the futex hash bucket, must be locked by the caller @@ -1459,11 +1532,10 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, int detect, ktime_t *time, int trylock) { struct hrtimer_sleeper timeout, *to = NULL; - struct task_struct *curr = current; struct futex_hash_bucket *hb; u32 uval; struct futex_q q; - int ret; + int res, ret; if (refill_pi_state_cache()) return -ENOMEM; @@ -1527,71 +1599,21 @@ retry_private: } spin_lock(q.lock_ptr); - - if (!ret) { - /* - * Got the lock. We might not be the anticipated owner - * if we did a lock-steal - fix up the PI-state in - * that case: - */ - if (q.pi_state->owner != curr) - ret = fixup_pi_state_owner(uaddr, &q, curr, fshared); - } else { - /* - * Catch the rare case, where the lock was released - * when we were on the way back before we locked the - * hash bucket. - */ - if (q.pi_state->owner == curr) { - /* - * Try to get the rt_mutex now. This might - * fail as some other task acquired the - * rt_mutex after we removed ourself from the - * rt_mutex waiters list. - */ - if (rt_mutex_trylock(&q.pi_state->pi_mutex)) - ret = 0; - else { - /* - * pi_state is incorrect, some other - * task did a lock steal and we - * returned due to timeout or signal - * without taking the rt_mutex. Too - * late. We can access the - * rt_mutex_owner without locking, as - * the other task is now blocked on - * the hash bucket lock. Fix the state - * up. - */ - struct task_struct *owner; - int res; - - owner = rt_mutex_owner(&q.pi_state->pi_mutex); - res = fixup_pi_state_owner(uaddr, &q, owner, - fshared); - - /* propagate -EFAULT, if the fixup failed */ - if (res) - ret = res; - } - } else { - /* - * Paranoia check. If we did not take the lock - * in the trylock above, then we should not be - * the owner of the rtmutex, neither the real - * nor the pending one: - */ - if (rt_mutex_owner(&q.pi_state->pi_mutex) == curr) - printk(KERN_ERR "futex_lock_pi: ret = %d " - "pi-mutex: %p pi-state %p\n", ret, - q.pi_state->pi_mutex.owner, - q.pi_state->owner); - } - } + /* + * Fixup the pi_state owner and possibly acquire the lock if we + * haven't already. + */ + res = fixup_owner(uaddr, fshared, &q, !ret); + /* + * If fixup_owner() returned an error, proprogate that. If it acquired + * the lock, clear our -ETIMEDOUT or -EINTR. + */ + if (res) + ret = (res < 0) ? res : 0; /* - * If fixup_pi_state_owner() faulted and was unable to handle the - * fault, unlock it and return the fault to userspace. + * If fixup_owner() faulted and was unable to handle the fault, unlock + * it and return the fault to userspace. */ if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) rt_mutex_unlock(&q.pi_state->pi_mutex); @@ -1599,9 +1621,7 @@ retry_private: /* Unqueue and drop the lock */ unqueue_me_pi(&q); - if (to) - destroy_hrtimer_on_stack(&to->timer); - return ret != -EINTR ? ret : -ERESTARTNOINTR; + goto out; out_unlock_put_key: queue_unlock(&q, hb); @@ -1611,7 +1631,7 @@ out_put_key: out: if (to) destroy_hrtimer_on_stack(&to->timer); - return ret; + return ret != -EINTR ? ret : -ERESTARTNOINTR; uaddr_faulted: /* -- cgit v1.2.3 From 8dac456a681bd94272ff50ecb31be6b669382c2b Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:40:12 -0700 Subject: rt_mutex: add proxy lock routines This patch is a prerequisite for futex requeue_pi. It basically splits rt_mutex_slowlock() right down the middle, just before the first call to schedule(). It further adds helper functions which make use of the split and provide the rt-mutex preliminaries for futex requeue_pi. Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/rtmutex.c | 240 +++++++++++++++++++++++++++++++++++++----------- kernel/rtmutex_common.h | 8 ++ 2 files changed, 195 insertions(+), 53 deletions(-) (limited to 'kernel') diff --git a/kernel/rtmutex.c b/kernel/rtmutex.c index 69d9cb921ffa..fec77e7e0562 100644 --- a/kernel/rtmutex.c +++ b/kernel/rtmutex.c @@ -300,7 +300,8 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task, * assigned pending owner [which might not have taken the * lock yet]: */ -static inline int try_to_steal_lock(struct rt_mutex *lock) +static inline int try_to_steal_lock(struct rt_mutex *lock, + struct task_struct *task) { struct task_struct *pendowner = rt_mutex_owner(lock); struct rt_mutex_waiter *next; @@ -309,11 +310,11 @@ static inline int try_to_steal_lock(struct rt_mutex *lock) if (!rt_mutex_owner_pending(lock)) return 0; - if (pendowner == current) + if (pendowner == task) return 1; spin_lock_irqsave(&pendowner->pi_lock, flags); - if (current->prio >= pendowner->prio) { + if (task->prio >= pendowner->prio) { spin_unlock_irqrestore(&pendowner->pi_lock, flags); return 0; } @@ -338,21 +339,21 @@ static inline int try_to_steal_lock(struct rt_mutex *lock) * We are going to steal the lock and a waiter was * enqueued on the pending owners pi_waiters queue. So * we have to enqueue this waiter into - * current->pi_waiters list. This covers the case, - * where current is boosted because it holds another + * task->pi_waiters list. This covers the case, + * where task is boosted because it holds another * lock and gets unboosted because the booster is * interrupted, so we would delay a waiter with higher - * priority as current->normal_prio. + * priority as task->normal_prio. * * Note: in the rare case of a SCHED_OTHER task changing * its priority and thus stealing the lock, next->task - * might be current: + * might be task: */ - if (likely(next->task != current)) { - spin_lock_irqsave(¤t->pi_lock, flags); - plist_add(&next->pi_list_entry, ¤t->pi_waiters); - __rt_mutex_adjust_prio(current); - spin_unlock_irqrestore(¤t->pi_lock, flags); + if (likely(next->task != task)) { + spin_lock_irqsave(&task->pi_lock, flags); + plist_add(&next->pi_list_entry, &task->pi_waiters); + __rt_mutex_adjust_prio(task); + spin_unlock_irqrestore(&task->pi_lock, flags); } return 1; } @@ -389,7 +390,7 @@ static int try_to_take_rt_mutex(struct rt_mutex *lock) */ mark_rt_mutex_waiters(lock); - if (rt_mutex_owner(lock) && !try_to_steal_lock(lock)) + if (rt_mutex_owner(lock) && !try_to_steal_lock(lock, current)) return 0; /* We got the lock. */ @@ -411,6 +412,7 @@ static int try_to_take_rt_mutex(struct rt_mutex *lock) */ static int task_blocks_on_rt_mutex(struct rt_mutex *lock, struct rt_mutex_waiter *waiter, + struct task_struct *task, int detect_deadlock) { struct task_struct *owner = rt_mutex_owner(lock); @@ -418,21 +420,21 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock, unsigned long flags; int chain_walk = 0, res; - spin_lock_irqsave(¤t->pi_lock, flags); - __rt_mutex_adjust_prio(current); - waiter->task = current; + spin_lock_irqsave(&task->pi_lock, flags); + __rt_mutex_adjust_prio(task); + waiter->task = task; waiter->lock = lock; - plist_node_init(&waiter->list_entry, current->prio); - plist_node_init(&waiter->pi_list_entry, current->prio); + plist_node_init(&waiter->list_entry, task->prio); + plist_node_init(&waiter->pi_list_entry, task->prio); /* Get the top priority waiter on the lock */ if (rt_mutex_has_waiters(lock)) top_waiter = rt_mutex_top_waiter(lock); plist_add(&waiter->list_entry, &lock->wait_list); - current->pi_blocked_on = waiter; + task->pi_blocked_on = waiter; - spin_unlock_irqrestore(¤t->pi_lock, flags); + spin_unlock_irqrestore(&task->pi_lock, flags); if (waiter == rt_mutex_top_waiter(lock)) { spin_lock_irqsave(&owner->pi_lock, flags); @@ -460,7 +462,7 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock, spin_unlock(&lock->wait_lock); res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock, waiter, - current); + task); spin_lock(&lock->wait_lock); @@ -605,37 +607,25 @@ void rt_mutex_adjust_pi(struct task_struct *task) rt_mutex_adjust_prio_chain(task, 0, NULL, NULL, task); } -/* - * Slow path lock function: +/** + * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop + * @lock: the rt_mutex to take + * @state: the state the task should block in (TASK_INTERRUPTIBLE + * or TASK_UNINTERRUPTIBLE) + * @timeout: the pre-initialized and started timer, or NULL for none + * @waiter: the pre-initialized rt_mutex_waiter + * @detect_deadlock: passed to task_blocks_on_rt_mutex + * + * lock->wait_lock must be held by the caller. */ static int __sched -rt_mutex_slowlock(struct rt_mutex *lock, int state, - struct hrtimer_sleeper *timeout, - int detect_deadlock) +__rt_mutex_slowlock(struct rt_mutex *lock, int state, + struct hrtimer_sleeper *timeout, + struct rt_mutex_waiter *waiter, + int detect_deadlock) { - struct rt_mutex_waiter waiter; int ret = 0; - debug_rt_mutex_init_waiter(&waiter); - waiter.task = NULL; - - spin_lock(&lock->wait_lock); - - /* Try to acquire the lock again: */ - if (try_to_take_rt_mutex(lock)) { - spin_unlock(&lock->wait_lock); - return 0; - } - - set_current_state(state); - - /* Setup the timer, when timeout != NULL */ - if (unlikely(timeout)) { - hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); - if (!hrtimer_active(&timeout->timer)) - timeout->task = NULL; - } - for (;;) { /* Try to acquire the lock: */ if (try_to_take_rt_mutex(lock)) @@ -656,19 +646,19 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state, } /* - * waiter.task is NULL the first time we come here and + * waiter->task is NULL the first time we come here and * when we have been woken up by the previous owner * but the lock got stolen by a higher prio task. */ - if (!waiter.task) { - ret = task_blocks_on_rt_mutex(lock, &waiter, + if (!waiter->task) { + ret = task_blocks_on_rt_mutex(lock, waiter, current, detect_deadlock); /* * If we got woken up by the owner then start loop * all over without going into schedule to try * to get the lock now: */ - if (unlikely(!waiter.task)) { + if (unlikely(!waiter->task)) { /* * Reset the return value. We might * have returned with -EDEADLK and the @@ -684,15 +674,52 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state, spin_unlock(&lock->wait_lock); - debug_rt_mutex_print_deadlock(&waiter); + debug_rt_mutex_print_deadlock(waiter); - if (waiter.task) + if (waiter->task) schedule_rt_mutex(lock); spin_lock(&lock->wait_lock); set_current_state(state); } + return ret; +} + +/* + * Slow path lock function: + */ +static int __sched +rt_mutex_slowlock(struct rt_mutex *lock, int state, + struct hrtimer_sleeper *timeout, + int detect_deadlock) +{ + struct rt_mutex_waiter waiter; + int ret = 0; + + debug_rt_mutex_init_waiter(&waiter); + waiter.task = NULL; + + spin_lock(&lock->wait_lock); + + /* Try to acquire the lock again: */ + if (try_to_take_rt_mutex(lock)) { + spin_unlock(&lock->wait_lock); + return 0; + } + + set_current_state(state); + + /* Setup the timer, when timeout != NULL */ + if (unlikely(timeout)) { + hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); + if (!hrtimer_active(&timeout->timer)) + timeout->task = NULL; + } + + ret = __rt_mutex_slowlock(lock, state, timeout, &waiter, + detect_deadlock); + set_current_state(TASK_RUNNING); if (unlikely(waiter.task)) @@ -985,6 +1012,59 @@ void rt_mutex_proxy_unlock(struct rt_mutex *lock, rt_mutex_deadlock_account_unlock(proxy_owner); } +/** + * rt_mutex_start_proxy_lock() - Start lock acquisition for another task + * @lock: the rt_mutex to take + * @waiter: the pre-initialized rt_mutex_waiter + * @task: the task to prepare + * @detect_deadlock: perform deadlock detection (1) or not (0) + * + * Returns: + * 0 - task blocked on lock + * 1 - acquired the lock for task, caller should wake it up + * <0 - error + * + * Special API call for FUTEX_REQUEUE_PI support. + */ +int rt_mutex_start_proxy_lock(struct rt_mutex *lock, + struct rt_mutex_waiter *waiter, + struct task_struct *task, int detect_deadlock) +{ + int ret; + + spin_lock(&lock->wait_lock); + + mark_rt_mutex_waiters(lock); + + if (!rt_mutex_owner(lock) || try_to_steal_lock(lock, task)) { + /* We got the lock for task. */ + debug_rt_mutex_lock(lock); + + rt_mutex_set_owner(lock, task, 0); + + rt_mutex_deadlock_account_lock(lock, task); + return 1; + } + + ret = task_blocks_on_rt_mutex(lock, waiter, task, detect_deadlock); + + + if (ret && !waiter->task) { + /* + * Reset the return value. We might have + * returned with -EDEADLK and the owner + * released the lock while we were walking the + * pi chain. Let the waiter sort it out. + */ + ret = 0; + } + spin_unlock(&lock->wait_lock); + + debug_rt_mutex_print_deadlock(waiter); + + return ret; +} + /** * rt_mutex_next_owner - return the next owner of the lock * @@ -1004,3 +1084,57 @@ struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock) return rt_mutex_top_waiter(lock)->task; } + +/** + * rt_mutex_finish_proxy_lock() - Complete lock acquisition + * @lock: the rt_mutex we were woken on + * @to: the timeout, null if none. hrtimer should already have + * been started. + * @waiter: the pre-initialized rt_mutex_waiter + * @detect_deadlock: perform deadlock detection (1) or not (0) + * + * Complete the lock acquisition started our behalf by another thread. + * + * Returns: + * 0 - success + * <0 - error, one of -EINTR, -ETIMEDOUT, or -EDEADLK + * + * Special API call for PI-futex requeue support + */ +int rt_mutex_finish_proxy_lock(struct rt_mutex *lock, + struct hrtimer_sleeper *to, + struct rt_mutex_waiter *waiter, + int detect_deadlock) +{ + int ret; + + spin_lock(&lock->wait_lock); + + set_current_state(TASK_INTERRUPTIBLE); + + ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter, + detect_deadlock); + + set_current_state(TASK_RUNNING); + + if (unlikely(waiter->task)) + remove_waiter(lock, waiter); + + /* + * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might + * have to fix that up. + */ + fixup_rt_mutex_waiters(lock); + + spin_unlock(&lock->wait_lock); + + /* + * Readjust priority, when we did not get the lock. We might have been + * the pending owner and boosted. Since we did not take the lock, the + * PI boost has to go. + */ + if (unlikely(ret)) + rt_mutex_adjust_prio(current); + + return ret; +} diff --git a/kernel/rtmutex_common.h b/kernel/rtmutex_common.h index e124bf5800ea..97a2f81866af 100644 --- a/kernel/rtmutex_common.h +++ b/kernel/rtmutex_common.h @@ -120,6 +120,14 @@ extern void rt_mutex_init_proxy_locked(struct rt_mutex *lock, struct task_struct *proxy_owner); extern void rt_mutex_proxy_unlock(struct rt_mutex *lock, struct task_struct *proxy_owner); +extern int rt_mutex_start_proxy_lock(struct rt_mutex *lock, + struct rt_mutex_waiter *waiter, + struct task_struct *task, + int detect_deadlock); +extern int rt_mutex_finish_proxy_lock(struct rt_mutex *lock, + struct hrtimer_sleeper *to, + struct rt_mutex_waiter *waiter, + int detect_deadlock); #ifdef CONFIG_DEBUG_RT_MUTEXES # include "rtmutex-debug.h" -- cgit v1.2.3 From a72188d8a64ebe74722f1cf7ffac41b41ffdba21 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:40:22 -0700 Subject: futex: add FUTEX_HAS_TIMEOUT flag to restart.futex.flags Currently restart is only used if there is a timeout. The requeue_pi functionality requires restarting to futex_lock_pi() on signal after wakeup in futex_wait_requeue_pi() regardless of if there was a timeout or not. Using 0 for the timeout value is confusing as that could indicate an expired timer. The flag makes this explicit. While the check is not technically needed in futex_wait_restart(), doing so makes the code consistent with and will avoid confusion should the need arise to restart wait without a timeout. Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index af831fbb7fb4..6b597cf33b02 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1252,6 +1252,7 @@ handle_fault: */ #define FLAGS_SHARED 0x01 #define FLAGS_CLOCKRT 0x02 +#define FLAGS_HAS_TIMEOUT 0x04 static long futex_wait_restart(struct restart_block *restart); @@ -1486,7 +1487,7 @@ retry_private: restart->futex.val = val; restart->futex.time = abs_time->tv64; restart->futex.bitset = bitset; - restart->futex.flags = 0; + restart->futex.flags = FLAGS_HAS_TIMEOUT; if (fshared) restart->futex.flags |= FLAGS_SHARED; @@ -1510,13 +1511,16 @@ static long futex_wait_restart(struct restart_block *restart) { u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; int fshared = 0; - ktime_t t; + ktime_t t, *tp = NULL; - t.tv64 = restart->futex.time; + if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { + t.tv64 = restart->futex.time; + tp = &t; + } restart->fn = do_no_restart_syscall; if (restart->futex.flags & FLAGS_SHARED) fshared = 1; - return (long)futex_wait(uaddr, fshared, restart->futex.val, &t, + return (long)futex_wait(uaddr, fshared, restart->futex.val, tp, restart->futex.bitset, restart->futex.flags & FLAGS_CLOCKRT); } -- cgit v1.2.3 From 9121e4783cd5c7e2a407763f3b61c2d573891133 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:40:31 -0700 Subject: futex: distangle futex_requeue() futex_requeue() is getting a bit long-winded, and will be getting more so after the requeue_pi patch. Factor out the actual requeueing into a nicely contained inline function to reduce function length and improve legibility. Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 43 +++++++++++++++++++++++++++++-------------- 1 file changed, 29 insertions(+), 14 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 6b597cf33b02..e76942e2a79f 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -940,6 +940,34 @@ out: return ret; } +/** + * requeue_futex() - Requeue a futex_q from one hb to another + * @q: the futex_q to requeue + * @hb1: the source hash_bucket + * @hb2: the target hash_bucket + * @key2: the new key for the requeued futex_q + */ +static inline +void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, + struct futex_hash_bucket *hb2, union futex_key *key2) +{ + + /* + * If key1 and key2 hash to the same bucket, no need to + * requeue. + */ + if (likely(&hb1->chain != &hb2->chain)) { + plist_del(&q->list, &hb1->chain); + plist_add(&q->list, &hb2->chain); + q->lock_ptr = &hb2->lock; +#ifdef CONFIG_DEBUG_PI_LIST + q->list.plist.lock = &hb2->lock; +#endif + } + get_futex_key_refs(key2); + q->key = *key2; +} + /* * Requeue all waiters hashed on one physical page to another * physical page. @@ -999,20 +1027,7 @@ retry_private: if (++ret <= nr_wake) { wake_futex(this); } else { - /* - * If key1 and key2 hash to the same bucket, no need to - * requeue. - */ - if (likely(head1 != &hb2->chain)) { - plist_del(&this->list, &hb1->chain); - plist_add(&this->list, &hb2->chain); - this->lock_ptr = &hb2->lock; -#ifdef CONFIG_DEBUG_PI_LIST - this->list.plist.lock = &hb2->lock; -#endif - } - this->key = key2; - get_futex_key_refs(&key2); + requeue_futex(this, hb1, hb2, &key2); drop_count++; if (ret - nr_wake >= nr_requeue) -- cgit v1.2.3 From f801073f87aa22ddf0e9146355fec3993163790f Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:40:40 -0700 Subject: futex: split out futex value validation code Refactor the code to validate the expected futex value in order to reuse it with the requeue_pi code. Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 116 +++++++++++++++++++++++++++++++++++---------------------- 1 file changed, 72 insertions(+), 44 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index e76942e2a79f..dbe857aa4381 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1398,42 +1398,29 @@ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, __set_current_state(TASK_RUNNING); } -static int futex_wait(u32 __user *uaddr, int fshared, - u32 val, ktime_t *abs_time, u32 bitset, int clockrt) +/** + * futex_wait_setup() - Prepare to wait on a futex + * @uaddr: the futex userspace address + * @val: the expected value + * @fshared: whether the futex is shared (1) or not (0) + * @q: the associated futex_q + * @hb: storage for hash_bucket pointer to be returned to caller + * + * Setup the futex_q and locate the hash_bucket. Get the futex value and + * compare it with the expected value. Handle atomic faults internally. + * Return with the hb lock held and a q.key reference on success, and unlocked + * with no q.key reference on failure. + * + * Returns: + * 0 - uaddr contains val and hb has been locked + * <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlcoked + */ +static int futex_wait_setup(u32 __user *uaddr, u32 val, int fshared, + struct futex_q *q, struct futex_hash_bucket **hb) { - struct hrtimer_sleeper timeout, *to = NULL; - DECLARE_WAITQUEUE(wait, current); - struct restart_block *restart; - struct futex_hash_bucket *hb; - struct futex_q q; u32 uval; int ret; - if (!bitset) - return -EINVAL; - - q.pi_state = NULL; - q.bitset = bitset; - - if (abs_time) { - to = &timeout; - - hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : - CLOCK_MONOTONIC, HRTIMER_MODE_ABS); - hrtimer_init_sleeper(to, current); - hrtimer_set_expires_range_ns(&to->timer, *abs_time, - current->timer_slack_ns); - } - -retry: - q.key = FUTEX_KEY_INIT; - ret = get_futex_key(uaddr, fshared, &q.key); - if (unlikely(ret != 0)) - goto out; - -retry_private: - hb = queue_lock(&q); - /* * Access the page AFTER the hash-bucket is locked. * Order is important: @@ -1450,33 +1437,74 @@ retry_private: * A consequence is that futex_wait() can return zero and absorb * a wakeup when *uaddr != val on entry to the syscall. This is * rare, but normal. - * - * For shared futexes, we hold the mmap semaphore, so the mapping - * cannot have changed since we looked it up in get_futex_key. */ +retry: + q->key = FUTEX_KEY_INIT; + ret = get_futex_key(uaddr, fshared, &q->key); + if (unlikely(ret != 0)) + goto out; + +retry_private: + *hb = queue_lock(q); + ret = get_futex_value_locked(&uval, uaddr); - if (unlikely(ret)) { - queue_unlock(&q, hb); + if (ret) { + queue_unlock(q, *hb); ret = get_user(uval, uaddr); if (ret) - goto out_put_key; + goto out; if (!fshared) goto retry_private; - put_futex_key(fshared, &q.key); + put_futex_key(fshared, &q->key); goto retry; } - ret = -EWOULDBLOCK; - /* Only actually queue if *uaddr contained val. */ - if (unlikely(uval != val)) { - queue_unlock(&q, hb); - goto out_put_key; + if (uval != val) { + queue_unlock(q, *hb); + ret = -EWOULDBLOCK; } +out: + if (ret) + put_futex_key(fshared, &q->key); + return ret; +} + +static int futex_wait(u32 __user *uaddr, int fshared, + u32 val, ktime_t *abs_time, u32 bitset, int clockrt) +{ + struct hrtimer_sleeper timeout, *to = NULL; + DECLARE_WAITQUEUE(wait, current); + struct restart_block *restart; + struct futex_hash_bucket *hb; + struct futex_q q; + int ret; + + if (!bitset) + return -EINVAL; + + q.pi_state = NULL; + q.bitset = bitset; + + if (abs_time) { + to = &timeout; + + hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : + CLOCK_MONOTONIC, HRTIMER_MODE_ABS); + hrtimer_init_sleeper(to, current); + hrtimer_set_expires_range_ns(&to->timer, *abs_time, + current->timer_slack_ns); + } + + /* Prepare to wait on uaddr. */ + ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); + if (ret) + goto out; + /* queue_me and wait for wakeup, timeout, or a signal. */ futex_wait_queue_me(hb, &q, to, &wait); -- cgit v1.2.3 From 52400ba946759af28442dee6265c5c0180ac7122 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 3 Apr 2009 13:40:49 -0700 Subject: futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart Reviewed-by: Thomas Gleixner Signed-off-by: Thomas Gleixner --- kernel/futex.c | 519 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 500 insertions(+), 19 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index dbe857aa4381..185c981d89e3 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -19,6 +19,10 @@ * PRIVATE futexes by Eric Dumazet * Copyright (C) 2007 Eric Dumazet * + * Requeue-PI support by Darren Hart + * Copyright (C) IBM Corporation, 2009 + * Thanks to Thomas Gleixner for conceptual design and careful reviews. + * * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly * enough at me, Linus for the original (flawed) idea, Matthew * Kirkwood for proof-of-concept implementation. @@ -109,6 +113,9 @@ struct futex_q { struct futex_pi_state *pi_state; struct task_struct *task; + /* rt_waiter storage for requeue_pi: */ + struct rt_mutex_waiter *rt_waiter; + /* Bitset for the optional bitmasked wakeup */ u32 bitset; }; @@ -827,7 +834,7 @@ static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) plist_for_each_entry_safe(this, next, head, list) { if (match_futex (&this->key, &key)) { - if (this->pi_state) { + if (this->pi_state || this->rt_waiter) { ret = -EINVAL; break; } @@ -968,20 +975,138 @@ void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, q->key = *key2; } -/* - * Requeue all waiters hashed on one physical page to another - * physical page. +/** + * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue + * q: the futex_q + * key: the key of the requeue target futex + * + * During futex_requeue, with requeue_pi=1, it is possible to acquire the + * target futex if it is uncontended or via a lock steal. Set the futex_q key + * to the requeue target futex so the waiter can detect the wakeup on the right + * futex, but remove it from the hb and NULL the rt_waiter so it can detect + * atomic lock acquisition. Must be called with the q->lock_ptr held. + */ +static inline +void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key) +{ + drop_futex_key_refs(&q->key); + get_futex_key_refs(key); + q->key = *key; + + WARN_ON(plist_node_empty(&q->list)); + plist_del(&q->list, &q->list.plist); + + WARN_ON(!q->rt_waiter); + q->rt_waiter = NULL; + + wake_up(&q->waiter); +} + +/** + * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter + * @pifutex: the user address of the to futex + * @hb1: the from futex hash bucket, must be locked by the caller + * @hb2: the to futex hash bucket, must be locked by the caller + * @key1: the from futex key + * @key2: the to futex key + * + * Try and get the lock on behalf of the top waiter if we can do it atomically. + * Wake the top waiter if we succeed. hb1 and hb2 must be held by the caller. + * + * Returns: + * 0 - failed to acquire the lock atomicly + * 1 - acquired the lock + * <0 - error + */ +static int futex_proxy_trylock_atomic(u32 __user *pifutex, + struct futex_hash_bucket *hb1, + struct futex_hash_bucket *hb2, + union futex_key *key1, union futex_key *key2, + struct futex_pi_state **ps) +{ + struct futex_q *top_waiter; + u32 curval; + int ret; + + if (get_futex_value_locked(&curval, pifutex)) + return -EFAULT; + + top_waiter = futex_top_waiter(hb1, key1); + + /* There are no waiters, nothing for us to do. */ + if (!top_waiter) + return 0; + + /* + * Either take the lock for top_waiter or set the FUTEX_WAITERS bit. + * The pi_state is returned in ps in contended cases. + */ + ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task); + if (ret == 1) + requeue_pi_wake_futex(top_waiter, key2); + + return ret; +} + +/** + * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 + * uaddr1: source futex user address + * uaddr2: target futex user address + * nr_wake: number of waiters to wake (must be 1 for requeue_pi) + * nr_requeue: number of waiters to requeue (0-INT_MAX) + * requeue_pi: if we are attempting to requeue from a non-pi futex to a + * pi futex (pi to pi requeue is not supported) + * + * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire + * uaddr2 atomically on behalf of the top waiter. + * + * Returns: + * >=0 - on success, the number of tasks requeued or woken + * <0 - on error */ static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, - int nr_wake, int nr_requeue, u32 *cmpval) + int nr_wake, int nr_requeue, u32 *cmpval, + int requeue_pi) { union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; + int drop_count = 0, task_count = 0, ret; + struct futex_pi_state *pi_state = NULL; struct futex_hash_bucket *hb1, *hb2; struct plist_head *head1; struct futex_q *this, *next; - int ret, drop_count = 0; + u32 curval2; + + if (requeue_pi) { + /* + * requeue_pi requires a pi_state, try to allocate it now + * without any locks in case it fails. + */ + if (refill_pi_state_cache()) + return -ENOMEM; + /* + * requeue_pi must wake as many tasks as it can, up to nr_wake + * + nr_requeue, since it acquires the rt_mutex prior to + * returning to userspace, so as to not leave the rt_mutex with + * waiters and no owner. However, second and third wake-ups + * cannot be predicted as they involve race conditions with the + * first wake and a fault while looking up the pi_state. Both + * pthread_cond_signal() and pthread_cond_broadcast() should + * use nr_wake=1. + */ + if (nr_wake != 1) + return -EINVAL; + } retry: + if (pi_state != NULL) { + /* + * We will have to lookup the pi_state again, so free this one + * to keep the accounting correct. + */ + free_pi_state(pi_state); + pi_state = NULL; + } + ret = get_futex_key(uaddr1, fshared, &key1); if (unlikely(ret != 0)) goto out; @@ -1020,19 +1145,94 @@ retry_private: } } + if (requeue_pi && (task_count - nr_wake < nr_requeue)) { + /* Attempt to acquire uaddr2 and wake the top_waiter. */ + ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, + &key2, &pi_state); + + /* + * At this point the top_waiter has either taken uaddr2 or is + * waiting on it. If the former, then the pi_state will not + * exist yet, look it up one more time to ensure we have a + * reference to it. + */ + if (ret == 1) { + WARN_ON(pi_state); + task_count++; + ret = get_futex_value_locked(&curval2, uaddr2); + if (!ret) + ret = lookup_pi_state(curval2, hb2, &key2, + &pi_state); + } + + switch (ret) { + case 0: + break; + case -EFAULT: + double_unlock_hb(hb1, hb2); + put_futex_key(fshared, &key2); + put_futex_key(fshared, &key1); + ret = get_user(curval2, uaddr2); + if (!ret) + goto retry; + goto out; + case -EAGAIN: + /* The owner was exiting, try again. */ + double_unlock_hb(hb1, hb2); + put_futex_key(fshared, &key2); + put_futex_key(fshared, &key1); + cond_resched(); + goto retry; + default: + goto out_unlock; + } + } + head1 = &hb1->chain; plist_for_each_entry_safe(this, next, head1, list) { - if (!match_futex (&this->key, &key1)) + if (task_count - nr_wake >= nr_requeue) + break; + + if (!match_futex(&this->key, &key1)) continue; - if (++ret <= nr_wake) { + + WARN_ON(!requeue_pi && this->rt_waiter); + WARN_ON(requeue_pi && !this->rt_waiter); + + /* + * Wake nr_wake waiters. For requeue_pi, if we acquired the + * lock, we already woke the top_waiter. If not, it will be + * woken by futex_unlock_pi(). + */ + if (++task_count <= nr_wake && !requeue_pi) { wake_futex(this); - } else { - requeue_futex(this, hb1, hb2, &key2); - drop_count++; + continue; + } - if (ret - nr_wake >= nr_requeue) - break; + /* + * Requeue nr_requeue waiters and possibly one more in the case + * of requeue_pi if we couldn't acquire the lock atomically. + */ + if (requeue_pi) { + /* Prepare the waiter to take the rt_mutex. */ + atomic_inc(&pi_state->refcount); + this->pi_state = pi_state; + ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, + this->rt_waiter, + this->task, 1); + if (ret == 1) { + /* We got the lock. */ + requeue_pi_wake_futex(this, &key2); + continue; + } else if (ret) { + /* -EDEADLK */ + this->pi_state = NULL; + free_pi_state(pi_state); + goto out_unlock; + } } + requeue_futex(this, hb1, hb2, &key2); + drop_count++; } out_unlock: @@ -1047,7 +1247,9 @@ out_put_keys: out_put_key1: put_futex_key(fshared, &key1); out: - return ret; + if (pi_state != NULL) + free_pi_state(pi_state); + return ret ? ret : task_count; } /* The key must be already stored in q->key. */ @@ -1270,6 +1472,7 @@ handle_fault: #define FLAGS_HAS_TIMEOUT 0x04 static long futex_wait_restart(struct restart_block *restart); +static long futex_lock_pi_restart(struct restart_block *restart); /** * fixup_owner() - Post lock pi_state and corner case management @@ -1489,6 +1692,7 @@ static int futex_wait(u32 __user *uaddr, int fshared, q.pi_state = NULL; q.bitset = bitset; + q.rt_waiter = NULL; if (abs_time) { to = &timeout; @@ -1596,6 +1800,7 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, } q.pi_state = NULL; + q.rt_waiter = NULL; retry: q.key = FUTEX_KEY_INIT; ret = get_futex_key(uaddr, fshared, &q.key); @@ -1701,6 +1906,20 @@ uaddr_faulted: goto retry; } +static long futex_lock_pi_restart(struct restart_block *restart) +{ + u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; + ktime_t t, *tp = NULL; + int fshared = restart->futex.flags & FLAGS_SHARED; + + if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { + t.tv64 = restart->futex.time; + tp = &t; + } + restart->fn = do_no_restart_syscall; + + return (long)futex_lock_pi(uaddr, fshared, restart->futex.val, tp, 0); +} /* * Userspace attempted a TID -> 0 atomic transition, and failed. @@ -1803,6 +2022,253 @@ pi_faulted: return ret; } +/** + * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex + * @hb: the hash_bucket futex_q was original enqueued on + * @q: the futex_q woken while waiting to be requeued + * @key2: the futex_key of the requeue target futex + * @timeout: the timeout associated with the wait (NULL if none) + * + * Detect if the task was woken on the initial futex as opposed to the requeue + * target futex. If so, determine if it was a timeout or a signal that caused + * the wakeup and return the appropriate error code to the caller. Must be + * called with the hb lock held. + * + * Returns + * 0 - no early wakeup detected + * <0 - -ETIMEDOUT or -ERESTARTSYS (FIXME: or ERESTARTNOINTR?) + */ +static inline +int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, + struct futex_q *q, union futex_key *key2, + struct hrtimer_sleeper *timeout) +{ + int ret = 0; + + /* + * With the hb lock held, we avoid races while we process the wakeup. + * We only need to hold hb (and not hb2) to ensure atomicity as the + * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. + * It can't be requeued from uaddr2 to something else since we don't + * support a PI aware source futex for requeue. + */ + if (!match_futex(&q->key, key2)) { + WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); + /* + * We were woken prior to requeue by a timeout or a signal. + * Unqueue the futex_q and determine which it was. + */ + plist_del(&q->list, &q->list.plist); + drop_futex_key_refs(&q->key); + + if (timeout && !timeout->task) + ret = -ETIMEDOUT; + else { + /* + * We expect signal_pending(current), but another + * thread may have handled it for us already. + */ + /* FIXME: ERESTARTSYS or ERESTARTNOINTR? Do we care if + * the user specified SA_RESTART or not? */ + ret = -ERESTARTSYS; + } + } + return ret; +} + +/** + * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 + * @uaddr: the futex we initialyl wait on (non-pi) + * @fshared: whether the futexes are shared (1) or not (0). They must be + * the same type, no requeueing from private to shared, etc. + * @val: the expected value of uaddr + * @abs_time: absolute timeout + * @bitset: 32 bit wakeup bitset set by userspace, defaults to all. + * @clockrt: whether to use CLOCK_REALTIME (1) or CLOCK_MONOTONIC (0) + * @uaddr2: the pi futex we will take prior to returning to user-space + * + * The caller will wait on uaddr and will be requeued by futex_requeue() to + * uaddr2 which must be PI aware. Normal wakeup will wake on uaddr2 and + * complete the acquisition of the rt_mutex prior to returning to userspace. + * This ensures the rt_mutex maintains an owner when it has waiters; without + * one, the pi logic wouldn't know which task to boost/deboost, if there was a + * need to. + * + * We call schedule in futex_wait_queue_me() when we enqueue and return there + * via the following: + * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() + * 2) wakeup on uaddr2 after a requeue and subsequent unlock + * 3) signal (before or after requeue) + * 4) timeout (before or after requeue) + * + * If 3, we setup a restart_block with futex_wait_requeue_pi() as the function. + * + * If 2, we may then block on trying to take the rt_mutex and return via: + * 5) successful lock + * 6) signal + * 7) timeout + * 8) other lock acquisition failure + * + * If 6, we setup a restart_block with futex_lock_pi() as the function. + * + * If 4 or 7, we cleanup and return with -ETIMEDOUT. + * + * Returns: + * 0 - On success + * <0 - On error + */ +static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, + u32 val, ktime_t *abs_time, u32 bitset, + int clockrt, u32 __user *uaddr2) +{ + struct hrtimer_sleeper timeout, *to = NULL; + struct rt_mutex_waiter rt_waiter; + struct rt_mutex *pi_mutex = NULL; + DECLARE_WAITQUEUE(wait, current); + struct restart_block *restart; + struct futex_hash_bucket *hb; + union futex_key key2; + struct futex_q q; + int res, ret; + u32 uval; + + if (!bitset) + return -EINVAL; + + if (abs_time) { + to = &timeout; + hrtimer_init_on_stack(&to->timer, clockrt ? CLOCK_REALTIME : + CLOCK_MONOTONIC, HRTIMER_MODE_ABS); + hrtimer_init_sleeper(to, current); + hrtimer_set_expires_range_ns(&to->timer, *abs_time, + current->timer_slack_ns); + } + + /* + * The waiter is allocated on our stack, manipulated by the requeue + * code while we sleep on uaddr. + */ + debug_rt_mutex_init_waiter(&rt_waiter); + rt_waiter.task = NULL; + + q.pi_state = NULL; + q.bitset = bitset; + q.rt_waiter = &rt_waiter; + + key2 = FUTEX_KEY_INIT; + ret = get_futex_key(uaddr2, fshared, &key2); + if (unlikely(ret != 0)) + goto out; + + /* Prepare to wait on uaddr. */ + ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); + if (ret) { + put_futex_key(fshared, &key2); + goto out; + } + + /* Queue the futex_q, drop the hb lock, wait for wakeup. */ + futex_wait_queue_me(hb, &q, to, &wait); + + spin_lock(&hb->lock); + ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); + spin_unlock(&hb->lock); + if (ret) + goto out_put_keys; + + /* + * In order for us to be here, we know our q.key == key2, and since + * we took the hb->lock above, we also know that futex_requeue() has + * completed and we no longer have to concern ourselves with a wakeup + * race with the atomic proxy lock acquition by the requeue code. + */ + + /* Check if the requeue code acquired the second futex for us. */ + if (!q.rt_waiter) { + /* + * Got the lock. We might not be the anticipated owner if we + * did a lock-steal - fix up the PI-state in that case. + */ + if (q.pi_state && (q.pi_state->owner != current)) { + spin_lock(q.lock_ptr); + ret = fixup_pi_state_owner(uaddr2, &q, current, + fshared); + spin_unlock(q.lock_ptr); + } + } else { + /* + * We have been woken up by futex_unlock_pi(), a timeout, or a + * signal. futex_unlock_pi() will not destroy the lock_ptr nor + * the pi_state. + */ + WARN_ON(!&q.pi_state); + pi_mutex = &q.pi_state->pi_mutex; + ret = rt_mutex_finish_proxy_lock(pi_mutex, to, &rt_waiter, 1); + debug_rt_mutex_free_waiter(&rt_waiter); + + spin_lock(q.lock_ptr); + /* + * Fixup the pi_state owner and possibly acquire the lock if we + * haven't already. + */ + res = fixup_owner(uaddr2, fshared, &q, !ret); + /* + * If fixup_owner() returned an error, proprogate that. If it + * acquired the lock, clear our -ETIMEDOUT or -EINTR. + */ + if (res) + ret = (res < 0) ? res : 0; + + /* Unqueue and drop the lock. */ + unqueue_me_pi(&q); + } + + /* + * If fixup_pi_state_owner() faulted and was unable to handle the + * fault, unlock the rt_mutex and return the fault to userspace. + */ + if (ret == -EFAULT) { + if (rt_mutex_owner(pi_mutex) == current) + rt_mutex_unlock(pi_mutex); + } else if (ret == -EINTR) { + ret = -EFAULT; + if (get_user(uval, uaddr2)) + goto out_put_keys; + + /* + * We've already been requeued, so restart by calling + * futex_lock_pi() directly, rather then returning to this + * function. + */ + ret = -ERESTART_RESTARTBLOCK; + restart = ¤t_thread_info()->restart_block; + restart->fn = futex_lock_pi_restart; + restart->futex.uaddr = (u32 *)uaddr2; + restart->futex.val = uval; + restart->futex.flags = 0; + if (abs_time) { + restart->futex.flags |= FLAGS_HAS_TIMEOUT; + restart->futex.time = abs_time->tv64; + } + + if (fshared) + restart->futex.flags |= FLAGS_SHARED; + if (clockrt) + restart->futex.flags |= FLAGS_CLOCKRT; + } + +out_put_keys: + put_futex_key(fshared, &q.key); + put_futex_key(fshared, &key2); + +out: + if (to) { + hrtimer_cancel(&to->timer); + destroy_hrtimer_on_stack(&to->timer); + } + return ret; +} + /* * Support for robust futexes: the kernel cleans up held futexes at * thread exit time. @@ -2025,7 +2491,7 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, fshared = 1; clockrt = op & FUTEX_CLOCK_REALTIME; - if (clockrt && cmd != FUTEX_WAIT_BITSET) + if (clockrt && cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI) return -ENOSYS; switch (cmd) { @@ -2040,10 +2506,11 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, ret = futex_wake(uaddr, fshared, val, val3); break; case FUTEX_REQUEUE: - ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL); + ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 0); break; case FUTEX_CMP_REQUEUE: - ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3); + ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, + 0); break; case FUTEX_WAKE_OP: ret = futex_wake_op(uaddr, fshared, uaddr2, val, val2, val3); @@ -2060,6 +2527,18 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, if (futex_cmpxchg_enabled) ret = futex_lock_pi(uaddr, fshared, 0, timeout, 1); break; + case FUTEX_WAIT_REQUEUE_PI: + val3 = FUTEX_BITSET_MATCH_ANY; + ret = futex_wait_requeue_pi(uaddr, fshared, val, timeout, val3, + clockrt, uaddr2); + break; + case FUTEX_REQUEUE_PI: + ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 1); + break; + case FUTEX_CMP_REQUEUE_PI: + ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, + 1); + break; default: ret = -ENOSYS; } @@ -2077,7 +2556,8 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, int cmd = op & FUTEX_CMD_MASK; if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI || - cmd == FUTEX_WAIT_BITSET)) { + cmd == FUTEX_WAIT_BITSET || + cmd == FUTEX_WAIT_REQUEUE_PI)) { if (copy_from_user(&ts, utime, sizeof(ts)) != 0) return -EFAULT; if (!timespec_valid(&ts)) @@ -2089,10 +2569,11 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, tp = &t; } /* - * requeue parameter in 'utime' if cmd == FUTEX_REQUEUE. + * requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*. * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. */ if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || + cmd == FUTEX_REQUEUE_PI || cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) val2 = (u32) (unsigned long) utime; -- cgit v1.2.3 From 9c03d88e328d5f28f13191622c2ea1349c36b799 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:00 +0200 Subject: perf_counter: add more context information Change the callchain context entries to u16, so as to gain some space. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094517.457320003@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 2a5d4f525567..727624db5078 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1657,9 +1657,7 @@ void perf_counter_do_pending(void) * Callchain support -- arch specific */ -struct perf_callchain_entry * -__attribute__((weak)) -perf_callchain(struct pt_regs *regs) +__weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) { return NULL; } @@ -1819,7 +1817,7 @@ void perf_counter_output(struct perf_counter *counter, callchain = perf_callchain(regs); if (callchain) { - callchain_size = (2 + callchain->nr) * sizeof(u64); + callchain_size = (1 + callchain->nr) * sizeof(u64); header.type |= __PERF_EVENT_CALLCHAIN; header.size += callchain_size; -- cgit v1.2.3 From 3c446b3d3b38f991f97e9d2df0ad26a60a94dcff Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:01 +0200 Subject: perf_counter: SIGIO support Provide support for fcntl() I/O availability signals. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094517.579788800@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 727624db5078..c58cc64319e1 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1526,6 +1526,22 @@ out: return ret; } +static int perf_fasync(int fd, struct file *filp, int on) +{ + struct perf_counter *counter = filp->private_data; + struct inode *inode = filp->f_path.dentry->d_inode; + int retval; + + mutex_lock(&inode->i_mutex); + retval = fasync_helper(fd, filp, on, &counter->fasync); + mutex_unlock(&inode->i_mutex); + + if (retval < 0) + return retval; + + return 0; +} + static const struct file_operations perf_fops = { .release = perf_release, .read = perf_read, @@ -1533,6 +1549,7 @@ static const struct file_operations perf_fops = { .unlocked_ioctl = perf_ioctl, .compat_ioctl = perf_ioctl, .mmap = perf_mmap, + .fasync = perf_fasync, }; /* @@ -1549,7 +1566,7 @@ void perf_counter_wakeup(struct perf_counter *counter) rcu_read_lock(); data = rcu_dereference(counter->data); if (data) { - (void)atomic_xchg(&data->wakeup, POLL_IN); + atomic_set(&data->wakeup, POLL_IN); /* * Ensure all data writes are issued before updating the * user-space data head information. The matching rmb() @@ -1561,6 +1578,7 @@ void perf_counter_wakeup(struct perf_counter *counter) rcu_read_unlock(); wake_up_all(&counter->waitq); + kill_fasync(&counter->fasync, SIGIO, POLL_IN); } /* -- cgit v1.2.3 From 671dec5daf3b3c43c5777be282f00120a44cf37f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:02 +0200 Subject: perf_counter: generalize pending infrastructure Prepare the pending infrastructure to do more than wakeups. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094517.634732847@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 53 +++++++++++++++++++++++++++++++-------------------- 1 file changed, 32 insertions(+), 21 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c58cc64319e1..0a2ade2e4f11 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1581,6 +1581,14 @@ void perf_counter_wakeup(struct perf_counter *counter) kill_fasync(&counter->fasync, SIGIO, POLL_IN); } +static void perf_pending_wakeup(struct perf_pending_entry *entry) +{ + struct perf_counter *counter = container_of(entry, + struct perf_counter, pending); + + perf_counter_wakeup(counter); +} + /* * Pending wakeups * @@ -1590,45 +1598,47 @@ void perf_counter_wakeup(struct perf_counter *counter) * single linked list and use cmpxchg() to add entries lockless. */ -#define PENDING_TAIL ((struct perf_wakeup_entry *)-1UL) +#define PENDING_TAIL ((struct perf_pending_entry *)-1UL) -static DEFINE_PER_CPU(struct perf_wakeup_entry *, perf_wakeup_head) = { +static DEFINE_PER_CPU(struct perf_pending_entry *, perf_pending_head) = { PENDING_TAIL, }; -static void perf_pending_queue(struct perf_counter *counter) +static void perf_pending_queue(struct perf_pending_entry *entry, + void (*func)(struct perf_pending_entry *)) { - struct perf_wakeup_entry **head; - struct perf_wakeup_entry *prev, *next; + struct perf_pending_entry **head; - if (cmpxchg(&counter->wakeup.next, NULL, PENDING_TAIL) != NULL) + if (cmpxchg(&entry->next, NULL, PENDING_TAIL) != NULL) return; - head = &get_cpu_var(perf_wakeup_head); + entry->func = func; + + head = &get_cpu_var(perf_pending_head); do { - prev = counter->wakeup.next = *head; - next = &counter->wakeup; - } while (cmpxchg(head, prev, next) != prev); + entry->next = *head; + } while (cmpxchg(head, entry->next, entry) != entry->next); set_perf_counter_pending(); - put_cpu_var(perf_wakeup_head); + put_cpu_var(perf_pending_head); } static int __perf_pending_run(void) { - struct perf_wakeup_entry *list; + struct perf_pending_entry *list; int nr = 0; - list = xchg(&__get_cpu_var(perf_wakeup_head), PENDING_TAIL); + list = xchg(&__get_cpu_var(perf_pending_head), PENDING_TAIL); while (list != PENDING_TAIL) { - struct perf_counter *counter = container_of(list, - struct perf_counter, wakeup); + void (*func)(struct perf_pending_entry *); + struct perf_pending_entry *entry = list; list = list->next; - counter->wakeup.next = NULL; + func = entry->func; + entry->next = NULL; /* * Ensure we observe the unqueue before we issue the wakeup, * so that we won't be waiting forever. @@ -1636,7 +1646,7 @@ static int __perf_pending_run(void) */ smp_wmb(); - perf_counter_wakeup(counter); + func(entry); nr++; } @@ -1658,7 +1668,7 @@ static inline int perf_not_pending(struct perf_counter *counter) * so that we do not miss the wakeup. -- see perf_pending_handle() */ smp_rmb(); - return counter->wakeup.next == NULL; + return counter->pending.next == NULL; } static void perf_pending_sync(struct perf_counter *counter) @@ -1695,9 +1705,10 @@ struct perf_output_handle { static inline void __perf_output_wakeup(struct perf_output_handle *handle) { - if (handle->nmi) - perf_pending_queue(handle->counter); - else + if (handle->nmi) { + perf_pending_queue(&handle->counter->pending, + perf_pending_wakeup); + } else perf_counter_wakeup(handle->counter); } -- cgit v1.2.3 From f6c7d5fe58b4846ee0cb4b98b6042489705eced4 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:04 +0200 Subject: perf_counter: theres more to overflow than writing events Prepare for more generic overflow handling. The new perf_counter_overflow() method will handle the generic bits of the counter overflow, and can return a !0 return value, in which case the counter should be (soft) disabled, so that it won't count until it's properly disabled. XXX: do powerpc and swcounter Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094517.812109629@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 29 +++++++++++++++++++++++------ 1 file changed, 23 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0a2ade2e4f11..195e976eb07d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1800,8 +1800,8 @@ static void perf_output_end(struct perf_output_handle *handle) rcu_read_unlock(); } -void perf_counter_output(struct perf_counter *counter, - int nmi, struct pt_regs *regs) +static void perf_counter_output(struct perf_counter *counter, + int nmi, struct pt_regs *regs) { int ret; u64 record_type = counter->hw_event.record_type; @@ -2033,6 +2033,17 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, perf_counter_mmap_event(&mmap_event); } +/* + * Generic counter overflow handling. + */ + +int perf_counter_overflow(struct perf_counter *counter, + int nmi, struct pt_regs *regs) +{ + perf_counter_output(counter, nmi, regs); + return 0; +} + /* * Generic software counter infrastructure */ @@ -2077,6 +2088,7 @@ static void perf_swcounter_set_period(struct perf_counter *counter) static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) { + enum hrtimer_restart ret = HRTIMER_RESTART; struct perf_counter *counter; struct pt_regs *regs; @@ -2092,12 +2104,14 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) !counter->hw_event.exclude_user) regs = task_pt_regs(current); - if (regs) - perf_counter_output(counter, 0, regs); + if (regs) { + if (perf_counter_overflow(counter, 0, regs)) + ret = HRTIMER_NORESTART; + } hrtimer_forward_now(hrtimer, ns_to_ktime(counter->hw.irq_period)); - return HRTIMER_RESTART; + return ret; } static void perf_swcounter_overflow(struct perf_counter *counter, @@ -2105,7 +2119,10 @@ static void perf_swcounter_overflow(struct perf_counter *counter, { perf_swcounter_update(counter); perf_swcounter_set_period(counter); - perf_counter_output(counter, nmi, regs); + if (perf_counter_overflow(counter, nmi, regs)) + /* soft-disable the counter */ + ; + } static int perf_swcounter_match(struct perf_counter *counter, -- cgit v1.2.3 From ebb3c4c4cb81d64cc041356915ec015e2c57092a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:05 +0200 Subject: perf_counter: fix the mlock accounting Reading through the code I saw I forgot the finish the mlock accounting. Do so now. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094517.899767331@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 33 +++++++++++++++++++++------------ 1 file changed, 21 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 195e976eb07d..c841563de043 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1461,13 +1461,14 @@ static void perf_mmap_close(struct vm_area_struct *vma) if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { + vma->vm_mm->locked_vm -= counter->data->nr_pages + 1; perf_mmap_data_free(counter); mutex_unlock(&counter->mmap_mutex); } } static struct vm_operations_struct perf_mmap_vmops = { - .open = perf_mmap_open, + .open = perf_mmap_open, .close = perf_mmap_close, .fault = perf_mmap_fault, }; @@ -1499,24 +1500,32 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) if (vma->vm_pgoff != 0) return -EINVAL; - locked = vma_size >> PAGE_SHIFT; - locked += vma->vm_mm->locked_vm; + mutex_lock(&counter->mmap_mutex); + if (atomic_inc_not_zero(&counter->mmap_count)) { + if (nr_pages != counter->data->nr_pages) + ret = -EINVAL; + goto unlock; + } + + locked = vma->vm_mm->locked_vm; + locked += nr_pages + 1; lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; lock_limit >>= PAGE_SHIFT; - if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) - return -EPERM; - - mutex_lock(&counter->mmap_mutex); - if (atomic_inc_not_zero(&counter->mmap_count)) - goto out; + if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { + ret = -EPERM; + goto unlock; + } WARN_ON(counter->data); ret = perf_mmap_data_alloc(counter, nr_pages); - if (!ret) - atomic_set(&counter->mmap_count, 1); -out: + if (ret) + goto unlock; + + atomic_set(&counter->mmap_count, 1); + vma->vm_mm->locked_vm += nr_pages + 1; +unlock: mutex_unlock(&counter->mmap_mutex); vma->vm_flags &= ~VM_MAYWRITE; -- cgit v1.2.3 From 339f7c90b8a2f3aa2dd4267e79f797999e8a3c59 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:06 +0200 Subject: perf_counter: PERF_RECORD_TIME By popular request, provide means to log a timestamp along with the counter overflow event. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.024173282@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c841563de043..19990d1f0215 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1826,6 +1826,7 @@ static void perf_counter_output(struct perf_counter *counter, } group_entry; struct perf_callchain_entry *callchain = NULL; int callchain_size = 0; + u64 time; header.type = PERF_EVENT_COUNTER_OVERFLOW; header.size = sizeof(header); @@ -1862,6 +1863,16 @@ static void perf_counter_output(struct perf_counter *counter, } } + if (record_type & PERF_RECORD_TIME) { + /* + * Maybe do better on x86 and provide cpu_clock_nmi() + */ + time = sched_clock(); + + header.type |= __PERF_EVENT_TIME; + header.size += sizeof(u64); + } + ret = perf_output_begin(&handle, counter, header.size, nmi); if (ret) return; @@ -1895,6 +1906,9 @@ static void perf_counter_output(struct perf_counter *counter, if (callchain) perf_output_copy(&handle, callchain, callchain_size); + if (record_type & PERF_RECORD_TIME) + perf_output_put(&handle, time); + perf_output_end(&handle); } -- cgit v1.2.3 From 79f146415623fe74f39af67c0f6adc208939a410 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:07 +0200 Subject: perf_counter: counter overflow limit Provide means to auto-disable the counter after 'n' overflow events. Create the counter with hw_event.disabled = 1, and then issue an ioctl(fd, PREF_COUNTER_IOC_REFRESH, n); to set the limit and enable the counter. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.083139737@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 51 +++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 41 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 19990d1f0215..c05e10354bc9 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -744,6 +744,12 @@ static void perf_counter_enable(struct perf_counter *counter) spin_unlock_irq(&ctx->lock); } +static void perf_counter_refresh(struct perf_counter *counter, int refresh) +{ + atomic_add(refresh, &counter->event_limit); + perf_counter_enable(counter); +} + /* * Enable a counter and all its children. */ @@ -1311,6 +1317,9 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) case PERF_COUNTER_IOC_DISABLE: perf_counter_disable_family(counter); break; + case PERF_COUNTER_IOC_REFRESH: + perf_counter_refresh(counter, arg); + break; default: err = -ENOTTY; } @@ -1590,14 +1599,6 @@ void perf_counter_wakeup(struct perf_counter *counter) kill_fasync(&counter->fasync, SIGIO, POLL_IN); } -static void perf_pending_wakeup(struct perf_pending_entry *entry) -{ - struct perf_counter *counter = container_of(entry, - struct perf_counter, pending); - - perf_counter_wakeup(counter); -} - /* * Pending wakeups * @@ -1607,6 +1608,22 @@ static void perf_pending_wakeup(struct perf_pending_entry *entry) * single linked list and use cmpxchg() to add entries lockless. */ +static void perf_pending_counter(struct perf_pending_entry *entry) +{ + struct perf_counter *counter = container_of(entry, + struct perf_counter, pending); + + if (counter->pending_disable) { + counter->pending_disable = 0; + perf_counter_disable(counter); + } + + if (counter->pending_wakeup) { + counter->pending_wakeup = 0; + perf_counter_wakeup(counter); + } +} + #define PENDING_TAIL ((struct perf_pending_entry *)-1UL) static DEFINE_PER_CPU(struct perf_pending_entry *, perf_pending_head) = { @@ -1715,8 +1732,9 @@ struct perf_output_handle { static inline void __perf_output_wakeup(struct perf_output_handle *handle) { if (handle->nmi) { + handle->counter->pending_wakeup = 1; perf_pending_queue(&handle->counter->pending, - perf_pending_wakeup); + perf_pending_counter); } else perf_counter_wakeup(handle->counter); } @@ -2063,8 +2081,21 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, int perf_counter_overflow(struct perf_counter *counter, int nmi, struct pt_regs *regs) { + int events = atomic_read(&counter->event_limit); + int ret = 0; + + if (events && atomic_dec_and_test(&counter->event_limit)) { + ret = 1; + if (nmi) { + counter->pending_disable = 1; + perf_pending_queue(&counter->pending, + perf_pending_counter); + } else + perf_counter_disable(counter); + } + perf_counter_output(counter, nmi, regs); - return 0; + return ret; } /* -- cgit v1.2.3 From 4c9e25428ff46b968a30f1dfafdba550cb6e4141 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:09 +0200 Subject: perf_counter: change event definition Currently the definition of an event is slightly ambiguous. We have wakeup events, for poll() and SIGIO, which are either generated when a record crosses a page boundary (hw_events.wakeup_events == 0), or every wakeup_events new records. Now a record can be either a counter overflow record, or a number of different things, like the mmap PROT_EXEC region notifications. Then there is the PERF_COUNTER_IOC_REFRESH event limit, which only considers counter overflows. This patch changes then wakeup_events and SIGIO notification to only consider overflow events. Furthermore it changes the SIGIO notification to report SIGHUP when the event limit is reached and the counter will be disabled. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.266679874@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c05e10354bc9..8c8eaf0625f9 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1596,7 +1596,11 @@ void perf_counter_wakeup(struct perf_counter *counter) rcu_read_unlock(); wake_up_all(&counter->waitq); - kill_fasync(&counter->fasync, SIGIO, POLL_IN); + + if (counter->pending_kill) { + kill_fasync(&counter->fasync, SIGIO, counter->pending_kill); + counter->pending_kill = 0; + } } /* @@ -1727,6 +1731,7 @@ struct perf_output_handle { unsigned int head; int wakeup; int nmi; + int overflow; }; static inline void __perf_output_wakeup(struct perf_output_handle *handle) @@ -1741,7 +1746,7 @@ static inline void __perf_output_wakeup(struct perf_output_handle *handle) static int perf_output_begin(struct perf_output_handle *handle, struct perf_counter *counter, unsigned int size, - int nmi) + int nmi, int overflow) { struct perf_mmap_data *data; unsigned int offset, head; @@ -1751,8 +1756,9 @@ static int perf_output_begin(struct perf_output_handle *handle, if (!data) goto out; - handle->counter = counter; - handle->nmi = nmi; + handle->counter = counter; + handle->nmi = nmi; + handle->overflow = overflow; if (!data->nr_pages) goto fail; @@ -1816,7 +1822,7 @@ static void perf_output_end(struct perf_output_handle *handle) { int wakeup_events = handle->counter->hw_event.wakeup_events; - if (wakeup_events) { + if (handle->overflow && wakeup_events) { int events = atomic_inc_return(&handle->data->events); if (events >= wakeup_events) { atomic_sub(wakeup_events, &handle->data->events); @@ -1891,7 +1897,7 @@ static void perf_counter_output(struct perf_counter *counter, header.size += sizeof(u64); } - ret = perf_output_begin(&handle, counter, header.size, nmi); + ret = perf_output_begin(&handle, counter, header.size, nmi, 1); if (ret) return; @@ -1955,7 +1961,7 @@ static void perf_counter_mmap_output(struct perf_counter *counter, { struct perf_output_handle handle; int size = mmap_event->event.header.size; - int ret = perf_output_begin(&handle, counter, size, 0); + int ret = perf_output_begin(&handle, counter, size, 0, 0); if (ret) return; @@ -2084,8 +2090,10 @@ int perf_counter_overflow(struct perf_counter *counter, int events = atomic_read(&counter->event_limit); int ret = 0; + counter->pending_kill = POLL_IN; if (events && atomic_dec_and_test(&counter->event_limit)) { ret = 1; + counter->pending_kill = POLL_HUP; if (nmi) { counter->pending_disable = 1; perf_pending_queue(&counter->pending, -- cgit v1.2.3 From 4af4998b8aa35600f4c4a4f3c3a23baca6081d02 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:10 +0200 Subject: perf_counter: rework context time Since perf_counter_context is switched along with tasks, we can maintain the context time without using the task runtime clock. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.353552838@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 78 ++++++++++++++++++++++----------------------------- 1 file changed, 34 insertions(+), 44 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 8c8eaf0625f9..84d85ab4e161 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -117,7 +117,7 @@ counter_sched_out(struct perf_counter *counter, return; counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_stopped = ctx->time_now; + counter->tstamp_stopped = ctx->time; counter->hw_ops->disable(counter); counter->oncpu = -1; @@ -253,27 +253,20 @@ retry: spin_unlock_irq(&ctx->lock); } -/* - * Get the current time for this context. - * If this is a task context, we use the task's task clock, - * or for a per-cpu context, we use the cpu clock. - */ -static u64 get_context_time(struct perf_counter_context *ctx, int update) +static inline u64 perf_clock(void) { - struct task_struct *curr = ctx->task; - - if (!curr) - return cpu_clock(smp_processor_id()); - - return __task_delta_exec(curr, update) + curr->se.sum_exec_runtime; + return cpu_clock(smp_processor_id()); } /* * Update the record of the current time in a context. */ -static void update_context_time(struct perf_counter_context *ctx, int update) +static void update_context_time(struct perf_counter_context *ctx) { - ctx->time_now = get_context_time(ctx, update) - ctx->time_lost; + u64 now = perf_clock(); + + ctx->time += now - ctx->timestamp; + ctx->timestamp = now; } /* @@ -284,15 +277,17 @@ static void update_counter_times(struct perf_counter *counter) struct perf_counter_context *ctx = counter->ctx; u64 run_end; - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { - counter->total_time_enabled = ctx->time_now - - counter->tstamp_enabled; - if (counter->state == PERF_COUNTER_STATE_INACTIVE) - run_end = counter->tstamp_stopped; - else - run_end = ctx->time_now; - counter->total_time_running = run_end - counter->tstamp_running; - } + if (counter->state < PERF_COUNTER_STATE_INACTIVE) + return; + + counter->total_time_enabled = ctx->time - counter->tstamp_enabled; + + if (counter->state == PERF_COUNTER_STATE_INACTIVE) + run_end = counter->tstamp_stopped; + else + run_end = ctx->time; + + counter->total_time_running = run_end - counter->tstamp_running; } /* @@ -332,7 +327,7 @@ static void __perf_counter_disable(void *info) * If it is in error state, leave it in error state. */ if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { - update_context_time(ctx, 1); + update_context_time(ctx); update_counter_times(counter); if (counter == counter->group_leader) group_sched_out(counter, cpuctx, ctx); @@ -426,7 +421,7 @@ counter_sched_in(struct perf_counter *counter, return -EAGAIN; } - counter->tstamp_running += ctx->time_now - counter->tstamp_stopped; + counter->tstamp_running += ctx->time - counter->tstamp_stopped; if (!is_software_counter(counter)) cpuctx->active_oncpu++; @@ -493,9 +488,9 @@ static void add_counter_to_ctx(struct perf_counter *counter, list_add_counter(counter, ctx); ctx->nr_counters++; counter->prev_state = PERF_COUNTER_STATE_OFF; - counter->tstamp_enabled = ctx->time_now; - counter->tstamp_running = ctx->time_now; - counter->tstamp_stopped = ctx->time_now; + counter->tstamp_enabled = ctx->time; + counter->tstamp_running = ctx->time; + counter->tstamp_stopped = ctx->time; } /* @@ -522,7 +517,7 @@ static void __perf_install_in_context(void *info) curr_rq_lock_irq_save(&flags); spin_lock(&ctx->lock); - update_context_time(ctx, 1); + update_context_time(ctx); /* * Protect the list operation against NMI by disabling the @@ -648,13 +643,13 @@ static void __perf_counter_enable(void *info) curr_rq_lock_irq_save(&flags); spin_lock(&ctx->lock); - update_context_time(ctx, 1); + update_context_time(ctx); counter->prev_state = counter->state; if (counter->state >= PERF_COUNTER_STATE_INACTIVE) goto unlock; counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = ctx->time_now - counter->total_time_enabled; + counter->tstamp_enabled = ctx->time - counter->total_time_enabled; /* * If the counter is in a group and isn't the group leader, @@ -737,8 +732,8 @@ static void perf_counter_enable(struct perf_counter *counter) */ if (counter->state == PERF_COUNTER_STATE_OFF) { counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = ctx->time_now - - counter->total_time_enabled; + counter->tstamp_enabled = + ctx->time - counter->total_time_enabled; } out: spin_unlock_irq(&ctx->lock); @@ -778,7 +773,7 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, ctx->is_active = 0; if (likely(!ctx->nr_counters)) goto out; - update_context_time(ctx, 0); + update_context_time(ctx); flags = hw_perf_save_disable(); if (ctx->nr_active) { @@ -883,12 +878,7 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, if (likely(!ctx->nr_counters)) goto out; - /* - * Add any time since the last sched_out to the lost time - * so it doesn't get included in the total_time_enabled and - * total_time_running measures for counters in the context. - */ - ctx->time_lost = get_context_time(ctx, 0) - ctx->time_now; + ctx->timestamp = perf_clock(); flags = hw_perf_save_disable(); @@ -1043,8 +1033,8 @@ int perf_counter_task_enable(void) if (counter->state > PERF_COUNTER_STATE_OFF) continue; counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = ctx->time_now - - counter->total_time_enabled; + counter->tstamp_enabled = + ctx->time - counter->total_time_enabled; counter->hw_event.disabled = 0; } hw_perf_restore(perf_flags); @@ -1113,7 +1103,7 @@ static void __read(void *info) curr_rq_lock_irq_save(&flags); if (ctx->is_active) - update_context_time(ctx, 1); + update_context_time(ctx); counter->hw_ops->read(counter); update_counter_times(counter); curr_rq_unlock_irq_restore(&flags); -- cgit v1.2.3 From a39d6f2556c4a19f58f538c6aa28bf8faca4fcb8 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:11 +0200 Subject: perf_counter: rework the task clock software counter Rework the task clock software counter to use the context time instead of the task runtime clock, this removes the last such user. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.445450972@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 42 ++++++++++++------------------------------ 1 file changed, 12 insertions(+), 30 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 84d85ab4e161..56b7eb53d673 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -974,9 +974,6 @@ int perf_counter_task_disable(void) curr_rq_lock_irq_save(&flags); cpu = smp_processor_id(); - /* force the update of the task clock: */ - __task_delta_exec(curr, 1); - perf_counter_task_sched_out(curr, cpu); spin_lock(&ctx->lock); @@ -1017,9 +1014,6 @@ int perf_counter_task_enable(void) curr_rq_lock_irq_save(&flags); cpu = smp_processor_id(); - /* force the update of the task clock: */ - __task_delta_exec(curr, 1); - perf_counter_task_sched_out(curr, cpu); spin_lock(&ctx->lock); @@ -2347,38 +2341,28 @@ static const struct hw_perf_counter_ops perf_ops_cpu_clock = { * Software counter: task time clock */ -/* - * Called from within the scheduler: - */ -static u64 task_clock_perf_counter_val(struct perf_counter *counter, int update) -{ - struct task_struct *curr = counter->task; - u64 delta; - - delta = __task_delta_exec(curr, update); - - return curr->se.sum_exec_runtime + delta; -} - -static void task_clock_perf_counter_update(struct perf_counter *counter, u64 now) +static void task_clock_perf_counter_update(struct perf_counter *counter) { - u64 prev; + u64 prev, now; s64 delta; - prev = atomic64_read(&counter->hw.prev_count); - - atomic64_set(&counter->hw.prev_count, now); + update_context_time(counter->ctx); + now = counter->ctx->time; + prev = atomic64_xchg(&counter->hw.prev_count, now); delta = now - prev; - atomic64_add(delta, &counter->count); } static int task_clock_perf_counter_enable(struct perf_counter *counter) { struct hw_perf_counter *hwc = &counter->hw; + u64 now; + + update_context_time(counter->ctx); + now = counter->ctx->time; - atomic64_set(&hwc->prev_count, task_clock_perf_counter_val(counter, 0)); + atomic64_set(&hwc->prev_count, now); hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hwc->hrtimer.function = perf_swcounter_hrtimer; if (hwc->irq_period) { @@ -2393,14 +2377,12 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) static void task_clock_perf_counter_disable(struct perf_counter *counter) { hrtimer_cancel(&counter->hw.hrtimer); - task_clock_perf_counter_update(counter, - task_clock_perf_counter_val(counter, 0)); + task_clock_perf_counter_update(counter); } static void task_clock_perf_counter_read(struct perf_counter *counter) { - task_clock_perf_counter_update(counter, - task_clock_perf_counter_val(counter, 1)); + task_clock_perf_counter_update(counter); } static const struct hw_perf_counter_ops perf_ops_task_clock = { -- cgit v1.2.3 From 849691a6cd40270ff5f4a8846d5f6bf8df663ffc Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:12 +0200 Subject: perf_counter: remove rq->lock usage Now that all the task runtime clock users are gone, remove the ugly rq->lock usage from perf counters, which solves the nasty deadlock seen when a software task clock counter was read from an NMI overflow context. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.531137582@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 42 ++++++++++++++++-------------------------- kernel/sched.c | 20 -------------------- 2 files changed, 16 insertions(+), 46 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 56b7eb53d673..f4f7596f7841 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -172,8 +172,7 @@ static void __perf_counter_remove_from_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - curr_rq_lock_irq_save(&flags); - spin_lock(&ctx->lock); + spin_lock_irqsave(&ctx->lock, flags); counter_sched_out(counter, cpuctx, ctx); @@ -198,8 +197,7 @@ static void __perf_counter_remove_from_context(void *info) perf_max_counters - perf_reserved_percpu); } - spin_unlock(&ctx->lock); - curr_rq_unlock_irq_restore(&flags); + spin_unlock_irqrestore(&ctx->lock, flags); } @@ -319,8 +317,7 @@ static void __perf_counter_disable(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - curr_rq_lock_irq_save(&flags); - spin_lock(&ctx->lock); + spin_lock_irqsave(&ctx->lock, flags); /* * If the counter is on, turn it off. @@ -336,8 +333,7 @@ static void __perf_counter_disable(void *info) counter->state = PERF_COUNTER_STATE_OFF; } - spin_unlock(&ctx->lock); - curr_rq_unlock_irq_restore(&flags); + spin_unlock_irqrestore(&ctx->lock, flags); } /* @@ -515,8 +511,7 @@ static void __perf_install_in_context(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - curr_rq_lock_irq_save(&flags); - spin_lock(&ctx->lock); + spin_lock_irqsave(&ctx->lock, flags); update_context_time(ctx); /* @@ -565,8 +560,7 @@ static void __perf_install_in_context(void *info) unlock: hw_perf_restore(perf_flags); - spin_unlock(&ctx->lock); - curr_rq_unlock_irq_restore(&flags); + spin_unlock_irqrestore(&ctx->lock, flags); } /* @@ -641,8 +635,7 @@ static void __perf_counter_enable(void *info) if (ctx->task && cpuctx->task_ctx != ctx) return; - curr_rq_lock_irq_save(&flags); - spin_lock(&ctx->lock); + spin_lock_irqsave(&ctx->lock, flags); update_context_time(ctx); counter->prev_state = counter->state; @@ -678,8 +671,7 @@ static void __perf_counter_enable(void *info) } unlock: - spin_unlock(&ctx->lock); - curr_rq_unlock_irq_restore(&flags); + spin_unlock_irqrestore(&ctx->lock, flags); } /* @@ -971,7 +963,7 @@ int perf_counter_task_disable(void) if (likely(!ctx->nr_counters)) return 0; - curr_rq_lock_irq_save(&flags); + local_irq_save(flags); cpu = smp_processor_id(); perf_counter_task_sched_out(curr, cpu); @@ -992,9 +984,7 @@ int perf_counter_task_disable(void) hw_perf_restore(perf_flags); - spin_unlock(&ctx->lock); - - curr_rq_unlock_irq_restore(&flags); + spin_unlock_irqrestore(&ctx->lock, flags); return 0; } @@ -1011,7 +1001,7 @@ int perf_counter_task_enable(void) if (likely(!ctx->nr_counters)) return 0; - curr_rq_lock_irq_save(&flags); + local_irq_save(flags); cpu = smp_processor_id(); perf_counter_task_sched_out(curr, cpu); @@ -1037,7 +1027,7 @@ int perf_counter_task_enable(void) perf_counter_task_sched_in(curr, cpu); - curr_rq_unlock_irq_restore(&flags); + local_irq_restore(flags); return 0; } @@ -1095,12 +1085,12 @@ static void __read(void *info) struct perf_counter_context *ctx = counter->ctx; unsigned long flags; - curr_rq_lock_irq_save(&flags); + local_irq_save(flags); if (ctx->is_active) update_context_time(ctx); counter->hw_ops->read(counter); update_counter_times(counter); - curr_rq_unlock_irq_restore(&flags); + local_irq_restore(flags); } static u64 perf_counter_read(struct perf_counter *counter) @@ -2890,7 +2880,7 @@ __perf_counter_exit_task(struct task_struct *child, * Be careful about zapping the list - IRQ/NMI context * could still be processing it: */ - curr_rq_lock_irq_save(&flags); + local_irq_save(flags); perf_flags = hw_perf_save_disable(); cpuctx = &__get_cpu_var(perf_cpu_context); @@ -2903,7 +2893,7 @@ __perf_counter_exit_task(struct task_struct *child, child_ctx->nr_counters--; hw_perf_restore(perf_flags); - curr_rq_unlock_irq_restore(&flags); + local_irq_restore(flags); } parent_counter = child_counter->parent; diff --git a/kernel/sched.c b/kernel/sched.c index f76e3c0188a2..0de2f814fb18 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -997,26 +997,6 @@ static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) } } -void curr_rq_lock_irq_save(unsigned long *flags) - __acquires(rq->lock) -{ - struct rq *rq; - - local_irq_save(*flags); - rq = cpu_rq(smp_processor_id()); - spin_lock(&rq->lock); -} - -void curr_rq_unlock_irq_restore(unsigned long *flags) - __releases(rq->lock) -{ - struct rq *rq; - - rq = cpu_rq(smp_processor_id()); - spin_unlock(&rq->lock); - local_irq_restore(*flags); -} - void task_rq_unlock_wait(struct task_struct *p) { struct rq *rq = task_rq(p); -- cgit v1.2.3 From bce379bf358738ab8efc8cda2594a299ac685085 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 6 Apr 2009 11:45:13 +0200 Subject: perf_counter: minimize context time updates Push the update_context_time() calls up the stack so that we get less invokations and thereby a less noisy output: before: # ./perfstat -e 1:0 -e 1:1 -e 1:1 -e 1:1 -l ls > /dev/null Performance counter stats for 'ls': 10.163691 cpu clock ticks (msecs) (scaled from 98.94%) 10.215360 task clock ticks (msecs) (scaled from 98.18%) 10.185549 task clock ticks (msecs) (scaled from 98.53%) 10.183581 task clock ticks (msecs) (scaled from 98.71%) Wall-clock time elapsed: 11.912858 msecs after: # ./perfstat -e 1:0 -e 1:1 -e 1:1 -e 1:1 -l ls > /dev/null Performance counter stats for 'ls': 9.316630 cpu clock ticks (msecs) 9.280789 task clock ticks (msecs) 9.280789 task clock ticks (msecs) 9.280789 task clock ticks (msecs) Wall-clock time elapsed: 9.574872 msecs Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090406094518.618876874@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f4f7596f7841..863703b3158f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -319,6 +319,8 @@ static void __perf_counter_disable(void *info) spin_lock_irqsave(&ctx->lock, flags); + update_context_time(ctx); + /* * If the counter is on, turn it off. * If it is in error state, leave it in error state. @@ -797,6 +799,8 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) if (likely(!cpuctx->task_ctx)) return; + update_context_time(ctx); + regs = task_pt_regs(task); perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs); __perf_counter_sched_out(ctx, cpuctx); @@ -2336,7 +2340,6 @@ static void task_clock_perf_counter_update(struct perf_counter *counter) u64 prev, now; s64 delta; - update_context_time(counter->ctx); now = counter->ctx->time; prev = atomic64_xchg(&counter->hw.prev_count, now); @@ -2349,7 +2352,6 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) struct hw_perf_counter *hwc = &counter->hw; u64 now; - update_context_time(counter->ctx); now = counter->ctx->time; atomic64_set(&hwc->prev_count, now); @@ -2372,6 +2374,7 @@ static void task_clock_perf_counter_disable(struct perf_counter *counter) static void task_clock_perf_counter_read(struct perf_counter *counter) { + update_context_time(counter->ctx); task_clock_perf_counter_update(counter); } -- cgit v1.2.3 From bab5bc9e857638880facef76e4b4c3fa807f8c73 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Tue, 7 Apr 2009 23:23:50 -0700 Subject: futex: fixup unlocked requeue pi case Thomas's testing caught a problem when the requeue target futex is unowned and multiple tasks are requeued to it. This patch ensures the FUTEX_WAITERS bit gets set if futex_requeue() will requeue one or more tasks in addition to the one acquiring the lock. Signed-off-by: Darren Hart Signed-off-by: Thomas Gleixner --- kernel/futex.c | 65 +++++++++++++++++++++++++++++++++++++++------------------- 1 file changed, 44 insertions(+), 21 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 185c981d89e3..041bf3ac4be9 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -565,12 +565,14 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, /** * futex_lock_pi_atomic() - atomic work required to acquire a pi aware futex - * @uaddr: the pi futex user address - * @hb: the pi futex hash bucket - * @key: the futex key associated with uaddr and hb - * @ps: the pi_state pointer where we store the result of the lookup - * @task: the task to perform the atomic lock work for. This will be - * "current" except in the case of requeue pi. + * @uaddr: the pi futex user address + * @hb: the pi futex hash bucket + * @key: the futex key associated with uaddr and hb + * @ps: the pi_state pointer where we store the result of the + * lookup + * @task: the task to perform the atomic lock work for. This will + * be "current" except in the case of requeue pi. + * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) * * Returns: * 0 - ready to wait @@ -582,7 +584,7 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb, static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, union futex_key *key, struct futex_pi_state **ps, - struct task_struct *task) + struct task_struct *task, int set_waiters) { int lock_taken, ret, ownerdied = 0; u32 uval, newval, curval; @@ -596,6 +598,8 @@ retry: * the locks. It will most likely not succeed. */ newval = task_pid_vnr(task); + if (set_waiters) + newval |= FUTEX_WAITERS; curval = cmpxchg_futex_value_locked(uaddr, 0, newval); @@ -1004,14 +1008,18 @@ void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key) /** * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter - * @pifutex: the user address of the to futex - * @hb1: the from futex hash bucket, must be locked by the caller - * @hb2: the to futex hash bucket, must be locked by the caller - * @key1: the from futex key - * @key2: the to futex key + * @pifutex: the user address of the to futex + * @hb1: the from futex hash bucket, must be locked by the caller + * @hb2: the to futex hash bucket, must be locked by the caller + * @key1: the from futex key + * @key2: the to futex key + * @ps: address to store the pi_state pointer + * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) * * Try and get the lock on behalf of the top waiter if we can do it atomically. - * Wake the top waiter if we succeed. hb1 and hb2 must be held by the caller. + * Wake the top waiter if we succeed. If the caller specified set_waiters, + * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. + * hb1 and hb2 must be held by the caller. * * Returns: * 0 - failed to acquire the lock atomicly @@ -1022,15 +1030,23 @@ static int futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2, union futex_key *key1, union futex_key *key2, - struct futex_pi_state **ps) + struct futex_pi_state **ps, int set_waiters) { - struct futex_q *top_waiter; + struct futex_q *top_waiter = NULL; u32 curval; int ret; if (get_futex_value_locked(&curval, pifutex)) return -EFAULT; + /* + * Find the top_waiter and determine if there are additional waiters. + * If the caller intends to requeue more than 1 waiter to pifutex, + * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, + * as we have means to handle the possible fault. If not, don't set + * the bit unecessarily as it will force the subsequent unlock to enter + * the kernel. + */ top_waiter = futex_top_waiter(hb1, key1); /* There are no waiters, nothing for us to do. */ @@ -1038,10 +1054,12 @@ static int futex_proxy_trylock_atomic(u32 __user *pifutex, return 0; /* - * Either take the lock for top_waiter or set the FUTEX_WAITERS bit. - * The pi_state is returned in ps in contended cases. + * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in + * the contended case or if set_waiters is 1. The pi_state is returned + * in ps in contended cases. */ - ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task); + ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, + set_waiters); if (ret == 1) requeue_pi_wake_futex(top_waiter, key2); @@ -1146,9 +1164,14 @@ retry_private: } if (requeue_pi && (task_count - nr_wake < nr_requeue)) { - /* Attempt to acquire uaddr2 and wake the top_waiter. */ + /* + * Attempt to acquire uaddr2 and wake the top waiter. If we + * intend to requeue waiters, force setting the FUTEX_WAITERS + * bit. We force this here where we are able to easily handle + * faults rather in the requeue loop below. + */ ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, - &key2, &pi_state); + &key2, &pi_state, nr_requeue); /* * At this point the top_waiter has either taken uaddr2 or is @@ -1810,7 +1833,7 @@ retry: retry_private: hb = queue_lock(&q); - ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current); + ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, 0); if (unlikely(ret)) { switch (ret) { case 1: -- cgit v1.2.3 From e30e08f65c7ef6c230424264f09c3d53f117f58b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 8 Apr 2009 15:01:25 +0200 Subject: perf_counter: fix NMI race in task clock We should not be updating ctx->time from NMI context, work around that. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090408130408.681326666@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 25 ++++++++++++++++--------- 1 file changed, 16 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 863703b3158f..84a39081344c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -319,8 +319,6 @@ static void __perf_counter_disable(void *info) spin_lock_irqsave(&ctx->lock, flags); - update_context_time(ctx); - /* * If the counter is on, turn it off. * If it is in error state, leave it in error state. @@ -2335,13 +2333,11 @@ static const struct hw_perf_counter_ops perf_ops_cpu_clock = { * Software counter: task time clock */ -static void task_clock_perf_counter_update(struct perf_counter *counter) +static void task_clock_perf_counter_update(struct perf_counter *counter, u64 now) { - u64 prev, now; + u64 prev; s64 delta; - now = counter->ctx->time; - prev = atomic64_xchg(&counter->hw.prev_count, now); delta = now - prev; atomic64_add(delta, &counter->count); @@ -2369,13 +2365,24 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) static void task_clock_perf_counter_disable(struct perf_counter *counter) { hrtimer_cancel(&counter->hw.hrtimer); - task_clock_perf_counter_update(counter); + task_clock_perf_counter_update(counter, counter->ctx->time); + } static void task_clock_perf_counter_read(struct perf_counter *counter) { - update_context_time(counter->ctx); - task_clock_perf_counter_update(counter); + u64 time; + + if (!in_nmi()) { + update_context_time(counter->ctx); + time = counter->ctx->time; + } else { + u64 now = perf_clock(); + u64 delta = now - counter->ctx->timestamp; + time = counter->ctx->time + delta; + } + + task_clock_perf_counter_update(counter, time); } static const struct hw_perf_counter_ops perf_ops_task_clock = { -- cgit v1.2.3 From 6fab01927e8bdbbc77bafba2abb4810c5591ad52 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 8 Apr 2009 15:01:26 +0200 Subject: perf_counter: provide misc bits in the event header Limit the size of each record to 64k (or should we count in multiples of u64 and have a 512K limit?), this gives 16 bits or spare room in the header, which we can use for misc bits, so as to not have to grow the record with u64 every time we have a few bits to report. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090408130408.769271806@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 84a39081344c..4af98f943d3b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1831,6 +1831,9 @@ static void perf_counter_output(struct perf_counter *counter, header.type = PERF_EVENT_COUNTER_OVERFLOW; header.size = sizeof(header); + header.misc = user_mode(regs) ? + PERF_EVENT_MISC_USER : PERF_EVENT_MISC_KERNEL; + if (record_type & PERF_RECORD_IP) { ip = instruction_pointer(regs); header.type |= __PERF_EVENT_IP; -- cgit v1.2.3 From 6b6e5486b3a168f0328c82a8d4376caf901472b1 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 8 Apr 2009 15:01:27 +0200 Subject: perf_counter: use misc field to widen type Push the PERF_EVENT_COUNTER_OVERFLOW bit into the misc field so that we can have the full 32bit for PERF_RECORD_ bits. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090408130408.891867663@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4af98f943d3b..bf12df6f3538 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1828,15 +1828,16 @@ static void perf_counter_output(struct perf_counter *counter, int callchain_size = 0; u64 time; - header.type = PERF_EVENT_COUNTER_OVERFLOW; + header.type = 0; header.size = sizeof(header); - header.misc = user_mode(regs) ? + header.misc = PERF_EVENT_MISC_OVERFLOW; + header.misc |= user_mode(regs) ? PERF_EVENT_MISC_USER : PERF_EVENT_MISC_KERNEL; if (record_type & PERF_RECORD_IP) { ip = instruction_pointer(regs); - header.type |= __PERF_EVENT_IP; + header.type |= PERF_RECORD_IP; header.size += sizeof(ip); } @@ -1845,12 +1846,12 @@ static void perf_counter_output(struct perf_counter *counter, tid_entry.pid = current->group_leader->pid; tid_entry.tid = current->pid; - header.type |= __PERF_EVENT_TID; + header.type |= PERF_RECORD_TID; header.size += sizeof(tid_entry); } if (record_type & PERF_RECORD_GROUP) { - header.type |= __PERF_EVENT_GROUP; + header.type |= PERF_RECORD_GROUP; header.size += sizeof(u64) + counter->nr_siblings * sizeof(group_entry); } @@ -1861,7 +1862,7 @@ static void perf_counter_output(struct perf_counter *counter, if (callchain) { callchain_size = (1 + callchain->nr) * sizeof(u64); - header.type |= __PERF_EVENT_CALLCHAIN; + header.type |= PERF_RECORD_CALLCHAIN; header.size += callchain_size; } } @@ -1872,7 +1873,7 @@ static void perf_counter_output(struct perf_counter *counter, */ time = sched_clock(); - header.type |= __PERF_EVENT_TIME; + header.type |= PERF_RECORD_TIME; header.size += sizeof(u64); } -- cgit v1.2.3 From 8d1b2d9361b494bfc761700c348c65ebbe3deb5b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 8 Apr 2009 15:01:30 +0200 Subject: perf_counter: track task-comm data Similar to the mmap data stream, add one that tracks the task COMM field, so that the userspace reporting knows what to call a task. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090408130409.127422406@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 93 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 93 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index bf12df6f3538..2d4aebb2982b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1916,6 +1916,99 @@ static void perf_counter_output(struct perf_counter *counter, perf_output_end(&handle); } +/* + * comm tracking + */ + +struct perf_comm_event { + struct task_struct *task; + char *comm; + int comm_size; + + struct { + struct perf_event_header header; + + u32 pid; + u32 tid; + } event; +}; + +static void perf_counter_comm_output(struct perf_counter *counter, + struct perf_comm_event *comm_event) +{ + struct perf_output_handle handle; + int size = comm_event->event.header.size; + int ret = perf_output_begin(&handle, counter, size, 0, 0); + + if (ret) + return; + + perf_output_put(&handle, comm_event->event); + perf_output_copy(&handle, comm_event->comm, + comm_event->comm_size); + perf_output_end(&handle); +} + +static int perf_counter_comm_match(struct perf_counter *counter, + struct perf_comm_event *comm_event) +{ + if (counter->hw_event.comm && + comm_event->event.header.type == PERF_EVENT_COMM) + return 1; + + return 0; +} + +static void perf_counter_comm_ctx(struct perf_counter_context *ctx, + struct perf_comm_event *comm_event) +{ + struct perf_counter *counter; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { + if (perf_counter_comm_match(counter, comm_event)) + perf_counter_comm_output(counter, comm_event); + } + rcu_read_unlock(); +} + +static void perf_counter_comm_event(struct perf_comm_event *comm_event) +{ + struct perf_cpu_context *cpuctx; + unsigned int size; + char *comm = comm_event->task->comm; + + size = ALIGN(strlen(comm), sizeof(u64)); + + comm_event->comm = comm; + comm_event->comm_size = size; + + comm_event->event.header.size = sizeof(comm_event->event) + size; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_counter_comm_ctx(&cpuctx->ctx, comm_event); + put_cpu_var(perf_cpu_context); + + perf_counter_comm_ctx(¤t->perf_counter_ctx, comm_event); +} + +void perf_counter_comm(struct task_struct *task) +{ + struct perf_comm_event comm_event = { + .task = task, + .event = { + .header = { .type = PERF_EVENT_COMM, }, + .pid = task->group_leader->pid, + .tid = task->pid, + }, + }; + + perf_counter_comm_event(&comm_event); +} + /* * mmap tracking */ -- cgit v1.2.3 From 4d855457d84b819fefcd1cd1b0a2a0a0ec475c07 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 8 Apr 2009 15:01:32 +0200 Subject: perf_counter: move PERF_RECORD_TIME Move PERF_RECORD_TIME so that all the fixed length items come before the variable length ones. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090408130409.307926436@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 26 +++++++++++++------------- 1 file changed, 13 insertions(+), 13 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 2d4aebb2982b..4dc8600d2825 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1850,6 +1850,16 @@ static void perf_counter_output(struct perf_counter *counter, header.size += sizeof(tid_entry); } + if (record_type & PERF_RECORD_TIME) { + /* + * Maybe do better on x86 and provide cpu_clock_nmi() + */ + time = sched_clock(); + + header.type |= PERF_RECORD_TIME; + header.size += sizeof(u64); + } + if (record_type & PERF_RECORD_GROUP) { header.type |= PERF_RECORD_GROUP; header.size += sizeof(u64) + @@ -1867,16 +1877,6 @@ static void perf_counter_output(struct perf_counter *counter, } } - if (record_type & PERF_RECORD_TIME) { - /* - * Maybe do better on x86 and provide cpu_clock_nmi() - */ - time = sched_clock(); - - header.type |= PERF_RECORD_TIME; - header.size += sizeof(u64); - } - ret = perf_output_begin(&handle, counter, header.size, nmi, 1); if (ret) return; @@ -1889,6 +1889,9 @@ static void perf_counter_output(struct perf_counter *counter, if (record_type & PERF_RECORD_TID) perf_output_put(&handle, tid_entry); + if (record_type & PERF_RECORD_TIME) + perf_output_put(&handle, time); + if (record_type & PERF_RECORD_GROUP) { struct perf_counter *leader, *sub; u64 nr = counter->nr_siblings; @@ -1910,9 +1913,6 @@ static void perf_counter_output(struct perf_counter *counter, if (callchain) perf_output_copy(&handle, callchain, callchain_size); - if (record_type & PERF_RECORD_TIME) - perf_output_put(&handle, time); - perf_output_end(&handle); } -- cgit v1.2.3 From 78f13e9525ba777da25c4ddab89f28e9366a8b7c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 8 Apr 2009 15:01:33 +0200 Subject: perf_counter: allow for data addresses to be recorded Paul suggested we allow for data addresses to be recorded along with the traditional IPs as power can provide these. For now, only the software pagefault events provide data addresses, but in the future power might as well for some events. x86 doesn't seem capable of providing this atm. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090408130409.394816925@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 46 +++++++++++++++++++++++++++++----------------- 1 file changed, 29 insertions(+), 17 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4dc8600d2825..321c57e3556f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -800,7 +800,7 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) update_context_time(ctx); regs = task_pt_regs(task); - perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs); + perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs, 0); __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; @@ -1810,7 +1810,7 @@ static void perf_output_end(struct perf_output_handle *handle) } static void perf_counter_output(struct perf_counter *counter, - int nmi, struct pt_regs *regs) + int nmi, struct pt_regs *regs, u64 addr) { int ret; u64 record_type = counter->hw_event.record_type; @@ -1860,6 +1860,11 @@ static void perf_counter_output(struct perf_counter *counter, header.size += sizeof(u64); } + if (record_type & PERF_RECORD_ADDR) { + header.type |= PERF_RECORD_ADDR; + header.size += sizeof(u64); + } + if (record_type & PERF_RECORD_GROUP) { header.type |= PERF_RECORD_GROUP; header.size += sizeof(u64) + @@ -1892,6 +1897,9 @@ static void perf_counter_output(struct perf_counter *counter, if (record_type & PERF_RECORD_TIME) perf_output_put(&handle, time); + if (record_type & PERF_RECORD_ADDR) + perf_output_put(&handle, addr); + if (record_type & PERF_RECORD_GROUP) { struct perf_counter *leader, *sub; u64 nr = counter->nr_siblings; @@ -2158,7 +2166,7 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, */ int perf_counter_overflow(struct perf_counter *counter, - int nmi, struct pt_regs *regs) + int nmi, struct pt_regs *regs, u64 addr) { int events = atomic_read(&counter->event_limit); int ret = 0; @@ -2175,7 +2183,7 @@ int perf_counter_overflow(struct perf_counter *counter, perf_counter_disable(counter); } - perf_counter_output(counter, nmi, regs); + perf_counter_output(counter, nmi, regs, addr); return ret; } @@ -2240,7 +2248,7 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) regs = task_pt_regs(current); if (regs) { - if (perf_counter_overflow(counter, 0, regs)) + if (perf_counter_overflow(counter, 0, regs, 0)) ret = HRTIMER_NORESTART; } @@ -2250,11 +2258,11 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) } static void perf_swcounter_overflow(struct perf_counter *counter, - int nmi, struct pt_regs *regs) + int nmi, struct pt_regs *regs, u64 addr) { perf_swcounter_update(counter); perf_swcounter_set_period(counter); - if (perf_counter_overflow(counter, nmi, regs)) + if (perf_counter_overflow(counter, nmi, regs, addr)) /* soft-disable the counter */ ; @@ -2286,16 +2294,17 @@ static int perf_swcounter_match(struct perf_counter *counter, } static void perf_swcounter_add(struct perf_counter *counter, u64 nr, - int nmi, struct pt_regs *regs) + int nmi, struct pt_regs *regs, u64 addr) { int neg = atomic64_add_negative(nr, &counter->hw.count); if (counter->hw.irq_period && !neg) - perf_swcounter_overflow(counter, nmi, regs); + perf_swcounter_overflow(counter, nmi, regs, addr); } static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, enum perf_event_types type, u32 event, - u64 nr, int nmi, struct pt_regs *regs) + u64 nr, int nmi, struct pt_regs *regs, + u64 addr) { struct perf_counter *counter; @@ -2305,7 +2314,7 @@ static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { if (perf_swcounter_match(counter, type, event, regs)) - perf_swcounter_add(counter, nr, nmi, regs); + perf_swcounter_add(counter, nr, nmi, regs, addr); } rcu_read_unlock(); } @@ -2325,7 +2334,8 @@ static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx) } static void __perf_swcounter_event(enum perf_event_types type, u32 event, - u64 nr, int nmi, struct pt_regs *regs) + u64 nr, int nmi, struct pt_regs *regs, + u64 addr) { struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); int *recursion = perf_swcounter_recursion_context(cpuctx); @@ -2336,10 +2346,11 @@ static void __perf_swcounter_event(enum perf_event_types type, u32 event, (*recursion)++; barrier(); - perf_swcounter_ctx_event(&cpuctx->ctx, type, event, nr, nmi, regs); + perf_swcounter_ctx_event(&cpuctx->ctx, type, event, + nr, nmi, regs, addr); if (cpuctx->task_ctx) { perf_swcounter_ctx_event(cpuctx->task_ctx, type, event, - nr, nmi, regs); + nr, nmi, regs, addr); } barrier(); @@ -2349,9 +2360,10 @@ out: put_cpu_var(perf_cpu_context); } -void perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs) +void +perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { - __perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, regs); + __perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, regs, addr); } static void perf_swcounter_read(struct perf_counter *counter) @@ -2548,7 +2560,7 @@ void perf_tpcounter_event(int event_id) if (!regs) regs = task_pt_regs(current); - __perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs); + __perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs, 0); } extern int ftrace_profile_enable(int); -- cgit v1.2.3 From 002f128b473fb82f454654be5081b0919ee01ab2 Mon Sep 17 00:00:00 2001 From: Paul Turner Date: Wed, 8 Apr 2009 15:29:43 -0700 Subject: sched: remove redundant hierarchy walk in check_preempt_wakeup Impact: micro-optimization Under group scheduling we traverse up until we are at common siblings to make the wakeup comparison on. At this point however, they should have the same parent so continuing to check up the tree is redundant. Signed-off-by: Paul Turner Acked-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/sched_fair.c | 13 +++---------- 1 file changed, 3 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c index 3816f217f119..5f9650e8fe75 100644 --- a/kernel/sched_fair.c +++ b/kernel/sched_fair.c @@ -1487,17 +1487,10 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync) find_matching_se(&se, &pse); - while (se) { - BUG_ON(!pse); + BUG_ON(!pse); - if (wakeup_preempt_entity(se, pse) == 1) { - resched_task(curr); - break; - } - - se = parent_entity(se); - pse = parent_entity(pse); - } + if (wakeup_preempt_entity(se, pse) == 1) + resched_task(curr); } static struct task_struct *pick_next_task_fair(struct rq *rq) -- cgit v1.2.3 From 888fcee066a2f4abd0d0bc9418c0535f9b01e6e5 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 9 Apr 2009 09:48:22 +0200 Subject: perf_counter: fix off task->comm by one strlen() does not include the \0. Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 321c57e3556f..b07195bbd228 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1989,7 +1989,7 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) unsigned int size; char *comm = comm_event->task->comm; - size = ALIGN(strlen(comm), sizeof(u64)); + size = ALIGN(strlen(comm)+1, sizeof(u64)); comm_event->comm = comm; comm_event->comm_size = size; @@ -2109,7 +2109,7 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) } got_name: - size = ALIGN(strlen(name), sizeof(u64)); + size = ALIGN(strlen(name)+1, sizeof(u64)); mmap_event->file_name = name; mmap_event->file_size = size; -- cgit v1.2.3 From 9ee318a7825929bc3734110b83ae8e20e53d9de3 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 9 Apr 2009 10:53:44 +0200 Subject: perf_counter: optimize mmap/comm tracking Impact: performance optimization The mmap/comm tracking code does quite a lot of work before it discovers there's no interest in it, avoid that by keeping a counter. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090409085524.427173196@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 39 ++++++++++++++++++++++++++++++++++++--- 1 file changed, 36 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index b07195bbd228..76376ecb23b5 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -38,6 +38,10 @@ int perf_max_counters __read_mostly = 1; static int perf_reserved_percpu __read_mostly; static int perf_overcommit __read_mostly = 1; +static atomic_t nr_mmap_tracking __read_mostly; +static atomic_t nr_munmap_tracking __read_mostly; +static atomic_t nr_comm_tracking __read_mostly; + /* * Mutex for (sysadmin-configurable) counter reservations: */ @@ -1186,6 +1190,13 @@ static void free_counter(struct perf_counter *counter) { perf_pending_sync(counter); + if (counter->hw_event.mmap) + atomic_dec(&nr_mmap_tracking); + if (counter->hw_event.munmap) + atomic_dec(&nr_munmap_tracking); + if (counter->hw_event.comm) + atomic_dec(&nr_comm_tracking); + if (counter->destroy) counter->destroy(counter); @@ -2005,7 +2016,12 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) void perf_counter_comm(struct task_struct *task) { - struct perf_comm_event comm_event = { + struct perf_comm_event comm_event; + + if (!atomic_read(&nr_comm_tracking)) + return; + + comm_event = (struct perf_comm_event){ .task = task, .event = { .header = { .type = PERF_EVENT_COMM, }, @@ -2128,7 +2144,12 @@ got_name: void perf_counter_mmap(unsigned long addr, unsigned long len, unsigned long pgoff, struct file *file) { - struct perf_mmap_event mmap_event = { + struct perf_mmap_event mmap_event; + + if (!atomic_read(&nr_mmap_tracking)) + return; + + mmap_event = (struct perf_mmap_event){ .file = file, .event = { .header = { .type = PERF_EVENT_MMAP, }, @@ -2146,7 +2167,12 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, void perf_counter_munmap(unsigned long addr, unsigned long len, unsigned long pgoff, struct file *file) { - struct perf_mmap_event mmap_event = { + struct perf_mmap_event mmap_event; + + if (!atomic_read(&nr_munmap_tracking)) + return; + + mmap_event = (struct perf_mmap_event){ .file = file, .event = { .header = { .type = PERF_EVENT_MUNMAP, }, @@ -2725,6 +2751,13 @@ done: counter->hw_ops = hw_ops; + if (counter->hw_event.mmap) + atomic_inc(&nr_mmap_tracking); + if (counter->hw_event.munmap) + atomic_inc(&nr_munmap_tracking); + if (counter->hw_event.comm) + atomic_inc(&nr_comm_tracking); + return counter; } -- cgit v1.2.3 From 1ccd15497869f3ed83b5225d410df53a96e52757 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 9 Apr 2009 10:53:45 +0200 Subject: perf_counter: sysctl for system wide perf counters Impact: add sysctl for paranoid/relaxed perfcounters policy Allow the use of system wide perf counters to everybody, but provide a sysctl to disable it for the paranoid security minded. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090409085524.514046352@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 +++- kernel/sysctl.c | 11 +++++++++++ 2 files changed, 14 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 76376ecb23b5..7efb7ebaaae0 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -42,6 +42,8 @@ static atomic_t nr_mmap_tracking __read_mostly; static atomic_t nr_munmap_tracking __read_mostly; static atomic_t nr_comm_tracking __read_mostly; +int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ + /* * Mutex for (sysadmin-configurable) counter reservations: */ @@ -1132,7 +1134,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) */ if (cpu != -1) { /* Must be root to operate on a CPU counter: */ - if (!capable(CAP_SYS_ADMIN)) + if (sysctl_perf_counter_priv && !capable(CAP_SYS_ADMIN)) return ERR_PTR(-EACCES); if (cpu < 0 || cpu > num_possible_cpus()) diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 4286b62b34a0..8ba457838d95 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -49,6 +49,7 @@ #include #include #include +#include #include #include @@ -920,6 +921,16 @@ static struct ctl_table kern_table[] = { .child = slow_work_sysctls, }, #endif +#ifdef CONFIG_PERF_COUNTERS + { + .ctl_name = CTL_UNNUMBERED, + .procname = "perf_counter_privileged", + .data = &sysctl_perf_counter_priv, + .maxlen = sizeof(sysctl_perf_counter_priv), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, +#endif /* * NOTE: do not add new entries to this table unless you have read * Documentation/sysctl/ctl_unnumbered.txt -- cgit v1.2.3 From d3d21c412d8525eb2e208d990ab5eee5fb0fe03d Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 9 Apr 2009 10:53:46 +0200 Subject: perf_counter: log full path names Impact: fix perf-report output for /home mounted binaries, etc. dentry_path() only provide path-names up to the mount root, which is unsuited for out purpose, use d_path() instead. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090409085524.601794134@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7efb7ebaaae0..7f9521c3c01b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2116,7 +2116,7 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) name = strncpy(tmp, "//enomem", sizeof(tmp)); goto got_name; } - name = dentry_path(file->f_dentry, buf, PATH_MAX); + name = d_path(&file->f_path, buf, PATH_MAX); if (IS_ERR(name)) { name = strncpy(tmp, "//toolong", sizeof(tmp)); goto got_name; -- cgit v1.2.3 From a5a2a0c7fa039c59619bc908b3b1ed24734d442a Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Fri, 10 Apr 2009 09:50:05 -0700 Subject: futex: fix futex_wait_setup key handling If the get_futex_key() call were to fail, the existing code would try and put_futex_key() prior to returning. This patch makes sure we only put_futex_key() if get_futex_key() succeeded. Reported-by: Clark Williams Signed-off-by: Darren Hart LKML-Reference: <20090410165005.14342.16973.stgit@Aeon> Signed-off-by: Thomas Gleixner --- kernel/futex.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 041bf3ac4be9..6d2daa46f9ff 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1668,7 +1668,7 @@ retry: q->key = FUTEX_KEY_INIT; ret = get_futex_key(uaddr, fshared, &q->key); if (unlikely(ret != 0)) - goto out; + return ret; retry_private: *hb = queue_lock(q); -- cgit v1.2.3 From 7ba5c840e64d4a967379f1ae3eca73278180b11d Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Mon, 13 Apr 2009 21:31:17 -0700 Subject: rcu: Add __rcu_pending tracing to hierarchical RCU Add tracing to __rcu_pending() to provide information on why RCU processing was kicked off. This is helpful for debugging hierarchical RCU, and might also be helpful in learning how hierarchical RCU operates. Located-by: Anton Blanchard Tested-by: Anton Blanchard Signed-off-by: Paul E. McKenney Cc: anton@samba.org Cc: akpm@linux-foundation.org Cc: dipankar@in.ibm.com Cc: manfred@colorfullife.com Cc: cl@linux-foundation.org Cc: josht@linux.vnet.ibm.com Cc: schamp@sgi.com Cc: niv@us.ibm.com Cc: dvhltc@us.ibm.com Cc: ego@in.ibm.com Cc: laijs@cn.fujitsu.com Cc: rostedt@goodmis.org Cc: peterz@infradead.org Cc: penberg@cs.helsinki.fi Cc: andi@firstfloor.org Cc: "Paul E. McKenney" LKML-Reference: <1239683479943-git-send-email-> Signed-off-by: Ingo Molnar --- kernel/rcutree.c | 25 +++++++++++++++----- kernel/rcutree_trace.c | 64 +++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 82 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/rcutree.c b/kernel/rcutree.c index d2a372fb0b9b..0dccfbba6d26 100644 --- a/kernel/rcutree.c +++ b/kernel/rcutree.c @@ -1259,31 +1259,44 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) check_cpu_stall(rsp, rdp); /* Is the RCU core waiting for a quiescent state from this CPU? */ - if (rdp->qs_pending) + if (rdp->qs_pending) { + rdp->n_rp_qs_pending++; return 1; + } /* Does this CPU have callbacks ready to invoke? */ - if (cpu_has_callbacks_ready_to_invoke(rdp)) + if (cpu_has_callbacks_ready_to_invoke(rdp)) { + rdp->n_rp_cb_ready++; return 1; + } /* Has RCU gone idle with this CPU needing another grace period? */ - if (cpu_needs_another_gp(rsp, rdp)) + if (cpu_needs_another_gp(rsp, rdp)) { + rdp->n_rp_cpu_needs_gp++; return 1; + } /* Has another RCU grace period completed? */ - if (ACCESS_ONCE(rsp->completed) != rdp->completed) /* outside of lock */ + if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */ + rdp->n_rp_gp_completed++; return 1; + } /* Has a new RCU grace period started? */ - if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) /* outside of lock */ + if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */ + rdp->n_rp_gp_started++; return 1; + } /* Has an RCU GP gone long enough to send resched IPIs &c? */ if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && - ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) + ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { + rdp->n_rp_need_fqs++; return 1; + } /* nothing to do */ + rdp->n_rp_need_nothing++; return 0; } diff --git a/kernel/rcutree_trace.c b/kernel/rcutree_trace.c index 4b1875ba9404..fe1dcdbf1ca3 100644 --- a/kernel/rcutree_trace.c +++ b/kernel/rcutree_trace.c @@ -213,7 +213,63 @@ static struct file_operations rcugp_fops = { .release = single_release, }; -static struct dentry *rcudir, *datadir, *datadir_csv, *hierdir, *gpdir; +static void print_one_rcu_pending(struct seq_file *m, struct rcu_data *rdp) +{ + seq_printf(m, "%3d%cnp=%ld " + "qsp=%ld cbr=%ld cng=%ld gpc=%ld gps=%ld nf=%ld nn=%ld\n", + rdp->cpu, + cpu_is_offline(rdp->cpu) ? '!' : ' ', + rdp->n_rcu_pending, + rdp->n_rp_qs_pending, + rdp->n_rp_cb_ready, + rdp->n_rp_cpu_needs_gp, + rdp->n_rp_gp_completed, + rdp->n_rp_gp_started, + rdp->n_rp_need_fqs, + rdp->n_rp_need_nothing); +} + +static void print_rcu_pendings(struct seq_file *m, struct rcu_state *rsp) +{ + int cpu; + struct rcu_data *rdp; + + for_each_possible_cpu(cpu) { + rdp = rsp->rda[cpu]; + if (rdp->beenonline) + print_one_rcu_pending(m, rdp); + } +} + +static int show_rcu_pending(struct seq_file *m, void *unused) +{ + seq_puts(m, "rcu:\n"); + print_rcu_pendings(m, &rcu_state); + seq_puts(m, "rcu_bh:\n"); + print_rcu_pendings(m, &rcu_bh_state); + return 0; +} + +static int rcu_pending_open(struct inode *inode, struct file *file) +{ + return single_open(file, show_rcu_pending, NULL); +} + +static struct file_operations rcu_pending_fops = { + .owner = THIS_MODULE, + .open = rcu_pending_open, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, +}; + +static struct dentry *rcudir; +static struct dentry *datadir; +static struct dentry *datadir_csv; +static struct dentry *gpdir; +static struct dentry *hierdir; +static struct dentry *rcu_pendingdir; + static int __init rcuclassic_trace_init(void) { rcudir = debugfs_create_dir("rcu", NULL); @@ -238,6 +294,11 @@ static int __init rcuclassic_trace_init(void) NULL, &rcuhier_fops); if (!hierdir) goto free_out; + + rcu_pendingdir = debugfs_create_file("rcu_pending", 0444, rcudir, + NULL, &rcu_pending_fops); + if (!rcu_pendingdir) + goto free_out; return 0; free_out: if (datadir) @@ -257,6 +318,7 @@ static void __exit rcuclassic_trace_cleanup(void) debugfs_remove(datadir_csv); debugfs_remove(gpdir); debugfs_remove(hierdir); + debugfs_remove(rcu_pendingdir); debugfs_remove(rcudir); } -- cgit v1.2.3 From f711f6090a81cbd396b63de90f415d33f563af9b Mon Sep 17 00:00:00 2001 From: Gautham R Shenoy Date: Tue, 14 Apr 2009 10:25:30 +0530 Subject: sched: Nominate idle load balancer from a semi-idle package. Currently the nomination of idle-load balancer is done by choosing the first idle cpu in the nohz.cpu_mask. This may not be power-efficient, since such an idle cpu could come from a completely idle core/package thereby preventing the whole core/package from being in a low-power state. For eg, consider a quad-core dual package system. The cpu numbering need not be sequential and can something like [0, 2, 4, 6] and [1, 3, 5, 7]. With sched_mc/smt_power_savings and the power-aware IRQ balance, we try to keep as fewer Packages/Cores active. But the current idle load balancer logic goes against this by choosing the first_cpu in the nohz.cpu_mask and not taking the system topology into consideration. Improve the algorithm to nominate the idle load balancer from a semi idle cores/packages thereby increasing the probability of the cores/packages being in deeper sleep states for longer duration. The algorithm is activated only when sched_mc/smt_power_savings != 0. Signed-off-by: Gautham R Shenoy Acked-by: Peter Zijlstra LKML-Reference: <20090414045530.7645.12175.stgit@sofia.in.ibm.com> Signed-off-by: Ingo Molnar --- kernel/sched.c | 127 +++++++++++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 118 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 5724508c3b66..b0fefa300b40 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -4240,10 +4240,126 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) static struct { atomic_t load_balancer; cpumask_var_t cpu_mask; + cpumask_var_t ilb_grp_nohz_mask; } nohz ____cacheline_aligned = { .load_balancer = ATOMIC_INIT(-1), }; +#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) +/** + * lowest_flag_domain - Return lowest sched_domain containing flag. + * @cpu: The cpu whose lowest level of sched domain is to + * be returned. + * @flag: The flag to check for the lowest sched_domain + * for the given cpu. + * + * Returns the lowest sched_domain of a cpu which contains the given flag. + */ +static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) +{ + struct sched_domain *sd; + + for_each_domain(cpu, sd) + if (sd && (sd->flags & flag)) + break; + + return sd; +} + +/** + * for_each_flag_domain - Iterates over sched_domains containing the flag. + * @cpu: The cpu whose domains we're iterating over. + * @sd: variable holding the value of the power_savings_sd + * for cpu. + * @flag: The flag to filter the sched_domains to be iterated. + * + * Iterates over all the scheduler domains for a given cpu that has the 'flag' + * set, starting from the lowest sched_domain to the highest. + */ +#define for_each_flag_domain(cpu, sd, flag) \ + for (sd = lowest_flag_domain(cpu, flag); \ + (sd && (sd->flags & flag)); sd = sd->parent) + +/** + * is_semi_idle_group - Checks if the given sched_group is semi-idle. + * @ilb_group: group to be checked for semi-idleness + * + * Returns: 1 if the group is semi-idle. 0 otherwise. + * + * We define a sched_group to be semi idle if it has atleast one idle-CPU + * and atleast one non-idle CPU. This helper function checks if the given + * sched_group is semi-idle or not. + */ +static inline int is_semi_idle_group(struct sched_group *ilb_group) +{ + cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, + sched_group_cpus(ilb_group)); + + /* + * A sched_group is semi-idle when it has atleast one busy cpu + * and atleast one idle cpu. + */ + if (cpumask_empty(nohz.ilb_grp_nohz_mask)) + return 0; + + if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) + return 0; + + return 1; +} +/** + * find_new_ilb - Finds the optimum idle load balancer for nomination. + * @cpu: The cpu which is nominating a new idle_load_balancer. + * + * Returns: Returns the id of the idle load balancer if it exists, + * Else, returns >= nr_cpu_ids. + * + * This algorithm picks the idle load balancer such that it belongs to a + * semi-idle powersavings sched_domain. The idea is to try and avoid + * completely idle packages/cores just for the purpose of idle load balancing + * when there are other idle cpu's which are better suited for that job. + */ +static int find_new_ilb(int cpu) +{ + struct sched_domain *sd; + struct sched_group *ilb_group; + + /* + * Have idle load balancer selection from semi-idle packages only + * when power-aware load balancing is enabled + */ + if (!(sched_smt_power_savings || sched_mc_power_savings)) + goto out_done; + + /* + * Optimize for the case when we have no idle CPUs or only one + * idle CPU. Don't walk the sched_domain hierarchy in such cases + */ + if (cpumask_weight(nohz.cpu_mask) < 2) + goto out_done; + + for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { + ilb_group = sd->groups; + + do { + if (is_semi_idle_group(ilb_group)) + return cpumask_first(nohz.ilb_grp_nohz_mask); + + ilb_group = ilb_group->next; + + } while (ilb_group != sd->groups); + } + +out_done: + return cpumask_first(nohz.cpu_mask); +} +#else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ +static inline int find_new_ilb(int call_cpu) +{ + return first_cpu(nohz.cpu_mask); +} +#endif + /* * This routine will try to nominate the ilb (idle load balancing) * owner among the cpus whose ticks are stopped. ilb owner will do the idle @@ -4468,15 +4584,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu) } if (atomic_read(&nohz.load_balancer) == -1) { - /* - * simple selection for now: Nominate the - * first cpu in the nohz list to be the next - * ilb owner. - * - * TBD: Traverse the sched domains and nominate - * the nearest cpu in the nohz.cpu_mask. - */ - int ilb = cpumask_first(nohz.cpu_mask); + int ilb = find_new_ilb(cpu); if (ilb < nr_cpu_ids) resched_cpu(ilb); @@ -9051,6 +9159,7 @@ void __init sched_init(void) #ifdef CONFIG_SMP #ifdef CONFIG_NO_HZ alloc_bootmem_cpumask_var(&nohz.cpu_mask); + alloc_bootmem_cpumask_var(&nohz.ilb_grp_nohz_mask); #endif alloc_bootmem_cpumask_var(&cpu_isolated_map); #endif /* SMP */ -- cgit v1.2.3 From e790fb0ba64bfec158e1219d899cb588275d12ab Mon Sep 17 00:00:00 2001 From: Gautham R Shenoy Date: Tue, 14 Apr 2009 10:25:35 +0530 Subject: sched: Nominate a power-efficient ilb in select_nohz_balancer() The CPU that first goes idle becomes the idle-load-balancer and remains that until either it picks up a task or till all the CPUs of the system goes idle. Optimize this further to allow it to relinquish it's post once all it's siblings in the power-aware sched_domain go idle, thereby allowing the whole package-core to go idle. While relinquising the post, nominate another an idle-load balancer from a semi-idle core/package. Signed-off-by: Gautham R Shenoy Acked-by: Peter Zijlstra LKML-Reference: <20090414045535.7645.31641.stgit@sofia.in.ibm.com> Signed-off-by: Ingo Molnar --- kernel/sched.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index b0fefa300b40..36d213bca473 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -4414,8 +4414,24 @@ int select_nohz_load_balancer(int stop_tick) /* make me the ilb owner */ if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) return 1; - } else if (atomic_read(&nohz.load_balancer) == cpu) + } else if (atomic_read(&nohz.load_balancer) == cpu) { + int new_ilb; + + if (!(sched_smt_power_savings || + sched_mc_power_savings)) + return 1; + /* + * Check to see if there is a more power-efficient + * ilb. + */ + new_ilb = find_new_ilb(cpu); + if (new_ilb < nr_cpu_ids && new_ilb != cpu) { + atomic_set(&nohz.load_balancer, -1); + resched_cpu(new_ilb); + return 0; + } return 1; + } } else { if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) return 0; -- cgit v1.2.3 From 78ddb08feb7d4fbe3c0a9931804c51ee58be4023 Mon Sep 17 00:00:00 2001 From: Johannes Weiner Date: Tue, 14 Apr 2009 16:53:05 +0200 Subject: wait: don't use __wake_up_common() '777c6c5 wait: prevent exclusive waiter starvation' made __wake_up_common() global to be used from abort_exclusive_wait(). It was needed to do a wake-up with the waitqueue lock held while passing down a key to the wake-up function. Since '4ede816 epoll keyed wakeups: add __wake_up_locked_key() and __wake_up_sync_key()' there is an appropriate wrapper for this case: __wake_up_locked_key(). Use it here and make __wake_up_common() private to the scheduler again. Signed-off-by: Johannes Weiner Cc: Andrew Morton Cc: Peter Zijlstra LKML-Reference: <1239720785-19661-1-git-send-email-hannes@cmpxchg.org> Signed-off-by: Ingo Molnar --- kernel/sched.c | 2 +- kernel/wait.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 36d213bca473..92b4b56ad093 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -5345,7 +5345,7 @@ EXPORT_SYMBOL(default_wake_function); * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns * zero in this (rare) case, and we handle it by continuing to scan the queue. */ -void __wake_up_common(wait_queue_head_t *q, unsigned int mode, +static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, int nr_exclusive, int sync, void *key) { wait_queue_t *curr, *next; diff --git a/kernel/wait.c b/kernel/wait.c index 42a2dbc181c8..ea7c3b4275cf 100644 --- a/kernel/wait.c +++ b/kernel/wait.c @@ -154,7 +154,7 @@ void abort_exclusive_wait(wait_queue_head_t *q, wait_queue_t *wait, if (!list_empty(&wait->task_list)) list_del_init(&wait->task_list); else if (waitqueue_active(q)) - __wake_up_common(q, mode, 1, 0, key); + __wake_up_locked_key(q, mode, key); spin_unlock_irqrestore(&q->lock, flags); } EXPORT_SYMBOL(abort_exclusive_wait); -- cgit v1.2.3 From 19e4529ee7345079eeacc8e40cf69a304a64dc23 Mon Sep 17 00:00:00 2001 From: Stephen Rothwell Date: Tue, 14 Apr 2009 17:27:18 +1000 Subject: modules: Fix up build when CONFIG_MODULE_UNLOAD=n. Commit 3d43321b7015387cfebbe26436d0e9d299162ea1 ("modules: sysctl to block module loading") introduces a modules_disabled variable that is only defined if CONFIG_MODULE_UNLOAD is enabled, despite being used in other places. This moves it up and fixes up the build. CC kernel/module.o kernel/module.c: In function 'sys_init_module': kernel/module.c:2401: error: 'modules_disabled' undeclared (first use in this function) kernel/module.c:2401: error: (Each undeclared identifier is reported only once kernel/module.c:2401: error: for each function it appears in.) make[1]: *** [kernel/module.o] Error 1 make: *** [kernel/module.o] Error 2 Signed-off-by: Paul Mundt Signed-off-by: James Morris --- kernel/module.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/module.c b/kernel/module.c index eeb3f7b1383c..ee7ab612dafa 100644 --- a/kernel/module.c +++ b/kernel/module.c @@ -71,6 +71,9 @@ static DEFINE_MUTEX(module_mutex); static LIST_HEAD(modules); +/* Block module loading/unloading? */ +int modules_disabled = 0; + /* Waiting for a module to finish initializing? */ static DECLARE_WAIT_QUEUE_HEAD(module_wq); @@ -778,9 +781,6 @@ static void wait_for_zero_refcount(struct module *mod) mutex_lock(&module_mutex); } -/* Block module loading/unloading? */ -int modules_disabled = 0; - SYSCALL_DEFINE2(delete_module, const char __user *, name_user, unsigned int, flags) { -- cgit v1.2.3 From 13318a7186d8e0ae08c996ea4111a945e7789772 Mon Sep 17 00:00:00 2001 From: Miao Xie Date: Wed, 15 Apr 2009 09:59:10 +0800 Subject: sched: use group_first_cpu() instead of cpumask_first(sched_group_cpus()) Impact: cleanup This patch changes cpumask_first(sched_group_cpus()) to group_first_cpu() for maintainability. Signed-off-by: Miao Xie Acked-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/sched.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 92b4b56ad093..7601ceebf7ce 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -7995,7 +7995,7 @@ static void init_numa_sched_groups_power(struct sched_group *group_head) struct sched_domain *sd; sd = &per_cpu(phys_domains, j).sd; - if (j != cpumask_first(sched_group_cpus(sd->groups))) { + if (j != group_first_cpu(sd->groups)) { /* * Only add "power" once for each * physical package. @@ -8073,7 +8073,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd) WARN_ON(!sd || !sd->groups); - if (cpu != cpumask_first(sched_group_cpus(sd->groups))) + if (cpu != group_first_cpu(sd->groups)) return; child = sd->child; -- cgit v1.2.3 From ff7b1b4f000cea84f071c1b6aa2918b2119d66f1 Mon Sep 17 00:00:00 2001 From: Steven Whitehouse Date: Wed, 15 Apr 2009 16:55:05 +0100 Subject: perfcounters: export perf_tpcounter_event Needed for modular tracepoint support. Signed-off-by: Steven Whitehouse Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7f9521c3c01b..09396098dd0d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2590,6 +2590,7 @@ void perf_tpcounter_event(int event_id) __perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, 1, 1, regs, 0); } +EXPORT_SYMBOL_GPL(perf_tpcounter_event); extern int ftrace_profile_enable(int); extern void ftrace_profile_disable(int); -- cgit v1.2.3 From f1f9b3b1795da8625e0e6096813c9d18d4a344ce Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 20 Apr 2009 20:38:21 +0200 Subject: perfcounters, sched: remove __task_delta_exec() This function was left orphan by the latest round of sw-counter cleanups. [ Impact: remove unused kernel function ] Signed-off-by: Ingo Molnar --- kernel/sched.c | 23 ----------------------- 1 file changed, 23 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index b66a08c2480e..a69278eef425 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -4546,29 +4546,6 @@ DEFINE_PER_CPU(struct kernel_stat, kstat); EXPORT_PER_CPU_SYMBOL(kstat); -/* - * Return any ns on the sched_clock that have not yet been banked in - * @p in case that task is currently running. - */ -unsigned long long __task_delta_exec(struct task_struct *p, int update) -{ - s64 delta_exec; - struct rq *rq; - - rq = task_rq(p); - WARN_ON_ONCE(!runqueue_is_locked()); - WARN_ON_ONCE(!task_current(rq, p)); - - if (update) - update_rq_clock(rq); - - delta_exec = rq->clock - p->se.exec_start; - - WARN_ON_ONCE(delta_exec < 0); - - return delta_exec; -} - /* * Return any ns on the sched_clock that have not yet been banked in * @p in case that task is currently running. -- cgit v1.2.3 From ff743345bf7685a207868048a70e23164c4785e5 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 13 Mar 2009 12:21:26 +0100 Subject: sched: remove extra call overhead for schedule() Lai Jiangshan's patch reminded me that I promised Nick to remove that extra call overhead in schedule(). Signed-off-by: Peter Zijlstra LKML-Reference: <20090313112300.927414207@chello.nl> Signed-off-by: Ingo Molnar --- kernel/mutex.c | 4 +++- kernel/sched.c | 12 ++++-------- 2 files changed, 7 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/mutex.c b/kernel/mutex.c index 5d79781394a3..e1fb73510409 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -248,7 +248,9 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass, /* didnt get the lock, go to sleep: */ spin_unlock_mutex(&lock->wait_lock, flags); - __schedule(); + preempt_enable_no_resched(); + schedule(); + preempt_disable(); spin_lock_mutex(&lock->wait_lock, flags); } diff --git a/kernel/sched.c b/kernel/sched.c index 7601ceebf7ce..797f6fdabadf 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -5131,13 +5131,15 @@ pick_next_task(struct rq *rq) /* * schedule() is the main scheduler function. */ -asmlinkage void __sched __schedule(void) +asmlinkage void __sched schedule(void) { struct task_struct *prev, *next; unsigned long *switch_count; struct rq *rq; int cpu; +need_resched: + preempt_disable(); cpu = smp_processor_id(); rq = cpu_rq(cpu); rcu_qsctr_inc(cpu); @@ -5194,15 +5196,9 @@ need_resched_nonpreemptible: if (unlikely(reacquire_kernel_lock(current) < 0)) goto need_resched_nonpreemptible; -} -asmlinkage void __sched schedule(void) -{ -need_resched: - preempt_disable(); - __schedule(); preempt_enable_no_resched(); - if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) + if (need_resched()) goto need_resched; } EXPORT_SYMBOL(schedule); -- cgit v1.2.3 From 6e29ec5701e9d44fa02b96c1c5c45f7516182b65 Mon Sep 17 00:00:00 2001 From: Gautham R Shenoy Date: Tue, 21 Apr 2009 08:40:49 +0530 Subject: sched: Replace first_cpu() with cpumask_first() in ILB nomination code Stephen Rothwell reported this build warning: > kernel/sched.c: In function 'find_new_ilb': > kernel/sched.c:4355: warning: passing argument 1 of '__first_cpu' from incompatible pointer type > > Possibly caused by commit f711f6090a81cbd396b63de90f415d33f563af9b > ("sched: Nominate idle load balancer from a semi-idle package") from > the sched tree. Should this call to first_cpu be cpumask_first? For !(CONFIG_SCHED_MC || CONFIG_SCHED_SMT), find_new_ilb() nominates the Idle load balancer as the first cpu from the nohz.cpu_mask. This code uses the older API first_cpu(). Replace it with cpumask_first(), which is the correct API here. [ Impact: cleanup, address build warning ] Reported-by: Stephen Rothwell Signed-off-by: Gautham R Shenoy Cc: Rusty Russell LKML-Reference: <20090421031049.GA4140@in.ibm.com> Signed-off-by: Ingo Molnar --- kernel/sched.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 797f6fdabadf..54d67b94f1a9 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -4356,7 +4356,7 @@ out_done: #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ static inline int find_new_ilb(int call_cpu) { - return first_cpu(nohz.cpu_mask); + return cpumask_first(nohz.cpu_mask); } #endif -- cgit v1.2.3 From 9ec4fa271faf2db3b8e1419c998da1ca6b094eb6 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Mon, 27 Apr 2009 17:57:18 -0700 Subject: irq, cpumask: correct CPUMASKS_OFFSTACK typo and fix fallout CPUMASKS_OFFSTACK is not defined anywhere (it is CPUMASK_OFFSTACK). It is a typo and init_allocate_desc_masks() is called before it set affinity to all cpus... Split init_alloc_desc_masks() into all_desc_masks() and init_desc_masks(). Also use CPUMASK_OFFSTACK in alloc_desc_masks(). [ Impact: fix smp_affinity copying/setup when moving irq_desc between CPUs ] Signed-off-by: Yinghai Lu Acked-by: Rusty Russell Cc: Andrew Morton Cc: Suresh Siddha Cc: "Eric W. Biederman" LKML-Reference: <49F6546E.3040406@kernel.org> Signed-off-by: Ingo Molnar --- kernel/irq/handle.c | 9 ++++++--- kernel/irq/numa_migrate.c | 2 +- 2 files changed, 7 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index d82142be8dd2..882c79800107 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -115,10 +115,11 @@ static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu) printk(KERN_ERR "can not alloc kstat_irqs\n"); BUG_ON(1); } - if (!init_alloc_desc_masks(desc, cpu, false)) { + if (!alloc_desc_masks(desc, cpu, false)) { printk(KERN_ERR "can not alloc irq_desc cpumasks\n"); BUG_ON(1); } + init_desc_masks(desc); arch_init_chip_data(desc, cpu); } @@ -169,7 +170,8 @@ int __init early_irq_init(void) desc[i].irq = i; desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids; lockdep_set_class(&desc[i].lock, &irq_desc_lock_class); - init_alloc_desc_masks(&desc[i], 0, true); + alloc_desc_masks(&desc[i], 0, true); + init_desc_masks(&desc[i]); irq_desc_ptrs[i] = desc + i; } @@ -256,7 +258,8 @@ int __init early_irq_init(void) for (i = 0; i < count; i++) { desc[i].irq = i; - init_alloc_desc_masks(&desc[i], 0, true); + alloc_desc_masks(&desc[i], 0, true); + init_desc_masks(&desc[i]); desc[i].kstat_irqs = kstat_irqs_all[i]; } return arch_early_irq_init(); diff --git a/kernel/irq/numa_migrate.c b/kernel/irq/numa_migrate.c index 44bbdcbaf8d2..5760d7251626 100644 --- a/kernel/irq/numa_migrate.c +++ b/kernel/irq/numa_migrate.c @@ -37,7 +37,7 @@ static bool init_copy_one_irq_desc(int irq, struct irq_desc *old_desc, struct irq_desc *desc, int cpu) { memcpy(desc, old_desc, sizeof(struct irq_desc)); - if (!init_alloc_desc_masks(desc, cpu, false)) { + if (!alloc_desc_masks(desc, cpu, false)) { printk(KERN_ERR "irq %d: can not get new irq_desc cpumask " "for migration.\n", irq); return false; -- cgit v1.2.3 From fcef5911c7ea89b80d5bfc727f402f37c9eefd57 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Mon, 27 Apr 2009 17:58:23 -0700 Subject: x86/irq: remove leftover code from NUMA_MIGRATE_IRQ_DESC The original feature of migrating irq_desc dynamic was too fragile and was causing problems: it caused crashes on systems with lots of cards with MSI-X when user-space irq-balancer was enabled. We now have new patches that create irq_desc according to device numa node. This patch removes the leftover bits of the dynamic balancer. [ Impact: remove dead code ] Signed-off-by: Yinghai Lu Cc: Andrew Morton Cc: Suresh Siddha Cc: "Eric W. Biederman" Cc: Rusty Russell LKML-Reference: <49F654AF.8000808@kernel.org> Signed-off-by: Ingo Molnar --- kernel/irq/Makefile | 2 +- kernel/irq/chip.c | 12 ++---------- kernel/irq/handle.c | 9 ++------- kernel/irq/numa_migrate.c | 2 -- 4 files changed, 5 insertions(+), 20 deletions(-) (limited to 'kernel') diff --git a/kernel/irq/Makefile b/kernel/irq/Makefile index 3394f8f52964..2f065277f8ee 100644 --- a/kernel/irq/Makefile +++ b/kernel/irq/Makefile @@ -3,5 +3,5 @@ obj-y := handle.o manage.o spurious.o resend.o chip.o devres.o obj-$(CONFIG_GENERIC_IRQ_PROBE) += autoprobe.o obj-$(CONFIG_PROC_FS) += proc.o obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o -obj-$(CONFIG_NUMA_MIGRATE_IRQ_DESC) += numa_migrate.o +obj-$(CONFIG_SPARSE_IRQ) += numa_migrate.o obj-$(CONFIG_PM_SLEEP) += pm.o diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c index c687ba4363f2..13c68e71b726 100644 --- a/kernel/irq/chip.c +++ b/kernel/irq/chip.c @@ -359,7 +359,6 @@ handle_level_irq(unsigned int irq, struct irq_desc *desc) spin_lock(&desc->lock); mask_ack_irq(desc, irq); - desc = irq_remap_to_desc(irq, desc); if (unlikely(desc->status & IRQ_INPROGRESS)) goto out_unlock; @@ -438,7 +437,6 @@ handle_fasteoi_irq(unsigned int irq, struct irq_desc *desc) desc->status &= ~IRQ_INPROGRESS; out: desc->chip->eoi(irq); - desc = irq_remap_to_desc(irq, desc); spin_unlock(&desc->lock); } @@ -475,7 +473,6 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc) !desc->action)) { desc->status |= (IRQ_PENDING | IRQ_MASKED); mask_ack_irq(desc, irq); - desc = irq_remap_to_desc(irq, desc); goto out_unlock; } kstat_incr_irqs_this_cpu(irq, desc); @@ -483,7 +480,6 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc) /* Start handling the irq */ if (desc->chip->ack) desc->chip->ack(irq); - desc = irq_remap_to_desc(irq, desc); /* Mark the IRQ currently in progress.*/ desc->status |= IRQ_INPROGRESS; @@ -544,10 +540,8 @@ handle_percpu_irq(unsigned int irq, struct irq_desc *desc) if (!noirqdebug) note_interrupt(irq, desc, action_ret); - if (desc->chip->eoi) { + if (desc->chip->eoi) desc->chip->eoi(irq); - desc = irq_remap_to_desc(irq, desc); - } } void @@ -582,10 +576,8 @@ __set_irq_handler(unsigned int irq, irq_flow_handler_t handle, int is_chained, /* Uninstall? */ if (handle == handle_bad_irq) { - if (desc->chip != &no_irq_chip) { + if (desc->chip != &no_irq_chip) mask_ack_irq(desc, irq); - desc = irq_remap_to_desc(irq, desc); - } desc->status |= IRQ_DISABLED; desc->depth = 1; } diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index 882c79800107..3e0cbc44bd73 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -458,11 +458,8 @@ unsigned int __do_IRQ(unsigned int irq) /* * No locking required for CPU-local interrupts: */ - if (desc->chip->ack) { + if (desc->chip->ack) desc->chip->ack(irq); - /* get new one */ - desc = irq_remap_to_desc(irq, desc); - } if (likely(!(desc->status & IRQ_DISABLED))) { action_ret = handle_IRQ_event(irq, desc->action); if (!noirqdebug) @@ -473,10 +470,8 @@ unsigned int __do_IRQ(unsigned int irq) } spin_lock(&desc->lock); - if (desc->chip->ack) { + if (desc->chip->ack) desc->chip->ack(irq); - desc = irq_remap_to_desc(irq, desc); - } /* * REPLAY is when Linux resends an IRQ that was dropped earlier * WAITING is used by probe to mark irqs that are being tested diff --git a/kernel/irq/numa_migrate.c b/kernel/irq/numa_migrate.c index 5760d7251626..ce72bc3f4ced 100644 --- a/kernel/irq/numa_migrate.c +++ b/kernel/irq/numa_migrate.c @@ -97,9 +97,7 @@ static struct irq_desc *__real_move_irq_desc(struct irq_desc *old_desc, /* free the old one */ free_one_irq_desc(old_desc, desc); - spin_unlock(&old_desc->lock); kfree(old_desc); - spin_lock(&desc->lock); return desc; -- cgit v1.2.3 From 57b150cce8e004ddd36330490a68bfb59b7271e9 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Mon, 27 Apr 2009 17:59:53 -0700 Subject: irq: only update affinity if ->set_affinity() is sucessfull irq_set_affinity() and move_masked_irq() try to assign affinity before calling chip set_affinity(). Some archs are assigning it in ->set_affinity() again. We do something like: cpumask_cpy(desc->affinity, mask); desc->chip->set_affinity(mask); But in the failure path, affinity should not be touched - otherwise we'll end up with a different affinity mask despite the failure to migrate the IRQ. So try to update the afffinity only if set_affinity returns with 0. Also call irq_set_thread_affinity accordingly. v2: update after "irq, x86: Remove IRQ_DISABLED check in process context IRQ move" v3: according to Ingo, change set_affinity() in irq_chip should return int. v4: update comments by removing moving irq_desc code. [ Impact: fix /proc/irq/*/smp_affinity setting corner case bug ] Signed-off-by: Yinghai Lu Cc: Andrew Morton Cc: Suresh Siddha Cc: "Eric W. Biederman" Cc: Rusty Russell LKML-Reference: <49F65509.60307@kernel.org> Signed-off-by: Ingo Molnar --- kernel/irq/internals.h | 3 +++ kernel/irq/manage.c | 17 +++++++++++------ kernel/irq/migration.c | 14 +++++++++----- 3 files changed, 23 insertions(+), 11 deletions(-) (limited to 'kernel') diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h index 01ce20eab38f..de5f412f6a92 100644 --- a/kernel/irq/internals.h +++ b/kernel/irq/internals.h @@ -42,6 +42,9 @@ static inline void unregister_handler_proc(unsigned int irq, extern int irq_select_affinity_usr(unsigned int irq); +extern void +irq_set_thread_affinity(struct irq_desc *desc, const struct cpumask *cpumask); + /* * Debugging printout: */ diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index 2734eca59243..aaf5c9d05770 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -80,7 +80,7 @@ int irq_can_set_affinity(unsigned int irq) return 1; } -static void +void irq_set_thread_affinity(struct irq_desc *desc, const struct cpumask *cpumask) { struct irqaction *action = desc->action; @@ -109,17 +109,22 @@ int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask) spin_lock_irqsave(&desc->lock, flags); #ifdef CONFIG_GENERIC_PENDING_IRQ - if (desc->status & IRQ_MOVE_PCNTXT) - desc->chip->set_affinity(irq, cpumask); + if (desc->status & IRQ_MOVE_PCNTXT) { + if (!desc->chip->set_affinity(irq, cpumask)) { + cpumask_copy(desc->affinity, cpumask); + irq_set_thread_affinity(desc, cpumask); + } + } else { desc->status |= IRQ_MOVE_PENDING; cpumask_copy(desc->pending_mask, cpumask); } #else - cpumask_copy(desc->affinity, cpumask); - desc->chip->set_affinity(irq, cpumask); + if (!desc->chip->set_affinity(irq, cpumask)) { + cpumask_copy(desc->affinity, cpumask); + irq_set_thread_affinity(desc, cpumask); + } #endif - irq_set_thread_affinity(desc, cpumask); desc->status |= IRQ_AFFINITY_SET; spin_unlock_irqrestore(&desc->lock, flags); return 0; diff --git a/kernel/irq/migration.c b/kernel/irq/migration.c index e05ad9be43b7..cfe767ca1545 100644 --- a/kernel/irq/migration.c +++ b/kernel/irq/migration.c @@ -1,5 +1,8 @@ #include +#include + +#include "internals.h" void move_masked_irq(int irq) { @@ -39,11 +42,12 @@ void move_masked_irq(int irq) * masking the irqs. */ if (likely(cpumask_any_and(desc->pending_mask, cpu_online_mask) - < nr_cpu_ids)) { - cpumask_and(desc->affinity, - desc->pending_mask, cpu_online_mask); - desc->chip->set_affinity(irq, desc->affinity); - } + < nr_cpu_ids)) + if (!desc->chip->set_affinity(irq, desc->pending_mask)) { + cpumask_copy(desc->affinity, desc->pending_mask); + irq_set_thread_affinity(desc, desc->pending_mask); + } + cpumask_clear(desc->pending_mask); } -- cgit v1.2.3 From 85ac16d033370caf6f48d743c8dc8103700f5cc5 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Mon, 27 Apr 2009 18:00:38 -0700 Subject: x86/irq: change irq_desc_alloc() to take node instead of cpu This simplifies the node awareness of the code. All our allocators only deal with a NUMA node ID locality not with CPU ids anyway - so there's no need to maintain (and transform) a CPU id all across the IRq layer. v2: keep move_irq_desc related [ Impact: cleanup, prepare IRQ code to be NUMA-aware ] Signed-off-by: Yinghai Lu Cc: Andrew Morton Cc: Suresh Siddha Cc: "Eric W. Biederman" Cc: Rusty Russell Cc: Jeremy Fitzhardinge LKML-Reference: <49F65536.2020300@kernel.org> Signed-off-by: Ingo Molnar --- kernel/irq/handle.c | 28 +++++++++++----------------- kernel/irq/internals.h | 2 +- kernel/irq/numa_migrate.c | 36 ++++++++++++------------------------ kernel/softirq.c | 2 +- 4 files changed, 25 insertions(+), 43 deletions(-) (limited to 'kernel') diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index 3e0cbc44bd73..a6368db2618b 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -81,12 +81,10 @@ static struct irq_desc irq_desc_init = { .lock = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock), }; -void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr) +void init_kstat_irqs(struct irq_desc *desc, int node, int nr) { - int node; void *ptr; - node = cpu_to_node(cpu); ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), GFP_ATOMIC, node); /* @@ -94,33 +92,32 @@ void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr) * init_copy_kstat_irqs() could still use old one */ if (ptr) { - printk(KERN_DEBUG " alloc kstat_irqs on cpu %d node %d\n", - cpu, node); + printk(KERN_DEBUG " alloc kstat_irqs on node %d\n", node); desc->kstat_irqs = ptr; } } -static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu) +static void init_one_irq_desc(int irq, struct irq_desc *desc, int node) { memcpy(desc, &irq_desc_init, sizeof(struct irq_desc)); spin_lock_init(&desc->lock); desc->irq = irq; #ifdef CONFIG_SMP - desc->cpu = cpu; + desc->node = node; #endif lockdep_set_class(&desc->lock, &irq_desc_lock_class); - init_kstat_irqs(desc, cpu, nr_cpu_ids); + init_kstat_irqs(desc, node, nr_cpu_ids); if (!desc->kstat_irqs) { printk(KERN_ERR "can not alloc kstat_irqs\n"); BUG_ON(1); } - if (!alloc_desc_masks(desc, cpu, false)) { + if (!alloc_desc_masks(desc, node, false)) { printk(KERN_ERR "can not alloc irq_desc cpumasks\n"); BUG_ON(1); } init_desc_masks(desc); - arch_init_chip_data(desc, cpu); + arch_init_chip_data(desc, node); } /* @@ -189,11 +186,10 @@ struct irq_desc *irq_to_desc(unsigned int irq) return NULL; } -struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu) +struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node) { struct irq_desc *desc; unsigned long flags; - int node; if (irq >= nr_irqs) { WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n", @@ -212,15 +208,13 @@ struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu) if (desc) goto out_unlock; - node = cpu_to_node(cpu); desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node); - printk(KERN_DEBUG " alloc irq_desc for %d on cpu %d node %d\n", - irq, cpu, node); + printk(KERN_DEBUG " alloc irq_desc for %d on node %d\n", irq, node); if (!desc) { printk(KERN_ERR "can not alloc irq_desc\n"); BUG_ON(1); } - init_one_irq_desc(irq, desc, cpu); + init_one_irq_desc(irq, desc, node); irq_desc_ptrs[irq] = desc; @@ -270,7 +264,7 @@ struct irq_desc *irq_to_desc(unsigned int irq) return (irq < NR_IRQS) ? irq_desc + irq : NULL; } -struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu) +struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node) { return irq_to_desc(irq); } diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h index de5f412f6a92..73468253143b 100644 --- a/kernel/irq/internals.h +++ b/kernel/irq/internals.h @@ -16,7 +16,7 @@ extern void __disable_irq(struct irq_desc *desc, unsigned int irq, bool susp); extern void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume); extern struct lock_class_key irq_desc_lock_class; -extern void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr); +extern void init_kstat_irqs(struct irq_desc *desc, int node, int nr); extern void clear_kstat_irqs(struct irq_desc *desc); extern spinlock_t sparse_irq_lock; diff --git a/kernel/irq/numa_migrate.c b/kernel/irq/numa_migrate.c index ce72bc3f4ced..2f69bee57bf2 100644 --- a/kernel/irq/numa_migrate.c +++ b/kernel/irq/numa_migrate.c @@ -15,9 +15,9 @@ static void init_copy_kstat_irqs(struct irq_desc *old_desc, struct irq_desc *desc, - int cpu, int nr) + int node, int nr) { - init_kstat_irqs(desc, cpu, nr); + init_kstat_irqs(desc, node, nr); if (desc->kstat_irqs != old_desc->kstat_irqs) memcpy(desc->kstat_irqs, old_desc->kstat_irqs, @@ -34,20 +34,20 @@ static void free_kstat_irqs(struct irq_desc *old_desc, struct irq_desc *desc) } static bool init_copy_one_irq_desc(int irq, struct irq_desc *old_desc, - struct irq_desc *desc, int cpu) + struct irq_desc *desc, int node) { memcpy(desc, old_desc, sizeof(struct irq_desc)); - if (!alloc_desc_masks(desc, cpu, false)) { + if (!alloc_desc_masks(desc, node, false)) { printk(KERN_ERR "irq %d: can not get new irq_desc cpumask " "for migration.\n", irq); return false; } spin_lock_init(&desc->lock); - desc->cpu = cpu; + desc->node = node; lockdep_set_class(&desc->lock, &irq_desc_lock_class); - init_copy_kstat_irqs(old_desc, desc, cpu, nr_cpu_ids); + init_copy_kstat_irqs(old_desc, desc, node, nr_cpu_ids); init_copy_desc_masks(old_desc, desc); - arch_init_copy_chip_data(old_desc, desc, cpu); + arch_init_copy_chip_data(old_desc, desc, node); return true; } @@ -59,12 +59,11 @@ static void free_one_irq_desc(struct irq_desc *old_desc, struct irq_desc *desc) } static struct irq_desc *__real_move_irq_desc(struct irq_desc *old_desc, - int cpu) + int node) { struct irq_desc *desc; unsigned int irq; unsigned long flags; - int node; irq = old_desc->irq; @@ -76,7 +75,6 @@ static struct irq_desc *__real_move_irq_desc(struct irq_desc *old_desc, if (desc && old_desc != desc) goto out_unlock; - node = cpu_to_node(cpu); desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node); if (!desc) { printk(KERN_ERR "irq %d: can not get new irq_desc " @@ -85,7 +83,7 @@ static struct irq_desc *__real_move_irq_desc(struct irq_desc *old_desc, desc = old_desc; goto out_unlock; } - if (!init_copy_one_irq_desc(irq, old_desc, desc, cpu)) { + if (!init_copy_one_irq_desc(irq, old_desc, desc, node)) { /* still use old one */ kfree(desc); desc = old_desc; @@ -107,24 +105,14 @@ out_unlock: return desc; } -struct irq_desc *move_irq_desc(struct irq_desc *desc, int cpu) +struct irq_desc *move_irq_desc(struct irq_desc *desc, int node) { - int old_cpu; - int node, old_node; - /* those all static, do move them */ if (desc->irq < NR_IRQS_LEGACY) return desc; - old_cpu = desc->cpu; - if (old_cpu != cpu) { - node = cpu_to_node(cpu); - old_node = cpu_to_node(old_cpu); - if (old_node != node) - desc = __real_move_irq_desc(desc, cpu); - else - desc->cpu = cpu; - } + if (desc->node != node) + desc = __real_move_irq_desc(desc, node); return desc; } diff --git a/kernel/softirq.c b/kernel/softirq.c index b525dd348511..f674f332a024 100644 --- a/kernel/softirq.c +++ b/kernel/softirq.c @@ -828,7 +828,7 @@ int __init __weak arch_early_irq_init(void) return 0; } -int __weak arch_init_chip_data(struct irq_desc *desc, int cpu) +int __weak arch_init_chip_data(struct irq_desc *desc, int node) { return 0; } -- cgit v1.2.3 From 50fa610a3b6ba7cf91d7a92229177dfaff2b81a1 Mon Sep 17 00:00:00 2001 From: David Howells Date: Tue, 28 Apr 2009 15:01:38 +0100 Subject: sched: Document memory barriers implied by sleep/wake-up primitives Add a section to the memory barriers document to note the implied memory barriers of sleep primitives (set_current_state() and wrappers) and wake-up primitives (wake_up() and co.). Also extend the in-code comments on the wake_up() functions to note these implied barriers. [ Impact: add documentation ] Signed-off-by: David Howells Cc: Oleg Nesterov Cc: Linus Torvalds Cc: Andrew Morton LKML-Reference: <20090428140138.1192.94723.stgit@warthog.procyon.org.uk> Signed-off-by: Ingo Molnar --- kernel/sched.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index b902e587a3a0..fd0c2cee3f35 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2458,6 +2458,17 @@ out: return success; } +/** + * wake_up_process - Wake up a specific process + * @p: The process to be woken up. + * + * Attempt to wake up the nominated process and move it to the set of runnable + * processes. Returns 1 if the process was woken up, 0 if it was already + * running. + * + * It may be assumed that this function implies a write memory barrier before + * changing the task state if and only if any tasks are woken up. + */ int wake_up_process(struct task_struct *p) { return try_to_wake_up(p, TASK_ALL, 0); @@ -5241,6 +5252,9 @@ void __wake_up_common(wait_queue_head_t *q, unsigned int mode, * @mode: which threads * @nr_exclusive: how many wake-one or wake-many threads to wake up * @key: is directly passed to the wakeup function + * + * It may be assumed that this function implies a write memory barrier before + * changing the task state if and only if any tasks are woken up. */ void __wake_up(wait_queue_head_t *q, unsigned int mode, int nr_exclusive, void *key) @@ -5279,6 +5293,9 @@ void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) * with each other. This can prevent needless bouncing between CPUs. * * On UP it can prevent extra preemption. + * + * It may be assumed that this function implies a write memory barrier before + * changing the task state if and only if any tasks are woken up. */ void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, int nr_exclusive, void *key) @@ -5315,6 +5332,9 @@ EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ * awakened in the same order in which they were queued. * * See also complete_all(), wait_for_completion() and related routines. + * + * It may be assumed that this function implies a write memory barrier before + * changing the task state if and only if any tasks are woken up. */ void complete(struct completion *x) { @@ -5332,6 +5352,9 @@ EXPORT_SYMBOL(complete); * @x: holds the state of this particular completion * * This will wake up all threads waiting on this particular completion event. + * + * It may be assumed that this function implies a write memory barrier before + * changing the task state if and only if any tasks are woken up. */ void complete_all(struct completion *x) { -- cgit v1.2.3 From 4aeb0b4239bb3b67ed402cb9cef3e000c892cadf Mon Sep 17 00:00:00 2001 From: Robert Richter Date: Wed, 29 Apr 2009 12:47:03 +0200 Subject: perfcounters: rename struct hw_perf_counter_ops into struct pmu This patch renames struct hw_perf_counter_ops into struct pmu. It introduces a structure to describe a cpu specific pmu (performance monitoring unit). It may contain ops and data. The new name of the structure fits better, is shorter, and thus better to handle. Where it was appropriate, names of function and variable have been changed too. [ Impact: cleanup ] Signed-off-by: Robert Richter Cc: Paul Mackerras Acked-by: Peter Zijlstra LKML-Reference: <1241002046-8832-7-git-send-email-robert.richter@amd.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 68 ++++++++++++++++++++++++--------------------------- 1 file changed, 32 insertions(+), 36 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 09396098dd0d..582108addefa 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -52,8 +52,7 @@ static DEFINE_MUTEX(perf_resource_mutex); /* * Architecture provided APIs - weak aliases: */ -extern __weak const struct hw_perf_counter_ops * -hw_perf_counter_init(struct perf_counter *counter) +extern __weak const struct pmu *hw_perf_counter_init(struct perf_counter *counter) { return NULL; } @@ -124,7 +123,7 @@ counter_sched_out(struct perf_counter *counter, counter->state = PERF_COUNTER_STATE_INACTIVE; counter->tstamp_stopped = ctx->time; - counter->hw_ops->disable(counter); + counter->pmu->disable(counter); counter->oncpu = -1; if (!is_software_counter(counter)) @@ -417,7 +416,7 @@ counter_sched_in(struct perf_counter *counter, */ smp_wmb(); - if (counter->hw_ops->enable(counter)) { + if (counter->pmu->enable(counter)) { counter->state = PERF_COUNTER_STATE_INACTIVE; counter->oncpu = -1; return -EAGAIN; @@ -1096,7 +1095,7 @@ static void __read(void *info) local_irq_save(flags); if (ctx->is_active) update_context_time(ctx); - counter->hw_ops->read(counter); + counter->pmu->read(counter); update_counter_times(counter); local_irq_restore(flags); } @@ -1922,7 +1921,7 @@ static void perf_counter_output(struct perf_counter *counter, leader = counter->group_leader; list_for_each_entry(sub, &leader->sibling_list, list_entry) { if (sub != counter) - sub->hw_ops->read(sub); + sub->pmu->read(sub); group_entry.event = sub->hw_event.config; group_entry.counter = atomic64_read(&sub->count); @@ -2264,7 +2263,7 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) struct pt_regs *regs; counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); - counter->hw_ops->read(counter); + counter->pmu->read(counter); regs = get_irq_regs(); /* @@ -2410,7 +2409,7 @@ static void perf_swcounter_disable(struct perf_counter *counter) perf_swcounter_update(counter); } -static const struct hw_perf_counter_ops perf_ops_generic = { +static const struct pmu perf_ops_generic = { .enable = perf_swcounter_enable, .disable = perf_swcounter_disable, .read = perf_swcounter_read, @@ -2460,7 +2459,7 @@ static void cpu_clock_perf_counter_read(struct perf_counter *counter) cpu_clock_perf_counter_update(counter); } -static const struct hw_perf_counter_ops perf_ops_cpu_clock = { +static const struct pmu perf_ops_cpu_clock = { .enable = cpu_clock_perf_counter_enable, .disable = cpu_clock_perf_counter_disable, .read = cpu_clock_perf_counter_read, @@ -2522,7 +2521,7 @@ static void task_clock_perf_counter_read(struct perf_counter *counter) task_clock_perf_counter_update(counter, time); } -static const struct hw_perf_counter_ops perf_ops_task_clock = { +static const struct pmu perf_ops_task_clock = { .enable = task_clock_perf_counter_enable, .disable = task_clock_perf_counter_disable, .read = task_clock_perf_counter_read, @@ -2574,7 +2573,7 @@ static void cpu_migrations_perf_counter_disable(struct perf_counter *counter) cpu_migrations_perf_counter_update(counter); } -static const struct hw_perf_counter_ops perf_ops_cpu_migrations = { +static const struct pmu perf_ops_cpu_migrations = { .enable = cpu_migrations_perf_counter_enable, .disable = cpu_migrations_perf_counter_disable, .read = cpu_migrations_perf_counter_read, @@ -2600,8 +2599,7 @@ static void tp_perf_counter_destroy(struct perf_counter *counter) ftrace_profile_disable(perf_event_id(&counter->hw_event)); } -static const struct hw_perf_counter_ops * -tp_perf_counter_init(struct perf_counter *counter) +static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) { int event_id = perf_event_id(&counter->hw_event); int ret; @@ -2616,18 +2614,16 @@ tp_perf_counter_init(struct perf_counter *counter) return &perf_ops_generic; } #else -static const struct hw_perf_counter_ops * -tp_perf_counter_init(struct perf_counter *counter) +static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) { return NULL; } #endif -static const struct hw_perf_counter_ops * -sw_perf_counter_init(struct perf_counter *counter) +static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) { struct perf_counter_hw_event *hw_event = &counter->hw_event; - const struct hw_perf_counter_ops *hw_ops = NULL; + const struct pmu *pmu = NULL; struct hw_perf_counter *hwc = &counter->hw; /* @@ -2639,7 +2635,7 @@ sw_perf_counter_init(struct perf_counter *counter) */ switch (perf_event_id(&counter->hw_event)) { case PERF_COUNT_CPU_CLOCK: - hw_ops = &perf_ops_cpu_clock; + pmu = &perf_ops_cpu_clock; if (hw_event->irq_period && hw_event->irq_period < 10000) hw_event->irq_period = 10000; @@ -2650,9 +2646,9 @@ sw_perf_counter_init(struct perf_counter *counter) * use the cpu_clock counter instead. */ if (counter->ctx->task) - hw_ops = &perf_ops_task_clock; + pmu = &perf_ops_task_clock; else - hw_ops = &perf_ops_cpu_clock; + pmu = &perf_ops_cpu_clock; if (hw_event->irq_period && hw_event->irq_period < 10000) hw_event->irq_period = 10000; @@ -2661,18 +2657,18 @@ sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_PAGE_FAULTS_MIN: case PERF_COUNT_PAGE_FAULTS_MAJ: case PERF_COUNT_CONTEXT_SWITCHES: - hw_ops = &perf_ops_generic; + pmu = &perf_ops_generic; break; case PERF_COUNT_CPU_MIGRATIONS: if (!counter->hw_event.exclude_kernel) - hw_ops = &perf_ops_cpu_migrations; + pmu = &perf_ops_cpu_migrations; break; } - if (hw_ops) + if (pmu) hwc->irq_period = hw_event->irq_period; - return hw_ops; + return pmu; } /* @@ -2685,7 +2681,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, struct perf_counter *group_leader, gfp_t gfpflags) { - const struct hw_perf_counter_ops *hw_ops; + const struct pmu *pmu; struct perf_counter *counter; long err; @@ -2713,46 +2709,46 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->cpu = cpu; counter->hw_event = *hw_event; counter->group_leader = group_leader; - counter->hw_ops = NULL; + counter->pmu = NULL; counter->ctx = ctx; counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) counter->state = PERF_COUNTER_STATE_OFF; - hw_ops = NULL; + pmu = NULL; if (perf_event_raw(hw_event)) { - hw_ops = hw_perf_counter_init(counter); + pmu = hw_perf_counter_init(counter); goto done; } switch (perf_event_type(hw_event)) { case PERF_TYPE_HARDWARE: - hw_ops = hw_perf_counter_init(counter); + pmu = hw_perf_counter_init(counter); break; case PERF_TYPE_SOFTWARE: - hw_ops = sw_perf_counter_init(counter); + pmu = sw_perf_counter_init(counter); break; case PERF_TYPE_TRACEPOINT: - hw_ops = tp_perf_counter_init(counter); + pmu = tp_perf_counter_init(counter); break; } done: err = 0; - if (!hw_ops) + if (!pmu) err = -EINVAL; - else if (IS_ERR(hw_ops)) - err = PTR_ERR(hw_ops); + else if (IS_ERR(pmu)) + err = PTR_ERR(pmu); if (err) { kfree(counter); return ERR_PTR(err); } - counter->hw_ops = hw_ops; + counter->pmu = pmu; if (counter->hw_event.mmap) atomic_inc(&nr_mmap_tracking); -- cgit v1.2.3 From 98144511427c192e4249ff66a3f9debc55c59411 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 29 Apr 2009 14:52:50 +0200 Subject: perf_counter: add/update copyrights Acked-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 582108addefa..a95a171e608c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1,9 +1,9 @@ /* * Performance counter core code * - * Copyright(C) 2008 Thomas Gleixner - * Copyright(C) 2008 Red Hat, Inc., Ingo Molnar - * + * Copyright (C) 2008 Thomas Gleixner + * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar + * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra * * For licensing details see kernel-base/COPYING */ -- cgit v1.2.3 From 23b94b967f118bef941369238f33c8140be46539 Mon Sep 17 00:00:00 2001 From: Luis Henriques Date: Wed, 29 Apr 2009 21:54:51 +0100 Subject: locking, rtmutex.c: Documentation cleanup Two minor updates on functions documentation: - Updated documentation for function rt_mutex_unlock(), which contained an incorrect name - Removed extra '*' from comment in function rt_mutex_destroy() [ Impact: cleanup ] Signed-off-by: Luis Henriques Cc: Steven Rostedt LKML-Reference: <20090429205451.GA23154@hades.domain.com> Signed-off-by: Ingo Molnar --- kernel/rtmutex.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/rtmutex.c b/kernel/rtmutex.c index 69d9cb921ffa..013882e83497 100644 --- a/kernel/rtmutex.c +++ b/kernel/rtmutex.c @@ -864,9 +864,9 @@ int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock, EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible); /** - * rt_mutex_lock_interruptible_ktime - lock a rt_mutex interruptible - * the timeout structure is provided - * by the caller + * rt_mutex_timed_lock - lock a rt_mutex interruptible + * the timeout structure is provided + * by the caller * * @lock: the rt_mutex to be locked * @timeout: timeout structure or NULL (no timeout) @@ -913,7 +913,7 @@ void __sched rt_mutex_unlock(struct rt_mutex *lock) } EXPORT_SYMBOL_GPL(rt_mutex_unlock); -/*** +/** * rt_mutex_destroy - mark a mutex unusable * @lock: the mutex to be destroyed * -- cgit v1.2.3 From 3bcac0263f0b45e67a64034ebcb69eb9abb742f4 Mon Sep 17 00:00:00 2001 From: David Howells Date: Wed, 29 Apr 2009 13:45:05 +0100 Subject: SELinux: Don't flush inherited SIGKILL during execve() Don't flush inherited SIGKILL during execve() in SELinux's post cred commit hook. This isn't really a security problem: if the SIGKILL came before the credentials were changed, then we were right to receive it at the time, and should honour it; if it came after the creds were changed, then we definitely should honour it; and in any case, all that will happen is that the process will be scrapped before it ever returns to userspace. Signed-off-by: David Howells Signed-off-by: Oleg Nesterov Signed-off-by: James Morris --- kernel/signal.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/signal.c b/kernel/signal.c index 1c8814481a11..f93efec14ff5 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -238,14 +238,19 @@ void flush_sigqueue(struct sigpending *queue) /* * Flush all pending signals for a task. */ +void __flush_signals(struct task_struct *t) +{ + clear_tsk_thread_flag(t, TIF_SIGPENDING); + flush_sigqueue(&t->pending); + flush_sigqueue(&t->signal->shared_pending); +} + void flush_signals(struct task_struct *t) { unsigned long flags; spin_lock_irqsave(&t->sighand->siglock, flags); - clear_tsk_thread_flag(t, TIF_SIGPENDING); - flush_sigqueue(&t->pending); - flush_sigqueue(&t->signal->shared_pending); + __flush_signals(t); spin_unlock_irqrestore(&t->sighand->siglock, flags); } -- cgit v1.2.3 From c5dd016cdf0a040e1de0b691e274fbfe642b2cdc Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Thu, 30 Apr 2009 09:48:16 +1000 Subject: perf_counter: update copyright notice This adds my name to the list of copyright holders on the core perf_counter.c, since I have contributed a significant amount of the code in there. Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Robert Richter LKML-Reference: <18936.59200.888049.746658@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index a95a171e608c..75f2b6c82392 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -4,6 +4,7 @@ * Copyright (C) 2008 Thomas Gleixner * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra + * Copyright © 2009 Paul Mackerras, IBM Corp. * * For licensing details see kernel-base/COPYING */ -- cgit v1.2.3 From a511e3f968c462a55ef58697257f5347c73d306e Mon Sep 17 00:00:00 2001 From: Andrew Morton Date: Wed, 29 Apr 2009 15:59:58 -0700 Subject: mutex: add atomic_dec_and_mutex_lock(), fix include/linux/mutex.h:136: warning: 'mutex_lock' declared inline after being called include/linux/mutex.h:136: warning: previous declaration of 'mutex_lock' was here uninline it. [ Impact: clean up and uninline, address compiler warning ] Signed-off-by: Andrew Morton Cc: Al Viro Cc: Christoph Hellwig Cc: Eric Paris Cc: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <200904292318.n3TNIsi6028340@imap1.linux-foundation.org> Signed-off-by: Ingo Molnar --- kernel/mutex.c | 25 ++++++++++++++++++++++++- 1 file changed, 24 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/mutex.c b/kernel/mutex.c index 507cf2b5e9f1..e2d25e9e62ae 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -471,5 +471,28 @@ int __sched mutex_trylock(struct mutex *lock) return ret; } - EXPORT_SYMBOL(mutex_trylock); + +/** + * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0 + * @cnt: the atomic which we are to dec + * @lock: the mutex to return holding if we dec to 0 + * + * return true and hold lock if we dec to 0, return false otherwise + */ +int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock) +{ + /* dec if we can't possibly hit 0 */ + if (atomic_add_unless(cnt, -1, 1)) + return 0; + /* we might hit 0, so take the lock */ + mutex_lock(lock); + if (!atomic_dec_and_test(cnt)) { + /* when we actually did the dec, we didn't hit 0 */ + mutex_unlock(lock); + return 0; + } + /* we hit 0, and we hold the lock */ + return 1; +} +EXPORT_SYMBOL(atomic_dec_and_mutex_lock); -- cgit v1.2.3 From ba9c22f2c01cf5c88beed5a6b9e07d42e10bd358 Mon Sep 17 00:00:00 2001 From: Darren Hart Date: Mon, 20 Apr 2009 22:22:22 -0700 Subject: futex: remove FUTEX_REQUEUE_PI (non CMP) The new requeue PI futex op codes were modeled after the existing FUTEX_REQUEUE and FUTEX_CMP_REQUEUE calls. I was unaware at the time that FUTEX_REQUEUE was only around for compatibility reasons and shouldn't be used in new code. Ulrich Drepper elaborates on this in his Futexes are Tricky paper: http://people.redhat.com/drepper/futex.pdf. The deprecated call doesn't catch changes to the futex corresponding to the destination futex which can lead to deadlock. Therefor, I feel it best to remove FUTEX_REQUEUE_PI and leave only FUTEX_CMP_REQUEUE_PI as there are not yet any existing users of the API. This patch does change the OP code value of FUTEX_CMP_REQUEUE_PI to 12 from 13. Since my test case is the only known user of this API, I felt this was the right thing to do, rather than leave a hole in the enumeration. I chose to continue using the _CMP_ modifier in the OP code to make it explicit to the user that the test is being done. Builds, boots, and ran several hundred iterations requeue_pi.c. Signed-off-by: Darren Hart LKML-Reference: <49ED580E.1050502@us.ibm.com> Signed-off-by: Thomas Gleixner --- kernel/futex.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 6d2daa46f9ff..aec8bf89bf4e 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -2555,9 +2555,6 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, ret = futex_wait_requeue_pi(uaddr, fshared, val, timeout, val3, clockrt, uaddr2); break; - case FUTEX_REQUEUE_PI: - ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, NULL, 1); - break; case FUTEX_CMP_REQUEUE_PI: ret = futex_requeue(uaddr, fshared, uaddr2, val, val2, &val3, 1); @@ -2596,8 +2593,7 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, * number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP. */ if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE || - cmd == FUTEX_REQUEUE_PI || cmd == FUTEX_CMP_REQUEUE_PI || - cmd == FUTEX_WAKE_OP) + cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP) val2 = (u32) (unsigned long) utime; return do_futex(uaddr, op, val, tp, uaddr2, val2, val3); -- cgit v1.2.3 From 30b4ae8a4498543863501f707879b7220b649602 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 4 Apr 2009 21:01:01 +0000 Subject: signals: split do_tkill Split out the code from do_tkill to make it reusable by the follow up patch which implements sys_rt_tgsigqueueinfo Signed-off-by: Thomas Gleixner Reviewed-by: Oleg Nesterov --- kernel/signal.c | 30 ++++++++++++++++++------------ 1 file changed, 18 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/signal.c b/kernel/signal.c index d8034737db4c..56d27acad87e 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -2278,24 +2278,17 @@ SYSCALL_DEFINE2(kill, pid_t, pid, int, sig) return kill_something_info(sig, &info, pid); } -static int do_tkill(pid_t tgid, pid_t pid, int sig) +static int +do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info) { - int error; - struct siginfo info; struct task_struct *p; unsigned long flags; - - error = -ESRCH; - info.si_signo = sig; - info.si_errno = 0; - info.si_code = SI_TKILL; - info.si_pid = task_tgid_vnr(current); - info.si_uid = current_uid(); + int error = -ESRCH; rcu_read_lock(); p = find_task_by_vpid(pid); if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) { - error = check_kill_permission(sig, &info, p); + error = check_kill_permission(sig, info, p); /* * The null signal is a permissions and process existence * probe. No signal is actually delivered. @@ -2305,7 +2298,7 @@ static int do_tkill(pid_t tgid, pid_t pid, int sig) * signal is private anyway. */ if (!error && sig && lock_task_sighand(p, &flags)) { - error = specific_send_sig_info(sig, &info, p); + error = specific_send_sig_info(sig, info, p); unlock_task_sighand(p, &flags); } } @@ -2314,6 +2307,19 @@ static int do_tkill(pid_t tgid, pid_t pid, int sig) return error; } +static int do_tkill(pid_t tgid, pid_t pid, int sig) +{ + struct siginfo info; + + info.si_signo = sig; + info.si_errno = 0; + info.si_code = SI_TKILL; + info.si_pid = task_tgid_vnr(current); + info.si_uid = current_uid(); + + return do_send_specific(tgid, pid, sig, &info); +} + /** * sys_tgkill - send signal to one specific thread * @tgid: the thread group ID of the thread -- cgit v1.2.3 From 62ab4505e3efaf67784f84059e0fb9cedb1728ea Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 4 Apr 2009 21:01:06 +0000 Subject: signals: implement sys_rt_tgsigqueueinfo sys_kill has the per thread counterpart sys_tgkill. sigqueueinfo is missing a thread directed counterpart. Such an interface is important for migrating applications from other OSes which have the per thread delivery implemented. Signed-off-by: Thomas Gleixner Reviewed-by: Oleg Nesterov Acked-by: Roland McGrath Acked-by: Ulrich Drepper --- kernel/compat.c | 11 +++++++++++ kernel/signal.c | 26 ++++++++++++++++++++++++++ 2 files changed, 37 insertions(+) (limited to 'kernel') diff --git a/kernel/compat.c b/kernel/compat.c index 42d56544460f..f6c204f07ea6 100644 --- a/kernel/compat.c +++ b/kernel/compat.c @@ -882,6 +882,17 @@ compat_sys_rt_sigtimedwait (compat_sigset_t __user *uthese, } +asmlinkage long +compat_sys_rt_tgsigqueueinfo(compat_pid_t tgid, compat_pid_t pid, int sig, + struct compat_siginfo __user *uinfo) +{ + siginfo_t info; + + if (copy_siginfo_from_user32(&info, uinfo)) + return -EFAULT; + return do_rt_tgsigqueueinfo(tgid, pid, sig, &info); +} + #ifdef __ARCH_WANT_COMPAT_SYS_TIME /* compat_time_t is a 32 bit "long" and needs to get converted. */ diff --git a/kernel/signal.c b/kernel/signal.c index 56d27acad87e..f79b3b9f8375 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -2369,6 +2369,32 @@ SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig, return kill_proc_info(sig, &info, pid); } +long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info) +{ + /* This is only valid for single tasks */ + if (pid <= 0 || tgid <= 0) + return -EINVAL; + + /* Not even root can pretend to send signals from the kernel. + Nor can they impersonate a kill(), which adds source info. */ + if (info->si_code >= 0) + return -EPERM; + info->si_signo = sig; + + return do_send_specific(tgid, pid, sig, info); +} + +SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig, + siginfo_t __user *, uinfo) +{ + siginfo_t info; + + if (copy_from_user(&info, uinfo, sizeof(siginfo_t))) + return -EFAULT; + + return do_rt_tgsigqueueinfo(tgid, pid, sig, &info); +} + int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact) { struct task_struct *t = current; -- cgit v1.2.3 From 78a3d9d5654a7fd99cf8b2ab06b9497b9c7aad64 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Wed, 29 Apr 2009 18:01:23 +0200 Subject: do_wait: do take security_task_wait() into account I was never able to understand what should we actually do when security_task_wait() fails, but the current code doesn't look right. If ->task_wait() returns the error, we update *notask_error correctly. But then we either reap the child (despite the fact this was forbidden) or clear *notask_error (and hide the securiy policy problems). This patch assumes that "stolen by ptrace" doesn't matter. If selinux denies the child we should ignore it but make sure we report -EACCESS instead of -ECHLD if there are no other eligible children. Signed-off-by: Oleg Nesterov Acked-by: Roland McGrath Signed-off-by: James Morris --- kernel/exit.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index 167e1e3ad7c6..d2e8239ea187 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -1582,6 +1582,7 @@ static int wait_consider_task(struct task_struct *parent, int ptrace, */ if (*notask_error) *notask_error = ret; + return 0; } if (likely(!ptrace) && unlikely(p->ptrace)) { -- cgit v1.2.3 From c33a0bc4e41ef169d6e807d8abb9502544b518e5 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 1 May 2009 12:23:16 +0200 Subject: perf_counter: fix race in perf_output_* When two (or more) contexts output to the same buffer, it is possible to observe half written output. Suppose we have CPU0 doing perf_counter_mmap(), CPU1 doing perf_counter_overflow(). If CPU1 does a wakeup and exposes head to user-space, then CPU2 can observe the data CPU0 is still writing. [ Impact: fix occasionally corrupted profiling records ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090501102533.007821627@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 130 +++++++++++++++++++++++++++++++++++++++----------- 1 file changed, 101 insertions(+), 29 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 75f2b6c82392..8660ae579530 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1279,14 +1279,12 @@ static unsigned int perf_poll(struct file *file, poll_table *wait) { struct perf_counter *counter = file->private_data; struct perf_mmap_data *data; - unsigned int events; + unsigned int events = POLL_HUP; rcu_read_lock(); data = rcu_dereference(counter->data); if (data) - events = atomic_xchg(&data->wakeup, 0); - else - events = POLL_HUP; + events = atomic_xchg(&data->poll, 0); rcu_read_unlock(); poll_wait(file, &counter->waitq, wait); @@ -1568,22 +1566,6 @@ static const struct file_operations perf_fops = { void perf_counter_wakeup(struct perf_counter *counter) { - struct perf_mmap_data *data; - - rcu_read_lock(); - data = rcu_dereference(counter->data); - if (data) { - atomic_set(&data->wakeup, POLL_IN); - /* - * Ensure all data writes are issued before updating the - * user-space data head information. The matching rmb() - * will be in userspace after reading this value. - */ - smp_wmb(); - data->user_page->data_head = atomic_read(&data->head); - } - rcu_read_unlock(); - wake_up_all(&counter->waitq); if (counter->pending_kill) { @@ -1721,10 +1703,14 @@ struct perf_output_handle { int wakeup; int nmi; int overflow; + int locked; + unsigned long flags; }; -static inline void __perf_output_wakeup(struct perf_output_handle *handle) +static void perf_output_wakeup(struct perf_output_handle *handle) { + atomic_set(&handle->data->poll, POLL_IN); + if (handle->nmi) { handle->counter->pending_wakeup = 1; perf_pending_queue(&handle->counter->pending, @@ -1733,6 +1719,86 @@ static inline void __perf_output_wakeup(struct perf_output_handle *handle) perf_counter_wakeup(handle->counter); } +/* + * Curious locking construct. + * + * We need to ensure a later event doesn't publish a head when a former + * event isn't done writing. However since we need to deal with NMIs we + * cannot fully serialize things. + * + * What we do is serialize between CPUs so we only have to deal with NMI + * nesting on a single CPU. + * + * We only publish the head (and generate a wakeup) when the outer-most + * event completes. + */ +static void perf_output_lock(struct perf_output_handle *handle) +{ + struct perf_mmap_data *data = handle->data; + int cpu; + + handle->locked = 0; + + local_irq_save(handle->flags); + cpu = smp_processor_id(); + + if (in_nmi() && atomic_read(&data->lock) == cpu) + return; + + while (atomic_cmpxchg(&data->lock, 0, cpu) != 0) + cpu_relax(); + + handle->locked = 1; +} + +static void perf_output_unlock(struct perf_output_handle *handle) +{ + struct perf_mmap_data *data = handle->data; + int head, cpu; + + if (handle->wakeup) + data->wakeup_head = data->head; + + if (!handle->locked) + goto out; + +again: + /* + * The xchg implies a full barrier that ensures all writes are done + * before we publish the new head, matched by a rmb() in userspace when + * reading this position. + */ + while ((head = atomic_xchg(&data->wakeup_head, 0))) { + data->user_page->data_head = head; + handle->wakeup = 1; + } + + /* + * NMI can happen here, which means we can miss a wakeup_head update. + */ + + cpu = atomic_xchg(&data->lock, 0); + WARN_ON_ONCE(cpu != smp_processor_id()); + + /* + * Therefore we have to validate we did not indeed do so. + */ + if (unlikely(atomic_read(&data->wakeup_head))) { + /* + * Since we had it locked, we can lock it again. + */ + while (atomic_cmpxchg(&data->lock, 0, cpu) != 0) + cpu_relax(); + + goto again; + } + + if (handle->wakeup) + perf_output_wakeup(handle); +out: + local_irq_restore(handle->flags); +} + static int perf_output_begin(struct perf_output_handle *handle, struct perf_counter *counter, unsigned int size, int nmi, int overflow) @@ -1745,6 +1811,7 @@ static int perf_output_begin(struct perf_output_handle *handle, if (!data) goto out; + handle->data = data; handle->counter = counter; handle->nmi = nmi; handle->overflow = overflow; @@ -1752,12 +1819,13 @@ static int perf_output_begin(struct perf_output_handle *handle, if (!data->nr_pages) goto fail; + perf_output_lock(handle); + do { offset = head = atomic_read(&data->head); head += size; } while (atomic_cmpxchg(&data->head, offset, head) != offset); - handle->data = data; handle->offset = offset; handle->head = head; handle->wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); @@ -1765,7 +1833,7 @@ static int perf_output_begin(struct perf_output_handle *handle, return 0; fail: - __perf_output_wakeup(handle); + perf_output_wakeup(handle); out: rcu_read_unlock(); @@ -1809,16 +1877,20 @@ static void perf_output_copy(struct perf_output_handle *handle, static void perf_output_end(struct perf_output_handle *handle) { - int wakeup_events = handle->counter->hw_event.wakeup_events; + struct perf_counter *counter = handle->counter; + struct perf_mmap_data *data = handle->data; + + int wakeup_events = counter->hw_event.wakeup_events; if (handle->overflow && wakeup_events) { - int events = atomic_inc_return(&handle->data->events); + int events = atomic_inc_return(&data->events); if (events >= wakeup_events) { - atomic_sub(wakeup_events, &handle->data->events); - __perf_output_wakeup(handle); + atomic_sub(wakeup_events, &data->events); + handle->wakeup = 1; } - } else if (handle->wakeup) - __perf_output_wakeup(handle); + } + + perf_output_unlock(handle); rcu_read_unlock(); } -- cgit v1.2.3 From 15e957d08dd4a841359cfec59ecb74041e0097aa Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Thu, 30 Apr 2009 01:17:50 -0700 Subject: x86/irq: use move_irq_desc() in create_irq_nr() move_irq_desc() will try to move irq_desc to the home node if the allocated one is not correct, in create_irq_nr(). ( This can happen on devices that are on different nodes that are using MSI, when drivers are loaded and unloaded randomly. ) v2: fix non-smp build v3: add NUMA_IRQ_DESC to eliminate #ifdefs [ Impact: improve irq descriptor locality on NUMA systems ] Signed-off-by: Yinghai Lu Cc: Andrew Morton Cc: Suresh Siddha Cc: "Eric W. Biederman" Cc: Rusty Russell LKML-Reference: <49F95EAE.2050903@kernel.org> Signed-off-by: Ingo Molnar --- kernel/irq/Makefile | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/irq/Makefile b/kernel/irq/Makefile index 2f065277f8ee..7d047808419d 100644 --- a/kernel/irq/Makefile +++ b/kernel/irq/Makefile @@ -3,5 +3,5 @@ obj-y := handle.o manage.o spurious.o resend.o chip.o devres.o obj-$(CONFIG_GENERIC_IRQ_PROBE) += autoprobe.o obj-$(CONFIG_PROC_FS) += proc.o obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o -obj-$(CONFIG_SPARSE_IRQ) += numa_migrate.o +obj-$(CONFIG_NUMA_IRQ_DESC) += numa_migrate.o obj-$(CONFIG_PM_SLEEP) += pm.o -- cgit v1.2.3 From a52f5c5620673c292cb159205bf0e1eb5af1985b Mon Sep 17 00:00:00 2001 From: Dmitri Vorobiev Date: Fri, 1 May 2009 13:10:21 -0700 Subject: clockevents: tick_broadcast_device can become static The variable tick_broadcast_device is not used outside of the file where it is defined, so let's make it static. Signed-off-by: Dmitri Vorobiev Signed-off-by: Andrew Morton Signed-off-by: Thomas Gleixner --- kernel/time/tick-broadcast.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c index 118a3b3b3f9a..877dbedc3118 100644 --- a/kernel/time/tick-broadcast.c +++ b/kernel/time/tick-broadcast.c @@ -27,7 +27,7 @@ * timer stops in C3 state. */ -struct tick_device tick_broadcast_device; +static struct tick_device tick_broadcast_device; /* FIXME: Use cpumask_var_t. */ static DECLARE_BITMAP(tick_broadcast_mask, NR_CPUS); static DECLARE_BITMAP(tmpmask, NR_CPUS); -- cgit v1.2.3 From a04198887658e1d8ae25f5420035c057cb170e67 Mon Sep 17 00:00:00 2001 From: Jon Hunter Date: Fri, 1 May 2009 13:10:23 -0700 Subject: timers: allow deferrable timers for intervals tv2-tv5 to be deferred In the current kernel implementation only kernel timers for time interval tv1 are being deferred. This patch allows any timer that is configured as deferrable to be defer regardless of time interval. This patch was previously discussed in http://marc.info/?l=linux-kernel&m=123196343531966&w=2 and was acked by Venki Pallipadi, the author of the original deferrable timer patch. Signed-off-by: Jon Hunter Acked-by: Venkatesh Pallipadi Signed-off-by: Andrew Morton Signed-off-by: Thomas Gleixner --- kernel/timer.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/timer.c b/kernel/timer.c index cffffad01c31..5c1e84beaf4a 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -1015,6 +1015,9 @@ cascade: index = slot = timer_jiffies & TVN_MASK; do { list_for_each_entry(nte, varp->vec + slot, entry) { + if (tbase_get_deferrable(nte->base)) + continue; + found = 1; if (time_before(nte->expires, expires)) expires = nte->expires; -- cgit v1.2.3 From a25cbd045a2ffc42787d4dbcbb9c7118f5f42732 Mon Sep 17 00:00:00 2001 From: Magnus Damm Date: Fri, 1 May 2009 14:45:46 +0900 Subject: clocksource: setup mult_orig in clocksource_enable() Setup clocksource mult_orig in clocksource_enable(). Clocksource drivers can save power by using keeping the device clock disabled while the clocksource is unused. In practice this means that the enable() and disable() callbacks perform clk_enable() and clk_disable(). The enable() callback may also use clk_get_rate() to get the clock rate from the clock framework. This information can then be used to calculate the shift and mult variables. Currently the mult_orig variable is setup from mult at registration time only. This is conflicting with the above case since the clock is disabled and the mult variable is not yet calculated at the time of registration. Moving the mult_orig setup code to clocksource_enable() allows us to both handle the common case with no enable() callback and the mult-changed-after-enable() case. [ Impact: allow dynamic clock source usage ] Signed-off-by: Magnus Damm LKML-Reference: <20090501054546.8193.10688.sendpatchset@rx1.opensource.se> Signed-off-by: Thomas Gleixner --- kernel/time/clocksource.c | 3 --- 1 file changed, 3 deletions(-) (limited to 'kernel') diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c index ecfd7b5187e0..80189f6f1c5a 100644 --- a/kernel/time/clocksource.c +++ b/kernel/time/clocksource.c @@ -402,9 +402,6 @@ int clocksource_register(struct clocksource *c) unsigned long flags; int ret; - /* save mult_orig on registration */ - c->mult_orig = c->mult; - spin_lock_irqsave(&clocksource_lock, flags); ret = clocksource_enqueue(c); if (!ret) -- cgit v1.2.3 From 7d27558c4138ac6b3684dea35c2f4379b940a7dd Mon Sep 17 00:00:00 2001 From: john stultz Date: Fri, 1 May 2009 13:10:26 -0700 Subject: timekeeping: create arch_gettimeoffset infrastructure Some arches don't supply their own clocksource. This is mainly the case in architectures that get their inter-tick times by reading the counter on their interval timer. Since these timers wrap every tick, they're not really useful as clocksources. Wrapping them to act like one is possible but not very efficient. So we provide a callout these arches can implement for use with the jiffies clocksource to provide finer then tick granular time. [ Impact: ease the migration to generic time keeping ] Signed-off-by: John Stultz Signed-off-by: Andrew Morton Signed-off-by: Thomas Gleixner --- kernel/time/timekeeping.c | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'kernel') diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index 687dff49f6e7..e97c50f8458b 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -77,6 +77,10 @@ static void clocksource_forward_now(void) clock->cycle_last = cycle_now; nsec = cyc2ns(clock, cycle_delta); + + /* If arch requires, add in gettimeoffset() */ + nsec += arch_gettimeoffset(); + timespec_add_ns(&xtime, nsec); nsec = ((s64)cycle_delta * clock->mult_orig) >> clock->shift; @@ -111,6 +115,9 @@ void getnstimeofday(struct timespec *ts) /* convert to nanoseconds: */ nsecs = cyc2ns(clock, cycle_delta); + /* If arch requires, add in gettimeoffset() */ + nsecs += arch_gettimeoffset(); + } while (read_seqretry(&xtime_lock, seq)); timespec_add_ns(ts, nsecs); -- cgit v1.2.3 From c81fc2c331b8514ad112054cd2d87e6ec132286b Mon Sep 17 00:00:00 2001 From: Magnus Damm Date: Fri, 1 May 2009 14:52:47 +0900 Subject: clockevent: export register_device and delta2ns Export the following symbols using EXPORT_SYMBOL_GPL: - clockevent_delta2ns - clockevents_register_device This allows us to build SuperH clockevent and clocksource drivers as modules, see drivers/clocksource/sh_*.c [ Impact: allow modular build of clockevent drivers ] Signed-off-by: Magnus Damm LKML-Reference: <20090501055247.8286.64067.sendpatchset@rx1.opensource.se> Signed-off-by: Thomas Gleixner --- kernel/time/clockevents.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c index d13be216a790..3948fa644a2d 100644 --- a/kernel/time/clockevents.c +++ b/kernel/time/clockevents.c @@ -54,6 +54,7 @@ unsigned long clockevent_delta2ns(unsigned long latch, return (unsigned long) clc; } +EXPORT_SYMBOL_GPL(clockevent_delta2ns); /** * clockevents_set_mode - set the operating mode of a clock event device @@ -187,6 +188,7 @@ void clockevents_register_device(struct clock_event_device *dev) spin_unlock(&clockevents_lock); } +EXPORT_SYMBOL_GPL(clockevents_register_device); /* * Noop handler when we shut down an event device -- cgit v1.2.3 From b82914ce33146186d554b0f5c41e4e13693614ce Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 4 May 2009 18:54:32 +0200 Subject: perf_counter: round-robin per-CPU counters too This used to be unstable when we had the rq->lock dependencies, but now that they are that of the past we can turn on percpu counter RR too. [ Impact: handle counter over-commit for per-CPU counters too ] LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 8660ae579530..b9679c36bcc2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1069,18 +1069,14 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); struct perf_counter_context *ctx = &curr->perf_counter_ctx; - const int rotate_percpu = 0; - if (rotate_percpu) - perf_counter_cpu_sched_out(cpuctx); + perf_counter_cpu_sched_out(cpuctx); perf_counter_task_sched_out(curr, cpu); - if (rotate_percpu) - rotate_ctx(&cpuctx->ctx); + rotate_ctx(&cpuctx->ctx); rotate_ctx(ctx); - if (rotate_percpu) - perf_counter_cpu_sched_in(cpuctx, cpu); + perf_counter_cpu_sched_in(cpuctx, cpu); perf_counter_task_sched_in(curr, cpu); } -- cgit v1.2.3 From 0d905bca23aca5c86a10ee101bcd3b1abbd40b25 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 4 May 2009 19:13:30 +0200 Subject: perf_counter: initialize the per-cpu context earlier percpu scheduling for perfcounters wants to take the context lock, but that lock first needs to be initialized. Currently it is an early_initcall() - but that is too late, the task tick runs much sooner than that. Call it explicitly from the scheduler init sequence instead. [ Impact: fix access-before-init crash ] LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 5 +---- kernel/sched.c | 5 ++++- 2 files changed, 5 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index b9679c36bcc2..fcdafa234a5d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3265,15 +3265,12 @@ static struct notifier_block __cpuinitdata perf_cpu_nb = { .notifier_call = perf_cpu_notify, }; -static int __init perf_counter_init(void) +void __init perf_counter_init(void) { perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, (void *)(long)smp_processor_id()); register_cpu_notifier(&perf_cpu_nb); - - return 0; } -early_initcall(perf_counter_init); static ssize_t perf_show_reserve_percpu(struct sysdev_class *class, char *buf) { diff --git a/kernel/sched.c b/kernel/sched.c index 2f600e30dcf0..a728976a3a6c 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -39,6 +39,7 @@ #include #include #include +#include #include #include #include @@ -8996,7 +8997,7 @@ void __init sched_init(void) * 1024) and two child groups A0 and A1 (of weight 1024 each), * then A0's share of the cpu resource is: * - * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% + * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% * * We achieve this by letting init_task_group's tasks sit * directly in rq->cfs (i.e init_task_group->se[] = NULL). @@ -9097,6 +9098,8 @@ void __init sched_init(void) alloc_bootmem_cpumask_var(&cpu_isolated_map); #endif /* SMP */ + perf_counter_init(); + scheduler_running = 1; } -- cgit v1.2.3 From 1dce8d99b85aba6eddb8b8260baea944922e6fe7 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 4 May 2009 19:23:18 +0200 Subject: perf_counter: convert perf_resource_mutex to a spinlock Now percpu counters can be initialized very early. But the init sequence uses mutex_lock(). Fortunately, perf_resource_mutex should be a spinlock anyway, so convert it. [ Impact: fix crash due to early init mutex use ] LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index fcdafa234a5d..5f86a1156c94 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -46,9 +46,9 @@ static atomic_t nr_comm_tracking __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ /* - * Mutex for (sysadmin-configurable) counter reservations: + * Lock for (sysadmin-configurable) counter reservations: */ -static DEFINE_MUTEX(perf_resource_mutex); +static DEFINE_SPINLOCK(perf_resource_lock); /* * Architecture provided APIs - weak aliases: @@ -3207,9 +3207,9 @@ static void __cpuinit perf_counter_init_cpu(int cpu) cpuctx = &per_cpu(perf_cpu_context, cpu); __perf_counter_init_context(&cpuctx->ctx, NULL); - mutex_lock(&perf_resource_mutex); + spin_lock(&perf_resource_lock); cpuctx->max_pertask = perf_max_counters - perf_reserved_percpu; - mutex_unlock(&perf_resource_mutex); + spin_unlock(&perf_resource_lock); hw_perf_counter_setup(cpu); } @@ -3292,7 +3292,7 @@ perf_set_reserve_percpu(struct sysdev_class *class, if (val > perf_max_counters) return -EINVAL; - mutex_lock(&perf_resource_mutex); + spin_lock(&perf_resource_lock); perf_reserved_percpu = val; for_each_online_cpu(cpu) { cpuctx = &per_cpu(perf_cpu_context, cpu); @@ -3302,7 +3302,7 @@ perf_set_reserve_percpu(struct sysdev_class *class, cpuctx->max_pertask = mpt; spin_unlock_irq(&cpuctx->ctx.lock); } - mutex_unlock(&perf_resource_mutex); + spin_unlock(&perf_resource_lock); return count; } @@ -3324,9 +3324,9 @@ perf_set_overcommit(struct sysdev_class *class, const char *buf, size_t count) if (val > 1) return -EINVAL; - mutex_lock(&perf_resource_mutex); + spin_lock(&perf_resource_lock); perf_overcommit = val; - mutex_unlock(&perf_resource_mutex); + spin_unlock(&perf_resource_lock); return count; } -- cgit v1.2.3 From 60aa605dfce2976e54fa76e805ab0f221372d4d9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 May 2009 17:50:21 +0200 Subject: sched: rt: document the risk of small values in the bandwidth settings Thomas noted that we should disallow sysctl_sched_rt_runtime == 0 for (!RT_GROUP) since the root group always has some RT tasks in it. Further, update the documentation to inspire clue. [ Impact: exclude corner-case sysctl_sched_rt_runtime value ] Reported-by: Thomas Gleixner Signed-off-by: Peter Zijlstra LKML-Reference: <20090505155436.863098054@chello.nl> Signed-off-by: Ingo Molnar --- kernel/sched.c | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 54d67b94f1a9..2a43a581ead3 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -9917,6 +9917,13 @@ static int sched_rt_global_constraints(void) if (sysctl_sched_rt_period <= 0) return -EINVAL; + /* + * There's always some RT tasks in the root group + * -- migration, kstopmachine etc.. + */ + if (sysctl_sched_rt_runtime == 0) + return -EBUSY; + spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); for_each_possible_cpu(i) { struct rt_rq *rt_rq = &cpu_rq(i)->rt; -- cgit v1.2.3 From c66de4a5be7913247bd83d79168f8e4420c9cfbc Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 May 2009 17:50:22 +0200 Subject: perf_counter: uncouple data_head updates from wakeups Keep data_head up-to-date irrespective of notifications. This fixes the case where you disable a counter and don't get a notification for the last few pending events, and it also allows polling usage. [ Impact: increase precision of perfcounter mmap-ed fields ] Suggested-by: Corey Ashford Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20090505155436.925084300@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 20 +++++++++----------- 1 file changed, 9 insertions(+), 11 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 5f86a1156c94..ba5e921e1f36 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1696,7 +1696,6 @@ struct perf_output_handle { struct perf_mmap_data *data; unsigned int offset; unsigned int head; - int wakeup; int nmi; int overflow; int locked; @@ -1752,8 +1751,7 @@ static void perf_output_unlock(struct perf_output_handle *handle) struct perf_mmap_data *data = handle->data; int head, cpu; - if (handle->wakeup) - data->wakeup_head = data->head; + data->done_head = data->head; if (!handle->locked) goto out; @@ -1764,13 +1762,11 @@ again: * before we publish the new head, matched by a rmb() in userspace when * reading this position. */ - while ((head = atomic_xchg(&data->wakeup_head, 0))) { + while ((head = atomic_xchg(&data->done_head, 0))) data->user_page->data_head = head; - handle->wakeup = 1; - } /* - * NMI can happen here, which means we can miss a wakeup_head update. + * NMI can happen here, which means we can miss a done_head update. */ cpu = atomic_xchg(&data->lock, 0); @@ -1779,7 +1775,7 @@ again: /* * Therefore we have to validate we did not indeed do so. */ - if (unlikely(atomic_read(&data->wakeup_head))) { + if (unlikely(atomic_read(&data->done_head))) { /* * Since we had it locked, we can lock it again. */ @@ -1789,7 +1785,7 @@ again: goto again; } - if (handle->wakeup) + if (atomic_xchg(&data->wakeup, 0)) perf_output_wakeup(handle); out: local_irq_restore(handle->flags); @@ -1824,7 +1820,9 @@ static int perf_output_begin(struct perf_output_handle *handle, handle->offset = offset; handle->head = head; - handle->wakeup = (offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT); + + if ((offset >> PAGE_SHIFT) != (head >> PAGE_SHIFT)) + atomic_set(&data->wakeup, 1); return 0; @@ -1882,7 +1880,7 @@ static void perf_output_end(struct perf_output_handle *handle) int events = atomic_inc_return(&data->events); if (events >= wakeup_events) { atomic_sub(wakeup_events, &data->events); - handle->wakeup = 1; + atomic_set(&data->wakeup, 1); } } -- cgit v1.2.3 From 6de6a7b95705b859b61430fa3afa1403034eb3e6 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 May 2009 17:50:23 +0200 Subject: perf_counter: add ioctl(PERF_COUNTER_IOC_RESET) Provide a way to reset an existing counter - this eases PAPI libraries around perfcounters. Similar to read() it doesn't collapse pending child counters. [ Impact: new perfcounter fd ioctl method to reset counters ] Suggested-by: Corey Ashford Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20090505155437.022272933@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ba5e921e1f36..6e6834e0587e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1288,6 +1288,11 @@ static unsigned int perf_poll(struct file *file, poll_table *wait) return events; } +static void perf_counter_reset(struct perf_counter *counter) +{ + atomic_set(&counter->count, 0); +} + static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) { struct perf_counter *counter = file->private_data; @@ -1303,6 +1308,9 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) case PERF_COUNTER_IOC_REFRESH: perf_counter_refresh(counter, arg); break; + case PERF_COUNTER_IOC_RESET: + perf_counter_reset(counter); + break; default: err = -ENOTTY; } -- cgit v1.2.3 From c5078f78b455fbf67ea71442c7e7ca8acf9ff095 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 May 2009 17:50:24 +0200 Subject: perf_counter: provide an mlock threshold Provide a threshold to relax the mlock accounting, increasing usability. Each counter gets perf_counter_mlock_kb for free. [ Impact: allow more mmap buffering ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090505155437.112113632@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 15 +++++++++++---- kernel/sysctl.c | 8 ++++++++ 2 files changed, 19 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 6e6834e0587e..2d1342738305 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -44,6 +44,7 @@ static atomic_t nr_munmap_tracking __read_mostly; static atomic_t nr_comm_tracking __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ +int sysctl_perf_counter_mlock __read_mostly = 128; /* 'free' kb per counter */ /* * Lock for (sysadmin-configurable) counter reservations: @@ -1461,7 +1462,7 @@ static void perf_mmap_close(struct vm_area_struct *vma) if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { - vma->vm_mm->locked_vm -= counter->data->nr_pages + 1; + vma->vm_mm->locked_vm -= counter->data->nr_locked; perf_mmap_data_free(counter); mutex_unlock(&counter->mmap_mutex); } @@ -1480,6 +1481,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) unsigned long nr_pages; unsigned long locked, lock_limit; int ret = 0; + long extra; if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE)) return -EINVAL; @@ -1507,8 +1509,12 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; } - locked = vma->vm_mm->locked_vm; - locked += nr_pages + 1; + extra = nr_pages /* + 1 only account the data pages */; + extra -= sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); + if (extra < 0) + extra = 0; + + locked = vma->vm_mm->locked_vm + extra; lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; lock_limit >>= PAGE_SHIFT; @@ -1524,7 +1530,8 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; atomic_set(&counter->mmap_count, 1); - vma->vm_mm->locked_vm += nr_pages + 1; + vma->vm_mm->locked_vm += extra; + counter->data->nr_locked = extra; unlock: mutex_unlock(&counter->mmap_mutex); diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 8203d70928d5..3b05c2b088d2 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -920,6 +920,14 @@ static struct ctl_table kern_table[] = { .mode = 0644, .proc_handler = &proc_dointvec, }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "perf_counter_mlock_kb", + .data = &sysctl_perf_counter_mlock, + .maxlen = sizeof(sysctl_perf_counter_mlock), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, #endif /* * NOTE: do not add new entries to this table unless you have read -- cgit v1.2.3 From 22c1558e51c210787c6cf75d8905246fc91ec030 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 May 2009 17:50:25 +0200 Subject: perf_counter: fix the output lock Use -1 instead of 0 as unlocked, since 0 is a valid cpu number. ( This is not an issue right now but will be once we allow multiple counters to output to the same mmap area. ) [ Impact: prepare code for multi-counter profile output ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090505155437.232686598@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 2d1342738305..c881afef997b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1409,6 +1409,7 @@ static int perf_mmap_data_alloc(struct perf_counter *counter, int nr_pages) } data->nr_pages = nr_pages; + atomic_set(&data->lock, -1); rcu_assign_pointer(counter->data, data); @@ -1755,7 +1756,7 @@ static void perf_output_lock(struct perf_output_handle *handle) if (in_nmi() && atomic_read(&data->lock) == cpu) return; - while (atomic_cmpxchg(&data->lock, 0, cpu) != 0) + while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) cpu_relax(); handle->locked = 1; @@ -1784,7 +1785,7 @@ again: * NMI can happen here, which means we can miss a done_head update. */ - cpu = atomic_xchg(&data->lock, 0); + cpu = atomic_xchg(&data->lock, -1); WARN_ON_ONCE(cpu != smp_processor_id()); /* @@ -1794,7 +1795,7 @@ again: /* * Since we had it locked, we can lock it again. */ - while (atomic_cmpxchg(&data->lock, 0, cpu) != 0) + while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) cpu_relax(); goto again; -- cgit v1.2.3 From 2023b359214bbc5bad31571cf50d7fb83b535c0a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 5 May 2009 17:50:26 +0200 Subject: perf_counter: inheritable sample counters Redirect the output to the parent counter and put in some sanity checks. [ Impact: new perfcounter feature - inherited sampling counters ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090505155437.331556171@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 32 ++++++++++++++++++++++++++++++-- 1 file changed, 30 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c881afef997b..60e55f0b48f4 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -738,10 +738,18 @@ static void perf_counter_enable(struct perf_counter *counter) spin_unlock_irq(&ctx->lock); } -static void perf_counter_refresh(struct perf_counter *counter, int refresh) +static int perf_counter_refresh(struct perf_counter *counter, int refresh) { + /* + * not supported on inherited counters + */ + if (counter->hw_event.inherit) + return -EINVAL; + atomic_add(refresh, &counter->event_limit); perf_counter_enable(counter); + + return 0; } /* @@ -1307,7 +1315,7 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) perf_counter_disable_family(counter); break; case PERF_COUNTER_IOC_REFRESH: - perf_counter_refresh(counter, arg); + err = perf_counter_refresh(counter, arg); break; case PERF_COUNTER_IOC_RESET: perf_counter_reset(counter); @@ -1814,6 +1822,12 @@ static int perf_output_begin(struct perf_output_handle *handle, struct perf_mmap_data *data; unsigned int offset, head; + /* + * For inherited counters we send all the output towards the parent. + */ + if (counter->parent) + counter = counter->parent; + rcu_read_lock(); data = rcu_dereference(counter->data); if (!data) @@ -1995,6 +2009,9 @@ static void perf_counter_output(struct perf_counter *counter, if (record_type & PERF_RECORD_ADDR) perf_output_put(&handle, addr); + /* + * XXX PERF_RECORD_GROUP vs inherited counters seems difficult. + */ if (record_type & PERF_RECORD_GROUP) { struct perf_counter *leader, *sub; u64 nr = counter->nr_siblings; @@ -2281,6 +2298,11 @@ int perf_counter_overflow(struct perf_counter *counter, int events = atomic_read(&counter->event_limit); int ret = 0; + /* + * XXX event_limit might not quite work as expected on inherited + * counters + */ + counter->pending_kill = POLL_IN; if (events && atomic_dec_and_test(&counter->event_limit)) { ret = 1; @@ -2801,6 +2823,12 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, pmu = NULL; + /* + * we currently do not support PERF_RECORD_GROUP on inherited counters + */ + if (hw_event->inherit && (hw_event->record_type & PERF_RECORD_GROUP)) + goto done; + if (perf_event_raw(hw_event)) { pmu = hw_perf_counter_init(counter); goto done; -- cgit v1.2.3 From 41c51c98f588edcdf6141cff1895df738e03ddd4 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Sun, 3 May 2009 23:11:18 +0200 Subject: rcu: rcu_sched_grace_period(): kill the bogus flush_signals() As a kernel thread, rcu_sched_grace_period() runs with all signals ignored. It can never receive a signal even if it sleeps in TASK_INTERRUPTIBLE, it needs the explicit allow_signal() to be visible for signals. [ Impact: reduce kernel size, remove dead code ] Signed-off-by: Oleg Nesterov Reviewed-by: Paul E. McKenney Cc: Andrew Morton LKML-Reference: <20090503211118.GA22973@redhat.com> Signed-off-by: Ingo Molnar --- kernel/rcupreempt.c | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/rcupreempt.c b/kernel/rcupreempt.c index ce97a4df64d3..beb0e659adcc 100644 --- a/kernel/rcupreempt.c +++ b/kernel/rcupreempt.c @@ -1356,17 +1356,11 @@ static int rcu_sched_grace_period(void *arg) rcu_ctrlblk.sched_sleep = rcu_sched_sleeping; spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags); - ret = 0; + ret = 0; /* unused */ __wait_event_interruptible(rcu_ctrlblk.sched_wq, rcu_ctrlblk.sched_sleep != rcu_sched_sleeping, ret); - /* - * Signals would prevent us from sleeping, and we cannot - * do much with them in any case. So flush them. - */ - if (ret) - flush_signals(current); couldsleepnext = 0; } while (!kthread_should_stop()); -- cgit v1.2.3 From aa47b7e0f89b9998dad4d1667447e8cb7703ff4e Mon Sep 17 00:00:00 2001 From: David Rientjes Date: Mon, 4 May 2009 01:38:05 -0700 Subject: sched: emit thread info flags with stack trace When a thread is oom killed and fails to exit, it's helpful to know which threads have access to memory reserves if the machine livelocks. This is done by testing for the TIF_MEMDIE thread info flag and should be displayed alongside stack traces to identify tasks that have access to such reserves but are still stuck allocating pages, for instance. It would probably be helpful in other cases as well, so all thread info flags are emitted when showing a task. ( v2: fix warning reported by Stephen Rothwell ) [ Impact: extend debug printout info ] Signed-off-by: David Rientjes Cc: Peter Zijlstra Cc: Stephen Rothwell LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/sched.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 2a43a581ead3..5aa63f50c696 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -6610,8 +6610,9 @@ void sched_show_task(struct task_struct *p) #ifdef CONFIG_DEBUG_STACK_USAGE free = stack_not_used(p); #endif - printk(KERN_CONT "%5lu %5d %6d\n", free, - task_pid_nr(p), task_pid_nr(p->real_parent)); + printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, + task_pid_nr(p), task_pid_nr(p->real_parent), + (unsigned long)task_thread_info(p)->flags); show_stack(p, NULL); } -- cgit v1.2.3 From 7fc23a5380797012e92a9633169440f2f4a21253 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 8 May 2009 18:52:21 +0200 Subject: perf_counter: optimize perf_counter_task_tick() perf_counter_task_tick() does way too much work to find out there's nothing to do. Provide an easy short-circuit for the normal case where there are no counters on the system. [ Impact: micro-optimization ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090508170028.750619201@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 60e55f0b48f4..fdb0d2421276 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -39,6 +39,7 @@ int perf_max_counters __read_mostly = 1; static int perf_reserved_percpu __read_mostly; static int perf_overcommit __read_mostly = 1; +static atomic_t nr_counters __read_mostly; static atomic_t nr_mmap_tracking __read_mostly; static atomic_t nr_munmap_tracking __read_mostly; static atomic_t nr_comm_tracking __read_mostly; @@ -1076,8 +1077,14 @@ static void rotate_ctx(struct perf_counter_context *ctx) void perf_counter_task_tick(struct task_struct *curr, int cpu) { - struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_cpu_context *cpuctx; + struct perf_counter_context *ctx; + + if (!atomic_read(&nr_counters)) + return; + + cpuctx = &per_cpu(perf_cpu_context, cpu); + ctx = &curr->perf_counter_ctx; perf_counter_cpu_sched_out(cpuctx); perf_counter_task_sched_out(curr, cpu); @@ -1197,6 +1204,7 @@ static void free_counter(struct perf_counter *counter) { perf_pending_sync(counter); + atomic_dec(&nr_counters); if (counter->hw_event.mmap) atomic_dec(&nr_mmap_tracking); if (counter->hw_event.munmap) @@ -2861,6 +2869,7 @@ done: counter->pmu = pmu; + atomic_inc(&nr_counters); if (counter->hw_event.mmap) atomic_inc(&nr_mmap_tracking); if (counter->hw_event.munmap) -- cgit v1.2.3 From 3df5edad87a998273aa5a9a8c728c05d855ad00e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 8 May 2009 18:52:22 +0200 Subject: perf_counter: rework ioctl()s Corey noticed that ioctl()s on grouped counters didn't work on the whole group. This extends the ioctl() interface to take a second argument that is interpreted as a flags field. We then provide PERF_IOC_FLAG_GROUP to toggle the behaviour. Having this flag gives the greatest flexibility, allowing you to individually enable/disable/reset counters in a group, or all together. [ Impact: fix group counter enable/disable semantics ] Reported-by: Corey Ashford Signed-off-by: Peter Zijlstra Cc: Paul Mackerras LKML-Reference: <20090508170028.837558214@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 104 ++++++++++++++++++++++++++++---------------------- 1 file changed, 58 insertions(+), 46 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index fdb0d2421276..f4883f1f47eb 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -82,7 +82,7 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) * add it straight to the context's counter list, or to the group * leader's sibling list: */ - if (counter->group_leader == counter) + if (group_leader == counter) list_add_tail(&counter->list_entry, &ctx->counter_list); else { list_add_tail(&counter->list_entry, &group_leader->sibling_list); @@ -385,24 +385,6 @@ static void perf_counter_disable(struct perf_counter *counter) spin_unlock_irq(&ctx->lock); } -/* - * Disable a counter and all its children. - */ -static void perf_counter_disable_family(struct perf_counter *counter) -{ - struct perf_counter *child; - - perf_counter_disable(counter); - - /* - * Lock the mutex to protect the list of children - */ - mutex_lock(&counter->mutex); - list_for_each_entry(child, &counter->child_list, child_list) - perf_counter_disable(child); - mutex_unlock(&counter->mutex); -} - static int counter_sched_in(struct perf_counter *counter, struct perf_cpu_context *cpuctx, @@ -753,24 +735,6 @@ static int perf_counter_refresh(struct perf_counter *counter, int refresh) return 0; } -/* - * Enable a counter and all its children. - */ -static void perf_counter_enable_family(struct perf_counter *counter) -{ - struct perf_counter *child; - - perf_counter_enable(counter); - - /* - * Lock the mutex to protect the list of children - */ - mutex_lock(&counter->mutex); - list_for_each_entry(child, &counter->child_list, child_list) - perf_counter_enable(child); - mutex_unlock(&counter->mutex); -} - void __perf_counter_sched_out(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx) { @@ -1307,31 +1271,79 @@ static unsigned int perf_poll(struct file *file, poll_table *wait) static void perf_counter_reset(struct perf_counter *counter) { + (void)perf_counter_read(counter); atomic_set(&counter->count, 0); + perf_counter_update_userpage(counter); +} + +static void perf_counter_for_each_sibling(struct perf_counter *counter, + void (*func)(struct perf_counter *)) +{ + struct perf_counter_context *ctx = counter->ctx; + struct perf_counter *sibling; + + spin_lock_irq(&ctx->lock); + counter = counter->group_leader; + + func(counter); + list_for_each_entry(sibling, &counter->sibling_list, list_entry) + func(sibling); + spin_unlock_irq(&ctx->lock); +} + +static void perf_counter_for_each_child(struct perf_counter *counter, + void (*func)(struct perf_counter *)) +{ + struct perf_counter *child; + + mutex_lock(&counter->mutex); + func(counter); + list_for_each_entry(child, &counter->child_list, child_list) + func(child); + mutex_unlock(&counter->mutex); +} + +static void perf_counter_for_each(struct perf_counter *counter, + void (*func)(struct perf_counter *)) +{ + struct perf_counter *child; + + mutex_lock(&counter->mutex); + perf_counter_for_each_sibling(counter, func); + list_for_each_entry(child, &counter->child_list, child_list) + perf_counter_for_each_sibling(child, func); + mutex_unlock(&counter->mutex); } static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) { struct perf_counter *counter = file->private_data; - int err = 0; + void (*func)(struct perf_counter *); + u32 flags = arg; switch (cmd) { case PERF_COUNTER_IOC_ENABLE: - perf_counter_enable_family(counter); + func = perf_counter_enable; break; case PERF_COUNTER_IOC_DISABLE: - perf_counter_disable_family(counter); - break; - case PERF_COUNTER_IOC_REFRESH: - err = perf_counter_refresh(counter, arg); + func = perf_counter_disable; break; case PERF_COUNTER_IOC_RESET: - perf_counter_reset(counter); + func = perf_counter_reset; break; + + case PERF_COUNTER_IOC_REFRESH: + return perf_counter_refresh(counter, arg); default: - err = -ENOTTY; + return -ENOTTY; } - return err; + + if (flags & PERF_IOC_FLAG_GROUP) + perf_counter_for_each(counter, func); + else + perf_counter_for_each_child(counter, func); + + return 0; } /* -- cgit v1.2.3 From a85f61abe11a46553c4562e74edb27ebc782aeb7 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 8 May 2009 18:52:23 +0200 Subject: perf_counter: add PERF_RECORD_CONFIG Much like CONFIG_RECORD_GROUP records the hw_event.config to identify the values, allow to record this for all counters. [ Impact: extend perfcounter output record format ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090508170028.923228280@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f4883f1f47eb..c615f52aa408 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1994,6 +1994,11 @@ static void perf_counter_output(struct perf_counter *counter, header.size += sizeof(u64); } + if (record_type & PERF_RECORD_CONFIG) { + header.type |= PERF_RECORD_CONFIG; + header.size += sizeof(u64); + } + if (record_type & PERF_RECORD_GROUP) { header.type |= PERF_RECORD_GROUP; header.size += sizeof(u64) + @@ -2029,6 +2034,9 @@ static void perf_counter_output(struct perf_counter *counter, if (record_type & PERF_RECORD_ADDR) perf_output_put(&handle, addr); + if (record_type & PERF_RECORD_CONFIG) + perf_output_put(&handle, counter->hw_event.config); + /* * XXX PERF_RECORD_GROUP vs inherited counters seems difficult. */ -- cgit v1.2.3 From f370e1e2f195ec1e6420e26fc83e0319595db578 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 8 May 2009 18:52:24 +0200 Subject: perf_counter: add PERF_RECORD_CPU Allow recording the CPU number the event was generated on. RFC: this leaves a u32 as reserved, should we fill in the node_id() there, or leave this open for future extention, as userspace can already easily do the cpu->node mapping if needed. [ Impact: extend perfcounter output record format ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: <20090508170029.008627711@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c615f52aa408..d850a1fb8d4c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1956,6 +1956,9 @@ static void perf_counter_output(struct perf_counter *counter, struct perf_callchain_entry *callchain = NULL; int callchain_size = 0; u64 time; + struct { + u32 cpu, reserved; + } cpu_entry; header.type = 0; header.size = sizeof(header); @@ -1999,6 +2002,13 @@ static void perf_counter_output(struct perf_counter *counter, header.size += sizeof(u64); } + if (record_type & PERF_RECORD_CPU) { + header.type |= PERF_RECORD_CPU; + header.size += sizeof(cpu_entry); + + cpu_entry.cpu = raw_smp_processor_id(); + } + if (record_type & PERF_RECORD_GROUP) { header.type |= PERF_RECORD_GROUP; header.size += sizeof(u64) + @@ -2037,6 +2047,9 @@ static void perf_counter_output(struct perf_counter *counter, if (record_type & PERF_RECORD_CONFIG) perf_output_put(&handle, counter->hw_event.config); + if (record_type & PERF_RECORD_CPU) + perf_output_put(&handle, cpu_entry); + /* * XXX PERF_RECORD_GROUP vs inherited counters seems difficult. */ -- cgit v1.2.3 From 92d23f703c608fcb2c8edd74a3fd0f4031e18606 Mon Sep 17 00:00:00 2001 From: Ron Date: Fri, 8 May 2009 22:54:49 +0930 Subject: sched: Fix fallback sched_clock()'s offset when using jiffies Account for the initial offset to the jiffy count. [ Impact: fix printk timestamps on architectures using fallback sched_clock() ] Signed-off-by: Ron Lee Cc: Andrew Morton Signed-off-by: Ingo Molnar --- kernel/sched_clock.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/sched_clock.c b/kernel/sched_clock.c index 819f17ac796e..e1d16c9a7680 100644 --- a/kernel/sched_clock.c +++ b/kernel/sched_clock.c @@ -38,7 +38,8 @@ */ unsigned long long __attribute__((weak)) sched_clock(void) { - return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); + return (unsigned long long)(jiffies - INITIAL_JIFFIES) + * (NSEC_PER_SEC / HZ); } static __read_mostly int sched_clock_running; -- cgit v1.2.3 From 5e751e992f3fb08ba35e1ca8095ec8fbf9eda523 Mon Sep 17 00:00:00 2001 From: David Howells Date: Fri, 8 May 2009 13:55:22 +0100 Subject: CRED: Rename cred_exec_mutex to reflect that it's a guard against ptrace Rename cred_exec_mutex to reflect that it's a guard against foreign intervention on a process's credential state, such as is made by ptrace(). The attachment of a debugger to a process affects execve()'s calculation of the new credential state - _and_ also setprocattr()'s calculation of that state. Signed-off-by: David Howells Signed-off-by: James Morris --- kernel/cred.c | 4 ++-- kernel/ptrace.c | 9 +++++---- 2 files changed, 7 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/cred.c b/kernel/cred.c index 3a039189d707..1bb4d7e5d616 100644 --- a/kernel/cred.c +++ b/kernel/cred.c @@ -167,7 +167,7 @@ EXPORT_SYMBOL(prepare_creds); /* * Prepare credentials for current to perform an execve() - * - The caller must hold current->cred_exec_mutex + * - The caller must hold current->cred_guard_mutex */ struct cred *prepare_exec_creds(void) { @@ -276,7 +276,7 @@ int copy_creds(struct task_struct *p, unsigned long clone_flags) struct cred *new; int ret; - mutex_init(&p->cred_exec_mutex); + mutex_init(&p->cred_guard_mutex); if ( #ifdef CONFIG_KEYS diff --git a/kernel/ptrace.c b/kernel/ptrace.c index 0692ab5a0d67..27ac80298bfa 100644 --- a/kernel/ptrace.c +++ b/kernel/ptrace.c @@ -185,10 +185,11 @@ int ptrace_attach(struct task_struct *task) if (same_thread_group(task, current)) goto out; - /* Protect exec's credential calculations against our interference; - * SUID, SGID and LSM creds get determined differently under ptrace. + /* Protect the target's credential calculations against our + * interference; SUID, SGID and LSM creds get determined differently + * under ptrace. */ - retval = mutex_lock_interruptible(&task->cred_exec_mutex); + retval = mutex_lock_interruptible(&task->cred_guard_mutex); if (retval < 0) goto out; @@ -232,7 +233,7 @@ repeat: bad: write_unlock_irqrestore(&tasklist_lock, flags); task_unlock(task); - mutex_unlock(&task->cred_exec_mutex); + mutex_unlock(&task->cred_guard_mutex); out: return retval; } -- cgit v1.2.3 From 5b93629b4509c03ffa87a9316412fedf6f58cb37 Mon Sep 17 00:00:00 2001 From: Tejun Heo Date: Thu, 7 May 2009 22:24:38 +0900 Subject: block: implement blk_rq_pos/[cur_]sectors() and convert obvious ones Implement accessors - blk_rq_pos(), blk_rq_sectors() and blk_rq_cur_sectors() which return rq->hard_sector, rq->hard_nr_sectors and rq->hard_cur_sectors respectively and convert direct references of the said fields to the accessors. This is in preparation of request data length handling cleanup. Geert : suggested adding const to struct request * parameter to accessors Sergei : spotted error in patch description [ Impact: cleanup ] Signed-off-by: Tejun Heo Acked-by: Geert Uytterhoeven Acked-by: Stephen Rothwell Tested-by: Grant Likely Acked-by: Grant Likely Ackec-by: Sergei Shtylyov Cc: Bartlomiej Zolnierkiewicz Cc: Borislav Petkov Cc: James Bottomley Signed-off-by: Jens Axboe --- kernel/trace/blktrace.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c index 921ef5d1f0ba..42f1c11e754c 100644 --- a/kernel/trace/blktrace.c +++ b/kernel/trace/blktrace.c @@ -646,7 +646,7 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq, rq->cmd_len, rq->cmd); } else { what |= BLK_TC_ACT(BLK_TC_FS); - __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, + __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_sectors(rq) << 9, rw, what, rq->errors, 0, NULL); } } @@ -857,7 +857,7 @@ void blk_add_driver_data(struct request_queue *q, __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA, rq->errors, len, data); else - __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, + __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_sectors(rq) << 9, 0, BLK_TA_DRV_DATA, rq->errors, len, data); } EXPORT_SYMBOL_GPL(blk_add_driver_data); -- cgit v1.2.3 From 2e46e8b27aa57c6bd34b3102b40ee4d0144b4fab Mon Sep 17 00:00:00 2001 From: Tejun Heo Date: Thu, 7 May 2009 22:24:41 +0900 Subject: block: drop request->hard_* and *nr_sectors struct request has had a few different ways to represent some properties of a request. ->hard_* represent block layer's view of the request progress (completion cursor) and the ones without the prefix are supposed to represent the issue cursor and allowed to be updated as necessary by the low level drivers. The thing is that as block layer supports partial completion, the two cursors really aren't necessary and only cause confusion. In addition, manual management of request detail from low level drivers is cumbersome and error-prone at the very least. Another interesting duplicate fields are rq->[hard_]nr_sectors and rq->{hard_cur|current}_nr_sectors against rq->data_len and rq->bio->bi_size. This is more convoluted than the hard_ case. rq->[hard_]nr_sectors are initialized for requests with bio but blk_rq_bytes() uses it only for !pc requests. rq->data_len is initialized for all request but blk_rq_bytes() uses it only for pc requests. This causes good amount of confusion throughout block layer and its drivers and determining the request length has been a bit of black magic which may or may not work depending on circumstances and what the specific LLD is actually doing. rq->{hard_cur|current}_nr_sectors represent the number of sectors in the contiguous data area at the front. This is mainly used by drivers which transfers data by walking request segment-by-segment. This value always equals rq->bio->bi_size >> 9. However, data length for pc requests may not be multiple of 512 bytes and using this field becomes a bit confusing. In general, having multiple fields to represent the same property leads only to confusion and subtle bugs. With recent block low level driver cleanups, no driver is accessing or manipulating these duplicate fields directly. Drop all the duplicates. Now rq->sector means the current sector, rq->data_len the current total length and rq->bio->bi_size the current segment length. Everything else is defined in terms of these three and available only through accessors. * blk_recalc_rq_sectors() is collapsed into blk_update_request() and now handles pc and fs requests equally other than rq->sector update. This means that now pc requests can use partial completion too (no in-kernel user yet tho). * bio_cur_sectors() is replaced with bio_cur_bytes() as block layer now uses byte count as the primary data length. * blk_rq_pos() is now guranteed to be always correct. In-block users converted. * blk_rq_bytes() is now guaranteed to be always valid as is blk_rq_sectors(). In-block users converted. * blk_rq_sectors() is now guaranteed to equal blk_rq_bytes() >> 9. More convenient one is used. * blk_rq_bytes() and blk_rq_cur_bytes() are now inlined and take const pointer to request. [ Impact: API cleanup, single way to represent one property of a request ] Signed-off-by: Tejun Heo Cc: Boaz Harrosh Signed-off-by: Jens Axboe --- kernel/trace/blktrace.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c index 42f1c11e754c..5708a14bee54 100644 --- a/kernel/trace/blktrace.c +++ b/kernel/trace/blktrace.c @@ -642,12 +642,12 @@ static void blk_add_trace_rq(struct request_queue *q, struct request *rq, if (blk_pc_request(rq)) { what |= BLK_TC_ACT(BLK_TC_PC); - __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, - rq->cmd_len, rq->cmd); + __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw, + what, rq->errors, rq->cmd_len, rq->cmd); } else { what |= BLK_TC_ACT(BLK_TC_FS); - __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_sectors(rq) << 9, - rw, what, rq->errors, 0, NULL); + __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw, + what, rq->errors, 0, NULL); } } @@ -854,11 +854,11 @@ void blk_add_driver_data(struct request_queue *q, return; if (blk_pc_request(rq)) - __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA, - rq->errors, len, data); + __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, + BLK_TA_DRV_DATA, rq->errors, len, data); else - __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_sectors(rq) << 9, - 0, BLK_TA_DRV_DATA, rq->errors, len, data); + __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, + BLK_TA_DRV_DATA, rq->errors, len, data); } EXPORT_SYMBOL_GPL(blk_add_driver_data); -- cgit v1.2.3 From 6751b71ea2c7ab8c0d65f01973a3fc8ea16992f4 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 11 May 2009 12:08:02 +1000 Subject: perf_counter: Put whole group on when enabling group leader Currently, if you have a group where the leader is disabled and there are siblings that are enabled, and then you enable the leader, we only put the leader on the PMU, and not its enabled siblings. This is incorrect, since the enabled group members should be all on or all off at any given point. This fixes it by adding a call to group_sched_in in __perf_counter_enable in the case where we're enabling a group leader. To avoid the need for a forward declaration this also moves group_sched_in up before __perf_counter_enable. The actual content of group_sched_in is unchanged by this patch. [ Impact: fix bug in counter enable code ] Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Corey Ashford LKML-Reference: <18951.34946.451546.691693@drongo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 99 ++++++++++++++++++++++++++------------------------- 1 file changed, 51 insertions(+), 48 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d850a1fb8d4c..a5bdc93ac477 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -419,6 +419,54 @@ counter_sched_in(struct perf_counter *counter, return 0; } +static int +group_sched_in(struct perf_counter *group_counter, + struct perf_cpu_context *cpuctx, + struct perf_counter_context *ctx, + int cpu) +{ + struct perf_counter *counter, *partial_group; + int ret; + + if (group_counter->state == PERF_COUNTER_STATE_OFF) + return 0; + + ret = hw_perf_group_sched_in(group_counter, cpuctx, ctx, cpu); + if (ret) + return ret < 0 ? ret : 0; + + group_counter->prev_state = group_counter->state; + if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) + return -EAGAIN; + + /* + * Schedule in siblings as one group (if any): + */ + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { + counter->prev_state = counter->state; + if (counter_sched_in(counter, cpuctx, ctx, cpu)) { + partial_group = counter; + goto group_error; + } + } + + return 0; + +group_error: + /* + * Groups can be scheduled in as one unit only, so undo any + * partial group before returning: + */ + list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { + if (counter == partial_group) + break; + counter_sched_out(counter, cpuctx, ctx); + } + counter_sched_out(group_counter, cpuctx, ctx); + + return -EAGAIN; +} + /* * Return 1 for a group consisting entirely of software counters, * 0 if the group contains any hardware counters. @@ -643,6 +691,9 @@ static void __perf_counter_enable(void *info) if (!group_can_go_on(counter, cpuctx, 1)) err = -EEXIST; + else if (counter == leader) + err = group_sched_in(counter, cpuctx, ctx, + smp_processor_id()); else err = counter_sched_in(counter, cpuctx, ctx, smp_processor_id()); @@ -791,54 +842,6 @@ static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx) __perf_counter_sched_out(&cpuctx->ctx, cpuctx); } -static int -group_sched_in(struct perf_counter *group_counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, - int cpu) -{ - struct perf_counter *counter, *partial_group; - int ret; - - if (group_counter->state == PERF_COUNTER_STATE_OFF) - return 0; - - ret = hw_perf_group_sched_in(group_counter, cpuctx, ctx, cpu); - if (ret) - return ret < 0 ? ret : 0; - - group_counter->prev_state = group_counter->state; - if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) - return -EAGAIN; - - /* - * Schedule in siblings as one group (if any): - */ - list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { - counter->prev_state = counter->state; - if (counter_sched_in(counter, cpuctx, ctx, cpu)) { - partial_group = counter; - goto group_error; - } - } - - return 0; - -group_error: - /* - * Groups can be scheduled in as one unit only, so undo any - * partial group before returning: - */ - list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { - if (counter == partial_group) - break; - counter_sched_out(counter, cpuctx, ctx); - } - counter_sched_out(group_counter, cpuctx, ctx); - - return -EAGAIN; -} - static void __perf_counter_sched_in(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx, int cpu) -- cgit v1.2.3 From a08b159fc243dbfe415250466d24cfc5010deee5 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 11 May 2009 15:46:10 +1000 Subject: perf_counter: don't count scheduler ticks as context switches The context-switch software counter gives inflated values at present because each scheduler tick and each process-wide counter enable/disable prctl gets counted as a context switch. This happens because perf_counter_task_tick, perf_counter_task_disable and perf_counter_task_enable all call perf_counter_task_sched_out, which calls perf_swcounter_event to record a context switch event. This fixes it by introducing a variant of perf_counter_task_sched_out with two underscores in front for internal use within the perf_counter code, and makes perf_counter_task_{tick,disable,enable} call it. This variant doesn't record a context switch event, and takes a struct perf_counter_context *. This adds the new variant rather than changing the behaviour or interface of perf_counter_task_sched_out because that is called from other code. [ Impact: fix inflated context-switch event counts ] Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Corey Ashford LKML-Reference: <18951.48034.485580.498953@drongo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index a5bdc93ac477..7373b96bc36c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -837,6 +837,14 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) cpuctx->task_ctx = NULL; } +static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + + __perf_counter_sched_out(ctx, cpuctx); + cpuctx->task_ctx = NULL; +} + static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx) { __perf_counter_sched_out(&cpuctx->ctx, cpuctx); @@ -943,15 +951,13 @@ int perf_counter_task_disable(void) struct perf_counter *counter; unsigned long flags; u64 perf_flags; - int cpu; if (likely(!ctx->nr_counters)) return 0; local_irq_save(flags); - cpu = smp_processor_id(); - perf_counter_task_sched_out(curr, cpu); + __perf_counter_task_sched_out(ctx); spin_lock(&ctx->lock); @@ -989,7 +995,7 @@ int perf_counter_task_enable(void) local_irq_save(flags); cpu = smp_processor_id(); - perf_counter_task_sched_out(curr, cpu); + __perf_counter_task_sched_out(ctx); spin_lock(&ctx->lock); @@ -1054,7 +1060,7 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) ctx = &curr->perf_counter_ctx; perf_counter_cpu_sched_out(cpuctx); - perf_counter_task_sched_out(curr, cpu); + __perf_counter_task_sched_out(ctx); rotate_ctx(&cpuctx->ctx); rotate_ctx(ctx); -- cgit v1.2.3 From 615a3f1e055ac9b0ae74e1f935a12ea2cfe2a2ad Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 11 May 2009 15:50:21 +1000 Subject: perf_counter: call atomic64_set for counter->count A compile warning triggered because we are calling atomic_set(&counter->count). But since counter->count is an atomic64_t, we have to use atomic64_set. So the count can be set short, resulting in the reset ioctl only resetting the low word. [ Impact: clear counter properly during the reset ioctl ] Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Corey Ashford LKML-Reference: <18951.48285.270311.981806@drongo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7373b96bc36c..5ea0240adab2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1281,7 +1281,7 @@ static unsigned int perf_poll(struct file *file, poll_table *wait) static void perf_counter_reset(struct perf_counter *counter) { (void)perf_counter_read(counter); - atomic_set(&counter->count, 0); + atomic64_set(&counter->count, 0); perf_counter_update_userpage(counter); } -- cgit v1.2.3 From 5031296c57024a78ddad4edfc993367dbf4abb98 Mon Sep 17 00:00:00 2001 From: "H. Peter Anvin" Date: Thu, 7 May 2009 16:54:11 -0700 Subject: x86: add extension fields for bootloader type and version A long ago, in days of yore, it all began with a god named Thor. There were vikings and boats and some plans for a Linux kernel header. Unfortunately, a single 8-bit field was used for bootloader type and version. This has generally worked without *too* much pain, but we're getting close to flat running out of ID fields. Add extension fields for both type and version. The type will be extended if it the old field is 0xE; the version is a simple MSB extension. Keep /proc/sys/kernel/bootloader_type containing (type << 4) + (ver & 0xf) for backwards compatiblity, but also add /proc/sys/kernel/bootloader_version which contains the full version number. [ Impact: new feature to support more bootloaders ] Signed-off-by: H. Peter Anvin --- kernel/sysctl.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'kernel') diff --git a/kernel/sysctl.c b/kernel/sysctl.c index e3d2c7dd59b9..cf91c9317b26 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -727,6 +727,14 @@ static struct ctl_table kern_table[] = { .mode = 0444, .proc_handler = &proc_dointvec, }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "bootloader_version", + .data = &bootloader_version, + .maxlen = sizeof (int), + .mode = 0444, + .proc_handler = &proc_dointvec, + }, { .ctl_name = CTL_UNNUMBERED, .procname = "kstack_depth_to_print", -- cgit v1.2.3 From e758a33d6fc5b9d6a3ae489863d04fcecad8120b Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Tue, 12 May 2009 21:59:01 +1000 Subject: perf_counter: call hw_perf_save_disable/restore around group_sched_in I noticed that when enabling a group via the PERF_COUNTER_IOC_ENABLE ioctl on the group leader, the counters weren't enabled and counting immediately on return from the ioctl, but did start counting a little while later (presumably after a context switch). The reason was that __perf_counter_enable calls group_sched_in which calls hw_perf_group_sched_in, which on powerpc assumes that the caller has called hw_perf_save_disable already. Until commit 46d686c6 ("perf_counter: put whole group on when enabling group leader") it was true that all callers of group_sched_in had called hw_perf_save_disable first, and the powerpc hw_perf_group_sched_in relies on that (there isn't an x86 version). This fixes the problem by putting calls to hw_perf_save_disable / hw_perf_restore around the calls to group_sched_in and counter_sched_in in __perf_counter_enable. Having the calls to hw_perf_save_disable/restore around the counter_sched_in call is harmless and makes this call consistent with the other call sites of counter_sched_in, which have all called hw_perf_save_disable first. [ Impact: more precise counter group disable/enable functionality ] Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Corey Ashford LKML-Reference: <18953.25733.53359.147452@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 19 ++++++++++++------- 1 file changed, 12 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 5ea0240adab2..ff166c11b69a 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -663,6 +663,7 @@ static void __perf_counter_enable(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter_context *ctx = counter->ctx; struct perf_counter *leader = counter->group_leader; + unsigned long pmuflags; unsigned long flags; int err; @@ -689,14 +690,18 @@ static void __perf_counter_enable(void *info) if (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE) goto unlock; - if (!group_can_go_on(counter, cpuctx, 1)) + if (!group_can_go_on(counter, cpuctx, 1)) { err = -EEXIST; - else if (counter == leader) - err = group_sched_in(counter, cpuctx, ctx, - smp_processor_id()); - else - err = counter_sched_in(counter, cpuctx, ctx, - smp_processor_id()); + } else { + pmuflags = hw_perf_save_disable(); + if (counter == leader) + err = group_sched_in(counter, cpuctx, ctx, + smp_processor_id()); + else + err = counter_sched_in(counter, cpuctx, ctx, + smp_processor_id()); + hw_perf_restore(pmuflags); + } if (err) { /* -- cgit v1.2.3 From d80c19df5fcceb8c741e96f09f275c2da719efef Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 12 May 2009 16:29:13 +0200 Subject: lockdep: increase MAX_LOCKDEP_ENTRIES and MAX_LOCKDEP_CHAINS Now that lockdep coverage has increased it has become easier to run out of entries: [ 21.401387] BUG: MAX_LOCKDEP_ENTRIES too low! [ 21.402007] turning off the locking correctness validator. [ 21.402007] Pid: 1555, comm: S99local Not tainted 2.6.30-rc5-tip #2 [ 21.402007] Call Trace: [ 21.402007] [] add_lock_to_list+0x53/0xba [ 21.402007] [] ? lookup_mnt+0x19/0x53 [ 21.402007] [] check_prev_add+0x14b/0x1c7 [ 21.402007] [] validate_chain+0x474/0x52a [ 21.402007] [] __lock_acquire+0x342/0x3c7 [ 21.402007] [] lock_acquire+0xc1/0xe5 [ 21.402007] [] ? lookup_mnt+0x19/0x53 [ 21.402007] [] _spin_lock+0x31/0x66 Double the size - as we've done in the past. [ Impact: allow lockdep to cover more locks ] Acked-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/lockdep_internals.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/lockdep_internals.h b/kernel/lockdep_internals.h index a2cc7e9a6e84..699a2ac3a0d7 100644 --- a/kernel/lockdep_internals.h +++ b/kernel/lockdep_internals.h @@ -54,9 +54,9 @@ enum { * table (if it's not there yet), and we check it for lock order * conflicts and deadlocks. */ -#define MAX_LOCKDEP_ENTRIES 8192UL +#define MAX_LOCKDEP_ENTRIES 16384UL -#define MAX_LOCKDEP_CHAINS_BITS 14 +#define MAX_LOCKDEP_CHAINS_BITS 15 #define MAX_LOCKDEP_CHAINS (1UL << MAX_LOCKDEP_CHAINS_BITS) #define MAX_LOCKDEP_CHAIN_HLOCKS (MAX_LOCKDEP_CHAINS*5) -- cgit v1.2.3 From 597d0275736dad9c3bda6f0a00a1c477dc0f37b1 Mon Sep 17 00:00:00 2001 From: Arun R Bharadwaj Date: Thu, 16 Apr 2009 12:13:26 +0530 Subject: timers: Framework for identifying pinned timers * Arun R Bharadwaj [2009-04-16 12:11:36]: This patch creates a new framework for identifying cpu-pinned timers and hrtimers. This framework is needed because pinned timers are expected to fire on the same CPU on which they are queued. So it is essential to identify these and not migrate them, in case there are any. For regular timers, the currently existing add_timer_on() can be used queue pinned timers and subsequently mod_timer_pinned() can be used to modify the 'expires' field. For hrtimers, new modes HRTIMER_ABS_PINNED and HRTIMER_REL_PINNED are added to queue cpu-pinned hrtimer. [ tglx: use .._PINNED mode argument instead of creating tons of new functions ] Signed-off-by: Arun R Bharadwaj Signed-off-by: Thomas Gleixner --- kernel/hrtimer.c | 7 ++++--- kernel/timer.c | 31 +++++++++++++++++++++++++++---- 2 files changed, 31 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index cb8a15c19583..c71bcd549241 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -193,7 +193,8 @@ struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, * Switch the timer base to the current CPU when possible. */ static inline struct hrtimer_clock_base * -switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base) +switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, + int pinned) { struct hrtimer_clock_base *new_base; struct hrtimer_cpu_base *new_cpu_base; @@ -907,9 +908,9 @@ int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, ret = remove_hrtimer(timer, base); /* Switch the timer base, if necessary: */ - new_base = switch_hrtimer_base(timer, base); + new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); - if (mode == HRTIMER_MODE_REL) { + if (mode & HRTIMER_MODE_REL) { tim = ktime_add_safe(tim, new_base->get_time()); /* * CONFIG_TIME_LOW_RES is a temporary way for architectures diff --git a/kernel/timer.c b/kernel/timer.c index 5c1e84beaf4a..3424dfd11d50 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -604,7 +604,8 @@ static struct tvec_base *lock_timer_base(struct timer_list *timer, } static inline int -__mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only) +__mod_timer(struct timer_list *timer, unsigned long expires, + bool pending_only, int pinned) { struct tvec_base *base, *new_base; unsigned long flags; @@ -668,7 +669,7 @@ out_unlock: */ int mod_timer_pending(struct timer_list *timer, unsigned long expires) { - return __mod_timer(timer, expires, true); + return __mod_timer(timer, expires, true, TIMER_NOT_PINNED); } EXPORT_SYMBOL(mod_timer_pending); @@ -702,10 +703,32 @@ int mod_timer(struct timer_list *timer, unsigned long expires) if (timer->expires == expires && timer_pending(timer)) return 1; - return __mod_timer(timer, expires, false); + return __mod_timer(timer, expires, false, TIMER_NOT_PINNED); } EXPORT_SYMBOL(mod_timer); +/** + * mod_timer_pinned - modify a timer's timeout + * @timer: the timer to be modified + * @expires: new timeout in jiffies + * + * mod_timer_pinned() is a way to update the expire field of an + * active timer (if the timer is inactive it will be activated) + * and not allow the timer to be migrated to a different CPU. + * + * mod_timer_pinned(timer, expires) is equivalent to: + * + * del_timer(timer); timer->expires = expires; add_timer(timer); + */ +int mod_timer_pinned(struct timer_list *timer, unsigned long expires) +{ + if (timer->expires == expires && timer_pending(timer)) + return 1; + + return __mod_timer(timer, expires, false, TIMER_PINNED); +} +EXPORT_SYMBOL(mod_timer_pinned); + /** * add_timer - start a timer * @timer: the timer to be added @@ -1356,7 +1379,7 @@ signed long __sched schedule_timeout(signed long timeout) expire = timeout + jiffies; setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); - __mod_timer(&timer, expire, false); + __mod_timer(&timer, expire, false, TIMER_NOT_PINNED); schedule(); del_singleshot_timer_sync(&timer); -- cgit v1.2.3 From 5c333864a6ba811052d52ef14fbed056b9ac3512 Mon Sep 17 00:00:00 2001 From: Arun R Bharadwaj Date: Thu, 16 Apr 2009 12:14:37 +0530 Subject: timers: Identifying the existing pinned timers * Arun R Bharadwaj [2009-04-16 12:11:36]: The following pinned hrtimers have been identified and marked: 1)sched_rt_period_timer 2)tick_sched_timer 3)stack_trace_timer_fn [ tglx: fixup the hrtimer pinned mode ] Signed-off-by: Arun R Bharadwaj Signed-off-by: Thomas Gleixner --- kernel/sched.c | 4 ++-- kernel/time/tick-sched.c | 7 ++++--- kernel/trace/trace_sysprof.c | 3 ++- 3 files changed, 8 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index b902e587a3a0..9c5b4d3f97ab 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -244,7 +244,7 @@ static void start_rt_bandwidth(struct rt_bandwidth *rt_b) hard = hrtimer_get_expires(&rt_b->rt_period_timer); delta = ktime_to_ns(ktime_sub(hard, soft)); __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, - HRTIMER_MODE_ABS, 0); + HRTIMER_MODE_ABS_PINNED, 0); } spin_unlock(&rt_b->rt_runtime_lock); } @@ -1154,7 +1154,7 @@ static __init void init_hrtick(void) static void hrtick_start(struct rq *rq, u64 delay) { __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, - HRTIMER_MODE_REL, 0); + HRTIMER_MODE_REL_PINNED, 0); } static inline void init_hrtick(void) diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c index d3f1ef4d5cbe..2aff39c6f10c 100644 --- a/kernel/time/tick-sched.c +++ b/kernel/time/tick-sched.c @@ -349,7 +349,7 @@ void tick_nohz_stop_sched_tick(int inidle) if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { hrtimer_start(&ts->sched_timer, expires, - HRTIMER_MODE_ABS); + HRTIMER_MODE_ABS_PINNED); /* Check, if the timer was already in the past */ if (hrtimer_active(&ts->sched_timer)) goto out; @@ -395,7 +395,7 @@ static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { hrtimer_start_expires(&ts->sched_timer, - HRTIMER_MODE_ABS); + HRTIMER_MODE_ABS_PINNED); /* Check, if the timer was already in the past */ if (hrtimer_active(&ts->sched_timer)) break; @@ -698,7 +698,8 @@ void tick_setup_sched_timer(void) for (;;) { hrtimer_forward(&ts->sched_timer, now, tick_period); - hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS); + hrtimer_start_expires(&ts->sched_timer, + HRTIMER_MODE_ABS_PINNED); /* Check, if the timer was already in the past */ if (hrtimer_active(&ts->sched_timer)) break; diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index 91fd19c2149f..d180554bc935 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -203,7 +203,8 @@ static void start_stack_timer(void *unused) hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hrtimer->function = stack_trace_timer_fn; - hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL); + hrtimer_start(hrtimer, ns_to_ktime(sample_period), + HRTIMER_MODE_REL_PINNED); } static void start_stack_timers(void) -- cgit v1.2.3 From cd1bb94b4a0531e8211a3774f17de831f8285f76 Mon Sep 17 00:00:00 2001 From: Arun R Bharadwaj Date: Thu, 16 Apr 2009 12:15:34 +0530 Subject: timers: /proc/sys sysctl hook to enable timer migration * Arun R Bharadwaj [2009-04-16 12:11:36]: This patch creates the /proc/sys sysctl interface at /proc/sys/kernel/timer_migration Timer migration is enabled by default. To disable timer migration, when CONFIG_SCHED_DEBUG = y, echo 0 > /proc/sys/kernel/timer_migration Signed-off-by: Arun R Bharadwaj Signed-off-by: Thomas Gleixner --- kernel/sched.c | 2 ++ kernel/sysctl.c | 8 ++++++++ 2 files changed, 10 insertions(+) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 9c5b4d3f97ab..7f1dd56af863 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -8731,6 +8731,8 @@ void __init sched_init_smp(void) } #endif /* CONFIG_SMP */ +const_debug unsigned int sysctl_timer_migration = 1; + int in_sched_functions(unsigned long addr) { return in_lock_functions(addr) || diff --git a/kernel/sysctl.c b/kernel/sysctl.c index e3d2c7dd59b9..b3ce58137303 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -324,6 +324,14 @@ static struct ctl_table kern_table[] = { .mode = 0644, .proc_handler = &proc_dointvec, }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "timer_migration", + .data = &sysctl_timer_migration, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, #endif { .ctl_name = CTL_UNNUMBERED, -- cgit v1.2.3 From eea08f32adb3f97553d49a4f79a119833036000a Mon Sep 17 00:00:00 2001 From: Arun R Bharadwaj Date: Thu, 16 Apr 2009 12:16:41 +0530 Subject: timers: Logic to move non pinned timers * Arun R Bharadwaj [2009-04-16 12:11:36]: This patch migrates all non pinned timers and hrtimers to the current idle load balancer, from all the idle CPUs. Timers firing on busy CPUs are not migrated. While migrating hrtimers, care should be taken to check if migrating a hrtimer would result in a latency or not. So we compare the expiry of the hrtimer with the next timer interrupt on the target cpu and migrate the hrtimer only if it expires *after* the next interrupt on the target cpu. So, added a clockevents_get_next_event() helper function to return the next_event on the target cpu's clock_event_device. [ tglx: cleanups and simplifications ] Signed-off-by: Arun R Bharadwaj Signed-off-by: Thomas Gleixner --- kernel/hrtimer.c | 51 +++++++++++++++++++++++++++++++++++++++++++++-- kernel/sched.c | 5 +++++ kernel/time/clockevents.c | 12 +++++++++++ kernel/timer.c | 17 +++++++++++++--- 4 files changed, 80 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index c71bcd549241..b675a67c9ac3 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -43,6 +43,8 @@ #include #include #include +#include +#include #include @@ -198,8 +200,19 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, { struct hrtimer_clock_base *new_base; struct hrtimer_cpu_base *new_cpu_base; + int cpu, preferred_cpu = -1; + + cpu = smp_processor_id(); +#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP) + if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) { + preferred_cpu = get_nohz_load_balancer(); + if (preferred_cpu >= 0) + cpu = preferred_cpu; + } +#endif - new_cpu_base = &__get_cpu_var(hrtimer_bases); +again: + new_cpu_base = &per_cpu(hrtimer_bases, cpu); new_base = &new_cpu_base->clock_base[base->index]; if (base != new_base) { @@ -219,6 +232,40 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, timer->base = NULL; spin_unlock(&base->cpu_base->lock); spin_lock(&new_base->cpu_base->lock); + + /* Optimized away for NOHZ=n SMP=n */ + if (cpu == preferred_cpu) { + /* Calculate clock monotonic expiry time */ +#ifdef CONFIG_HIGH_RES_TIMERS + ktime_t expires = ktime_sub(hrtimer_get_expires(timer), + new_base->offset); +#else + ktime_t expires = hrtimer_get_expires(timer); +#endif + + /* + * Get the next event on target cpu from the + * clock events layer. + * This covers the highres=off nohz=on case as well. + */ + ktime_t next = clockevents_get_next_event(cpu); + + ktime_t delta = ktime_sub(expires, next); + + /* + * We do not migrate the timer when it is expiring + * before the next event on the target cpu because + * we cannot reprogram the target cpu hardware and + * we would cause it to fire late. + */ + if (delta.tv64 < 0) { + cpu = smp_processor_id(); + spin_unlock(&new_base->cpu_base->lock); + spin_lock(&base->cpu_base->lock); + timer->base = base; + goto again; + } + } timer->base = new_base; } return new_base; @@ -236,7 +283,7 @@ lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) return base; } -# define switch_hrtimer_base(t, b) (b) +# define switch_hrtimer_base(t, b, p) (b) #endif /* !CONFIG_SMP */ diff --git a/kernel/sched.c b/kernel/sched.c index 7f1dd56af863..9fe3774a0fd3 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -4244,6 +4244,11 @@ static struct { .load_balancer = ATOMIC_INIT(-1), }; +int get_nohz_load_balancer(void) +{ + return atomic_read(&nohz.load_balancer); +} + /* * This routine will try to nominate the ilb (idle load balancing) * owner among the cpus whose ticks are stopped. ilb owner will do the idle diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c index d13be216a790..ab20ded013bd 100644 --- a/kernel/time/clockevents.c +++ b/kernel/time/clockevents.c @@ -18,6 +18,7 @@ #include #include #include +#include /* The registered clock event devices */ static LIST_HEAD(clockevent_devices); @@ -251,4 +252,15 @@ void clockevents_notify(unsigned long reason, void *arg) spin_unlock(&clockevents_lock); } EXPORT_SYMBOL_GPL(clockevents_notify); + +ktime_t clockevents_get_next_event(int cpu) +{ + struct tick_device *td; + struct clock_event_device *dev; + + td = &per_cpu(tick_cpu_device, cpu); + dev = td->evtdev; + + return dev->next_event; +} #endif diff --git a/kernel/timer.c b/kernel/timer.c index 3424dfd11d50..3f841db5edf9 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -37,6 +37,7 @@ #include #include #include +#include #include #include @@ -609,9 +610,7 @@ __mod_timer(struct timer_list *timer, unsigned long expires, { struct tvec_base *base, *new_base; unsigned long flags; - int ret; - - ret = 0; + int ret = 0 , cpu; timer_stats_timer_set_start_info(timer); BUG_ON(!timer->function); @@ -630,6 +629,18 @@ __mod_timer(struct timer_list *timer, unsigned long expires, new_base = __get_cpu_var(tvec_bases); + cpu = smp_processor_id(); + +#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP) + if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) { + int preferred_cpu = get_nohz_load_balancer(); + + if (preferred_cpu >= 0) + cpu = preferred_cpu; + } +#endif + new_base = per_cpu(tvec_bases, cpu); + if (base != new_base) { /* * We are trying to schedule the timer on the local CPU. -- cgit v1.2.3 From 53020fe81eecd0b7be295868ce5850ef8f41074e Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 13 May 2009 21:26:19 +0200 Subject: perf_counter: Fix perf_output_copy() WARN to account for overflow The simple reservation test in perf_output_copy() failed to take unsigned int overflow into account, fix this. [ Impact: fix false positive warning with more than 4GB of profiling data ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ff166c11b69a..985be0b662af 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1927,7 +1927,11 @@ static void perf_output_copy(struct perf_output_handle *handle, handle->offset = offset; - WARN_ON_ONCE(handle->offset > handle->head); + /* + * Check we didn't copy past our reservation window, taking the + * possible unsigned int wrap into account. + */ + WARN_ON_ONCE(((int)(handle->head - handle->offset)) < 0); } #define perf_output_put(handle, x) \ -- cgit v1.2.3 From 9e35ad388bea89f7d6f375af4c0ae98803688666 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 13 May 2009 16:21:38 +0200 Subject: perf_counter: Rework the perf counter disable/enable The current disable/enable mechanism is: token = hw_perf_save_disable(); ... /* do bits */ ... hw_perf_restore(token); This works well, provided that the use nests properly. Except we don't. x86 NMI/INT throttling has non-nested use of this, breaking things. Therefore provide a reference counter disable/enable interface, where the first disable disables the hardware, and the last enable enables the hardware again. [ Impact: refactor, simplify the PMU disable/enable logic ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 76 +++++++++++++++++++++++++++++++-------------------- 1 file changed, 47 insertions(+), 29 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 985be0b662af..e814ff04d7ca 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -60,8 +60,9 @@ extern __weak const struct pmu *hw_perf_counter_init(struct perf_counter *counte return NULL; } -u64 __weak hw_perf_save_disable(void) { return 0; } -void __weak hw_perf_restore(u64 ctrl) { barrier(); } +void __weak hw_perf_disable(void) { barrier(); } +void __weak hw_perf_enable(void) { barrier(); } + void __weak hw_perf_counter_setup(int cpu) { barrier(); } int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, struct perf_cpu_context *cpuctx, @@ -72,6 +73,32 @@ int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, void __weak perf_counter_print_debug(void) { } +static DEFINE_PER_CPU(int, disable_count); + +void __perf_disable(void) +{ + __get_cpu_var(disable_count)++; +} + +bool __perf_enable(void) +{ + return !--__get_cpu_var(disable_count); +} + +void perf_disable(void) +{ + __perf_disable(); + hw_perf_disable(); +} +EXPORT_SYMBOL_GPL(perf_disable); /* ACPI idle */ + +void perf_enable(void) +{ + if (__perf_enable()) + hw_perf_enable(); +} +EXPORT_SYMBOL_GPL(perf_enable); /* ACPI idle */ + static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { @@ -170,7 +197,6 @@ static void __perf_counter_remove_from_context(void *info) struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; unsigned long flags; - u64 perf_flags; /* * If this is a task context, we need to check whether it is @@ -191,9 +217,9 @@ static void __perf_counter_remove_from_context(void *info) * Protect the list operation against NMI by disabling the * counters on a global level. NOP for non NMI based counters. */ - perf_flags = hw_perf_save_disable(); + perf_disable(); list_del_counter(counter, ctx); - hw_perf_restore(perf_flags); + perf_enable(); if (!ctx->task) { /* @@ -538,7 +564,6 @@ static void __perf_install_in_context(void *info) struct perf_counter *leader = counter->group_leader; int cpu = smp_processor_id(); unsigned long flags; - u64 perf_flags; int err; /* @@ -556,7 +581,7 @@ static void __perf_install_in_context(void *info) * Protect the list operation against NMI by disabling the * counters on a global level. NOP for non NMI based counters. */ - perf_flags = hw_perf_save_disable(); + perf_disable(); add_counter_to_ctx(counter, ctx); @@ -596,7 +621,7 @@ static void __perf_install_in_context(void *info) cpuctx->max_pertask--; unlock: - hw_perf_restore(perf_flags); + perf_enable(); spin_unlock_irqrestore(&ctx->lock, flags); } @@ -663,7 +688,6 @@ static void __perf_counter_enable(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter_context *ctx = counter->ctx; struct perf_counter *leader = counter->group_leader; - unsigned long pmuflags; unsigned long flags; int err; @@ -693,14 +717,14 @@ static void __perf_counter_enable(void *info) if (!group_can_go_on(counter, cpuctx, 1)) { err = -EEXIST; } else { - pmuflags = hw_perf_save_disable(); + perf_disable(); if (counter == leader) err = group_sched_in(counter, cpuctx, ctx, smp_processor_id()); else err = counter_sched_in(counter, cpuctx, ctx, smp_processor_id()); - hw_perf_restore(pmuflags); + perf_enable(); } if (err) { @@ -795,7 +819,6 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx) { struct perf_counter *counter; - u64 flags; spin_lock(&ctx->lock); ctx->is_active = 0; @@ -803,12 +826,12 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, goto out; update_context_time(ctx); - flags = hw_perf_save_disable(); + perf_disable(); if (ctx->nr_active) { list_for_each_entry(counter, &ctx->counter_list, list_entry) group_sched_out(counter, cpuctx, ctx); } - hw_perf_restore(flags); + perf_enable(); out: spin_unlock(&ctx->lock); } @@ -860,7 +883,6 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, struct perf_cpu_context *cpuctx, int cpu) { struct perf_counter *counter; - u64 flags; int can_add_hw = 1; spin_lock(&ctx->lock); @@ -870,7 +892,7 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, ctx->timestamp = perf_clock(); - flags = hw_perf_save_disable(); + perf_disable(); /* * First go through the list and put on any pinned groups @@ -917,7 +939,7 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, can_add_hw = 0; } } - hw_perf_restore(flags); + perf_enable(); out: spin_unlock(&ctx->lock); } @@ -955,7 +977,6 @@ int perf_counter_task_disable(void) struct perf_counter_context *ctx = &curr->perf_counter_ctx; struct perf_counter *counter; unsigned long flags; - u64 perf_flags; if (likely(!ctx->nr_counters)) return 0; @@ -969,7 +990,7 @@ int perf_counter_task_disable(void) /* * Disable all the counters: */ - perf_flags = hw_perf_save_disable(); + perf_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { if (counter->state != PERF_COUNTER_STATE_ERROR) { @@ -978,7 +999,7 @@ int perf_counter_task_disable(void) } } - hw_perf_restore(perf_flags); + perf_enable(); spin_unlock_irqrestore(&ctx->lock, flags); @@ -991,7 +1012,6 @@ int perf_counter_task_enable(void) struct perf_counter_context *ctx = &curr->perf_counter_ctx; struct perf_counter *counter; unsigned long flags; - u64 perf_flags; int cpu; if (likely(!ctx->nr_counters)) @@ -1007,7 +1027,7 @@ int perf_counter_task_enable(void) /* * Disable all the counters: */ - perf_flags = hw_perf_save_disable(); + perf_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { if (counter->state > PERF_COUNTER_STATE_OFF) @@ -1017,7 +1037,7 @@ int perf_counter_task_enable(void) ctx->time - counter->total_time_enabled; counter->hw_event.disabled = 0; } - hw_perf_restore(perf_flags); + perf_enable(); spin_unlock(&ctx->lock); @@ -1034,7 +1054,6 @@ int perf_counter_task_enable(void) static void rotate_ctx(struct perf_counter_context *ctx) { struct perf_counter *counter; - u64 perf_flags; if (!ctx->nr_counters) return; @@ -1043,12 +1062,12 @@ static void rotate_ctx(struct perf_counter_context *ctx) /* * Rotate the first entry last (works just fine for group counters too): */ - perf_flags = hw_perf_save_disable(); + perf_disable(); list_for_each_entry(counter, &ctx->counter_list, list_entry) { list_move_tail(&counter->list_entry, &ctx->counter_list); break; } - hw_perf_restore(perf_flags); + perf_enable(); spin_unlock(&ctx->lock); } @@ -3194,7 +3213,6 @@ __perf_counter_exit_task(struct task_struct *child, } else { struct perf_cpu_context *cpuctx; unsigned long flags; - u64 perf_flags; /* * Disable and unlink this counter. @@ -3203,7 +3221,7 @@ __perf_counter_exit_task(struct task_struct *child, * could still be processing it: */ local_irq_save(flags); - perf_flags = hw_perf_save_disable(); + perf_disable(); cpuctx = &__get_cpu_var(perf_cpu_context); @@ -3214,7 +3232,7 @@ __perf_counter_exit_task(struct task_struct *child, child_ctx->nr_counters--; - hw_perf_restore(perf_flags); + perf_enable(); local_irq_restore(flags); } -- cgit v1.2.3 From f1a11e0576c7a73d759d05d776692b2b2d37172b Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 5 May 2009 19:21:40 +0200 Subject: futex: remove the wait queue The waitqueue which is used in struct futex_q is a leftover from the futexfd implementation. There is no need to use a waitqueue at all, as the waiting task is the only user of it. The waitqueue just adds additional locking and a loop in the wake up path which both can be avoided. We have already a task reference in struct futex_q which is used for PI futexes. Use it for normal futexes as well and just wake up the task directly. The logic of signalling the futex wakeup via setting q->lock_ptr to NULL is kept with the difference that we set it NULL before doing the wakeup. This opens an exit race window vs. a non futex wake up of the to be woken up task, which we prevent with get_task_struct / put_task_struct on the waiter. [ Impact: simplification ] Signed-off-by: Thomas Gleixner --- kernel/futex.c | 58 +++++++++++++++++++++++++--------------------------------- 1 file changed, 25 insertions(+), 33 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index aec8bf89bf4e..157bfcd725b8 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -100,8 +100,8 @@ struct futex_pi_state { */ struct futex_q { struct plist_node list; - /* There can only be a single waiter */ - wait_queue_head_t waiter; + /* Waiter reference */ + struct task_struct *task; /* Which hash list lock to use: */ spinlock_t *lock_ptr; @@ -111,7 +111,6 @@ struct futex_q { /* Optional priority inheritance state: */ struct futex_pi_state *pi_state; - struct task_struct *task; /* rt_waiter storage for requeue_pi: */ struct rt_mutex_waiter *rt_waiter; @@ -694,22 +693,29 @@ retry: */ static void wake_futex(struct futex_q *q) { - plist_del(&q->list, &q->list.plist); + struct task_struct *p = q->task; + /* - * The lock in wake_up_all() is a crucial memory barrier after the - * plist_del() and also before assigning to q->lock_ptr. + * We set q->lock_ptr = NULL _before_ we wake up the task. If + * a non futex wake up happens on another CPU then the task + * might exit and p would dereference a non existing task + * struct. Prevent this by holding a reference on p across the + * wake up. */ - wake_up(&q->waiter); + get_task_struct(p); + + plist_del(&q->list, &q->list.plist); /* - * The waiting task can free the futex_q as soon as this is written, - * without taking any locks. This must come last. - * - * A memory barrier is required here to prevent the following store to - * lock_ptr from getting ahead of the wakeup. Clearing the lock at the - * end of wake_up() does not prevent this store from moving. + * The waiting task can free the futex_q as soon as + * q->lock_ptr = NULL is written, without taking any locks. A + * memory barrier is required here to prevent the following + * store to lock_ptr from getting ahead of the plist_del. */ smp_wmb(); q->lock_ptr = NULL; + + wake_up_state(p, TASK_NORMAL); + put_task_struct(p); } static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this) @@ -1003,7 +1009,7 @@ void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key) WARN_ON(!q->rt_waiter); q->rt_waiter = NULL; - wake_up(&q->waiter); + wake_up_state(q->task, TASK_NORMAL); } /** @@ -1280,8 +1286,6 @@ static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) { struct futex_hash_bucket *hb; - init_waitqueue_head(&q->waiter); - get_futex_key_refs(&q->key); hb = hash_futex(&q->key); q->lock_ptr = &hb->lock; @@ -1575,11 +1579,9 @@ out: * @hb: the futex hash bucket, must be locked by the caller * @q: the futex_q to queue up on * @timeout: the prepared hrtimer_sleeper, or null for no timeout - * @wait: the wait_queue to add to the futex_q after queueing in the hb */ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, - struct hrtimer_sleeper *timeout, - wait_queue_t *wait) + struct hrtimer_sleeper *timeout) { queue_me(q, hb); @@ -1587,19 +1589,11 @@ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, * There might have been scheduling since the queue_me(), as we * cannot hold a spinlock across the get_user() in case it * faults, and we cannot just set TASK_INTERRUPTIBLE state when - * queueing ourselves into the futex hash. This code thus has to + * queueing ourselves into the futex hash. This code thus has to * rely on the futex_wake() code removing us from hash when it * wakes us up. */ - - /* add_wait_queue is the barrier after __set_current_state. */ - __set_current_state(TASK_INTERRUPTIBLE); - - /* - * Add current as the futex_q waiter. We don't remove ourselves from - * the wait_queue because we are the only user of it. - */ - add_wait_queue(&q->waiter, wait); + set_current_state(TASK_INTERRUPTIBLE); /* Arm the timer */ if (timeout) { @@ -1704,7 +1698,6 @@ static int futex_wait(u32 __user *uaddr, int fshared, u32 val, ktime_t *abs_time, u32 bitset, int clockrt) { struct hrtimer_sleeper timeout, *to = NULL; - DECLARE_WAITQUEUE(wait, current); struct restart_block *restart; struct futex_hash_bucket *hb; struct futex_q q; @@ -1733,7 +1726,7 @@ static int futex_wait(u32 __user *uaddr, int fshared, goto out; /* queue_me and wait for wakeup, timeout, or a signal. */ - futex_wait_queue_me(hb, &q, to, &wait); + futex_wait_queue_me(hb, &q, to); /* If we were woken (and unqueued), we succeeded, whatever. */ ret = 0; @@ -2147,7 +2140,6 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, struct hrtimer_sleeper timeout, *to = NULL; struct rt_mutex_waiter rt_waiter; struct rt_mutex *pi_mutex = NULL; - DECLARE_WAITQUEUE(wait, current); struct restart_block *restart; struct futex_hash_bucket *hb; union futex_key key2; @@ -2191,7 +2183,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, } /* Queue the futex_q, drop the hb lock, wait for wakeup. */ - futex_wait_queue_me(hb, &q, to, &wait); + futex_wait_queue_me(hb, &q, to); spin_lock(&hb->lock); ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); -- cgit v1.2.3 From 548e1ddf255b4ebfb4ef20c08936fd8d4deb3bd9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 15:19:26 +0200 Subject: perf_counter: remove perf_disable/enable exports Now that ACPI idle doesn't use it anymore, remove the exports. [ Impact: remove dead code/data ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo LKML-Reference: <20090515132018.429826617@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index e814ff04d7ca..0173738dd548 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -90,14 +90,12 @@ void perf_disable(void) __perf_disable(); hw_perf_disable(); } -EXPORT_SYMBOL_GPL(perf_disable); /* ACPI idle */ void perf_enable(void) { if (__perf_enable()) hw_perf_enable(); } -EXPORT_SYMBOL_GPL(perf_enable); /* ACPI idle */ static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) -- cgit v1.2.3 From 789f90fcf6b0b54e655740e9396c954378542c79 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 15:19:27 +0200 Subject: perf_counter: per user mlock gift Instead of a per-process mlock gift for perf-counters, use a per-user gift so that there is less of a DoS potential. [ Impact: allow less worst-case unprivileged memory consumption ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo LKML-Reference: <20090515132018.496182835@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0173738dd548..93f4a0e4b873 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -45,7 +45,7 @@ static atomic_t nr_munmap_tracking __read_mostly; static atomic_t nr_comm_tracking __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ -int sysctl_perf_counter_mlock __read_mostly = 128; /* 'free' kb per counter */ +int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ /* * Lock for (sysadmin-configurable) counter reservations: @@ -1522,6 +1522,9 @@ static void perf_mmap_close(struct vm_area_struct *vma) if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { + struct user_struct *user = current_user(); + + atomic_long_sub(counter->data->nr_pages + 1, &user->locked_vm); vma->vm_mm->locked_vm -= counter->data->nr_locked; perf_mmap_data_free(counter); mutex_unlock(&counter->mmap_mutex); @@ -1537,11 +1540,13 @@ static struct vm_operations_struct perf_mmap_vmops = { static int perf_mmap(struct file *file, struct vm_area_struct *vma) { struct perf_counter *counter = file->private_data; + struct user_struct *user = current_user(); unsigned long vma_size; unsigned long nr_pages; + unsigned long user_locked, user_lock_limit; unsigned long locked, lock_limit; + long user_extra, extra; int ret = 0; - long extra; if (!(vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_WRITE)) return -EINVAL; @@ -1569,15 +1574,17 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; } - extra = nr_pages /* + 1 only account the data pages */; - extra -= sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); - if (extra < 0) - extra = 0; + user_extra = nr_pages + 1; + user_lock_limit = sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); + user_locked = atomic_long_read(&user->locked_vm) + user_extra; - locked = vma->vm_mm->locked_vm + extra; + extra = 0; + if (user_locked > user_lock_limit) + extra = user_locked - user_lock_limit; lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; lock_limit >>= PAGE_SHIFT; + locked = vma->vm_mm->locked_vm + extra; if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) { ret = -EPERM; @@ -1590,6 +1597,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) goto unlock; atomic_set(&counter->mmap_count, 1); + atomic_long_add(user_extra, &user->locked_vm); vma->vm_mm->locked_vm += extra; counter->data->nr_locked = extra; unlock: -- cgit v1.2.3 From 60db5e09c13109b13830cc9dcae688003fd39e79 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 15:19:28 +0200 Subject: perf_counter: frequency based adaptive irq_period Instead of specifying the irq_period for a counter, provide a target interrupt frequency and dynamically adapt the irq_period to match this frequency. [ Impact: new perf-counter attribute/feature ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo LKML-Reference: <20090515132018.646195868@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 63 +++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 51 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 93f4a0e4b873..0ad1db4f3d65 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1046,6 +1046,38 @@ int perf_counter_task_enable(void) return 0; } +void perf_adjust_freq(struct perf_counter_context *ctx) +{ + struct perf_counter *counter; + u64 irq_period; + u64 events, period; + s64 delta; + + spin_lock(&ctx->lock); + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (counter->state != PERF_COUNTER_STATE_ACTIVE) + continue; + + if (!counter->hw_event.freq || !counter->hw_event.irq_freq) + continue; + + events = HZ * counter->hw.interrupts * counter->hw.irq_period; + period = div64_u64(events, counter->hw_event.irq_freq); + + delta = (s64)(1 + period - counter->hw.irq_period); + delta >>= 1; + + irq_period = counter->hw.irq_period + delta; + + if (!irq_period) + irq_period = 1; + + counter->hw.irq_period = irq_period; + counter->hw.interrupts = 0; + } + spin_unlock(&ctx->lock); +} + /* * Round-robin a context's counters: */ @@ -1081,6 +1113,9 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) cpuctx = &per_cpu(perf_cpu_context, cpu); ctx = &curr->perf_counter_ctx; + perf_adjust_freq(&cpuctx->ctx); + perf_adjust_freq(ctx); + perf_counter_cpu_sched_out(cpuctx); __perf_counter_task_sched_out(ctx); @@ -2382,6 +2417,8 @@ int perf_counter_overflow(struct perf_counter *counter, int events = atomic_read(&counter->event_limit); int ret = 0; + counter->hw.interrupts++; + /* * XXX event_limit might not quite work as expected on inherited * counters @@ -2450,6 +2487,7 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) enum hrtimer_restart ret = HRTIMER_RESTART; struct perf_counter *counter; struct pt_regs *regs; + u64 period; counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); counter->pmu->read(counter); @@ -2468,7 +2506,8 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) ret = HRTIMER_NORESTART; } - hrtimer_forward_now(hrtimer, ns_to_ktime(counter->hw.irq_period)); + period = max_t(u64, 10000, counter->hw.irq_period); + hrtimer_forward_now(hrtimer, ns_to_ktime(period)); return ret; } @@ -2629,8 +2668,9 @@ static int cpu_clock_perf_counter_enable(struct perf_counter *counter) hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hwc->hrtimer.function = perf_swcounter_hrtimer; if (hwc->irq_period) { + u64 period = max_t(u64, 10000, hwc->irq_period); __hrtimer_start_range_ns(&hwc->hrtimer, - ns_to_ktime(hwc->irq_period), 0, + ns_to_ktime(period), 0, HRTIMER_MODE_REL, 0); } @@ -2679,8 +2719,9 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hwc->hrtimer.function = perf_swcounter_hrtimer; if (hwc->irq_period) { + u64 period = max_t(u64, 10000, hwc->irq_period); __hrtimer_start_range_ns(&hwc->hrtimer, - ns_to_ktime(hwc->irq_period), 0, + ns_to_ktime(period), 0, HRTIMER_MODE_REL, 0); } @@ -2811,9 +2852,7 @@ static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) { - struct perf_counter_hw_event *hw_event = &counter->hw_event; const struct pmu *pmu = NULL; - struct hw_perf_counter *hwc = &counter->hw; /* * Software counters (currently) can't in general distinguish @@ -2826,8 +2865,6 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_CPU_CLOCK: pmu = &perf_ops_cpu_clock; - if (hw_event->irq_period && hw_event->irq_period < 10000) - hw_event->irq_period = 10000; break; case PERF_COUNT_TASK_CLOCK: /* @@ -2839,8 +2876,6 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) else pmu = &perf_ops_cpu_clock; - if (hw_event->irq_period && hw_event->irq_period < 10000) - hw_event->irq_period = 10000; break; case PERF_COUNT_PAGE_FAULTS: case PERF_COUNT_PAGE_FAULTS_MIN: @@ -2854,9 +2889,6 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) break; } - if (pmu) - hwc->irq_period = hw_event->irq_period; - return pmu; } @@ -2872,6 +2904,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, { const struct pmu *pmu; struct perf_counter *counter; + struct hw_perf_counter *hwc; long err; counter = kzalloc(sizeof(*counter), gfpflags); @@ -2907,6 +2940,12 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, pmu = NULL; + hwc = &counter->hw; + if (hw_event->freq && hw_event->irq_freq) + hwc->irq_period = TICK_NSEC / hw_event->irq_freq; + else + hwc->irq_period = hw_event->irq_period; + /* * we currently do not support PERF_RECORD_GROUP on inherited counters */ -- cgit v1.2.3 From dce48a84adf1806676319f6f480e30a6daa012f9 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 11 Apr 2009 10:43:41 +0200 Subject: sched, timers: move calc_load() to scheduler Dimitri Sivanich noticed that xtime_lock is held write locked across calc_load() which iterates over all online CPUs. That can cause long latencies for xtime_lock readers on large SMP systems. The load average calculation is an rough estimate anyway so there is no real need to protect the readers vs. the update. It's not a problem when the avenrun array is updated while a reader copies the values. Instead of iterating over all online CPUs let the scheduler_tick code update the number of active tasks shortly before the avenrun update happens. The avenrun update itself is handled by the CPU which calls do_timer(). [ Impact: reduce xtime_lock write locked section ] Signed-off-by: Thomas Gleixner Acked-by: Peter Zijlstra --- kernel/sched.c | 84 +++++++++++++++++++++++++++++++++++++++++------ kernel/sched_idletask.c | 3 +- kernel/time/timekeeping.c | 2 +- kernel/timer.c | 54 ++---------------------------- 4 files changed, 79 insertions(+), 64 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 8908d190a348..f4eb88153bd1 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -630,6 +630,10 @@ struct rq { struct list_head migration_queue; #endif + /* calc_load related fields */ + unsigned long calc_load_update; + long calc_load_active; + #ifdef CONFIG_SCHED_HRTICK #ifdef CONFIG_SMP int hrtick_csd_pending; @@ -1728,6 +1732,8 @@ static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) } #endif +static void calc_load_account_active(struct rq *this_rq); + #include "sched_stats.h" #include "sched_idletask.c" #include "sched_fair.c" @@ -2856,19 +2862,57 @@ unsigned long nr_iowait(void) return sum; } -unsigned long nr_active(void) +/* Variables and functions for calc_load */ +static atomic_long_t calc_load_tasks; +static unsigned long calc_load_update; +unsigned long avenrun[3]; +EXPORT_SYMBOL(avenrun); + +static unsigned long +calc_load(unsigned long load, unsigned long exp, unsigned long active) { - unsigned long i, running = 0, uninterruptible = 0; + load *= exp; + load += active * (FIXED_1 - exp); + return load >> FSHIFT; +} - for_each_online_cpu(i) { - running += cpu_rq(i)->nr_running; - uninterruptible += cpu_rq(i)->nr_uninterruptible; - } +/* + * calc_load - update the avenrun load estimates 10 ticks after the + * CPUs have updated calc_load_tasks. + */ +void calc_global_load(void) +{ + unsigned long upd = calc_load_update + 10; + long active; + + if (time_before(jiffies, upd)) + return; - if (unlikely((long)uninterruptible < 0)) - uninterruptible = 0; + active = atomic_long_read(&calc_load_tasks); + active = active > 0 ? active * FIXED_1 : 0; - return running + uninterruptible; + avenrun[0] = calc_load(avenrun[0], EXP_1, active); + avenrun[1] = calc_load(avenrun[1], EXP_5, active); + avenrun[2] = calc_load(avenrun[2], EXP_15, active); + + calc_load_update += LOAD_FREQ; +} + +/* + * Either called from update_cpu_load() or from a cpu going idle + */ +static void calc_load_account_active(struct rq *this_rq) +{ + long nr_active, delta; + + nr_active = this_rq->nr_running; + nr_active += (long) this_rq->nr_uninterruptible; + + if (nr_active != this_rq->calc_load_active) { + delta = nr_active - this_rq->calc_load_active; + this_rq->calc_load_active = nr_active; + atomic_long_add(delta, &calc_load_tasks); + } } /* @@ -2899,6 +2943,11 @@ static void update_cpu_load(struct rq *this_rq) new_load += scale-1; this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; } + + if (time_after_eq(jiffies, this_rq->calc_load_update)) { + this_rq->calc_load_update += LOAD_FREQ; + calc_load_account_active(this_rq); + } } #ifdef CONFIG_SMP @@ -7091,6 +7140,14 @@ static void migrate_dead_tasks(unsigned int dead_cpu) } } + +/* + * remove the tasks which were accounted by rq from calc_load_tasks. + */ +static void calc_global_load_remove(struct rq *rq) +{ + atomic_long_sub(rq->calc_load_active, &calc_load_tasks); +} #endif /* CONFIG_HOTPLUG_CPU */ #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) @@ -7325,6 +7382,8 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) /* Update our root-domain */ rq = cpu_rq(cpu); spin_lock_irqsave(&rq->lock, flags); + rq->calc_load_update = calc_load_update; + rq->calc_load_active = 0; if (rq->rd) { BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); @@ -7364,7 +7423,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) cpuset_unlock(); migrate_nr_uninterruptible(rq); BUG_ON(rq->nr_running != 0); - + calc_global_load_remove(rq); /* * No need to migrate the tasks: it was best-effort if * they didn't take sched_hotcpu_mutex. Just wake up @@ -9059,6 +9118,8 @@ void __init sched_init(void) rq = cpu_rq(i); spin_lock_init(&rq->lock); rq->nr_running = 0; + rq->calc_load_active = 0; + rq->calc_load_update = jiffies + LOAD_FREQ; init_cfs_rq(&rq->cfs, rq); init_rt_rq(&rq->rt, rq); #ifdef CONFIG_FAIR_GROUP_SCHED @@ -9166,6 +9227,9 @@ void __init sched_init(void) * when this runqueue becomes "idle". */ init_idle(current, smp_processor_id()); + + calc_load_update = jiffies + LOAD_FREQ; + /* * During early bootup we pretend to be a normal task: */ diff --git a/kernel/sched_idletask.c b/kernel/sched_idletask.c index 8a21a2e28c13..499672c10cbd 100644 --- a/kernel/sched_idletask.c +++ b/kernel/sched_idletask.c @@ -22,7 +22,8 @@ static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int sy static struct task_struct *pick_next_task_idle(struct rq *rq) { schedstat_inc(rq, sched_goidle); - + /* adjust the active tasks as we might go into a long sleep */ + calc_load_account_active(rq); return rq->idle; } diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index 687dff49f6e7..52a8bf8931f3 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -22,7 +22,7 @@ /* * This read-write spinlock protects us from races in SMP while - * playing with xtime and avenrun. + * playing with xtime. */ __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); diff --git a/kernel/timer.c b/kernel/timer.c index cffffad01c31..6a21d7af9620 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -1122,47 +1122,6 @@ void update_process_times(int user_tick) run_posix_cpu_timers(p); } -/* - * Nr of active tasks - counted in fixed-point numbers - */ -static unsigned long count_active_tasks(void) -{ - return nr_active() * FIXED_1; -} - -/* - * Hmm.. Changed this, as the GNU make sources (load.c) seems to - * imply that avenrun[] is the standard name for this kind of thing. - * Nothing else seems to be standardized: the fractional size etc - * all seem to differ on different machines. - * - * Requires xtime_lock to access. - */ -unsigned long avenrun[3]; - -EXPORT_SYMBOL(avenrun); - -/* - * calc_load - given tick count, update the avenrun load estimates. - * This is called while holding a write_lock on xtime_lock. - */ -static inline void calc_load(unsigned long ticks) -{ - unsigned long active_tasks; /* fixed-point */ - static int count = LOAD_FREQ; - - count -= ticks; - if (unlikely(count < 0)) { - active_tasks = count_active_tasks(); - do { - CALC_LOAD(avenrun[0], EXP_1, active_tasks); - CALC_LOAD(avenrun[1], EXP_5, active_tasks); - CALC_LOAD(avenrun[2], EXP_15, active_tasks); - count += LOAD_FREQ; - } while (count < 0); - } -} - /* * This function runs timers and the timer-tq in bottom half context. */ @@ -1186,16 +1145,6 @@ void run_local_timers(void) softlockup_tick(); } -/* - * Called by the timer interrupt. xtime_lock must already be taken - * by the timer IRQ! - */ -static inline void update_times(unsigned long ticks) -{ - update_wall_time(); - calc_load(ticks); -} - /* * The 64-bit jiffies value is not atomic - you MUST NOT read it * without sampling the sequence number in xtime_lock. @@ -1205,7 +1154,8 @@ static inline void update_times(unsigned long ticks) void do_timer(unsigned long ticks) { jiffies_64 += ticks; - update_times(ticks); + update_wall_time(); + calc_global_load(); } #ifdef __ARCH_WANT_SYS_ALARM -- cgit v1.2.3 From 2d02494f5a90f2e4b3c4c6acc85ec94674cdc431 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 2 May 2009 20:08:52 +0200 Subject: sched, timers: cleanup avenrun users avenrun is an rough estimate so we don't have to worry about consistency of the three avenrun values. Remove the xtime lock dependency and provide a function to scale the values. Cleanup the users. [ Impact: cleanup ] Signed-off-by: Thomas Gleixner Acked-by: Peter Zijlstra --- kernel/sched.c | 15 +++++++++++++++ kernel/timer.c | 32 ++++++-------------------------- 2 files changed, 21 insertions(+), 26 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index f4eb88153bd1..497c09ba61e7 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2868,6 +2868,21 @@ static unsigned long calc_load_update; unsigned long avenrun[3]; EXPORT_SYMBOL(avenrun); +/** + * get_avenrun - get the load average array + * @loads: pointer to dest load array + * @offset: offset to add + * @shift: shift count to shift the result left + * + * These values are estimates at best, so no need for locking. + */ +void get_avenrun(unsigned long *loads, unsigned long offset, int shift) +{ + loads[0] = (avenrun[0] + offset) << shift; + loads[1] = (avenrun[1] + offset) << shift; + loads[2] = (avenrun[2] + offset) << shift; +} + static unsigned long calc_load(unsigned long load, unsigned long exp, unsigned long active) { diff --git a/kernel/timer.c b/kernel/timer.c index 6a21d7af9620..a26ed294f938 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -1356,37 +1356,17 @@ int do_sysinfo(struct sysinfo *info) { unsigned long mem_total, sav_total; unsigned int mem_unit, bitcount; - unsigned long seq; + struct timespec tp; memset(info, 0, sizeof(struct sysinfo)); - do { - struct timespec tp; - seq = read_seqbegin(&xtime_lock); - - /* - * This is annoying. The below is the same thing - * posix_get_clock_monotonic() does, but it wants to - * take the lock which we want to cover the loads stuff - * too. - */ - - getnstimeofday(&tp); - tp.tv_sec += wall_to_monotonic.tv_sec; - tp.tv_nsec += wall_to_monotonic.tv_nsec; - monotonic_to_bootbased(&tp); - if (tp.tv_nsec - NSEC_PER_SEC >= 0) { - tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC; - tp.tv_sec++; - } - info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); + ktime_get_ts(&tp); + monotonic_to_bootbased(&tp); + info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0); - info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT); - info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT); - info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT); + get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT); - info->procs = nr_threads; - } while (read_seqretry(&xtime_lock, seq)); + info->procs = nr_threads; si_meminfo(info); si_swapinfo(info); -- cgit v1.2.3 From 2e569d36729c8105ae066a9b105068305442cc77 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 15:37:47 +0200 Subject: perf_counter: frequency based adaptive irq_period, 32-bit fix fix: kernel/built-in.o: In function `perf_counter_alloc': perf_counter.c:(.text+0x7ddc7): undefined reference to `__udivdi3' [ Impact: build fix on 32-bit systems ] Reported-by: Ingo Molnar Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner LKML-Reference: <1242394667.6642.1887.camel@laptop> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0ad1db4f3d65..728a595399b0 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2942,7 +2942,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, hwc = &counter->hw; if (hw_event->freq && hw_event->irq_freq) - hwc->irq_period = TICK_NSEC / hw_event->irq_freq; + hwc->irq_period = div64_u64(TICK_NSEC, hw_event->irq_freq); else hwc->irq_period = hw_event->irq_period; -- cgit v1.2.3 From 9d23a90a67261e73b2fcac04d8ca963c6b496afb Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Thu, 14 May 2009 21:48:08 +1000 Subject: perf_counter: allow arch to supply event misc flags and instruction pointer At present the values we put in overflow events for the misc flags indicating processor mode and the instruction pointer are obtained using the standard user_mode() and instruction_pointer() functions. Those functions tell you where the performance monitor interrupt was taken, which might not be exactly where the counter overflow occurred, for example because interrupts were disabled at the point where the overflow occurred, or because the processor had many instructions in flight and chose to complete some more instructions beyond the one that caused the counter overflow. Some architectures (e.g. powerpc) can supply more precise information about where the counter overflow occurred and the processor mode at that point. This introduces new functions, perf_misc_flags() and perf_instruction_pointer(), which arch code can override to provide more precise information if available. They have default implementations which are identical to the existing code. This also adds a new misc flag value, PERF_EVENT_MISC_HYPERVISOR, for the case where a counter overflow occurred in the hypervisor. We encode the processor mode in the 2 bits previously used to indicate user or kernel mode; the values for user and kernel mode are unchanged and hypervisor mode is indicated by both bits being set. [ Impact: generalize perfcounter core facilities ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo LKML-Reference: <18956.1272.818511.561835@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 728a595399b0..57840a94b163 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2042,11 +2042,10 @@ static void perf_counter_output(struct perf_counter *counter, header.size = sizeof(header); header.misc = PERF_EVENT_MISC_OVERFLOW; - header.misc |= user_mode(regs) ? - PERF_EVENT_MISC_USER : PERF_EVENT_MISC_KERNEL; + header.misc |= perf_misc_flags(regs); if (record_type & PERF_RECORD_IP) { - ip = instruction_pointer(regs); + ip = perf_instruction_pointer(regs); header.type |= PERF_RECORD_IP; header.size += sizeof(ip); } -- cgit v1.2.3 From 88fc86c283d9c3854e67e4155808027bc2519eb6 Mon Sep 17 00:00:00 2001 From: GeunSik Lim Date: Thu, 14 May 2009 17:23:38 +0900 Subject: tracing: Append prompt in /debug/tracing/README file append prompt in /debug/tracing/README file. This is trivial issue. Fix typo Mini Howto file(README) for ftrace. [ Impact: cleanup ] Signed-off-by: GeunSik Lim Acked-by: Steven Rostedt Cc: williams LKML-Reference: <1242289418.31161.45.camel@centos51> Signed-off-by: Ingo Molnar --- kernel/trace/trace.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index a884c09006c4..cda81ec58d9f 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2380,7 +2380,7 @@ static const char readme_msg[] = "# echo print-parent > /debug/tracing/trace_options\n" "# echo 1 > /debug/tracing/tracing_enabled\n" "# cat /debug/tracing/trace > /tmp/trace.txt\n" - "echo 0 > /debug/tracing/tracing_enabled\n" + "# echo 0 > /debug/tracing/tracing_enabled\n" ; static ssize_t -- cgit v1.2.3 From 4484079d517c2b6521621be0b1ea246ccc55c7d7 Mon Sep 17 00:00:00 2001 From: Bjorn Helgaas Date: Fri, 15 May 2009 23:30:50 +0200 Subject: PM: check sysdev_suspend(PMSG_FREEZE) return value Check the return value of sysdev_suspend(). I think this was a typo. Without this change, the following "if" check is always false. I also changed the error message so it's distinguishable from the similar message a few lines above. Signed-off-by: Bjorn Helgaas Acked-by: Pavel Machek Signed-off-by: Rafael J. Wysocki --- kernel/power/disk.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/power/disk.c b/kernel/power/disk.c index e71ca9cd81b2..b0dc9e7a0d17 100644 --- a/kernel/power/disk.c +++ b/kernel/power/disk.c @@ -241,9 +241,9 @@ static int create_image(int platform_mode) local_irq_disable(); - sysdev_suspend(PMSG_FREEZE); + error = sysdev_suspend(PMSG_FREEZE); if (error) { - printk(KERN_ERR "PM: Some devices failed to power down, " + printk(KERN_ERR "PM: Some system devices failed to power down, " "aborting hibernation\n"); goto Enable_irqs; } -- cgit v1.2.3 From 0f6f49a8cd0163fdb1723ed29f01fc65177108dc Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Sat, 16 May 2009 13:41:28 -0700 Subject: Fix caller information for warn_slowpath_null Ian Campbell noticed that since "Eliminate thousands of warnings with gcc 3.2 build" (commit 57adc4d2dbf968fdbe516359688094eef4d46581) all WARN_ON()'s currently appear to come from warn_slowpath_null(), eg: WARNING: at kernel/softirq.c:143 warn_slowpath_null+0x1c/0x20() because now that warn_slowpath_null() is in the call path, the __builtin_return_address(0) returns that, rather than the place that caused the warning. Fix this by splitting up the warn_slowpath_null/fmt cases differently, using a common helper function, and getting the return address in the right place. This also happens to avoid the unnecessary stack usage for the non-stdargs case, and just generally cleans things up. Make the function name printout use %pS while at it. Cc: Ian Campbell Cc: Jesper Nilsson Cc: Johannes Weiner Cc: Arjan van de Ven Cc: Andi Kleen Cc: Hugh Dickins Cc: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/panic.c | 35 ++++++++++++++++++++--------------- 1 file changed, 20 insertions(+), 15 deletions(-) (limited to 'kernel') diff --git a/kernel/panic.c b/kernel/panic.c index 874ecf1307ae..984b3ecbd72c 100644 --- a/kernel/panic.c +++ b/kernel/panic.c @@ -340,39 +340,44 @@ void oops_exit(void) } #ifdef WANT_WARN_ON_SLOWPATH -void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...) -{ +struct slowpath_args { + const char *fmt; va_list args; - char function[KSYM_SYMBOL_LEN]; - unsigned long caller = (unsigned long)__builtin_return_address(0); - const char *board; +}; - sprint_symbol(function, caller); +static void warn_slowpath_common(const char *file, int line, void *caller, struct slowpath_args *args) +{ + const char *board; printk(KERN_WARNING "------------[ cut here ]------------\n"); - printk(KERN_WARNING "WARNING: at %s:%d %s()\n", file, - line, function); + printk(KERN_WARNING "WARNING: at %s:%d %pS()\n", file, line, caller); board = dmi_get_system_info(DMI_PRODUCT_NAME); if (board) printk(KERN_WARNING "Hardware name: %s\n", board); - if (*fmt) { - va_start(args, fmt); - vprintk(fmt, args); - va_end(args); - } + if (args) + vprintk(args->fmt, args->args); print_modules(); dump_stack(); print_oops_end_marker(); add_taint(TAINT_WARN); } + +void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...) +{ + struct slowpath_args args; + + args.fmt = fmt; + va_start(args.args, fmt); + warn_slowpath_common(file, line, __builtin_return_address(0), &args); + va_end(args.args); +} EXPORT_SYMBOL(warn_slowpath_fmt); void warn_slowpath_null(const char *file, int line) { - static const char *empty = ""; - warn_slowpath_fmt(file, line, empty); + warn_slowpath_common(file, line, __builtin_return_address(0), NULL); } EXPORT_SYMBOL(warn_slowpath_null); #endif -- cgit v1.2.3 From 8bc2095951517e2c74b8aeeca4685ddd6b16ed4b Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 20:45:59 +0200 Subject: perf_counter: Fix inheritance cleanup code Clean up code that open-coded the list_{add,del}_counter() code in __perf_counter_exit_task() which consequently diverged. This could lead to software counter crashes. Also, fold the ctx->nr_counter inc/dec into those functions and clean up some of the related code. [ Impact: fix potential sw counter crash, cleanup ] Signed-off-by: Peter Zijlstra Cc: Srivatsa Vaddagiri Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: Marcelo Tosatti Signed-off-by: Ingo Molnar Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 32 +++++++++++++++----------------- 1 file changed, 15 insertions(+), 17 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 57840a94b163..59a926d04baf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -115,6 +115,7 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) } list_add_rcu(&counter->event_entry, &ctx->event_list); + ctx->nr_counters++; } static void @@ -122,6 +123,8 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { struct perf_counter *sibling, *tmp; + ctx->nr_counters--; + list_del_init(&counter->list_entry); list_del_rcu(&counter->event_entry); @@ -209,7 +212,6 @@ static void __perf_counter_remove_from_context(void *info) counter_sched_out(counter, cpuctx, ctx); counter->task = NULL; - ctx->nr_counters--; /* * Protect the list operation against NMI by disabling the @@ -276,7 +278,6 @@ retry: * succeed. */ if (!list_empty(&counter->list_entry)) { - ctx->nr_counters--; list_del_counter(counter, ctx); counter->task = NULL; } @@ -544,7 +545,6 @@ static void add_counter_to_ctx(struct perf_counter *counter, struct perf_counter_context *ctx) { list_add_counter(counter, ctx); - ctx->nr_counters++; counter->prev_state = PERF_COUNTER_STATE_OFF; counter->tstamp_enabled = ctx->time; counter->tstamp_running = ctx->time; @@ -3206,9 +3206,8 @@ static int inherit_group(struct perf_counter *parent_counter, static void sync_child_counter(struct perf_counter *child_counter, struct perf_counter *parent_counter) { - u64 parent_val, child_val; + u64 child_val; - parent_val = atomic64_read(&parent_counter->count); child_val = atomic64_read(&child_counter->count); /* @@ -3240,7 +3239,6 @@ __perf_counter_exit_task(struct task_struct *child, struct perf_counter_context *child_ctx) { struct perf_counter *parent_counter; - struct perf_counter *sub, *tmp; /* * If we do not self-reap then we have to wait for the @@ -3252,8 +3250,8 @@ __perf_counter_exit_task(struct task_struct *child, */ if (child != current) { wait_task_inactive(child, 0); - list_del_init(&child_counter->list_entry); update_counter_times(child_counter); + list_del_counter(child_counter, child_ctx); } else { struct perf_cpu_context *cpuctx; unsigned long flags; @@ -3272,9 +3270,7 @@ __perf_counter_exit_task(struct task_struct *child, group_sched_out(child_counter, cpuctx, child_ctx); update_counter_times(child_counter); - list_del_init(&child_counter->list_entry); - - child_ctx->nr_counters--; + list_del_counter(child_counter, child_ctx); perf_enable(); local_irq_restore(flags); @@ -3288,13 +3284,6 @@ __perf_counter_exit_task(struct task_struct *child, */ if (parent_counter) { sync_child_counter(child_counter, parent_counter); - list_for_each_entry_safe(sub, tmp, &child_counter->sibling_list, - list_entry) { - if (sub->parent) { - sync_child_counter(sub, sub->parent); - free_counter(sub); - } - } free_counter(child_counter); } } @@ -3315,9 +3304,18 @@ void perf_counter_exit_task(struct task_struct *child) if (likely(!child_ctx->nr_counters)) return; +again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, list_entry) __perf_counter_exit_task(child, child_counter, child_ctx); + + /* + * If the last counter was a group counter, it will have appended all + * its siblings to the list, but we obtained 'tmp' before that which + * will still point to the list head terminating the iteration. + */ + if (!list_empty(&child_ctx->counter_list)) + goto again; } /* -- cgit v1.2.3 From 856d56b9e5de650a64a6c41c17aaed702b55d578 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 15 May 2009 20:45:59 +0200 Subject: perf_counter: Fix counter inheritance Srivatsa Vaddagiri reported that a Java workload triggers this warning in kernel/exit.c: WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list)); Add the inherited counter propagation on self-detach, this could cause counter leaks and incomplete stats in threaded code like the below: #include #include void *thread(void *arg) { sleep(5); return NULL; } void main(void) { pthread_t thr; pthread_create(&thr, NULL, thread, NULL); } Reported-by: Srivatsa Vaddagiri Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: Marcelo Tosatti Signed-off-by: Ingo Molnar --- kernel/exit.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index 4741376c8dec..16d74f13a3e7 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -128,6 +128,12 @@ static void __exit_signal(struct task_struct *tsk) sig = NULL; /* Marker for below. */ } + /* + * Flush inherited counters to the parent - before the parent + * gets woken up by child-exit notifications. + */ + perf_counter_exit_task(tsk); + __unhash_process(tsk); /* -- cgit v1.2.3 From 0203026b58b4299ba7281c0b4b417207c1f05d0e Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 17 May 2009 11:24:08 +0200 Subject: perf_counter: fix threaded task exit Flushing counters in __exit_signal() with irqs disabled is not a good idea as perf_counter_exit_task() acquires mutexes. So flush it before acquiring the tasklist lock. (Note, we still need a fix for when the PID has been unhashed.) [ Impact: fix crash with inherited counters ] Cc: Peter Zijlstra Cc: Srivatsa Vaddagiri Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: Marcelo Tosatti Signed-off-by: Ingo Molnar --- kernel/exit.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index 16d74f13a3e7..73affd35e76d 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -128,12 +128,6 @@ static void __exit_signal(struct task_struct *tsk) sig = NULL; /* Marker for below. */ } - /* - * Flush inherited counters to the parent - before the parent - * gets woken up by child-exit notifications. - */ - perf_counter_exit_task(tsk); - __unhash_process(tsk); /* @@ -183,6 +177,13 @@ repeat: atomic_dec(&__task_cred(p)->user->processes); proc_flush_task(p); + + /* + * Flush inherited counters to the parent - before the parent + * gets woken up by child-exit notifications. + */ + perf_counter_exit_task(p); + write_lock_irq(&tasklist_lock); tracehook_finish_release_task(p); __exit_signal(p); -- cgit v1.2.3 From 4200efd9acda4accf24640f1e77d24fdcdb524df Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 19 May 2009 09:22:19 +0200 Subject: sched: properly define the sched_group::cpumask and sched_domain::span fields Properly document the variable-size structure tricks we are doing wrt. struct sched_group and sched_domain, and use the field[0] GCC extension instead of defining a vla array. Dont use unions for this, as pointed out by Linus. [ Impact: cleanup, un-confuse Sparse and LLVM ] Reported-by: Jeff Garzik Acked-by: Linus Torvalds LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/sched.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 497c09ba61e7..228acae8821f 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -7948,8 +7948,9 @@ int sched_smt_power_savings = 0, sched_mc_power_savings = 0; /* * The cpus mask in sched_group and sched_domain hangs off the end. - * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space - * for nr_cpu_ids < CONFIG_NR_CPUS. + * + * ( See the the comments in include/linux/sched.h:struct sched_group + * and struct sched_domain. ) */ struct static_sched_group { struct sched_group sg; -- cgit v1.2.3 From 64d1304a64477629cb16b75491a77bafe6f86963 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 18 May 2009 21:20:10 +0200 Subject: futex: setup writeable mapping for futex ops which modify user space data The futex code installs a read only mapping via get_user_pages_fast() even if the futex op function has to modify user space data. The eventual fault was fixed up by futex_handle_fault() which walked the VMA with mmap_sem held. After the cleanup patches which removed the mmap_sem dependency of the futex code commit 4dc5b7a36a49eff97050894cf1b3a9a02523717 (futex: clean up fault logic) removed the private VMA walk logic from the futex code. This change results in a stale RO mapping which is not fixed up. Instead of reintroducing the previous fault logic we set up the mapping in get_user_pages_fast() read/write for all operations which modify user space data. Also handle private futexes in the same way and make the current unconditional access_ok(VERIFY_WRITE) depend on the futex op. Reported-by: Andreas Schwab Signed-off-by: Thomas Gleixner CC: stable@kernel.org --- kernel/futex.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index eef8cd26b5e5..d546b2d53a62 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -193,6 +193,7 @@ static void drop_futex_key_refs(union futex_key *key) * @uaddr: virtual address of the futex * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED * @key: address where result is stored. + * @rw: mapping needs to be read/write (values: VERIFY_READ, VERIFY_WRITE) * * Returns a negative error code or 0 * The key words are stored in *key on success. @@ -203,7 +204,8 @@ static void drop_futex_key_refs(union futex_key *key) * * lock_page() might sleep, the caller should not hold a spinlock. */ -static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) +static int +get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key, int rw) { unsigned long address = (unsigned long)uaddr; struct mm_struct *mm = current->mm; @@ -226,7 +228,7 @@ static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) * but access_ok() should be faster than find_vma() */ if (!fshared) { - if (unlikely(!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))) + if (unlikely(!access_ok(rw, uaddr, sizeof(u32)))) return -EFAULT; key->private.mm = mm; key->private.address = address; @@ -235,7 +237,7 @@ static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) } again: - err = get_user_pages_fast(address, 1, 0, &page); + err = get_user_pages_fast(address, 1, rw == VERIFY_WRITE, &page); if (err < 0) return err; @@ -677,7 +679,7 @@ static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) if (!bitset) return -EINVAL; - ret = get_futex_key(uaddr, fshared, &key); + ret = get_futex_key(uaddr, fshared, &key, VERIFY_READ); if (unlikely(ret != 0)) goto out; @@ -723,10 +725,10 @@ futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, int ret, op_ret; retry: - ret = get_futex_key(uaddr1, fshared, &key1); + ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); if (unlikely(ret != 0)) goto out; - ret = get_futex_key(uaddr2, fshared, &key2); + ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_WRITE); if (unlikely(ret != 0)) goto out_put_key1; @@ -814,10 +816,10 @@ static int futex_requeue(u32 __user *uaddr1, int fshared, u32 __user *uaddr2, int ret, drop_count = 0; retry: - ret = get_futex_key(uaddr1, fshared, &key1); + ret = get_futex_key(uaddr1, fshared, &key1, VERIFY_READ); if (unlikely(ret != 0)) goto out; - ret = get_futex_key(uaddr2, fshared, &key2); + ret = get_futex_key(uaddr2, fshared, &key2, VERIFY_READ); if (unlikely(ret != 0)) goto out_put_key1; @@ -1140,7 +1142,7 @@ static int futex_wait(u32 __user *uaddr, int fshared, q.bitset = bitset; retry: q.key = FUTEX_KEY_INIT; - ret = get_futex_key(uaddr, fshared, &q.key); + ret = get_futex_key(uaddr, fshared, &q.key, VERIFY_READ); if (unlikely(ret != 0)) goto out; @@ -1330,7 +1332,7 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared, q.pi_state = NULL; retry: q.key = FUTEX_KEY_INIT; - ret = get_futex_key(uaddr, fshared, &q.key); + ret = get_futex_key(uaddr, fshared, &q.key, VERIFY_WRITE); if (unlikely(ret != 0)) goto out; @@ -1594,7 +1596,7 @@ retry: if ((uval & FUTEX_TID_MASK) != task_pid_vnr(current)) return -EPERM; - ret = get_futex_key(uaddr, fshared, &key); + ret = get_futex_key(uaddr, fshared, &key, VERIFY_WRITE); if (unlikely(ret != 0)) goto out; -- cgit v1.2.3 From 33b2fb303fe7f6b08bbb32f708e67b96eaa94a7a Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 17 May 2009 11:08:41 +0200 Subject: perf_counter: fix counter freeing logic Fix counter lifetime bugs which explain the crashes reported by Marcelo Tosatti and Arnaldo Carvalho de Melo. The new rule is: flushing + freeing is only done for a task's own counters, never for other tasks. [ Impact: fix crashes/lockups with inherited counters ] Reported-by: Arnaldo Carvalho de Melo Reported-by: Marcelo Tosatti Acked-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Signed-off-by: Ingo Molnar --- kernel/exit.c | 19 +++++++------------ kernel/perf_counter.c | 2 ++ 2 files changed, 9 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index 73affd35e76d..f9dfedd94af0 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -178,12 +178,6 @@ repeat: proc_flush_task(p); - /* - * Flush inherited counters to the parent - before the parent - * gets woken up by child-exit notifications. - */ - perf_counter_exit_task(p); - write_lock_irq(&tasklist_lock); tracehook_finish_release_task(p); __exit_signal(p); @@ -985,6 +979,13 @@ NORET_TYPE void do_exit(long code) module_put(tsk->binfmt->module); proc_exit_connector(tsk); + + /* + * Flush inherited counters to the parent - before the parent + * gets woken up by child-exit notifications. + */ + perf_counter_exit_task(tsk); + exit_notify(tsk, group_dead); #ifdef CONFIG_NUMA mpol_put(tsk->mempolicy); @@ -1257,12 +1258,6 @@ static int wait_task_zombie(struct task_struct *p, int options, */ read_unlock(&tasklist_lock); - /* - * Flush inherited counters to the parent - before the parent - * gets woken up by child-exit notifications. - */ - perf_counter_exit_task(p); - retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0; status = (p->signal->flags & SIGNAL_GROUP_EXIT) ? p->signal->group_exit_code : p->exit_code; diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 59a926d04baf..7af16d1c480f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3299,6 +3299,8 @@ void perf_counter_exit_task(struct task_struct *child) struct perf_counter *child_counter, *tmp; struct perf_counter_context *child_ctx; + WARN_ON_ONCE(child != current); + child_ctx = &child->perf_counter_ctx; if (likely(!child_ctx->nr_counters)) -- cgit v1.2.3 From c44d70a340554a33071339064a303ac0f1a31623 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 17 May 2009 11:24:08 +0200 Subject: perf_counter: fix counter inheritance race Context rotation should not occur when we are in the middle of walking the counter list when inheriting counters ... [ Impact: fix occasionally incorrect perf stat results ] Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: Marcelo Tosatti Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7af16d1c480f..4d8f97375f3a 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1120,7 +1120,8 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) __perf_counter_task_sched_out(ctx); rotate_ctx(&cpuctx->ctx); - rotate_ctx(ctx); + if (ctx->rr_allowed) + rotate_ctx(ctx); perf_counter_cpu_sched_in(cpuctx, cpu); perf_counter_task_sched_in(curr, cpu); @@ -3108,6 +3109,7 @@ __perf_counter_init_context(struct perf_counter_context *ctx, mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->counter_list); INIT_LIST_HEAD(&ctx->event_list); + ctx->rr_allowed = 1; ctx->task = task; } @@ -3348,6 +3350,9 @@ void perf_counter_init_task(struct task_struct *child) */ mutex_lock(&parent_ctx->mutex); + parent_ctx->rr_allowed = 0; + barrier(); /* irqs */ + /* * We dont have to disable NMIs - we are only looking at * the list, not manipulating it: @@ -3361,6 +3366,9 @@ void perf_counter_init_task(struct task_struct *child) break; } + barrier(); /* irqs */ + parent_ctx->rr_allowed = 1; + mutex_unlock(&parent_ctx->mutex); } -- cgit v1.2.3 From c8b15a706d921baed3195407e4f55270112bb3c6 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Wed, 20 May 2009 09:18:50 +0200 Subject: futex: cleanup error exit Reuse the put_key_ref(key2) call in the exit path. Signed-off-by: Thomas Gleixner --- kernel/futex.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 476603afd147..381125a9f1e0 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -2185,10 +2185,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, /* Prepare to wait on uaddr. */ ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); - if (ret) { - put_futex_key(fshared, &key2); - goto out; - } + if (ret) + goto out_key2; /* Queue the futex_q, drop the hb lock, wait for wakeup. */ futex_wait_queue_me(hb, &q, to); @@ -2282,6 +2280,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, out_put_keys: put_futex_key(fshared, &q.key); +out_key2: put_futex_key(fshared, &key2); out: -- cgit v1.2.3 From 1c840c14906d4ddf66c1f4f5daea059aad951c82 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Wed, 20 May 2009 09:22:40 +0200 Subject: futex: fix restart for early wakeup in futex_wait_requeue_pi() The futex_wait_requeue_pi op should restart unconditionally like futex_lock_pi. The user of that function e.g. pthread_cond_wait can not be interrupted so we do not care about the SA_RESTART flag of the signal. Clean up the FIXMEs. Signed-off-by: Thomas Gleixner --- kernel/futex.c | 13 +++---------- 1 file changed, 3 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 381125a9f1e0..2aa216e5b594 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -2060,7 +2060,7 @@ pi_faulted: * * Returns * 0 - no early wakeup detected - * <0 - -ETIMEDOUT or -ERESTARTSYS (FIXME: or ERESTARTNOINTR?) + * <0 - -ETIMEDOUT or -ERESTARTNOINTR */ static inline int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, @@ -2087,15 +2087,8 @@ int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, if (timeout && !timeout->task) ret = -ETIMEDOUT; - else { - /* - * We expect signal_pending(current), but another - * thread may have handled it for us already. - */ - /* FIXME: ERESTARTSYS or ERESTARTNOINTR? Do we care if - * the user specified SA_RESTART or not? */ - ret = -ERESTARTSYS; - } + else + ret = -ERESTARTNOINTR; } return ret; } -- cgit v1.2.3 From 2070887fdeacd9c13f3e805e3f0086c9f22a4d93 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Tue, 19 May 2009 23:04:59 +0200 Subject: futex: fix restart in wait_requeue_pi If the waiter has been requeued to the outer PI futex and is interrupted by a signal and the thread handles the signal then ERESTART_RESTARTBLOCK is changed to EINTR and the restart block is discarded. That way we return an unexcpected EINTR to user space instead of ending up in futex_lock_pi_restart. But we do not need to restart the syscall because we know that the condition has changed since we have been requeued. If we would simply restart the syscall then we would drop out via the comparison of the user space value with EWOULDBLOCK. The user space side needs to handle EWOULDBLOCK anyway as the enqueueing on the inner futex can race with a requeue/wake. So we can simply return EWOULDBLOCK to user space which also signals that we did not take the outer futex and let user space handle it in the same way it has to handle the requeue/wake race. Signed-off-by: Thomas Gleixner --- kernel/futex.c | 49 +++++++++---------------------------------------- 1 file changed, 9 insertions(+), 40 deletions(-) (limited to 'kernel') diff --git a/kernel/futex.c b/kernel/futex.c index 2aa216e5b594..80b5ce716596 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1507,7 +1507,6 @@ handle_fault: #define FLAGS_HAS_TIMEOUT 0x04 static long futex_wait_restart(struct restart_block *restart); -static long futex_lock_pi_restart(struct restart_block *restart); /** * fixup_owner() - Post lock pi_state and corner case management @@ -1930,21 +1929,6 @@ uaddr_faulted: goto retry; } -static long futex_lock_pi_restart(struct restart_block *restart) -{ - u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; - ktime_t t, *tp = NULL; - int fshared = restart->futex.flags & FLAGS_SHARED; - - if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { - t.tv64 = restart->futex.time; - tp = &t; - } - restart->fn = do_no_restart_syscall; - - return (long)futex_lock_pi(uaddr, fshared, restart->futex.val, tp, 0); -} - /* * Userspace attempted a TID -> 0 atomic transition, and failed. * This is the in-kernel slowpath: we look up the PI state (if any), @@ -2141,12 +2125,10 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, struct hrtimer_sleeper timeout, *to = NULL; struct rt_mutex_waiter rt_waiter; struct rt_mutex *pi_mutex = NULL; - struct restart_block *restart; struct futex_hash_bucket *hb; union futex_key key2; struct futex_q q; int res, ret; - u32 uval; if (!bitset) return -EINVAL; @@ -2245,30 +2227,17 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared, if (rt_mutex_owner(pi_mutex) == current) rt_mutex_unlock(pi_mutex); } else if (ret == -EINTR) { - ret = -EFAULT; - if (get_user(uval, uaddr2)) - goto out_put_keys; - /* - * We've already been requeued, so restart by calling - * futex_lock_pi() directly, rather then returning to this - * function. + * We've already been requeued, but we have no way to + * restart by calling futex_lock_pi() directly. We + * could restart the syscall, but that will look at + * the user space value and return right away. So we + * drop back with EWOULDBLOCK to tell user space that + * "val" has been changed. That's the same what the + * restart of the syscall would do in + * futex_wait_setup(). */ - ret = -ERESTART_RESTARTBLOCK; - restart = ¤t_thread_info()->restart_block; - restart->fn = futex_lock_pi_restart; - restart->futex.uaddr = (u32 *)uaddr2; - restart->futex.val = uval; - restart->futex.flags = 0; - if (abs_time) { - restart->futex.flags |= FLAGS_HAS_TIMEOUT; - restart->futex.time = abs_time->tv64; - } - - if (fshared) - restart->futex.flags |= FLAGS_SHARED; - if (clockrt) - restart->futex.flags |= FLAGS_CLOCKRT; + ret = -EWOULDBLOCK; } out_put_keys: -- cgit v1.2.3 From d7b629a34fc4134a43c730b5f0197855dc4948d0 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 20 May 2009 12:21:19 +0200 Subject: perf_counter: Solve the rotate_ctx vs inherit race differently Instead of disabling RR scheduling of the counters, use a different list that does not get rotated to iterate the counters on inheritance. [ Impact: cleanup, optimization ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: Marcelo Tosatti Cc: John Kacur LKML-Reference: <20090520102553.237504544@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 15 +++++---------- 1 file changed, 5 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4d8f97375f3a..64113e6d1942 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1120,8 +1120,7 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) __perf_counter_task_sched_out(ctx); rotate_ctx(&cpuctx->ctx); - if (ctx->rr_allowed) - rotate_ctx(ctx); + rotate_ctx(ctx); perf_counter_cpu_sched_in(cpuctx, cpu); perf_counter_task_sched_in(curr, cpu); @@ -3109,7 +3108,6 @@ __perf_counter_init_context(struct perf_counter_context *ctx, mutex_init(&ctx->mutex); INIT_LIST_HEAD(&ctx->counter_list); INIT_LIST_HEAD(&ctx->event_list); - ctx->rr_allowed = 1; ctx->task = task; } @@ -3350,14 +3348,14 @@ void perf_counter_init_task(struct task_struct *child) */ mutex_lock(&parent_ctx->mutex); - parent_ctx->rr_allowed = 0; - barrier(); /* irqs */ - /* * We dont have to disable NMIs - we are only looking at * the list, not manipulating it: */ - list_for_each_entry(counter, &parent_ctx->counter_list, list_entry) { + list_for_each_entry_rcu(counter, &parent_ctx->event_list, event_entry) { + if (counter != counter->group_leader) + continue; + if (!counter->hw_event.inherit) continue; @@ -3366,9 +3364,6 @@ void perf_counter_init_task(struct task_struct *child) break; } - barrier(); /* irqs */ - parent_ctx->rr_allowed = 1; - mutex_unlock(&parent_ctx->mutex); } -- cgit v1.2.3 From 26b119bc811a73bac6ecf95bdf284bf31c7955f0 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 20 May 2009 12:21:20 +0200 Subject: perf_counter: Log irq_period changes For the dynamic irq_period code, log whenever we change the period so that analyzing code can normalize the event flow. [ Impact: add new feature to allow more precise profiling ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: Marcelo Tosatti Cc: John Kacur LKML-Reference: <20090520102553.298769743@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 40 +++++++++++++++++++++++++++++++++++++++- 1 file changed, 39 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 64113e6d1942..db02eb16c777 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1046,7 +1046,9 @@ int perf_counter_task_enable(void) return 0; } -void perf_adjust_freq(struct perf_counter_context *ctx) +static void perf_log_period(struct perf_counter *counter, u64 period); + +static void perf_adjust_freq(struct perf_counter_context *ctx) { struct perf_counter *counter; u64 irq_period; @@ -1072,6 +1074,8 @@ void perf_adjust_freq(struct perf_counter_context *ctx) if (!irq_period) irq_period = 1; + perf_log_period(counter, irq_period); + counter->hw.irq_period = irq_period; counter->hw.interrupts = 0; } @@ -2406,6 +2410,40 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, perf_counter_mmap_event(&mmap_event); } +/* + * + */ + +static void perf_log_period(struct perf_counter *counter, u64 period) +{ + struct perf_output_handle handle; + int ret; + + struct { + struct perf_event_header header; + u64 time; + u64 period; + } freq_event = { + .header = { + .type = PERF_EVENT_PERIOD, + .misc = 0, + .size = sizeof(freq_event), + }, + .time = sched_clock(), + .period = period, + }; + + if (counter->hw.irq_period == period) + return; + + ret = perf_output_begin(&handle, counter, sizeof(freq_event), 0, 0); + if (ret) + return; + + perf_output_put(&handle, freq_event); + perf_output_end(&handle); +} + /* * Generic counter overflow handling. */ -- cgit v1.2.3 From b986d7ec0f8b7ea3cc7366d80a137fbe839df227 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 20 May 2009 12:21:21 +0200 Subject: perf_counter: Optimize disable of time based sw counters Currently we call hrtimer_cancel() unconditionally on disable of time based software counters. Avoid when possible. [ Impact: micro-optimize the code ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: Marcelo Tosatti Cc: John Kacur LKML-Reference: <20090520102553.388185031@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index db02eb16c777..473ed2cafbfc 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2716,7 +2716,8 @@ static int cpu_clock_perf_counter_enable(struct perf_counter *counter) static void cpu_clock_perf_counter_disable(struct perf_counter *counter) { - hrtimer_cancel(&counter->hw.hrtimer); + if (counter->hw.irq_period) + hrtimer_cancel(&counter->hw.hrtimer); cpu_clock_perf_counter_update(counter); } @@ -2767,7 +2768,8 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) static void task_clock_perf_counter_disable(struct perf_counter *counter) { - hrtimer_cancel(&counter->hw.hrtimer); + if (counter->hw.irq_period) + hrtimer_cancel(&counter->hw.hrtimer); task_clock_perf_counter_update(counter, counter->ctx->time); } -- cgit v1.2.3 From afedadf23a2c90f3ba0d963282cbe6a6be129494 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 20 May 2009 12:21:22 +0200 Subject: perf_counter: Optimize sched in/out of counters Avoid a function call for !group counters by directly calling the counter function. [ Impact: micro-optimize the code ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: Marcelo Tosatti Cc: John Kacur LKML-Reference: <20090520102553.511933670@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 25 +++++++++++++++++++------ 1 file changed, 19 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 473ed2cafbfc..69d4de815963 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -826,8 +826,12 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, perf_disable(); if (ctx->nr_active) { - list_for_each_entry(counter, &ctx->counter_list, list_entry) - group_sched_out(counter, cpuctx, ctx); + list_for_each_entry(counter, &ctx->counter_list, list_entry) { + if (counter != counter->group_leader) + counter_sched_out(counter, cpuctx, ctx); + else + group_sched_out(counter, cpuctx, ctx); + } } perf_enable(); out: @@ -903,8 +907,12 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, if (counter->cpu != -1 && counter->cpu != cpu) continue; - if (group_can_go_on(counter, cpuctx, 1)) - group_sched_in(counter, cpuctx, ctx, cpu); + if (counter != counter->group_leader) + counter_sched_in(counter, cpuctx, ctx, cpu); + else { + if (group_can_go_on(counter, cpuctx, 1)) + group_sched_in(counter, cpuctx, ctx, cpu); + } /* * If this pinned group hasn't been scheduled, @@ -932,9 +940,14 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, if (counter->cpu != -1 && counter->cpu != cpu) continue; - if (group_can_go_on(counter, cpuctx, can_add_hw)) { - if (group_sched_in(counter, cpuctx, ctx, cpu)) + if (counter != counter->group_leader) { + if (counter_sched_in(counter, cpuctx, ctx, cpu)) can_add_hw = 0; + } else { + if (group_can_go_on(counter, cpuctx, can_add_hw)) { + if (group_sched_in(counter, cpuctx, ctx, cpu)) + can_add_hw = 0; + } } } perf_enable(); -- cgit v1.2.3 From 34adc8062227f41b04ade0ff3fbd1dbe3002669e Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 20 May 2009 20:13:28 +0200 Subject: perf_counter: Fix context removal deadlock Disable the PMU globally before removing a counter from a context. This fixes the following lockup: [22081.741922] ------------[ cut here ]------------ [22081.746668] WARNING: at arch/x86/kernel/cpu/perf_counter.c:803 intel_pmu_handle_irq+0x9b/0x24e() [22081.755624] Hardware name: X8DTN [22081.758903] perfcounters: irq loop stuck! [22081.762985] Modules linked in: [22081.766136] Pid: 11082, comm: perf Not tainted 2.6.30-rc6-tip #226 [22081.772432] Call Trace: [22081.774940] [] ? intel_pmu_handle_irq+0x9b/0x24e [22081.781993] [] ? intel_pmu_handle_irq+0x9b/0x24e [22081.788368] [] ? warn_slowpath_common+0x77/0xa3 [22081.794649] [] ? warn_slowpath_fmt+0x40/0x45 [22081.800696] [] ? intel_pmu_handle_irq+0x9b/0x24e [22081.807080] [] ? perf_counter_nmi_handler+0x3f/0x4a [22081.813751] [] ? notifier_call_chain+0x58/0x86 [22081.819951] [] ? notify_die+0x2d/0x32 [22081.825392] [] ? do_nmi+0x8e/0x242 [22081.830538] [] ? nmi+0x1a/0x20 [22081.835342] [] ? selinux_file_free_security+0x0/0x1a [22081.842105] [] ? x86_pmu_disable_counter+0x15/0x41 [22081.848673] <> [] ? x86_pmu_disable+0x86/0x103 [22081.855512] [] ? __perf_counter_remove_from_context+0x0/0xfe [22081.862926] [] ? counter_sched_out+0x30/0xce [22081.868909] [] ? __perf_counter_remove_from_context+0x59/0xfe [22081.876382] [] ? smp_call_function_single+0x6c/0xe6 [22081.882955] [] ? perf_release+0x86/0x14c [22081.888600] [] ? __fput+0xe7/0x195 [22081.893718] [] ? filp_close+0x5b/0x62 [22081.899107] [] ? put_files_struct+0x64/0xc2 [22081.905031] [] ? do_exit+0x1e2/0x6ef [22081.910360] [] ? _spin_lock_irqsave+0x9/0xe [22081.916292] [] ? do_group_exit+0x67/0x93 [22081.921953] [] ? sys_exit_group+0x12/0x16 [22081.927759] [] ? system_call_fastpath+0x16/0x1b [22081.934076] ---[ end trace 3a3936ce3e1b4505 ]--- And could potentially also fix the lockup reported by Marcelo Tosatti. Also, print more debug info in case of a detected lockup. [ Impact: fix lockup ] Reported-by: Marcelo Tosatti Acked-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Thomas Gleixner LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 69d4de815963..08584c16049f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -208,18 +208,17 @@ static void __perf_counter_remove_from_context(void *info) return; spin_lock_irqsave(&ctx->lock, flags); + /* + * Protect the list operation against NMI by disabling the + * counters on a global level. + */ + perf_disable(); counter_sched_out(counter, cpuctx, ctx); counter->task = NULL; - /* - * Protect the list operation against NMI by disabling the - * counters on a global level. NOP for non NMI based counters. - */ - perf_disable(); list_del_counter(counter, ctx); - perf_enable(); if (!ctx->task) { /* @@ -231,6 +230,7 @@ static void __perf_counter_remove_from_context(void *info) perf_max_counters - perf_reserved_percpu); } + perf_enable(); spin_unlock_irqrestore(&ctx->lock, flags); } -- cgit v1.2.3 From a63eaf34ae60bdb067a354cc8def2e8f4a01f5f4 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 22 May 2009 14:17:31 +1000 Subject: perf_counter: Dynamically allocate tasks' perf_counter_context struct This replaces the struct perf_counter_context in the task_struct with a pointer to a dynamically allocated perf_counter_context struct. The main reason for doing is this is to allow us to transfer a perf_counter_context from one task to another when we do lazy PMU switching in a later patch. This has a few side-benefits: the task_struct becomes a little smaller, we save some memory because only tasks that have perf_counters attached get a perf_counter_context allocated for them, and we can remove the inclusion of in sched.h, meaning that we don't end up recompiling nearly everything whenever perf_counter.h changes. The perf_counter_context structures are reference-counted and freed when the last reference is dropped. A context can have references from its task and the counters on its task. Counters can outlive the task so it is possible that a context will be freed well after its task has exited. Contexts are allocated on fork if the parent had a context, or otherwise the first time that a per-task counter is created on a task. In the latter case, we set the context pointer in the task struct locklessly using an atomic compare-and-exchange operation in case we raced with some other task in creating a context for the subject task. This also removes the task pointer from the perf_counter struct. The task pointer was not used anywhere and would make it harder to move a context from one task to another. Anything that needed to know which task a counter was attached to was already using counter->ctx->task. The __perf_counter_init_context function moves up in perf_counter.c so that it can be called from find_get_context, and now initializes the refcount, but is otherwise unchanged. We were potentially calling list_del_counter twice: once from __perf_counter_exit_task when the task exits and once from __perf_counter_remove_from_context when the counter's fd gets closed. This adds a check in list_del_counter so it doesn't do anything if the counter has already been removed from the lists. Since perf_counter_task_sched_in doesn't do anything if the task doesn't have a context, and leaves cpuctx->task_ctx = NULL, this adds code to __perf_install_in_context to set cpuctx->task_ctx if necessary, i.e. in the case where the current task adds the first counter to itself and thus creates a context for itself. This also adds similar code to __perf_counter_enable to handle a similar situation which can arise when the counters have been disabled using prctl; that also leaves cpuctx->task_ctx = NULL. [ Impact: refactor counter context management to prepare for new feature ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo LKML-Reference: <18966.10075.781053.231153@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/exit.c | 3 +- kernel/fork.c | 1 + kernel/perf_counter.c | 218 +++++++++++++++++++++++++++++++------------------- 3 files changed, 139 insertions(+), 83 deletions(-) (limited to 'kernel') diff --git a/kernel/exit.c b/kernel/exit.c index f9dfedd94af0..99ad4063ee4a 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -48,6 +48,7 @@ #include #include #include +#include #include #include @@ -159,7 +160,7 @@ static void delayed_put_task_struct(struct rcu_head *rhp) struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); #ifdef CONFIG_PERF_COUNTERS - WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list)); + WARN_ON_ONCE(tsk->perf_counter_ctxp); #endif trace_sched_process_free(tsk); put_task_struct(tsk); diff --git a/kernel/fork.c b/kernel/fork.c index d32fef4d38e5..e72a09f5355b 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -63,6 +63,7 @@ #include #include #include +#include #include #include diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 08584c16049f..06ea3eae886e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -97,6 +97,17 @@ void perf_enable(void) hw_perf_enable(); } +static void get_ctx(struct perf_counter_context *ctx) +{ + atomic_inc(&ctx->refcount); +} + +static void put_ctx(struct perf_counter_context *ctx) +{ + if (atomic_dec_and_test(&ctx->refcount)) + kfree(ctx); +} + static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { @@ -118,11 +129,17 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) ctx->nr_counters++; } +/* + * Remove a counter from the lists for its context. + * Must be called with counter->mutex and ctx->mutex held. + */ static void list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { struct perf_counter *sibling, *tmp; + if (list_empty(&counter->list_entry)) + return; ctx->nr_counters--; list_del_init(&counter->list_entry); @@ -216,8 +233,6 @@ static void __perf_counter_remove_from_context(void *info) counter_sched_out(counter, cpuctx, ctx); - counter->task = NULL; - list_del_counter(counter, ctx); if (!ctx->task) { @@ -279,7 +294,6 @@ retry: */ if (!list_empty(&counter->list_entry)) { list_del_counter(counter, ctx); - counter->task = NULL; } spin_unlock_irq(&ctx->lock); } @@ -568,11 +582,17 @@ static void __perf_install_in_context(void *info) * If this is a task context, we need to check whether it is * the current task context of this cpu. If not it has been * scheduled out before the smp call arrived. + * Or possibly this is the right context but it isn't + * on this cpu because it had no counters. */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; + if (ctx->task && cpuctx->task_ctx != ctx) { + if (cpuctx->task_ctx || ctx->task != current) + return; + cpuctx->task_ctx = ctx; + } spin_lock_irqsave(&ctx->lock, flags); + ctx->is_active = 1; update_context_time(ctx); /* @@ -653,7 +673,6 @@ perf_install_in_context(struct perf_counter_context *ctx, return; } - counter->task = task; retry: task_oncpu_function_call(task, __perf_install_in_context, counter); @@ -693,10 +712,14 @@ static void __perf_counter_enable(void *info) * If this is a per-task counter, need to check whether this * counter's task is the current task on this cpu. */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; + if (ctx->task && cpuctx->task_ctx != ctx) { + if (cpuctx->task_ctx || ctx->task != current) + return; + cpuctx->task_ctx = ctx; + } spin_lock_irqsave(&ctx->lock, flags); + ctx->is_active = 1; update_context_time(ctx); counter->prev_state = counter->state; @@ -852,10 +875,10 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, void perf_counter_task_sched_out(struct task_struct *task, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct perf_counter_context *ctx = task->perf_counter_ctxp; struct pt_regs *regs; - if (likely(!cpuctx->task_ctx)) + if (likely(!ctx || !cpuctx->task_ctx)) return; update_context_time(ctx); @@ -871,6 +894,8 @@ static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + if (!cpuctx->task_ctx) + return; __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; } @@ -969,8 +994,10 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, void perf_counter_task_sched_in(struct task_struct *task, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &task->perf_counter_ctx; + struct perf_counter_context *ctx = task->perf_counter_ctxp; + if (likely(!ctx)) + return; __perf_counter_sched_in(ctx, cpuctx, cpu); cpuctx->task_ctx = ctx; } @@ -985,11 +1012,11 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) int perf_counter_task_disable(void) { struct task_struct *curr = current; - struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; unsigned long flags; - if (likely(!ctx->nr_counters)) + if (!ctx || !ctx->nr_counters) return 0; local_irq_save(flags); @@ -1020,12 +1047,12 @@ int perf_counter_task_disable(void) int perf_counter_task_enable(void) { struct task_struct *curr = current; - struct perf_counter_context *ctx = &curr->perf_counter_ctx; + struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; unsigned long flags; int cpu; - if (likely(!ctx->nr_counters)) + if (!ctx || !ctx->nr_counters) return 0; local_irq_save(flags); @@ -1128,19 +1155,23 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) return; cpuctx = &per_cpu(perf_cpu_context, cpu); - ctx = &curr->perf_counter_ctx; + ctx = curr->perf_counter_ctxp; perf_adjust_freq(&cpuctx->ctx); - perf_adjust_freq(ctx); + if (ctx) + perf_adjust_freq(ctx); perf_counter_cpu_sched_out(cpuctx); - __perf_counter_task_sched_out(ctx); + if (ctx) + __perf_counter_task_sched_out(ctx); rotate_ctx(&cpuctx->ctx); - rotate_ctx(ctx); + if (ctx) + rotate_ctx(ctx); perf_counter_cpu_sched_in(cpuctx, cpu); - perf_counter_task_sched_in(curr, cpu); + if (ctx) + perf_counter_task_sched_in(curr, cpu); } /* @@ -1176,6 +1207,22 @@ static u64 perf_counter_read(struct perf_counter *counter) return atomic64_read(&counter->count); } +/* + * Initialize the perf_counter context in a task_struct: + */ +static void +__perf_counter_init_context(struct perf_counter_context *ctx, + struct task_struct *task) +{ + memset(ctx, 0, sizeof(*ctx)); + spin_lock_init(&ctx->lock); + mutex_init(&ctx->mutex); + INIT_LIST_HEAD(&ctx->counter_list); + INIT_LIST_HEAD(&ctx->event_list); + atomic_set(&ctx->refcount, 1); + ctx->task = task; +} + static void put_context(struct perf_counter_context *ctx) { if (ctx->task) @@ -1186,6 +1233,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) { struct perf_cpu_context *cpuctx; struct perf_counter_context *ctx; + struct perf_counter_context *tctx; struct task_struct *task; /* @@ -1225,15 +1273,36 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) if (!task) return ERR_PTR(-ESRCH); - ctx = &task->perf_counter_ctx; - ctx->task = task; - /* Reuse ptrace permission checks for now. */ if (!ptrace_may_access(task, PTRACE_MODE_READ)) { - put_context(ctx); + put_task_struct(task); return ERR_PTR(-EACCES); } + ctx = task->perf_counter_ctxp; + if (!ctx) { + ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); + if (!ctx) { + put_task_struct(task); + return ERR_PTR(-ENOMEM); + } + __perf_counter_init_context(ctx, task); + /* + * Make sure other cpus see correct values for *ctx + * once task->perf_counter_ctxp is visible to them. + */ + smp_wmb(); + tctx = cmpxchg(&task->perf_counter_ctxp, NULL, ctx); + if (tctx) { + /* + * We raced with some other task; use + * the context they set. + */ + kfree(ctx); + ctx = tctx; + } + } + return ctx; } @@ -1242,6 +1311,7 @@ static void free_counter_rcu(struct rcu_head *head) struct perf_counter *counter; counter = container_of(head, struct perf_counter, rcu_head); + put_ctx(counter->ctx); kfree(counter); } @@ -2247,7 +2317,7 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) perf_counter_comm_ctx(&cpuctx->ctx, comm_event); put_cpu_var(perf_cpu_context); - perf_counter_comm_ctx(¤t->perf_counter_ctx, comm_event); + perf_counter_comm_ctx(current->perf_counter_ctxp, comm_event); } void perf_counter_comm(struct task_struct *task) @@ -2256,7 +2326,9 @@ void perf_counter_comm(struct task_struct *task) if (!atomic_read(&nr_comm_tracking)) return; - + if (!current->perf_counter_ctxp) + return; + comm_event = (struct perf_comm_event){ .task = task, .event = { @@ -2372,7 +2444,7 @@ got_name: perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event); put_cpu_var(perf_cpu_context); - perf_counter_mmap_ctx(¤t->perf_counter_ctx, mmap_event); + perf_counter_mmap_ctx(current->perf_counter_ctxp, mmap_event); kfree(buf); } @@ -2384,6 +2456,8 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, if (!atomic_read(&nr_mmap_tracking)) return; + if (!current->perf_counter_ctxp) + return; mmap_event = (struct perf_mmap_event){ .file = file, @@ -2985,6 +3059,7 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->group_leader = group_leader; counter->pmu = NULL; counter->ctx = ctx; + get_ctx(ctx); counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) @@ -3149,21 +3224,6 @@ err_put_context: goto out_fput; } -/* - * Initialize the perf_counter context in a task_struct: - */ -static void -__perf_counter_init_context(struct perf_counter_context *ctx, - struct task_struct *task) -{ - memset(ctx, 0, sizeof(*ctx)); - spin_lock_init(&ctx->lock); - mutex_init(&ctx->mutex); - INIT_LIST_HEAD(&ctx->counter_list); - INIT_LIST_HEAD(&ctx->event_list); - ctx->task = task; -} - /* * inherit a counter from parent task to child task: */ @@ -3195,7 +3255,6 @@ inherit_counter(struct perf_counter *parent_counter, /* * Link it up in the child's context: */ - child_counter->task = child; add_counter_to_ctx(child_counter, child_ctx); child_counter->parent = parent_counter; @@ -3294,40 +3353,15 @@ __perf_counter_exit_task(struct task_struct *child, struct perf_counter *parent_counter; /* - * If we do not self-reap then we have to wait for the - * child task to unschedule (it will happen for sure), - * so that its counter is at its final count. (This - * condition triggers rarely - child tasks usually get - * off their CPU before the parent has a chance to - * get this far into the reaping action) + * Protect against concurrent operations on child_counter + * due its fd getting closed, etc. */ - if (child != current) { - wait_task_inactive(child, 0); - update_counter_times(child_counter); - list_del_counter(child_counter, child_ctx); - } else { - struct perf_cpu_context *cpuctx; - unsigned long flags; - - /* - * Disable and unlink this counter. - * - * Be careful about zapping the list - IRQ/NMI context - * could still be processing it: - */ - local_irq_save(flags); - perf_disable(); - - cpuctx = &__get_cpu_var(perf_cpu_context); + mutex_lock(&child_counter->mutex); - group_sched_out(child_counter, cpuctx, child_ctx); - update_counter_times(child_counter); + update_counter_times(child_counter); + list_del_counter(child_counter, child_ctx); - list_del_counter(child_counter, child_ctx); - - perf_enable(); - local_irq_restore(flags); - } + mutex_unlock(&child_counter->mutex); parent_counter = child_counter->parent; /* @@ -3346,19 +3380,29 @@ __perf_counter_exit_task(struct task_struct *child, * * Note: we may be running in child context, but the PID is not hashed * anymore so new counters will not be added. + * (XXX not sure that is true when we get called from flush_old_exec. + * -- paulus) */ void perf_counter_exit_task(struct task_struct *child) { struct perf_counter *child_counter, *tmp; struct perf_counter_context *child_ctx; + unsigned long flags; WARN_ON_ONCE(child != current); - child_ctx = &child->perf_counter_ctx; + child_ctx = child->perf_counter_ctxp; - if (likely(!child_ctx->nr_counters)) + if (likely(!child_ctx)) return; + local_irq_save(flags); + __perf_counter_task_sched_out(child_ctx); + child->perf_counter_ctxp = NULL; + local_irq_restore(flags); + + mutex_lock(&child_ctx->mutex); + again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, list_entry) @@ -3371,6 +3415,10 @@ again: */ if (!list_empty(&child_ctx->counter_list)) goto again; + + mutex_unlock(&child_ctx->mutex); + + put_ctx(child_ctx); } /* @@ -3382,19 +3430,25 @@ void perf_counter_init_task(struct task_struct *child) struct perf_counter *counter; struct task_struct *parent = current; - child_ctx = &child->perf_counter_ctx; - parent_ctx = &parent->perf_counter_ctx; - - __perf_counter_init_context(child_ctx, child); + child->perf_counter_ctxp = NULL; /* * This is executed from the parent task context, so inherit - * counters that have been marked for cloning: + * counters that have been marked for cloning. + * First allocate and initialize a context for the child. */ - if (likely(!parent_ctx->nr_counters)) + child_ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); + if (!child_ctx) + return; + + parent_ctx = parent->perf_counter_ctxp; + if (likely(!parent_ctx || !parent_ctx->nr_counters)) return; + __perf_counter_init_context(child_ctx, child); + child->perf_counter_ctxp = child_ctx; + /* * Lock the parent list. No need to lock the child - not PID * hashed yet and not running, so nobody can access it. -- cgit v1.2.3 From 564c2b210add41df9a3a5aaa365c1d97cff6110d Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 22 May 2009 14:27:22 +1000 Subject: perf_counter: Optimize context switch between identical inherited contexts When monitoring a process and its descendants with a set of inherited counters, we can often get the situation in a context switch where both the old (outgoing) and new (incoming) process have the same set of counters, and their values are ultimately going to be added together. In that situation it doesn't matter which set of counters are used to count the activity for the new process, so there is really no need to go through the process of reading the hardware counters and updating the old task's counters and then setting up the PMU for the new task. This optimizes the context switch in this situation. Instead of scheduling out the perf_counter_context for the old task and scheduling in the new context, we simply transfer the old context to the new task and keep using it without interruption. The new context gets transferred to the old task. This means that both tasks still have a valid perf_counter_context, so no special case is introduced when the old task gets scheduled in again, either on this CPU or another CPU. The equivalence of contexts is detected by keeping a pointer in each cloned context pointing to the context it was cloned from. To cope with the situation where a context is changed by adding or removing counters after it has been cloned, we also keep a generation number on each context which is incremented every time a context is changed. When a context is cloned we take a copy of the parent's generation number, and two cloned contexts are equivalent only if they have the same parent and the same generation number. In order that the parent context pointer remains valid (and is not reused), we increment the parent context's reference count for each context cloned from it. Since we don't have individual fds for the counters in a cloned context, the only thing that can make two clones of a given parent different after they have been cloned is enabling or disabling all counters with prctl. To account for this, we keep a count of the number of enabled counters in each context. Two contexts must have the same number of enabled counters to be considered equivalent. Here are some measurements of the context switch time as measured with the lat_ctx benchmark from lmbench, comparing the times obtained with and without this patch series: -----Unmodified----- With this patch series Counters: none 2 HW 4H+4S none 2 HW 4H+4S 2 processes: Average 3.44 6.45 11.24 3.12 3.39 3.60 St dev 0.04 0.04 0.13 0.05 0.17 0.19 8 processes: Average 6.45 8.79 14.00 5.57 6.23 7.57 St dev 1.27 1.04 0.88 1.42 1.46 1.42 32 processes: Average 5.56 8.43 13.78 5.28 5.55 7.15 St dev 0.41 0.47 0.53 0.54 0.57 0.81 The numbers are the mean and standard deviation of 20 runs of lat_ctx. The "none" columns are lat_ctx run directly without any counters. The "2 HW" columns are with lat_ctx run under perfstat, counting cycles and instructions. The "4H+4S" columns are lat_ctx run under perfstat with 4 hardware counters and 4 software counters (cycles, instructions, cache references, cache misses, task clock, context switch, cpu migrations, and page faults). [ Impact: performance optimization of counter context-switches ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo LKML-Reference: <18966.10666.517218.332164@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 109 +++++++++++++++++++++++++++++++++++++++++++------- kernel/sched.c | 2 +- 2 files changed, 96 insertions(+), 15 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 06ea3eae886e..c10055416dea 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -104,8 +104,11 @@ static void get_ctx(struct perf_counter_context *ctx) static void put_ctx(struct perf_counter_context *ctx) { - if (atomic_dec_and_test(&ctx->refcount)) + if (atomic_dec_and_test(&ctx->refcount)) { + if (ctx->parent_ctx) + put_ctx(ctx->parent_ctx); kfree(ctx); + } } static void @@ -127,6 +130,8 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_add_rcu(&counter->event_entry, &ctx->event_list); ctx->nr_counters++; + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) + ctx->nr_enabled++; } /* @@ -141,6 +146,8 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) if (list_empty(&counter->list_entry)) return; ctx->nr_counters--; + if (counter->state >= PERF_COUNTER_STATE_INACTIVE) + ctx->nr_enabled--; list_del_init(&counter->list_entry); list_del_rcu(&counter->event_entry); @@ -203,6 +210,22 @@ group_sched_out(struct perf_counter *group_counter, cpuctx->exclusive = 0; } +/* + * Mark this context as not being a clone of another. + * Called when counters are added to or removed from this context. + * We also increment our generation number so that anything that + * was cloned from this context before this will not match anything + * cloned from this context after this. + */ +static void unclone_ctx(struct perf_counter_context *ctx) +{ + ++ctx->generation; + if (!ctx->parent_ctx) + return; + put_ctx(ctx->parent_ctx); + ctx->parent_ctx = NULL; +} + /* * Cross CPU call to remove a performance counter * @@ -263,6 +286,7 @@ static void perf_counter_remove_from_context(struct perf_counter *counter) struct perf_counter_context *ctx = counter->ctx; struct task_struct *task = ctx->task; + unclone_ctx(ctx); if (!task) { /* * Per cpu counters are removed via an smp call and @@ -378,6 +402,7 @@ static void __perf_counter_disable(void *info) else counter_sched_out(counter, cpuctx, ctx); counter->state = PERF_COUNTER_STATE_OFF; + ctx->nr_enabled--; } spin_unlock_irqrestore(&ctx->lock, flags); @@ -419,6 +444,7 @@ static void perf_counter_disable(struct perf_counter *counter) if (counter->state == PERF_COUNTER_STATE_INACTIVE) { update_counter_times(counter); counter->state = PERF_COUNTER_STATE_OFF; + ctx->nr_enabled--; } spin_unlock_irq(&ctx->lock); @@ -727,6 +753,7 @@ static void __perf_counter_enable(void *info) goto unlock; counter->state = PERF_COUNTER_STATE_INACTIVE; counter->tstamp_enabled = ctx->time - counter->total_time_enabled; + ctx->nr_enabled++; /* * If the counter is in a group and isn't the group leader, @@ -817,6 +844,7 @@ static void perf_counter_enable(struct perf_counter *counter) counter->state = PERF_COUNTER_STATE_INACTIVE; counter->tstamp_enabled = ctx->time - counter->total_time_enabled; + ctx->nr_enabled++; } out: spin_unlock_irq(&ctx->lock); @@ -861,6 +889,25 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, spin_unlock(&ctx->lock); } +/* + * Test whether two contexts are equivalent, i.e. whether they + * have both been cloned from the same version of the same context + * and they both have the same number of enabled counters. + * If the number of enabled counters is the same, then the set + * of enabled counters should be the same, because these are both + * inherited contexts, therefore we can't access individual counters + * in them directly with an fd; we can only enable/disable all + * counters via prctl, or enable/disable all counters in a family + * via ioctl, which will have the same effect on both contexts. + */ +static int context_equiv(struct perf_counter_context *ctx1, + struct perf_counter_context *ctx2) +{ + return ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx + && ctx1->parent_gen == ctx2->parent_gen + && ctx1->nr_enabled == ctx2->nr_enabled; +} + /* * Called from scheduler to remove the counters of the current task, * with interrupts disabled. @@ -872,10 +919,12 @@ void __perf_counter_sched_out(struct perf_counter_context *ctx, * accessing the counter control register. If a NMI hits, then it will * not restart the counter. */ -void perf_counter_task_sched_out(struct task_struct *task, int cpu) +void perf_counter_task_sched_out(struct task_struct *task, + struct task_struct *next, int cpu) { struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); struct perf_counter_context *ctx = task->perf_counter_ctxp; + struct perf_counter_context *next_ctx; struct pt_regs *regs; if (likely(!ctx || !cpuctx->task_ctx)) @@ -885,6 +934,16 @@ void perf_counter_task_sched_out(struct task_struct *task, int cpu) regs = task_pt_regs(task); perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs, 0); + + next_ctx = next->perf_counter_ctxp; + if (next_ctx && context_equiv(ctx, next_ctx)) { + task->perf_counter_ctxp = next_ctx; + next->perf_counter_ctxp = ctx; + ctx->task = next; + next_ctx->task = task; + return; + } + __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; @@ -998,6 +1057,8 @@ void perf_counter_task_sched_in(struct task_struct *task, int cpu) if (likely(!ctx)) return; + if (cpuctx->task_ctx == ctx) + return; __perf_counter_sched_in(ctx, cpuctx, cpu); cpuctx->task_ctx = ctx; } @@ -3252,6 +3313,16 @@ inherit_counter(struct perf_counter *parent_counter, if (IS_ERR(child_counter)) return child_counter; + /* + * Make the child state follow the state of the parent counter, + * not its hw_event.disabled bit. We hold the parent's mutex, + * so we won't race with perf_counter_{en,dis}able_family. + */ + if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE) + child_counter->state = PERF_COUNTER_STATE_INACTIVE; + else + child_counter->state = PERF_COUNTER_STATE_OFF; + /* * Link it up in the child's context: */ @@ -3277,16 +3348,6 @@ inherit_counter(struct perf_counter *parent_counter, mutex_lock(&parent_counter->mutex); list_add_tail(&child_counter->child_list, &parent_counter->child_list); - /* - * Make the child state follow the state of the parent counter, - * not its hw_event.disabled bit. We hold the parent's mutex, - * so we won't race with perf_counter_{en,dis}able_family. - */ - if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE) - child_counter->state = PERF_COUNTER_STATE_INACTIVE; - else - child_counter->state = PERF_COUNTER_STATE_OFF; - mutex_unlock(&parent_counter->mutex); return child_counter; @@ -3429,6 +3490,7 @@ void perf_counter_init_task(struct task_struct *child) struct perf_counter_context *child_ctx, *parent_ctx; struct perf_counter *counter; struct task_struct *parent = current; + int inherited_all = 1; child->perf_counter_ctxp = NULL; @@ -3463,12 +3525,31 @@ void perf_counter_init_task(struct task_struct *child) if (counter != counter->group_leader) continue; - if (!counter->hw_event.inherit) + if (!counter->hw_event.inherit) { + inherited_all = 0; continue; + } if (inherit_group(counter, parent, - parent_ctx, child, child_ctx)) + parent_ctx, child, child_ctx)) { + inherited_all = 0; break; + } + } + + if (inherited_all) { + /* + * Mark the child context as a clone of the parent + * context, or of whatever the parent is a clone of. + */ + if (parent_ctx->parent_ctx) { + child_ctx->parent_ctx = parent_ctx->parent_ctx; + child_ctx->parent_gen = parent_ctx->parent_gen; + } else { + child_ctx->parent_ctx = parent_ctx; + child_ctx->parent_gen = parent_ctx->generation; + } + get_ctx(child_ctx->parent_ctx); } mutex_unlock(&parent_ctx->mutex); diff --git a/kernel/sched.c b/kernel/sched.c index 419a39d0988f..4c0d58bce6b2 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -5091,7 +5091,7 @@ need_resched_nonpreemptible: if (likely(prev != next)) { sched_info_switch(prev, next); - perf_counter_task_sched_out(prev, cpu); + perf_counter_task_sched_out(prev, next, cpu); rq->nr_switches++; rq->curr = next; -- cgit v1.2.3 From 948cd52906baf1f92aeea2f9b5c515db1b2e592a Mon Sep 17 00:00:00 2001 From: Paul Mundt Date: Fri, 22 May 2009 10:40:09 +0900 Subject: sparseirq: Allow early irq_desc allocation Presently non-legacy IRQs have their irq_desc allocated with kzalloc_node(). This assumes that all callers of irq_to_desc_node_alloc() will be sufficiently late in the boot process that kmalloc is available. While porting sparseirq support to sh this blew up immediately, as at the time that we register the CPU's interrupt vector map only bootmem is available. Check slab_is_available() to work out which path to use. [ Impact: fix SH early boot crash with sparseirq enabled ] Signed-off-by: Paul Mundt Acked-by: Yinghai Lu Cc: Andrew Morton Cc: Mel Gorman LKML-Reference: <20090522014008.GA2806@linux-sh.org> Signed-off-by: Ingo Molnar --- kernel/irq/handle.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index a3c671e0f165..18041a254d32 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -11,6 +11,7 @@ */ #include +#include #include #include #include @@ -81,11 +82,16 @@ static struct irq_desc irq_desc_init = { .lock = __SPIN_LOCK_UNLOCKED(irq_desc_init.lock), }; -void init_kstat_irqs(struct irq_desc *desc, int node, int nr) +void __ref init_kstat_irqs(struct irq_desc *desc, int node, int nr) { void *ptr; - ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), GFP_ATOMIC, node); + if (slab_is_available()) + ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), + GFP_ATOMIC, node); + else + ptr = alloc_bootmem_node(NODE_DATA(node), + nr * sizeof(*desc->kstat_irqs)); /* * don't overwite if can not get new one @@ -186,7 +192,7 @@ struct irq_desc *irq_to_desc(unsigned int irq) return NULL; } -struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node) +struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node) { struct irq_desc *desc; unsigned long flags; @@ -208,7 +214,11 @@ struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node) if (desc) goto out_unlock; - desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node); + if (slab_is_available()) + desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node); + else + desc = alloc_bootmem_node(NODE_DATA(node), sizeof(*desc)); + printk(KERN_DEBUG " alloc irq_desc for %d on node %d\n", irq, node); if (!desc) { printk(KERN_ERR "can not alloc irq_desc\n"); -- cgit v1.2.3 From e220d2dcb944c5c488b6855d15ec66d76900514f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:28:55 +0200 Subject: perf_counter: Fix dynamic irq_period logging We call perf_adjust_freq() from perf_counter_task_tick() which is is called under the rq->lock causing lock recursion. However, it's no longer required to be called under the rq->lock, so remove it from under it. Also, fix up some related comments. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163012.476197912@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 3 ++- kernel/sched.c | 3 ++- 2 files changed, 4 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c10055416dea..2f410ea2cb39 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2559,7 +2559,8 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, } /* - * + * Log irq_period changes so that analyzing tools can re-normalize the + * event flow. */ static void perf_log_period(struct perf_counter *counter, u64 period) diff --git a/kernel/sched.c b/kernel/sched.c index 4c0d58bce6b2..ad079f07c9c8 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -4875,9 +4875,10 @@ void scheduler_tick(void) update_rq_clock(rq); update_cpu_load(rq); curr->sched_class->task_tick(rq, curr, 0); - perf_counter_task_tick(curr, cpu); spin_unlock(&rq->lock); + perf_counter_task_tick(curr, cpu); + #ifdef CONFIG_SMP rq->idle_at_tick = idle_cpu(cpu); trigger_load_balance(rq, cpu); -- cgit v1.2.3 From fccc714b3148ab9741fafc1e90c3876d50df6093 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:28:56 +0200 Subject: perf_counter: Sanitize counter->mutex s/counter->mutex/counter->child_mutex/ and make sure its only used to protect child_list. The usage in __perf_counter_exit_task() doesn't appear to be problematic since ctx->mutex also covers anything related to fd tear-down. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163012.533186528@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 47 +++++++++++++++++++---------------------------- 1 file changed, 19 insertions(+), 28 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 2f410ea2cb39..679c3b5bb7d4 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -111,6 +111,10 @@ static void put_ctx(struct perf_counter_context *ctx) } } +/* + * Add a counter from the lists for its context. + * Must be called with ctx->mutex and ctx->lock held. + */ static void list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) { @@ -136,7 +140,7 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) /* * Remove a counter from the lists for its context. - * Must be called with counter->mutex and ctx->mutex held. + * Must be called with ctx->mutex and ctx->lock held. */ static void list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) @@ -276,7 +280,7 @@ static void __perf_counter_remove_from_context(void *info) /* * Remove the counter from a task's (or a CPU's) list of counters. * - * Must be called with counter->mutex and ctx->mutex held. + * Must be called with ctx->mutex held. * * CPU counters are removed with a smp call. For task counters we only * call when the task is on a CPU. @@ -1407,11 +1411,7 @@ static int perf_release(struct inode *inode, struct file *file) file->private_data = NULL; mutex_lock(&ctx->mutex); - mutex_lock(&counter->mutex); - perf_counter_remove_from_context(counter); - - mutex_unlock(&counter->mutex); mutex_unlock(&ctx->mutex); free_counter(counter); @@ -1437,7 +1437,7 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (counter->state == PERF_COUNTER_STATE_ERROR) return 0; - mutex_lock(&counter->mutex); + mutex_lock(&counter->child_mutex); values[0] = perf_counter_read(counter); n = 1; if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) @@ -1446,7 +1446,7 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) values[n++] = counter->total_time_running + atomic64_read(&counter->child_total_time_running); - mutex_unlock(&counter->mutex); + mutex_unlock(&counter->child_mutex); if (count < n * sizeof(u64)) return -EINVAL; @@ -1510,11 +1510,11 @@ static void perf_counter_for_each_child(struct perf_counter *counter, { struct perf_counter *child; - mutex_lock(&counter->mutex); + mutex_lock(&counter->child_mutex); func(counter); list_for_each_entry(child, &counter->child_list, child_list) func(child); - mutex_unlock(&counter->mutex); + mutex_unlock(&counter->child_mutex); } static void perf_counter_for_each(struct perf_counter *counter, @@ -1522,11 +1522,11 @@ static void perf_counter_for_each(struct perf_counter *counter, { struct perf_counter *child; - mutex_lock(&counter->mutex); + mutex_lock(&counter->child_mutex); perf_counter_for_each_sibling(counter, func); list_for_each_entry(child, &counter->child_list, child_list) perf_counter_for_each_sibling(child, func); - mutex_unlock(&counter->mutex); + mutex_unlock(&counter->child_mutex); } static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) @@ -3106,7 +3106,9 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, if (!group_leader) group_leader = counter; - mutex_init(&counter->mutex); + mutex_init(&counter->child_mutex); + INIT_LIST_HEAD(&counter->child_list); + INIT_LIST_HEAD(&counter->list_entry); INIT_LIST_HEAD(&counter->event_entry); INIT_LIST_HEAD(&counter->sibling_list); @@ -3114,8 +3116,6 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, mutex_init(&counter->mmap_mutex); - INIT_LIST_HEAD(&counter->child_list); - counter->cpu = cpu; counter->hw_event = *hw_event; counter->group_leader = group_leader; @@ -3346,10 +3346,9 @@ inherit_counter(struct perf_counter *parent_counter, /* * Link this into the parent counter's child list */ - mutex_lock(&parent_counter->mutex); + mutex_lock(&parent_counter->child_mutex); list_add_tail(&child_counter->child_list, &parent_counter->child_list); - - mutex_unlock(&parent_counter->mutex); + mutex_unlock(&parent_counter->child_mutex); return child_counter; } @@ -3396,9 +3395,9 @@ static void sync_child_counter(struct perf_counter *child_counter, /* * Remove this counter from the parent's list */ - mutex_lock(&parent_counter->mutex); + mutex_lock(&parent_counter->child_mutex); list_del_init(&child_counter->child_list); - mutex_unlock(&parent_counter->mutex); + mutex_unlock(&parent_counter->child_mutex); /* * Release the parent counter, if this was the last @@ -3414,17 +3413,9 @@ __perf_counter_exit_task(struct task_struct *child, { struct perf_counter *parent_counter; - /* - * Protect against concurrent operations on child_counter - * due its fd getting closed, etc. - */ - mutex_lock(&child_counter->mutex); - update_counter_times(child_counter); list_del_counter(child_counter, child_ctx); - mutex_unlock(&child_counter->mutex); - parent_counter = child_counter->parent; /* * It can happen that parent exits first, and has counters -- cgit v1.2.3 From 682076ae1de0aba9c2da509f7b19dc03e30a6e1f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:28:57 +0200 Subject: perf_counter: Sanitize context locking Ensure we're consistent with the context locks. context->mutex context->lock list_{add,del}_counter(); so that either lock is sufficient to stabilize the context. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163012.618790733@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 679c3b5bb7d4..d162d2f0b270 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -597,6 +597,8 @@ static void add_counter_to_ctx(struct perf_counter *counter, /* * Cross CPU call to install and enable a performance counter + * + * Must be called with ctx->mutex held */ static void __perf_install_in_context(void *info) { @@ -1496,13 +1498,13 @@ static void perf_counter_for_each_sibling(struct perf_counter *counter, struct perf_counter_context *ctx = counter->ctx; struct perf_counter *sibling; - spin_lock_irq(&ctx->lock); + mutex_lock(&ctx->mutex); counter = counter->group_leader; func(counter); list_for_each_entry(sibling, &counter->sibling_list, list_entry) func(sibling); - spin_unlock_irq(&ctx->lock); + mutex_unlock(&ctx->mutex); } static void perf_counter_for_each_child(struct perf_counter *counter, @@ -3414,7 +3416,10 @@ __perf_counter_exit_task(struct task_struct *child, struct perf_counter *parent_counter; update_counter_times(child_counter); + + spin_lock_irq(&child_ctx->lock); list_del_counter(child_counter, child_ctx); + spin_unlock_irq(&child_ctx->lock); parent_counter = child_counter->parent; /* -- cgit v1.2.3 From aa9c67f53d1969cf1db4c9c2db3a78c4ceb96469 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:28:59 +0200 Subject: perf_counter: Simplify context cleanup Use perf_counter_remove_from_context() to remove counters from the context. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163012.796275849@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index d162d2f0b270..0e97f8961333 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3416,10 +3416,7 @@ __perf_counter_exit_task(struct task_struct *child, struct perf_counter *parent_counter; update_counter_times(child_counter); - - spin_lock_irq(&child_ctx->lock); - list_del_counter(child_counter, child_ctx); - spin_unlock_irq(&child_ctx->lock); + perf_counter_remove_from_context(child_counter); parent_counter = child_counter->parent; /* -- cgit v1.2.3 From 082ff5a2767a0679ee543f14883adbafb631ffbe Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:29:00 +0200 Subject: perf_counter: Change pctrl() behaviour Instead of en/dis-abling all counters acting on a particular task, en/dis- able all counters we created. [ v2: fix crash on first counter enable ] Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163012.916937244@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 87 ++++++++++++++------------------------------------- 1 file changed, 24 insertions(+), 63 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0e97f8961333..4c86a6369764 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1076,79 +1076,26 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) __perf_counter_sched_in(ctx, cpuctx, cpu); } -int perf_counter_task_disable(void) +int perf_counter_task_enable(void) { - struct task_struct *curr = current; - struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; - unsigned long flags; - - if (!ctx || !ctx->nr_counters) - return 0; - - local_irq_save(flags); - __perf_counter_task_sched_out(ctx); - - spin_lock(&ctx->lock); - - /* - * Disable all the counters: - */ - perf_disable(); - - list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->state != PERF_COUNTER_STATE_ERROR) { - update_group_times(counter); - counter->state = PERF_COUNTER_STATE_OFF; - } - } - - perf_enable(); - - spin_unlock_irqrestore(&ctx->lock, flags); + mutex_lock(¤t->perf_counter_mutex); + list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) + perf_counter_enable(counter); + mutex_unlock(¤t->perf_counter_mutex); return 0; } -int perf_counter_task_enable(void) +int perf_counter_task_disable(void) { - struct task_struct *curr = current; - struct perf_counter_context *ctx = curr->perf_counter_ctxp; struct perf_counter *counter; - unsigned long flags; - int cpu; - - if (!ctx || !ctx->nr_counters) - return 0; - - local_irq_save(flags); - cpu = smp_processor_id(); - - __perf_counter_task_sched_out(ctx); - - spin_lock(&ctx->lock); - /* - * Disable all the counters: - */ - perf_disable(); - - list_for_each_entry(counter, &ctx->counter_list, list_entry) { - if (counter->state > PERF_COUNTER_STATE_OFF) - continue; - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = - ctx->time - counter->total_time_enabled; - counter->hw_event.disabled = 0; - } - perf_enable(); - - spin_unlock(&ctx->lock); - - perf_counter_task_sched_in(curr, cpu); - - local_irq_restore(flags); + mutex_lock(¤t->perf_counter_mutex); + list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) + perf_counter_disable(counter); + mutex_unlock(¤t->perf_counter_mutex); return 0; } @@ -1416,6 +1363,11 @@ static int perf_release(struct inode *inode, struct file *file) perf_counter_remove_from_context(counter); mutex_unlock(&ctx->mutex); + mutex_lock(&counter->owner->perf_counter_mutex); + list_del_init(&counter->owner_entry); + mutex_unlock(&counter->owner->perf_counter_mutex); + put_task_struct(counter->owner); + free_counter(counter); put_context(ctx); @@ -3272,6 +3224,12 @@ SYSCALL_DEFINE5(perf_counter_open, perf_install_in_context(ctx, counter, cpu); mutex_unlock(&ctx->mutex); + counter->owner = current; + get_task_struct(current); + mutex_lock(¤t->perf_counter_mutex); + list_add_tail(&counter->owner_entry, ¤t->perf_counter_list); + mutex_unlock(¤t->perf_counter_mutex); + fput_light(counter_file, fput_needed2); out_fput: @@ -3488,6 +3446,9 @@ void perf_counter_init_task(struct task_struct *child) child->perf_counter_ctxp = NULL; + mutex_init(&child->perf_counter_mutex); + INIT_LIST_HEAD(&child->perf_counter_list); + /* * This is executed from the parent task context, so inherit * counters that have been marked for cloning. -- cgit v1.2.3 From 475c55797323b67435083f6e2eb8ee670f6410ec Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Sat, 23 May 2009 18:29:01 +0200 Subject: perf_counter: Remove perf_counter_context::nr_enabled now that pctrl() no longer disables other people's counters, remove the PMU cache code that deals with that. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090523163013.032998331@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 11 +---------- 1 file changed, 1 insertion(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4c86a6369764..cb4062559b47 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -134,8 +134,6 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) list_add_rcu(&counter->event_entry, &ctx->event_list); ctx->nr_counters++; - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) - ctx->nr_enabled++; } /* @@ -150,8 +148,6 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) if (list_empty(&counter->list_entry)) return; ctx->nr_counters--; - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) - ctx->nr_enabled--; list_del_init(&counter->list_entry); list_del_rcu(&counter->event_entry); @@ -406,7 +402,6 @@ static void __perf_counter_disable(void *info) else counter_sched_out(counter, cpuctx, ctx); counter->state = PERF_COUNTER_STATE_OFF; - ctx->nr_enabled--; } spin_unlock_irqrestore(&ctx->lock, flags); @@ -448,7 +443,6 @@ static void perf_counter_disable(struct perf_counter *counter) if (counter->state == PERF_COUNTER_STATE_INACTIVE) { update_counter_times(counter); counter->state = PERF_COUNTER_STATE_OFF; - ctx->nr_enabled--; } spin_unlock_irq(&ctx->lock); @@ -759,7 +753,6 @@ static void __perf_counter_enable(void *info) goto unlock; counter->state = PERF_COUNTER_STATE_INACTIVE; counter->tstamp_enabled = ctx->time - counter->total_time_enabled; - ctx->nr_enabled++; /* * If the counter is in a group and isn't the group leader, @@ -850,7 +843,6 @@ static void perf_counter_enable(struct perf_counter *counter) counter->state = PERF_COUNTER_STATE_INACTIVE; counter->tstamp_enabled = ctx->time - counter->total_time_enabled; - ctx->nr_enabled++; } out: spin_unlock_irq(&ctx->lock); @@ -910,8 +902,7 @@ static int context_equiv(struct perf_counter_context *ctx1, struct perf_counter_context *ctx2) { return ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx - && ctx1->parent_gen == ctx2->parent_gen - && ctx1->nr_enabled == ctx2->nr_enabled; + && ctx1->parent_gen == ctx2->parent_gen; } /* -- cgit v1.2.3 From a3862d3f814ce7dfca9eed56ac23d29db3aee8d5 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 24 May 2009 09:02:37 +0200 Subject: perf_counter: Increase mmap limit In a default 'perf top' run the tool will create a counter for each online CPU. With enough CPUs this will eventually exhaust the default limit. So scale it up with the number of online CPUs. Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur Cc: Mike Galbraith LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index cb4062559b47..6cdf8248eda2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1704,6 +1704,12 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) user_extra = nr_pages + 1; user_lock_limit = sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); + + /* + * Increase the limit linearly with more CPUs: + */ + user_lock_limit *= num_online_cpus(); + user_locked = atomic_long_read(&user->locked_vm) + user_extra; extra = 0; -- cgit v1.2.3 From 32bdfac5462d777f35b00838893c4f87baf23efe Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Sun, 24 May 2009 21:15:07 +0200 Subject: PM: Do not hold dpm_list_mtx while disabling/enabling nonboot CPUs We shouldn't hold dpm_list_mtx while executing [disable|enable]_nonboot_cpus(), because theoretically this may lead to a deadlock as shown by the following example (provided by Johannes Berg): CPU 3 CPU 2 CPU 1 suspend/hibernate something: rtnl_lock() device_pm_lock() -> mutex_lock(&dpm_list_mtx) mutex_lock(&dpm_list_mtx) linkwatch_work -> rtnl_lock() disable_nonboot_cpus() -> flush CPU 3 workqueue Fortunately, device drivers are supposed to stop any activities that might lead to the registration of new device objects way before disable_nonboot_cpus() is called, so it shouldn't be necessary to hold dpm_list_mtx over the entire late part of device suspend and early part of device resume. Thus, during the late suspend and the early resume of devices acquire dpm_list_mtx only when dpm_list is going to be traversed and release it right after that. This patch is reported to fix the regressions tracked as http://bugzilla.kernel.org/show_bug.cgi?id=13245. Signed-off-by: Rafael J. Wysocki Acked-by: Alan Stern Reported-by: Miles Lane Tested-by: Ming Lei --- kernel/kexec.c | 2 -- kernel/power/disk.c | 21 +++------------------ kernel/power/main.c | 7 +------ 3 files changed, 4 insertions(+), 26 deletions(-) (limited to 'kernel') diff --git a/kernel/kexec.c b/kernel/kexec.c index 5a758c6e4950..e4983770913b 100644 --- a/kernel/kexec.c +++ b/kernel/kexec.c @@ -1451,7 +1451,6 @@ int kernel_kexec(void) error = device_suspend(PMSG_FREEZE); if (error) goto Resume_console; - device_pm_lock(); /* At this point, device_suspend() has been called, * but *not* device_power_down(). We *must* * device_power_down() now. Otherwise, drivers for @@ -1489,7 +1488,6 @@ int kernel_kexec(void) enable_nonboot_cpus(); device_power_up(PMSG_RESTORE); Resume_devices: - device_pm_unlock(); device_resume(PMSG_RESTORE); Resume_console: resume_console(); diff --git a/kernel/power/disk.c b/kernel/power/disk.c index b0dc9e7a0d17..5cb080e7eebd 100644 --- a/kernel/power/disk.c +++ b/kernel/power/disk.c @@ -215,8 +215,6 @@ static int create_image(int platform_mode) if (error) return error; - device_pm_lock(); - /* At this point, device_suspend() has been called, but *not* * device_power_down(). We *must* call device_power_down() now. * Otherwise, drivers for some devices (e.g. interrupt controllers) @@ -227,7 +225,7 @@ static int create_image(int platform_mode) if (error) { printk(KERN_ERR "PM: Some devices failed to power down, " "aborting hibernation\n"); - goto Unlock; + return error; } error = platform_pre_snapshot(platform_mode); @@ -280,9 +278,6 @@ static int create_image(int platform_mode) device_power_up(in_suspend ? (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); - Unlock: - device_pm_unlock(); - return error; } @@ -344,13 +339,11 @@ static int resume_target_kernel(bool platform_mode) { int error; - device_pm_lock(); - error = device_power_down(PMSG_QUIESCE); if (error) { printk(KERN_ERR "PM: Some devices failed to power down, " "aborting resume\n"); - goto Unlock; + return error; } error = platform_pre_restore(platform_mode); @@ -403,9 +396,6 @@ static int resume_target_kernel(bool platform_mode) device_power_up(PMSG_RECOVER); - Unlock: - device_pm_unlock(); - return error; } @@ -464,11 +454,9 @@ int hibernation_platform_enter(void) goto Resume_devices; } - device_pm_lock(); - error = device_power_down(PMSG_HIBERNATE); if (error) - goto Unlock; + goto Resume_devices; error = hibernation_ops->prepare(); if (error) @@ -493,9 +481,6 @@ int hibernation_platform_enter(void) device_power_up(PMSG_RESTORE); - Unlock: - device_pm_unlock(); - Resume_devices: entering_platform_hibernation = false; device_resume(PMSG_RESTORE); diff --git a/kernel/power/main.c b/kernel/power/main.c index f99ed6a75eac..868028280d13 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -289,12 +289,10 @@ static int suspend_enter(suspend_state_t state) { int error; - device_pm_lock(); - if (suspend_ops->prepare) { error = suspend_ops->prepare(); if (error) - goto Done; + return error; } error = device_power_down(PMSG_SUSPEND); @@ -343,9 +341,6 @@ static int suspend_enter(suspend_state_t state) if (suspend_ops->finish) suspend_ops->finish(); - Done: - device_pm_unlock(); - return error; } -- cgit v1.2.3 From d5a877e8dd409d8c702986d06485c374b705d340 Mon Sep 17 00:00:00 2001 From: James Bottomley Date: Sun, 24 May 2009 13:03:43 -0700 Subject: async: make sure independent async domains can't accidentally entangle The problem occurs when async_synchronize_full_domain() is called when the async_pending list is not empty. This will cause lowest_running() to return the cookie of the first entry on the async_pending list, which might be nothing at all to do with the domain being asked for and thus cause the domain synchronization to wait for an unrelated domain. This can cause a deadlock if domain synchronization is used from one domain to wait for another. Fix by running over the async_pending list to see if any pending items actually belong to our domain (and return their cookies if they do). Signed-off-by: James Bottomley Signed-off-by: Arjan van de Ven Signed-off-by: Linus Torvalds --- kernel/async.c | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/async.c b/kernel/async.c index 968ef9457d4e..50540301ed0f 100644 --- a/kernel/async.c +++ b/kernel/async.c @@ -92,19 +92,23 @@ extern int initcall_debug; static async_cookie_t __lowest_in_progress(struct list_head *running) { struct async_entry *entry; + async_cookie_t ret = next_cookie; /* begin with "infinity" value */ + if (!list_empty(running)) { entry = list_first_entry(running, struct async_entry, list); - return entry->cookie; - } else if (!list_empty(&async_pending)) { - entry = list_first_entry(&async_pending, - struct async_entry, list); - return entry->cookie; - } else { - /* nothing in progress... next_cookie is "infinity" */ - return next_cookie; + ret = entry->cookie; } + if (!list_empty(&async_pending)) { + list_for_each_entry(entry, &async_pending, list) + if (entry->running == running) { + ret = entry->cookie; + break; + } + } + + return ret; } static async_cookie_t lowest_in_progress(struct list_head *running) -- cgit v1.2.3 From e4cbb4e3ac8b09fdb11e39e5a5611bfab0a7cd1a Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 19 May 2009 15:50:30 +0200 Subject: perf_counter: Move child perfcounter init to after scheduler init Initialize a task's perfcounters (inherit from parent, etc.) after the child task's scheduler fields have been initialized already. [ Impact: cleanup ] Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur Cc: Mike Galbraith LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/fork.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index e72a09f5355b..675e01e9072a 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -984,7 +984,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, goto fork_out; rt_mutex_init_task(p); - perf_counter_init_task(p); #ifdef CONFIG_PROVE_LOCKING DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); @@ -1096,6 +1095,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, /* Perform scheduler related setup. Assign this task to a CPU. */ sched_fork(p, clone_flags); + perf_counter_init_task(p); if ((retval = audit_alloc(p))) goto bad_fork_cleanup_policy; -- cgit v1.2.3 From 771d7cde144d87f2d1fbee4da3c6234d61f7e42a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 25 May 2009 14:45:26 +0200 Subject: perf_counter: Make pctrl() affect inherited counters too Paul noted that the new ptcrl() didn't work on child counters. Reported-by: Paul Mackerras Signed-off-by: Peter Zijlstra Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525124600.203151469@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 48 ++++++++++++++++++++++++------------------------ 1 file changed, 24 insertions(+), 24 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 6cdf8248eda2..217dbcce2ebd 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1067,30 +1067,6 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) __perf_counter_sched_in(ctx, cpuctx, cpu); } -int perf_counter_task_enable(void) -{ - struct perf_counter *counter; - - mutex_lock(¤t->perf_counter_mutex); - list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) - perf_counter_enable(counter); - mutex_unlock(¤t->perf_counter_mutex); - - return 0; -} - -int perf_counter_task_disable(void) -{ - struct perf_counter *counter; - - mutex_lock(¤t->perf_counter_mutex); - list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) - perf_counter_disable(counter); - mutex_unlock(¤t->perf_counter_mutex); - - return 0; -} - static void perf_log_period(struct perf_counter *counter, u64 period); static void perf_adjust_freq(struct perf_counter_context *ctx) @@ -1505,6 +1481,30 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) return 0; } +int perf_counter_task_enable(void) +{ + struct perf_counter *counter; + + mutex_lock(¤t->perf_counter_mutex); + list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) + perf_counter_for_each_child(counter, perf_counter_enable); + mutex_unlock(¤t->perf_counter_mutex); + + return 0; +} + +int perf_counter_task_disable(void) +{ + struct perf_counter *counter; + + mutex_lock(¤t->perf_counter_mutex); + list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) + perf_counter_for_each_child(counter, perf_counter_disable); + mutex_unlock(¤t->perf_counter_mutex); + + return 0; +} + /* * Callers need to ensure there can be no nesting of this function, otherwise * the seqlock logic goes bad. We can not serialize this because the arch -- cgit v1.2.3 From 6ab423e0eaca827fbd201ca4ae7d4f8573a366b2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 25 May 2009 14:45:27 +0200 Subject: perf_counter: Propagate inheritance failures down the fork() path Fail fork() when we fail inheritance for some reason (-ENOMEM most likely). Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525124600.324656474@chello.nl> Signed-off-by: Ingo Molnar --- kernel/fork.c | 6 +++++- kernel/perf_counter.c | 20 ++++++++++++-------- 2 files changed, 17 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index 675e01e9072a..c07c3335ceac 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1095,7 +1095,10 @@ static struct task_struct *copy_process(unsigned long clone_flags, /* Perform scheduler related setup. Assign this task to a CPU. */ sched_fork(p, clone_flags); - perf_counter_init_task(p); + + retval = perf_counter_init_task(p); + if (retval) + goto bad_fork_cleanup_policy; if ((retval = audit_alloc(p))) goto bad_fork_cleanup_policy; @@ -1295,6 +1298,7 @@ bad_fork_cleanup_semundo: bad_fork_cleanup_audit: audit_free(p); bad_fork_cleanup_policy: + perf_counter_exit_task(p); #ifdef CONFIG_NUMA mpol_put(p->mempolicy); bad_fork_cleanup_cgroup: diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 217dbcce2ebd..7a7a144870ef 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3434,18 +3434,23 @@ again: /* * Initialize the perf_counter context in task_struct */ -void perf_counter_init_task(struct task_struct *child) +int perf_counter_init_task(struct task_struct *child) { struct perf_counter_context *child_ctx, *parent_ctx; struct perf_counter *counter; struct task_struct *parent = current; int inherited_all = 1; + int ret = 0; child->perf_counter_ctxp = NULL; mutex_init(&child->perf_counter_mutex); INIT_LIST_HEAD(&child->perf_counter_list); + parent_ctx = parent->perf_counter_ctxp; + if (likely(!parent_ctx || !parent_ctx->nr_counters)) + return 0; + /* * This is executed from the parent task context, so inherit * counters that have been marked for cloning. @@ -3454,11 +3459,7 @@ void perf_counter_init_task(struct task_struct *child) child_ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); if (!child_ctx) - return; - - parent_ctx = parent->perf_counter_ctxp; - if (likely(!parent_ctx || !parent_ctx->nr_counters)) - return; + return -ENOMEM; __perf_counter_init_context(child_ctx, child); child->perf_counter_ctxp = child_ctx; @@ -3482,8 +3483,9 @@ void perf_counter_init_task(struct task_struct *child) continue; } - if (inherit_group(counter, parent, - parent_ctx, child, child_ctx)) { + ret = inherit_group(counter, parent, parent_ctx, + child, child_ctx); + if (ret) { inherited_all = 0; break; } @@ -3505,6 +3507,8 @@ void perf_counter_init_task(struct task_struct *child) } mutex_unlock(&parent_ctx->mutex); + + return ret; } static void __cpuinit perf_counter_init_cpu(int cpu) -- cgit v1.2.3 From 10989fb2451763fae6f42d85fa6106c8fd010cf5 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 25 May 2009 14:45:28 +0200 Subject: perf_counter: Fix PERF_COUNTER_CONTEXT_SWITCHES for cpu counters Ingo noticed that cpu counters had 0 context switches, even though there was plenty scheduling on the cpu. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525124600.419025548@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 7a7a144870ef..14b1fe984832 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -924,14 +924,13 @@ void perf_counter_task_sched_out(struct task_struct *task, struct perf_counter_context *next_ctx; struct pt_regs *regs; + regs = task_pt_regs(task); + perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs, 0); + if (likely(!ctx || !cpuctx->task_ctx)) return; update_context_time(ctx); - - regs = task_pt_regs(task); - perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs, 0); - next_ctx = next->perf_counter_ctxp; if (next_ctx && context_equiv(ctx, next_ctx)) { task->perf_counter_ctxp = next_ctx; -- cgit v1.2.3 From a78ac3258782f3e64cb40beb5990808e1febcc0c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 25 May 2009 17:39:05 +0200 Subject: perf_counter: Generic per counter interrupt throttle Introduce a generic per counter interrupt throttle. This uses the perf_counter_overflow() quick disable to throttle a specific counter when its going too fast when a pmu->unthrottle() method is provided which can undo the quick disable. Power needs to implement both the quick disable and the unthrottle method. Signed-off-by: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525153931.703093461@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++---- kernel/sysctl.c | 8 +++++++ 2 files changed, 63 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 14b1fe984832..ec9c4007a7f9 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -46,6 +46,7 @@ static atomic_t nr_comm_tracking __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ +int sysctl_perf_counter_limit __read_mostly = 100000; /* max NMIs per second */ /* * Lock for (sysadmin-configurable) counter reservations: @@ -1066,12 +1067,15 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) __perf_counter_sched_in(ctx, cpuctx, cpu); } +#define MAX_INTERRUPTS (~0ULL) + +static void perf_log_throttle(struct perf_counter *counter, int enable); static void perf_log_period(struct perf_counter *counter, u64 period); static void perf_adjust_freq(struct perf_counter_context *ctx) { struct perf_counter *counter; - u64 irq_period; + u64 interrupts, irq_period; u64 events, period; s64 delta; @@ -1080,10 +1084,19 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) if (counter->state != PERF_COUNTER_STATE_ACTIVE) continue; + interrupts = counter->hw.interrupts; + counter->hw.interrupts = 0; + + if (interrupts == MAX_INTERRUPTS) { + perf_log_throttle(counter, 1); + counter->pmu->unthrottle(counter); + interrupts = 2*sysctl_perf_counter_limit/HZ; + } + if (!counter->hw_event.freq || !counter->hw_event.irq_freq) continue; - events = HZ * counter->hw.interrupts * counter->hw.irq_period; + events = HZ * interrupts * counter->hw.irq_period; period = div64_u64(events, counter->hw_event.irq_freq); delta = (s64)(1 + period - counter->hw.irq_period); @@ -1097,7 +1110,6 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) perf_log_period(counter, irq_period); counter->hw.irq_period = irq_period; - counter->hw.interrupts = 0; } spin_unlock(&ctx->lock); } @@ -2543,6 +2555,35 @@ static void perf_log_period(struct perf_counter *counter, u64 period) perf_output_end(&handle); } +/* + * IRQ throttle logging + */ + +static void perf_log_throttle(struct perf_counter *counter, int enable) +{ + struct perf_output_handle handle; + int ret; + + struct { + struct perf_event_header header; + u64 time; + } throttle_event = { + .header = { + .type = PERF_EVENT_THROTTLE + 1, + .misc = 0, + .size = sizeof(throttle_event), + }, + .time = sched_clock(), + }; + + ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 0, 0); + if (ret) + return; + + perf_output_put(&handle, throttle_event); + perf_output_end(&handle); +} + /* * Generic counter overflow handling. */ @@ -2551,9 +2592,19 @@ int perf_counter_overflow(struct perf_counter *counter, int nmi, struct pt_regs *regs, u64 addr) { int events = atomic_read(&counter->event_limit); + int throttle = counter->pmu->unthrottle != NULL; int ret = 0; - counter->hw.interrupts++; + if (!throttle) { + counter->hw.interrupts++; + } else if (counter->hw.interrupts != MAX_INTERRUPTS) { + counter->hw.interrupts++; + if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) { + counter->hw.interrupts = MAX_INTERRUPTS; + perf_log_throttle(counter, 0); + ret = 1; + } + } /* * XXX event_limit might not quite work as expected on inherited diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 3cb1849f5989..0c4bf863afa3 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -930,6 +930,14 @@ static struct ctl_table kern_table[] = { .mode = 0644, .proc_handler = &proc_dointvec, }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "perf_counter_int_limit", + .data = &sysctl_perf_counter_limit, + .maxlen = sizeof(sysctl_perf_counter_limit), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, #endif /* * NOTE: do not add new entries to this table unless you have read -- cgit v1.2.3 From 0127c3ea082ee9f1034789b978dfc7fd83254617 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 25 May 2009 22:03:26 +0200 Subject: perf_counter: fix warning & lockup - remove bogus warning - fix wakeup from NMI path lockup - also fix up whitespace noise in perf_counter.h Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Corey Ashford Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <20090525153931.703093461@chello.nl> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ec9c4007a7f9..070f92d3232a 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2576,7 +2576,7 @@ static void perf_log_throttle(struct perf_counter *counter, int enable) .time = sched_clock(), }; - ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 0, 0); + ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 1, 0); if (ret) return; @@ -3449,8 +3449,6 @@ void perf_counter_exit_task(struct task_struct *child) struct perf_counter_context *child_ctx; unsigned long flags; - WARN_ON_ONCE(child != current); - child_ctx = child->perf_counter_ctxp; if (likely(!child_ctx)) -- cgit v1.2.3 From 329d876d6fd326109f191ae0fb2798b8834fb70b Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Tue, 26 May 2009 08:10:00 +0200 Subject: perf_counter: Initialize ->oncpu properly This shouldnt matter normally (and i have not seen any misbehavior), because active counters always have a proper ->oncpu value - but nevertheless initialize the field properly to -1. [ Impact: cleanup ] Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 070f92d3232a..367299f91aaf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3122,6 +3122,8 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->group_leader = group_leader; counter->pmu = NULL; counter->ctx = ctx; + counter->oncpu = -1; + get_ctx(ctx); counter->state = PERF_COUNTER_STATE_INACTIVE; -- cgit v1.2.3 From ab2b7ebaad16226c9a5e85c5f384d19fa58a7459 Mon Sep 17 00:00:00 2001 From: Tetsuo Handa Date: Tue, 26 May 2009 09:11:03 +0100 Subject: kmod: Release sub_info on cred allocation failure. call_usermodehelper_setup() forgot to kfree(sub_info) when prepare_usermodehelper_creds() failed. Signed-off-by: Tetsuo Handa Signed-off-by: David Howells Signed-off-by: Linus Torvalds --- kernel/kmod.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/kmod.c b/kernel/kmod.c index b750675251e5..7e95bedb2bfc 100644 --- a/kernel/kmod.c +++ b/kernel/kmod.c @@ -370,8 +370,10 @@ struct subprocess_info *call_usermodehelper_setup(char *path, char **argv, sub_info->argv = argv; sub_info->envp = envp; sub_info->cred = prepare_usermodehelper_creds(); - if (!sub_info->cred) + if (!sub_info->cred) { + kfree(sub_info); return NULL; + } out: return sub_info; -- cgit v1.2.3 From c93f7669098eb97c5376e5396e3dfb734c17df4f Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Thu, 28 May 2009 22:18:17 +1000 Subject: perf_counter: Fix race in attaching counters to tasks and exiting Commit 564c2b21 ("perf_counter: Optimize context switch between identical inherited contexts") introduced a race where it is possible that a counter being attached to a task could get attached to the wrong task, if the task is one that has inherited its context from another task via fork. This happens because the optimized context switch could switch the context to another task after find_get_context has read task->perf_counter_ctxp. In fact, it's possible that the context could then get freed, if the other task then exits. This fixes the problem by protecting both the context switch and the critical code in find_get_context with spinlocks. The context switch locks the cxt->lock of both the outgoing and incoming contexts before swapping them. That means that once code such as find_get_context has obtained the spinlock for the context associated with a task, the context can't get swapped to another task. However, the context may have been swapped in the interval between reading task->perf_counter_ctxp and getting the lock, so it is necessary to check and retry. To make sure that none of the contexts being looked at in find_get_context can get freed, this changes the context freeing code to use RCU. Thus an rcu_read_lock() is sufficient to ensure that no contexts can get freed. This part of the patch is lifted from a patch posted by Peter Zijlstra. This also adds a check to make sure that we can't add a counter to a task that is exiting. There is also a race between perf_counter_exit_task and find_get_context; this solves the race by moving the get_ctx that was in perf_counter_alloc into the locked region in find_get_context, so that once find_get_context has got the context for a task, it won't get freed even if the task calls perf_counter_exit_task. It doesn't matter if new top-level (non-inherited) counters get attached to the context after perf_counter_exit_task has detached the context from the task. They will just stay there and never get scheduled in until the counters' fds get closed, and then perf_release will remove them from the context and eventually free the context. With this, we are now doing the unclone in find_get_context rather than when a counter was added to or removed from a context (actually, we were missing the unclone_ctx() call when adding a counter to a context). We don't need to unclone when removing a counter from a context because we have no way to remove a counter from a cloned context. This also takes out the smp_wmb() in find_get_context, which Peter Zijlstra pointed out was unnecessary because the cmpxchg implies a full barrier anyway. Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra Cc: Corey Ashford Cc: Mike Galbraith Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <18974.33033.667187.273886@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 205 +++++++++++++++++++++++++++++++++++--------------- 1 file changed, 145 insertions(+), 60 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 367299f91aaf..52e5a15321d8 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -103,12 +103,22 @@ static void get_ctx(struct perf_counter_context *ctx) atomic_inc(&ctx->refcount); } +static void free_ctx(struct rcu_head *head) +{ + struct perf_counter_context *ctx; + + ctx = container_of(head, struct perf_counter_context, rcu_head); + kfree(ctx); +} + static void put_ctx(struct perf_counter_context *ctx) { if (atomic_dec_and_test(&ctx->refcount)) { if (ctx->parent_ctx) put_ctx(ctx->parent_ctx); - kfree(ctx); + if (ctx->task) + put_task_struct(ctx->task); + call_rcu(&ctx->rcu_head, free_ctx); } } @@ -211,22 +221,6 @@ group_sched_out(struct perf_counter *group_counter, cpuctx->exclusive = 0; } -/* - * Mark this context as not being a clone of another. - * Called when counters are added to or removed from this context. - * We also increment our generation number so that anything that - * was cloned from this context before this will not match anything - * cloned from this context after this. - */ -static void unclone_ctx(struct perf_counter_context *ctx) -{ - ++ctx->generation; - if (!ctx->parent_ctx) - return; - put_ctx(ctx->parent_ctx); - ctx->parent_ctx = NULL; -} - /* * Cross CPU call to remove a performance counter * @@ -281,13 +275,19 @@ static void __perf_counter_remove_from_context(void *info) * * CPU counters are removed with a smp call. For task counters we only * call when the task is on a CPU. + * + * If counter->ctx is a cloned context, callers must make sure that + * every task struct that counter->ctx->task could possibly point to + * remains valid. This is OK when called from perf_release since + * that only calls us on the top-level context, which can't be a clone. + * When called from perf_counter_exit_task, it's OK because the + * context has been detached from its task. */ static void perf_counter_remove_from_context(struct perf_counter *counter) { struct perf_counter_context *ctx = counter->ctx; struct task_struct *task = ctx->task; - unclone_ctx(ctx); if (!task) { /* * Per cpu counters are removed via an smp call and @@ -410,6 +410,16 @@ static void __perf_counter_disable(void *info) /* * Disable a counter. + * + * If counter->ctx is a cloned context, callers must make sure that + * every task struct that counter->ctx->task could possibly point to + * remains valid. This condition is satisifed when called through + * perf_counter_for_each_child or perf_counter_for_each because they + * hold the top-level counter's child_mutex, so any descendant that + * goes to exit will block in sync_child_counter. + * When called from perf_pending_counter it's OK because counter->ctx + * is the current context on this CPU and preemption is disabled, + * hence we can't get into perf_counter_task_sched_out for this context. */ static void perf_counter_disable(struct perf_counter *counter) { @@ -794,6 +804,12 @@ static void __perf_counter_enable(void *info) /* * Enable a counter. + * + * If counter->ctx is a cloned context, callers must make sure that + * every task struct that counter->ctx->task could possibly point to + * remains valid. This condition is satisfied when called through + * perf_counter_for_each_child or perf_counter_for_each as described + * for perf_counter_disable. */ static void perf_counter_enable(struct perf_counter *counter) { @@ -923,7 +939,9 @@ void perf_counter_task_sched_out(struct task_struct *task, struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); struct perf_counter_context *ctx = task->perf_counter_ctxp; struct perf_counter_context *next_ctx; + struct perf_counter_context *parent; struct pt_regs *regs; + int do_switch = 1; regs = task_pt_regs(task); perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs, 0); @@ -932,18 +950,39 @@ void perf_counter_task_sched_out(struct task_struct *task, return; update_context_time(ctx); + + rcu_read_lock(); + parent = rcu_dereference(ctx->parent_ctx); next_ctx = next->perf_counter_ctxp; - if (next_ctx && context_equiv(ctx, next_ctx)) { - task->perf_counter_ctxp = next_ctx; - next->perf_counter_ctxp = ctx; - ctx->task = next; - next_ctx->task = task; - return; + if (parent && next_ctx && + rcu_dereference(next_ctx->parent_ctx) == parent) { + /* + * Looks like the two contexts are clones, so we might be + * able to optimize the context switch. We lock both + * contexts and check that they are clones under the + * lock (including re-checking that neither has been + * uncloned in the meantime). It doesn't matter which + * order we take the locks because no other cpu could + * be trying to lock both of these tasks. + */ + spin_lock(&ctx->lock); + spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); + if (context_equiv(ctx, next_ctx)) { + task->perf_counter_ctxp = next_ctx; + next->perf_counter_ctxp = ctx; + ctx->task = next; + next_ctx->task = task; + do_switch = 0; + } + spin_unlock(&next_ctx->lock); + spin_unlock(&ctx->lock); } + rcu_read_unlock(); - __perf_counter_sched_out(ctx, cpuctx); - - cpuctx->task_ctx = NULL; + if (do_switch) { + __perf_counter_sched_out(ctx, cpuctx); + cpuctx->task_ctx = NULL; + } } static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) @@ -1215,18 +1254,13 @@ __perf_counter_init_context(struct perf_counter_context *ctx, ctx->task = task; } -static void put_context(struct perf_counter_context *ctx) -{ - if (ctx->task) - put_task_struct(ctx->task); -} - static struct perf_counter_context *find_get_context(pid_t pid, int cpu) { struct perf_cpu_context *cpuctx; struct perf_counter_context *ctx; - struct perf_counter_context *tctx; + struct perf_counter_context *parent_ctx; struct task_struct *task; + int err; /* * If cpu is not a wildcard then this is a percpu counter: @@ -1249,6 +1283,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) cpuctx = &per_cpu(perf_cpu_context, cpu); ctx = &cpuctx->ctx; + get_ctx(ctx); return ctx; } @@ -1265,37 +1300,79 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) if (!task) return ERR_PTR(-ESRCH); + /* + * Can't attach counters to a dying task. + */ + err = -ESRCH; + if (task->flags & PF_EXITING) + goto errout; + /* Reuse ptrace permission checks for now. */ - if (!ptrace_may_access(task, PTRACE_MODE_READ)) { - put_task_struct(task); - return ERR_PTR(-EACCES); + err = -EACCES; + if (!ptrace_may_access(task, PTRACE_MODE_READ)) + goto errout; + + retry_lock: + rcu_read_lock(); + retry: + ctx = rcu_dereference(task->perf_counter_ctxp); + if (ctx) { + /* + * If this context is a clone of another, it might + * get swapped for another underneath us by + * perf_counter_task_sched_out, though the + * rcu_read_lock() protects us from any context + * getting freed. Lock the context and check if it + * got swapped before we could get the lock, and retry + * if so. If we locked the right context, then it + * can't get swapped on us any more and we can + * unclone it if necessary. + * Once it's not a clone things will be stable. + */ + spin_lock_irq(&ctx->lock); + if (ctx != rcu_dereference(task->perf_counter_ctxp)) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + parent_ctx = ctx->parent_ctx; + if (parent_ctx) { + put_ctx(parent_ctx); + ctx->parent_ctx = NULL; /* no longer a clone */ + } + ++ctx->generation; + /* + * Get an extra reference before dropping the lock so that + * this context won't get freed if the task exits. + */ + get_ctx(ctx); + spin_unlock_irq(&ctx->lock); } + rcu_read_unlock(); - ctx = task->perf_counter_ctxp; if (!ctx) { ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); - if (!ctx) { - put_task_struct(task); - return ERR_PTR(-ENOMEM); - } + err = -ENOMEM; + if (!ctx) + goto errout; __perf_counter_init_context(ctx, task); - /* - * Make sure other cpus see correct values for *ctx - * once task->perf_counter_ctxp is visible to them. - */ - smp_wmb(); - tctx = cmpxchg(&task->perf_counter_ctxp, NULL, ctx); - if (tctx) { + get_ctx(ctx); + if (cmpxchg(&task->perf_counter_ctxp, NULL, ctx)) { /* * We raced with some other task; use * the context they set. */ kfree(ctx); - ctx = tctx; + goto retry_lock; } + get_task_struct(task); } + put_task_struct(task); return ctx; + + errout: + put_task_struct(task); + return ERR_PTR(err); } static void free_counter_rcu(struct rcu_head *head) @@ -1303,7 +1380,6 @@ static void free_counter_rcu(struct rcu_head *head) struct perf_counter *counter; counter = container_of(head, struct perf_counter, rcu_head); - put_ctx(counter->ctx); kfree(counter); } @@ -1324,6 +1400,7 @@ static void free_counter(struct perf_counter *counter) if (counter->destroy) counter->destroy(counter); + put_ctx(counter->ctx); call_rcu(&counter->rcu_head, free_counter_rcu); } @@ -1347,7 +1424,6 @@ static int perf_release(struct inode *inode, struct file *file) put_task_struct(counter->owner); free_counter(counter); - put_context(ctx); return 0; } @@ -1437,6 +1513,12 @@ static void perf_counter_for_each_sibling(struct perf_counter *counter, mutex_unlock(&ctx->mutex); } +/* + * Holding the top-level counter's child_mutex means that any + * descendant process that has inherited this counter will block + * in sync_child_counter if it goes to exit, thus satisfying the + * task existence requirements of perf_counter_enable/disable. + */ static void perf_counter_for_each_child(struct perf_counter *counter, void (*func)(struct perf_counter *)) { @@ -3124,8 +3206,6 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, counter->ctx = ctx; counter->oncpu = -1; - get_ctx(ctx); - counter->state = PERF_COUNTER_STATE_INACTIVE; if (hw_event->disabled) counter->state = PERF_COUNTER_STATE_OFF; @@ -3290,7 +3370,7 @@ err_free_put_context: kfree(counter); err_put_context: - put_context(ctx); + put_ctx(ctx); goto out_fput; } @@ -3322,6 +3402,7 @@ inherit_counter(struct perf_counter *parent_counter, group_leader, GFP_KERNEL); if (IS_ERR(child_counter)) return child_counter; + get_ctx(child_ctx); /* * Make the child state follow the state of the parent counter, @@ -3439,11 +3520,6 @@ __perf_counter_exit_task(struct task_struct *child, /* * When a child task exits, feed back counter values to parent counters. - * - * Note: we may be running in child context, but the PID is not hashed - * anymore so new counters will not be added. - * (XXX not sure that is true when we get called from flush_old_exec. - * -- paulus) */ void perf_counter_exit_task(struct task_struct *child) { @@ -3458,7 +3534,15 @@ void perf_counter_exit_task(struct task_struct *child) local_irq_save(flags); __perf_counter_task_sched_out(child_ctx); + + /* + * Take the context lock here so that if find_get_context is + * reading child->perf_counter_ctxp, we wait until it has + * incremented the context's refcount before we do put_ctx below. + */ + spin_lock(&child_ctx->lock); child->perf_counter_ctxp = NULL; + spin_unlock(&child_ctx->lock); local_irq_restore(flags); mutex_lock(&child_ctx->mutex); @@ -3513,6 +3597,7 @@ int perf_counter_init_task(struct task_struct *child) __perf_counter_init_context(child_ctx, child); child->perf_counter_ctxp = child_ctx; + get_task_struct(child); /* * Lock the parent list. No need to lock the child - not PID -- cgit v1.2.3 From a9862e0560866eadbc59b84867492004da436516 Mon Sep 17 00:00:00 2001 From: Andi Kleen Date: Tue, 19 May 2009 22:49:07 +0200 Subject: Export add_timer_on for modules Needed in followon patch. Signed-off-by: Andi Kleen Signed-off-by: H. Peter Anvin Signed-off-by: Hidetoshi Seto Signed-off-by: H. Peter Anvin --- kernel/timer.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/timer.c b/kernel/timer.c index cffffad01c31..e2c47b82ac36 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -756,6 +756,7 @@ void add_timer_on(struct timer_list *timer, int cpu) wake_up_idle_cpu(cpu); spin_unlock_irqrestore(&base->lock, flags); } +EXPORT_SYMBOL_GPL(add_timer_on); /** * del_timer - deactive a timer. -- cgit v1.2.3 From ad3a37de81c45f6c20d410ece86004b98f7b6d84 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 29 May 2009 16:06:20 +1000 Subject: perf_counter: Don't swap contexts containing locked mutex Peter Zijlstra pointed out that under some circumstances, we can take the mutex in a context or a counter and then swap that context or counter to another task, potentially leading to lock order inversions or the mutexes not protecting what they are supposed to protect. This fixes the problem by making sure that we never take a mutex in a context or counter which could get swapped to another task. Most of the cases where we take a mutex is on a top-level counter or context, i.e. a counter which has an fd associated with it or a context that contains such a counter. This adds WARN_ON_ONCE statements to verify that. The two cases where we need to take the mutex on a context that is a clone of another are in perf_counter_exit_task and perf_counter_init_task. The perf_counter_exit_task case is solved by uncloning the context before starting to remove the counters from it. The perf_counter_init_task is a little trickier; we temporarily disable context swapping for the parent (forking) task by setting its ctx->parent_gen to the all-1s value after locking the context, if it is a cloned context, and restore the ctx->parent_gen value at the end if the context didn't get uncloned in the meantime. This also moves the increment of the context generation count to be within the same critical section, protected by the context mutex, that adds the new counter to the context. That way, taking the mutex is sufficient to ensure that both the counter list and the generation count are stable. [ Impact: fix hangs, races with inherited and PID counters ] Signed-off-by: Paul Mackerras Acked-by: Peter Zijlstra LKML-Reference: <18975.31580.520676.619896@drongo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 89 +++++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 79 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 52e5a15321d8..db843f812a60 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -919,7 +919,8 @@ static int context_equiv(struct perf_counter_context *ctx1, struct perf_counter_context *ctx2) { return ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx - && ctx1->parent_gen == ctx2->parent_gen; + && ctx1->parent_gen == ctx2->parent_gen + && ctx1->parent_gen != ~0ull; } /* @@ -1339,7 +1340,6 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) put_ctx(parent_ctx); ctx->parent_ctx = NULL; /* no longer a clone */ } - ++ctx->generation; /* * Get an extra reference before dropping the lock so that * this context won't get freed if the task exits. @@ -1414,6 +1414,7 @@ static int perf_release(struct inode *inode, struct file *file) file->private_data = NULL; + WARN_ON_ONCE(ctx->parent_ctx); mutex_lock(&ctx->mutex); perf_counter_remove_from_context(counter); mutex_unlock(&ctx->mutex); @@ -1445,6 +1446,7 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (counter->state == PERF_COUNTER_STATE_ERROR) return 0; + WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->child_mutex); values[0] = perf_counter_read(counter); n = 1; @@ -1504,6 +1506,7 @@ static void perf_counter_for_each_sibling(struct perf_counter *counter, struct perf_counter_context *ctx = counter->ctx; struct perf_counter *sibling; + WARN_ON_ONCE(ctx->parent_ctx); mutex_lock(&ctx->mutex); counter = counter->group_leader; @@ -1524,6 +1527,7 @@ static void perf_counter_for_each_child(struct perf_counter *counter, { struct perf_counter *child; + WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->child_mutex); func(counter); list_for_each_entry(child, &counter->child_list, child_list) @@ -1536,6 +1540,7 @@ static void perf_counter_for_each(struct perf_counter *counter, { struct perf_counter *child; + WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->child_mutex); perf_counter_for_each_sibling(counter, func); list_for_each_entry(child, &counter->child_list, child_list) @@ -1741,6 +1746,7 @@ static void perf_mmap_close(struct vm_area_struct *vma) { struct perf_counter *counter = vma->vm_file->private_data; + WARN_ON_ONCE(counter->ctx->parent_ctx); if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { struct user_struct *user = current_user(); @@ -1788,6 +1794,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma) if (vma->vm_pgoff != 0) return -EINVAL; + WARN_ON_ONCE(counter->ctx->parent_ctx); mutex_lock(&counter->mmap_mutex); if (atomic_inc_not_zero(&counter->mmap_count)) { if (nr_pages != counter->data->nr_pages) @@ -3349,8 +3356,10 @@ SYSCALL_DEFINE5(perf_counter_open, goto err_free_put_context; counter->filp = counter_file; + WARN_ON_ONCE(ctx->parent_ctx); mutex_lock(&ctx->mutex); perf_install_in_context(ctx, counter, cpu); + ++ctx->generation; mutex_unlock(&ctx->mutex); counter->owner = current; @@ -3436,6 +3445,7 @@ inherit_counter(struct perf_counter *parent_counter, /* * Link this into the parent counter's child list */ + WARN_ON_ONCE(parent_counter->ctx->parent_ctx); mutex_lock(&parent_counter->child_mutex); list_add_tail(&child_counter->child_list, &parent_counter->child_list); mutex_unlock(&parent_counter->child_mutex); @@ -3485,6 +3495,7 @@ static void sync_child_counter(struct perf_counter *child_counter, /* * Remove this counter from the parent's list */ + WARN_ON_ONCE(parent_counter->ctx->parent_ctx); mutex_lock(&parent_counter->child_mutex); list_del_init(&child_counter->child_list); mutex_unlock(&parent_counter->child_mutex); @@ -3527,12 +3538,17 @@ void perf_counter_exit_task(struct task_struct *child) struct perf_counter_context *child_ctx; unsigned long flags; - child_ctx = child->perf_counter_ctxp; - - if (likely(!child_ctx)) + if (likely(!child->perf_counter_ctxp)) return; local_irq_save(flags); + /* + * We can't reschedule here because interrupts are disabled, + * and either child is current or it is a task that can't be + * scheduled, so we are now safe from rescheduling changing + * our context. + */ + child_ctx = child->perf_counter_ctxp; __perf_counter_task_sched_out(child_ctx); /* @@ -3542,6 +3558,15 @@ void perf_counter_exit_task(struct task_struct *child) */ spin_lock(&child_ctx->lock); child->perf_counter_ctxp = NULL; + if (child_ctx->parent_ctx) { + /* + * This context is a clone; unclone it so it can't get + * swapped to another process while we're removing all + * the counters from it. + */ + put_ctx(child_ctx->parent_ctx); + child_ctx->parent_ctx = NULL; + } spin_unlock(&child_ctx->lock); local_irq_restore(flags); @@ -3571,9 +3596,11 @@ again: int perf_counter_init_task(struct task_struct *child) { struct perf_counter_context *child_ctx, *parent_ctx; + struct perf_counter_context *cloned_ctx; struct perf_counter *counter; struct task_struct *parent = current; int inherited_all = 1; + u64 cloned_gen; int ret = 0; child->perf_counter_ctxp = NULL; @@ -3581,8 +3608,7 @@ int perf_counter_init_task(struct task_struct *child) mutex_init(&child->perf_counter_mutex); INIT_LIST_HEAD(&child->perf_counter_list); - parent_ctx = parent->perf_counter_ctxp; - if (likely(!parent_ctx || !parent_ctx->nr_counters)) + if (likely(!parent->perf_counter_ctxp)) return 0; /* @@ -3599,6 +3625,34 @@ int perf_counter_init_task(struct task_struct *child) child->perf_counter_ctxp = child_ctx; get_task_struct(child); + /* + * If the parent's context is a clone, temporarily set its + * parent_gen to an impossible value (all 1s) so it won't get + * swapped under us. The rcu_read_lock makes sure that + * parent_ctx continues to exist even if it gets swapped to + * another process and then freed while we are trying to get + * its lock. + */ + rcu_read_lock(); + retry: + parent_ctx = rcu_dereference(parent->perf_counter_ctxp); + /* + * No need to check if parent_ctx != NULL here; since we saw + * it non-NULL earlier, the only reason for it to become NULL + * is if we exit, and since we're currently in the middle of + * a fork we can't be exiting at the same time. + */ + spin_lock_irq(&parent_ctx->lock); + if (parent_ctx != rcu_dereference(parent->perf_counter_ctxp)) { + spin_unlock_irq(&parent_ctx->lock); + goto retry; + } + cloned_gen = parent_ctx->parent_gen; + if (parent_ctx->parent_ctx) + parent_ctx->parent_gen = ~0ull; + spin_unlock_irq(&parent_ctx->lock); + rcu_read_unlock(); + /* * Lock the parent list. No need to lock the child - not PID * hashed yet and not running, so nobody can access it. @@ -3630,10 +3684,15 @@ int perf_counter_init_task(struct task_struct *child) /* * Mark the child context as a clone of the parent * context, or of whatever the parent is a clone of. + * Note that if the parent is a clone, it could get + * uncloned at any point, but that doesn't matter + * because the list of counters and the generation + * count can't have changed since we took the mutex. */ - if (parent_ctx->parent_ctx) { - child_ctx->parent_ctx = parent_ctx->parent_ctx; - child_ctx->parent_gen = parent_ctx->parent_gen; + cloned_ctx = rcu_dereference(parent_ctx->parent_ctx); + if (cloned_ctx) { + child_ctx->parent_ctx = cloned_ctx; + child_ctx->parent_gen = cloned_gen; } else { child_ctx->parent_ctx = parent_ctx; child_ctx->parent_gen = parent_ctx->generation; @@ -3643,6 +3702,16 @@ int perf_counter_init_task(struct task_struct *child) mutex_unlock(&parent_ctx->mutex); + /* + * Restore the clone status of the parent. + */ + if (parent_ctx->parent_ctx) { + spin_lock_irq(&parent_ctx->lock); + if (parent_ctx->parent_ctx) + parent_ctx->parent_gen = cloned_gen; + spin_unlock_irq(&parent_ctx->lock); + } + return ret; } -- cgit v1.2.3 From 3f4dee227348daac32f36daad9a91059efd0723e Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 29 May 2009 11:25:09 +0200 Subject: perf_counter: Fix cpuctx->task_ctx races Peter noticed that we are sometimes reading cpuctx->task_ctx with interrupts enabled. Noticed-by: Peter Zijlstra Acked-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index db843f812a60..eb346048f00f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -234,15 +234,18 @@ static void __perf_counter_remove_from_context(void *info) struct perf_counter_context *ctx = counter->ctx; unsigned long flags; + local_irq_save(flags); /* * If this is a task context, we need to check whether it is * the current task context of this cpu. If not it has been * scheduled out before the smp call arrived. */ - if (ctx->task && cpuctx->task_ctx != ctx) + if (ctx->task && cpuctx->task_ctx != ctx) { + local_irq_restore(flags); return; + } - spin_lock_irqsave(&ctx->lock, flags); + spin_lock(&ctx->lock); /* * Protect the list operation against NMI by disabling the * counters on a global level. @@ -382,14 +385,17 @@ static void __perf_counter_disable(void *info) struct perf_counter_context *ctx = counter->ctx; unsigned long flags; + local_irq_save(flags); /* * If this is a per-task counter, need to check whether this * counter's task is the current task on this cpu. */ - if (ctx->task && cpuctx->task_ctx != ctx) + if (ctx->task && cpuctx->task_ctx != ctx) { + local_irq_restore(flags); return; + } - spin_lock_irqsave(&ctx->lock, flags); + spin_lock(&ctx->lock); /* * If the counter is on, turn it off. @@ -615,6 +621,7 @@ static void __perf_install_in_context(void *info) unsigned long flags; int err; + local_irq_save(flags); /* * If this is a task context, we need to check whether it is * the current task context of this cpu. If not it has been @@ -623,12 +630,14 @@ static void __perf_install_in_context(void *info) * on this cpu because it had no counters. */ if (ctx->task && cpuctx->task_ctx != ctx) { - if (cpuctx->task_ctx || ctx->task != current) + if (cpuctx->task_ctx || ctx->task != current) { + local_irq_restore(flags); return; + } cpuctx->task_ctx = ctx; } - spin_lock_irqsave(&ctx->lock, flags); + spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); @@ -745,17 +754,20 @@ static void __perf_counter_enable(void *info) unsigned long flags; int err; + local_irq_save(flags); /* * If this is a per-task counter, need to check whether this * counter's task is the current task on this cpu. */ if (ctx->task && cpuctx->task_ctx != ctx) { - if (cpuctx->task_ctx || ctx->task != current) + if (cpuctx->task_ctx || ctx->task != current) { + local_irq_restore(flags); return; + } cpuctx->task_ctx = ctx; } - spin_lock_irqsave(&ctx->lock, flags); + spin_lock(&ctx->lock); ctx->is_active = 1; update_context_time(ctx); -- cgit v1.2.3 From 012b84dae17126d8b5d159173091eb3db5a2bc43 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sun, 17 May 2009 11:08:41 +0200 Subject: perf_counter: Robustify counter-free logic This fixes a nasty crash and highlights a bug that we were freeing failed-fork() counters incorrectly. (the fix for that will come separately) [ Impact: fix crashes/lockups with inherited counters ] Acked-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index eb346048f00f..616c52426b32 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1004,6 +1004,10 @@ static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) if (!cpuctx->task_ctx) return; + + if (WARN_ON_ONCE(ctx != cpuctx->task_ctx)) + return; + __perf_counter_sched_out(ctx, cpuctx); cpuctx->task_ctx = NULL; } -- cgit v1.2.3 From efb3d17240d80e27508d238809168120fe4b93a4 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 29 May 2009 14:25:58 +0200 Subject: perf_counter: Fix COMM and MMAP events for cpu wide counters Commit a63eaf34ae6 ("perf_counter: Dynamically allocate tasks' perf_counter_context struct") broke COMM and MMAP notification for cpu wide counters by dropping out early if there was no task context, thereby also not iterating the cpu context. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 616c52426b32..58d6d198faa2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2443,9 +2443,9 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) cpuctx = &get_cpu_var(perf_cpu_context); perf_counter_comm_ctx(&cpuctx->ctx, comm_event); + if (cpuctx->task_ctx) + perf_counter_comm_ctx(cpuctx->task_ctx, comm_event); put_cpu_var(perf_cpu_context); - - perf_counter_comm_ctx(current->perf_counter_ctxp, comm_event); } void perf_counter_comm(struct task_struct *task) @@ -2454,8 +2454,6 @@ void perf_counter_comm(struct task_struct *task) if (!atomic_read(&nr_comm_tracking)) return; - if (!current->perf_counter_ctxp) - return; comm_event = (struct perf_comm_event){ .task = task, @@ -2570,10 +2568,10 @@ got_name: cpuctx = &get_cpu_var(perf_cpu_context); perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event); + if (cpuctx->task_ctx) + perf_counter_mmap_ctx(cpuctx->task_ctx, mmap_event); put_cpu_var(perf_cpu_context); - perf_counter_mmap_ctx(current->perf_counter_ctxp, mmap_event); - kfree(buf); } @@ -2584,8 +2582,6 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, if (!atomic_read(&nr_mmap_tracking)) return; - if (!current->perf_counter_ctxp) - return; mmap_event = (struct perf_mmap_event){ .file = file, -- cgit v1.2.3 From 665c2142a94202881a3c11cbaee6506cb10ada2d Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 29 May 2009 14:51:57 +0200 Subject: perf_counter: Clean up task_ctx vs interrupts Remove the local_irq_save() etc.. in routines that are smp function calls, or have IRQs disabled by other means. Then change the COMM, MMAP, and swcounter context iteration to current->perf_counter_ctxp and RCU, since it really doesn't matter which context they iterate, they're all folded. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 82 +++++++++++++++++++++++++++++++-------------------- 1 file changed, 50 insertions(+), 32 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 58d6d198faa2..0c000d305e0e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -232,18 +232,14 @@ static void __perf_counter_remove_from_context(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter *counter = info; struct perf_counter_context *ctx = counter->ctx; - unsigned long flags; - local_irq_save(flags); /* * If this is a task context, we need to check whether it is * the current task context of this cpu. If not it has been * scheduled out before the smp call arrived. */ - if (ctx->task && cpuctx->task_ctx != ctx) { - local_irq_restore(flags); + if (ctx->task && cpuctx->task_ctx != ctx) return; - } spin_lock(&ctx->lock); /* @@ -267,7 +263,7 @@ static void __perf_counter_remove_from_context(void *info) } perf_enable(); - spin_unlock_irqrestore(&ctx->lock, flags); + spin_unlock(&ctx->lock); } @@ -383,17 +379,13 @@ static void __perf_counter_disable(void *info) struct perf_counter *counter = info; struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter_context *ctx = counter->ctx; - unsigned long flags; - local_irq_save(flags); /* * If this is a per-task counter, need to check whether this * counter's task is the current task on this cpu. */ - if (ctx->task && cpuctx->task_ctx != ctx) { - local_irq_restore(flags); + if (ctx->task && cpuctx->task_ctx != ctx) return; - } spin_lock(&ctx->lock); @@ -411,7 +403,7 @@ static void __perf_counter_disable(void *info) counter->state = PERF_COUNTER_STATE_OFF; } - spin_unlock_irqrestore(&ctx->lock, flags); + spin_unlock(&ctx->lock); } /* @@ -618,10 +610,8 @@ static void __perf_install_in_context(void *info) struct perf_counter_context *ctx = counter->ctx; struct perf_counter *leader = counter->group_leader; int cpu = smp_processor_id(); - unsigned long flags; int err; - local_irq_save(flags); /* * If this is a task context, we need to check whether it is * the current task context of this cpu. If not it has been @@ -630,10 +620,8 @@ static void __perf_install_in_context(void *info) * on this cpu because it had no counters. */ if (ctx->task && cpuctx->task_ctx != ctx) { - if (cpuctx->task_ctx || ctx->task != current) { - local_irq_restore(flags); + if (cpuctx->task_ctx || ctx->task != current) return; - } cpuctx->task_ctx = ctx; } @@ -687,7 +675,7 @@ static void __perf_install_in_context(void *info) unlock: perf_enable(); - spin_unlock_irqrestore(&ctx->lock, flags); + spin_unlock(&ctx->lock); } /* @@ -751,19 +739,15 @@ static void __perf_counter_enable(void *info) struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); struct perf_counter_context *ctx = counter->ctx; struct perf_counter *leader = counter->group_leader; - unsigned long flags; int err; - local_irq_save(flags); /* * If this is a per-task counter, need to check whether this * counter's task is the current task on this cpu. */ if (ctx->task && cpuctx->task_ctx != ctx) { - if (cpuctx->task_ctx || ctx->task != current) { - local_irq_restore(flags); + if (cpuctx->task_ctx || ctx->task != current) return; - } cpuctx->task_ctx = ctx; } @@ -811,7 +795,7 @@ static void __perf_counter_enable(void *info) } unlock: - spin_unlock_irqrestore(&ctx->lock, flags); + spin_unlock(&ctx->lock); } /* @@ -981,6 +965,10 @@ void perf_counter_task_sched_out(struct task_struct *task, spin_lock(&ctx->lock); spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); if (context_equiv(ctx, next_ctx)) { + /* + * XXX do we need a memory barrier of sorts + * wrt to rcu_dereference() of perf_counter_ctxp + */ task->perf_counter_ctxp = next_ctx; next->perf_counter_ctxp = ctx; ctx->task = next; @@ -998,6 +986,9 @@ void perf_counter_task_sched_out(struct task_struct *task, } } +/* + * Called with IRQs disabled + */ static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) { struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); @@ -1012,6 +1003,9 @@ static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) cpuctx->task_ctx = NULL; } +/* + * Called with IRQs disabled + */ static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx) { __perf_counter_sched_out(&cpuctx->ctx, cpuctx); @@ -2431,6 +2425,7 @@ static void perf_counter_comm_ctx(struct perf_counter_context *ctx, static void perf_counter_comm_event(struct perf_comm_event *comm_event) { struct perf_cpu_context *cpuctx; + struct perf_counter_context *ctx; unsigned int size; char *comm = comm_event->task->comm; @@ -2443,9 +2438,17 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event) cpuctx = &get_cpu_var(perf_cpu_context); perf_counter_comm_ctx(&cpuctx->ctx, comm_event); - if (cpuctx->task_ctx) - perf_counter_comm_ctx(cpuctx->task_ctx, comm_event); put_cpu_var(perf_cpu_context); + + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_counter_ctxp); + if (ctx) + perf_counter_comm_ctx(ctx, comm_event); + rcu_read_unlock(); } void perf_counter_comm(struct task_struct *task) @@ -2536,6 +2539,7 @@ static void perf_counter_mmap_ctx(struct perf_counter_context *ctx, static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) { struct perf_cpu_context *cpuctx; + struct perf_counter_context *ctx; struct file *file = mmap_event->file; unsigned int size; char tmp[16]; @@ -2568,10 +2572,18 @@ got_name: cpuctx = &get_cpu_var(perf_cpu_context); perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event); - if (cpuctx->task_ctx) - perf_counter_mmap_ctx(cpuctx->task_ctx, mmap_event); put_cpu_var(perf_cpu_context); + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_counter_ctxp); + if (ctx) + perf_counter_mmap_ctx(ctx, mmap_event); + rcu_read_unlock(); + kfree(buf); } @@ -2882,6 +2894,7 @@ static void __perf_swcounter_event(enum perf_event_types type, u32 event, { struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); int *recursion = perf_swcounter_recursion_context(cpuctx); + struct perf_counter_context *ctx; if (*recursion) goto out; @@ -2891,10 +2904,15 @@ static void __perf_swcounter_event(enum perf_event_types type, u32 event, perf_swcounter_ctx_event(&cpuctx->ctx, type, event, nr, nmi, regs, addr); - if (cpuctx->task_ctx) { - perf_swcounter_ctx_event(cpuctx->task_ctx, type, event, - nr, nmi, regs, addr); - } + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_counter_ctxp); + if (ctx) + perf_swcounter_ctx_event(ctx, type, event, nr, nmi, regs, addr); + rcu_read_unlock(); barrier(); (*recursion)--; -- cgit v1.2.3 From bbbee90829304d156c12b171c0ac7e6e1aba8b90 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 29 May 2009 14:25:58 +0200 Subject: perf_counter: Ammend cleanup in fork() fail When fork() fails we cannot use perf_counter_exit_task() since that assumes to operate on current. Write a new helper that cleans up unused/clean contexts. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/fork.c | 2 +- kernel/perf_counter.c | 43 ++++++++++++++++++++++++++++++++++++++++--- 2 files changed, 41 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index c07c3335ceac..23bf757ed321 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1298,7 +1298,7 @@ bad_fork_cleanup_semundo: bad_fork_cleanup_audit: audit_free(p); bad_fork_cleanup_policy: - perf_counter_exit_task(p); + perf_counter_free_task(p); #ifdef CONFIG_NUMA mpol_put(p->mempolicy); bad_fork_cleanup_cgroup: diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0c000d305e0e..79c3f26541d3 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3538,8 +3538,7 @@ static void sync_child_counter(struct perf_counter *child_counter, } static void -__perf_counter_exit_task(struct task_struct *child, - struct perf_counter *child_counter, +__perf_counter_exit_task(struct perf_counter *child_counter, struct perf_counter_context *child_ctx) { struct perf_counter *parent_counter; @@ -3605,7 +3604,7 @@ void perf_counter_exit_task(struct task_struct *child) again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, list_entry) - __perf_counter_exit_task(child, child_counter, child_ctx); + __perf_counter_exit_task(child_counter, child_ctx); /* * If the last counter was a group counter, it will have appended all @@ -3620,6 +3619,44 @@ again: put_ctx(child_ctx); } +/* + * free an unexposed, unused context as created by inheritance by + * init_task below, used by fork() in case of fail. + */ +void perf_counter_free_task(struct task_struct *task) +{ + struct perf_counter_context *ctx = task->perf_counter_ctxp; + struct perf_counter *counter, *tmp; + + if (!ctx) + return; + + mutex_lock(&ctx->mutex); +again: + list_for_each_entry_safe(counter, tmp, &ctx->counter_list, list_entry) { + struct perf_counter *parent = counter->parent; + + if (WARN_ON_ONCE(!parent)) + continue; + + mutex_lock(&parent->child_mutex); + list_del_init(&counter->child_list); + mutex_unlock(&parent->child_mutex); + + fput(parent->filp); + + list_del_counter(counter, ctx); + free_counter(counter); + } + + if (!list_empty(&ctx->counter_list)) + goto again; + + mutex_unlock(&ctx->mutex); + + put_ctx(ctx); +} + /* * Initialize the perf_counter context in task_struct */ -- cgit v1.2.3 From 25346b93ca079080c9cb23331db5c4f6404e8530 Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 1 Jun 2009 17:48:12 +1000 Subject: perf_counter: Provide functions for locking and pinning the context for a task This abstracts out the code for locking the context associated with a task. Because the context might get transferred from one task to another concurrently, we have to check after locking the context that it is still the right context for the task and retry if not. This was open-coded in find_get_context() and perf_counter_init_task(). This adds a further function for pinning the context for a task, i.e. marking it so it can't be transferred to another task. This adds a 'pin_count' field to struct perf_counter_context to indicate that a context is pinned, instead of the previous method of setting the parent_gen count to all 1s. Pinning the context with a pin_count is easier to undo and doesn't require saving the parent_gen value. This also adds a perf_unpin_context() to undo the effect of perf_pin_task_context() and changes perf_counter_init_task to use it. Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <18979.34748.755674.596386@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 128 +++++++++++++++++++++++++++++--------------------- 1 file changed, 74 insertions(+), 54 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 79c3f26541d3..da8dfef4b472 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -122,6 +122,69 @@ static void put_ctx(struct perf_counter_context *ctx) } } +/* + * Get the perf_counter_context for a task and lock it. + * This has to cope with with the fact that until it is locked, + * the context could get moved to another task. + */ +static struct perf_counter_context *perf_lock_task_context( + struct task_struct *task, unsigned long *flags) +{ + struct perf_counter_context *ctx; + + rcu_read_lock(); + retry: + ctx = rcu_dereference(task->perf_counter_ctxp); + if (ctx) { + /* + * If this context is a clone of another, it might + * get swapped for another underneath us by + * perf_counter_task_sched_out, though the + * rcu_read_lock() protects us from any context + * getting freed. Lock the context and check if it + * got swapped before we could get the lock, and retry + * if so. If we locked the right context, then it + * can't get swapped on us any more. + */ + spin_lock_irqsave(&ctx->lock, *flags); + if (ctx != rcu_dereference(task->perf_counter_ctxp)) { + spin_unlock_irqrestore(&ctx->lock, *flags); + goto retry; + } + } + rcu_read_unlock(); + return ctx; +} + +/* + * Get the context for a task and increment its pin_count so it + * can't get swapped to another task. This also increments its + * reference count so that the context can't get freed. + */ +static struct perf_counter_context *perf_pin_task_context(struct task_struct *task) +{ + struct perf_counter_context *ctx; + unsigned long flags; + + ctx = perf_lock_task_context(task, &flags); + if (ctx) { + ++ctx->pin_count; + get_ctx(ctx); + spin_unlock_irqrestore(&ctx->lock, flags); + } + return ctx; +} + +static void perf_unpin_context(struct perf_counter_context *ctx) +{ + unsigned long flags; + + spin_lock_irqsave(&ctx->lock, flags); + --ctx->pin_count; + spin_unlock_irqrestore(&ctx->lock, flags); + put_ctx(ctx); +} + /* * Add a counter from the lists for its context. * Must be called with ctx->mutex and ctx->lock held. @@ -916,7 +979,7 @@ static int context_equiv(struct perf_counter_context *ctx1, { return ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx && ctx1->parent_gen == ctx2->parent_gen - && ctx1->parent_gen != ~0ull; + && !ctx1->pin_count && !ctx2->pin_count; } /* @@ -1271,6 +1334,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) struct perf_counter_context *ctx; struct perf_counter_context *parent_ctx; struct task_struct *task; + unsigned long flags; int err; /* @@ -1323,28 +1387,9 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) if (!ptrace_may_access(task, PTRACE_MODE_READ)) goto errout; - retry_lock: - rcu_read_lock(); retry: - ctx = rcu_dereference(task->perf_counter_ctxp); + ctx = perf_lock_task_context(task, &flags); if (ctx) { - /* - * If this context is a clone of another, it might - * get swapped for another underneath us by - * perf_counter_task_sched_out, though the - * rcu_read_lock() protects us from any context - * getting freed. Lock the context and check if it - * got swapped before we could get the lock, and retry - * if so. If we locked the right context, then it - * can't get swapped on us any more and we can - * unclone it if necessary. - * Once it's not a clone things will be stable. - */ - spin_lock_irq(&ctx->lock); - if (ctx != rcu_dereference(task->perf_counter_ctxp)) { - spin_unlock_irq(&ctx->lock); - goto retry; - } parent_ctx = ctx->parent_ctx; if (parent_ctx) { put_ctx(parent_ctx); @@ -1355,9 +1400,8 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) * this context won't get freed if the task exits. */ get_ctx(ctx); - spin_unlock_irq(&ctx->lock); + spin_unlock_irqrestore(&ctx->lock, flags); } - rcu_read_unlock(); if (!ctx) { ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); @@ -1372,7 +1416,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) * the context they set. */ kfree(ctx); - goto retry_lock; + goto retry; } get_task_struct(task); } @@ -3667,7 +3711,6 @@ int perf_counter_init_task(struct task_struct *child) struct perf_counter *counter; struct task_struct *parent = current; int inherited_all = 1; - u64 cloned_gen; int ret = 0; child->perf_counter_ctxp = NULL; @@ -3693,32 +3736,17 @@ int perf_counter_init_task(struct task_struct *child) get_task_struct(child); /* - * If the parent's context is a clone, temporarily set its - * parent_gen to an impossible value (all 1s) so it won't get - * swapped under us. The rcu_read_lock makes sure that - * parent_ctx continues to exist even if it gets swapped to - * another process and then freed while we are trying to get - * its lock. + * If the parent's context is a clone, pin it so it won't get + * swapped under us. */ - rcu_read_lock(); - retry: - parent_ctx = rcu_dereference(parent->perf_counter_ctxp); + parent_ctx = perf_pin_task_context(parent); + /* * No need to check if parent_ctx != NULL here; since we saw * it non-NULL earlier, the only reason for it to become NULL * is if we exit, and since we're currently in the middle of * a fork we can't be exiting at the same time. */ - spin_lock_irq(&parent_ctx->lock); - if (parent_ctx != rcu_dereference(parent->perf_counter_ctxp)) { - spin_unlock_irq(&parent_ctx->lock); - goto retry; - } - cloned_gen = parent_ctx->parent_gen; - if (parent_ctx->parent_ctx) - parent_ctx->parent_gen = ~0ull; - spin_unlock_irq(&parent_ctx->lock); - rcu_read_unlock(); /* * Lock the parent list. No need to lock the child - not PID @@ -3759,7 +3787,7 @@ int perf_counter_init_task(struct task_struct *child) cloned_ctx = rcu_dereference(parent_ctx->parent_ctx); if (cloned_ctx) { child_ctx->parent_ctx = cloned_ctx; - child_ctx->parent_gen = cloned_gen; + child_ctx->parent_gen = parent_ctx->parent_gen; } else { child_ctx->parent_ctx = parent_ctx; child_ctx->parent_gen = parent_ctx->generation; @@ -3769,15 +3797,7 @@ int perf_counter_init_task(struct task_struct *child) mutex_unlock(&parent_ctx->mutex); - /* - * Restore the clone status of the parent. - */ - if (parent_ctx->parent_ctx) { - spin_lock_irq(&parent_ctx->lock); - if (parent_ctx->parent_ctx) - parent_ctx->parent_gen = cloned_gen; - spin_unlock_irq(&parent_ctx->lock); - } + perf_unpin_context(parent_ctx); return ret; } -- cgit v1.2.3 From 880ca15adf2392770a68047e7a98e076ff4d21da Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 1 Jun 2009 17:49:14 +1000 Subject: perf_counter: Allow software counters to count while task is not running This changes perf_swcounter_match() so that per-task software counters can count events that occur while their associated task is not running. This will allow us to use the generic software counter code for counting task migrations, which can occur while the task is not scheduled in. To do this, we have to distinguish between the situations where the counter is inactive because its task has been scheduled out, and those where the counter is inactive because it is part of a group that was not able to go on the PMU. In the former case we want the counter to count, but not in the latter case. If the context is active, we have the latter case. If the context is inactive then we need to know whether the counter was counting when the context was last active, which we can determine by comparing its ->tstamp_stopped timestamp with the context's timestamp. This also folds three checks in perf_swcounter_match, checking perf_event_raw(), perf_event_type() and perf_event_id() individually, into a single 64-bit comparison on counter->hw_event.config, as an optimization. Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <18979.34810.259718.955621@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 48 ++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 42 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index da8dfef4b472..ff8b4636f845 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2867,20 +2867,56 @@ static void perf_swcounter_overflow(struct perf_counter *counter, } +static int perf_swcounter_is_counting(struct perf_counter *counter) +{ + struct perf_counter_context *ctx; + unsigned long flags; + int count; + + if (counter->state == PERF_COUNTER_STATE_ACTIVE) + return 1; + + if (counter->state != PERF_COUNTER_STATE_INACTIVE) + return 0; + + /* + * If the counter is inactive, it could be just because + * its task is scheduled out, or because it's in a group + * which could not go on the PMU. We want to count in + * the first case but not the second. If the context is + * currently active then an inactive software counter must + * be the second case. If it's not currently active then + * we need to know whether the counter was active when the + * context was last active, which we can determine by + * comparing counter->tstamp_stopped with ctx->time. + * + * We are within an RCU read-side critical section, + * which protects the existence of *ctx. + */ + ctx = counter->ctx; + spin_lock_irqsave(&ctx->lock, flags); + count = 1; + /* Re-check state now we have the lock */ + if (counter->state < PERF_COUNTER_STATE_INACTIVE || + counter->ctx->is_active || + counter->tstamp_stopped < ctx->time) + count = 0; + spin_unlock_irqrestore(&ctx->lock, flags); + return count; +} + static int perf_swcounter_match(struct perf_counter *counter, enum perf_event_types type, u32 event, struct pt_regs *regs) { - if (counter->state != PERF_COUNTER_STATE_ACTIVE) - return 0; + u64 event_config; - if (perf_event_raw(&counter->hw_event)) - return 0; + event_config = ((u64) type << PERF_COUNTER_TYPE_SHIFT) | event; - if (perf_event_type(&counter->hw_event) != type) + if (!perf_swcounter_is_counting(counter)) return 0; - if (perf_event_id(&counter->hw_event) != event) + if (counter->hw_event.config != event_config) return 0; if (counter->hw_event.exclude_user && user_mode(regs)) -- cgit v1.2.3 From 22a4f650d686eeaac3629dae1c4294381485efdf Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 1 Jun 2009 10:13:37 +0200 Subject: perf_counter: Tidy up style details - whitespace fixlets - make local variable definitions more consistent [ Impact: cleanup ] Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 39 +++++++++++++++++++++------------------ 1 file changed, 21 insertions(+), 18 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ff8b4636f845..df319c48c52b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -16,8 +16,9 @@ #include #include #include -#include +#include #include +#include #include #include #include @@ -26,7 +27,6 @@ #include #include #include -#include #include @@ -65,7 +65,9 @@ void __weak hw_perf_disable(void) { barrier(); } void __weak hw_perf_enable(void) { barrier(); } void __weak hw_perf_counter_setup(int cpu) { barrier(); } -int __weak hw_perf_group_sched_in(struct perf_counter *group_leader, + +int __weak +hw_perf_group_sched_in(struct perf_counter *group_leader, struct perf_cpu_context *cpuctx, struct perf_counter_context *ctx, int cpu) { @@ -127,8 +129,8 @@ static void put_ctx(struct perf_counter_context *ctx) * This has to cope with with the fact that until it is locked, * the context could get moved to another task. */ -static struct perf_counter_context *perf_lock_task_context( - struct task_struct *task, unsigned long *flags) +static struct perf_counter_context * +perf_lock_task_context(struct task_struct *task, unsigned long *flags) { struct perf_counter_context *ctx; @@ -1330,9 +1332,9 @@ __perf_counter_init_context(struct perf_counter_context *ctx, static struct perf_counter_context *find_get_context(pid_t pid, int cpu) { - struct perf_cpu_context *cpuctx; - struct perf_counter_context *ctx; struct perf_counter_context *parent_ctx; + struct perf_counter_context *ctx; + struct perf_cpu_context *cpuctx; struct task_struct *task; unsigned long flags; int err; @@ -1664,8 +1666,8 @@ int perf_counter_task_disable(void) */ void perf_counter_update_userpage(struct perf_counter *counter) { - struct perf_mmap_data *data; struct perf_counter_mmap_page *userpg; + struct perf_mmap_data *data; rcu_read_lock(); data = rcu_dereference(counter->data); @@ -1769,10 +1771,11 @@ fail: static void __perf_mmap_data_free(struct rcu_head *rcu_head) { - struct perf_mmap_data *data = container_of(rcu_head, - struct perf_mmap_data, rcu_head); + struct perf_mmap_data *data; int i; + data = container_of(rcu_head, struct perf_mmap_data, rcu_head); + free_page((unsigned long)data->user_page); for (i = 0; i < data->nr_pages; i++) free_page((unsigned long)data->data_pages[i]); @@ -1801,8 +1804,7 @@ static void perf_mmap_close(struct vm_area_struct *vma) struct perf_counter *counter = vma->vm_file->private_data; WARN_ON_ONCE(counter->ctx->parent_ctx); - if (atomic_dec_and_mutex_lock(&counter->mmap_count, - &counter->mmap_mutex)) { + if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { struct user_struct *user = current_user(); atomic_long_sub(counter->data->nr_pages + 1, &user->locked_vm); @@ -1821,11 +1823,11 @@ static struct vm_operations_struct perf_mmap_vmops = { static int perf_mmap(struct file *file, struct vm_area_struct *vma) { struct perf_counter *counter = file->private_data; + unsigned long user_locked, user_lock_limit; struct user_struct *user = current_user(); + unsigned long locked, lock_limit; unsigned long vma_size; unsigned long nr_pages; - unsigned long user_locked, user_lock_limit; - unsigned long locked, lock_limit; long user_extra, extra; int ret = 0; @@ -1900,8 +1902,8 @@ unlock: static int perf_fasync(int fd, struct file *filp, int on) { - struct perf_counter *counter = filp->private_data; struct inode *inode = filp->f_path.dentry->d_inode; + struct perf_counter *counter = filp->private_data; int retval; mutex_lock(&inode->i_mutex); @@ -2412,8 +2414,8 @@ static void perf_counter_output(struct perf_counter *counter, */ struct perf_comm_event { - struct task_struct *task; - char *comm; + struct task_struct *task; + char *comm; int comm_size; struct { @@ -2932,6 +2934,7 @@ static void perf_swcounter_add(struct perf_counter *counter, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { int neg = atomic64_add_negative(nr, &counter->hw.count); + if (counter->hw.irq_period && !neg) perf_swcounter_overflow(counter, nmi, regs, addr); } @@ -3526,7 +3529,7 @@ inherit_counter(struct perf_counter *parent_counter, /* * Make the child state follow the state of the parent counter, * not its hw_event.disabled bit. We hold the parent's mutex, - * so we won't race with perf_counter_{en,dis}able_family. + * so we won't race with perf_counter_{en, dis}able_family. */ if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE) child_counter->state = PERF_COUNTER_STATE_INACTIVE; -- cgit v1.2.3 From 2af15d6a44b871ad4c2a651302374cde8f335480 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Thu, 28 May 2009 13:37:24 -0400 Subject: ftrace: add kernel command line function filtering When using ftrace=function on the command line to trace functions on boot up, one can not filter out functions that are commonly called. This patch adds two new ftrace command line commands. ftrace_notrace=function-list ftrace_filter=function-list Where function-list is a comma separated list of functions to filter. The ftrace_notrace will make the functions listed not be included in the function tracing, and ftrace_filter will only trace the functions listed. These two act the same as the debugfs/tracing/set_ftrace_notrace and debugfs/tracing/set_ftrace_filter respectively. The simple glob expressions that are allowed by the filter files can also be used by the command line interface. ftrace_notrace=rcu*,*lock,*spin* Will not trace any function that starts with rcu, ends with lock, or has the word spin in it. Note, if the self tests are enabled, they may interfere with the filtering set by the command lines. Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 42 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 42 insertions(+) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 140699a9a8a7..2074e5b7766b 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -32,6 +32,7 @@ #include #include +#include #include "trace_output.h" #include "trace_stat.h" @@ -2369,6 +2370,45 @@ void ftrace_set_notrace(unsigned char *buf, int len, int reset) ftrace_set_regex(buf, len, reset, 0); } +/* + * command line interface to allow users to set filters on boot up. + */ +#define FTRACE_FILTER_SIZE COMMAND_LINE_SIZE +static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata; +static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata; + +static int __init set_ftrace_notrace(char *str) +{ + strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE); + return 1; +} +__setup("ftrace_notrace=", set_ftrace_notrace); + +static int __init set_ftrace_filter(char *str) +{ + strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE); + return 1; +} +__setup("ftrace_filter=", set_ftrace_filter); + +static void __init set_ftrace_early_filter(char *buf, int enable) +{ + char *func; + + while (buf) { + func = strsep(&buf, ","); + ftrace_set_regex(func, strlen(func), 0, enable); + } +} + +static void __init set_ftrace_early_filters(void) +{ + if (ftrace_filter_buf[0]) + set_ftrace_early_filter(ftrace_filter_buf, 1); + if (ftrace_notrace_buf[0]) + set_ftrace_early_filter(ftrace_notrace_buf, 0); +} + static int ftrace_regex_release(struct inode *inode, struct file *file, int enable) { @@ -2829,6 +2869,8 @@ void __init ftrace_init(void) if (ret) pr_warning("Failed to register trace ftrace module notifier\n"); + set_ftrace_early_filters(); + return; failed: ftrace_disabled = 1; -- cgit v1.2.3 From 5e0a093910876882f91f1d4b8a1635a099e6c7ba Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Thu, 28 May 2009 15:50:13 -0400 Subject: tracing: fix config options to not show when automatically selected There are two options that are selected by all tracers, but we want to have those options available when no tracer is selected. These are The event tracer and sched switch tracer. The are enabled by all tracers, but if a tracer is not selected we want the options to appear. All tracers including them select TRACING. Thus what we would like to do is: config EVENT_TRACER bool "prompt" depends on TRACING select TRACING But that gives us a bug in the kbuild system since we just created a circular dependency. We only want the prompt to show when TRACING is off. This patch adds GENERIC_TRACER that all tracers will select instead of TRACING. The two options (sched switch and event tracer) will select TRACING directly and depend on !GENERIC_TRACER. This solves the cicular dependency. [ Impact: hide options that are selected by default ] Signed-off-by: Steven Rostedt --- kernel/trace/Kconfig | 43 ++++++++++++++++++++++++++++--------------- 1 file changed, 28 insertions(+), 15 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index a508b9d2adb8..6e55cc3ac49d 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -56,6 +56,13 @@ config CONTEXT_SWITCH_TRACER select MARKERS bool +# All tracer options should select GENERIC_TRACER. For those options that are +# enabled by all tracers (context switch and event tracer) they select TRACING. +# This allows those options to appear when no other tracer is selected. But the +# options do not appear when something else selects it. We need the two options +# GENERIC_TRACER and TRACING to avoid circular dependencies to accomplish the +# hidding of the automatic options options. + config TRACING bool select DEBUG_FS @@ -66,6 +73,10 @@ config TRACING select BINARY_PRINTF select EVENT_TRACING +config GENERIC_TRACER + bool + select TRACING + # # Minimum requirements an architecture has to meet for us to # be able to offer generic tracing facilities: @@ -95,7 +106,7 @@ config FUNCTION_TRACER depends on HAVE_FUNCTION_TRACER select FRAME_POINTER select KALLSYMS - select TRACING + select GENERIC_TRACER select CONTEXT_SWITCH_TRACER help Enable the kernel to trace every kernel function. This is done @@ -126,7 +137,7 @@ config IRQSOFF_TRACER depends on TRACE_IRQFLAGS_SUPPORT depends on GENERIC_TIME select TRACE_IRQFLAGS - select TRACING + select GENERIC_TRACER select TRACER_MAX_TRACE help This option measures the time spent in irqs-off critical @@ -147,7 +158,7 @@ config PREEMPT_TRACER default n depends on GENERIC_TIME depends on PREEMPT - select TRACING + select GENERIC_TRACER select TRACER_MAX_TRACE help This option measures the time spent in preemption off critical @@ -166,7 +177,7 @@ config PREEMPT_TRACER config SYSPROF_TRACER bool "Sysprof Tracer" depends on X86 - select TRACING + select GENERIC_TRACER select CONTEXT_SWITCH_TRACER help This tracer provides the trace needed by the 'Sysprof' userspace @@ -174,7 +185,7 @@ config SYSPROF_TRACER config SCHED_TRACER bool "Scheduling Latency Tracer" - select TRACING + select GENERIC_TRACER select CONTEXT_SWITCH_TRACER select TRACER_MAX_TRACE help @@ -183,6 +194,7 @@ config SCHED_TRACER config ENABLE_CONTEXT_SWITCH_TRACER bool "Trace process context switches" + depends on !GENERIC_TRACER select TRACING select CONTEXT_SWITCH_TRACER help @@ -191,6 +203,7 @@ config ENABLE_CONTEXT_SWITCH_TRACER config ENABLE_EVENT_TRACING bool "Trace various events in the kernel" + depends on !GENERIC_TRACER select TRACING help This tracer hooks to various trace points in the kernel @@ -204,14 +217,14 @@ config ENABLE_EVENT_TRACING config FTRACE_SYSCALLS bool "Trace syscalls" depends on HAVE_FTRACE_SYSCALLS - select TRACING + select GENERIC_TRACER select KALLSYMS help Basic tracer to catch the syscall entry and exit events. config BOOT_TRACER bool "Trace boot initcalls" - select TRACING + select GENERIC_TRACER select CONTEXT_SWITCH_TRACER help This tracer helps developers to optimize boot times: it records @@ -228,7 +241,7 @@ config BOOT_TRACER config TRACE_BRANCH_PROFILING bool - select TRACING + select GENERIC_TRACER choice prompt "Branch Profiling" @@ -308,7 +321,7 @@ config BRANCH_TRACER config POWER_TRACER bool "Trace power consumption behavior" depends on X86 - select TRACING + select GENERIC_TRACER help This tracer helps developers to analyze and optimize the kernels power management decisions, specifically the C-state and P-state @@ -342,14 +355,14 @@ config STACK_TRACER config HW_BRANCH_TRACER depends on HAVE_HW_BRANCH_TRACER bool "Trace hw branches" - select TRACING + select GENERIC_TRACER help This tracer records all branches on the system in a circular buffer giving access to the last N branches for each cpu. config KMEMTRACE bool "Trace SLAB allocations" - select TRACING + select GENERIC_TRACER help kmemtrace provides tracing for slab allocator functions, such as kmalloc, kfree, kmem_cache_alloc, kmem_cache_free etc.. Collected @@ -369,7 +382,7 @@ config KMEMTRACE config WORKQUEUE_TRACER bool "Trace workqueues" - select TRACING + select GENERIC_TRACER help The workqueue tracer provides some statistical informations about each cpu workqueue thread such as the number of the @@ -385,7 +398,7 @@ config BLK_DEV_IO_TRACE select RELAY select DEBUG_FS select TRACEPOINTS - select TRACING + select GENERIC_TRACER select STACKTRACE help Say Y here if you want to be able to trace the block layer actions @@ -446,7 +459,7 @@ config FTRACE_SELFTEST config FTRACE_STARTUP_TEST bool "Perform a startup test on ftrace" - depends on TRACING + depends on GENERIC_TRACER select FTRACE_SELFTEST help This option performs a series of startup tests on ftrace. On bootup @@ -457,7 +470,7 @@ config FTRACE_STARTUP_TEST config MMIOTRACE bool "Memory mapped IO tracing" depends on HAVE_MMIOTRACE_SUPPORT && PCI - select TRACING + select GENERIC_TRACER help Mmiotrace traces Memory Mapped I/O access and is meant for debugging and reverse engineering. It is called from the ioremap -- cgit v1.2.3 From 897f17a65389a26509bd0c79a9812d1c9ea8ea6f Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Thu, 28 May 2009 16:31:21 -0400 Subject: tracing: combine the default tracers into one config Both event tracer and sched switch plugin are selected by default by all generic tracers. But if no generic tracer is enabled, their options appear. But ether one of them will select the other, thus it only makes sense to have the default tracers be selected by one option. [ Impact: clean up kconfig menu ] Signed-off-by: Steven Rostedt --- kernel/trace/Kconfig | 19 +++---------------- 1 file changed, 3 insertions(+), 16 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index 6e55cc3ac49d..4a13e5a01ce3 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -192,27 +192,14 @@ config SCHED_TRACER This tracer tracks the latency of the highest priority task to be scheduled in, starting from the point it has woken up. -config ENABLE_CONTEXT_SWITCH_TRACER - bool "Trace process context switches" - depends on !GENERIC_TRACER - select TRACING - select CONTEXT_SWITCH_TRACER - help - This tracer gets called from the context switch and records - all switching of tasks. - -config ENABLE_EVENT_TRACING - bool "Trace various events in the kernel" +config ENABLE_DEFAULT_TRACERS + bool "Trace process context switches and events" depends on !GENERIC_TRACER select TRACING help This tracer hooks to various trace points in the kernel allowing the user to pick and choose which trace point they - want to trace. - - Note, all tracers enable event tracing. This option is - only a convenience to enable event tracing when no other - tracers are selected. + want to trace. It also includes the sched_switch tracer plugin. config FTRACE_SYSCALLS bool "Trace syscalls" -- cgit v1.2.3 From 7fcb7c472f455d1711eb5a7633204dba8800a6d6 Mon Sep 17 00:00:00 2001 From: Li Zefan Date: Mon, 1 Jun 2009 15:35:46 +0800 Subject: tracing/events: introduce __dynamic_array() __string() is limited: - it's a char array, but we may want to define array with other types - a source string should be available, but we may just know the string size We introduce __dynamic_array() to break those limitations, and __string() becomes a wrapper of it. As a side effect, now __get_str() can be used in TP_fast_assign but not only TP_print. Take XFS for example, we have the string length in the dirent, but the string itself is not NULL-terminated, so __dynamic_array() can be used: TRACE_EVENT(xfs_dir2, TP_PROTO(struct xfs_da_args *args), TP_ARGS(args), TP_STRUCT__entry( __field(int, namelen) __dynamic_array(char, name, args->namelen + 1) ... ), TP_fast_assign( char *name = __get_str(name); if (args->namelen) memcpy(name, args->name, args->namelen); name[args->namelen] = '\0'; __entry->namelen = args->namelen; ), TP_printk("name %.*s namelen %d", __entry->namelen ? __get_str(name) : NULL __entry->namelen) ); [ Impact: allow defining dynamic size arrays ] Signed-off-by: Li Zefan LKML-Reference: <4A2384D2.3080403@cn.fujitsu.com> Signed-off-by: Steven Rostedt --- kernel/trace/trace_events_filter.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c index a7430b16d243..db6e54bdb596 100644 --- a/kernel/trace/trace_events_filter.c +++ b/kernel/trace/trace_events_filter.c @@ -478,12 +478,12 @@ enum { static int is_string_field(const char *type) { + if (strstr(type, "__data_loc") && strstr(type, "char")) + return FILTER_DYN_STRING; + if (strchr(type, '[') && strstr(type, "char")) return FILTER_STATIC_STRING; - if (!strcmp(type, "__str_loc")) - return FILTER_DYN_STRING; - return 0; } -- cgit v1.2.3 From ec081ddc3d90aab35bc0de19a358b964978837cf Mon Sep 17 00:00:00 2001 From: Steven Whitehouse Date: Mon, 1 Jun 2009 15:53:35 +0100 Subject: tracing: add exports to use __print_symbolic and __print_flags from a module A patch to allow the use of __print_symbolic and __print_flags from a module. This allows the current GFS2 tracing patch to build. Signed-off-by: Steven Whitehouse LKML-Reference: <1243868015.29604.542.camel@localhost.localdomain> Signed-off-by: Steven Rostedt --- kernel/trace/trace_output.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index c12d95db2f56..0fe3b223f7ed 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -17,6 +17,7 @@ static DECLARE_RWSEM(trace_event_mutex); DEFINE_PER_CPU(struct trace_seq, ftrace_event_seq); +EXPORT_PER_CPU_SYMBOL(ftrace_event_seq); static struct hlist_head event_hash[EVENT_HASHSIZE] __read_mostly; @@ -250,6 +251,7 @@ ftrace_print_flags_seq(struct trace_seq *p, const char *delim, return p->buffer; } +EXPORT_SYMBOL(ftrace_print_flags_seq); const char * ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val, @@ -275,6 +277,7 @@ ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val, return p->buffer; } +EXPORT_SYMBOL(ftrace_print_symbols_seq); #ifdef CONFIG_KRETPROBES static inline const char *kretprobed(const char *name) -- cgit v1.2.3 From 112f38a7e36e9d688b389507136bf3af3e6d159b Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Mon, 1 Jun 2009 15:16:05 -0400 Subject: tracing: make trace pipe recognize latency format flag The trace_pipe did not recognize the latency format flag and would produce different output than the trace file. The problem was partly due that the trace flags in the iterator was not set as well as the trace_pipe zeros out part of the iterator (including the flags) to be able to use the same routines as the trace file. trace_flags of the iterator should not cause any problems when not zeroed out by for trace_pipe. Reported-by: Johannes Berg Signed-off-by: Steven Rostedt --- kernel/trace/trace.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index a3a8a87d7e91..cae34c69752f 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2826,6 +2826,9 @@ static int tracing_open_pipe(struct inode *inode, struct file *filp) /* trace pipe does not show start of buffer */ cpumask_setall(iter->started); + if (trace_flags & TRACE_ITER_LATENCY_FMT) + iter->iter_flags |= TRACE_FILE_LAT_FMT; + iter->cpu_file = cpu_file; iter->tr = &global_trace; mutex_init(&iter->mutex); -- cgit v1.2.3 From 0f6ce3de4ef6ff940308087c49760d068851c1a7 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Mon, 1 Jun 2009 21:51:28 -0400 Subject: ftrace: do not profile functions when disabled A race was found that if one were to enable and disable the function profiler repeatedly, then the system can panic. This was because a profiled function may be preempted just before disabling interrupts. While the profiler is disabled and then reenabled, the preempted function could start again, and access the hash as it is being initialized. This just adds a check in the irq disabled part to check if the profiler is enabled, and if it is not then it will just exit. When the system is disabled, the profile_enabled variable is cleared before calling the unregistering of the function profiler. This unregistering calls stop machine which also acts as a synchronize schedule. [ Impact: fix panic in enabling/disabling function profiler ] Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 2074e5b7766b..d6973dfadb36 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -599,7 +599,7 @@ function_profile_call(unsigned long ip, unsigned long parent_ip) local_irq_save(flags); stat = &__get_cpu_var(ftrace_profile_stats); - if (!stat->hash) + if (!stat->hash || !ftrace_profile_enabled) goto out; rec = ftrace_find_profiled_func(stat, ip); @@ -630,7 +630,7 @@ static void profile_graph_return(struct ftrace_graph_ret *trace) local_irq_save(flags); stat = &__get_cpu_var(ftrace_profile_stats); - if (!stat->hash) + if (!stat->hash || !ftrace_profile_enabled) goto out; calltime = trace->rettime - trace->calltime; @@ -724,6 +724,10 @@ ftrace_profile_write(struct file *filp, const char __user *ubuf, ftrace_profile_enabled = 1; } else { ftrace_profile_enabled = 0; + /* + * unregister_ftrace_profiler calls stop_machine + * so this acts like an synchronize_sched. + */ unregister_ftrace_profiler(); } } -- cgit v1.2.3 From f38b082081bf69a06fffb8b32a175999e2320c5b Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Tue, 2 Jun 2009 21:05:16 +1000 Subject: perf_counter: Initialize per-cpu context earlier on cpu up This arranges for perf_counter's notifier for cpu hotplug operations to be called earlier than the migration notifier in sched.c by increasing its priority to 20, compared to the 10 for the migration notifier. The reason for doing this is that a subsequent commit to convert the cpu migration counter to use the generic swcounter infrastructure will add a call into the perf_counter subsystem when tasks get migrated. Therefore the perf_counter subsystem needs a chance to initialize its per-cpu data for the new cpu before it can get called from the migration code. This also adds a comment to the migration notifier noting that its priority needs to be lower than that of the perf_counter notifier. Signed-off-by: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <18981.1900.792795.836858@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ++++ kernel/sched.c | 6 ++++-- 2 files changed, 8 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index df319c48c52b..8d2653f137e9 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3902,8 +3902,12 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) return NOTIFY_OK; } +/* + * This has to have a higher priority than migration_notifier in sched.c. + */ static struct notifier_block __cpuinitdata perf_cpu_nb = { .notifier_call = perf_cpu_notify, + .priority = 20, }; void __init perf_counter_init(void) diff --git a/kernel/sched.c b/kernel/sched.c index ad079f07c9c8..3226cc132e9f 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -7319,8 +7319,10 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) return NOTIFY_OK; } -/* Register at highest priority so that task migration (migrate_all_tasks) - * happens before everything else. +/* + * Register at high priority so that task migration (migrate_all_tasks) + * happens before everything else. This has to be lower priority than + * the notifier in the perf_counter subsystem, though. */ static struct notifier_block __cpuinitdata migration_notifier = { .notifier_call = migration_call, -- cgit v1.2.3 From 3f731ca60afc29f5bcdb5fd2a04391466313a9ac Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 1 Jun 2009 17:52:30 +1000 Subject: perf_counter: Fix cpu migration counter This fixes the cpu migration software counter to count correctly even when contexts get swapped from one task to another. Previously the cpu migration counts reported by perf stat were bogus, ranging from negative to several thousand for a single "lat_ctx 2 8 32" run. With this patch the cpu migration count reported for "lat_ctx 2 8 32" is almost always between 35 and 44. This fixes the problem by adding a call into the perf_counter code from set_task_cpu when tasks are migrated. This enables us to use the generic swcounter code (with some modifications) for the cpu migration counter. This modifies the swcounter code to allow a NULL regs pointer to be passed in to perf_swcounter_ctx_event() etc. The cpu migration counter does this because there isn't necessarily a pt_regs struct for the task available. In this case, the counter will not have interrupt capability - but the migration counter didn't have interrupt capability before, so this is no loss. Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <18979.35006.819769.416327@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 74 +++++++++++++++------------------------------------ kernel/sched.c | 1 + 2 files changed, 22 insertions(+), 53 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 8d2653f137e9..cd94cf3bf9e2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2921,11 +2921,13 @@ static int perf_swcounter_match(struct perf_counter *counter, if (counter->hw_event.config != event_config) return 0; - if (counter->hw_event.exclude_user && user_mode(regs)) - return 0; + if (regs) { + if (counter->hw_event.exclude_user && user_mode(regs)) + return 0; - if (counter->hw_event.exclude_kernel && !user_mode(regs)) - return 0; + if (counter->hw_event.exclude_kernel && !user_mode(regs)) + return 0; + } return 1; } @@ -2935,7 +2937,7 @@ static void perf_swcounter_add(struct perf_counter *counter, u64 nr, { int neg = atomic64_add_negative(nr, &counter->hw.count); - if (counter->hw.irq_period && !neg) + if (counter->hw.irq_period && !neg && regs) perf_swcounter_overflow(counter, nmi, regs, addr); } @@ -3151,55 +3153,24 @@ static const struct pmu perf_ops_task_clock = { /* * Software counter: cpu migrations */ - -static inline u64 get_cpu_migrations(struct perf_counter *counter) -{ - struct task_struct *curr = counter->ctx->task; - - if (curr) - return curr->se.nr_migrations; - return cpu_nr_migrations(smp_processor_id()); -} - -static void cpu_migrations_perf_counter_update(struct perf_counter *counter) -{ - u64 prev, now; - s64 delta; - - prev = atomic64_read(&counter->hw.prev_count); - now = get_cpu_migrations(counter); - - atomic64_set(&counter->hw.prev_count, now); - - delta = now - prev; - - atomic64_add(delta, &counter->count); -} - -static void cpu_migrations_perf_counter_read(struct perf_counter *counter) +void perf_counter_task_migration(struct task_struct *task, int cpu) { - cpu_migrations_perf_counter_update(counter); -} + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_counter_context *ctx; -static int cpu_migrations_perf_counter_enable(struct perf_counter *counter) -{ - if (counter->prev_state <= PERF_COUNTER_STATE_OFF) - atomic64_set(&counter->hw.prev_count, - get_cpu_migrations(counter)); - return 0; -} + perf_swcounter_ctx_event(&cpuctx->ctx, PERF_TYPE_SOFTWARE, + PERF_COUNT_CPU_MIGRATIONS, + 1, 1, NULL, 0); -static void cpu_migrations_perf_counter_disable(struct perf_counter *counter) -{ - cpu_migrations_perf_counter_update(counter); + ctx = perf_pin_task_context(task); + if (ctx) { + perf_swcounter_ctx_event(ctx, PERF_TYPE_SOFTWARE, + PERF_COUNT_CPU_MIGRATIONS, + 1, 1, NULL, 0); + perf_unpin_context(ctx); + } } -static const struct pmu perf_ops_cpu_migrations = { - .enable = cpu_migrations_perf_counter_enable, - .disable = cpu_migrations_perf_counter_disable, - .read = cpu_migrations_perf_counter_read, -}; - #ifdef CONFIG_EVENT_PROFILE void perf_tpcounter_event(int event_id) { @@ -3272,11 +3243,8 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) case PERF_COUNT_PAGE_FAULTS_MIN: case PERF_COUNT_PAGE_FAULTS_MAJ: case PERF_COUNT_CONTEXT_SWITCHES: - pmu = &perf_ops_generic; - break; case PERF_COUNT_CPU_MIGRATIONS: - if (!counter->hw_event.exclude_kernel) - pmu = &perf_ops_cpu_migrations; + pmu = &perf_ops_generic; break; } diff --git a/kernel/sched.c b/kernel/sched.c index 3226cc132e9f..8d43347a0c0d 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -1977,6 +1977,7 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) if (task_hot(p, old_rq->clock, NULL)) schedstat_inc(p, se.nr_forced2_migrations); #endif + perf_counter_task_migration(p, new_cpu); } p->se.vruntime -= old_cfsrq->min_vruntime - new_cfsrq->min_vruntime; -- cgit v1.2.3 From bf4e0ed3d027ce581be18496036862131b5f32aa Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Mon, 1 Jun 2009 17:53:16 +1000 Subject: perf_counter: Remove unused prev_state field This removes the prev_state field of struct perf_counter since it is now unused. It was only used by the cpu migration counter, which doesn't use it any more. Signed-off-by: Paul Mackerras Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: John Kacur LKML-Reference: <18979.35052.915728.626374@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 ---- 1 file changed, 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index cd94cf3bf9e2..fbed4d28ad7d 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -572,7 +572,6 @@ group_sched_in(struct perf_counter *group_counter, if (ret) return ret < 0 ? ret : 0; - group_counter->prev_state = group_counter->state; if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) return -EAGAIN; @@ -580,7 +579,6 @@ group_sched_in(struct perf_counter *group_counter, * Schedule in siblings as one group (if any): */ list_for_each_entry(counter, &group_counter->sibling_list, list_entry) { - counter->prev_state = counter->state; if (counter_sched_in(counter, cpuctx, ctx, cpu)) { partial_group = counter; goto group_error; @@ -657,7 +655,6 @@ static void add_counter_to_ctx(struct perf_counter *counter, struct perf_counter_context *ctx) { list_add_counter(counter, ctx); - counter->prev_state = PERF_COUNTER_STATE_OFF; counter->tstamp_enabled = ctx->time; counter->tstamp_running = ctx->time; counter->tstamp_stopped = ctx->time; @@ -820,7 +817,6 @@ static void __perf_counter_enable(void *info) ctx->is_active = 1; update_context_time(ctx); - counter->prev_state = counter->state; if (counter->state >= PERF_COUNTER_STATE_INACTIVE) goto unlock; counter->state = PERF_COUNTER_STATE_INACTIVE; -- cgit v1.2.3 From 709e50cf870e61745b39552044aa6c7c38e4f9e0 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 2 Jun 2009 14:13:15 +0200 Subject: perf_counter: Use PID namespaces properly Stop using task_struct::pid and start using PID namespaces. PIDs will be reported in the PID namespace of the monitoring task at the moment of counter creation. Signed-off-by: Peter Zijlstra Cc: Eric W. Biederman Cc: Oleg Nesterov Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 42 ++++++++++++++++++++++++++++++++++-------- 1 file changed, 34 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index fbed4d28ad7d..caa012cfe49a 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1432,6 +1432,8 @@ static void free_counter_rcu(struct rcu_head *head) struct perf_counter *counter; counter = container_of(head, struct perf_counter, rcu_head); + if (counter->ns) + put_pid_ns(counter->ns); kfree(counter); } @@ -2267,6 +2269,28 @@ static void perf_output_end(struct perf_output_handle *handle) rcu_read_unlock(); } +static u32 perf_counter_pid(struct perf_counter *counter, struct task_struct *p) +{ + /* + * only top level counters have the pid namespace they were created in + */ + if (counter->parent) + counter = counter->parent; + + return task_tgid_nr_ns(p, counter->ns); +} + +static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p) +{ + /* + * only top level counters have the pid namespace they were created in + */ + if (counter->parent) + counter = counter->parent; + + return task_pid_nr_ns(p, counter->ns); +} + static void perf_counter_output(struct perf_counter *counter, int nmi, struct pt_regs *regs, u64 addr) { @@ -2303,8 +2327,8 @@ static void perf_counter_output(struct perf_counter *counter, if (record_type & PERF_RECORD_TID) { /* namespace issues */ - tid_entry.pid = current->group_leader->pid; - tid_entry.tid = current->pid; + tid_entry.pid = perf_counter_pid(counter, current); + tid_entry.tid = perf_counter_tid(counter, current); header.type |= PERF_RECORD_TID; header.size += sizeof(tid_entry); @@ -2432,6 +2456,9 @@ static void perf_counter_comm_output(struct perf_counter *counter, if (ret) return; + comm_event->event.pid = perf_counter_pid(counter, comm_event->task); + comm_event->event.tid = perf_counter_tid(counter, comm_event->task); + perf_output_put(&handle, comm_event->event); perf_output_copy(&handle, comm_event->comm, comm_event->comm_size); @@ -2504,8 +2531,6 @@ void perf_counter_comm(struct task_struct *task) .task = task, .event = { .header = { .type = PERF_EVENT_COMM, }, - .pid = task->group_leader->pid, - .tid = task->pid, }, }; @@ -2542,6 +2567,9 @@ static void perf_counter_mmap_output(struct perf_counter *counter, if (ret) return; + mmap_event->event.pid = perf_counter_pid(counter, current); + mmap_event->event.tid = perf_counter_tid(counter, current); + perf_output_put(&handle, mmap_event->event); perf_output_copy(&handle, mmap_event->file_name, mmap_event->file_size); @@ -2641,8 +2669,6 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, .file = file, .event = { .header = { .type = PERF_EVENT_MMAP, }, - .pid = current->group_leader->pid, - .tid = current->pid, .start = addr, .len = len, .pgoff = pgoff, @@ -2664,8 +2690,6 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, .file = file, .event = { .header = { .type = PERF_EVENT_MUNMAP, }, - .pid = current->group_leader->pid, - .tid = current->pid, .start = addr, .len = len, .pgoff = pgoff, @@ -3445,6 +3469,8 @@ SYSCALL_DEFINE5(perf_counter_open, list_add_tail(&counter->owner_entry, ¤t->perf_counter_list); mutex_unlock(¤t->perf_counter_mutex); + counter->ns = get_pid_ns(current->nsproxy->pid_ns); + fput_light(counter_file, fput_needed2); out_fput: -- cgit v1.2.3 From 179c498ae2998461fe436437a74dc29036fc7dcc Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 2 Jun 2009 12:03:19 -0400 Subject: function-graph: only allocate init tasks if it was not already done When the function graph tracer is enabled, it calls the initialization needed for the init tasks that would be called on all created tasks. The problem is that this is called every time the function graph tracer is enabled, and the ret_stack is allocated for the idle tasks each time. Thus, the old ret_stack is lost and a memory leak is created. This is also dangerous because if an interrupt happened on another CPU with the init task and the ret_stack is replaced, we then lose all the return pointers for the interrupt, and a crash would take place. [ Impact: fix memory leak and possible crash due to race ] Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index f1ed080406c3..ebff62ef40be 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -2643,8 +2643,10 @@ static int start_graph_tracing(void) return -ENOMEM; /* The cpu_boot init_task->ret_stack will never be freed */ - for_each_online_cpu(cpu) - ftrace_graph_init_task(idle_task(cpu)); + for_each_online_cpu(cpu) { + if (!idle_task(cpu)->ret_stack) + ftrace_graph_init_task(idle_task(cpu)); + } do { ret = alloc_retstack_tasklist(ret_stack_list); -- cgit v1.2.3 From 82310a3272d5a2a7652f5649ad8a55f58c8f74d9 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 2 Jun 2009 12:26:07 -0400 Subject: function-graph: enable the stack after initialization of other variables The function graph tracer checks if the task_struct has ret_stack defined to know if it is OK or not to use it. The initialization is done for all tasks by one process, but the idle tasks use the same initialization used by new tasks. If an interrupt happens on an idle task that just had the ret_stack created, but before the rest of the initialization took place, then we can corrupt the return address of the functions. This patch moves the setting of the task_struct's ret_stack to after the other variables have been initialized. [ Impact: prevent kernel panic on idle task when starting function graph ] Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 9 +++++++-- kernel/trace/trace_functions_graph.c | 6 ++++++ 2 files changed, 13 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index ebff62ef40be..20e066065eb3 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -2739,15 +2739,20 @@ void unregister_ftrace_graph(void) void ftrace_graph_init_task(struct task_struct *t) { if (atomic_read(&ftrace_graph_active)) { - t->ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH + struct ftrace_ret_stack *ret_stack; + + ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH * sizeof(struct ftrace_ret_stack), GFP_KERNEL); - if (!t->ret_stack) + if (!ret_stack) return; t->curr_ret_stack = -1; atomic_set(&t->tracing_graph_pause, 0); atomic_set(&t->trace_overrun, 0); t->ftrace_timestamp = 0; + /* make curr_ret_stack visable before we add the ret_stack */ + smp_wmb(); + t->ret_stack = ret_stack; } else t->ret_stack = NULL; } diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c index d28687e7b3a7..baeb5fe36108 100644 --- a/kernel/trace/trace_functions_graph.c +++ b/kernel/trace/trace_functions_graph.c @@ -65,6 +65,12 @@ ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth) if (!current->ret_stack) return -EBUSY; + /* + * We must make sure the ret_stack is tested before we read + * anything else. + */ + smp_rmb(); + /* The return trace stack is full */ if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) { atomic_inc(¤t->trace_overrun); -- cgit v1.2.3 From 26c01624a2a40f8a4ddf6449b65c9b1c418d0e72 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 2 Jun 2009 14:01:19 -0400 Subject: function-graph: add memory barriers for accessing task's ret_stack The code that handles the tasks ret_stack allocation for every task assumes that only an interrupt can cause issues (even though interrupts are disabled). In reality, the code is allocating the ret_stack for tasks that may be running on other CPUs and there are not efficient memory barriers to handle this case. [ Impact: prevent crash due to using of uninitialized ret_stack variables ] Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 20e066065eb3..1664d3f33d38 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -2580,12 +2580,12 @@ static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list) } if (t->ret_stack == NULL) { - t->curr_ret_stack = -1; - /* Make sure IRQs see the -1 first: */ - barrier(); - t->ret_stack = ret_stack_list[start++]; atomic_set(&t->tracing_graph_pause, 0); atomic_set(&t->trace_overrun, 0); + t->curr_ret_stack = -1; + /* Make sure the tasks see the -1 first: */ + smp_wmb(); + t->ret_stack = ret_stack_list[start++]; } } while_each_thread(g, t); -- cgit v1.2.3 From 8e5799b1ad2a0567fdfaaf0e91b40efee010f2c1 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 2 Jun 2009 15:08:15 +0200 Subject: perf_counter: Add unique counter id Stephan raised the issue that we currently cannot distinguish between similar counters within a group (PERF_RECORD_GROUP uses the config value as identifier). Therefore, generate a new ID for each counter using a global u64 sequence counter. Reported-by: Stephane Eranian Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index caa012cfe49a..978ecfcc7aaf 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1510,6 +1510,8 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) values[n++] = counter->total_time_running + atomic64_read(&counter->child_total_time_running); + if (counter->hw_event.read_format & PERF_FORMAT_ID) + values[n++] = counter->id; mutex_unlock(&counter->child_mutex); if (count < n * sizeof(u64)) @@ -2303,7 +2305,7 @@ static void perf_counter_output(struct perf_counter *counter, u32 pid, tid; } tid_entry; struct { - u64 event; + u64 id; u64 counter; } group_entry; struct perf_callchain_entry *callchain = NULL; @@ -2416,7 +2418,7 @@ static void perf_counter_output(struct perf_counter *counter, if (sub != counter) sub->pmu->read(sub); - group_entry.event = sub->hw_event.config; + group_entry.id = sub->id; group_entry.counter = atomic64_read(&sub->count); perf_output_put(&handle, group_entry); @@ -3375,6 +3377,8 @@ done: return counter; } +static atomic64_t perf_counter_id; + /** * sys_perf_counter_open - open a performance counter, associate it to a task/cpu * @@ -3470,6 +3474,7 @@ SYSCALL_DEFINE5(perf_counter_open, mutex_unlock(¤t->perf_counter_mutex); counter->ns = get_pid_ns(current->nsproxy->pid_ns); + counter->id = atomic64_inc_return(&perf_counter_id); fput_light(counter_file, fput_needed2); -- cgit v1.2.3 From b23f3325ed465f1bd914384884269af0d106778c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 2 Jun 2009 15:13:03 +0200 Subject: perf_counter: Rename various fields A few renames: s/irq_period/sample_period/ s/irq_freq/sample_freq/ s/PERF_RECORD_/PERF_SAMPLE_/ s/record_type/sample_type/ And change both the new sample_type and read_format to u64. Reported-by: Stephane Eranian Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 104 +++++++++++++++++++++++++------------------------- 1 file changed, 52 insertions(+), 52 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 978ecfcc7aaf..5ecd9981c035 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1186,7 +1186,7 @@ static void perf_log_period(struct perf_counter *counter, u64 period); static void perf_adjust_freq(struct perf_counter_context *ctx) { struct perf_counter *counter; - u64 interrupts, irq_period; + u64 interrupts, sample_period; u64 events, period; s64 delta; @@ -1204,23 +1204,23 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) interrupts = 2*sysctl_perf_counter_limit/HZ; } - if (!counter->hw_event.freq || !counter->hw_event.irq_freq) + if (!counter->hw_event.freq || !counter->hw_event.sample_freq) continue; - events = HZ * interrupts * counter->hw.irq_period; - period = div64_u64(events, counter->hw_event.irq_freq); + events = HZ * interrupts * counter->hw.sample_period; + period = div64_u64(events, counter->hw_event.sample_freq); - delta = (s64)(1 + period - counter->hw.irq_period); + delta = (s64)(1 + period - counter->hw.sample_period); delta >>= 1; - irq_period = counter->hw.irq_period + delta; + sample_period = counter->hw.sample_period + delta; - if (!irq_period) - irq_period = 1; + if (!sample_period) + sample_period = 1; - perf_log_period(counter, irq_period); + perf_log_period(counter, sample_period); - counter->hw.irq_period = irq_period; + counter->hw.sample_period = sample_period; } spin_unlock(&ctx->lock); } @@ -2297,7 +2297,7 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, struct pt_regs *regs, u64 addr) { int ret; - u64 record_type = counter->hw_event.record_type; + u64 sample_type = counter->hw_event.sample_type; struct perf_output_handle handle; struct perf_event_header header; u64 ip; @@ -2321,61 +2321,61 @@ static void perf_counter_output(struct perf_counter *counter, header.misc = PERF_EVENT_MISC_OVERFLOW; header.misc |= perf_misc_flags(regs); - if (record_type & PERF_RECORD_IP) { + if (sample_type & PERF_SAMPLE_IP) { ip = perf_instruction_pointer(regs); - header.type |= PERF_RECORD_IP; + header.type |= PERF_SAMPLE_IP; header.size += sizeof(ip); } - if (record_type & PERF_RECORD_TID) { + if (sample_type & PERF_SAMPLE_TID) { /* namespace issues */ tid_entry.pid = perf_counter_pid(counter, current); tid_entry.tid = perf_counter_tid(counter, current); - header.type |= PERF_RECORD_TID; + header.type |= PERF_SAMPLE_TID; header.size += sizeof(tid_entry); } - if (record_type & PERF_RECORD_TIME) { + if (sample_type & PERF_SAMPLE_TIME) { /* * Maybe do better on x86 and provide cpu_clock_nmi() */ time = sched_clock(); - header.type |= PERF_RECORD_TIME; + header.type |= PERF_SAMPLE_TIME; header.size += sizeof(u64); } - if (record_type & PERF_RECORD_ADDR) { - header.type |= PERF_RECORD_ADDR; + if (sample_type & PERF_SAMPLE_ADDR) { + header.type |= PERF_SAMPLE_ADDR; header.size += sizeof(u64); } - if (record_type & PERF_RECORD_CONFIG) { - header.type |= PERF_RECORD_CONFIG; + if (sample_type & PERF_SAMPLE_CONFIG) { + header.type |= PERF_SAMPLE_CONFIG; header.size += sizeof(u64); } - if (record_type & PERF_RECORD_CPU) { - header.type |= PERF_RECORD_CPU; + if (sample_type & PERF_SAMPLE_CPU) { + header.type |= PERF_SAMPLE_CPU; header.size += sizeof(cpu_entry); cpu_entry.cpu = raw_smp_processor_id(); } - if (record_type & PERF_RECORD_GROUP) { - header.type |= PERF_RECORD_GROUP; + if (sample_type & PERF_SAMPLE_GROUP) { + header.type |= PERF_SAMPLE_GROUP; header.size += sizeof(u64) + counter->nr_siblings * sizeof(group_entry); } - if (record_type & PERF_RECORD_CALLCHAIN) { + if (sample_type & PERF_SAMPLE_CALLCHAIN) { callchain = perf_callchain(regs); if (callchain) { callchain_size = (1 + callchain->nr) * sizeof(u64); - header.type |= PERF_RECORD_CALLCHAIN; + header.type |= PERF_SAMPLE_CALLCHAIN; header.size += callchain_size; } } @@ -2386,28 +2386,28 @@ static void perf_counter_output(struct perf_counter *counter, perf_output_put(&handle, header); - if (record_type & PERF_RECORD_IP) + if (sample_type & PERF_SAMPLE_IP) perf_output_put(&handle, ip); - if (record_type & PERF_RECORD_TID) + if (sample_type & PERF_SAMPLE_TID) perf_output_put(&handle, tid_entry); - if (record_type & PERF_RECORD_TIME) + if (sample_type & PERF_SAMPLE_TIME) perf_output_put(&handle, time); - if (record_type & PERF_RECORD_ADDR) + if (sample_type & PERF_SAMPLE_ADDR) perf_output_put(&handle, addr); - if (record_type & PERF_RECORD_CONFIG) + if (sample_type & PERF_SAMPLE_CONFIG) perf_output_put(&handle, counter->hw_event.config); - if (record_type & PERF_RECORD_CPU) + if (sample_type & PERF_SAMPLE_CPU) perf_output_put(&handle, cpu_entry); /* - * XXX PERF_RECORD_GROUP vs inherited counters seems difficult. + * XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult. */ - if (record_type & PERF_RECORD_GROUP) { + if (sample_type & PERF_SAMPLE_GROUP) { struct perf_counter *leader, *sub; u64 nr = counter->nr_siblings; @@ -2702,7 +2702,7 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, } /* - * Log irq_period changes so that analyzing tools can re-normalize the + * Log sample_period changes so that analyzing tools can re-normalize the * event flow. */ @@ -2725,7 +2725,7 @@ static void perf_log_period(struct perf_counter *counter, u64 period) .period = period, }; - if (counter->hw.irq_period == period) + if (counter->hw.sample_period == period) return; ret = perf_output_begin(&handle, counter, sizeof(freq_event), 0, 0); @@ -2834,7 +2834,7 @@ static void perf_swcounter_set_period(struct perf_counter *counter) { struct hw_perf_counter *hwc = &counter->hw; s64 left = atomic64_read(&hwc->period_left); - s64 period = hwc->irq_period; + s64 period = hwc->sample_period; if (unlikely(left <= -period)) { left = period; @@ -2874,7 +2874,7 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) ret = HRTIMER_NORESTART; } - period = max_t(u64, 10000, counter->hw.irq_period); + period = max_t(u64, 10000, counter->hw.sample_period); hrtimer_forward_now(hrtimer, ns_to_ktime(period)); return ret; @@ -2959,7 +2959,7 @@ static void perf_swcounter_add(struct perf_counter *counter, u64 nr, { int neg = atomic64_add_negative(nr, &counter->hw.count); - if (counter->hw.irq_period && !neg && regs) + if (counter->hw.sample_period && !neg && regs) perf_swcounter_overflow(counter, nmi, regs, addr); } @@ -3080,8 +3080,8 @@ static int cpu_clock_perf_counter_enable(struct perf_counter *counter) atomic64_set(&hwc->prev_count, cpu_clock(cpu)); hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hwc->hrtimer.function = perf_swcounter_hrtimer; - if (hwc->irq_period) { - u64 period = max_t(u64, 10000, hwc->irq_period); + if (hwc->sample_period) { + u64 period = max_t(u64, 10000, hwc->sample_period); __hrtimer_start_range_ns(&hwc->hrtimer, ns_to_ktime(period), 0, HRTIMER_MODE_REL, 0); @@ -3092,7 +3092,7 @@ static int cpu_clock_perf_counter_enable(struct perf_counter *counter) static void cpu_clock_perf_counter_disable(struct perf_counter *counter) { - if (counter->hw.irq_period) + if (counter->hw.sample_period) hrtimer_cancel(&counter->hw.hrtimer); cpu_clock_perf_counter_update(counter); } @@ -3132,8 +3132,8 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) atomic64_set(&hwc->prev_count, now); hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); hwc->hrtimer.function = perf_swcounter_hrtimer; - if (hwc->irq_period) { - u64 period = max_t(u64, 10000, hwc->irq_period); + if (hwc->sample_period) { + u64 period = max_t(u64, 10000, hwc->sample_period); __hrtimer_start_range_ns(&hwc->hrtimer, ns_to_ktime(period), 0, HRTIMER_MODE_REL, 0); @@ -3144,7 +3144,7 @@ static int task_clock_perf_counter_enable(struct perf_counter *counter) static void task_clock_perf_counter_disable(struct perf_counter *counter) { - if (counter->hw.irq_period) + if (counter->hw.sample_period) hrtimer_cancel(&counter->hw.hrtimer); task_clock_perf_counter_update(counter, counter->ctx->time); @@ -3223,7 +3223,7 @@ static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) return NULL; counter->destroy = tp_perf_counter_destroy; - counter->hw.irq_period = counter->hw_event.irq_period; + counter->hw.sample_period = counter->hw_event.sample_period; return &perf_ops_generic; } @@ -3323,15 +3323,15 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, pmu = NULL; hwc = &counter->hw; - if (hw_event->freq && hw_event->irq_freq) - hwc->irq_period = div64_u64(TICK_NSEC, hw_event->irq_freq); + if (hw_event->freq && hw_event->sample_freq) + hwc->sample_period = div64_u64(TICK_NSEC, hw_event->sample_freq); else - hwc->irq_period = hw_event->irq_period; + hwc->sample_period = hw_event->sample_period; /* - * we currently do not support PERF_RECORD_GROUP on inherited counters + * we currently do not support PERF_SAMPLE_GROUP on inherited counters */ - if (hw_event->inherit && (hw_event->record_type & PERF_RECORD_GROUP)) + if (hw_event->inherit && (hw_event->sample_type & PERF_SAMPLE_GROUP)) goto done; if (perf_event_raw(hw_event)) { -- cgit v1.2.3 From 8e3747c13c39246c7e46def7cf495d9d21d4c5f9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 2 Jun 2009 16:16:02 +0200 Subject: perf_counter: Change data head from u32 to u64 Since some people worried that 4G might not be a large enough as an mmap data window, extend it to 64 bit for capable platforms. Reported-by: Stephane Eranian Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 5ecd9981c035..3f11a2bc6c79 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2067,8 +2067,8 @@ __weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) struct perf_output_handle { struct perf_counter *counter; struct perf_mmap_data *data; - unsigned int offset; - unsigned int head; + unsigned long head; + unsigned long offset; int nmi; int overflow; int locked; @@ -2122,7 +2122,8 @@ static void perf_output_lock(struct perf_output_handle *handle) static void perf_output_unlock(struct perf_output_handle *handle) { struct perf_mmap_data *data = handle->data; - int head, cpu; + unsigned long head; + int cpu; data->done_head = data->head; @@ -2135,7 +2136,7 @@ again: * before we publish the new head, matched by a rmb() in userspace when * reading this position. */ - while ((head = atomic_xchg(&data->done_head, 0))) + while ((head = atomic_long_xchg(&data->done_head, 0))) data->user_page->data_head = head; /* @@ -2148,7 +2149,7 @@ again: /* * Therefore we have to validate we did not indeed do so. */ - if (unlikely(atomic_read(&data->done_head))) { + if (unlikely(atomic_long_read(&data->done_head))) { /* * Since we had it locked, we can lock it again. */ @@ -2195,7 +2196,7 @@ static int perf_output_begin(struct perf_output_handle *handle, do { offset = head = atomic_read(&data->head); head += size; - } while (atomic_cmpxchg(&data->head, offset, head) != offset); + } while (atomic_long_cmpxchg(&data->head, offset, head) != offset); handle->offset = offset; handle->head = head; @@ -2246,7 +2247,7 @@ static void perf_output_copy(struct perf_output_handle *handle, * Check we didn't copy past our reservation window, taking the * possible unsigned int wrap into account. */ - WARN_ON_ONCE(((int)(handle->head - handle->offset)) < 0); + WARN_ON_ONCE(((long)(handle->head - handle->offset)) < 0); } #define perf_output_put(handle, x) \ -- cgit v1.2.3 From 08247e31ca79b8f02cce47b7e8120797a8726606 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 2 Jun 2009 16:46:57 +0200 Subject: perf_counter: Add ioctl for changing the sample period/frequency Reported-by: Stephane Eranian Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 41 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 3f11a2bc6c79..abe2f3b6c424 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1604,6 +1604,43 @@ static void perf_counter_for_each(struct perf_counter *counter, mutex_unlock(&counter->child_mutex); } +static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) +{ + struct perf_counter_context *ctx = counter->ctx; + unsigned long size; + int ret = 0; + u64 value; + + if (!counter->hw_event.sample_period) + return -EINVAL; + + size = copy_from_user(&value, arg, sizeof(value)); + if (size != sizeof(value)) + return -EFAULT; + + if (!value) + return -EINVAL; + + spin_lock_irq(&ctx->lock); + if (counter->hw_event.freq) { + if (value > sysctl_perf_counter_limit) { + ret = -EINVAL; + goto unlock; + } + + counter->hw_event.sample_freq = value; + } else { + counter->hw_event.sample_period = value; + counter->hw.sample_period = value; + + perf_log_period(counter, value); + } +unlock: + spin_unlock_irq(&ctx->lock); + + return ret; +} + static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) { struct perf_counter *counter = file->private_data; @@ -1623,6 +1660,10 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) case PERF_COUNTER_IOC_REFRESH: return perf_counter_refresh(counter, arg); + + case PERF_COUNTER_IOC_PERIOD: + return perf_counter_period(counter, (u64 __user *)arg); + default: return -ENOTTY; } -- cgit v1.2.3 From 0d48696f87e3618b0d35bd3e4e9d7c188d51e7de Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Tue, 2 Jun 2009 19:22:16 +0200 Subject: perf_counter: Rename perf_counter_hw_event => perf_counter_attr The structure isn't hw only and when I read event, I think about those things that fall out the other end. Rename the thing. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur Cc: Stephane Eranian LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 116 +++++++++++++++++++++++++------------------------- 1 file changed, 58 insertions(+), 58 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index abe2f3b6c424..317cef78a388 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -260,7 +260,7 @@ counter_sched_out(struct perf_counter *counter, if (!is_software_counter(counter)) cpuctx->active_oncpu--; ctx->nr_active--; - if (counter->hw_event.exclusive || !cpuctx->active_oncpu) + if (counter->attr.exclusive || !cpuctx->active_oncpu) cpuctx->exclusive = 0; } @@ -282,7 +282,7 @@ group_sched_out(struct perf_counter *group_counter, list_for_each_entry(counter, &group_counter->sibling_list, list_entry) counter_sched_out(counter, cpuctx, ctx); - if (group_counter->hw_event.exclusive) + if (group_counter->attr.exclusive) cpuctx->exclusive = 0; } @@ -550,7 +550,7 @@ counter_sched_in(struct perf_counter *counter, cpuctx->active_oncpu++; ctx->nr_active++; - if (counter->hw_event.exclusive) + if (counter->attr.exclusive) cpuctx->exclusive = 1; return 0; @@ -642,7 +642,7 @@ static int group_can_go_on(struct perf_counter *counter, * If this group is exclusive and there are already * counters on the CPU, it can't go on. */ - if (counter->hw_event.exclusive && cpuctx->active_oncpu) + if (counter->attr.exclusive && cpuctx->active_oncpu) return 0; /* * Otherwise, try to add it if all previous groups were able @@ -725,7 +725,7 @@ static void __perf_install_in_context(void *info) */ if (leader != counter) group_sched_out(leader, cpuctx, ctx); - if (leader->hw_event.pinned) { + if (leader->attr.pinned) { update_group_times(leader); leader->state = PERF_COUNTER_STATE_ERROR; } @@ -849,7 +849,7 @@ static void __perf_counter_enable(void *info) */ if (leader != counter) group_sched_out(leader, cpuctx, ctx); - if (leader->hw_event.pinned) { + if (leader->attr.pinned) { update_group_times(leader); leader->state = PERF_COUNTER_STATE_ERROR; } @@ -927,7 +927,7 @@ static int perf_counter_refresh(struct perf_counter *counter, int refresh) /* * not supported on inherited counters */ - if (counter->hw_event.inherit) + if (counter->attr.inherit) return -EINVAL; atomic_add(refresh, &counter->event_limit); @@ -1094,7 +1094,7 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, */ list_for_each_entry(counter, &ctx->counter_list, list_entry) { if (counter->state <= PERF_COUNTER_STATE_OFF || - !counter->hw_event.pinned) + !counter->attr.pinned) continue; if (counter->cpu != -1 && counter->cpu != cpu) continue; @@ -1122,7 +1122,7 @@ __perf_counter_sched_in(struct perf_counter_context *ctx, * ignore pinned counters since we did them already. */ if (counter->state <= PERF_COUNTER_STATE_OFF || - counter->hw_event.pinned) + counter->attr.pinned) continue; /* @@ -1204,11 +1204,11 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) interrupts = 2*sysctl_perf_counter_limit/HZ; } - if (!counter->hw_event.freq || !counter->hw_event.sample_freq) + if (!counter->attr.freq || !counter->attr.sample_freq) continue; events = HZ * interrupts * counter->hw.sample_period; - period = div64_u64(events, counter->hw_event.sample_freq); + period = div64_u64(events, counter->attr.sample_freq); delta = (s64)(1 + period - counter->hw.sample_period); delta >>= 1; @@ -1444,11 +1444,11 @@ static void free_counter(struct perf_counter *counter) perf_pending_sync(counter); atomic_dec(&nr_counters); - if (counter->hw_event.mmap) + if (counter->attr.mmap) atomic_dec(&nr_mmap_tracking); - if (counter->hw_event.munmap) + if (counter->attr.munmap) atomic_dec(&nr_munmap_tracking); - if (counter->hw_event.comm) + if (counter->attr.comm) atomic_dec(&nr_comm_tracking); if (counter->destroy) @@ -1504,13 +1504,13 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) mutex_lock(&counter->child_mutex); values[0] = perf_counter_read(counter); n = 1; - if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) values[n++] = counter->total_time_enabled + atomic64_read(&counter->child_total_time_enabled); - if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) values[n++] = counter->total_time_running + atomic64_read(&counter->child_total_time_running); - if (counter->hw_event.read_format & PERF_FORMAT_ID) + if (counter->attr.read_format & PERF_FORMAT_ID) values[n++] = counter->id; mutex_unlock(&counter->child_mutex); @@ -1611,7 +1611,7 @@ static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) int ret = 0; u64 value; - if (!counter->hw_event.sample_period) + if (!counter->attr.sample_period) return -EINVAL; size = copy_from_user(&value, arg, sizeof(value)); @@ -1622,15 +1622,15 @@ static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) return -EINVAL; spin_lock_irq(&ctx->lock); - if (counter->hw_event.freq) { + if (counter->attr.freq) { if (value > sysctl_perf_counter_limit) { ret = -EINVAL; goto unlock; } - counter->hw_event.sample_freq = value; + counter->attr.sample_freq = value; } else { - counter->hw_event.sample_period = value; + counter->attr.sample_period = value; counter->hw.sample_period = value; perf_log_period(counter, value); @@ -2299,7 +2299,7 @@ static void perf_output_end(struct perf_output_handle *handle) struct perf_counter *counter = handle->counter; struct perf_mmap_data *data = handle->data; - int wakeup_events = counter->hw_event.wakeup_events; + int wakeup_events = counter->attr.wakeup_events; if (handle->overflow && wakeup_events) { int events = atomic_inc_return(&data->events); @@ -2339,7 +2339,7 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, struct pt_regs *regs, u64 addr) { int ret; - u64 sample_type = counter->hw_event.sample_type; + u64 sample_type = counter->attr.sample_type; struct perf_output_handle handle; struct perf_event_header header; u64 ip; @@ -2441,7 +2441,7 @@ static void perf_counter_output(struct perf_counter *counter, perf_output_put(&handle, addr); if (sample_type & PERF_SAMPLE_CONFIG) - perf_output_put(&handle, counter->hw_event.config); + perf_output_put(&handle, counter->attr.config); if (sample_type & PERF_SAMPLE_CPU) perf_output_put(&handle, cpu_entry); @@ -2512,7 +2512,7 @@ static void perf_counter_comm_output(struct perf_counter *counter, static int perf_counter_comm_match(struct perf_counter *counter, struct perf_comm_event *comm_event) { - if (counter->hw_event.comm && + if (counter->attr.comm && comm_event->event.header.type == PERF_EVENT_COMM) return 1; @@ -2623,11 +2623,11 @@ static void perf_counter_mmap_output(struct perf_counter *counter, static int perf_counter_mmap_match(struct perf_counter *counter, struct perf_mmap_event *mmap_event) { - if (counter->hw_event.mmap && + if (counter->attr.mmap && mmap_event->event.header.type == PERF_EVENT_MMAP) return 1; - if (counter->hw_event.munmap && + if (counter->attr.munmap && mmap_event->event.header.type == PERF_EVENT_MUNMAP) return 1; @@ -2907,8 +2907,8 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) * In case we exclude kernel IPs or are somehow not in interrupt * context, provide the next best thing, the user IP. */ - if ((counter->hw_event.exclude_kernel || !regs) && - !counter->hw_event.exclude_user) + if ((counter->attr.exclude_kernel || !regs) && + !counter->attr.exclude_user) regs = task_pt_regs(current); if (regs) { @@ -2982,14 +2982,14 @@ static int perf_swcounter_match(struct perf_counter *counter, if (!perf_swcounter_is_counting(counter)) return 0; - if (counter->hw_event.config != event_config) + if (counter->attr.config != event_config) return 0; if (regs) { - if (counter->hw_event.exclude_user && user_mode(regs)) + if (counter->attr.exclude_user && user_mode(regs)) return 0; - if (counter->hw_event.exclude_kernel && !user_mode(regs)) + if (counter->attr.exclude_kernel && !user_mode(regs)) return 0; } @@ -3252,12 +3252,12 @@ extern void ftrace_profile_disable(int); static void tp_perf_counter_destroy(struct perf_counter *counter) { - ftrace_profile_disable(perf_event_id(&counter->hw_event)); + ftrace_profile_disable(perf_event_id(&counter->attr)); } static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) { - int event_id = perf_event_id(&counter->hw_event); + int event_id = perf_event_id(&counter->attr); int ret; ret = ftrace_profile_enable(event_id); @@ -3265,7 +3265,7 @@ static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) return NULL; counter->destroy = tp_perf_counter_destroy; - counter->hw.sample_period = counter->hw_event.sample_period; + counter->hw.sample_period = counter->attr.sample_period; return &perf_ops_generic; } @@ -3287,7 +3287,7 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) * to be kernel events, and page faults are never hypervisor * events. */ - switch (perf_event_id(&counter->hw_event)) { + switch (perf_event_id(&counter->attr)) { case PERF_COUNT_CPU_CLOCK: pmu = &perf_ops_cpu_clock; @@ -3319,7 +3319,7 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) * Allocate and initialize a counter structure */ static struct perf_counter * -perf_counter_alloc(struct perf_counter_hw_event *hw_event, +perf_counter_alloc(struct perf_counter_attr *attr, int cpu, struct perf_counter_context *ctx, struct perf_counter *group_leader, @@ -3352,36 +3352,36 @@ perf_counter_alloc(struct perf_counter_hw_event *hw_event, mutex_init(&counter->mmap_mutex); counter->cpu = cpu; - counter->hw_event = *hw_event; + counter->attr = *attr; counter->group_leader = group_leader; counter->pmu = NULL; counter->ctx = ctx; counter->oncpu = -1; counter->state = PERF_COUNTER_STATE_INACTIVE; - if (hw_event->disabled) + if (attr->disabled) counter->state = PERF_COUNTER_STATE_OFF; pmu = NULL; hwc = &counter->hw; - if (hw_event->freq && hw_event->sample_freq) - hwc->sample_period = div64_u64(TICK_NSEC, hw_event->sample_freq); + if (attr->freq && attr->sample_freq) + hwc->sample_period = div64_u64(TICK_NSEC, attr->sample_freq); else - hwc->sample_period = hw_event->sample_period; + hwc->sample_period = attr->sample_period; /* * we currently do not support PERF_SAMPLE_GROUP on inherited counters */ - if (hw_event->inherit && (hw_event->sample_type & PERF_SAMPLE_GROUP)) + if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP)) goto done; - if (perf_event_raw(hw_event)) { + if (perf_event_raw(attr)) { pmu = hw_perf_counter_init(counter); goto done; } - switch (perf_event_type(hw_event)) { + switch (perf_event_type(attr)) { case PERF_TYPE_HARDWARE: pmu = hw_perf_counter_init(counter); break; @@ -3409,11 +3409,11 @@ done: counter->pmu = pmu; atomic_inc(&nr_counters); - if (counter->hw_event.mmap) + if (counter->attr.mmap) atomic_inc(&nr_mmap_tracking); - if (counter->hw_event.munmap) + if (counter->attr.munmap) atomic_inc(&nr_munmap_tracking); - if (counter->hw_event.comm) + if (counter->attr.comm) atomic_inc(&nr_comm_tracking); return counter; @@ -3424,17 +3424,17 @@ static atomic64_t perf_counter_id; /** * sys_perf_counter_open - open a performance counter, associate it to a task/cpu * - * @hw_event_uptr: event type attributes for monitoring/sampling + * @attr_uptr: event type attributes for monitoring/sampling * @pid: target pid * @cpu: target cpu * @group_fd: group leader counter fd */ SYSCALL_DEFINE5(perf_counter_open, - const struct perf_counter_hw_event __user *, hw_event_uptr, + const struct perf_counter_attr __user *, attr_uptr, pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) { struct perf_counter *counter, *group_leader; - struct perf_counter_hw_event hw_event; + struct perf_counter_attr attr; struct perf_counter_context *ctx; struct file *counter_file = NULL; struct file *group_file = NULL; @@ -3446,7 +3446,7 @@ SYSCALL_DEFINE5(perf_counter_open, if (flags) return -EINVAL; - if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0) + if (copy_from_user(&attr, attr_uptr, sizeof(attr)) != 0) return -EFAULT; /* @@ -3484,11 +3484,11 @@ SYSCALL_DEFINE5(perf_counter_open, /* * Only a group leader can be exclusive or pinned */ - if (hw_event.exclusive || hw_event.pinned) + if (attr.exclusive || attr.pinned) goto err_put_context; } - counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader, + counter = perf_counter_alloc(&attr, cpu, ctx, group_leader, GFP_KERNEL); ret = PTR_ERR(counter); if (IS_ERR(counter)) @@ -3556,7 +3556,7 @@ inherit_counter(struct perf_counter *parent_counter, if (parent_counter->parent) parent_counter = parent_counter->parent; - child_counter = perf_counter_alloc(&parent_counter->hw_event, + child_counter = perf_counter_alloc(&parent_counter->attr, parent_counter->cpu, child_ctx, group_leader, GFP_KERNEL); if (IS_ERR(child_counter)) @@ -3565,7 +3565,7 @@ inherit_counter(struct perf_counter *parent_counter, /* * Make the child state follow the state of the parent counter, - * not its hw_event.disabled bit. We hold the parent's mutex, + * not its attr.disabled bit. We hold the parent's mutex, * so we won't race with perf_counter_{en, dis}able_family. */ if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE) @@ -3582,7 +3582,7 @@ inherit_counter(struct perf_counter *parent_counter, /* * inherit into child's child as well: */ - child_counter->hw_event.inherit = 1; + child_counter->attr.inherit = 1; /* * Get a reference to the parent filp - we will fput it @@ -3838,7 +3838,7 @@ int perf_counter_init_task(struct task_struct *child) if (counter != counter->group_leader) continue; - if (!counter->hw_event.inherit) { + if (!counter->attr.inherit) { inherited_all = 0; continue; } -- cgit v1.2.3 From f7e8b616ed1cc6f790b82324bce8a2a60295e5c2 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 2 Jun 2009 16:39:48 -0400 Subject: function-graph: move initialization of new tasks up in fork When the function graph tracer is enabled, all new tasks must allocate a ret_stack to place the return address of functions. This is because the function graph tracer will replace the real return address with a call to the tracing of the exit function. This initialization happens in fork, but it happens too late. If fork fails, then it will call free_task and that calls the freeing of this ret_stack. But before initialization happens, the new (failed) task points to its parents ret_stack. If a fork failure happens during the function trace, it would be catastrophic for the parent. Also, there's no need to call ftrace_graph_exit_task from fork, since it is called by free_task which fork calls on failure. [ Impact: prevent crash during failed fork running function graph tracer ] Signed-off-by: Steven Rostedt --- kernel/fork.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index b9e2edd00726..c4b1e35c430b 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -982,6 +982,8 @@ static struct task_struct *copy_process(unsigned long clone_flags, if (!p) goto fork_out; + ftrace_graph_init_task(p); + rt_mutex_init_task(p); #ifdef CONFIG_PROVE_LOCKING @@ -1131,8 +1133,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, } } - ftrace_graph_init_task(p); - p->pid = pid_nr(pid); p->tgid = p->pid; if (clone_flags & CLONE_THREAD) @@ -1141,7 +1141,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, if (current->nsproxy != p->nsproxy) { retval = ns_cgroup_clone(p, pid); if (retval) - goto bad_fork_free_graph; + goto bad_fork_free_pid; } p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; @@ -1233,7 +1233,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, spin_unlock(¤t->sighand->siglock); write_unlock_irq(&tasklist_lock); retval = -ERESTARTNOINTR; - goto bad_fork_free_graph; + goto bad_fork_free_pid; } if (clone_flags & CLONE_THREAD) { @@ -1268,8 +1268,6 @@ static struct task_struct *copy_process(unsigned long clone_flags, cgroup_post_fork(p); return p; -bad_fork_free_graph: - ftrace_graph_exit_task(p); bad_fork_free_pid: if (pid != &init_struct_pid) free_pid(pid); -- cgit v1.2.3 From 84047e360af0394ac5861d433f26bbcf30f77dd1 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 2 Jun 2009 16:51:55 -0400 Subject: function-graph: always initialize task ret_stack On creating a new task while running the function graph tracer, if we fail to allocate the ret_stack, and then fail the fork, the code will free the parent ret_stack. This is because the child duplicated the parent and currently points to the parent's ret_stack. This patch always initializes the task's ret_stack to NULL. [ Impact: prevent crash of parent on low memory during fork ] Signed-off-by: Steven Rostedt --- kernel/trace/ftrace.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 1664d3f33d38..bb081f37cac7 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -2738,6 +2738,9 @@ void unregister_ftrace_graph(void) /* Allocate a return stack for newly created task */ void ftrace_graph_init_task(struct task_struct *t) { + /* Make sure we do not use the parent ret_stack */ + t->ret_stack = NULL; + if (atomic_read(&ftrace_graph_active)) { struct ftrace_ret_stack *ret_stack; @@ -2753,8 +2756,7 @@ void ftrace_graph_init_task(struct task_struct *t) /* make curr_ret_stack visable before we add the ret_stack */ smp_wmb(); t->ret_stack = ret_stack; - } else - t->ret_stack = NULL; + } } void ftrace_graph_exit_task(struct task_struct *t) -- cgit v1.2.3 From 226f62fdd53d5b2c74e242aa11f6ad43d0285d3f Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 3 Jun 2009 11:23:56 +0200 Subject: perf_counter: Add a comm hook for pure fork()s I noticed missing COMM events and found that we missed reporting them for pure forks. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/fork.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index 23bf757ed321..b7d7a9f0bd7a 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1412,6 +1412,12 @@ long do_fork(unsigned long clone_flags, if (clone_flags & CLONE_VFORK) { p->vfork_done = &vfork; init_completion(&vfork); + } else { + /* + * vfork will do an exec which will call + * set_task_comm() + */ + perf_counter_comm(p); } audit_finish_fork(p); -- cgit v1.2.3 From a96bbc16418bc691317f265d6bf98ba941ca9c1a Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 3 Jun 2009 14:01:36 +0200 Subject: perf_counter: Fix race in counter initialization We need the PID namespace and counter ID available when the counter overflows and we need to generate a sample event. [ Impact: fix kernel crash with high-frequency sampling ] Reported-by: Ingo Molnar Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner Cc: John Kacur LKML-Reference: [ fixed a further crash and cleaned up the initialization a bit ] Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 25 ++++++++++++++----------- 1 file changed, 14 insertions(+), 11 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 317cef78a388..ab4455447f84 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -48,6 +48,8 @@ int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ int sysctl_perf_counter_limit __read_mostly = 100000; /* max NMIs per second */ +static atomic64_t perf_counter_id; + /* * Lock for (sysadmin-configurable) counter reservations: */ @@ -3351,14 +3353,18 @@ perf_counter_alloc(struct perf_counter_attr *attr, mutex_init(&counter->mmap_mutex); - counter->cpu = cpu; + counter->cpu = cpu; counter->attr = *attr; - counter->group_leader = group_leader; - counter->pmu = NULL; - counter->ctx = ctx; - counter->oncpu = -1; + counter->group_leader = group_leader; + counter->pmu = NULL; + counter->ctx = ctx; + counter->oncpu = -1; + + counter->ns = get_pid_ns(current->nsproxy->pid_ns); + counter->id = atomic64_inc_return(&perf_counter_id); + + counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->state = PERF_COUNTER_STATE_INACTIVE; if (attr->disabled) counter->state = PERF_COUNTER_STATE_OFF; @@ -3402,6 +3408,8 @@ done: err = PTR_ERR(pmu); if (err) { + if (counter->ns) + put_pid_ns(counter->ns); kfree(counter); return ERR_PTR(err); } @@ -3419,8 +3427,6 @@ done: return counter; } -static atomic64_t perf_counter_id; - /** * sys_perf_counter_open - open a performance counter, associate it to a task/cpu * @@ -3515,9 +3521,6 @@ SYSCALL_DEFINE5(perf_counter_open, list_add_tail(&counter->owner_entry, ¤t->perf_counter_list); mutex_unlock(¤t->perf_counter_mutex); - counter->ns = get_pid_ns(current->nsproxy->pid_ns); - counter->id = atomic64_inc_return(&perf_counter_id); - fput_light(counter_file, fput_needed2); out_fput: -- cgit v1.2.3 From a2023556409cf7fec5d67a26f7fcfa57c5a4086d Mon Sep 17 00:00:00 2001 From: Tim Bird Date: Tue, 2 Jun 2009 17:06:54 -0700 Subject: ring-buffer: fix bug in ring_buffer_discard_commit There's a bug in ring_buffer_discard_commit. The wrong pointer is being compared in order to check if the event can be freed from the buffer rather than discarded (i.e. marked as PAD). I noticed this when I was working on duration filtering. The bug is not deadly - it just results in lots of wasted space in the buffer. All filtered events are left in the buffer and marked as discarded, rather than being removed from the buffer to make space for other events. Unfortunately, when I fixed this bug, I got errors doing a filtered function trace. Multiple TIME_EXTEND events pile up in the buffer, and trigger the following loop overage warning in rb_iter_peek(): again: ... if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) return NULL; I'm not sure what the best way is to fix this. I don't know if I should extend the loop threshhold, or if I should make the test more complex (ignore TIME_EXTEND events), or just get rid of this loop check completely. Note that if I implement a workaround for this, then I see another problem from rb_advance_iter(). I haven't tracked that one down yet. In general, it seems like the case of removing filtered events has not been working properly, and so some assumptions about buffer invariant conditions need to be revisited. Here's the patch for the simple fix: Compare correct pointer for checking if an event can be freed rather than left as discarded in the buffer. Signed-off-by: Tim Bird LKML-Reference: <4A25BE9E.5090909@am.sony.com> Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 16b24d49604c..945302368691 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1708,7 +1708,7 @@ void ring_buffer_discard_commit(struct ring_buffer *buffer, bpage = cpu_buffer->tail_page; - if (bpage == (void *)addr && rb_page_write(bpage) == old_index) { + if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { /* * This is on the tail page. It is possible that * a write could come in and move the tail page -- cgit v1.2.3 From edd813bffc62a980bb4fb9b1243f31c1cce78da3 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 2 Jun 2009 23:00:53 -0400 Subject: ring-buffer: try to discard unneeded timestamps There are times that a race may happen that we add a timestamp in a nested write. This timestamp would just contain a zero delta and serves no purpose. Now that we have a way to discard events, this patch will try to discard the timestamp instead of just wasting the space in the ring buffer. Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 67 ++++++++++++++++++++++++++++------------------ 1 file changed, 41 insertions(+), 26 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 945302368691..50926601a28d 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1335,6 +1335,38 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, return event; } +static inline int +rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, + struct ring_buffer_event *event) +{ + unsigned long new_index, old_index; + struct buffer_page *bpage; + unsigned long index; + unsigned long addr; + + new_index = rb_event_index(event); + old_index = new_index + rb_event_length(event); + addr = (unsigned long)event; + addr &= PAGE_MASK; + + bpage = cpu_buffer->tail_page; + + if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { + /* + * This is on the tail page. It is possible that + * a write could come in and move the tail page + * and write to the next page. That is fine + * because we just shorten what is on this page. + */ + index = local_cmpxchg(&bpage->write, old_index, new_index); + if (index == old_index) + return 1; + } + + /* could not discard */ + return 0; +} + static int rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, u64 *delta) @@ -1384,10 +1416,13 @@ rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, /* let the caller know this was the commit */ ret = 1; } else { - /* Darn, this is just wasted space */ - event->time_delta = 0; - event->array[0] = 0; - ret = 0; + /* Try to discard the event */ + if (!rb_try_to_discard(cpu_buffer, event)) { + /* Darn, this is just wasted space */ + event->time_delta = 0; + event->array[0] = 0; + ret = 0; + } } *delta = 0; @@ -1682,10 +1717,6 @@ void ring_buffer_discard_commit(struct ring_buffer *buffer, struct ring_buffer_event *event) { struct ring_buffer_per_cpu *cpu_buffer; - unsigned long new_index, old_index; - struct buffer_page *bpage; - unsigned long index; - unsigned long addr; int cpu; /* The event is discarded regardless */ @@ -1701,24 +1732,8 @@ void ring_buffer_discard_commit(struct ring_buffer *buffer, cpu = smp_processor_id(); cpu_buffer = buffer->buffers[cpu]; - new_index = rb_event_index(event); - old_index = new_index + rb_event_length(event); - addr = (unsigned long)event; - addr &= PAGE_MASK; - - bpage = cpu_buffer->tail_page; - - if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { - /* - * This is on the tail page. It is possible that - * a write could come in and move the tail page - * and write to the next page. That is fine - * because we just shorten what is on this page. - */ - index = local_cmpxchg(&bpage->write, old_index, new_index); - if (index == old_index) - goto out; - } + if (!rb_try_to_discard(cpu_buffer, event)) + goto out; /* * The commit is still visible by the reader, so we -- cgit v1.2.3 From ea05b57cc19234d8de9887c8a32c2e58e84b56ba Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Wed, 3 Jun 2009 09:30:10 -0400 Subject: ring-buffer: discard timestamps that are at the start of the buffer Every buffer page in the ring buffer includes its own time stamp. When an event is recorded to the ring buffer with a delta time greater than what can be held in the event header, a time stamp event is created. If the the create timestamp falls over to the next buffer page, it is redundant because the buffer page holds a full time stamp. This patch will try to discard the time stamp when it falls to the start of the next page. This change also fixes a issues with disarding events. If most events are discarded, timestamps will start to creep into the ring buffer. If we do not discard the timestamps then they can fill up the ring buffer over time and waste space. This change will keep time stamps from filling up over another page. If something is recorded in the buffer page, and the rest is filtered, then the time stamps can only fill up to the end of the page. [ Impact: prevent time stamps from filling ring buffer ] Reported-by: Tim Bird Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 37 +++++++++++++++++++++---------------- 1 file changed, 21 insertions(+), 16 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 50926601a28d..7102d7a2fadb 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -370,6 +370,9 @@ static inline int test_time_stamp(u64 delta) /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) +/* Max number of timestamps that can fit on a page */ +#define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP) + int ring_buffer_print_page_header(struct trace_seq *s) { struct buffer_data_page field; @@ -1409,8 +1412,12 @@ rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, event->array[0] = *delta >> TS_SHIFT; } else { cpu_buffer->commit_page->page->time_stamp = *ts; - event->time_delta = 0; - event->array[0] = 0; + /* try to discard, since we do not need this */ + if (!rb_try_to_discard(cpu_buffer, event)) { + /* nope, just zero it */ + event->time_delta = 0; + event->array[0] = 0; + } } cpu_buffer->write_stamp = *ts; /* let the caller know this was the commit */ @@ -2268,8 +2275,8 @@ static void rb_advance_iter(struct ring_buffer_iter *iter) * Check if we are at the end of the buffer. */ if (iter->head >= rb_page_size(iter->head_page)) { - if (RB_WARN_ON(buffer, - iter->head_page == cpu_buffer->commit_page)) + /* discarded commits can make the page empty */ + if (iter->head_page == cpu_buffer->commit_page) return; rb_inc_iter(iter); return; @@ -2312,12 +2319,10 @@ rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) /* * We repeat when a timestamp is encountered. It is possible * to get multiple timestamps from an interrupt entering just - * as one timestamp is about to be written. The max times - * that this can happen is the number of nested interrupts we - * can have. Nesting 10 deep of interrupts is clearly - * an anomaly. + * as one timestamp is about to be written, or from discarded + * commits. The most that we can have is the number on a single page. */ - if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) + if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) return NULL; reader = rb_get_reader_page(cpu_buffer); @@ -2383,14 +2388,14 @@ rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) again: /* - * We repeat when a timestamp is encountered. It is possible - * to get multiple timestamps from an interrupt entering just - * as one timestamp is about to be written. The max times - * that this can happen is the number of nested interrupts we - * can have. Nesting 10 deep of interrupts is clearly - * an anomaly. + * We repeat when a timestamp is encountered. + * We can get multiple timestamps by nested interrupts or also + * if filtering is on (discarding commits). Since discarding + * commits can be frequent we can get a lot of timestamps. + * But we limit them by not adding timestamps if they begin + * at the start of a page. */ - if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) + if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) return NULL; if (rb_per_cpu_empty(cpu_buffer)) -- cgit v1.2.3 From 083a63b48e4dd0a6a2d44216720076dc81ebb255 Mon Sep 17 00:00:00 2001 From: walimis Date: Wed, 3 Jun 2009 16:01:28 +0800 Subject: tracing/trace_stack: fix the number of entries in the header The last entry in the stack_dump_trace is ULONG_MAX, which is not a valid entry, but max_stack_trace.nr_entries has accounted for it. So when printing the header, we should decrease it by one. Before fix, print as following, for example: Depth Size Location (53 entries) <--- should be 52 ----- ---- -------- 0) 3264 108 update_wall_time+0x4d5/0x9a0 ... 51) 80 80 syscall_call+0x7/0xb ^^^ it's correct. Signed-off-by: walimis LKML-Reference: <1244016090-7814-1-git-send-email-walimisdev@gmail.com> Signed-off-by: Steven Rostedt --- kernel/trace/trace_stack.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c index 1796f00524e1..2d7aebd71dbd 100644 --- a/kernel/trace/trace_stack.c +++ b/kernel/trace/trace_stack.c @@ -265,7 +265,7 @@ static int t_show(struct seq_file *m, void *v) seq_printf(m, " Depth Size Location" " (%d entries)\n" " ----- ---- --------\n", - max_stack_trace.nr_entries); + max_stack_trace.nr_entries - 1); if (!stack_tracer_enabled && !max_stack_size) print_disabled(m); -- cgit v1.2.3 From f11b3f4e2932bfdcfc458ab8d1ece62724ceabfc Mon Sep 17 00:00:00 2001 From: walimis Date: Wed, 3 Jun 2009 16:01:29 +0800 Subject: tracing/events: fix output format of kernel stack According to "events/ftrace/kernel_stack/format", output format of kernel stack should use "=>" instead of "<=". The second problem is that we shouldn't skip the first entry in the stack, although it seems to be duplicated when used in the "function" tracer, but events also use it. If we skip the first one, we will drop the topmost entry of the stack. The last problem is that if the last entry is ULONG_MAX(0xffffffff), we should drop it, otherwise it will print a NULL name line. before fix: sh-1072 [000] 26.957239: sched_process_fork: parent sh:1072 child sh:1073 sh-1072 [000] 26.957262: <= syscall_call <= sh-1072 [000] 26.957744: sched_switch: task sh:1072 [120] (R) ==> sh:1073 [120] sh-1072 [000] 26.957752: <= preempt_schedule <= wake_up_new_task <= do_fork <= sys_clone <= syscall_call <= After fix: sh-1075 [000] 39.791848: sched_process_fork: parent sh:1075 child sh:1076 sh-1075 [000] 39.791871: => sys_clone => syscall_call sh-1075 [000] 39.792713: sched_switch: task sh:1075 [120] (R) ==> sh:1076 [120] sh-1075 [000] 39.792722: => schedule => preempt_schedule => wake_up_new_task => do_fork => sys_clone => syscall_call Signed-off-by: walimis LKML-Reference: <1244016090-7814-2-git-send-email-walimisdev@gmail.com> Signed-off-by: Steven Rostedt --- kernel/trace/trace_output.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 0fe3b223f7ed..64596a571609 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -975,16 +975,16 @@ static enum print_line_t trace_stack_print(struct trace_iterator *iter, trace_assign_type(field, iter->ent); + if (!trace_seq_puts(s, "\n")) + goto partial; for (i = 0; i < FTRACE_STACK_ENTRIES; i++) { - if (!field->caller[i]) + if (!field->caller[i] || (field->caller[i] == ULONG_MAX)) break; - if (i) { - if (!trace_seq_puts(s, " <= ")) - goto partial; + if (!trace_seq_puts(s, " => ")) + goto partial; - if (!seq_print_ip_sym(s, field->caller[i], flags)) - goto partial; - } + if (!seq_print_ip_sym(s, field->caller[i], flags)) + goto partial; if (!trace_seq_puts(s, "\n")) goto partial; } -- cgit v1.2.3 From 048dc50c5e7eada19ebabbad70b7966d14283d41 Mon Sep 17 00:00:00 2001 From: walimis Date: Wed, 3 Jun 2009 16:01:30 +0800 Subject: tracing/events: fix output format of user stack According to "events/ftrace/user_stack/format", fix the output of user stack. before fix: sh-1073 [000] 31.137561: <- <0804e33c> <- <080835c1> after fix: sh-1072 [000] 37.039329: => => <0804e33c> => <080835c1> Signed-off-by: walimis LKML-Reference: <1244016090-7814-3-git-send-email-walimisdev@gmail.com> Signed-off-by: Steven Rostedt --- kernel/trace/trace_output.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 64596a571609..8dadbbbd2d5c 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -389,17 +389,20 @@ seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s, if (ip == ULONG_MAX || !ret) break; - if (i && ret) - ret = trace_seq_puts(s, " <- "); + if (ret) + ret = trace_seq_puts(s, " => "); if (!ip) { if (ret) ret = trace_seq_puts(s, "??"); + if (ret) + ret = trace_seq_puts(s, "\n"); continue; } if (!ret) break; if (ret) ret = seq_print_user_ip(s, mm, ip, sym_flags); + ret = trace_seq_puts(s, "\n"); } if (mm) @@ -1012,10 +1015,10 @@ static enum print_line_t trace_user_stack_print(struct trace_iterator *iter, trace_assign_type(field, iter->ent); - if (!seq_print_userip_objs(field, s, flags)) + if (!trace_seq_putc(s, '\n')) goto partial; - if (!trace_seq_putc(s, '\n')) + if (!seq_print_userip_objs(field, s, flags)) goto partial; return TRACE_TYPE_HANDLED; -- cgit v1.2.3 From 56d8bd3f0b98972312cad683947ec90b21011199 Mon Sep 17 00:00:00 2001 From: Steven Whitehouse Date: Wed, 3 Jun 2009 14:52:03 +0100 Subject: tracing: fix multiple use of __print_flags and __print_symbolic Here is an updated patch to include the extra call to trace_seq_init() as requested. This is vs. the latest -tip tree and fixes the use of multiple __print_flags and __print_symbolic in a single tracer. Also tested to ensure its working now: mount.gfs2-2534 [000] 235.850587: gfs2_glock_queue: 8.7 glock 1:2 dequeue PR mount.gfs2-2534 [000] 235.850591: gfs2_demote_rq: 8.7 glock 1:0 demote EX to NL flags:DI mount.gfs2-2534 [000] 235.850591: gfs2_glock_queue: 8.7 glock 1:0 dequeue EX glock_workqueue-2529 [000] 235.850666: gfs2_glock_state_change: 8.7 glock 1:0 state EX => NL tgt:NL dmt:NL flags:lDpI glock_workqueue-2529 [000] 235.850672: gfs2_glock_put: 8.7 glock 1:0 state NL => IV flags:I Signed-off-by: Steven Whitehouse LKML-Reference: <1244037123.29604.603.camel@localhost.localdomain> Signed-off-by: Steven Rostedt --- kernel/trace/trace_output.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 8dadbbbd2d5c..8afeea412e77 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -223,10 +223,9 @@ ftrace_print_flags_seq(struct trace_seq *p, const char *delim, { unsigned long mask; const char *str; + const char *ret = p->buffer + p->len; int i; - trace_seq_init(p); - for (i = 0; flag_array[i].name && flags; i++) { mask = flag_array[i].mask; @@ -249,7 +248,7 @@ ftrace_print_flags_seq(struct trace_seq *p, const char *delim, trace_seq_putc(p, 0); - return p->buffer; + return ret; } EXPORT_SYMBOL(ftrace_print_flags_seq); @@ -258,8 +257,7 @@ ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val, const struct trace_print_flags *symbol_array) { int i; - - trace_seq_init(p); + const char *ret = p->buffer + p->len; for (i = 0; symbol_array[i].name; i++) { @@ -275,7 +273,7 @@ ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val, trace_seq_putc(p, 0); - return p->buffer; + return ret; } EXPORT_SYMBOL(ftrace_print_symbols_seq); -- cgit v1.2.3 From 563af16c30ede41eda2d614195d88e07f7c7103d Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Wed, 3 Jun 2009 11:10:44 -0400 Subject: tracing: add annotation to what type of stack trace is recorded The current method of printing out a stack trace is to add a new line and print out the trace: yum-updatesd-3120 [002] 573.691303: => do_softirq => irq_exit => smp_apic_timer_interrupt => apic_timer_interrupt This looks a bit awkward, and if we have both stack and user stack traces running, it would be nice to have a title to tell them apart, although it is easy to tell by the output. This patch adds an annotation to the start of the stack traces: init-1 [003] 929.304979: => user_path_at => vfs_fstatat => vfs_stat => sys_newstat => system_call_fastpath cat-3459 [002] 1016.824040: => <0000003aae6c0250> => <00007ffff4b06ae4> => <69636172742f6775> Signed-off-by: Steven Rostedt --- kernel/trace/trace_output.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 8afeea412e77..425725c1622d 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -976,7 +976,7 @@ static enum print_line_t trace_stack_print(struct trace_iterator *iter, trace_assign_type(field, iter->ent); - if (!trace_seq_puts(s, "\n")) + if (!trace_seq_puts(s, "\n")) goto partial; for (i = 0; i < FTRACE_STACK_ENTRIES; i++) { if (!field->caller[i] || (field->caller[i] == ULONG_MAX)) @@ -1013,7 +1013,7 @@ static enum print_line_t trace_user_stack_print(struct trace_iterator *iter, trace_assign_type(field, iter->ent); - if (!trace_seq_putc(s, '\n')) + if (!trace_seq_puts(s, "\n")) goto partial; if (!seq_print_userip_objs(field, s, flags)) -- cgit v1.2.3 From 128f048f0f0d2a477ad2555e7acd2ad15a1b6061 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 3 Jun 2009 22:19:36 +0200 Subject: perf_counter: Fix throttling lock-up Throttling logic is broken and we can lock up with too small hw sampling intervals. Make the throttling code more robust: disable counters even if we already disabled them. ( Also clean up whitespace damage i noticed while reading various pieces of code related to throttling. ) Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo Cc: Thomas Gleixner LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ab4455447f84..0bb03f15a5b6 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2822,11 +2822,20 @@ int perf_counter_overflow(struct perf_counter *counter, if (!throttle) { counter->hw.interrupts++; - } else if (counter->hw.interrupts != MAX_INTERRUPTS) { - counter->hw.interrupts++; - if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) { - counter->hw.interrupts = MAX_INTERRUPTS; - perf_log_throttle(counter, 0); + } else { + if (counter->hw.interrupts != MAX_INTERRUPTS) { + counter->hw.interrupts++; + if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) { + counter->hw.interrupts = MAX_INTERRUPTS; + perf_log_throttle(counter, 0); + ret = 1; + } + } else { + /* + * Keep re-disabling counters even though on the previous + * pass we disabled it - just in case we raced with a + * sched-in and the counter got enabled again: + */ ret = 1; } } -- cgit v1.2.3 From e0a94c2a63f2644826069044649669b5e7ca75d3 Mon Sep 17 00:00:00 2001 From: Christoph Lameter Date: Wed, 3 Jun 2009 16:04:31 -0400 Subject: security: use mmap_min_addr indepedently of security models This patch removes the dependency of mmap_min_addr on CONFIG_SECURITY. It also sets a default mmap_min_addr of 4096. mmapping of addresses below 4096 will only be possible for processes with CAP_SYS_RAWIO. Signed-off-by: Christoph Lameter Acked-by: Eric Paris Looks-ok-by: Linus Torvalds Signed-off-by: James Morris --- kernel/sysctl.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 149581fb48ab..45bd711a242e 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -1237,7 +1237,6 @@ static struct ctl_table vm_table[] = { .strategy = &sysctl_jiffies, }, #endif -#ifdef CONFIG_SECURITY { .ctl_name = CTL_UNNUMBERED, .procname = "mmap_min_addr", @@ -1246,7 +1245,6 @@ static struct ctl_table vm_table[] = { .mode = 0644, .proc_handler = &proc_doulongvec_minmax, }, -#endif #ifdef CONFIG_NUMA { .ctl_name = CTL_UNNUMBERED, -- cgit v1.2.3 From 60313ebed739b331e8e61079da27a11ee3b73a30 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 4 Jun 2009 16:53:44 +0200 Subject: perf_counter: Add fork event Create a fork event so that we can easily clone the comm and dso maps without having to generate all those events. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/fork.c | 4 +- kernel/perf_counter.c | 131 +++++++++++++++++++++++++++++++++++++++++++------- 2 files changed, 116 insertions(+), 19 deletions(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index b7d7a9f0bd7a..f4466ca37ece 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1412,12 +1412,12 @@ long do_fork(unsigned long clone_flags, if (clone_flags & CLONE_VFORK) { p->vfork_done = &vfork; init_completion(&vfork); - } else { + } else if (!(clone_flags & CLONE_VM)) { /* * vfork will do an exec which will call * set_task_comm() */ - perf_counter_comm(p); + perf_counter_fork(p); } audit_finish_fork(p); diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 0bb03f15a5b6..78c58623a0dd 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -40,9 +40,9 @@ static int perf_reserved_percpu __read_mostly; static int perf_overcommit __read_mostly = 1; static atomic_t nr_counters __read_mostly; -static atomic_t nr_mmap_tracking __read_mostly; -static atomic_t nr_munmap_tracking __read_mostly; -static atomic_t nr_comm_tracking __read_mostly; +static atomic_t nr_mmap_counters __read_mostly; +static atomic_t nr_munmap_counters __read_mostly; +static atomic_t nr_comm_counters __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ @@ -1447,11 +1447,11 @@ static void free_counter(struct perf_counter *counter) atomic_dec(&nr_counters); if (counter->attr.mmap) - atomic_dec(&nr_mmap_tracking); + atomic_dec(&nr_mmap_counters); if (counter->attr.munmap) - atomic_dec(&nr_munmap_tracking); + atomic_dec(&nr_munmap_counters); if (counter->attr.comm) - atomic_dec(&nr_comm_tracking); + atomic_dec(&nr_comm_counters); if (counter->destroy) counter->destroy(counter); @@ -2475,6 +2475,105 @@ static void perf_counter_output(struct perf_counter *counter, perf_output_end(&handle); } +/* + * fork tracking + */ + +struct perf_fork_event { + struct task_struct *task; + + struct { + struct perf_event_header header; + + u32 pid; + u32 ppid; + } event; +}; + +static void perf_counter_fork_output(struct perf_counter *counter, + struct perf_fork_event *fork_event) +{ + struct perf_output_handle handle; + int size = fork_event->event.header.size; + struct task_struct *task = fork_event->task; + int ret = perf_output_begin(&handle, counter, size, 0, 0); + + if (ret) + return; + + fork_event->event.pid = perf_counter_pid(counter, task); + fork_event->event.ppid = perf_counter_pid(counter, task->real_parent); + + perf_output_put(&handle, fork_event->event); + perf_output_end(&handle); +} + +static int perf_counter_fork_match(struct perf_counter *counter) +{ + if (counter->attr.comm || counter->attr.mmap || counter->attr.munmap) + return 1; + + return 0; +} + +static void perf_counter_fork_ctx(struct perf_counter_context *ctx, + struct perf_fork_event *fork_event) +{ + struct perf_counter *counter; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { + if (perf_counter_fork_match(counter)) + perf_counter_fork_output(counter, fork_event); + } + rcu_read_unlock(); +} + +static void perf_counter_fork_event(struct perf_fork_event *fork_event) +{ + struct perf_cpu_context *cpuctx; + struct perf_counter_context *ctx; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_counter_fork_ctx(&cpuctx->ctx, fork_event); + put_cpu_var(perf_cpu_context); + + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_counter_ctxp); + if (ctx) + perf_counter_fork_ctx(ctx, fork_event); + rcu_read_unlock(); +} + +void perf_counter_fork(struct task_struct *task) +{ + struct perf_fork_event fork_event; + + if (!atomic_read(&nr_comm_counters) && + !atomic_read(&nr_mmap_counters) && + !atomic_read(&nr_munmap_counters)) + return; + + fork_event = (struct perf_fork_event){ + .task = task, + .event = { + .header = { + .type = PERF_EVENT_FORK, + .size = sizeof(fork_event.event), + }, + }, + }; + + perf_counter_fork_event(&fork_event); +} + /* * comm tracking */ @@ -2511,11 +2610,9 @@ static void perf_counter_comm_output(struct perf_counter *counter, perf_output_end(&handle); } -static int perf_counter_comm_match(struct perf_counter *counter, - struct perf_comm_event *comm_event) +static int perf_counter_comm_match(struct perf_counter *counter) { - if (counter->attr.comm && - comm_event->event.header.type == PERF_EVENT_COMM) + if (counter->attr.comm) return 1; return 0; @@ -2531,7 +2628,7 @@ static void perf_counter_comm_ctx(struct perf_counter_context *ctx, rcu_read_lock(); list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_counter_comm_match(counter, comm_event)) + if (perf_counter_comm_match(counter)) perf_counter_comm_output(counter, comm_event); } rcu_read_unlock(); @@ -2570,7 +2667,7 @@ void perf_counter_comm(struct task_struct *task) { struct perf_comm_event comm_event; - if (!atomic_read(&nr_comm_tracking)) + if (!atomic_read(&nr_comm_counters)) return; comm_event = (struct perf_comm_event){ @@ -2708,7 +2805,7 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, { struct perf_mmap_event mmap_event; - if (!atomic_read(&nr_mmap_tracking)) + if (!atomic_read(&nr_mmap_counters)) return; mmap_event = (struct perf_mmap_event){ @@ -2729,7 +2826,7 @@ void perf_counter_munmap(unsigned long addr, unsigned long len, { struct perf_mmap_event mmap_event; - if (!atomic_read(&nr_munmap_tracking)) + if (!atomic_read(&nr_munmap_counters)) return; mmap_event = (struct perf_mmap_event){ @@ -3427,11 +3524,11 @@ done: atomic_inc(&nr_counters); if (counter->attr.mmap) - atomic_inc(&nr_mmap_tracking); + atomic_inc(&nr_mmap_counters); if (counter->attr.munmap) - atomic_inc(&nr_munmap_tracking); + atomic_inc(&nr_munmap_counters); if (counter->attr.comm) - atomic_inc(&nr_comm_tracking); + atomic_inc(&nr_comm_counters); return counter; } -- cgit v1.2.3 From d99e9446200c1ffab28cb0e39b76c34a2bfafd06 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 4 Jun 2009 17:08:58 +0200 Subject: perf_counter: Remove munmap stuff In name of keeping it simple, only track mmap events. Userspace will have to remove old overlapping maps when it encounters them. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 38 +++----------------------------------- 1 file changed, 3 insertions(+), 35 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 78c58623a0dd..195712e20d07 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -41,7 +41,6 @@ static int perf_overcommit __read_mostly = 1; static atomic_t nr_counters __read_mostly; static atomic_t nr_mmap_counters __read_mostly; -static atomic_t nr_munmap_counters __read_mostly; static atomic_t nr_comm_counters __read_mostly; int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ @@ -1448,8 +1447,6 @@ static void free_counter(struct perf_counter *counter) atomic_dec(&nr_counters); if (counter->attr.mmap) atomic_dec(&nr_mmap_counters); - if (counter->attr.munmap) - atomic_dec(&nr_munmap_counters); if (counter->attr.comm) atomic_dec(&nr_comm_counters); @@ -2510,7 +2507,7 @@ static void perf_counter_fork_output(struct perf_counter *counter, static int perf_counter_fork_match(struct perf_counter *counter) { - if (counter->attr.comm || counter->attr.mmap || counter->attr.munmap) + if (counter->attr.comm || counter->attr.mmap) return 1; return 0; @@ -2557,8 +2554,7 @@ void perf_counter_fork(struct task_struct *task) struct perf_fork_event fork_event; if (!atomic_read(&nr_comm_counters) && - !atomic_read(&nr_mmap_counters) && - !atomic_read(&nr_munmap_counters)) + !atomic_read(&nr_mmap_counters)) return; fork_event = (struct perf_fork_event){ @@ -2722,12 +2718,7 @@ static void perf_counter_mmap_output(struct perf_counter *counter, static int perf_counter_mmap_match(struct perf_counter *counter, struct perf_mmap_event *mmap_event) { - if (counter->attr.mmap && - mmap_event->event.header.type == PERF_EVENT_MMAP) - return 1; - - if (counter->attr.munmap && - mmap_event->event.header.type == PERF_EVENT_MUNMAP) + if (counter->attr.mmap) return 1; return 0; @@ -2821,27 +2812,6 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, perf_counter_mmap_event(&mmap_event); } -void perf_counter_munmap(unsigned long addr, unsigned long len, - unsigned long pgoff, struct file *file) -{ - struct perf_mmap_event mmap_event; - - if (!atomic_read(&nr_munmap_counters)) - return; - - mmap_event = (struct perf_mmap_event){ - .file = file, - .event = { - .header = { .type = PERF_EVENT_MUNMAP, }, - .start = addr, - .len = len, - .pgoff = pgoff, - }, - }; - - perf_counter_mmap_event(&mmap_event); -} - /* * Log sample_period changes so that analyzing tools can re-normalize the * event flow. @@ -3525,8 +3495,6 @@ done: atomic_inc(&nr_counters); if (counter->attr.mmap) atomic_inc(&nr_mmap_counters); - if (counter->attr.munmap) - atomic_inc(&nr_munmap_counters); if (counter->attr.comm) atomic_inc(&nr_comm_counters); -- cgit v1.2.3 From 087eb437051b3de817720f9c80c440fc9e7dcce8 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Thu, 4 Jun 2009 16:29:07 -0700 Subject: ptrace: tracehook_report_clone: fix false positives The "trace || CLONE_PTRACE" check in tracehook_report_clone() is not right, - If the untraced task does clone(CLONE_PTRACE) the new child is not traced, we must not queue SIGSTOP. - If we forked the traced task, but the tracer exits and untraces both the forking task and the new child (after copy_process() drops tasklist_lock), we should not queue SIGSTOP too. Change the code to check task_ptrace() != 0 instead. This is still racy, but the race is harmless. We can race with another tracer attaching to this child, or the tracer can exit and detach in parallel. But giwen that we didn't do wake_up_new_task() yet, the child must have the pending SIGSTOP anyway. Signed-off-by: Oleg Nesterov Acked-by: Roland McGrath Cc: Christoph Hellwig Cc: Ingo Molnar Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/fork.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index b9e2edd00726..875ffbdd96d0 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1409,7 +1409,7 @@ long do_fork(unsigned long clone_flags, } audit_finish_fork(p); - tracehook_report_clone(trace, regs, clone_flags, nr, p); + tracehook_report_clone(regs, clone_flags, nr, p); /* * We set PF_STARTING at creation in case tracing wants to -- cgit v1.2.3 From edaba2c5334492f82d39ec35637c6dea5176a977 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Thu, 4 Jun 2009 16:29:09 -0700 Subject: ptrace: revert "ptrace_detach: the wrong wakeup breaks the ERESTARTxxx logic" Commit 95a3540da9c81a5987be810e1d9a83640a366bd5 ("ptrace_detach: the wrong wakeup breaks the ERESTARTxxx logic") removed the "extra" wake_up_process() from ptrace_detach(), but as Jan pointed out this breaks the compatibility. I believe the changelog is right and this wake_up() is wrong in many ways, but GDB assumes that ptrace(PTRACE_DETACH, child, 0, 0) always wakes up the tracee. Despite the fact this breaks SIGNAL_STOP_STOPPED/group_stop_count logic, and despite the fact this wake_up_process() can break another assumption: PTRACE_DETACH with SIGSTOP should leave the tracee in TASK_STOPPED case. Because the untraced child can dequeue SIGSTOP and call do_signal_stop() before ptrace_detach() calls wake_up_process(). Revert this change for now. We need some fixes even if we we want to keep the current behaviour, but these fixes are not for 2.6.30. Signed-off-by: Oleg Nesterov Acked-by: Roland McGrath Cc: Jan Kratochvil Cc: Denys Vlasenko Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/ptrace.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/ptrace.c b/kernel/ptrace.c index 0692ab5a0d67..42c317874cfa 100644 --- a/kernel/ptrace.c +++ b/kernel/ptrace.c @@ -304,6 +304,8 @@ int ptrace_detach(struct task_struct *child, unsigned int data) if (child->ptrace) { child->exit_code = data; dead = __ptrace_detach(current, child); + if (!child->exit_state) + wake_up_process(child); } write_unlock_irq(&tasklist_lock); -- cgit v1.2.3 From 6dc5f2a41759987e35e757ef00192e7b424563bb Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Fri, 5 Jun 2009 12:36:28 +1000 Subject: perf_counter: Fix lockup with interrupting counters Commit 8e3747c1 ("perf_counter: Change data head from u32 to u64") changed the type of 'head' in struct perf_mmap_data from atomic_t to atomic_long_t, but missed converting one use of atomic_read on it to atomic_long_read. The effect of using atomic_read rather than atomic_long_read on powerpc (and other big-endian architectures) is that we get the high half of the 64-bit quantity, resulting in the cmpxchg retry loop in perf_output_begin spinning forever as soon as data->head becomes non-zero. On little-endian architectures such as x86 we would get the low half, resulting in a lockup once data->head becomes greater than 4G. This fixes it by using atomic_long_read rather than atomic_read. [ Impact: fix perfcounter lockup on PowerPC / big-endian systems ] Signed-off-by: Paul Mackerras Cc: Peter Zijlstra LKML-Reference: <18984.33964.21541.743096@cargo.ozlabs.ibm.com> Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 195712e20d07..a5d3e2aedd2f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2234,7 +2234,7 @@ static int perf_output_begin(struct perf_output_handle *handle, perf_output_lock(handle); do { - offset = head = atomic_read(&data->head); + offset = head = atomic_long_read(&data->head); head += size; } while (atomic_long_cmpxchg(&data->head, offset, head) != offset); -- cgit v1.2.3 From 089dd79db9264dc0da602bad45d42f1b3e7d1e07 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 5 Jun 2009 14:04:55 +0200 Subject: perf_counter: Generate mmap events for install_special_mapping() In order to track the vdso also generate mmap events for install_special_mapping(). Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 34 ++++++++++++++++++++++------------ 1 file changed, 22 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index a5d3e2aedd2f..37a5a241ca7e 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2255,7 +2255,7 @@ out: } static void perf_output_copy(struct perf_output_handle *handle, - void *buf, unsigned int len) + const void *buf, unsigned int len) { unsigned int pages_mask; unsigned int offset; @@ -2681,9 +2681,10 @@ void perf_counter_comm(struct task_struct *task) */ struct perf_mmap_event { - struct file *file; - char *file_name; - int file_size; + struct vm_area_struct *vma; + + const char *file_name; + int file_size; struct { struct perf_event_header header; @@ -2744,11 +2745,12 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) { struct perf_cpu_context *cpuctx; struct perf_counter_context *ctx; - struct file *file = mmap_event->file; + struct vm_area_struct *vma = mmap_event->vma; + struct file *file = vma->vm_file; unsigned int size; char tmp[16]; char *buf = NULL; - char *name; + const char *name; if (file) { buf = kzalloc(PATH_MAX, GFP_KERNEL); @@ -2762,6 +2764,15 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) goto got_name; } } else { + name = arch_vma_name(mmap_event->vma); + if (name) + goto got_name; + + if (!vma->vm_mm) { + name = strncpy(tmp, "[vdso]", sizeof(tmp)); + goto got_name; + } + name = strncpy(tmp, "//anon", sizeof(tmp)); goto got_name; } @@ -2791,8 +2802,7 @@ got_name: kfree(buf); } -void perf_counter_mmap(unsigned long addr, unsigned long len, - unsigned long pgoff, struct file *file) +void __perf_counter_mmap(struct vm_area_struct *vma) { struct perf_mmap_event mmap_event; @@ -2800,12 +2810,12 @@ void perf_counter_mmap(unsigned long addr, unsigned long len, return; mmap_event = (struct perf_mmap_event){ - .file = file, + .vma = vma, .event = { .header = { .type = PERF_EVENT_MMAP, }, - .start = addr, - .len = len, - .pgoff = pgoff, + .start = vma->vm_start, + .len = vma->vm_end - vma->vm_start, + .pgoff = vma->vm_pgoff, }, }; -- cgit v1.2.3 From ac4bcf889469ffbca88f234d3184452886a47905 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 5 Jun 2009 14:44:52 +0200 Subject: perf_counter: Change PERF_SAMPLE_CONFIG into PERF_SAMPLE_ID The purpose of PERF_SAMPLE_CONFIG was to identify the counters, since then we've added counter ids, use those instead. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 37a5a241ca7e..e75b91a76a58 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2392,8 +2392,8 @@ static void perf_counter_output(struct perf_counter *counter, header.size += sizeof(u64); } - if (sample_type & PERF_SAMPLE_CONFIG) { - header.type |= PERF_SAMPLE_CONFIG; + if (sample_type & PERF_SAMPLE_ID) { + header.type |= PERF_SAMPLE_ID; header.size += sizeof(u64); } @@ -2439,8 +2439,8 @@ static void perf_counter_output(struct perf_counter *counter, if (sample_type & PERF_SAMPLE_ADDR) perf_output_put(&handle, addr); - if (sample_type & PERF_SAMPLE_CONFIG) - perf_output_put(&handle, counter->attr.config); + if (sample_type & PERF_SAMPLE_ID) + perf_output_put(&handle, counter->id); if (sample_type & PERF_SAMPLE_CPU) perf_output_put(&handle, cpu_entry); -- cgit v1.2.3 From 689802b2d0536e72281dc959ab9cb34fb3c304cf Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 5 Jun 2009 15:05:43 +0200 Subject: perf_counter: Add PERF_SAMPLE_PERIOD In order to allow easy tracking of the period, also provide means of adding it to the sample data. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index e75b91a76a58..f8390668c391 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2404,6 +2404,11 @@ static void perf_counter_output(struct perf_counter *counter, cpu_entry.cpu = raw_smp_processor_id(); } + if (sample_type & PERF_SAMPLE_PERIOD) { + header.type |= PERF_SAMPLE_PERIOD; + header.size += sizeof(u64); + } + if (sample_type & PERF_SAMPLE_GROUP) { header.type |= PERF_SAMPLE_GROUP; header.size += sizeof(u64) + @@ -2445,6 +2450,9 @@ static void perf_counter_output(struct perf_counter *counter, if (sample_type & PERF_SAMPLE_CPU) perf_output_put(&handle, cpu_entry); + if (sample_type & PERF_SAMPLE_PERIOD) + perf_output_put(&handle, counter->hw.sample_period); + /* * XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult. */ @@ -2835,6 +2843,7 @@ static void perf_log_period(struct perf_counter *counter, u64 period) struct { struct perf_event_header header; u64 time; + u64 id; u64 period; } freq_event = { .header = { @@ -2843,6 +2852,7 @@ static void perf_log_period(struct perf_counter *counter, u64 period) .size = sizeof(freq_event), }, .time = sched_clock(), + .id = counter->id, .period = period, }; -- cgit v1.2.3 From 6a24ed6c6082ec65d19331a4bfa30c0512a1a822 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 5 Jun 2009 18:01:29 +0200 Subject: perf_counter: Fix frequency adjustment for < HZ Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 32 +++++++++++++++++++++++++------- 1 file changed, 25 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index f8390668c391..47c92fb927f2 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1187,8 +1187,9 @@ static void perf_log_period(struct perf_counter *counter, u64 period); static void perf_adjust_freq(struct perf_counter_context *ctx) { struct perf_counter *counter; + struct hw_perf_counter *hwc; u64 interrupts, sample_period; - u64 events, period; + u64 events, period, freq; s64 delta; spin_lock(&ctx->lock); @@ -1196,8 +1197,10 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) if (counter->state != PERF_COUNTER_STATE_ACTIVE) continue; - interrupts = counter->hw.interrupts; - counter->hw.interrupts = 0; + hwc = &counter->hw; + + interrupts = hwc->interrupts; + hwc->interrupts = 0; if (interrupts == MAX_INTERRUPTS) { perf_log_throttle(counter, 1); @@ -1208,20 +1211,35 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) if (!counter->attr.freq || !counter->attr.sample_freq) continue; - events = HZ * interrupts * counter->hw.sample_period; + if (counter->attr.sample_freq < HZ) { + freq = counter->attr.sample_freq; + + hwc->freq_count += freq; + hwc->freq_interrupts += interrupts; + + if (hwc->freq_count < HZ) + continue; + + interrupts = hwc->freq_interrupts; + hwc->freq_interrupts = 0; + hwc->freq_count -= HZ; + } else + freq = HZ; + + events = freq * interrupts * hwc->sample_period; period = div64_u64(events, counter->attr.sample_freq); - delta = (s64)(1 + period - counter->hw.sample_period); + delta = (s64)(1 + period - hwc->sample_period); delta >>= 1; - sample_period = counter->hw.sample_period + delta; + sample_period = hwc->sample_period + delta; if (!sample_period) sample_period = 1; perf_log_period(counter, sample_period); - counter->hw.sample_period = sample_period; + hwc->sample_period = sample_period; } spin_unlock(&ctx->lock); } -- cgit v1.2.3 From a21ca2cac582886a3e95c8bb84ff7c52d4d15e54 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Sat, 6 Jun 2009 09:58:57 +0200 Subject: perf_counter: Separate out attr->type from attr->config Counter type is a frequently used value and we do a lot of bit juggling by encoding and decoding it from attr->config. Clean this up by creating a separate attr->type field. Also clean up the various similarly complex user-space bits all around counter attribute management. The net improvement is significant, and it will be easier to add a new major type (which is what triggered this cleanup). (This changes the ABI, all tools are adapted.) (PowerPC build-tested.) Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 47c92fb927f2..75ae76796df1 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3091,14 +3091,12 @@ static int perf_swcounter_match(struct perf_counter *counter, enum perf_event_types type, u32 event, struct pt_regs *regs) { - u64 event_config; - - event_config = ((u64) type << PERF_COUNTER_TYPE_SHIFT) | event; - if (!perf_swcounter_is_counting(counter)) return 0; - if (counter->attr.config != event_config) + if (counter->attr.type != type) + return 0; + if (counter->attr.config != event) return 0; if (regs) { @@ -3403,7 +3401,7 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) * to be kernel events, and page faults are never hypervisor * events. */ - switch (perf_event_id(&counter->attr)) { + switch (counter->attr.config) { case PERF_COUNT_CPU_CLOCK: pmu = &perf_ops_cpu_clock; @@ -3496,12 +3494,12 @@ perf_counter_alloc(struct perf_counter_attr *attr, if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP)) goto done; - if (perf_event_raw(attr)) { + if (attr->type == PERF_TYPE_RAW) { pmu = hw_perf_counter_init(counter); goto done; } - switch (perf_event_type(attr)) { + switch (attr->type) { case PERF_TYPE_HARDWARE: pmu = hw_perf_counter_init(counter); break; -- cgit v1.2.3 From 8326f44da090d6d304d29b9fdc7fb3e20889e329 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 5 Jun 2009 20:22:46 +0200 Subject: perf_counter: Implement generalized cache event types Extend generic event enumeration with the PERF_TYPE_HW_CACHE method. This is a 3-dimensional space: { L1-D, L1-I, L2, ITLB, DTLB, BPU } x { load, store, prefetch } x { accesses, misses } User-space passes in the 3 coordinates and the kernel provides a counter. (if the hardware supports that type and if the combination makes sense.) Combinations that make no sense produce a -EINVAL. Combinations that are not supported by the hardware produce -ENOTSUP. Extend the tools to deal with this, and rewrite the event symbol parsing code with various popular aliases for the units and access methods above. So 'l1-cache-miss' and 'l1d-read-ops' are both valid aliases. ( x86 is supported for now, with the Nehalem event table filled in, and with Core2 and Atom having placeholder tables. ) Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Corey Ashford Cc: Marcelo Tosatti Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 75ae76796df1..5eacaaf3f9cd 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3501,6 +3501,7 @@ perf_counter_alloc(struct perf_counter_attr *attr, switch (attr->type) { case PERF_TYPE_HARDWARE: + case PERF_TYPE_HW_CACHE: pmu = hw_perf_counter_init(counter); break; -- cgit v1.2.3 From 3af968e066d593bc4dacc021715f3e95ddf0996f Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Mon, 8 Jun 2009 12:31:53 -0700 Subject: async: Fix lack of boot-time console due to insufficient synchronization Our async work synchronization was broken by "async: make sure independent async domains can't accidentally entangle" (commit d5a877e8dd409d8c702986d06485c374b705d340), because it would report the wrong lowest active async ID when there was both running and pending async work. This caused things like no being able to read the root filesystem, resulting in missing console devices and inability to run 'init', causing a boot-time panic. This fixes it by properly returning the lowest pending async ID: if there is any running async work, that will have a lower ID than any pending work, and we should _not_ look at the pending work list. There were alternative patches from Jaswinder and James, but this one also cleans up the code by removing the pointless 'ret' variable and the unnecesary testing for an empty list around 'for_each_entry()' (if the list is empty, the for_each_entry() thing just won't execute). Fixes-bug: http://bugzilla.kernel.org/show_bug.cgi?id=13474 Reported-and-tested-by: Chris Clayton Cc: Jaswinder Singh Rajput Cc: James Bottomley Cc: Arjan van de Ven Signed-off-by: Linus Torvalds --- kernel/async.c | 15 +++++---------- 1 file changed, 5 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/async.c b/kernel/async.c index 50540301ed0f..27235f5de198 100644 --- a/kernel/async.c +++ b/kernel/async.c @@ -92,23 +92,18 @@ extern int initcall_debug; static async_cookie_t __lowest_in_progress(struct list_head *running) { struct async_entry *entry; - async_cookie_t ret = next_cookie; /* begin with "infinity" value */ if (!list_empty(running)) { entry = list_first_entry(running, struct async_entry, list); - ret = entry->cookie; + return entry->cookie; } - if (!list_empty(&async_pending)) { - list_for_each_entry(entry, &async_pending, list) - if (entry->running == running) { - ret = entry->cookie; - break; - } - } + list_for_each_entry(entry, &async_pending, list) + if (entry->running == running) + return entry->cookie; - return ret; + return next_cookie; /* "infinity" value */ } static async_cookie_t lowest_in_progress(struct list_head *running) -- cgit v1.2.3 From 1f8a6a10fb9437eac3f516ea4324a19087872f30 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Mon, 8 Jun 2009 18:18:39 +0200 Subject: ring-buffer: pass in lockdep class key for reader_lock On Sun, 7 Jun 2009, Ingo Molnar wrote: > Testing tracer sched_switch: <6>Starting ring buffer hammer > PASSED > Testing tracer sysprof: PASSED > Testing tracer function: PASSED > Testing tracer irqsoff: > ============================================= > PASSED > Testing tracer preemptoff: PASSED > Testing tracer preemptirqsoff: [ INFO: possible recursive locking detected ] > PASSED > Testing tracer branch: 2.6.30-rc8-tip-01972-ge5b9078-dirty #5760 > --------------------------------------------- > rb_consumer/431 is trying to acquire lock: > (&cpu_buffer->reader_lock){......}, at: [] ring_buffer_reset_cpu+0x37/0x70 > > but task is already holding lock: > (&cpu_buffer->reader_lock){......}, at: [] ring_buffer_consume+0x7e/0xc0 > > other info that might help us debug this: > 1 lock held by rb_consumer/431: > #0: (&cpu_buffer->reader_lock){......}, at: [] ring_buffer_consume+0x7e/0xc0 The ring buffer is a generic structure, and can be used outside of ftrace. If ftrace traces within the use of the ring buffer, it can produce false positives with lockdep. This patch passes in a static lock key into the allocation of the ring buffer, so that different ring buffers will have their own lock class. Reported-by: Ingo Molnar Signed-off-by: Peter Zijlstra LKML-Reference: <1244477919.13761.9042.camel@twins> [ store key in ring buffer descriptor ] Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 7102d7a2fadb..22878b0d370c 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -426,6 +426,8 @@ struct ring_buffer { atomic_t record_disabled; cpumask_var_t cpumask; + struct lock_class_key *reader_lock_key; + struct mutex mutex; struct ring_buffer_per_cpu **buffers; @@ -565,6 +567,7 @@ rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) cpu_buffer->cpu = cpu; cpu_buffer->buffer = buffer; spin_lock_init(&cpu_buffer->reader_lock); + lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; INIT_LIST_HEAD(&cpu_buffer->pages); @@ -635,7 +638,8 @@ static int rb_cpu_notify(struct notifier_block *self, * when the buffer wraps. If this flag is not set, the buffer will * drop data when the tail hits the head. */ -struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) +struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, + struct lock_class_key *key) { struct ring_buffer *buffer; int bsize; @@ -658,6 +662,7 @@ struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); buffer->flags = flags; buffer->clock = trace_clock_local; + buffer->reader_lock_key = key; /* need at least two pages */ if (buffer->pages == 1) @@ -715,7 +720,7 @@ struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) kfree(buffer); return NULL; } -EXPORT_SYMBOL_GPL(ring_buffer_alloc); +EXPORT_SYMBOL_GPL(__ring_buffer_alloc); /** * ring_buffer_free - free a ring buffer. -- cgit v1.2.3 From eaa958402ea40851097d051f52ba1bb7a885efe9 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Sat, 6 Jun 2009 14:51:36 -0700 Subject: cpumask: alloc zeroed cpumask for static cpumask_var_ts These are defined as static cpumask_var_t so if MAXSMP is not used, they are cleared already. Avoid surprises when MAXSMP is enabled. Signed-off-by: Yinghai Lu Signed-off-by: Rusty Russell --- kernel/sched_cpupri.c | 2 +- kernel/sched_rt.c | 2 +- kernel/smp.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/sched_cpupri.c b/kernel/sched_cpupri.c index cdd3c89574cd..344712a5e3ed 100644 --- a/kernel/sched_cpupri.c +++ b/kernel/sched_cpupri.c @@ -165,7 +165,7 @@ int __init_refok cpupri_init(struct cpupri *cp, bool bootmem) vec->count = 0; if (bootmem) alloc_bootmem_cpumask_var(&vec->mask); - else if (!alloc_cpumask_var(&vec->mask, GFP_KERNEL)) + else if (!zalloc_cpumask_var(&vec->mask, GFP_KERNEL)) goto cleanup; } diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c index f2c66f8f9712..9bf0d2a73045 100644 --- a/kernel/sched_rt.c +++ b/kernel/sched_rt.c @@ -1591,7 +1591,7 @@ static inline void init_sched_rt_class(void) unsigned int i; for_each_possible_cpu(i) - alloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), + zalloc_cpumask_var_node(&per_cpu(local_cpu_mask, i), GFP_KERNEL, cpu_to_node(i)); } #endif /* CONFIG_SMP */ diff --git a/kernel/smp.c b/kernel/smp.c index 858baac568ee..ad63d8501207 100644 --- a/kernel/smp.c +++ b/kernel/smp.c @@ -52,7 +52,7 @@ hotplug_cfd(struct notifier_block *nfb, unsigned long action, void *hcpu) switch (action) { case CPU_UP_PREPARE: case CPU_UP_PREPARE_FROZEN: - if (!alloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL, + if (!zalloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL, cpu_to_node(cpu))) return NOTIFY_BAD; break; -- cgit v1.2.3 From f57a8a1911342265e7acdc190333c4e9235a6632 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Fri, 5 Jun 2009 14:11:30 -0400 Subject: ring-buffer: fix ret in rb_add_time_stamp The update of ret got mistakenly added to the if statement of rb_try_to_discard. The variable ret should be 1 on commit and zero otherwise. [ Impact: fix compiler warning and real bug ] Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 22878b0d370c..2e642b2b7253 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1433,8 +1433,8 @@ rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, /* Darn, this is just wasted space */ event->time_delta = 0; event->array[0] = 0; - ret = 0; } + ret = 0; } *delta = 0; -- cgit v1.2.3 From 55782138e47d9baf2f7d3a7af9e7cf42adf72c56 Mon Sep 17 00:00:00 2001 From: Li Zefan Date: Tue, 9 Jun 2009 13:43:05 +0800 Subject: tracing/events: convert block trace points to TRACE_EVENT() TRACE_EVENT is a more generic way to define tracepoints. Doing so adds these new capabilities to this tracepoint: - zero-copy and per-cpu splice() tracing - binary tracing without printf overhead - structured logging records exposed under /debug/tracing/events - trace events embedded in function tracer output and other plugins - user-defined, per tracepoint filter expressions ... Cons: - no dev_t info for the output of plug, unplug_timer and unplug_io events. no dev_t info for getrq and sleeprq events if bio == NULL. no dev_t info for rq_abort,...,rq_requeue events if rq->rq_disk == NULL. This is mainly because we can't get the deivce from a request queue. But this may change in the future. - A packet command is converted to a string in TP_assign, not TP_print. While blktrace do the convertion just before output. Since pc requests should be rather rare, this is not a big issue. - In blktrace, an event can have 2 different print formats, but a TRACE_EVENT has a unique format, which means we have some unused data in a trace entry. The overhead is minimized by using __dynamic_array() instead of __array(). I've benchmarked the ioctl blktrace vs the splice based TRACE_EVENT tracing: dd dd + ioctl blktrace dd + TRACE_EVENT (splice) 1 7.36s, 42.7 MB/s 7.50s, 42.0 MB/s 7.41s, 42.5 MB/s 2 7.43s, 42.3 MB/s 7.48s, 42.1 MB/s 7.43s, 42.4 MB/s 3 7.38s, 42.6 MB/s 7.45s, 42.2 MB/s 7.41s, 42.5 MB/s So the overhead of tracing is very small, and no regression when using those trace events vs blktrace. And the binary output of TRACE_EVENT is much smaller than blktrace: # ls -l -h -rw-r--r-- 1 root root 8.8M 06-09 13:24 sda.blktrace.0 -rw-r--r-- 1 root root 195K 06-09 13:24 sda.blktrace.1 -rw-r--r-- 1 root root 2.7M 06-09 13:25 trace_splice.out Following are some comparisons between TRACE_EVENT and blktrace: plug: kjournald-480 [000] 303.084981: block_plug: [kjournald] kjournald-480 [000] 303.084981: 8,0 P N [kjournald] unplug_io: kblockd/0-118 [000] 300.052973: block_unplug_io: [kblockd/0] 1 kblockd/0-118 [000] 300.052974: 8,0 U N [kblockd/0] 1 remap: kjournald-480 [000] 303.085042: block_remap: 8,0 W 102736992 + 8 <- (8,8) 33384 kjournald-480 [000] 303.085043: 8,0 A W 102736992 + 8 <- (8,8) 33384 bio_backmerge: kjournald-480 [000] 303.085086: block_bio_backmerge: 8,0 W 102737032 + 8 [kjournald] kjournald-480 [000] 303.085086: 8,0 M W 102737032 + 8 [kjournald] getrq: kjournald-480 [000] 303.084974: block_getrq: 8,0 W 102736984 + 8 [kjournald] kjournald-480 [000] 303.084975: 8,0 G W 102736984 + 8 [kjournald] bash-2066 [001] 1072.953770: 8,0 G N [bash] bash-2066 [001] 1072.953773: block_getrq: 0,0 N 0 + 0 [bash] rq_complete: konsole-2065 [001] 300.053184: block_rq_complete: 8,0 W () 103669040 + 16 [0] konsole-2065 [001] 300.053191: 8,0 C W 103669040 + 16 [0] ksoftirqd/1-7 [001] 1072.953811: 8,0 C N (5a 00 08 00 00 00 00 00 24 00) [0] ksoftirqd/1-7 [001] 1072.953813: block_rq_complete: 0,0 N (5a 00 08 00 00 00 00 00 24 00) 0 + 0 [0] rq_insert: kjournald-480 [000] 303.084985: block_rq_insert: 8,0 W 0 () 102736984 + 8 [kjournald] kjournald-480 [000] 303.084986: 8,0 I W 102736984 + 8 [kjournald] Changelog from v2 -> v3: - use the newly introduced __dynamic_array(). Changelog from v1 -> v2: - use __string() instead of __array() to minimize the memory required to store hex dump of rq->cmd(). - support large pc requests. - add missing blk_fill_rwbs_rq() in block_rq_requeue TRACE_EVENT. - some cleanups. Signed-off-by: Li Zefan LKML-Reference: <4A2DF669.5070905@cn.fujitsu.com> Signed-off-by: Steven Rostedt --- kernel/trace/Makefile | 5 +++- kernel/trace/blktrace.c | 78 ++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 81 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile index 06b85850fab4..844164dca90a 100644 --- a/kernel/trace/Makefile +++ b/kernel/trace/Makefile @@ -45,7 +45,10 @@ obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o obj-$(CONFIG_POWER_TRACER) += trace_power.o obj-$(CONFIG_KMEMTRACE) += kmemtrace.o obj-$(CONFIG_WORKQUEUE_TRACER) += trace_workqueue.o -obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o +obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o +ifeq ($(CONFIG_BLOCK),y) +obj-$(CONFIG_EVENT_TRACING) += blktrace.o +endif obj-$(CONFIG_EVENT_TRACING) += trace_events.o obj-$(CONFIG_EVENT_TRACING) += trace_export.o obj-$(CONFIG_FTRACE_SYSCALLS) += trace_syscalls.o diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c index e3abf55bc8e5..7bd6a9893c24 100644 --- a/kernel/trace/blktrace.c +++ b/kernel/trace/blktrace.c @@ -23,10 +23,14 @@ #include #include #include -#include #include + +#include + #include "trace_output.h" +#ifdef CONFIG_BLK_DEV_IO_TRACE + static unsigned int blktrace_seq __read_mostly = 1; static struct trace_array *blk_tr; @@ -1658,3 +1662,75 @@ int blk_trace_init_sysfs(struct device *dev) return sysfs_create_group(&dev->kobj, &blk_trace_attr_group); } +#endif /* CONFIG_BLK_DEV_IO_TRACE */ + +#ifdef CONFIG_EVENT_TRACING + +void blk_dump_cmd(char *buf, struct request *rq) +{ + int i, end; + int len = rq->cmd_len; + unsigned char *cmd = rq->cmd; + + if (!blk_pc_request(rq)) { + buf[0] = '\0'; + return; + } + + for (end = len - 1; end >= 0; end--) + if (cmd[end]) + break; + end++; + + for (i = 0; i < len; i++) { + buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]); + if (i == end && end != len - 1) { + sprintf(buf, " .."); + break; + } + } +} + +void blk_fill_rwbs(char *rwbs, u32 rw, int bytes) +{ + int i = 0; + + if (rw & WRITE) + rwbs[i++] = 'W'; + else if (rw & 1 << BIO_RW_DISCARD) + rwbs[i++] = 'D'; + else if (bytes) + rwbs[i++] = 'R'; + else + rwbs[i++] = 'N'; + + if (rw & 1 << BIO_RW_AHEAD) + rwbs[i++] = 'A'; + if (rw & 1 << BIO_RW_BARRIER) + rwbs[i++] = 'B'; + if (rw & 1 << BIO_RW_SYNCIO) + rwbs[i++] = 'S'; + if (rw & 1 << BIO_RW_META) + rwbs[i++] = 'M'; + + rwbs[i] = '\0'; +} + +void blk_fill_rwbs_rq(char *rwbs, struct request *rq) +{ + int rw = rq->cmd_flags & 0x03; + int bytes; + + if (blk_discard_rq(rq)) + rw |= (1 << BIO_RW_DISCARD); + + if (blk_pc_request(rq)) + bytes = rq->data_len; + else + bytes = rq->hard_nr_sectors << 9; + + blk_fill_rwbs(rwbs, rw, bytes); +} + +#endif /* CONFIG_EVENT_TRACING */ + -- cgit v1.2.3 From 725c624a58a10ef90a2ff889e122158fabf36147 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Mon, 8 Jun 2009 19:09:45 -0400 Subject: tracing: add trace_seq_vprint interface The code to update the print formats for events requires a vprintf format in the trace_seq. This patch adds that interface. Signed-off-by: Steven Rostedt --- kernel/trace/trace_output.c | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+) (limited to 'kernel') diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index 425725c1622d..c05aff465dc9 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -100,6 +100,38 @@ trace_seq_printf(struct trace_seq *s, const char *fmt, ...) } EXPORT_SYMBOL_GPL(trace_seq_printf); +/** + * trace_seq_vprintf - sequence printing of trace information + * @s: trace sequence descriptor + * @fmt: printf format string + * + * The tracer may use either sequence operations or its own + * copy to user routines. To simplify formating of a trace + * trace_seq_printf is used to store strings into a special + * buffer (@s). Then the output may be either used by + * the sequencer or pulled into another buffer. + */ +int +trace_seq_vprintf(struct trace_seq *s, const char *fmt, va_list args) +{ + int len = (PAGE_SIZE - 1) - s->len; + int ret; + + if (!len) + return 0; + + ret = vsnprintf(s->buffer + s->len, len, fmt, args); + + /* If we can't write it all, don't bother writing anything */ + if (ret >= len) + return 0; + + s->len += ret; + + return len; +} +EXPORT_SYMBOL_GPL(trace_seq_vprintf); + int trace_seq_bprintf(struct trace_seq *s, const char *fmt, const u32 *binary) { int len = (PAGE_SIZE - 1) - s->len; -- cgit v1.2.3 From ad6ccfad6f759a5d657dabe2071a8f2a503fcc84 Mon Sep 17 00:00:00 2001 From: Manish Katiyar Date: Tue, 12 May 2009 13:43:35 -0700 Subject: kernel/kallsyms.c: replace deprecated __initcall with device_initcall and fix whitespace Fix coding style whitespace issues and replace __initcall with device_initcall. Fixed multi-line comments as per coding style. Errors as reported by checkpatch.pl :- Before: total: 14 errors, 14 warnings, 487 lines checked After : total: 0 errors, 8 warnings, 507 lines checked Compile tested binary verified as :- Before: text data bss dec hex filename 2405 4 0 2409 969 kernel/kallsyms.o After : text data bss dec hex filename 2405 4 0 2409 969 kernel/kallsyms.o Signed-off-by: Manish Katiyar Signed-off-by: Andrew Morton Signed-off-by: Sam Ravnborg --- kernel/kallsyms.c | 134 +++++++++++++++++++++++++++++++----------------------- 1 file changed, 78 insertions(+), 56 deletions(-) (limited to 'kernel') diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c index 374faf9bfdc7..3a29dbe7898e 100644 --- a/kernel/kallsyms.c +++ b/kernel/kallsyms.c @@ -30,12 +30,16 @@ #define all_var 0 #endif -/* These will be re-linked against their real values during the second link stage */ +/* + * These will be re-linked against their real values + * during the second link stage. + */ extern const unsigned long kallsyms_addresses[] __attribute__((weak)); extern const u8 kallsyms_names[] __attribute__((weak)); -/* tell the compiler that the count isn't in the small data section if the arch - * has one (eg: FRV) +/* + * Tell the compiler that the count isn't in the small data section if the arch + * has one (eg: FRV). */ extern const unsigned long kallsyms_num_syms __attribute__((weak, section(".rodata"))); @@ -75,31 +79,37 @@ static int is_ksym_addr(unsigned long addr) return is_kernel_text(addr) || is_kernel_inittext(addr); } -/* expand a compressed symbol data into the resulting uncompressed string, - given the offset to where the symbol is in the compressed stream */ +/* + * Expand a compressed symbol data into the resulting uncompressed string, + * given the offset to where the symbol is in the compressed stream. + */ static unsigned int kallsyms_expand_symbol(unsigned int off, char *result) { int len, skipped_first = 0; const u8 *tptr, *data; - /* get the compressed symbol length from the first symbol byte */ + /* Get the compressed symbol length from the first symbol byte. */ data = &kallsyms_names[off]; len = *data; data++; - /* update the offset to return the offset for the next symbol on - * the compressed stream */ + /* + * Update the offset to return the offset for the next symbol on + * the compressed stream. + */ off += len + 1; - /* for every byte on the compressed symbol data, copy the table - entry for that byte */ - while(len) { - tptr = &kallsyms_token_table[ kallsyms_token_index[*data] ]; + /* + * For every byte on the compressed symbol data, copy the table + * entry for that byte. + */ + while (len) { + tptr = &kallsyms_token_table[kallsyms_token_index[*data]]; data++; len--; while (*tptr) { - if(skipped_first) { + if (skipped_first) { *result = *tptr; result++; } else @@ -110,36 +120,46 @@ static unsigned int kallsyms_expand_symbol(unsigned int off, char *result) *result = '\0'; - /* return to offset to the next symbol */ + /* Return to offset to the next symbol. */ return off; } -/* get symbol type information. This is encoded as a single char at the - * begining of the symbol name */ +/* + * Get symbol type information. This is encoded as a single char at the + * beginning of the symbol name. + */ static char kallsyms_get_symbol_type(unsigned int off) { - /* get just the first code, look it up in the token table, and return the - * first char from this token */ - return kallsyms_token_table[ kallsyms_token_index[ kallsyms_names[off+1] ] ]; + /* + * Get just the first code, look it up in the token table, + * and return the first char from this token. + */ + return kallsyms_token_table[kallsyms_token_index[kallsyms_names[off + 1]]]; } -/* find the offset on the compressed stream given and index in the - * kallsyms array */ +/* + * Find the offset on the compressed stream given and index in the + * kallsyms array. + */ static unsigned int get_symbol_offset(unsigned long pos) { const u8 *name; int i; - /* use the closest marker we have. We have markers every 256 positions, - * so that should be close enough */ - name = &kallsyms_names[ kallsyms_markers[pos>>8] ]; + /* + * Use the closest marker we have. We have markers every 256 positions, + * so that should be close enough. + */ + name = &kallsyms_names[kallsyms_markers[pos >> 8]]; - /* sequentially scan all the symbols up to the point we're searching for. - * Every symbol is stored in a [][ bytes of data] format, so we - * just need to add the len to the current pointer for every symbol we - * wish to skip */ - for(i = 0; i < (pos&0xFF); i++) + /* + * Sequentially scan all the symbols up to the point we're searching + * for. Every symbol is stored in a [][ bytes of data] format, + * so we just need to add the len to the current pointer for every + * symbol we wish to skip. + */ + for (i = 0; i < (pos & 0xFF); i++) name = name + (*name) + 1; return name - kallsyms_names; @@ -190,7 +210,7 @@ static unsigned long get_symbol_pos(unsigned long addr, /* This kernel should never had been booted. */ BUG_ON(!kallsyms_addresses); - /* do a binary search on the sorted kallsyms_addresses array */ + /* Do a binary search on the sorted kallsyms_addresses array. */ low = 0; high = kallsyms_num_syms; @@ -203,15 +223,15 @@ static unsigned long get_symbol_pos(unsigned long addr, } /* - * search for the first aliased symbol. Aliased - * symbols are symbols with the same address + * Search for the first aliased symbol. Aliased + * symbols are symbols with the same address. */ while (low && kallsyms_addresses[low-1] == kallsyms_addresses[low]) --low; symbol_start = kallsyms_addresses[low]; - /* Search for next non-aliased symbol */ + /* Search for next non-aliased symbol. */ for (i = low + 1; i < kallsyms_num_syms; i++) { if (kallsyms_addresses[i] > symbol_start) { symbol_end = kallsyms_addresses[i]; @@ -219,7 +239,7 @@ static unsigned long get_symbol_pos(unsigned long addr, } } - /* if we found no next symbol, we use the end of the section */ + /* If we found no next symbol, we use the end of the section. */ if (!symbol_end) { if (is_kernel_inittext(addr)) symbol_end = (unsigned long)_einittext; @@ -252,10 +272,10 @@ int kallsyms_lookup_size_offset(unsigned long addr, unsigned long *symbolsize, /* * Lookup an address - * - modname is set to NULL if it's in the kernel - * - we guarantee that the returned name is valid until we reschedule even if - * it resides in a module - * - we also guarantee that modname will be valid until rescheduled + * - modname is set to NULL if it's in the kernel. + * - We guarantee that the returned name is valid until we reschedule even if. + * It resides in a module. + * - We also guarantee that modname will be valid until rescheduled. */ const char *kallsyms_lookup(unsigned long addr, unsigned long *symbolsize, @@ -276,7 +296,7 @@ const char *kallsyms_lookup(unsigned long addr, return namebuf; } - /* see if it's in a module */ + /* See if it's in a module. */ return module_address_lookup(addr, symbolsize, offset, modname, namebuf); } @@ -294,7 +314,7 @@ int lookup_symbol_name(unsigned long addr, char *symname) kallsyms_expand_symbol(get_symbol_offset(pos), symname); return 0; } - /* see if it's in a module */ + /* See if it's in a module. */ return lookup_module_symbol_name(addr, symname); } @@ -313,7 +333,7 @@ int lookup_symbol_attrs(unsigned long addr, unsigned long *size, modname[0] = '\0'; return 0; } - /* see if it's in a module */ + /* See if it's in a module. */ return lookup_module_symbol_attrs(addr, size, offset, modname, name); } @@ -342,6 +362,7 @@ int sprint_symbol(char *buffer, unsigned long address) return len; } +EXPORT_SYMBOL_GPL(sprint_symbol); /* Look up a kernel symbol and print it to the kernel messages. */ void __print_symbol(const char *fmt, unsigned long address) @@ -352,13 +373,13 @@ void __print_symbol(const char *fmt, unsigned long address) printk(fmt, buffer); } +EXPORT_SYMBOL(__print_symbol); /* To avoid using get_symbol_offset for every symbol, we carry prefix along. */ -struct kallsym_iter -{ +struct kallsym_iter { loff_t pos; unsigned long value; - unsigned int nameoff; /* If iterating in core kernel symbols */ + unsigned int nameoff; /* If iterating in core kernel symbols. */ char type; char name[KSYM_NAME_LEN]; char module_name[MODULE_NAME_LEN]; @@ -404,7 +425,7 @@ static int update_iter(struct kallsym_iter *iter, loff_t pos) iter->pos = pos; return get_ksymbol_mod(iter); } - + /* If we're not on the desired position, reset to new position. */ if (pos != iter->pos) reset_iter(iter, pos); @@ -439,23 +460,25 @@ static int s_show(struct seq_file *m, void *p) { struct kallsym_iter *iter = m->private; - /* Some debugging symbols have no name. Ignore them. */ + /* Some debugging symbols have no name. Ignore them. */ if (!iter->name[0]) return 0; if (iter->module_name[0]) { char type; - /* Label it "global" if it is exported, - * "local" if not exported. */ + /* + * Label it "global" if it is exported, + * "local" if not exported. + */ type = iter->exported ? toupper(iter->type) : tolower(iter->type); seq_printf(m, "%0*lx %c %s\t[%s]\n", - (int)(2*sizeof(void*)), + (int)(2 * sizeof(void *)), iter->value, type, iter->name, iter->module_name); } else seq_printf(m, "%0*lx %c %s\n", - (int)(2*sizeof(void*)), + (int)(2 * sizeof(void *)), iter->value, iter->type, iter->name); return 0; } @@ -469,9 +492,11 @@ static const struct seq_operations kallsyms_op = { static int kallsyms_open(struct inode *inode, struct file *file) { - /* We keep iterator in m->private, since normal case is to + /* + * We keep iterator in m->private, since normal case is to * s_start from where we left off, so we avoid doing - * using get_symbol_offset for every symbol */ + * using get_symbol_offset for every symbol. + */ struct kallsym_iter *iter; int ret; @@ -500,7 +525,4 @@ static int __init kallsyms_init(void) proc_create("kallsyms", 0444, NULL, &kallsyms_operations); return 0; } -__initcall(kallsyms_init); - -EXPORT_SYMBOL(__print_symbol); -EXPORT_SYMBOL_GPL(sprint_symbol); +device_initcall(kallsyms_init); -- cgit v1.2.3 From 110bf2b764eb6026b868d84499263cb24b1bcc8d Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Tue, 9 Jun 2009 17:29:07 -0400 Subject: tracing: add protection around module events unload When reading the trace buffer, there is a race that when a module is unloaded it removes events that is stilled referenced in the buffers. This patch adds the protection around the unloading of the events from modules and the reading of the trace buffers. Signed-off-by: Steven Rostedt --- kernel/trace/trace_events.c | 4 +++- kernel/trace/trace_output.c | 15 ++++++++++++--- kernel/trace/trace_output.h | 4 ++++ 3 files changed, 19 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c index 6c81f9c21426..aa08be69a1b6 100644 --- a/kernel/trace/trace_events.c +++ b/kernel/trace/trace_events.c @@ -1050,12 +1050,13 @@ static void trace_module_remove_events(struct module *mod) struct ftrace_event_call *call, *p; bool found = false; + down_write(&trace_event_mutex); list_for_each_entry_safe(call, p, &ftrace_events, list) { if (call->mod == mod) { found = true; ftrace_event_enable_disable(call, 0); if (call->event) - unregister_ftrace_event(call->event); + __unregister_ftrace_event(call->event); debugfs_remove_recursive(call->dir); list_del(&call->list); trace_destroy_fields(call); @@ -1079,6 +1080,7 @@ static void trace_module_remove_events(struct module *mod) */ if (found) tracing_reset_current_online_cpus(); + up_write(&trace_event_mutex); } static int trace_module_notify(struct notifier_block *self, diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c index c05aff465dc9..7938f3ae93e3 100644 --- a/kernel/trace/trace_output.c +++ b/kernel/trace/trace_output.c @@ -14,7 +14,7 @@ /* must be a power of 2 */ #define EVENT_HASHSIZE 128 -static DECLARE_RWSEM(trace_event_mutex); +DECLARE_RWSEM(trace_event_mutex); DEFINE_PER_CPU(struct trace_seq, ftrace_event_seq); EXPORT_PER_CPU_SYMBOL(ftrace_event_seq); @@ -702,6 +702,16 @@ int register_ftrace_event(struct trace_event *event) } EXPORT_SYMBOL_GPL(register_ftrace_event); +/* + * Used by module code with the trace_event_mutex held for write. + */ +int __unregister_ftrace_event(struct trace_event *event) +{ + hlist_del(&event->node); + list_del(&event->list); + return 0; +} + /** * unregister_ftrace_event - remove a no longer used event * @event: the event to remove @@ -709,8 +719,7 @@ EXPORT_SYMBOL_GPL(register_ftrace_event); int unregister_ftrace_event(struct trace_event *event) { down_write(&trace_event_mutex); - hlist_del(&event->node); - list_del(&event->list); + __unregister_ftrace_event(event); up_write(&trace_event_mutex); return 0; diff --git a/kernel/trace/trace_output.h b/kernel/trace/trace_output.h index ac240e76eb01..d38bec4a9c30 100644 --- a/kernel/trace/trace_output.h +++ b/kernel/trace/trace_output.h @@ -27,6 +27,10 @@ extern struct trace_event *ftrace_find_event(int type); extern enum print_line_t trace_nop_print(struct trace_iterator *iter, int flags); +/* used by module unregistering */ +extern int __unregister_ftrace_event(struct trace_event *event); +extern struct rw_semaphore trace_event_mutex; + #define MAX_MEMHEX_BYTES 8 #define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1) -- cgit v1.2.3 From bd2b5b12849a3446abad0b25e920f86f5480b309 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 10 Jun 2009 13:40:57 +0200 Subject: perf_counter: More aggressive frequency adjustment Also employ the overflow handler to adjust the frequency, this results in a stable frequency in about 40~50 samples, instead of that many ticks. This also means we can start sampling at a sample period of 1 without running head-first into the throttle. It relies on sched_clock() to accurately measure the time difference between the overflow NMIs. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 130 ++++++++++++++++++++++++++++++++++---------------- 1 file changed, 88 insertions(+), 42 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 5eacaaf3f9cd..51c571ee4d0b 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1184,13 +1184,33 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) static void perf_log_throttle(struct perf_counter *counter, int enable); static void perf_log_period(struct perf_counter *counter, u64 period); -static void perf_adjust_freq(struct perf_counter_context *ctx) +static void perf_adjust_period(struct perf_counter *counter, u64 events) +{ + struct hw_perf_counter *hwc = &counter->hw; + u64 period, sample_period; + s64 delta; + + events *= hwc->sample_period; + period = div64_u64(events, counter->attr.sample_freq); + + delta = (s64)(period - hwc->sample_period); + delta = (delta + 7) / 8; /* low pass filter */ + + sample_period = hwc->sample_period + delta; + + if (!sample_period) + sample_period = 1; + + perf_log_period(counter, sample_period); + + hwc->sample_period = sample_period; +} + +static void perf_ctx_adjust_freq(struct perf_counter_context *ctx) { struct perf_counter *counter; struct hw_perf_counter *hwc; - u64 interrupts, sample_period; - u64 events, period, freq; - s64 delta; + u64 interrupts, freq; spin_lock(&ctx->lock); list_for_each_entry(counter, &ctx->counter_list, list_entry) { @@ -1202,6 +1222,9 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) interrupts = hwc->interrupts; hwc->interrupts = 0; + /* + * unthrottle counters on the tick + */ if (interrupts == MAX_INTERRUPTS) { perf_log_throttle(counter, 1); counter->pmu->unthrottle(counter); @@ -1211,6 +1234,9 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) if (!counter->attr.freq || !counter->attr.sample_freq) continue; + /* + * if the specified freq < HZ then we need to skip ticks + */ if (counter->attr.sample_freq < HZ) { freq = counter->attr.sample_freq; @@ -1226,20 +1252,20 @@ static void perf_adjust_freq(struct perf_counter_context *ctx) } else freq = HZ; - events = freq * interrupts * hwc->sample_period; - period = div64_u64(events, counter->attr.sample_freq); - - delta = (s64)(1 + period - hwc->sample_period); - delta >>= 1; - - sample_period = hwc->sample_period + delta; - - if (!sample_period) - sample_period = 1; + perf_adjust_period(counter, freq * interrupts); - perf_log_period(counter, sample_period); - - hwc->sample_period = sample_period; + /* + * In order to avoid being stalled by an (accidental) huge + * sample period, force reset the sample period if we didn't + * get any events in this freq period. + */ + if (!interrupts) { + perf_disable(); + counter->pmu->disable(counter); + atomic_set(&hwc->period_left, 0); + counter->pmu->enable(counter); + perf_enable(); + } } spin_unlock(&ctx->lock); } @@ -1279,9 +1305,9 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu) cpuctx = &per_cpu(perf_cpu_context, cpu); ctx = curr->perf_counter_ctxp; - perf_adjust_freq(&cpuctx->ctx); + perf_ctx_adjust_freq(&cpuctx->ctx); if (ctx) - perf_adjust_freq(ctx); + perf_ctx_adjust_freq(ctx); perf_counter_cpu_sched_out(cpuctx); if (ctx) @@ -1647,10 +1673,10 @@ static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) counter->attr.sample_freq = value; } else { + perf_log_period(counter, value); + counter->attr.sample_period = value; counter->hw.sample_period = value; - - perf_log_period(counter, value); } unlock: spin_unlock_irq(&ctx->lock); @@ -2853,35 +2879,41 @@ void __perf_counter_mmap(struct vm_area_struct *vma) * event flow. */ +struct freq_event { + struct perf_event_header header; + u64 time; + u64 id; + u64 period; +}; + static void perf_log_period(struct perf_counter *counter, u64 period) { struct perf_output_handle handle; + struct freq_event event; int ret; - struct { - struct perf_event_header header; - u64 time; - u64 id; - u64 period; - } freq_event = { + if (counter->hw.sample_period == period) + return; + + if (counter->attr.sample_type & PERF_SAMPLE_PERIOD) + return; + + event = (struct freq_event) { .header = { .type = PERF_EVENT_PERIOD, .misc = 0, - .size = sizeof(freq_event), + .size = sizeof(event), }, .time = sched_clock(), .id = counter->id, .period = period, }; - if (counter->hw.sample_period == period) - return; - - ret = perf_output_begin(&handle, counter, sizeof(freq_event), 0, 0); + ret = perf_output_begin(&handle, counter, sizeof(event), 1, 0); if (ret) return; - perf_output_put(&handle, freq_event); + perf_output_put(&handle, event); perf_output_end(&handle); } @@ -2923,15 +2955,16 @@ int perf_counter_overflow(struct perf_counter *counter, { int events = atomic_read(&counter->event_limit); int throttle = counter->pmu->unthrottle != NULL; + struct hw_perf_counter *hwc = &counter->hw; int ret = 0; if (!throttle) { - counter->hw.interrupts++; + hwc->interrupts++; } else { - if (counter->hw.interrupts != MAX_INTERRUPTS) { - counter->hw.interrupts++; - if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) { - counter->hw.interrupts = MAX_INTERRUPTS; + if (hwc->interrupts != MAX_INTERRUPTS) { + hwc->interrupts++; + if (HZ * hwc->interrupts > (u64)sysctl_perf_counter_limit) { + hwc->interrupts = MAX_INTERRUPTS; perf_log_throttle(counter, 0); ret = 1; } @@ -2945,6 +2978,16 @@ int perf_counter_overflow(struct perf_counter *counter, } } + if (counter->attr.freq) { + u64 now = sched_clock(); + s64 delta = now - hwc->freq_stamp; + + hwc->freq_stamp = now; + + if (delta > 0 && delta < TICK_NSEC) + perf_adjust_period(counter, NSEC_PER_SEC / (int)delta); + } + /* * XXX event_limit might not quite work as expected on inherited * counters @@ -3379,7 +3422,6 @@ static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) return NULL; counter->destroy = tp_perf_counter_destroy; - counter->hw.sample_period = counter->attr.sample_period; return &perf_ops_generic; } @@ -3483,10 +3525,11 @@ perf_counter_alloc(struct perf_counter_attr *attr, pmu = NULL; hwc = &counter->hw; + hwc->sample_period = attr->sample_period; if (attr->freq && attr->sample_freq) - hwc->sample_period = div64_u64(TICK_NSEC, attr->sample_freq); - else - hwc->sample_period = attr->sample_period; + hwc->sample_period = 1; + + atomic64_set(&hwc->period_left, hwc->sample_period); /* * we currently do not support PERF_SAMPLE_GROUP on inherited counters @@ -3687,6 +3730,9 @@ inherit_counter(struct perf_counter *parent_counter, else child_counter->state = PERF_COUNTER_STATE_OFF; + if (parent_counter->attr.freq) + child_counter->hw.sample_period = parent_counter->hw.sample_period; + /* * Link it up in the child's context: */ -- cgit v1.2.3 From 66fff22483d8542dfb4d61a28d21277bbde321e8 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 10 Jun 2009 22:53:37 +0200 Subject: perf_counter: Annotate exit ctx recursion Ever since Paul fixed it to unclone the context before taking the ctx->lock this became a false positive, annotate it away. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 51c571ee4d0b..ae591a1275a6 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3879,7 +3879,18 @@ void perf_counter_exit_task(struct task_struct *child) spin_unlock(&child_ctx->lock); local_irq_restore(flags); - mutex_lock(&child_ctx->mutex); + /* + * We can recurse on the same lock type through: + * + * __perf_counter_exit_task() + * sync_child_counter() + * fput(parent_counter->filp) + * perf_release() + * mutex_lock(&ctx->mutex) + * + * But since its the parent context it won't be the same instance. + */ + mutex_lock_nested(&child_ctx->mutex, SINGLE_DEPTH_NESTING); again: list_for_each_entry_safe(child_counter, tmp, &child_ctx->counter_list, -- cgit v1.2.3 From df1a132bf3d3508f863336c80a27806a2ac947e0 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 10 Jun 2009 21:02:22 +0200 Subject: perf_counter: Introduce struct for sample data For easy extension of the sample data, put it in a structure. Signed-off-by: Peter Zijlstra Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 38 ++++++++++++++++++++++---------------- 1 file changed, 22 insertions(+), 16 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ae591a1275a6..4fe85e804f43 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2378,8 +2378,8 @@ static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p) return task_pid_nr_ns(p, counter->ns); } -static void perf_counter_output(struct perf_counter *counter, - int nmi, struct pt_regs *regs, u64 addr) +static void perf_counter_output(struct perf_counter *counter, int nmi, + struct perf_sample_data *data) { int ret; u64 sample_type = counter->attr.sample_type; @@ -2404,10 +2404,10 @@ static void perf_counter_output(struct perf_counter *counter, header.size = sizeof(header); header.misc = PERF_EVENT_MISC_OVERFLOW; - header.misc |= perf_misc_flags(regs); + header.misc |= perf_misc_flags(data->regs); if (sample_type & PERF_SAMPLE_IP) { - ip = perf_instruction_pointer(regs); + ip = perf_instruction_pointer(data->regs); header.type |= PERF_SAMPLE_IP; header.size += sizeof(ip); } @@ -2460,7 +2460,7 @@ static void perf_counter_output(struct perf_counter *counter, } if (sample_type & PERF_SAMPLE_CALLCHAIN) { - callchain = perf_callchain(regs); + callchain = perf_callchain(data->regs); if (callchain) { callchain_size = (1 + callchain->nr) * sizeof(u64); @@ -2486,7 +2486,7 @@ static void perf_counter_output(struct perf_counter *counter, perf_output_put(&handle, time); if (sample_type & PERF_SAMPLE_ADDR) - perf_output_put(&handle, addr); + perf_output_put(&handle, data->addr); if (sample_type & PERF_SAMPLE_ID) perf_output_put(&handle, counter->id); @@ -2950,8 +2950,8 @@ static void perf_log_throttle(struct perf_counter *counter, int enable) * Generic counter overflow handling. */ -int perf_counter_overflow(struct perf_counter *counter, - int nmi, struct pt_regs *regs, u64 addr) +int perf_counter_overflow(struct perf_counter *counter, int nmi, + struct perf_sample_data *data) { int events = atomic_read(&counter->event_limit); int throttle = counter->pmu->unthrottle != NULL; @@ -3005,7 +3005,7 @@ int perf_counter_overflow(struct perf_counter *counter, perf_counter_disable(counter); } - perf_counter_output(counter, nmi, regs, addr); + perf_counter_output(counter, nmi, data); return ret; } @@ -3054,24 +3054,25 @@ static void perf_swcounter_set_period(struct perf_counter *counter) static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) { enum hrtimer_restart ret = HRTIMER_RESTART; + struct perf_sample_data data; struct perf_counter *counter; - struct pt_regs *regs; u64 period; counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); counter->pmu->read(counter); - regs = get_irq_regs(); + data.addr = 0; + data.regs = get_irq_regs(); /* * In case we exclude kernel IPs or are somehow not in interrupt * context, provide the next best thing, the user IP. */ - if ((counter->attr.exclude_kernel || !regs) && + if ((counter->attr.exclude_kernel || !data.regs) && !counter->attr.exclude_user) - regs = task_pt_regs(current); + data.regs = task_pt_regs(current); - if (regs) { - if (perf_counter_overflow(counter, 0, regs, 0)) + if (data.regs) { + if (perf_counter_overflow(counter, 0, &data)) ret = HRTIMER_NORESTART; } @@ -3084,9 +3085,14 @@ static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) static void perf_swcounter_overflow(struct perf_counter *counter, int nmi, struct pt_regs *regs, u64 addr) { + struct perf_sample_data data = { + .regs = regs, + .addr = addr, + }; + perf_swcounter_update(counter); perf_swcounter_set_period(counter); - if (perf_counter_overflow(counter, nmi, regs, addr)) + if (perf_counter_overflow(counter, nmi, &data)) /* soft-disable the counter */ ; -- cgit v1.2.3 From 9e350de37ac9607012fcf9c5314a28fbddf8f43c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 10 Jun 2009 21:34:59 +0200 Subject: perf_counter: Accurate period data We currently log hw.sample_period for PERF_SAMPLE_PERIOD, however this is incorrect. When we adjust the period, it will only take effect the next cycle but report it for the current cycle. So when we adjust the period for every cycle, we're always wrong. Solve this by keeping track of the last_period. Signed-off-by: Peter Zijlstra Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 4fe85e804f43..8b89b40bd0f0 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2495,7 +2495,7 @@ static void perf_counter_output(struct perf_counter *counter, int nmi, perf_output_put(&handle, cpu_entry); if (sample_type & PERF_SAMPLE_PERIOD) - perf_output_put(&handle, counter->hw.sample_period); + perf_output_put(&handle, data->period); /* * XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult. @@ -3040,11 +3040,13 @@ static void perf_swcounter_set_period(struct perf_counter *counter) if (unlikely(left <= -period)) { left = period; atomic64_set(&hwc->period_left, left); + hwc->last_period = period; } if (unlikely(left <= 0)) { left += period; atomic64_add(period, &hwc->period_left); + hwc->last_period = period; } atomic64_set(&hwc->prev_count, -left); @@ -3086,8 +3088,9 @@ static void perf_swcounter_overflow(struct perf_counter *counter, int nmi, struct pt_regs *regs, u64 addr) { struct perf_sample_data data = { - .regs = regs, - .addr = addr, + .regs = regs, + .addr = addr, + .period = counter->hw.last_period, }; perf_swcounter_update(counter); -- cgit v1.2.3 From 3f68535adad8dd89499505a65fb25d0e02d118cc Mon Sep 17 00:00:00 2001 From: john stultz Date: Wed, 21 Jan 2009 22:53:22 -0700 Subject: clocksource: sanity check sysfs clocksource changes Thomas, Andrew and Ingo pointed out that we don't have any safety checks in the clocksource sysfs entries to make sure sysadmins don't try to change the clocksource to a non high-res timer capable clocksource (such as jiffies) when high-res timers (HRT) is enabled. Doing so will likely hang a system. Correct this by filtering non HRT clocksources from available_clocksources and not accepting non HRT clocksources with HRT enabled. Signed-off-by: John Stultz Signed-off-by: Andrew Morton Signed-off-by: Thomas Gleixner --- kernel/hrtimer.c | 4 ++-- kernel/time/clocksource.c | 18 +++++++++++++++++- 2 files changed, 19 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index cb8a15c19583..1a70c18cdffe 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -476,7 +476,7 @@ static inline int hrtimer_is_hres_enabled(void) /* * Is the high resolution mode active ? */ -static inline int hrtimer_hres_active(void) +int hrtimer_hres_active(void) { return __get_cpu_var(hrtimer_bases).hres_active; } @@ -704,7 +704,7 @@ static int hrtimer_switch_to_hres(void) #else -static inline int hrtimer_hres_active(void) { return 0; } +int hrtimer_hres_active(void) { return 0; } static inline int hrtimer_is_hres_enabled(void) { return 0; } static inline int hrtimer_switch_to_hres(void) { return 0; } static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c index 80189f6f1c5a..18b9f5da4ee9 100644 --- a/kernel/time/clocksource.c +++ b/kernel/time/clocksource.c @@ -30,6 +30,7 @@ #include #include /* for spin_unlock_irq() using preempt_count() m68k */ #include +#include void timecounter_init(struct timecounter *tc, const struct cyclecounter *cc, @@ -509,6 +510,18 @@ static ssize_t sysfs_override_clocksource(struct sys_device *dev, } } + /* + * Check to make sure we don't switch to a non-HRT usable + * clocksource if HRT is enabled and running + */ + if (hrtimer_hres_active() && + !(ovr->flags & CLOCK_SOURCE_VALID_FOR_HRES)) { + printk(KERN_WARNING "%s clocksource is not HRT compatible. " + "Cannot switch while in HRT mode\n", ovr->name); + ovr = NULL; + override_name[0] = 0; + } + /* Reselect, when the override name has changed */ if (ovr != clocksource_override) { clocksource_override = ovr; @@ -537,7 +550,10 @@ sysfs_show_available_clocksources(struct sys_device *dev, spin_lock_irq(&clocksource_lock); list_for_each_entry(src, &clocksource_list, list) { - count += snprintf(buf + count, + /* Don't show non-HRES clocksource if HRES is enabled */ + if (!hrtimer_hres_active() || + (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) + count += snprintf(buf + count, max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "%s ", src->name); } -- cgit v1.2.3 From 0764771dab80d7b84b9a271bee7f1b21a04a3f0c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Jun 2009 11:18:36 +0200 Subject: perf_counter: More paranoia settings Rename the perf_counter_priv knob to perf_counter_paranoia (because priv can be read as private, as opposed to privileged) and provide one more level: 0 - permissive 1 - restrict cpu counters to privilidged contexts 2 - restrict kernel-mode code counting and profiling Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 25 +++++++++++++++++++++++-- kernel/sysctl.c | 6 +++--- 2 files changed, 26 insertions(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 8b89b40bd0f0..63f1987c1c1c 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -43,7 +43,23 @@ static atomic_t nr_counters __read_mostly; static atomic_t nr_mmap_counters __read_mostly; static atomic_t nr_comm_counters __read_mostly; -int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */ +/* + * 0 - not paranoid + * 1 - disallow cpu counters to unpriv + * 2 - disallow kernel profiling to unpriv + */ +int sysctl_perf_counter_paranoid __read_mostly; /* do we need to be privileged */ + +static inline bool perf_paranoid_cpu(void) +{ + return sysctl_perf_counter_paranoid > 0; +} + +static inline bool perf_paranoid_kernel(void) +{ + return sysctl_perf_counter_paranoid > 1; +} + int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ int sysctl_perf_counter_limit __read_mostly = 100000; /* max NMIs per second */ @@ -1385,7 +1401,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu) */ if (cpu != -1) { /* Must be root to operate on a CPU counter: */ - if (sysctl_perf_counter_priv && !capable(CAP_SYS_ADMIN)) + if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) return ERR_PTR(-EACCES); if (cpu < 0 || cpu > num_possible_cpus()) @@ -3618,6 +3634,11 @@ SYSCALL_DEFINE5(perf_counter_open, if (copy_from_user(&attr, attr_uptr, sizeof(attr)) != 0) return -EFAULT; + if (!attr.exclude_kernel) { + if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN)) + return -EACCES; + } + /* * Get the target context (task or percpu): */ diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 0c4bf863afa3..344a65981dee 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -916,9 +916,9 @@ static struct ctl_table kern_table[] = { #ifdef CONFIG_PERF_COUNTERS { .ctl_name = CTL_UNNUMBERED, - .procname = "perf_counter_privileged", - .data = &sysctl_perf_counter_priv, - .maxlen = sizeof(sysctl_perf_counter_priv), + .procname = "perf_counter_paranoid", + .data = &sysctl_perf_counter_paranoid, + .maxlen = sizeof(sysctl_perf_counter_paranoid), .mode = 0644, .proc_handler = &proc_dointvec, }, -- cgit v1.2.3 From df58ab24bf26b166874bfb18b3b5a2e0a8e63179 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Jun 2009 11:25:05 +0200 Subject: perf_counter: Rename perf_counter_limit sysctl Rename perf_counter_limit to perf_counter_max_sample_rate and prohibit creation of counters with a known higher sample frequency. Signed-off-by: Peter Zijlstra Cc: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 27 +++++++++++++++++++-------- kernel/sysctl.c | 6 +++--- 2 files changed, 22 insertions(+), 11 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 63f1987c1c1c..3b2829de5590 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -44,11 +44,12 @@ static atomic_t nr_mmap_counters __read_mostly; static atomic_t nr_comm_counters __read_mostly; /* - * 0 - not paranoid - * 1 - disallow cpu counters to unpriv - * 2 - disallow kernel profiling to unpriv + * perf counter paranoia level: + * 0 - not paranoid + * 1 - disallow cpu counters to unpriv + * 2 - disallow kernel profiling to unpriv */ -int sysctl_perf_counter_paranoid __read_mostly; /* do we need to be privileged */ +int sysctl_perf_counter_paranoid __read_mostly; static inline bool perf_paranoid_cpu(void) { @@ -61,7 +62,11 @@ static inline bool perf_paranoid_kernel(void) } int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ -int sysctl_perf_counter_limit __read_mostly = 100000; /* max NMIs per second */ + +/* + * max perf counter sample rate + */ +int sysctl_perf_counter_sample_rate __read_mostly = 100000; static atomic64_t perf_counter_id; @@ -1244,7 +1249,7 @@ static void perf_ctx_adjust_freq(struct perf_counter_context *ctx) if (interrupts == MAX_INTERRUPTS) { perf_log_throttle(counter, 1); counter->pmu->unthrottle(counter); - interrupts = 2*sysctl_perf_counter_limit/HZ; + interrupts = 2*sysctl_perf_counter_sample_rate/HZ; } if (!counter->attr.freq || !counter->attr.sample_freq) @@ -1682,7 +1687,7 @@ static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) spin_lock_irq(&ctx->lock); if (counter->attr.freq) { - if (value > sysctl_perf_counter_limit) { + if (value > sysctl_perf_counter_sample_rate) { ret = -EINVAL; goto unlock; } @@ -2979,7 +2984,8 @@ int perf_counter_overflow(struct perf_counter *counter, int nmi, } else { if (hwc->interrupts != MAX_INTERRUPTS) { hwc->interrupts++; - if (HZ * hwc->interrupts > (u64)sysctl_perf_counter_limit) { + if (HZ * hwc->interrupts > + (u64)sysctl_perf_counter_sample_rate) { hwc->interrupts = MAX_INTERRUPTS; perf_log_throttle(counter, 0); ret = 1; @@ -3639,6 +3645,11 @@ SYSCALL_DEFINE5(perf_counter_open, return -EACCES; } + if (attr.freq) { + if (attr.sample_freq > sysctl_perf_counter_sample_rate) + return -EINVAL; + } + /* * Get the target context (task or percpu): */ diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 344a65981dee..9fd4e436b696 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -932,9 +932,9 @@ static struct ctl_table kern_table[] = { }, { .ctl_name = CTL_UNNUMBERED, - .procname = "perf_counter_int_limit", - .data = &sysctl_perf_counter_limit, - .maxlen = sizeof(sysctl_perf_counter_limit), + .procname = "perf_counter_max_sample_rate", + .data = &sysctl_perf_counter_sample_rate, + .maxlen = sizeof(sysctl_perf_counter_sample_rate), .mode = 0644, .proc_handler = &proc_dointvec, }, -- cgit v1.2.3 From 1c432d899d32d36371ee4ee310fa3609cf0e5742 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Jun 2009 13:19:29 +0200 Subject: perf_counter: Rename enums Rename the perf enums to be in the 'perf_' namespace and strictly enumerate the ABI bits. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 3b2829de5590..c02535bed26f 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3162,7 +3162,7 @@ static int perf_swcounter_is_counting(struct perf_counter *counter) } static int perf_swcounter_match(struct perf_counter *counter, - enum perf_event_types type, + enum perf_type_id type, u32 event, struct pt_regs *regs) { if (!perf_swcounter_is_counting(counter)) @@ -3194,7 +3194,7 @@ static void perf_swcounter_add(struct perf_counter *counter, u64 nr, } static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, - enum perf_event_types type, u32 event, + enum perf_type_id type, u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { @@ -3225,7 +3225,7 @@ static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx) return &cpuctx->recursion[0]; } -static void __perf_swcounter_event(enum perf_event_types type, u32 event, +static void __perf_swcounter_event(enum perf_type_id type, u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr) { -- cgit v1.2.3 From f4dbfa8f3131a84257223393905f7efad0ca5996 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Jun 2009 14:06:28 +0200 Subject: perf_counter: Standardize event names Pure renames only, to PERF_COUNT_HW_* and PERF_COUNT_SW_*. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index c02535bed26f..8859b97390ec 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -1024,7 +1024,7 @@ void perf_counter_task_sched_out(struct task_struct *task, int do_switch = 1; regs = task_pt_regs(task); - perf_swcounter_event(PERF_COUNT_CONTEXT_SWITCHES, 1, 1, regs, 0); + perf_swcounter_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, regs, 0); if (likely(!ctx || !cpuctx->task_ctx)) return; @@ -3411,13 +3411,13 @@ void perf_counter_task_migration(struct task_struct *task, int cpu) struct perf_counter_context *ctx; perf_swcounter_ctx_event(&cpuctx->ctx, PERF_TYPE_SOFTWARE, - PERF_COUNT_CPU_MIGRATIONS, + PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); ctx = perf_pin_task_context(task); if (ctx) { perf_swcounter_ctx_event(ctx, PERF_TYPE_SOFTWARE, - PERF_COUNT_CPU_MIGRATIONS, + PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); perf_unpin_context(ctx); } @@ -3475,11 +3475,11 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) * events. */ switch (counter->attr.config) { - case PERF_COUNT_CPU_CLOCK: + case PERF_COUNT_SW_CPU_CLOCK: pmu = &perf_ops_cpu_clock; break; - case PERF_COUNT_TASK_CLOCK: + case PERF_COUNT_SW_TASK_CLOCK: /* * If the user instantiates this as a per-cpu counter, * use the cpu_clock counter instead. @@ -3490,11 +3490,11 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) pmu = &perf_ops_cpu_clock; break; - case PERF_COUNT_PAGE_FAULTS: - case PERF_COUNT_PAGE_FAULTS_MIN: - case PERF_COUNT_PAGE_FAULTS_MAJ: - case PERF_COUNT_CONTEXT_SWITCHES: - case PERF_COUNT_CPU_MIGRATIONS: + case PERF_COUNT_SW_PAGE_FAULTS: + case PERF_COUNT_SW_PAGE_FAULTS_MIN: + case PERF_COUNT_SW_PAGE_FAULTS_MAJ: + case PERF_COUNT_SW_CONTEXT_SWITCHES: + case PERF_COUNT_SW_CPU_MIGRATIONS: pmu = &perf_ops_generic; break; } -- cgit v1.2.3 From cca3f454a85ff42d426401bce7ac804541b2bd03 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Jun 2009 14:57:55 +0200 Subject: perf_counter: Add counter->id to the throttle event So as to be able to distuinguish between multiple counters. Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 8859b97390ec..ef5d8a5b2453 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -2950,13 +2950,15 @@ static void perf_log_throttle(struct perf_counter *counter, int enable) struct { struct perf_event_header header; u64 time; + u64 id; } throttle_event = { .header = { .type = PERF_EVENT_THROTTLE + 1, .misc = 0, .size = sizeof(throttle_event), }, - .time = sched_clock(), + .time = sched_clock(), + .id = counter->id, }; ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 1, 0); -- cgit v1.2.3 From 4f2294b6dc88d99295230d97fef2c9863cec44c3 Mon Sep 17 00:00:00 2001 From: Catalin Marinas Date: Thu, 11 Jun 2009 13:23:20 +0100 Subject: kmemleak: Add modules support This patch handles the kmemleak operations needed for modules loading so that memory allocations from inside a module are properly tracked. Signed-off-by: Catalin Marinas --- kernel/module.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 56 insertions(+) (limited to 'kernel') diff --git a/kernel/module.c b/kernel/module.c index 2383e60fcf3f..5cd55ab15daf 100644 --- a/kernel/module.c +++ b/kernel/module.c @@ -53,6 +53,7 @@ #include #include #include +#include #if 0 #define DEBUGP printk @@ -430,6 +431,7 @@ static void *percpu_modalloc(unsigned long size, unsigned long align, unsigned long extra; unsigned int i; void *ptr; + int cpu; if (align > PAGE_SIZE) { printk(KERN_WARNING "%s: per-cpu alignment %li > %li\n", @@ -459,6 +461,11 @@ static void *percpu_modalloc(unsigned long size, unsigned long align, if (!split_block(i, size)) return NULL; + /* add the per-cpu scanning areas */ + for_each_possible_cpu(cpu) + kmemleak_alloc(ptr + per_cpu_offset(cpu), size, 0, + GFP_KERNEL); + /* Mark allocated */ pcpu_size[i] = -pcpu_size[i]; return ptr; @@ -473,6 +480,7 @@ static void percpu_modfree(void *freeme) { unsigned int i; void *ptr = __per_cpu_start + block_size(pcpu_size[0]); + int cpu; /* First entry is core kernel percpu data. */ for (i = 1; i < pcpu_num_used; ptr += block_size(pcpu_size[i]), i++) { @@ -484,6 +492,10 @@ static void percpu_modfree(void *freeme) BUG(); free: + /* remove the per-cpu scanning areas */ + for_each_possible_cpu(cpu) + kmemleak_free(freeme + per_cpu_offset(cpu)); + /* Merge with previous? */ if (pcpu_size[i-1] >= 0) { pcpu_size[i-1] += pcpu_size[i]; @@ -1876,6 +1888,36 @@ static void *module_alloc_update_bounds(unsigned long size) return ret; } +#ifdef CONFIG_DEBUG_KMEMLEAK +static void kmemleak_load_module(struct module *mod, Elf_Ehdr *hdr, + Elf_Shdr *sechdrs, char *secstrings) +{ + unsigned int i; + + /* only scan the sections containing data */ + kmemleak_scan_area(mod->module_core, (unsigned long)mod - + (unsigned long)mod->module_core, + sizeof(struct module), GFP_KERNEL); + + for (i = 1; i < hdr->e_shnum; i++) { + if (!(sechdrs[i].sh_flags & SHF_ALLOC)) + continue; + if (strncmp(secstrings + sechdrs[i].sh_name, ".data", 5) != 0 + && strncmp(secstrings + sechdrs[i].sh_name, ".bss", 4) != 0) + continue; + + kmemleak_scan_area(mod->module_core, sechdrs[i].sh_addr - + (unsigned long)mod->module_core, + sechdrs[i].sh_size, GFP_KERNEL); + } +} +#else +static inline void kmemleak_load_module(struct module *mod, Elf_Ehdr *hdr, + Elf_Shdr *sechdrs, char *secstrings) +{ +} +#endif + /* Allocate and load the module: note that size of section 0 is always zero, and we rely on this for optional sections. */ static noinline struct module *load_module(void __user *umod, @@ -2046,6 +2088,12 @@ static noinline struct module *load_module(void __user *umod, /* Do the allocs. */ ptr = module_alloc_update_bounds(mod->core_size); + /* + * The pointer to this block is stored in the module structure + * which is inside the block. Just mark it as not being a + * leak. + */ + kmemleak_not_leak(ptr); if (!ptr) { err = -ENOMEM; goto free_percpu; @@ -2054,6 +2102,13 @@ static noinline struct module *load_module(void __user *umod, mod->module_core = ptr; ptr = module_alloc_update_bounds(mod->init_size); + /* + * The pointer to this block is stored in the module structure + * which is inside the block. This block doesn't need to be + * scanned as it contains data and code that will be freed + * after the module is initialized. + */ + kmemleak_ignore(ptr); if (!ptr && mod->init_size) { err = -ENOMEM; goto free_core; @@ -2084,6 +2139,7 @@ static noinline struct module *load_module(void __user *umod, } /* Module has been moved. */ mod = (void *)sechdrs[modindex].sh_addr; + kmemleak_load_module(mod, hdr, sechdrs, secstrings); #if defined(CONFIG_MODULE_UNLOAD) && defined(CONFIG_SMP) mod->refptr = percpu_modalloc(sizeof(local_t), __alignof__(local_t), -- cgit v1.2.3 From 36b7b6d465489c4754c4fd66fcec6086eba87896 Mon Sep 17 00:00:00 2001 From: Pekka Enberg Date: Wed, 10 Jun 2009 23:42:36 +0300 Subject: sched: use kzalloc() instead of the bootmem allocator Now that kmem_cache_init() happens before sched_init(), we should use kzalloc() and not the bootmem allocator. Signed-off-by: Pekka Enberg --- kernel/sched.c | 20 ++++++++------------ 1 file changed, 8 insertions(+), 12 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 14c447ae5d53..a9ff9533355b 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -68,7 +68,6 @@ #include #include #include -#include #include #include #include @@ -7782,21 +7781,18 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd) static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) { + gfp_t gfp = GFP_KERNEL; + memset(rd, 0, sizeof(*rd)); - if (bootmem) { - alloc_bootmem_cpumask_var(&def_root_domain.span); - alloc_bootmem_cpumask_var(&def_root_domain.online); - alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); - cpupri_init(&rd->cpupri, true); - return 0; - } + if (bootmem) + gfp = GFP_NOWAIT; - if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) + if (!alloc_cpumask_var(&rd->span, gfp)) goto out; - if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) + if (!alloc_cpumask_var(&rd->online, gfp)) goto free_span; - if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) + if (!alloc_cpumask_var(&rd->rto_mask, gfp)) goto free_online; if (cpupri_init(&rd->cpupri, false) != 0) @@ -9123,7 +9119,7 @@ void __init sched_init(void) * we use alloc_bootmem(). */ if (alloc_size) { - ptr = (unsigned long)alloc_bootmem(alloc_size); + ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); #ifdef CONFIG_FAIR_GROUP_SCHED init_task_group.se = (struct sched_entity **)ptr; -- cgit v1.2.3 From 38c7fed2f5ffee17e1fa3e0f78b0e1bf43d52d13 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Mon, 25 May 2009 15:10:58 +0300 Subject: x86: remove some alloc_bootmem_cpumask_var calling Now that we set up the slab allocator earlier, we can get rid of some alloc_bootmem_cpumask_var() calls in boot code. Cc: Ingo Molnar Cc: Johannes Weiner Cc: Linus Torvalds Signed-off-by: Yinghai Lu Signed-off-by: Pekka Enberg --- kernel/cpuset.c | 2 +- kernel/profile.c | 6 ------ 2 files changed, 1 insertion(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/cpuset.c b/kernel/cpuset.c index 026faccca869..d5a7e17474ee 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -1857,7 +1857,7 @@ struct cgroup_subsys cpuset_subsys = { int __init cpuset_init_early(void) { - alloc_bootmem_cpumask_var(&top_cpuset.cpus_allowed); + alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_NOWAIT); top_cpuset.mems_generation = cpuset_mems_generation++; return 0; diff --git a/kernel/profile.c b/kernel/profile.c index 7724e0409bae..28cf26ad2d24 100644 --- a/kernel/profile.c +++ b/kernel/profile.c @@ -111,12 +111,6 @@ int __ref profile_init(void) /* only text is profiled */ prof_len = (_etext - _stext) >> prof_shift; buffer_bytes = prof_len*sizeof(atomic_t); - if (!slab_is_available()) { - prof_buffer = alloc_bootmem(buffer_bytes); - alloc_bootmem_cpumask_var(&prof_cpu_mask); - cpumask_copy(prof_cpu_mask, cpu_possible_mask); - return 0; - } if (!alloc_cpumask_var(&prof_cpu_mask, GFP_KERNEL)) return -ENOMEM; -- cgit v1.2.3 From dad213aeb59718623fc59defeff95fe8c3feb8a0 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Thu, 28 May 2009 18:14:40 -0700 Subject: irq/cpumask: make memoryless node zero happy Don't hardcode to node zero for early boot IRQ setup memory allocations. [ penberg@cs.helsinki.fi: minor cleanups ] Cc: Ingo Molnar Cc: Johannes Weiner Cc: Linus Torvalds Signed-off-by: Yinghai Lu Signed-off-by: Pekka Enberg --- kernel/irq/handle.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index a60018402f42..e161999b6683 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -150,6 +150,7 @@ int __init early_irq_init(void) { struct irq_desc *desc; int legacy_count; + int node; int i; init_irq_default_affinity(); @@ -160,20 +161,20 @@ int __init early_irq_init(void) desc = irq_desc_legacy; legacy_count = ARRAY_SIZE(irq_desc_legacy); + node = first_online_node; /* allocate irq_desc_ptrs array based on nr_irqs */ irq_desc_ptrs = alloc_bootmem(nr_irqs * sizeof(void *)); /* allocate based on nr_cpu_ids */ - /* FIXME: invert kstat_irgs, and it'd be a per_cpu_alloc'd thing */ - kstat_irqs_legacy = alloc_bootmem(NR_IRQS_LEGACY * nr_cpu_ids * - sizeof(int)); + kstat_irqs_legacy = kzalloc_node(NR_IRQS_LEGACY * nr_cpu_ids * + sizeof(int), GFP_NOWAIT, node); for (i = 0; i < legacy_count; i++) { desc[i].irq = i; desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids; lockdep_set_class(&desc[i].lock, &irq_desc_lock_class); - alloc_desc_masks(&desc[i], 0, true); + alloc_desc_masks(&desc[i], node, true); init_desc_masks(&desc[i]); irq_desc_ptrs[i] = desc + i; } -- cgit v1.2.3 From 4bdddf8ff9bbb8aa7b4d7847586202bd25842c90 Mon Sep 17 00:00:00 2001 From: Pekka Enberg Date: Thu, 11 Jun 2009 08:35:27 +0300 Subject: sched: use alloc_cpumask_var() instead of alloc_bootmem_cpumask_var() Slab is initialized when sched_init() runs now so lets use alloc_cpumask_var(). Cc: Ingo Molnar Cc: Linus Torvalds Cc: Yinghai Lu Signed-off-by: Pekka Enberg --- kernel/sched.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index a9ff9533355b..12cc09cf5147 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -9310,13 +9310,13 @@ void __init sched_init(void) current->sched_class = &fair_sched_class; /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ - alloc_bootmem_cpumask_var(&nohz_cpu_mask); + alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); #ifdef CONFIG_SMP #ifdef CONFIG_NO_HZ - alloc_bootmem_cpumask_var(&nohz.cpu_mask); - alloc_bootmem_cpumask_var(&nohz.ilb_grp_nohz_mask); + alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); + alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); #endif - alloc_bootmem_cpumask_var(&cpu_isolated_map); + alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); #endif /* SMP */ scheduler_running = 1; -- cgit v1.2.3 From 0fb530291621c8b8a1b16abeeab05d9262489f71 Mon Sep 17 00:00:00 2001 From: Pekka Enberg Date: Thu, 11 Jun 2009 08:41:22 +0300 Subject: sched: use slab in cpupri_init() Lets not use the bootmem allocator in cpupri_init() as slab is already up when it is run. Cc: Ingo Molnar Cc: Linus Torvalds Cc: Yinghai Lu Signed-off-by: Pekka Enberg --- kernel/sched.c | 2 +- kernel/sched_cpupri.c | 8 +++++--- 2 files changed, 6 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index 12cc09cf5147..dcf2dc28931a 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -7795,7 +7795,7 @@ static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) if (!alloc_cpumask_var(&rd->rto_mask, gfp)) goto free_online; - if (cpupri_init(&rd->cpupri, false) != 0) + if (cpupri_init(&rd->cpupri, bootmem) != 0) goto free_rto_mask; return 0; diff --git a/kernel/sched_cpupri.c b/kernel/sched_cpupri.c index 344712a5e3ed..7deffc9f0e5f 100644 --- a/kernel/sched_cpupri.c +++ b/kernel/sched_cpupri.c @@ -154,8 +154,12 @@ void cpupri_set(struct cpupri *cp, int cpu, int newpri) */ int __init_refok cpupri_init(struct cpupri *cp, bool bootmem) { + gfp_t gfp = GFP_KERNEL; int i; + if (bootmem) + gfp = GFP_NOWAIT; + memset(cp, 0, sizeof(*cp)); for (i = 0; i < CPUPRI_NR_PRIORITIES; i++) { @@ -163,9 +167,7 @@ int __init_refok cpupri_init(struct cpupri *cp, bool bootmem) spin_lock_init(&vec->lock); vec->count = 0; - if (bootmem) - alloc_bootmem_cpumask_var(&vec->mask); - else if (!zalloc_cpumask_var(&vec->mask, GFP_KERNEL)) + if (!zalloc_cpumask_var(&vec->mask, gfp)) goto cleanup; } -- cgit v1.2.3 From 22fb4e71e646695c7e0f379ada66b372c2d1aa1a Mon Sep 17 00:00:00 2001 From: Pekka Enberg Date: Thu, 11 Jun 2009 14:46:49 +0300 Subject: irq: use kcalloc() instead of the bootmem allocator Fixes the following problem: [ 0.000000] Experimental hierarchical RCU init done. [ 0.000000] NR_IRQS:4352 nr_irqs:256 [ 0.000000] ------------[ cut here ]------------ [ 0.000000] WARNING: at mm/bootmem.c:537 alloc_arch_preferred_bootmem+0x40/0x7e() [ 0.000000] Hardware name: To Be Filled By O.E.M. [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #59709 [ 0.000000] Call Trace: [ 0.000000] [] ? alloc_arch_preferred_bootmem+0x40/0x7e [ 0.000000] [] warn_slowpath_common+0x88/0xcb [ 0.000000] [] warn_slowpath_null+0x27/0x3d [ 0.000000] [] alloc_arch_preferred_bootmem+0x40/0x7e [ 0.000000] [] ___alloc_bootmem_nopanic+0x4e/0xec [ 0.000000] [] ___alloc_bootmem+0x20/0x61 [ 0.000000] [] __alloc_bootmem+0x1e/0x34 [ 0.000000] [] early_irq_init+0x6d/0x118 [ 0.000000] [] ? early_idt_handler+0x0/0x71 [ 0.000000] [] start_kernel+0x192/0x394 [ 0.000000] [] ? early_idt_handler+0x0/0x71 [ 0.000000] [] x86_64_start_reservations+0xb4/0xcf [ 0.000000] [] ? __init_begin+0x0/0x140 [ 0.000000] [] x86_64_start_kernel+0x158/0x17b [ 0.000000] ---[ end trace a7919e7f17c0a725 ]--- [ 0.000000] Fast TSC calibration using PIT [ 0.000000] Detected 2002.510 MHz processor. [ 0.004000] Console: colour VGA+ 80x25 Reported-by: Ingo Molnar Signed-off-by: Pekka Enberg --- kernel/irq/handle.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index e161999b6683..104578541230 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -164,7 +164,7 @@ int __init early_irq_init(void) node = first_online_node; /* allocate irq_desc_ptrs array based on nr_irqs */ - irq_desc_ptrs = alloc_bootmem(nr_irqs * sizeof(void *)); + irq_desc_ptrs = kcalloc(nr_irqs, sizeof(void *), GFP_NOWAIT); /* allocate based on nr_cpu_ids */ kstat_irqs_legacy = kzalloc_node(NR_IRQS_LEGACY * nr_cpu_ids * -- cgit v1.2.3 From b415c49a864dab8ee90713833d642dd461eccae9 Mon Sep 17 00:00:00 2001 From: Oleg Nesterov Date: Thu, 11 Jun 2009 13:12:55 +0100 Subject: slow_work_thread() should do the exclusive wait slow_work_thread() sleeps on slow_work_thread_wq without WQ_FLAG_EXCLUSIVE, this means that slow_work_enqueue()->__wake_up(nr_exclusive => 1) wakes up all kslowd threads. This is not what we want, so we change slow_work_thread() to use prepare_to_wait_exclusive() instead. Signed-off-by: Oleg Nesterov Signed-off-by: David Howells Signed-off-by: Linus Torvalds --- kernel/slow-work.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/slow-work.c b/kernel/slow-work.c index b28d19135f43..521ed2004d63 100644 --- a/kernel/slow-work.c +++ b/kernel/slow-work.c @@ -372,8 +372,8 @@ static int slow_work_thread(void *_data) vsmax *= atomic_read(&slow_work_thread_count); vsmax /= 100; - prepare_to_wait(&slow_work_thread_wq, &wait, - TASK_INTERRUPTIBLE); + prepare_to_wait_exclusive(&slow_work_thread_wq, &wait, + TASK_INTERRUPTIBLE); if (!freezing(current) && !slow_work_threads_should_exit && !slow_work_available(vsmax) && -- cgit v1.2.3 From 589ff870ed60a9ebdd5ec99ec3f5afe1282fe151 Mon Sep 17 00:00:00 2001 From: Al Viro Date: Sat, 18 Apr 2009 03:28:19 -0400 Subject: Switch collect_mounts() to struct path Signed-off-by: Al Viro --- kernel/audit_tree.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c index 6e7351739a82..1f6396d76687 100644 --- a/kernel/audit_tree.c +++ b/kernel/audit_tree.c @@ -568,7 +568,7 @@ void audit_trim_trees(void) if (err) goto skip_it; - root_mnt = collect_mounts(path.mnt, path.dentry); + root_mnt = collect_mounts(&path); path_put(&path); if (!root_mnt) goto skip_it; @@ -660,7 +660,7 @@ int audit_add_tree_rule(struct audit_krule *rule) err = kern_path(tree->pathname, 0, &path); if (err) goto Err; - mnt = collect_mounts(path.mnt, path.dentry); + mnt = collect_mounts(&path); path_put(&path); if (!mnt) { err = -ENOMEM; @@ -720,7 +720,7 @@ int audit_tag_tree(char *old, char *new) err = kern_path(new, 0, &path); if (err) return err; - tagged = collect_mounts(path.mnt, path.dentry); + tagged = collect_mounts(&path); path_put(&path); if (!tagged) return -ENOMEM; -- cgit v1.2.3 From 337eb00a2c3a421999c39c94ce7e33545ee8baa7 Mon Sep 17 00:00:00 2001 From: Alessio Igor Bogani Date: Tue, 12 May 2009 15:10:54 +0200 Subject: Push BKL down into ->remount_fs() [xfs, btrfs, capifs, shmem don't need BKL, exempt] Signed-off-by: Alessio Igor Bogani Signed-off-by: Al Viro --- kernel/cgroup.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/cgroup.c b/kernel/cgroup.c index a7267bfd3765..3fb789f6df94 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c @@ -46,6 +46,7 @@ #include #include #include +#include #include @@ -900,6 +901,7 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data) struct cgroup *cgrp = &root->top_cgroup; struct cgroup_sb_opts opts; + lock_kernel(); mutex_lock(&cgrp->dentry->d_inode->i_mutex); mutex_lock(&cgroup_mutex); @@ -927,6 +929,7 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data) kfree(opts.release_agent); mutex_unlock(&cgroup_mutex); mutex_unlock(&cgrp->dentry->d_inode->i_mutex); + unlock_kernel(); return ret; } -- cgit v1.2.3 From 28be225b23b115573e0ecc8ef9996f42a1652f74 Mon Sep 17 00:00:00 2001 From: Yinghai Lu Date: Fri, 12 Jun 2009 11:33:02 +0300 Subject: irq: slab alloc for default irq_affinity Ingo had [ 0.000000] ------------[ cut here ]------------ [ 0.000000] WARNING: at mm/bootmem.c:537 alloc_arch_preferred_bootmem+0x2b/0x71() [ 0.000000] Hardware name: System Product Name [ 0.000000] Modules linked in: [ 0.000000] Pid: 0, comm: swapper Tainted: G W 2.6.30-tip-03087-g0bb2618-dirty #52506 [ 0.000000] Call Trace: [ 0.000000] [<81032588>] warn_slowpath_common+0x60/0x90 [ 0.000000] [<810325c5>] warn_slowpath_null+0xd/0x10 [ 0.000000] [<819d1bc0>] alloc_arch_preferred_bootmem+0x2b/0x71 [ 0.000000] [<819d1c31>] ___alloc_bootmem_nopanic+0x2b/0x9a [ 0.000000] [<81050a0a>] ? lock_release+0xac/0xb2 [ 0.000000] [<819d1d4c>] ___alloc_bootmem+0xe/0x2d [ 0.000000] [<819d1e9f>] __alloc_bootmem+0xa/0xc [ 0.000000] [<819d7c63>] alloc_bootmem_cpumask_var+0x21/0x26 [ 0.000000] [<819d0cc8>] early_irq_init+0x15/0x10d [ 0.000000] [<819bb75a>] start_kernel+0x167/0x326 [ 0.000000] [<819bb06b>] __init_begin+0x6b/0x70 [ 0.000000] ---[ end trace 4eaa2a86a8e2da23 ]--- [ 0.000000] NR_IRQS:2304 nr_irqs:424 [ 0.000000] CPU 0 irqstacks, hard=821e6000 soft=821e7000 we need to update init_irq_default_affinity Signed-off-by: Yinghai Lu Signed-off-by: Pekka Enberg --- kernel/irq/handle.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c index 104578541230..065205bdd920 100644 --- a/kernel/irq/handle.c +++ b/kernel/irq/handle.c @@ -45,7 +45,7 @@ void handle_bad_irq(unsigned int irq, struct irq_desc *desc) #if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_HARDIRQS) static void __init init_irq_default_affinity(void) { - alloc_bootmem_cpumask_var(&irq_default_affinity); + alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT); cpumask_setall(irq_default_affinity); } #else -- cgit v1.2.3 From 9a71af2c3627b379b7c31917a7f6ee0d29bc559b Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 21:46:53 -0600 Subject: module_param: invbool should take a 'bool', not an 'int' It takes an 'int' for historical reasons, and there are only two users: simply switch it over to bool. The other user (uvesafb.c) will get a (harmless-on-x86) warning until the next patch is applied. Cc: Brad Douglas Cc: Michal Januszewski Signed-off-by: Rusty Russell --- kernel/params.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/params.c b/kernel/params.c index de273ec85bd2..023abbf5f89f 100644 --- a/kernel/params.c +++ b/kernel/params.c @@ -272,13 +272,13 @@ int param_set_invbool(const char *val, struct kernel_param *kp) dummy.arg = &boolval; ret = param_set_bool(val, &dummy); if (ret == 0) - *(int *)kp->arg = !boolval; + *(bool *)kp->arg = !boolval; return ret; } int param_get_invbool(char *buffer, struct kernel_param *kp) { - return sprintf(buffer, "%c", (*(int *)kp->arg) ? 'N' : 'Y'); + return sprintf(buffer, "%c", (*(bool *)kp->arg) ? 'N' : 'Y'); } /* We break the rule and mangle the string. */ -- cgit v1.2.3 From 45fcc70c0b6ee0c508e1fdb5fef735c3546803f4 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 21:46:56 -0600 Subject: module_param: split perm field into flags and perm Impact: cleanup Rather than hack KPARAM_KMALLOCED into the perm field, separate it out. Since the perm field was 32 bits and only needs 16, we don't add bloat. Signed-off-by: Rusty Russell --- kernel/params.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) (limited to 'kernel') diff --git a/kernel/params.c b/kernel/params.c index 023abbf5f89f..b4660dc13dbc 100644 --- a/kernel/params.c +++ b/kernel/params.c @@ -24,9 +24,6 @@ #include #include -/* We abuse the high bits of "perm" to record whether we kmalloc'ed. */ -#define KPARAM_KMALLOCED 0x80000000 - #if 0 #define DEBUGP printk #else @@ -220,13 +217,13 @@ int param_set_charp(const char *val, struct kernel_param *kp) return -ENOSPC; } - if (kp->perm & KPARAM_KMALLOCED) + if (kp->flags & KPARAM_KMALLOCED) kfree(*(char **)kp->arg); /* This is a hack. We can't need to strdup in early boot, and we * don't need to; this mangled commandline is preserved. */ if (slab_is_available()) { - kp->perm |= KPARAM_KMALLOCED; + kp->flags |= KPARAM_KMALLOCED; *(char **)kp->arg = kstrdup(val, GFP_KERNEL); if (!kp->arg) return -ENOMEM; @@ -591,7 +588,7 @@ void destroy_params(const struct kernel_param *params, unsigned num) unsigned int i; for (i = 0; i < num; i++) - if (params[i].perm & KPARAM_KMALLOCED) + if (params[i].flags & KPARAM_KMALLOCED) kfree(*(char **)params[i].arg); } -- cgit v1.2.3 From fddd520122953550ec2c8b60e7ca0d0f0d115d97 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 21:46:57 -0600 Subject: module_param: allow 'bool' module_params to be bool, not just int. Impact: API cleanup For historical reasons, 'bool' parameters must be an int, not a bool. But there are around 600 users, so a conversion seems like useless churn. So we use __same_type() to distinguish, and handle both cases. Signed-off-by: Rusty Russell --- kernel/params.c | 33 ++++++++++++++++++++++++++------- 1 file changed, 26 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/params.c b/kernel/params.c index b4660dc13dbc..7f6912ced2ba 100644 --- a/kernel/params.c +++ b/kernel/params.c @@ -238,35 +238,54 @@ int param_get_charp(char *buffer, struct kernel_param *kp) return sprintf(buffer, "%s", *((char **)kp->arg)); } +/* Actually could be a bool or an int, for historical reasons. */ int param_set_bool(const char *val, struct kernel_param *kp) { + bool v; + /* No equals means "set"... */ if (!val) val = "1"; /* One of =[yYnN01] */ switch (val[0]) { case 'y': case 'Y': case '1': - *(int *)kp->arg = 1; - return 0; + v = true; + break; case 'n': case 'N': case '0': - *(int *)kp->arg = 0; - return 0; + v = false; + break; + default: + return -EINVAL; } - return -EINVAL; + + if (kp->flags & KPARAM_ISBOOL) + *(bool *)kp->arg = v; + else + *(int *)kp->arg = v; + return 0; } int param_get_bool(char *buffer, struct kernel_param *kp) { + bool val; + if (kp->flags & KPARAM_ISBOOL) + val = *(bool *)kp->arg; + else + val = *(int *)kp->arg; + /* Y and N chosen as being relatively non-coder friendly */ - return sprintf(buffer, "%c", (*(int *)kp->arg) ? 'Y' : 'N'); + return sprintf(buffer, "%c", val ? 'Y' : 'N'); } +/* This one must be bool. */ int param_set_invbool(const char *val, struct kernel_param *kp) { - int boolval, ret; + int ret; + bool boolval; struct kernel_param dummy; dummy.arg = &boolval; + dummy.flags = KPARAM_ISBOOL; ret = param_set_bool(val, &dummy); if (ret == 0) *(bool *)kp->arg = !boolval; -- cgit v1.2.3 From ad6561dffa17f17bb68d7207d422c26c381c4313 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 21:47:03 -0600 Subject: module: trim exception table on init free. It's theoretically possible that there are exception table entries which point into the (freed) init text of modules. These could cause future problems if other modules get loaded into that memory and cause an exception as we'd see the wrong fixup. The only case I know of is kvm-intel.ko (when CONFIG_CC_OPTIMIZE_FOR_SIZE=n). Amerigo fixed this long-standing FIXME in the x86 version, but this patch is more general. This implements trim_init_extable(); most archs are simple since they use the standard lib/extable.c sort code. Alpha and IA64 use relative addresses in their fixups, so thier trimming is a slight variation. Sparc32 is unique; it doesn't seem to define ARCH_HAS_SORT_EXTABLE, yet it defines its own sort_extable() which overrides the one in lib. It doesn't sort, so we have to mark deleted entries instead of actually trimming them. Inspired-by: Amerigo Wang Signed-off-by: Rusty Russell Cc: linux-alpha@vger.kernel.org Cc: sparclinux@vger.kernel.org Cc: linux-ia64@vger.kernel.org --- kernel/module.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/module.c b/kernel/module.c index 35f7de00bf0d..e4ab36ce7672 100644 --- a/kernel/module.c +++ b/kernel/module.c @@ -2455,6 +2455,7 @@ SYSCALL_DEFINE3(init_module, void __user *, umod, mutex_lock(&module_mutex); /* Drop initial reference. */ module_put(mod); + trim_init_extable(mod); module_free(mod, mod->module_init); mod->module_init = NULL; mod->init_size = 0; -- cgit v1.2.3 From 081fad86178ec0f64f32f1bd04cf4aad22714fb9 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Jun 2009 17:57:21 +0200 Subject: perf_counter: Remove PERF_TYPE_RAW special casing The PERF_TYPE_RAW special case seems superfluous these days. Remove it and add it to the switch() stmt like the others. [ Impact: cleanup ] Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index ef5d8a5b2453..663bbe015057 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3570,12 +3570,8 @@ perf_counter_alloc(struct perf_counter_attr *attr, if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP)) goto done; - if (attr->type == PERF_TYPE_RAW) { - pmu = hw_perf_counter_init(counter); - goto done; - } - switch (attr->type) { + case PERF_TYPE_RAW: case PERF_TYPE_HARDWARE: case PERF_TYPE_HW_CACHE: pmu = hw_perf_counter_init(counter); -- cgit v1.2.3 From 974802eaa1afdc87e00821df7020a2b3c6fee623 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 12 Jun 2009 12:46:55 +0200 Subject: perf_counter: Add forward/backward attribute ABI compatibility Provide for means of extending the perf_counter_attr in a 'natural' way. We allow growing the structure by appending fields at the end by specifying the full structure size inside it. When a new kernel sees a smaller (old) structure, it will 0 pad the tail. When an old kernel sees a larger (new) structure, it will verify the tail consists of 0s, otherwise fail. If we fail due to a size-mismatch, we return -E2BIG and write the kernel's native attribe size back into the provided structure. Furthermore, add some attribute verification, so that we'll fail counter creation when unknown bits are present (PERF_SAMPLE, PERF_FORMAT, or in the __reserved fields). (This ABI detail is introduced while keeping the existing syscall ABI.) Signed-off-by: Peter Zijlstra Cc: Mike Galbraith Cc: Paul Mackerras Cc: Arnaldo Carvalho de Melo LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_counter.c | 89 +++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 86 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c index 663bbe015057..29b685f551aa 100644 --- a/kernel/perf_counter.c +++ b/kernel/perf_counter.c @@ -3584,6 +3584,9 @@ perf_counter_alloc(struct perf_counter_attr *attr, case PERF_TYPE_TRACEPOINT: pmu = tp_perf_counter_init(counter); break; + + default: + break; } done: err = 0; @@ -3610,6 +3613,85 @@ done: return counter; } +static int perf_copy_attr(struct perf_counter_attr __user *uattr, + struct perf_counter_attr *attr) +{ + int ret; + u32 size; + + if (!access_ok(VERIFY_WRITE, uattr, PERF_ATTR_SIZE_VER0)) + return -EFAULT; + + /* + * zero the full structure, so that a short copy will be nice. + */ + memset(attr, 0, sizeof(*attr)); + + ret = get_user(size, &uattr->size); + if (ret) + return ret; + + if (size > PAGE_SIZE) /* silly large */ + goto err_size; + + if (!size) /* abi compat */ + size = PERF_ATTR_SIZE_VER0; + + if (size < PERF_ATTR_SIZE_VER0) + goto err_size; + + /* + * If we're handed a bigger struct than we know of, + * ensure all the unknown bits are 0. + */ + if (size > sizeof(*attr)) { + unsigned long val; + unsigned long __user *addr; + unsigned long __user *end; + + addr = PTR_ALIGN((void __user *)uattr + sizeof(*attr), + sizeof(unsigned long)); + end = PTR_ALIGN((void __user *)uattr + size, + sizeof(unsigned long)); + + for (; addr < end; addr += sizeof(unsigned long)) { + ret = get_user(val, addr); + if (ret) + return ret; + if (val) + goto err_size; + } + } + + ret = copy_from_user(attr, uattr, size); + if (ret) + return -EFAULT; + + /* + * If the type exists, the corresponding creation will verify + * the attr->config. + */ + if (attr->type >= PERF_TYPE_MAX) + return -EINVAL; + + if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) + return -EINVAL; + + if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) + return -EINVAL; + + if (attr->read_format & ~(PERF_FORMAT_MAX-1)) + return -EINVAL; + +out: + return ret; + +err_size: + put_user(sizeof(*attr), &uattr->size); + ret = -E2BIG; + goto out; +} + /** * sys_perf_counter_open - open a performance counter, associate it to a task/cpu * @@ -3619,7 +3701,7 @@ done: * @group_fd: group leader counter fd */ SYSCALL_DEFINE5(perf_counter_open, - const struct perf_counter_attr __user *, attr_uptr, + struct perf_counter_attr __user *, attr_uptr, pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) { struct perf_counter *counter, *group_leader; @@ -3635,8 +3717,9 @@ SYSCALL_DEFINE5(perf_counter_open, if (flags) return -EINVAL; - if (copy_from_user(&attr, attr_uptr, sizeof(attr)) != 0) - return -EFAULT; + ret = perf_copy_attr(attr_uptr, &attr); + if (ret) + return ret; if (!attr.exclude_kernel) { if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN)) -- cgit v1.2.3 From b43e352139f51216a8c56b0bd5fc3d4e05c65619 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 22:27:00 -0600 Subject: sched: export kick_process lguest needs kick_process: wake_up_process() does nothing if a process is running, which isn't sufficient (we need it in the kernel). And lguest support is usually modular. Signed-off-by: Rusty Russell Cc: Ingo Molnar --- kernel/sched.c | 1 + 1 file changed, 1 insertion(+) (limited to 'kernel') diff --git a/kernel/sched.c b/kernel/sched.c index f04aa9664504..8ec9d13140be 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -2192,6 +2192,7 @@ void kick_process(struct task_struct *p) smp_send_reschedule(cpu); preempt_enable(); } +EXPORT_SYMBOL_GPL(kick_process); /* * Return a low guess at the load of a migration-source cpu weighted -- cgit v1.2.3 From 1dc492a0a4470852cb451db1e00d580ce9fd7a28 Mon Sep 17 00:00:00 2001 From: Manish Katiyar Date: Sun, 22 Feb 2009 10:24:27 +0530 Subject: trivial: kernel/power/poweroff.c: whitespace fix Fix coding style whitespace fixes. Patch compile tested Before :- total: 1 errors, 0 warnings, 46 lines checked After total: 0 errors, 0 warnings, 46 lines checked Before :- text data bss dec hex filename 107 48 0 155 9b kernel/power/poweroff.o After text data bss dec hex filename 107 48 0 155 9b kernel/power/poweroff.o Signed-off-by: Manish Katiyar Signed-off-by: Jiri Kosina --- kernel/power/poweroff.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/power/poweroff.c b/kernel/power/poweroff.c index 97890831e1b5..e8b337006276 100644 --- a/kernel/power/poweroff.c +++ b/kernel/power/poweroff.c @@ -34,7 +34,7 @@ static struct sysrq_key_op sysrq_poweroff_op = { .handler = handle_poweroff, .help_msg = "powerOff", .action_msg = "Power Off", - .enable_mask = SYSRQ_ENABLE_BOOT, + .enable_mask = SYSRQ_ENABLE_BOOT, }; static int pm_sysrq_init(void) -- cgit v1.2.3 From 3ac49a1c9928b4a242b3cb1d83bc1d5c9b8fcb50 Mon Sep 17 00:00:00 2001 From: Jean Delvare Date: Thu, 4 Jun 2009 16:20:28 +0200 Subject: trivial: fix ETIMEOUT -> ETIMEDOUT typos fix ETIMEOUT -> ETIMEDOUT typos Signed-off-by: Jean Delvare Signed-off-by: Jiri Kosina --- kernel/rtmutex.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/rtmutex.c b/kernel/rtmutex.c index 820c5af44f3e..fcd107a78c5a 100644 --- a/kernel/rtmutex.c +++ b/kernel/rtmutex.c @@ -902,7 +902,7 @@ EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible); * Returns: * 0 on success * -EINTR when interrupted by a signal - * -ETIMEOUT when the timeout expired + * -ETIMEDOUT when the timeout expired * -EDEADLK when the lock would deadlock (when deadlock detection is on) */ int -- cgit v1.2.3 From e39a71ef80877f4e30d808af9acceec80f4d2f7c Mon Sep 17 00:00:00 2001 From: Magnus Damm Date: Fri, 15 May 2009 00:53:26 +0200 Subject: PM: Rename device_power_down/up() Rename the functions performing "_noirq" dev_pm_ops operations from device_power_down() and device_power_up() to device_suspend_noirq() and device_resume_noirq(). The new function names are chosen to show that the functions are responsible for calling the _noirq() versions to finalize the suspend/resume operation. The current function names do not perform power down/up anymore so the names may be misleading. Global function renames: - device_power_down() -> device_suspend_noirq() - device_power_up() -> device_resume_noirq() Static function renames: - suspend_device_noirq() -> __device_suspend_noirq() - resume_device_noirq() -> __device_resume_noirq() Signed-off-by: Magnus Damm Acked-by: Greg Kroah-Hartman Acked-by: Len Brown Signed-off-by: Rafael J. Wysocki --- kernel/kexec.c | 8 ++++---- kernel/power/disk.c | 16 ++++++++-------- kernel/power/main.c | 4 ++-- 3 files changed, 14 insertions(+), 14 deletions(-) (limited to 'kernel') diff --git a/kernel/kexec.c b/kernel/kexec.c index e4983770913b..5a3da87adae0 100644 --- a/kernel/kexec.c +++ b/kernel/kexec.c @@ -1452,13 +1452,13 @@ int kernel_kexec(void) if (error) goto Resume_console; /* At this point, device_suspend() has been called, - * but *not* device_power_down(). We *must* - * device_power_down() now. Otherwise, drivers for + * but *not* device_suspend_noirq(). We *must* call + * device_suspend_noirq() now. Otherwise, drivers for * some devices (e.g. interrupt controllers) become * desynchronized with the actual state of the * hardware at resume time, and evil weirdness ensues. */ - error = device_power_down(PMSG_FREEZE); + error = device_suspend_noirq(PMSG_FREEZE); if (error) goto Resume_devices; error = disable_nonboot_cpus(); @@ -1486,7 +1486,7 @@ int kernel_kexec(void) local_irq_enable(); Enable_cpus: enable_nonboot_cpus(); - device_power_up(PMSG_RESTORE); + device_resume_noirq(PMSG_RESTORE); Resume_devices: device_resume(PMSG_RESTORE); Resume_console: diff --git a/kernel/power/disk.c b/kernel/power/disk.c index 5cb080e7eebd..1c18bc894a2d 100644 --- a/kernel/power/disk.c +++ b/kernel/power/disk.c @@ -216,12 +216,12 @@ static int create_image(int platform_mode) return error; /* At this point, device_suspend() has been called, but *not* - * device_power_down(). We *must* call device_power_down() now. + * device_suspend_noirq(). We *must* call device_suspend_noirq() now. * Otherwise, drivers for some devices (e.g. interrupt controllers) * become desynchronized with the actual state of the hardware * at resume time, and evil weirdness ensues. */ - error = device_power_down(PMSG_FREEZE); + error = device_suspend_noirq(PMSG_FREEZE); if (error) { printk(KERN_ERR "PM: Some devices failed to power down, " "aborting hibernation\n"); @@ -262,7 +262,7 @@ static int create_image(int platform_mode) Power_up: sysdev_resume(); - /* NOTE: device_power_up() is just a resume() for devices + /* NOTE: device_resume_noirq() is just a resume() for devices * that suspended with irqs off ... no overall powerup. */ @@ -275,7 +275,7 @@ static int create_image(int platform_mode) Platform_finish: platform_finish(platform_mode); - device_power_up(in_suspend ? + device_resume_noirq(in_suspend ? (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); return error; @@ -339,7 +339,7 @@ static int resume_target_kernel(bool platform_mode) { int error; - error = device_power_down(PMSG_QUIESCE); + error = device_suspend_noirq(PMSG_QUIESCE); if (error) { printk(KERN_ERR "PM: Some devices failed to power down, " "aborting resume\n"); @@ -394,7 +394,7 @@ static int resume_target_kernel(bool platform_mode) Cleanup: platform_restore_cleanup(platform_mode); - device_power_up(PMSG_RECOVER); + device_resume_noirq(PMSG_RECOVER); return error; } @@ -454,7 +454,7 @@ int hibernation_platform_enter(void) goto Resume_devices; } - error = device_power_down(PMSG_HIBERNATE); + error = device_suspend_noirq(PMSG_HIBERNATE); if (error) goto Resume_devices; @@ -479,7 +479,7 @@ int hibernation_platform_enter(void) Platofrm_finish: hibernation_ops->finish(); - device_power_up(PMSG_RESTORE); + device_suspend_noirq(PMSG_RESTORE); Resume_devices: entering_platform_hibernation = false; diff --git a/kernel/power/main.c b/kernel/power/main.c index 868028280d13..2f6638ee03c0 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -295,7 +295,7 @@ static int suspend_enter(suspend_state_t state) return error; } - error = device_power_down(PMSG_SUSPEND); + error = device_suspend_noirq(PMSG_SUSPEND); if (error) { printk(KERN_ERR "PM: Some devices failed to power down\n"); goto Platfrom_finish; @@ -335,7 +335,7 @@ static int suspend_enter(suspend_state_t state) suspend_ops->wake(); Power_up_devices: - device_power_up(PMSG_RESUME); + device_resume_noirq(PMSG_RESUME); Platfrom_finish: if (suspend_ops->finish) -- cgit v1.2.3 From d161630297a20802d01c55847bfcba85d2118a9f Mon Sep 17 00:00:00 2001 From: Alan Stern Date: Sun, 24 May 2009 22:05:42 +0200 Subject: PM core: rename suspend and resume functions This patch (as1241) renames a bunch of functions in the PM core. Rather than go through a boring list of name changes, suffice it to say that in the end we have a bunch of pairs of functions: device_resume_noirq dpm_resume_noirq device_resume dpm_resume device_complete dpm_complete device_suspend_noirq dpm_suspend_noirq device_suspend dpm_suspend device_prepare dpm_prepare in which device_X does the X operation on a single device and dpm_X invokes device_X for all devices in the dpm_list. In addition, the old dpm_power_up and device_resume_noirq have been combined into a single function (dpm_resume_noirq). Lastly, dpm_suspend_start and dpm_resume_end are the renamed versions of the former top-level device_suspend and device_resume routines. Signed-off-by: Alan Stern Acked-by: Magnus Damm Signed-off-by: Rafael J. Wysocki --- kernel/kexec.c | 14 +++++++------- kernel/power/disk.c | 30 +++++++++++++++--------------- kernel/power/main.c | 8 ++++---- 3 files changed, 26 insertions(+), 26 deletions(-) (limited to 'kernel') diff --git a/kernel/kexec.c b/kernel/kexec.c index 5a3da87adae0..ae1c35201cc8 100644 --- a/kernel/kexec.c +++ b/kernel/kexec.c @@ -1448,17 +1448,17 @@ int kernel_kexec(void) goto Restore_console; } suspend_console(); - error = device_suspend(PMSG_FREEZE); + error = dpm_suspend_start(PMSG_FREEZE); if (error) goto Resume_console; - /* At this point, device_suspend() has been called, - * but *not* device_suspend_noirq(). We *must* call - * device_suspend_noirq() now. Otherwise, drivers for + /* At this point, dpm_suspend_start() has been called, + * but *not* dpm_suspend_noirq(). We *must* call + * dpm_suspend_noirq() now. Otherwise, drivers for * some devices (e.g. interrupt controllers) become * desynchronized with the actual state of the * hardware at resume time, and evil weirdness ensues. */ - error = device_suspend_noirq(PMSG_FREEZE); + error = dpm_suspend_noirq(PMSG_FREEZE); if (error) goto Resume_devices; error = disable_nonboot_cpus(); @@ -1486,9 +1486,9 @@ int kernel_kexec(void) local_irq_enable(); Enable_cpus: enable_nonboot_cpus(); - device_resume_noirq(PMSG_RESTORE); + dpm_resume_noirq(PMSG_RESTORE); Resume_devices: - device_resume(PMSG_RESTORE); + dpm_resume_end(PMSG_RESTORE); Resume_console: resume_console(); thaw_processes(); diff --git a/kernel/power/disk.c b/kernel/power/disk.c index 1c18bc894a2d..a9beba68b6c7 100644 --- a/kernel/power/disk.c +++ b/kernel/power/disk.c @@ -215,13 +215,13 @@ static int create_image(int platform_mode) if (error) return error; - /* At this point, device_suspend() has been called, but *not* - * device_suspend_noirq(). We *must* call device_suspend_noirq() now. + /* At this point, dpm_suspend_start() has been called, but *not* + * dpm_suspend_noirq(). We *must* call dpm_suspend_noirq() now. * Otherwise, drivers for some devices (e.g. interrupt controllers) * become desynchronized with the actual state of the hardware * at resume time, and evil weirdness ensues. */ - error = device_suspend_noirq(PMSG_FREEZE); + error = dpm_suspend_noirq(PMSG_FREEZE); if (error) { printk(KERN_ERR "PM: Some devices failed to power down, " "aborting hibernation\n"); @@ -262,7 +262,7 @@ static int create_image(int platform_mode) Power_up: sysdev_resume(); - /* NOTE: device_resume_noirq() is just a resume() for devices + /* NOTE: dpm_resume_noirq() is just a resume() for devices * that suspended with irqs off ... no overall powerup. */ @@ -275,7 +275,7 @@ static int create_image(int platform_mode) Platform_finish: platform_finish(platform_mode); - device_resume_noirq(in_suspend ? + dpm_resume_noirq(in_suspend ? (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); return error; @@ -304,7 +304,7 @@ int hibernation_snapshot(int platform_mode) goto Close; suspend_console(); - error = device_suspend(PMSG_FREEZE); + error = dpm_suspend_start(PMSG_FREEZE); if (error) goto Recover_platform; @@ -315,7 +315,7 @@ int hibernation_snapshot(int platform_mode) /* Control returns here after successful restore */ Resume_devices: - device_resume(in_suspend ? + dpm_resume_end(in_suspend ? (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); resume_console(); Close: @@ -339,7 +339,7 @@ static int resume_target_kernel(bool platform_mode) { int error; - error = device_suspend_noirq(PMSG_QUIESCE); + error = dpm_suspend_noirq(PMSG_QUIESCE); if (error) { printk(KERN_ERR "PM: Some devices failed to power down, " "aborting resume\n"); @@ -394,7 +394,7 @@ static int resume_target_kernel(bool platform_mode) Cleanup: platform_restore_cleanup(platform_mode); - device_resume_noirq(PMSG_RECOVER); + dpm_resume_noirq(PMSG_RECOVER); return error; } @@ -414,10 +414,10 @@ int hibernation_restore(int platform_mode) pm_prepare_console(); suspend_console(); - error = device_suspend(PMSG_QUIESCE); + error = dpm_suspend_start(PMSG_QUIESCE); if (!error) { error = resume_target_kernel(platform_mode); - device_resume(PMSG_RECOVER); + dpm_resume_end(PMSG_RECOVER); } resume_console(); pm_restore_console(); @@ -447,14 +447,14 @@ int hibernation_platform_enter(void) entering_platform_hibernation = true; suspend_console(); - error = device_suspend(PMSG_HIBERNATE); + error = dpm_suspend_start(PMSG_HIBERNATE); if (error) { if (hibernation_ops->recover) hibernation_ops->recover(); goto Resume_devices; } - error = device_suspend_noirq(PMSG_HIBERNATE); + error = dpm_suspend_noirq(PMSG_HIBERNATE); if (error) goto Resume_devices; @@ -479,11 +479,11 @@ int hibernation_platform_enter(void) Platofrm_finish: hibernation_ops->finish(); - device_suspend_noirq(PMSG_RESTORE); + dpm_suspend_noirq(PMSG_RESTORE); Resume_devices: entering_platform_hibernation = false; - device_resume(PMSG_RESTORE); + dpm_resume_end(PMSG_RESTORE); resume_console(); Close: diff --git a/kernel/power/main.c b/kernel/power/main.c index 2f6638ee03c0..46386b9f8dd1 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -295,7 +295,7 @@ static int suspend_enter(suspend_state_t state) return error; } - error = device_suspend_noirq(PMSG_SUSPEND); + error = dpm_suspend_noirq(PMSG_SUSPEND); if (error) { printk(KERN_ERR "PM: Some devices failed to power down\n"); goto Platfrom_finish; @@ -335,7 +335,7 @@ static int suspend_enter(suspend_state_t state) suspend_ops->wake(); Power_up_devices: - device_resume_noirq(PMSG_RESUME); + dpm_resume_noirq(PMSG_RESUME); Platfrom_finish: if (suspend_ops->finish) @@ -363,7 +363,7 @@ int suspend_devices_and_enter(suspend_state_t state) } suspend_console(); suspend_test_start(); - error = device_suspend(PMSG_SUSPEND); + error = dpm_suspend_start(PMSG_SUSPEND); if (error) { printk(KERN_ERR "PM: Some devices failed to suspend\n"); goto Recover_platform; @@ -376,7 +376,7 @@ int suspend_devices_and_enter(suspend_state_t state) Resume_devices: suspend_test_start(); - device_resume(PMSG_RESUME); + dpm_resume_end(PMSG_RESUME); suspend_test_finish("resume devices"); resume_console(); Close: -- cgit v1.2.3 From c6f37f12197ac3bd2e5a35f2f0e195ae63d437de Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Sun, 24 May 2009 22:16:31 +0200 Subject: PM/Suspend: Do not shrink memory before suspend Remove the shrinking of memory from the suspend-to-RAM code, where it is not really necessary. Signed-off-by: Rafael J. Wysocki Acked-by: Nigel Cunningham Acked-by: Wu Fengguang --- kernel/power/main.c | 20 +------------------- 1 file changed, 1 insertion(+), 19 deletions(-) (limited to 'kernel') diff --git a/kernel/power/main.c b/kernel/power/main.c index 46386b9f8dd1..2a19f347bd8a 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -188,9 +188,6 @@ static void suspend_test_finish(const char *label) #endif -/* This is just an arbitrary number */ -#define FREE_PAGE_NUMBER (100) - static struct platform_suspend_ops *suspend_ops; /** @@ -226,7 +223,6 @@ int suspend_valid_only_mem(suspend_state_t state) static int suspend_prepare(void) { int error; - unsigned int free_pages; if (!suspend_ops || !suspend_ops->enter) return -EPERM; @@ -241,24 +237,10 @@ static int suspend_prepare(void) if (error) goto Finish; - if (suspend_freeze_processes()) { - error = -EAGAIN; - goto Thaw; - } - - free_pages = global_page_state(NR_FREE_PAGES); - if (free_pages < FREE_PAGE_NUMBER) { - pr_debug("PM: free some memory\n"); - shrink_all_memory(FREE_PAGE_NUMBER - free_pages); - if (nr_free_pages() < FREE_PAGE_NUMBER) { - error = -ENOMEM; - printk(KERN_ERR "PM: No enough memory\n"); - } - } + error = suspend_freeze_processes(); if (!error) return 0; - Thaw: suspend_thaw_processes(); usermodehelper_enable(); Finish: -- cgit v1.2.3 From fe419535d82724314bbf1244a0e740e4ea1bd3ae Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Thu, 11 Jun 2009 23:11:17 +0200 Subject: PM/Hibernate: Move memory shrinking to snapshot.c (rev. 2) A future patch is going to modify the memory shrinking code so that it will make memory allocations to free memory instead of using an artificial memory shrinking mechanism for that. For this purpose it is convenient to move swsusp_shrink_memory() from kernel/power/swsusp.c to kernel/power/snapshot.c, because the new memory-shrinking code is going to use things that are local to kernel/power/snapshot.c . [rev. 2: Make some functions static and remove their headers from kernel/power/power.h] Signed-off-by: Rafael J. Wysocki Acked-by: Pavel Machek Acked-by: Wu Fengguang --- kernel/power/power.h | 4 +-- kernel/power/snapshot.c | 80 +++++++++++++++++++++++++++++++++++++++++++++++-- kernel/power/swsusp.c | 76 ---------------------------------------------- 3 files changed, 79 insertions(+), 81 deletions(-) (limited to 'kernel') diff --git a/kernel/power/power.h b/kernel/power/power.h index 46b5ec7a3afb..ec4dbdfb07b4 100644 --- a/kernel/power/power.h +++ b/kernel/power/power.h @@ -74,7 +74,7 @@ extern asmlinkage int swsusp_arch_resume(void); extern int create_basic_memory_bitmaps(void); extern void free_basic_memory_bitmaps(void); -extern unsigned int count_data_pages(void); +extern int swsusp_shrink_memory(void); /** * Auxiliary structure used for reading the snapshot image data and @@ -149,7 +149,6 @@ extern int swsusp_swap_in_use(void); /* kernel/power/disk.c */ extern int swsusp_check(void); -extern int swsusp_shrink_memory(void); extern void swsusp_free(void); extern int swsusp_read(unsigned int *flags_p); extern int swsusp_write(unsigned int flags); @@ -176,7 +175,6 @@ extern int pm_notifier_call_chain(unsigned long val); #endif #ifdef CONFIG_HIGHMEM -unsigned int count_highmem_pages(void); int restore_highmem(void); #else static inline unsigned int count_highmem_pages(void) { return 0; } diff --git a/kernel/power/snapshot.c b/kernel/power/snapshot.c index 33e2e4a819f9..523a451b45d3 100644 --- a/kernel/power/snapshot.c +++ b/kernel/power/snapshot.c @@ -39,6 +39,14 @@ static int swsusp_page_is_free(struct page *); static void swsusp_set_page_forbidden(struct page *); static void swsusp_unset_page_forbidden(struct page *); +/* + * Preferred image size in bytes (tunable via /sys/power/image_size). + * When it is set to N, swsusp will do its best to ensure the image + * size will not exceed N bytes, but if that is impossible, it will + * try to create the smallest image possible. + */ +unsigned long image_size = 500 * 1024 * 1024; + /* List of PBEs needed for restoring the pages that were allocated before * the suspend and included in the suspend image, but have also been * allocated by the "resume" kernel, so their contents cannot be written @@ -840,7 +848,7 @@ static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) * pages. */ -unsigned int count_highmem_pages(void) +static unsigned int count_highmem_pages(void) { struct zone *zone; unsigned int n = 0; @@ -902,7 +910,7 @@ static struct page *saveable_page(struct zone *zone, unsigned long pfn) * pages. */ -unsigned int count_data_pages(void) +static unsigned int count_data_pages(void) { struct zone *zone; unsigned long pfn, max_zone_pfn; @@ -1058,6 +1066,74 @@ void swsusp_free(void) buffer = NULL; } +/** + * swsusp_shrink_memory - Try to free as much memory as needed + * + * ... but do not OOM-kill anyone + * + * Notice: all userland should be stopped before it is called, or + * livelock is possible. + */ + +#define SHRINK_BITE 10000 +static inline unsigned long __shrink_memory(long tmp) +{ + if (tmp > SHRINK_BITE) + tmp = SHRINK_BITE; + return shrink_all_memory(tmp); +} + +int swsusp_shrink_memory(void) +{ + long tmp; + struct zone *zone; + unsigned long pages = 0; + unsigned int i = 0; + char *p = "-\\|/"; + struct timeval start, stop; + + printk(KERN_INFO "PM: Shrinking memory... "); + do_gettimeofday(&start); + do { + long size, highmem_size; + + highmem_size = count_highmem_pages(); + size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES; + tmp = size; + size += highmem_size; + for_each_populated_zone(zone) { + tmp += snapshot_additional_pages(zone); + if (is_highmem(zone)) { + highmem_size -= + zone_page_state(zone, NR_FREE_PAGES); + } else { + tmp -= zone_page_state(zone, NR_FREE_PAGES); + tmp += zone->lowmem_reserve[ZONE_NORMAL]; + } + } + + if (highmem_size < 0) + highmem_size = 0; + + tmp += highmem_size; + if (tmp > 0) { + tmp = __shrink_memory(tmp); + if (!tmp) + return -ENOMEM; + pages += tmp; + } else if (size > image_size / PAGE_SIZE) { + tmp = __shrink_memory(size - (image_size / PAGE_SIZE)); + pages += tmp; + } + printk("\b%c", p[i++%4]); + } while (tmp > 0); + do_gettimeofday(&stop); + printk("\bdone (%lu pages freed)\n", pages); + swsusp_show_speed(&start, &stop, pages, "Freed"); + + return 0; +} + #ifdef CONFIG_HIGHMEM /** * count_pages_for_highmem - compute the number of non-highmem pages diff --git a/kernel/power/swsusp.c b/kernel/power/swsusp.c index 78c35047586d..87b901cb3927 100644 --- a/kernel/power/swsusp.c +++ b/kernel/power/swsusp.c @@ -55,14 +55,6 @@ #include "power.h" -/* - * Preferred image size in bytes (tunable via /sys/power/image_size). - * When it is set to N, swsusp will do its best to ensure the image - * size will not exceed N bytes, but if that is impossible, it will - * try to create the smallest image possible. - */ -unsigned long image_size = 500 * 1024 * 1024; - int in_suspend __nosavedata = 0; /** @@ -195,74 +187,6 @@ void swsusp_show_speed(struct timeval *start, struct timeval *stop, kps / 1000, (kps % 1000) / 10); } -/** - * swsusp_shrink_memory - Try to free as much memory as needed - * - * ... but do not OOM-kill anyone - * - * Notice: all userland should be stopped before it is called, or - * livelock is possible. - */ - -#define SHRINK_BITE 10000 -static inline unsigned long __shrink_memory(long tmp) -{ - if (tmp > SHRINK_BITE) - tmp = SHRINK_BITE; - return shrink_all_memory(tmp); -} - -int swsusp_shrink_memory(void) -{ - long tmp; - struct zone *zone; - unsigned long pages = 0; - unsigned int i = 0; - char *p = "-\\|/"; - struct timeval start, stop; - - printk(KERN_INFO "PM: Shrinking memory... "); - do_gettimeofday(&start); - do { - long size, highmem_size; - - highmem_size = count_highmem_pages(); - size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES; - tmp = size; - size += highmem_size; - for_each_populated_zone(zone) { - tmp += snapshot_additional_pages(zone); - if (is_highmem(zone)) { - highmem_size -= - zone_page_state(zone, NR_FREE_PAGES); - } else { - tmp -= zone_page_state(zone, NR_FREE_PAGES); - tmp += zone->lowmem_reserve[ZONE_NORMAL]; - } - } - - if (highmem_size < 0) - highmem_size = 0; - - tmp += highmem_size; - if (tmp > 0) { - tmp = __shrink_memory(tmp); - if (!tmp) - return -ENOMEM; - pages += tmp; - } else if (size > image_size / PAGE_SIZE) { - tmp = __shrink_memory(size - (image_size / PAGE_SIZE)); - pages += tmp; - } - printk("\b%c", p[i++%4]); - } while (tmp > 0); - do_gettimeofday(&stop); - printk("\bdone (%lu pages freed)\n", pages); - swsusp_show_speed(&start, &stop, pages, "Freed"); - - return 0; -} - /* * Platforms, like ACPI, may want us to save some memory used by them during * hibernation and to restore the contents of this memory during the subsequent -- cgit v1.2.3 From a9d7052363a6e06bb623ed1876c56c7ca5b2c6d8 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Wed, 10 Jun 2009 01:27:12 +0200 Subject: PM: Separate suspend to RAM functionality from core Move the suspend to RAM and standby code from kernel/power/main.c to two separate files, kernel/power/suspend.c containing the basic functions and kernel/power/suspend_test.c containing the automatic suspend test facility based on the RTC clock alarm. There are no changes in functionality related to these modifications. Signed-off-by: Rafael J. Wysocki Acked-by: Pavel Machek --- kernel/power/Makefile | 2 + kernel/power/main.c | 503 -------------------------------------------- kernel/power/power.h | 17 +- kernel/power/suspend.c | 300 ++++++++++++++++++++++++++ kernel/power/suspend_test.c | 187 ++++++++++++++++ 5 files changed, 505 insertions(+), 504 deletions(-) create mode 100644 kernel/power/suspend.c create mode 100644 kernel/power/suspend_test.c (limited to 'kernel') diff --git a/kernel/power/Makefile b/kernel/power/Makefile index 720ea4f781bd..c4baf1b633c4 100644 --- a/kernel/power/Makefile +++ b/kernel/power/Makefile @@ -6,6 +6,8 @@ endif obj-$(CONFIG_PM) += main.o obj-$(CONFIG_PM_SLEEP) += console.o obj-$(CONFIG_FREEZER) += process.o +obj-$(CONFIG_SUSPEND) += suspend.o +obj-$(CONFIG_PM_TEST_SUSPEND) += suspend_test.o obj-$(CONFIG_HIBERNATION) += swsusp.o disk.o snapshot.o swap.o user.o obj-$(CONFIG_MAGIC_SYSRQ) += poweroff.o diff --git a/kernel/power/main.c b/kernel/power/main.c index 2a19f347bd8a..f710e36930cc 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -8,20 +8,9 @@ * */ -#include -#include #include #include -#include -#include -#include -#include -#include -#include #include -#include -#include -#include #include "power.h" @@ -119,355 +108,6 @@ power_attr(pm_test); #endif /* CONFIG_PM_SLEEP */ -#ifdef CONFIG_SUSPEND - -static int suspend_test(int level) -{ -#ifdef CONFIG_PM_DEBUG - if (pm_test_level == level) { - printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n"); - mdelay(5000); - return 1; - } -#endif /* !CONFIG_PM_DEBUG */ - return 0; -} - -#ifdef CONFIG_PM_TEST_SUSPEND - -/* - * We test the system suspend code by setting an RTC wakealarm a short - * time in the future, then suspending. Suspending the devices won't - * normally take long ... some systems only need a few milliseconds. - * - * The time it takes is system-specific though, so when we test this - * during system bootup we allow a LOT of time. - */ -#define TEST_SUSPEND_SECONDS 5 - -static unsigned long suspend_test_start_time; - -static void suspend_test_start(void) -{ - /* FIXME Use better timebase than "jiffies", ideally a clocksource. - * What we want is a hardware counter that will work correctly even - * during the irqs-are-off stages of the suspend/resume cycle... - */ - suspend_test_start_time = jiffies; -} - -static void suspend_test_finish(const char *label) -{ - long nj = jiffies - suspend_test_start_time; - unsigned msec; - - msec = jiffies_to_msecs(abs(nj)); - pr_info("PM: %s took %d.%03d seconds\n", label, - msec / 1000, msec % 1000); - - /* Warning on suspend means the RTC alarm period needs to be - * larger -- the system was sooo slooowwww to suspend that the - * alarm (should have) fired before the system went to sleep! - * - * Warning on either suspend or resume also means the system - * has some performance issues. The stack dump of a WARN_ON - * is more likely to get the right attention than a printk... - */ - WARN(msec > (TEST_SUSPEND_SECONDS * 1000), "Component: %s\n", label); -} - -#else - -static void suspend_test_start(void) -{ -} - -static void suspend_test_finish(const char *label) -{ -} - -#endif - -static struct platform_suspend_ops *suspend_ops; - -/** - * suspend_set_ops - Set the global suspend method table. - * @ops: Pointer to ops structure. - */ - -void suspend_set_ops(struct platform_suspend_ops *ops) -{ - mutex_lock(&pm_mutex); - suspend_ops = ops; - mutex_unlock(&pm_mutex); -} - -/** - * suspend_valid_only_mem - generic memory-only valid callback - * - * Platform drivers that implement mem suspend only and only need - * to check for that in their .valid callback can use this instead - * of rolling their own .valid callback. - */ -int suspend_valid_only_mem(suspend_state_t state) -{ - return state == PM_SUSPEND_MEM; -} - -/** - * suspend_prepare - Do prep work before entering low-power state. - * - * This is common code that is called for each state that we're entering. - * Run suspend notifiers, allocate a console and stop all processes. - */ -static int suspend_prepare(void) -{ - int error; - - if (!suspend_ops || !suspend_ops->enter) - return -EPERM; - - pm_prepare_console(); - - error = pm_notifier_call_chain(PM_SUSPEND_PREPARE); - if (error) - goto Finish; - - error = usermodehelper_disable(); - if (error) - goto Finish; - - error = suspend_freeze_processes(); - if (!error) - return 0; - - suspend_thaw_processes(); - usermodehelper_enable(); - Finish: - pm_notifier_call_chain(PM_POST_SUSPEND); - pm_restore_console(); - return error; -} - -/* default implementation */ -void __attribute__ ((weak)) arch_suspend_disable_irqs(void) -{ - local_irq_disable(); -} - -/* default implementation */ -void __attribute__ ((weak)) arch_suspend_enable_irqs(void) -{ - local_irq_enable(); -} - -/** - * suspend_enter - enter the desired system sleep state. - * @state: state to enter - * - * This function should be called after devices have been suspended. - */ -static int suspend_enter(suspend_state_t state) -{ - int error; - - if (suspend_ops->prepare) { - error = suspend_ops->prepare(); - if (error) - return error; - } - - error = dpm_suspend_noirq(PMSG_SUSPEND); - if (error) { - printk(KERN_ERR "PM: Some devices failed to power down\n"); - goto Platfrom_finish; - } - - if (suspend_ops->prepare_late) { - error = suspend_ops->prepare_late(); - if (error) - goto Power_up_devices; - } - - if (suspend_test(TEST_PLATFORM)) - goto Platform_wake; - - error = disable_nonboot_cpus(); - if (error || suspend_test(TEST_CPUS)) - goto Enable_cpus; - - arch_suspend_disable_irqs(); - BUG_ON(!irqs_disabled()); - - error = sysdev_suspend(PMSG_SUSPEND); - if (!error) { - if (!suspend_test(TEST_CORE)) - error = suspend_ops->enter(state); - sysdev_resume(); - } - - arch_suspend_enable_irqs(); - BUG_ON(irqs_disabled()); - - Enable_cpus: - enable_nonboot_cpus(); - - Platform_wake: - if (suspend_ops->wake) - suspend_ops->wake(); - - Power_up_devices: - dpm_resume_noirq(PMSG_RESUME); - - Platfrom_finish: - if (suspend_ops->finish) - suspend_ops->finish(); - - return error; -} - -/** - * suspend_devices_and_enter - suspend devices and enter the desired system - * sleep state. - * @state: state to enter - */ -int suspend_devices_and_enter(suspend_state_t state) -{ - int error; - - if (!suspend_ops) - return -ENOSYS; - - if (suspend_ops->begin) { - error = suspend_ops->begin(state); - if (error) - goto Close; - } - suspend_console(); - suspend_test_start(); - error = dpm_suspend_start(PMSG_SUSPEND); - if (error) { - printk(KERN_ERR "PM: Some devices failed to suspend\n"); - goto Recover_platform; - } - suspend_test_finish("suspend devices"); - if (suspend_test(TEST_DEVICES)) - goto Recover_platform; - - suspend_enter(state); - - Resume_devices: - suspend_test_start(); - dpm_resume_end(PMSG_RESUME); - suspend_test_finish("resume devices"); - resume_console(); - Close: - if (suspend_ops->end) - suspend_ops->end(); - return error; - - Recover_platform: - if (suspend_ops->recover) - suspend_ops->recover(); - goto Resume_devices; -} - -/** - * suspend_finish - Do final work before exiting suspend sequence. - * - * Call platform code to clean up, restart processes, and free the - * console that we've allocated. This is not called for suspend-to-disk. - */ -static void suspend_finish(void) -{ - suspend_thaw_processes(); - usermodehelper_enable(); - pm_notifier_call_chain(PM_POST_SUSPEND); - pm_restore_console(); -} - - - - -static const char * const pm_states[PM_SUSPEND_MAX] = { - [PM_SUSPEND_STANDBY] = "standby", - [PM_SUSPEND_MEM] = "mem", -}; - -static inline int valid_state(suspend_state_t state) -{ - /* All states need lowlevel support and need to be valid - * to the lowlevel implementation, no valid callback - * implies that none are valid. */ - if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state)) - return 0; - return 1; -} - - -/** - * enter_state - Do common work of entering low-power state. - * @state: pm_state structure for state we're entering. - * - * Make sure we're the only ones trying to enter a sleep state. Fail - * if someone has beat us to it, since we don't want anything weird to - * happen when we wake up. - * Then, do the setup for suspend, enter the state, and cleaup (after - * we've woken up). - */ -static int enter_state(suspend_state_t state) -{ - int error; - - if (!valid_state(state)) - return -ENODEV; - - if (!mutex_trylock(&pm_mutex)) - return -EBUSY; - - printk(KERN_INFO "PM: Syncing filesystems ... "); - sys_sync(); - printk("done.\n"); - - pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]); - error = suspend_prepare(); - if (error) - goto Unlock; - - if (suspend_test(TEST_FREEZER)) - goto Finish; - - pr_debug("PM: Entering %s sleep\n", pm_states[state]); - error = suspend_devices_and_enter(state); - - Finish: - pr_debug("PM: Finishing wakeup.\n"); - suspend_finish(); - Unlock: - mutex_unlock(&pm_mutex); - return error; -} - - -/** - * pm_suspend - Externally visible function for suspending system. - * @state: Enumerated value of state to enter. - * - * Determine whether or not value is within range, get state - * structure, and enter (above). - */ - -int pm_suspend(suspend_state_t state) -{ - if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX) - return enter_state(state); - return -EINVAL; -} - -EXPORT_SYMBOL(pm_suspend); - -#endif /* CONFIG_SUSPEND */ - struct kobject *power_kobj; /** @@ -480,7 +120,6 @@ struct kobject *power_kobj; * store() accepts one of those strings, translates it into the * proper enumerated value, and initiates a suspend transition. */ - static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf) { @@ -578,7 +217,6 @@ static struct attribute_group attr_group = { .attrs = g, }; - static int __init pm_init(void) { power_kobj = kobject_create_and_add("power", NULL); @@ -588,144 +226,3 @@ static int __init pm_init(void) } core_initcall(pm_init); - - -#ifdef CONFIG_PM_TEST_SUSPEND - -#include - -/* - * To test system suspend, we need a hands-off mechanism to resume the - * system. RTCs wake alarms are a common self-contained mechanism. - */ - -static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state) -{ - static char err_readtime[] __initdata = - KERN_ERR "PM: can't read %s time, err %d\n"; - static char err_wakealarm [] __initdata = - KERN_ERR "PM: can't set %s wakealarm, err %d\n"; - static char err_suspend[] __initdata = - KERN_ERR "PM: suspend test failed, error %d\n"; - static char info_test[] __initdata = - KERN_INFO "PM: test RTC wakeup from '%s' suspend\n"; - - unsigned long now; - struct rtc_wkalrm alm; - int status; - - /* this may fail if the RTC hasn't been initialized */ - status = rtc_read_time(rtc, &alm.time); - if (status < 0) { - printk(err_readtime, dev_name(&rtc->dev), status); - return; - } - rtc_tm_to_time(&alm.time, &now); - - memset(&alm, 0, sizeof alm); - rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time); - alm.enabled = true; - - status = rtc_set_alarm(rtc, &alm); - if (status < 0) { - printk(err_wakealarm, dev_name(&rtc->dev), status); - return; - } - - if (state == PM_SUSPEND_MEM) { - printk(info_test, pm_states[state]); - status = pm_suspend(state); - if (status == -ENODEV) - state = PM_SUSPEND_STANDBY; - } - if (state == PM_SUSPEND_STANDBY) { - printk(info_test, pm_states[state]); - status = pm_suspend(state); - } - if (status < 0) - printk(err_suspend, status); - - /* Some platforms can't detect that the alarm triggered the - * wakeup, or (accordingly) disable it after it afterwards. - * It's supposed to give oneshot behavior; cope. - */ - alm.enabled = false; - rtc_set_alarm(rtc, &alm); -} - -static int __init has_wakealarm(struct device *dev, void *name_ptr) -{ - struct rtc_device *candidate = to_rtc_device(dev); - - if (!candidate->ops->set_alarm) - return 0; - if (!device_may_wakeup(candidate->dev.parent)) - return 0; - - *(const char **)name_ptr = dev_name(dev); - return 1; -} - -/* - * Kernel options like "test_suspend=mem" force suspend/resume sanity tests - * at startup time. They're normally disabled, for faster boot and because - * we can't know which states really work on this particular system. - */ -static suspend_state_t test_state __initdata = PM_SUSPEND_ON; - -static char warn_bad_state[] __initdata = - KERN_WARNING "PM: can't test '%s' suspend state\n"; - -static int __init setup_test_suspend(char *value) -{ - unsigned i; - - /* "=mem" ==> "mem" */ - value++; - for (i = 0; i < PM_SUSPEND_MAX; i++) { - if (!pm_states[i]) - continue; - if (strcmp(pm_states[i], value) != 0) - continue; - test_state = (__force suspend_state_t) i; - return 0; - } - printk(warn_bad_state, value); - return 0; -} -__setup("test_suspend", setup_test_suspend); - -static int __init test_suspend(void) -{ - static char warn_no_rtc[] __initdata = - KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n"; - - char *pony = NULL; - struct rtc_device *rtc = NULL; - - /* PM is initialized by now; is that state testable? */ - if (test_state == PM_SUSPEND_ON) - goto done; - if (!valid_state(test_state)) { - printk(warn_bad_state, pm_states[test_state]); - goto done; - } - - /* RTCs have initialized by now too ... can we use one? */ - class_find_device(rtc_class, NULL, &pony, has_wakealarm); - if (pony) - rtc = rtc_class_open(pony); - if (!rtc) { - printk(warn_no_rtc); - goto done; - } - - /* go for it */ - test_wakealarm(rtc, test_state); - rtc_class_close(rtc); -done: - return 0; -} -late_initcall(test_suspend); - -#endif /* CONFIG_PM_TEST_SUSPEND */ diff --git a/kernel/power/power.h b/kernel/power/power.h index ec4dbdfb07b4..2bd98d9fc19e 100644 --- a/kernel/power/power.h +++ b/kernel/power/power.h @@ -160,15 +160,30 @@ extern void swsusp_show_speed(struct timeval *, struct timeval *, unsigned int, char *); #ifdef CONFIG_SUSPEND -/* kernel/power/main.c */ +/* kernel/power/suspend.c */ +extern const char *const pm_states[]; + +extern bool valid_state(suspend_state_t state); extern int suspend_devices_and_enter(suspend_state_t state); +extern int enter_state(suspend_state_t state); #else /* !CONFIG_SUSPEND */ static inline int suspend_devices_and_enter(suspend_state_t state) { return -ENOSYS; } +static inline int enter_state(suspend_state_t state) { return -ENOSYS; } +static inline bool valid_state(suspend_state_t state) { return false; } #endif /* !CONFIG_SUSPEND */ +#ifdef CONFIG_PM_TEST_SUSPEND +/* kernel/power/suspend_test.c */ +extern void suspend_test_start(void); +extern void suspend_test_finish(const char *label); +#else /* !CONFIG_PM_TEST_SUSPEND */ +static inline void suspend_test_start(void) {} +static inline void suspend_test_finish(const char *label) {} +#endif /* !CONFIG_PM_TEST_SUSPEND */ + #ifdef CONFIG_PM_SLEEP /* kernel/power/main.c */ extern int pm_notifier_call_chain(unsigned long val); diff --git a/kernel/power/suspend.c b/kernel/power/suspend.c new file mode 100644 index 000000000000..6f10dfc2d3e9 --- /dev/null +++ b/kernel/power/suspend.c @@ -0,0 +1,300 @@ +/* + * kernel/power/suspend.c - Suspend to RAM and standby functionality. + * + * Copyright (c) 2003 Patrick Mochel + * Copyright (c) 2003 Open Source Development Lab + * Copyright (c) 2009 Rafael J. Wysocki , Novell Inc. + * + * This file is released under the GPLv2. + */ + +#include +#include +#include +#include +#include +#include +#include + +#include "power.h" + +const char *const pm_states[PM_SUSPEND_MAX] = { + [PM_SUSPEND_STANDBY] = "standby", + [PM_SUSPEND_MEM] = "mem", +}; + +static struct platform_suspend_ops *suspend_ops; + +/** + * suspend_set_ops - Set the global suspend method table. + * @ops: Pointer to ops structure. + */ +void suspend_set_ops(struct platform_suspend_ops *ops) +{ + mutex_lock(&pm_mutex); + suspend_ops = ops; + mutex_unlock(&pm_mutex); +} + +bool valid_state(suspend_state_t state) +{ + /* + * All states need lowlevel support and need to be valid to the lowlevel + * implementation, no valid callback implies that none are valid. + */ + return suspend_ops && suspend_ops->valid && suspend_ops->valid(state); +} + +/** + * suspend_valid_only_mem - generic memory-only valid callback + * + * Platform drivers that implement mem suspend only and only need + * to check for that in their .valid callback can use this instead + * of rolling their own .valid callback. + */ +int suspend_valid_only_mem(suspend_state_t state) +{ + return state == PM_SUSPEND_MEM; +} + +static int suspend_test(int level) +{ +#ifdef CONFIG_PM_DEBUG + if (pm_test_level == level) { + printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n"); + mdelay(5000); + return 1; + } +#endif /* !CONFIG_PM_DEBUG */ + return 0; +} + +/** + * suspend_prepare - Do prep work before entering low-power state. + * + * This is common code that is called for each state that we're entering. + * Run suspend notifiers, allocate a console and stop all processes. + */ +static int suspend_prepare(void) +{ + int error; + + if (!suspend_ops || !suspend_ops->enter) + return -EPERM; + + pm_prepare_console(); + + error = pm_notifier_call_chain(PM_SUSPEND_PREPARE); + if (error) + goto Finish; + + error = usermodehelper_disable(); + if (error) + goto Finish; + + error = suspend_freeze_processes(); + if (!error) + return 0; + + suspend_thaw_processes(); + usermodehelper_enable(); + Finish: + pm_notifier_call_chain(PM_POST_SUSPEND); + pm_restore_console(); + return error; +} + +/* default implementation */ +void __attribute__ ((weak)) arch_suspend_disable_irqs(void) +{ + local_irq_disable(); +} + +/* default implementation */ +void __attribute__ ((weak)) arch_suspend_enable_irqs(void) +{ + local_irq_enable(); +} + +/** + * suspend_enter - enter the desired system sleep state. + * @state: state to enter + * + * This function should be called after devices have been suspended. + */ +static int suspend_enter(suspend_state_t state) +{ + int error; + + if (suspend_ops->prepare) { + error = suspend_ops->prepare(); + if (error) + return error; + } + + error = dpm_suspend_noirq(PMSG_SUSPEND); + if (error) { + printk(KERN_ERR "PM: Some devices failed to power down\n"); + goto Platfrom_finish; + } + + if (suspend_ops->prepare_late) { + error = suspend_ops->prepare_late(); + if (error) + goto Power_up_devices; + } + + if (suspend_test(TEST_PLATFORM)) + goto Platform_wake; + + error = disable_nonboot_cpus(); + if (error || suspend_test(TEST_CPUS)) + goto Enable_cpus; + + arch_suspend_disable_irqs(); + BUG_ON(!irqs_disabled()); + + error = sysdev_suspend(PMSG_SUSPEND); + if (!error) { + if (!suspend_test(TEST_CORE)) + error = suspend_ops->enter(state); + sysdev_resume(); + } + + arch_suspend_enable_irqs(); + BUG_ON(irqs_disabled()); + + Enable_cpus: + enable_nonboot_cpus(); + + Platform_wake: + if (suspend_ops->wake) + suspend_ops->wake(); + + Power_up_devices: + dpm_resume_noirq(PMSG_RESUME); + + Platfrom_finish: + if (suspend_ops->finish) + suspend_ops->finish(); + + return error; +} + +/** + * suspend_devices_and_enter - suspend devices and enter the desired system + * sleep state. + * @state: state to enter + */ +int suspend_devices_and_enter(suspend_state_t state) +{ + int error; + + if (!suspend_ops) + return -ENOSYS; + + if (suspend_ops->begin) { + error = suspend_ops->begin(state); + if (error) + goto Close; + } + suspend_console(); + suspend_test_start(); + error = dpm_suspend_start(PMSG_SUSPEND); + if (error) { + printk(KERN_ERR "PM: Some devices failed to suspend\n"); + goto Recover_platform; + } + suspend_test_finish("suspend devices"); + if (suspend_test(TEST_DEVICES)) + goto Recover_platform; + + suspend_enter(state); + + Resume_devices: + suspend_test_start(); + dpm_resume_end(PMSG_RESUME); + suspend_test_finish("resume devices"); + resume_console(); + Close: + if (suspend_ops->end) + suspend_ops->end(); + return error; + + Recover_platform: + if (suspend_ops->recover) + suspend_ops->recover(); + goto Resume_devices; +} + +/** + * suspend_finish - Do final work before exiting suspend sequence. + * + * Call platform code to clean up, restart processes, and free the + * console that we've allocated. This is not called for suspend-to-disk. + */ +static void suspend_finish(void) +{ + suspend_thaw_processes(); + usermodehelper_enable(); + pm_notifier_call_chain(PM_POST_SUSPEND); + pm_restore_console(); +} + +/** + * enter_state - Do common work of entering low-power state. + * @state: pm_state structure for state we're entering. + * + * Make sure we're the only ones trying to enter a sleep state. Fail + * if someone has beat us to it, since we don't want anything weird to + * happen when we wake up. + * Then, do the setup for suspend, enter the state, and cleaup (after + * we've woken up). + */ +int enter_state(suspend_state_t state) +{ + int error; + + if (!valid_state(state)) + return -ENODEV; + + if (!mutex_trylock(&pm_mutex)) + return -EBUSY; + + printk(KERN_INFO "PM: Syncing filesystems ... "); + sys_sync(); + printk("done.\n"); + + pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]); + error = suspend_prepare(); + if (error) + goto Unlock; + + if (suspend_test(TEST_FREEZER)) + goto Finish; + + pr_debug("PM: Entering %s sleep\n", pm_states[state]); + error = suspend_devices_and_enter(state); + + Finish: + pr_debug("PM: Finishing wakeup.\n"); + suspend_finish(); + Unlock: + mutex_unlock(&pm_mutex); + return error; +} + +/** + * pm_suspend - Externally visible function for suspending system. + * @state: Enumerated value of state to enter. + * + * Determine whether or not value is within range, get state + * structure, and enter (above). + */ +int pm_suspend(suspend_state_t state) +{ + if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX) + return enter_state(state); + return -EINVAL; +} +EXPORT_SYMBOL(pm_suspend); diff --git a/kernel/power/suspend_test.c b/kernel/power/suspend_test.c new file mode 100644 index 000000000000..17d8bb1acf9c --- /dev/null +++ b/kernel/power/suspend_test.c @@ -0,0 +1,187 @@ +/* + * kernel/power/suspend_test.c - Suspend to RAM and standby test facility. + * + * Copyright (c) 2009 Pavel Machek + * + * This file is released under the GPLv2. + */ + +#include +#include + +#include "power.h" + +/* + * We test the system suspend code by setting an RTC wakealarm a short + * time in the future, then suspending. Suspending the devices won't + * normally take long ... some systems only need a few milliseconds. + * + * The time it takes is system-specific though, so when we test this + * during system bootup we allow a LOT of time. + */ +#define TEST_SUSPEND_SECONDS 5 + +static unsigned long suspend_test_start_time; + +void suspend_test_start(void) +{ + /* FIXME Use better timebase than "jiffies", ideally a clocksource. + * What we want is a hardware counter that will work correctly even + * during the irqs-are-off stages of the suspend/resume cycle... + */ + suspend_test_start_time = jiffies; +} + +void suspend_test_finish(const char *label) +{ + long nj = jiffies - suspend_test_start_time; + unsigned msec; + + msec = jiffies_to_msecs(abs(nj)); + pr_info("PM: %s took %d.%03d seconds\n", label, + msec / 1000, msec % 1000); + + /* Warning on suspend means the RTC alarm period needs to be + * larger -- the system was sooo slooowwww to suspend that the + * alarm (should have) fired before the system went to sleep! + * + * Warning on either suspend or resume also means the system + * has some performance issues. The stack dump of a WARN_ON + * is more likely to get the right attention than a printk... + */ + WARN(msec > (TEST_SUSPEND_SECONDS * 1000), "Component: %s\n", label); +} + +/* + * To test system suspend, we need a hands-off mechanism to resume the + * system. RTCs wake alarms are a common self-contained mechanism. + */ + +static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state) +{ + static char err_readtime[] __initdata = + KERN_ERR "PM: can't read %s time, err %d\n"; + static char err_wakealarm [] __initdata = + KERN_ERR "PM: can't set %s wakealarm, err %d\n"; + static char err_suspend[] __initdata = + KERN_ERR "PM: suspend test failed, error %d\n"; + static char info_test[] __initdata = + KERN_INFO "PM: test RTC wakeup from '%s' suspend\n"; + + unsigned long now; + struct rtc_wkalrm alm; + int status; + + /* this may fail if the RTC hasn't been initialized */ + status = rtc_read_time(rtc, &alm.time); + if (status < 0) { + printk(err_readtime, dev_name(&rtc->dev), status); + return; + } + rtc_tm_to_time(&alm.time, &now); + + memset(&alm, 0, sizeof alm); + rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time); + alm.enabled = true; + + status = rtc_set_alarm(rtc, &alm); + if (status < 0) { + printk(err_wakealarm, dev_name(&rtc->dev), status); + return; + } + + if (state == PM_SUSPEND_MEM) { + printk(info_test, pm_states[state]); + status = pm_suspend(state); + if (status == -ENODEV) + state = PM_SUSPEND_STANDBY; + } + if (state == PM_SUSPEND_STANDBY) { + printk(info_test, pm_states[state]); + status = pm_suspend(state); + } + if (status < 0) + printk(err_suspend, status); + + /* Some platforms can't detect that the alarm triggered the + * wakeup, or (accordingly) disable it after it afterwards. + * It's supposed to give oneshot behavior; cope. + */ + alm.enabled = false; + rtc_set_alarm(rtc, &alm); +} + +static int __init has_wakealarm(struct device *dev, void *name_ptr) +{ + struct rtc_device *candidate = to_rtc_device(dev); + + if (!candidate->ops->set_alarm) + return 0; + if (!device_may_wakeup(candidate->dev.parent)) + return 0; + + *(const char **)name_ptr = dev_name(dev); + return 1; +} + +/* + * Kernel options like "test_suspend=mem" force suspend/resume sanity tests + * at startup time. They're normally disabled, for faster boot and because + * we can't know which states really work on this particular system. + */ +static suspend_state_t test_state __initdata = PM_SUSPEND_ON; + +static char warn_bad_state[] __initdata = + KERN_WARNING "PM: can't test '%s' suspend state\n"; + +static int __init setup_test_suspend(char *value) +{ + unsigned i; + + /* "=mem" ==> "mem" */ + value++; + for (i = 0; i < PM_SUSPEND_MAX; i++) { + if (!pm_states[i]) + continue; + if (strcmp(pm_states[i], value) != 0) + continue; + test_state = (__force suspend_state_t) i; + return 0; + } + printk(warn_bad_state, value); + return 0; +} +__setup("test_suspend", setup_test_suspend); + +static int __init test_suspend(void) +{ + static char warn_no_rtc[] __initdata = + KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n"; + + char *pony = NULL; + struct rtc_device *rtc = NULL; + + /* PM is initialized by now; is that state testable? */ + if (test_state == PM_SUSPEND_ON) + goto done; + if (!valid_state(test_state)) { + printk(warn_bad_state, pm_states[test_state]); + goto done; + } + + /* RTCs have initialized by now too ... can we use one? */ + class_find_device(rtc_class, NULL, &pony, has_wakealarm); + if (pony) + rtc = rtc_class_open(pony); + if (!rtc) { + printk(warn_no_rtc); + goto done; + } + + /* go for it */ + test_wakealarm(rtc, test_state); + rtc_class_close(rtc); +done: + return 0; +} +late_initcall(test_suspend); -- cgit v1.2.3 From 8b759b84c8b3c27ccc8dd787294636297b3ebb40 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Wed, 10 Jun 2009 01:27:49 +0200 Subject: PM/Hibernate: Rename disk.c to hibernate.c Change the name of kernel/power/disk.c to kernel/power/hibernate.c in analogy with the file names introduced by the changes that separated the suspend to RAM and standby funtionality from the common PM functions. Signed-off-by: Rafael J. Wysocki Acked-by: Pavel Machek --- kernel/power/Makefile | 2 +- kernel/power/disk.c | 955 ----------------------------------------------- kernel/power/hibernate.c | 955 +++++++++++++++++++++++++++++++++++++++++++++++ kernel/power/power.h | 4 +- 4 files changed, 958 insertions(+), 958 deletions(-) delete mode 100644 kernel/power/disk.c create mode 100644 kernel/power/hibernate.c (limited to 'kernel') diff --git a/kernel/power/Makefile b/kernel/power/Makefile index c4baf1b633c4..eadb17fc8f5e 100644 --- a/kernel/power/Makefile +++ b/kernel/power/Makefile @@ -8,6 +8,6 @@ obj-$(CONFIG_PM_SLEEP) += console.o obj-$(CONFIG_FREEZER) += process.o obj-$(CONFIG_SUSPEND) += suspend.o obj-$(CONFIG_PM_TEST_SUSPEND) += suspend_test.o -obj-$(CONFIG_HIBERNATION) += swsusp.o disk.o snapshot.o swap.o user.o +obj-$(CONFIG_HIBERNATION) += swsusp.o hibernate.o snapshot.o swap.o user.o obj-$(CONFIG_MAGIC_SYSRQ) += poweroff.o diff --git a/kernel/power/disk.c b/kernel/power/disk.c deleted file mode 100644 index a9beba68b6c7..000000000000 --- a/kernel/power/disk.c +++ /dev/null @@ -1,955 +0,0 @@ -/* - * kernel/power/disk.c - Suspend-to-disk support. - * - * Copyright (c) 2003 Patrick Mochel - * Copyright (c) 2003 Open Source Development Lab - * Copyright (c) 2004 Pavel Machek - * - * This file is released under the GPLv2. - * - */ - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#include "power.h" - - -static int noresume = 0; -static char resume_file[256] = CONFIG_PM_STD_PARTITION; -dev_t swsusp_resume_device; -sector_t swsusp_resume_block; - -enum { - HIBERNATION_INVALID, - HIBERNATION_PLATFORM, - HIBERNATION_TEST, - HIBERNATION_TESTPROC, - HIBERNATION_SHUTDOWN, - HIBERNATION_REBOOT, - /* keep last */ - __HIBERNATION_AFTER_LAST -}; -#define HIBERNATION_MAX (__HIBERNATION_AFTER_LAST-1) -#define HIBERNATION_FIRST (HIBERNATION_INVALID + 1) - -static int hibernation_mode = HIBERNATION_SHUTDOWN; - -static struct platform_hibernation_ops *hibernation_ops; - -/** - * hibernation_set_ops - set the global hibernate operations - * @ops: the hibernation operations to use in subsequent hibernation transitions - */ - -void hibernation_set_ops(struct platform_hibernation_ops *ops) -{ - if (ops && !(ops->begin && ops->end && ops->pre_snapshot - && ops->prepare && ops->finish && ops->enter && ops->pre_restore - && ops->restore_cleanup)) { - WARN_ON(1); - return; - } - mutex_lock(&pm_mutex); - hibernation_ops = ops; - if (ops) - hibernation_mode = HIBERNATION_PLATFORM; - else if (hibernation_mode == HIBERNATION_PLATFORM) - hibernation_mode = HIBERNATION_SHUTDOWN; - - mutex_unlock(&pm_mutex); -} - -static bool entering_platform_hibernation; - -bool system_entering_hibernation(void) -{ - return entering_platform_hibernation; -} -EXPORT_SYMBOL(system_entering_hibernation); - -#ifdef CONFIG_PM_DEBUG -static void hibernation_debug_sleep(void) -{ - printk(KERN_INFO "hibernation debug: Waiting for 5 seconds.\n"); - mdelay(5000); -} - -static int hibernation_testmode(int mode) -{ - if (hibernation_mode == mode) { - hibernation_debug_sleep(); - return 1; - } - return 0; -} - -static int hibernation_test(int level) -{ - if (pm_test_level == level) { - hibernation_debug_sleep(); - return 1; - } - return 0; -} -#else /* !CONFIG_PM_DEBUG */ -static int hibernation_testmode(int mode) { return 0; } -static int hibernation_test(int level) { return 0; } -#endif /* !CONFIG_PM_DEBUG */ - -/** - * platform_begin - tell the platform driver that we're starting - * hibernation - */ - -static int platform_begin(int platform_mode) -{ - return (platform_mode && hibernation_ops) ? - hibernation_ops->begin() : 0; -} - -/** - * platform_end - tell the platform driver that we've entered the - * working state - */ - -static void platform_end(int platform_mode) -{ - if (platform_mode && hibernation_ops) - hibernation_ops->end(); -} - -/** - * platform_pre_snapshot - prepare the machine for hibernation using the - * platform driver if so configured and return an error code if it fails - */ - -static int platform_pre_snapshot(int platform_mode) -{ - return (platform_mode && hibernation_ops) ? - hibernation_ops->pre_snapshot() : 0; -} - -/** - * platform_leave - prepare the machine for switching to the normal mode - * of operation using the platform driver (called with interrupts disabled) - */ - -static void platform_leave(int platform_mode) -{ - if (platform_mode && hibernation_ops) - hibernation_ops->leave(); -} - -/** - * platform_finish - switch the machine to the normal mode of operation - * using the platform driver (must be called after platform_prepare()) - */ - -static void platform_finish(int platform_mode) -{ - if (platform_mode && hibernation_ops) - hibernation_ops->finish(); -} - -/** - * platform_pre_restore - prepare the platform for the restoration from a - * hibernation image. If the restore fails after this function has been - * called, platform_restore_cleanup() must be called. - */ - -static int platform_pre_restore(int platform_mode) -{ - return (platform_mode && hibernation_ops) ? - hibernation_ops->pre_restore() : 0; -} - -/** - * platform_restore_cleanup - switch the platform to the normal mode of - * operation after a failing restore. If platform_pre_restore() has been - * called before the failing restore, this function must be called too, - * regardless of the result of platform_pre_restore(). - */ - -static void platform_restore_cleanup(int platform_mode) -{ - if (platform_mode && hibernation_ops) - hibernation_ops->restore_cleanup(); -} - -/** - * platform_recover - recover the platform from a failure to suspend - * devices. - */ - -static void platform_recover(int platform_mode) -{ - if (platform_mode && hibernation_ops && hibernation_ops->recover) - hibernation_ops->recover(); -} - -/** - * create_image - freeze devices that need to be frozen with interrupts - * off, create the hibernation image and thaw those devices. Control - * reappears in this routine after a restore. - */ - -static int create_image(int platform_mode) -{ - int error; - - error = arch_prepare_suspend(); - if (error) - return error; - - /* At this point, dpm_suspend_start() has been called, but *not* - * dpm_suspend_noirq(). We *must* call dpm_suspend_noirq() now. - * Otherwise, drivers for some devices (e.g. interrupt controllers) - * become desynchronized with the actual state of the hardware - * at resume time, and evil weirdness ensues. - */ - error = dpm_suspend_noirq(PMSG_FREEZE); - if (error) { - printk(KERN_ERR "PM: Some devices failed to power down, " - "aborting hibernation\n"); - return error; - } - - error = platform_pre_snapshot(platform_mode); - if (error || hibernation_test(TEST_PLATFORM)) - goto Platform_finish; - - error = disable_nonboot_cpus(); - if (error || hibernation_test(TEST_CPUS) - || hibernation_testmode(HIBERNATION_TEST)) - goto Enable_cpus; - - local_irq_disable(); - - error = sysdev_suspend(PMSG_FREEZE); - if (error) { - printk(KERN_ERR "PM: Some system devices failed to power down, " - "aborting hibernation\n"); - goto Enable_irqs; - } - - if (hibernation_test(TEST_CORE)) - goto Power_up; - - in_suspend = 1; - save_processor_state(); - error = swsusp_arch_suspend(); - if (error) - printk(KERN_ERR "PM: Error %d creating hibernation image\n", - error); - /* Restore control flow magically appears here */ - restore_processor_state(); - if (!in_suspend) - platform_leave(platform_mode); - - Power_up: - sysdev_resume(); - /* NOTE: dpm_resume_noirq() is just a resume() for devices - * that suspended with irqs off ... no overall powerup. - */ - - Enable_irqs: - local_irq_enable(); - - Enable_cpus: - enable_nonboot_cpus(); - - Platform_finish: - platform_finish(platform_mode); - - dpm_resume_noirq(in_suspend ? - (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); - - return error; -} - -/** - * hibernation_snapshot - quiesce devices and create the hibernation - * snapshot image. - * @platform_mode - if set, use the platform driver, if available, to - * prepare the platform firmware for the power transition. - * - * Must be called with pm_mutex held - */ - -int hibernation_snapshot(int platform_mode) -{ - int error; - - error = platform_begin(platform_mode); - if (error) - return error; - - /* Free memory before shutting down devices. */ - error = swsusp_shrink_memory(); - if (error) - goto Close; - - suspend_console(); - error = dpm_suspend_start(PMSG_FREEZE); - if (error) - goto Recover_platform; - - if (hibernation_test(TEST_DEVICES)) - goto Recover_platform; - - error = create_image(platform_mode); - /* Control returns here after successful restore */ - - Resume_devices: - dpm_resume_end(in_suspend ? - (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); - resume_console(); - Close: - platform_end(platform_mode); - return error; - - Recover_platform: - platform_recover(platform_mode); - goto Resume_devices; -} - -/** - * resume_target_kernel - prepare devices that need to be suspended with - * interrupts off, restore the contents of highmem that have not been - * restored yet from the image and run the low level code that will restore - * the remaining contents of memory and switch to the just restored target - * kernel. - */ - -static int resume_target_kernel(bool platform_mode) -{ - int error; - - error = dpm_suspend_noirq(PMSG_QUIESCE); - if (error) { - printk(KERN_ERR "PM: Some devices failed to power down, " - "aborting resume\n"); - return error; - } - - error = platform_pre_restore(platform_mode); - if (error) - goto Cleanup; - - error = disable_nonboot_cpus(); - if (error) - goto Enable_cpus; - - local_irq_disable(); - - error = sysdev_suspend(PMSG_QUIESCE); - if (error) - goto Enable_irqs; - - /* We'll ignore saved state, but this gets preempt count (etc) right */ - save_processor_state(); - error = restore_highmem(); - if (!error) { - error = swsusp_arch_resume(); - /* - * The code below is only ever reached in case of a failure. - * Otherwise execution continues at place where - * swsusp_arch_suspend() was called - */ - BUG_ON(!error); - /* This call to restore_highmem() undos the previous one */ - restore_highmem(); - } - /* - * The only reason why swsusp_arch_resume() can fail is memory being - * very tight, so we have to free it as soon as we can to avoid - * subsequent failures - */ - swsusp_free(); - restore_processor_state(); - touch_softlockup_watchdog(); - - sysdev_resume(); - - Enable_irqs: - local_irq_enable(); - - Enable_cpus: - enable_nonboot_cpus(); - - Cleanup: - platform_restore_cleanup(platform_mode); - - dpm_resume_noirq(PMSG_RECOVER); - - return error; -} - -/** - * hibernation_restore - quiesce devices and restore the hibernation - * snapshot image. If successful, control returns in hibernation_snaphot() - * @platform_mode - if set, use the platform driver, if available, to - * prepare the platform firmware for the transition. - * - * Must be called with pm_mutex held - */ - -int hibernation_restore(int platform_mode) -{ - int error; - - pm_prepare_console(); - suspend_console(); - error = dpm_suspend_start(PMSG_QUIESCE); - if (!error) { - error = resume_target_kernel(platform_mode); - dpm_resume_end(PMSG_RECOVER); - } - resume_console(); - pm_restore_console(); - return error; -} - -/** - * hibernation_platform_enter - enter the hibernation state using the - * platform driver (if available) - */ - -int hibernation_platform_enter(void) -{ - int error; - - if (!hibernation_ops) - return -ENOSYS; - - /* - * We have cancelled the power transition by running - * hibernation_ops->finish() before saving the image, so we should let - * the firmware know that we're going to enter the sleep state after all - */ - error = hibernation_ops->begin(); - if (error) - goto Close; - - entering_platform_hibernation = true; - suspend_console(); - error = dpm_suspend_start(PMSG_HIBERNATE); - if (error) { - if (hibernation_ops->recover) - hibernation_ops->recover(); - goto Resume_devices; - } - - error = dpm_suspend_noirq(PMSG_HIBERNATE); - if (error) - goto Resume_devices; - - error = hibernation_ops->prepare(); - if (error) - goto Platofrm_finish; - - error = disable_nonboot_cpus(); - if (error) - goto Platofrm_finish; - - local_irq_disable(); - sysdev_suspend(PMSG_HIBERNATE); - hibernation_ops->enter(); - /* We should never get here */ - while (1); - - /* - * We don't need to reenable the nonboot CPUs or resume consoles, since - * the system is going to be halted anyway. - */ - Platofrm_finish: - hibernation_ops->finish(); - - dpm_suspend_noirq(PMSG_RESTORE); - - Resume_devices: - entering_platform_hibernation = false; - dpm_resume_end(PMSG_RESTORE); - resume_console(); - - Close: - hibernation_ops->end(); - - return error; -} - -/** - * power_down - Shut the machine down for hibernation. - * - * Use the platform driver, if configured so; otherwise try - * to power off or reboot. - */ - -static void power_down(void) -{ - switch (hibernation_mode) { - case HIBERNATION_TEST: - case HIBERNATION_TESTPROC: - break; - case HIBERNATION_REBOOT: - kernel_restart(NULL); - break; - case HIBERNATION_PLATFORM: - hibernation_platform_enter(); - case HIBERNATION_SHUTDOWN: - kernel_power_off(); - break; - } - kernel_halt(); - /* - * Valid image is on the disk, if we continue we risk serious data - * corruption after resume. - */ - printk(KERN_CRIT "PM: Please power down manually\n"); - while(1); -} - -static int prepare_processes(void) -{ - int error = 0; - - if (freeze_processes()) { - error = -EBUSY; - thaw_processes(); - } - return error; -} - -/** - * hibernate - The granpappy of the built-in hibernation management - */ - -int hibernate(void) -{ - int error; - - mutex_lock(&pm_mutex); - /* The snapshot device should not be opened while we're running */ - if (!atomic_add_unless(&snapshot_device_available, -1, 0)) { - error = -EBUSY; - goto Unlock; - } - - pm_prepare_console(); - error = pm_notifier_call_chain(PM_HIBERNATION_PREPARE); - if (error) - goto Exit; - - error = usermodehelper_disable(); - if (error) - goto Exit; - - /* Allocate memory management structures */ - error = create_basic_memory_bitmaps(); - if (error) - goto Exit; - - printk(KERN_INFO "PM: Syncing filesystems ... "); - sys_sync(); - printk("done.\n"); - - error = prepare_processes(); - if (error) - goto Finish; - - if (hibernation_test(TEST_FREEZER)) - goto Thaw; - - if (hibernation_testmode(HIBERNATION_TESTPROC)) - goto Thaw; - - error = hibernation_snapshot(hibernation_mode == HIBERNATION_PLATFORM); - if (in_suspend && !error) { - unsigned int flags = 0; - - if (hibernation_mode == HIBERNATION_PLATFORM) - flags |= SF_PLATFORM_MODE; - pr_debug("PM: writing image.\n"); - error = swsusp_write(flags); - swsusp_free(); - if (!error) - power_down(); - } else { - pr_debug("PM: Image restored successfully.\n"); - swsusp_free(); - } - Thaw: - thaw_processes(); - Finish: - free_basic_memory_bitmaps(); - usermodehelper_enable(); - Exit: - pm_notifier_call_chain(PM_POST_HIBERNATION); - pm_restore_console(); - atomic_inc(&snapshot_device_available); - Unlock: - mutex_unlock(&pm_mutex); - return error; -} - - -/** - * software_resume - Resume from a saved image. - * - * Called as a late_initcall (so all devices are discovered and - * initialized), we call swsusp to see if we have a saved image or not. - * If so, we quiesce devices, the restore the saved image. We will - * return above (in hibernate() ) if everything goes well. - * Otherwise, we fail gracefully and return to the normally - * scheduled program. - * - */ - -static int software_resume(void) -{ - int error; - unsigned int flags; - - /* - * If the user said "noresume".. bail out early. - */ - if (noresume) - return 0; - - /* - * name_to_dev_t() below takes a sysfs buffer mutex when sysfs - * is configured into the kernel. Since the regular hibernate - * trigger path is via sysfs which takes a buffer mutex before - * calling hibernate functions (which take pm_mutex) this can - * cause lockdep to complain about a possible ABBA deadlock - * which cannot happen since we're in the boot code here and - * sysfs can't be invoked yet. Therefore, we use a subclass - * here to avoid lockdep complaining. - */ - mutex_lock_nested(&pm_mutex, SINGLE_DEPTH_NESTING); - - if (swsusp_resume_device) - goto Check_image; - - if (!strlen(resume_file)) { - error = -ENOENT; - goto Unlock; - } - - pr_debug("PM: Checking image partition %s\n", resume_file); - - /* Check if the device is there */ - swsusp_resume_device = name_to_dev_t(resume_file); - if (!swsusp_resume_device) { - /* - * Some device discovery might still be in progress; we need - * to wait for this to finish. - */ - wait_for_device_probe(); - /* - * We can't depend on SCSI devices being available after loading - * one of their modules until scsi_complete_async_scans() is - * called and the resume device usually is a SCSI one. - */ - scsi_complete_async_scans(); - - swsusp_resume_device = name_to_dev_t(resume_file); - if (!swsusp_resume_device) { - error = -ENODEV; - goto Unlock; - } - } - - Check_image: - pr_debug("PM: Resume from partition %d:%d\n", - MAJOR(swsusp_resume_device), MINOR(swsusp_resume_device)); - - pr_debug("PM: Checking hibernation image.\n"); - error = swsusp_check(); - if (error) - goto Unlock; - - /* The snapshot device should not be opened while we're running */ - if (!atomic_add_unless(&snapshot_device_available, -1, 0)) { - error = -EBUSY; - goto Unlock; - } - - pm_prepare_console(); - error = pm_notifier_call_chain(PM_RESTORE_PREPARE); - if (error) - goto Finish; - - error = usermodehelper_disable(); - if (error) - goto Finish; - - error = create_basic_memory_bitmaps(); - if (error) - goto Finish; - - pr_debug("PM: Preparing processes for restore.\n"); - error = prepare_processes(); - if (error) { - swsusp_close(FMODE_READ); - goto Done; - } - - pr_debug("PM: Reading hibernation image.\n"); - - error = swsusp_read(&flags); - if (!error) - hibernation_restore(flags & SF_PLATFORM_MODE); - - printk(KERN_ERR "PM: Restore failed, recovering.\n"); - swsusp_free(); - thaw_processes(); - Done: - free_basic_memory_bitmaps(); - usermodehelper_enable(); - Finish: - pm_notifier_call_chain(PM_POST_RESTORE); - pm_restore_console(); - atomic_inc(&snapshot_device_available); - /* For success case, the suspend path will release the lock */ - Unlock: - mutex_unlock(&pm_mutex); - pr_debug("PM: Resume from disk failed.\n"); - return error; -} - -late_initcall(software_resume); - - -static const char * const hibernation_modes[] = { - [HIBERNATION_PLATFORM] = "platform", - [HIBERNATION_SHUTDOWN] = "shutdown", - [HIBERNATION_REBOOT] = "reboot", - [HIBERNATION_TEST] = "test", - [HIBERNATION_TESTPROC] = "testproc", -}; - -/** - * disk - Control hibernation mode - * - * Suspend-to-disk can be handled in several ways. We have a few options - * for putting the system to sleep - using the platform driver (e.g. ACPI - * or other hibernation_ops), powering off the system or rebooting the - * system (for testing) as well as the two test modes. - * - * The system can support 'platform', and that is known a priori (and - * encoded by the presence of hibernation_ops). However, the user may - * choose 'shutdown' or 'reboot' as alternatives, as well as one fo the - * test modes, 'test' or 'testproc'. - * - * show() will display what the mode is currently set to. - * store() will accept one of - * - * 'platform' - * 'shutdown' - * 'reboot' - * 'test' - * 'testproc' - * - * It will only change to 'platform' if the system - * supports it (as determined by having hibernation_ops). - */ - -static ssize_t disk_show(struct kobject *kobj, struct kobj_attribute *attr, - char *buf) -{ - int i; - char *start = buf; - - for (i = HIBERNATION_FIRST; i <= HIBERNATION_MAX; i++) { - if (!hibernation_modes[i]) - continue; - switch (i) { - case HIBERNATION_SHUTDOWN: - case HIBERNATION_REBOOT: - case HIBERNATION_TEST: - case HIBERNATION_TESTPROC: - break; - case HIBERNATION_PLATFORM: - if (hibernation_ops) - break; - /* not a valid mode, continue with loop */ - continue; - } - if (i == hibernation_mode) - buf += sprintf(buf, "[%s] ", hibernation_modes[i]); - else - buf += sprintf(buf, "%s ", hibernation_modes[i]); - } - buf += sprintf(buf, "\n"); - return buf-start; -} - - -static ssize_t disk_store(struct kobject *kobj, struct kobj_attribute *attr, - const char *buf, size_t n) -{ - int error = 0; - int i; - int len; - char *p; - int mode = HIBERNATION_INVALID; - - p = memchr(buf, '\n', n); - len = p ? p - buf : n; - - mutex_lock(&pm_mutex); - for (i = HIBERNATION_FIRST; i <= HIBERNATION_MAX; i++) { - if (len == strlen(hibernation_modes[i]) - && !strncmp(buf, hibernation_modes[i], len)) { - mode = i; - break; - } - } - if (mode != HIBERNATION_INVALID) { - switch (mode) { - case HIBERNATION_SHUTDOWN: - case HIBERNATION_REBOOT: - case HIBERNATION_TEST: - case HIBERNATION_TESTPROC: - hibernation_mode = mode; - break; - case HIBERNATION_PLATFORM: - if (hibernation_ops) - hibernation_mode = mode; - else - error = -EINVAL; - } - } else - error = -EINVAL; - - if (!error) - pr_debug("PM: Hibernation mode set to '%s'\n", - hibernation_modes[mode]); - mutex_unlock(&pm_mutex); - return error ? error : n; -} - -power_attr(disk); - -static ssize_t resume_show(struct kobject *kobj, struct kobj_attribute *attr, - char *buf) -{ - return sprintf(buf,"%d:%d\n", MAJOR(swsusp_resume_device), - MINOR(swsusp_resume_device)); -} - -static ssize_t resume_store(struct kobject *kobj, struct kobj_attribute *attr, - const char *buf, size_t n) -{ - unsigned int maj, min; - dev_t res; - int ret = -EINVAL; - - if (sscanf(buf, "%u:%u", &maj, &min) != 2) - goto out; - - res = MKDEV(maj,min); - if (maj != MAJOR(res) || min != MINOR(res)) - goto out; - - mutex_lock(&pm_mutex); - swsusp_resume_device = res; - mutex_unlock(&pm_mutex); - printk(KERN_INFO "PM: Starting manual resume from disk\n"); - noresume = 0; - software_resume(); - ret = n; - out: - return ret; -} - -power_attr(resume); - -static ssize_t image_size_show(struct kobject *kobj, struct kobj_attribute *attr, - char *buf) -{ - return sprintf(buf, "%lu\n", image_size); -} - -static ssize_t image_size_store(struct kobject *kobj, struct kobj_attribute *attr, - const char *buf, size_t n) -{ - unsigned long size; - - if (sscanf(buf, "%lu", &size) == 1) { - image_size = size; - return n; - } - - return -EINVAL; -} - -power_attr(image_size); - -static struct attribute * g[] = { - &disk_attr.attr, - &resume_attr.attr, - &image_size_attr.attr, - NULL, -}; - - -static struct attribute_group attr_group = { - .attrs = g, -}; - - -static int __init pm_disk_init(void) -{ - return sysfs_create_group(power_kobj, &attr_group); -} - -core_initcall(pm_disk_init); - - -static int __init resume_setup(char *str) -{ - if (noresume) - return 1; - - strncpy( resume_file, str, 255 ); - return 1; -} - -static int __init resume_offset_setup(char *str) -{ - unsigned long long offset; - - if (noresume) - return 1; - - if (sscanf(str, "%llu", &offset) == 1) - swsusp_resume_block = offset; - - return 1; -} - -static int __init noresume_setup(char *str) -{ - noresume = 1; - return 1; -} - -__setup("noresume", noresume_setup); -__setup("resume_offset=", resume_offset_setup); -__setup("resume=", resume_setup); diff --git a/kernel/power/hibernate.c b/kernel/power/hibernate.c new file mode 100644 index 000000000000..81d2e7464893 --- /dev/null +++ b/kernel/power/hibernate.c @@ -0,0 +1,955 @@ +/* + * kernel/power/hibernate.c - Hibernation (a.k.a suspend-to-disk) support. + * + * Copyright (c) 2003 Patrick Mochel + * Copyright (c) 2003 Open Source Development Lab + * Copyright (c) 2004 Pavel Machek + * Copyright (c) 2009 Rafael J. Wysocki, Novell Inc. + * + * This file is released under the GPLv2. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "power.h" + + +static int noresume = 0; +static char resume_file[256] = CONFIG_PM_STD_PARTITION; +dev_t swsusp_resume_device; +sector_t swsusp_resume_block; + +enum { + HIBERNATION_INVALID, + HIBERNATION_PLATFORM, + HIBERNATION_TEST, + HIBERNATION_TESTPROC, + HIBERNATION_SHUTDOWN, + HIBERNATION_REBOOT, + /* keep last */ + __HIBERNATION_AFTER_LAST +}; +#define HIBERNATION_MAX (__HIBERNATION_AFTER_LAST-1) +#define HIBERNATION_FIRST (HIBERNATION_INVALID + 1) + +static int hibernation_mode = HIBERNATION_SHUTDOWN; + +static struct platform_hibernation_ops *hibernation_ops; + +/** + * hibernation_set_ops - set the global hibernate operations + * @ops: the hibernation operations to use in subsequent hibernation transitions + */ + +void hibernation_set_ops(struct platform_hibernation_ops *ops) +{ + if (ops && !(ops->begin && ops->end && ops->pre_snapshot + && ops->prepare && ops->finish && ops->enter && ops->pre_restore + && ops->restore_cleanup)) { + WARN_ON(1); + return; + } + mutex_lock(&pm_mutex); + hibernation_ops = ops; + if (ops) + hibernation_mode = HIBERNATION_PLATFORM; + else if (hibernation_mode == HIBERNATION_PLATFORM) + hibernation_mode = HIBERNATION_SHUTDOWN; + + mutex_unlock(&pm_mutex); +} + +static bool entering_platform_hibernation; + +bool system_entering_hibernation(void) +{ + return entering_platform_hibernation; +} +EXPORT_SYMBOL(system_entering_hibernation); + +#ifdef CONFIG_PM_DEBUG +static void hibernation_debug_sleep(void) +{ + printk(KERN_INFO "hibernation debug: Waiting for 5 seconds.\n"); + mdelay(5000); +} + +static int hibernation_testmode(int mode) +{ + if (hibernation_mode == mode) { + hibernation_debug_sleep(); + return 1; + } + return 0; +} + +static int hibernation_test(int level) +{ + if (pm_test_level == level) { + hibernation_debug_sleep(); + return 1; + } + return 0; +} +#else /* !CONFIG_PM_DEBUG */ +static int hibernation_testmode(int mode) { return 0; } +static int hibernation_test(int level) { return 0; } +#endif /* !CONFIG_PM_DEBUG */ + +/** + * platform_begin - tell the platform driver that we're starting + * hibernation + */ + +static int platform_begin(int platform_mode) +{ + return (platform_mode && hibernation_ops) ? + hibernation_ops->begin() : 0; +} + +/** + * platform_end - tell the platform driver that we've entered the + * working state + */ + +static void platform_end(int platform_mode) +{ + if (platform_mode && hibernation_ops) + hibernation_ops->end(); +} + +/** + * platform_pre_snapshot - prepare the machine for hibernation using the + * platform driver if so configured and return an error code if it fails + */ + +static int platform_pre_snapshot(int platform_mode) +{ + return (platform_mode && hibernation_ops) ? + hibernation_ops->pre_snapshot() : 0; +} + +/** + * platform_leave - prepare the machine for switching to the normal mode + * of operation using the platform driver (called with interrupts disabled) + */ + +static void platform_leave(int platform_mode) +{ + if (platform_mode && hibernation_ops) + hibernation_ops->leave(); +} + +/** + * platform_finish - switch the machine to the normal mode of operation + * using the platform driver (must be called after platform_prepare()) + */ + +static void platform_finish(int platform_mode) +{ + if (platform_mode && hibernation_ops) + hibernation_ops->finish(); +} + +/** + * platform_pre_restore - prepare the platform for the restoration from a + * hibernation image. If the restore fails after this function has been + * called, platform_restore_cleanup() must be called. + */ + +static int platform_pre_restore(int platform_mode) +{ + return (platform_mode && hibernation_ops) ? + hibernation_ops->pre_restore() : 0; +} + +/** + * platform_restore_cleanup - switch the platform to the normal mode of + * operation after a failing restore. If platform_pre_restore() has been + * called before the failing restore, this function must be called too, + * regardless of the result of platform_pre_restore(). + */ + +static void platform_restore_cleanup(int platform_mode) +{ + if (platform_mode && hibernation_ops) + hibernation_ops->restore_cleanup(); +} + +/** + * platform_recover - recover the platform from a failure to suspend + * devices. + */ + +static void platform_recover(int platform_mode) +{ + if (platform_mode && hibernation_ops && hibernation_ops->recover) + hibernation_ops->recover(); +} + +/** + * create_image - freeze devices that need to be frozen with interrupts + * off, create the hibernation image and thaw those devices. Control + * reappears in this routine after a restore. + */ + +static int create_image(int platform_mode) +{ + int error; + + error = arch_prepare_suspend(); + if (error) + return error; + + /* At this point, dpm_suspend_start() has been called, but *not* + * dpm_suspend_noirq(). We *must* call dpm_suspend_noirq() now. + * Otherwise, drivers for some devices (e.g. interrupt controllers) + * become desynchronized with the actual state of the hardware + * at resume time, and evil weirdness ensues. + */ + error = dpm_suspend_noirq(PMSG_FREEZE); + if (error) { + printk(KERN_ERR "PM: Some devices failed to power down, " + "aborting hibernation\n"); + return error; + } + + error = platform_pre_snapshot(platform_mode); + if (error || hibernation_test(TEST_PLATFORM)) + goto Platform_finish; + + error = disable_nonboot_cpus(); + if (error || hibernation_test(TEST_CPUS) + || hibernation_testmode(HIBERNATION_TEST)) + goto Enable_cpus; + + local_irq_disable(); + + error = sysdev_suspend(PMSG_FREEZE); + if (error) { + printk(KERN_ERR "PM: Some system devices failed to power down, " + "aborting hibernation\n"); + goto Enable_irqs; + } + + if (hibernation_test(TEST_CORE)) + goto Power_up; + + in_suspend = 1; + save_processor_state(); + error = swsusp_arch_suspend(); + if (error) + printk(KERN_ERR "PM: Error %d creating hibernation image\n", + error); + /* Restore control flow magically appears here */ + restore_processor_state(); + if (!in_suspend) + platform_leave(platform_mode); + + Power_up: + sysdev_resume(); + /* NOTE: dpm_resume_noirq() is just a resume() for devices + * that suspended with irqs off ... no overall powerup. + */ + + Enable_irqs: + local_irq_enable(); + + Enable_cpus: + enable_nonboot_cpus(); + + Platform_finish: + platform_finish(platform_mode); + + dpm_resume_noirq(in_suspend ? + (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); + + return error; +} + +/** + * hibernation_snapshot - quiesce devices and create the hibernation + * snapshot image. + * @platform_mode - if set, use the platform driver, if available, to + * prepare the platform firmware for the power transition. + * + * Must be called with pm_mutex held + */ + +int hibernation_snapshot(int platform_mode) +{ + int error; + + error = platform_begin(platform_mode); + if (error) + return error; + + /* Free memory before shutting down devices. */ + error = swsusp_shrink_memory(); + if (error) + goto Close; + + suspend_console(); + error = dpm_suspend_start(PMSG_FREEZE); + if (error) + goto Recover_platform; + + if (hibernation_test(TEST_DEVICES)) + goto Recover_platform; + + error = create_image(platform_mode); + /* Control returns here after successful restore */ + + Resume_devices: + dpm_resume_end(in_suspend ? + (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); + resume_console(); + Close: + platform_end(platform_mode); + return error; + + Recover_platform: + platform_recover(platform_mode); + goto Resume_devices; +} + +/** + * resume_target_kernel - prepare devices that need to be suspended with + * interrupts off, restore the contents of highmem that have not been + * restored yet from the image and run the low level code that will restore + * the remaining contents of memory and switch to the just restored target + * kernel. + */ + +static int resume_target_kernel(bool platform_mode) +{ + int error; + + error = dpm_suspend_noirq(PMSG_QUIESCE); + if (error) { + printk(KERN_ERR "PM: Some devices failed to power down, " + "aborting resume\n"); + return error; + } + + error = platform_pre_restore(platform_mode); + if (error) + goto Cleanup; + + error = disable_nonboot_cpus(); + if (error) + goto Enable_cpus; + + local_irq_disable(); + + error = sysdev_suspend(PMSG_QUIESCE); + if (error) + goto Enable_irqs; + + /* We'll ignore saved state, but this gets preempt count (etc) right */ + save_processor_state(); + error = restore_highmem(); + if (!error) { + error = swsusp_arch_resume(); + /* + * The code below is only ever reached in case of a failure. + * Otherwise execution continues at place where + * swsusp_arch_suspend() was called + */ + BUG_ON(!error); + /* This call to restore_highmem() undos the previous one */ + restore_highmem(); + } + /* + * The only reason why swsusp_arch_resume() can fail is memory being + * very tight, so we have to free it as soon as we can to avoid + * subsequent failures + */ + swsusp_free(); + restore_processor_state(); + touch_softlockup_watchdog(); + + sysdev_resume(); + + Enable_irqs: + local_irq_enable(); + + Enable_cpus: + enable_nonboot_cpus(); + + Cleanup: + platform_restore_cleanup(platform_mode); + + dpm_resume_noirq(PMSG_RECOVER); + + return error; +} + +/** + * hibernation_restore - quiesce devices and restore the hibernation + * snapshot image. If successful, control returns in hibernation_snaphot() + * @platform_mode - if set, use the platform driver, if available, to + * prepare the platform firmware for the transition. + * + * Must be called with pm_mutex held + */ + +int hibernation_restore(int platform_mode) +{ + int error; + + pm_prepare_console(); + suspend_console(); + error = dpm_suspend_start(PMSG_QUIESCE); + if (!error) { + error = resume_target_kernel(platform_mode); + dpm_resume_end(PMSG_RECOVER); + } + resume_console(); + pm_restore_console(); + return error; +} + +/** + * hibernation_platform_enter - enter the hibernation state using the + * platform driver (if available) + */ + +int hibernation_platform_enter(void) +{ + int error; + + if (!hibernation_ops) + return -ENOSYS; + + /* + * We have cancelled the power transition by running + * hibernation_ops->finish() before saving the image, so we should let + * the firmware know that we're going to enter the sleep state after all + */ + error = hibernation_ops->begin(); + if (error) + goto Close; + + entering_platform_hibernation = true; + suspend_console(); + error = dpm_suspend_start(PMSG_HIBERNATE); + if (error) { + if (hibernation_ops->recover) + hibernation_ops->recover(); + goto Resume_devices; + } + + error = dpm_suspend_noirq(PMSG_HIBERNATE); + if (error) + goto Resume_devices; + + error = hibernation_ops->prepare(); + if (error) + goto Platofrm_finish; + + error = disable_nonboot_cpus(); + if (error) + goto Platofrm_finish; + + local_irq_disable(); + sysdev_suspend(PMSG_HIBERNATE); + hibernation_ops->enter(); + /* We should never get here */ + while (1); + + /* + * We don't need to reenable the nonboot CPUs or resume consoles, since + * the system is going to be halted anyway. + */ + Platofrm_finish: + hibernation_ops->finish(); + + dpm_suspend_noirq(PMSG_RESTORE); + + Resume_devices: + entering_platform_hibernation = false; + dpm_resume_end(PMSG_RESTORE); + resume_console(); + + Close: + hibernation_ops->end(); + + return error; +} + +/** + * power_down - Shut the machine down for hibernation. + * + * Use the platform driver, if configured so; otherwise try + * to power off or reboot. + */ + +static void power_down(void) +{ + switch (hibernation_mode) { + case HIBERNATION_TEST: + case HIBERNATION_TESTPROC: + break; + case HIBERNATION_REBOOT: + kernel_restart(NULL); + break; + case HIBERNATION_PLATFORM: + hibernation_platform_enter(); + case HIBERNATION_SHUTDOWN: + kernel_power_off(); + break; + } + kernel_halt(); + /* + * Valid image is on the disk, if we continue we risk serious data + * corruption after resume. + */ + printk(KERN_CRIT "PM: Please power down manually\n"); + while(1); +} + +static int prepare_processes(void) +{ + int error = 0; + + if (freeze_processes()) { + error = -EBUSY; + thaw_processes(); + } + return error; +} + +/** + * hibernate - The granpappy of the built-in hibernation management + */ + +int hibernate(void) +{ + int error; + + mutex_lock(&pm_mutex); + /* The snapshot device should not be opened while we're running */ + if (!atomic_add_unless(&snapshot_device_available, -1, 0)) { + error = -EBUSY; + goto Unlock; + } + + pm_prepare_console(); + error = pm_notifier_call_chain(PM_HIBERNATION_PREPARE); + if (error) + goto Exit; + + error = usermodehelper_disable(); + if (error) + goto Exit; + + /* Allocate memory management structures */ + error = create_basic_memory_bitmaps(); + if (error) + goto Exit; + + printk(KERN_INFO "PM: Syncing filesystems ... "); + sys_sync(); + printk("done.\n"); + + error = prepare_processes(); + if (error) + goto Finish; + + if (hibernation_test(TEST_FREEZER)) + goto Thaw; + + if (hibernation_testmode(HIBERNATION_TESTPROC)) + goto Thaw; + + error = hibernation_snapshot(hibernation_mode == HIBERNATION_PLATFORM); + if (in_suspend && !error) { + unsigned int flags = 0; + + if (hibernation_mode == HIBERNATION_PLATFORM) + flags |= SF_PLATFORM_MODE; + pr_debug("PM: writing image.\n"); + error = swsusp_write(flags); + swsusp_free(); + if (!error) + power_down(); + } else { + pr_debug("PM: Image restored successfully.\n"); + swsusp_free(); + } + Thaw: + thaw_processes(); + Finish: + free_basic_memory_bitmaps(); + usermodehelper_enable(); + Exit: + pm_notifier_call_chain(PM_POST_HIBERNATION); + pm_restore_console(); + atomic_inc(&snapshot_device_available); + Unlock: + mutex_unlock(&pm_mutex); + return error; +} + + +/** + * software_resume - Resume from a saved image. + * + * Called as a late_initcall (so all devices are discovered and + * initialized), we call swsusp to see if we have a saved image or not. + * If so, we quiesce devices, the restore the saved image. We will + * return above (in hibernate() ) if everything goes well. + * Otherwise, we fail gracefully and return to the normally + * scheduled program. + * + */ + +static int software_resume(void) +{ + int error; + unsigned int flags; + + /* + * If the user said "noresume".. bail out early. + */ + if (noresume) + return 0; + + /* + * name_to_dev_t() below takes a sysfs buffer mutex when sysfs + * is configured into the kernel. Since the regular hibernate + * trigger path is via sysfs which takes a buffer mutex before + * calling hibernate functions (which take pm_mutex) this can + * cause lockdep to complain about a possible ABBA deadlock + * which cannot happen since we're in the boot code here and + * sysfs can't be invoked yet. Therefore, we use a subclass + * here to avoid lockdep complaining. + */ + mutex_lock_nested(&pm_mutex, SINGLE_DEPTH_NESTING); + + if (swsusp_resume_device) + goto Check_image; + + if (!strlen(resume_file)) { + error = -ENOENT; + goto Unlock; + } + + pr_debug("PM: Checking image partition %s\n", resume_file); + + /* Check if the device is there */ + swsusp_resume_device = name_to_dev_t(resume_file); + if (!swsusp_resume_device) { + /* + * Some device discovery might still be in progress; we need + * to wait for this to finish. + */ + wait_for_device_probe(); + /* + * We can't depend on SCSI devices being available after loading + * one of their modules until scsi_complete_async_scans() is + * called and the resume device usually is a SCSI one. + */ + scsi_complete_async_scans(); + + swsusp_resume_device = name_to_dev_t(resume_file); + if (!swsusp_resume_device) { + error = -ENODEV; + goto Unlock; + } + } + + Check_image: + pr_debug("PM: Resume from partition %d:%d\n", + MAJOR(swsusp_resume_device), MINOR(swsusp_resume_device)); + + pr_debug("PM: Checking hibernation image.\n"); + error = swsusp_check(); + if (error) + goto Unlock; + + /* The snapshot device should not be opened while we're running */ + if (!atomic_add_unless(&snapshot_device_available, -1, 0)) { + error = -EBUSY; + goto Unlock; + } + + pm_prepare_console(); + error = pm_notifier_call_chain(PM_RESTORE_PREPARE); + if (error) + goto Finish; + + error = usermodehelper_disable(); + if (error) + goto Finish; + + error = create_basic_memory_bitmaps(); + if (error) + goto Finish; + + pr_debug("PM: Preparing processes for restore.\n"); + error = prepare_processes(); + if (error) { + swsusp_close(FMODE_READ); + goto Done; + } + + pr_debug("PM: Reading hibernation image.\n"); + + error = swsusp_read(&flags); + if (!error) + hibernation_restore(flags & SF_PLATFORM_MODE); + + printk(KERN_ERR "PM: Restore failed, recovering.\n"); + swsusp_free(); + thaw_processes(); + Done: + free_basic_memory_bitmaps(); + usermodehelper_enable(); + Finish: + pm_notifier_call_chain(PM_POST_RESTORE); + pm_restore_console(); + atomic_inc(&snapshot_device_available); + /* For success case, the suspend path will release the lock */ + Unlock: + mutex_unlock(&pm_mutex); + pr_debug("PM: Resume from disk failed.\n"); + return error; +} + +late_initcall(software_resume); + + +static const char * const hibernation_modes[] = { + [HIBERNATION_PLATFORM] = "platform", + [HIBERNATION_SHUTDOWN] = "shutdown", + [HIBERNATION_REBOOT] = "reboot", + [HIBERNATION_TEST] = "test", + [HIBERNATION_TESTPROC] = "testproc", +}; + +/** + * disk - Control hibernation mode + * + * Suspend-to-disk can be handled in several ways. We have a few options + * for putting the system to sleep - using the platform driver (e.g. ACPI + * or other hibernation_ops), powering off the system or rebooting the + * system (for testing) as well as the two test modes. + * + * The system can support 'platform', and that is known a priori (and + * encoded by the presence of hibernation_ops). However, the user may + * choose 'shutdown' or 'reboot' as alternatives, as well as one fo the + * test modes, 'test' or 'testproc'. + * + * show() will display what the mode is currently set to. + * store() will accept one of + * + * 'platform' + * 'shutdown' + * 'reboot' + * 'test' + * 'testproc' + * + * It will only change to 'platform' if the system + * supports it (as determined by having hibernation_ops). + */ + +static ssize_t disk_show(struct kobject *kobj, struct kobj_attribute *attr, + char *buf) +{ + int i; + char *start = buf; + + for (i = HIBERNATION_FIRST; i <= HIBERNATION_MAX; i++) { + if (!hibernation_modes[i]) + continue; + switch (i) { + case HIBERNATION_SHUTDOWN: + case HIBERNATION_REBOOT: + case HIBERNATION_TEST: + case HIBERNATION_TESTPROC: + break; + case HIBERNATION_PLATFORM: + if (hibernation_ops) + break; + /* not a valid mode, continue with loop */ + continue; + } + if (i == hibernation_mode) + buf += sprintf(buf, "[%s] ", hibernation_modes[i]); + else + buf += sprintf(buf, "%s ", hibernation_modes[i]); + } + buf += sprintf(buf, "\n"); + return buf-start; +} + + +static ssize_t disk_store(struct kobject *kobj, struct kobj_attribute *attr, + const char *buf, size_t n) +{ + int error = 0; + int i; + int len; + char *p; + int mode = HIBERNATION_INVALID; + + p = memchr(buf, '\n', n); + len = p ? p - buf : n; + + mutex_lock(&pm_mutex); + for (i = HIBERNATION_FIRST; i <= HIBERNATION_MAX; i++) { + if (len == strlen(hibernation_modes[i]) + && !strncmp(buf, hibernation_modes[i], len)) { + mode = i; + break; + } + } + if (mode != HIBERNATION_INVALID) { + switch (mode) { + case HIBERNATION_SHUTDOWN: + case HIBERNATION_REBOOT: + case HIBERNATION_TEST: + case HIBERNATION_TESTPROC: + hibernation_mode = mode; + break; + case HIBERNATION_PLATFORM: + if (hibernation_ops) + hibernation_mode = mode; + else + error = -EINVAL; + } + } else + error = -EINVAL; + + if (!error) + pr_debug("PM: Hibernation mode set to '%s'\n", + hibernation_modes[mode]); + mutex_unlock(&pm_mutex); + return error ? error : n; +} + +power_attr(disk); + +static ssize_t resume_show(struct kobject *kobj, struct kobj_attribute *attr, + char *buf) +{ + return sprintf(buf,"%d:%d\n", MAJOR(swsusp_resume_device), + MINOR(swsusp_resume_device)); +} + +static ssize_t resume_store(struct kobject *kobj, struct kobj_attribute *attr, + const char *buf, size_t n) +{ + unsigned int maj, min; + dev_t res; + int ret = -EINVAL; + + if (sscanf(buf, "%u:%u", &maj, &min) != 2) + goto out; + + res = MKDEV(maj,min); + if (maj != MAJOR(res) || min != MINOR(res)) + goto out; + + mutex_lock(&pm_mutex); + swsusp_resume_device = res; + mutex_unlock(&pm_mutex); + printk(KERN_INFO "PM: Starting manual resume from disk\n"); + noresume = 0; + software_resume(); + ret = n; + out: + return ret; +} + +power_attr(resume); + +static ssize_t image_size_show(struct kobject *kobj, struct kobj_attribute *attr, + char *buf) +{ + return sprintf(buf, "%lu\n", image_size); +} + +static ssize_t image_size_store(struct kobject *kobj, struct kobj_attribute *attr, + const char *buf, size_t n) +{ + unsigned long size; + + if (sscanf(buf, "%lu", &size) == 1) { + image_size = size; + return n; + } + + return -EINVAL; +} + +power_attr(image_size); + +static struct attribute * g[] = { + &disk_attr.attr, + &resume_attr.attr, + &image_size_attr.attr, + NULL, +}; + + +static struct attribute_group attr_group = { + .attrs = g, +}; + + +static int __init pm_disk_init(void) +{ + return sysfs_create_group(power_kobj, &attr_group); +} + +core_initcall(pm_disk_init); + + +static int __init resume_setup(char *str) +{ + if (noresume) + return 1; + + strncpy( resume_file, str, 255 ); + return 1; +} + +static int __init resume_offset_setup(char *str) +{ + unsigned long long offset; + + if (noresume) + return 1; + + if (sscanf(str, "%llu", &offset) == 1) + swsusp_resume_block = offset; + + return 1; +} + +static int __init noresume_setup(char *str) +{ + noresume = 1; + return 1; +} + +__setup("noresume", noresume_setup); +__setup("resume_offset=", resume_offset_setup); +__setup("resume=", resume_setup); diff --git a/kernel/power/power.h b/kernel/power/power.h index 2bd98d9fc19e..26d5a26f82e3 100644 --- a/kernel/power/power.h +++ b/kernel/power/power.h @@ -45,7 +45,7 @@ static inline char *check_image_kernel(struct swsusp_info *info) */ #define SPARE_PAGES ((1024 * 1024) >> PAGE_SHIFT) -/* kernel/power/disk.c */ +/* kernel/power/hibernate.c */ extern int hibernation_snapshot(int platform_mode); extern int hibernation_restore(int platform_mode); extern int hibernation_platform_enter(void); @@ -147,7 +147,7 @@ extern int swsusp_swap_in_use(void); */ #define SF_PLATFORM_MODE 1 -/* kernel/power/disk.c */ +/* kernel/power/hibernate.c */ extern int swsusp_check(void); extern void swsusp_free(void); extern int swsusp_read(unsigned int *flags_p); -- cgit v1.2.3 From fce2b111fae9151a53dabb36513b398d03337a19 Mon Sep 17 00:00:00 2001 From: Cornelia Huck Date: Wed, 10 Jun 2009 01:28:19 +0200 Subject: PM/Hibernate: Move NVS routines into a seperate file (v2). The *_nvs_* routines in swsusp.c make use of the io*map() functions, which are only provided for HAS_IOMEM, thus breaking compilation if HAS_IOMEM is not set. Fix this by moving the *_nvs_* routines into hibernate_nvs.c, which is only compiled if HAS_IOMEM is set. [rjw: Change the name of the new file to hibernate_nvs.c, add the license line to the header comment.] Signed-off-by: Cornelia Huck Acked-by: Pavel Machek Signed-off-by: Rafael J. Wysocki --- kernel/power/Kconfig | 4 ++ kernel/power/Makefile | 1 + kernel/power/hibernate_nvs.c | 135 +++++++++++++++++++++++++++++++++++++++++++ kernel/power/swsusp.c | 122 -------------------------------------- 4 files changed, 140 insertions(+), 122 deletions(-) create mode 100644 kernel/power/hibernate_nvs.c (limited to 'kernel') diff --git a/kernel/power/Kconfig b/kernel/power/Kconfig index 23bd4daeb96b..72067cbdb37f 100644 --- a/kernel/power/Kconfig +++ b/kernel/power/Kconfig @@ -116,9 +116,13 @@ config SUSPEND_FREEZER Turning OFF this setting is NOT recommended! If in doubt, say Y. +config HIBERNATION_NVS + bool + config HIBERNATION bool "Hibernation (aka 'suspend to disk')" depends on PM && SWAP && ARCH_HIBERNATION_POSSIBLE + select HIBERNATION_NVS if HAS_IOMEM ---help--- Enable the suspend to disk (STD) functionality, which is usually called "hibernation" in user interfaces. STD checkpoints the diff --git a/kernel/power/Makefile b/kernel/power/Makefile index eadb17fc8f5e..c3b81c30e5d5 100644 --- a/kernel/power/Makefile +++ b/kernel/power/Makefile @@ -9,5 +9,6 @@ obj-$(CONFIG_FREEZER) += process.o obj-$(CONFIG_SUSPEND) += suspend.o obj-$(CONFIG_PM_TEST_SUSPEND) += suspend_test.o obj-$(CONFIG_HIBERNATION) += swsusp.o hibernate.o snapshot.o swap.o user.o +obj-$(CONFIG_HIBERNATION_NVS) += hibernate_nvs.o obj-$(CONFIG_MAGIC_SYSRQ) += poweroff.o diff --git a/kernel/power/hibernate_nvs.c b/kernel/power/hibernate_nvs.c new file mode 100644 index 000000000000..39ac698ef836 --- /dev/null +++ b/kernel/power/hibernate_nvs.c @@ -0,0 +1,135 @@ +/* + * linux/kernel/power/hibernate_nvs.c - Routines for handling NVS memory + * + * Copyright (C) 2008,2009 Rafael J. Wysocki , Novell Inc. + * + * This file is released under the GPLv2. + */ + +#include +#include +#include +#include +#include + +/* + * Platforms, like ACPI, may want us to save some memory used by them during + * hibernation and to restore the contents of this memory during the subsequent + * resume. The code below implements a mechanism allowing us to do that. + */ + +struct nvs_page { + unsigned long phys_start; + unsigned int size; + void *kaddr; + void *data; + struct list_head node; +}; + +static LIST_HEAD(nvs_list); + +/** + * hibernate_nvs_register - register platform NVS memory region to save + * @start - physical address of the region + * @size - size of the region + * + * The NVS region need not be page-aligned (both ends) and we arrange + * things so that the data from page-aligned addresses in this region will + * be copied into separate RAM pages. + */ +int hibernate_nvs_register(unsigned long start, unsigned long size) +{ + struct nvs_page *entry, *next; + + while (size > 0) { + unsigned int nr_bytes; + + entry = kzalloc(sizeof(struct nvs_page), GFP_KERNEL); + if (!entry) + goto Error; + + list_add_tail(&entry->node, &nvs_list); + entry->phys_start = start; + nr_bytes = PAGE_SIZE - (start & ~PAGE_MASK); + entry->size = (size < nr_bytes) ? size : nr_bytes; + + start += entry->size; + size -= entry->size; + } + return 0; + + Error: + list_for_each_entry_safe(entry, next, &nvs_list, node) { + list_del(&entry->node); + kfree(entry); + } + return -ENOMEM; +} + +/** + * hibernate_nvs_free - free data pages allocated for saving NVS regions + */ +void hibernate_nvs_free(void) +{ + struct nvs_page *entry; + + list_for_each_entry(entry, &nvs_list, node) + if (entry->data) { + free_page((unsigned long)entry->data); + entry->data = NULL; + if (entry->kaddr) { + iounmap(entry->kaddr); + entry->kaddr = NULL; + } + } +} + +/** + * hibernate_nvs_alloc - allocate memory necessary for saving NVS regions + */ +int hibernate_nvs_alloc(void) +{ + struct nvs_page *entry; + + list_for_each_entry(entry, &nvs_list, node) { + entry->data = (void *)__get_free_page(GFP_KERNEL); + if (!entry->data) { + hibernate_nvs_free(); + return -ENOMEM; + } + } + return 0; +} + +/** + * hibernate_nvs_save - save NVS memory regions + */ +void hibernate_nvs_save(void) +{ + struct nvs_page *entry; + + printk(KERN_INFO "PM: Saving platform NVS memory\n"); + + list_for_each_entry(entry, &nvs_list, node) + if (entry->data) { + entry->kaddr = ioremap(entry->phys_start, entry->size); + memcpy(entry->data, entry->kaddr, entry->size); + } +} + +/** + * hibernate_nvs_restore - restore NVS memory regions + * + * This function is going to be called with interrupts disabled, so it + * cannot iounmap the virtual addresses used to access the NVS region. + */ +void hibernate_nvs_restore(void) +{ + struct nvs_page *entry; + + printk(KERN_INFO "PM: Restoring platform NVS memory\n"); + + list_for_each_entry(entry, &nvs_list, node) + if (entry->data) + memcpy(entry->kaddr, entry->data, entry->size); +} diff --git a/kernel/power/swsusp.c b/kernel/power/swsusp.c index 87b901cb3927..6a07f4dbf2f8 100644 --- a/kernel/power/swsusp.c +++ b/kernel/power/swsusp.c @@ -186,125 +186,3 @@ void swsusp_show_speed(struct timeval *start, struct timeval *stop, centisecs / 100, centisecs % 100, kps / 1000, (kps % 1000) / 10); } - -/* - * Platforms, like ACPI, may want us to save some memory used by them during - * hibernation and to restore the contents of this memory during the subsequent - * resume. The code below implements a mechanism allowing us to do that. - */ - -struct nvs_page { - unsigned long phys_start; - unsigned int size; - void *kaddr; - void *data; - struct list_head node; -}; - -static LIST_HEAD(nvs_list); - -/** - * hibernate_nvs_register - register platform NVS memory region to save - * @start - physical address of the region - * @size - size of the region - * - * The NVS region need not be page-aligned (both ends) and we arrange - * things so that the data from page-aligned addresses in this region will - * be copied into separate RAM pages. - */ -int hibernate_nvs_register(unsigned long start, unsigned long size) -{ - struct nvs_page *entry, *next; - - while (size > 0) { - unsigned int nr_bytes; - - entry = kzalloc(sizeof(struct nvs_page), GFP_KERNEL); - if (!entry) - goto Error; - - list_add_tail(&entry->node, &nvs_list); - entry->phys_start = start; - nr_bytes = PAGE_SIZE - (start & ~PAGE_MASK); - entry->size = (size < nr_bytes) ? size : nr_bytes; - - start += entry->size; - size -= entry->size; - } - return 0; - - Error: - list_for_each_entry_safe(entry, next, &nvs_list, node) { - list_del(&entry->node); - kfree(entry); - } - return -ENOMEM; -} - -/** - * hibernate_nvs_free - free data pages allocated for saving NVS regions - */ -void hibernate_nvs_free(void) -{ - struct nvs_page *entry; - - list_for_each_entry(entry, &nvs_list, node) - if (entry->data) { - free_page((unsigned long)entry->data); - entry->data = NULL; - if (entry->kaddr) { - iounmap(entry->kaddr); - entry->kaddr = NULL; - } - } -} - -/** - * hibernate_nvs_alloc - allocate memory necessary for saving NVS regions - */ -int hibernate_nvs_alloc(void) -{ - struct nvs_page *entry; - - list_for_each_entry(entry, &nvs_list, node) { - entry->data = (void *)__get_free_page(GFP_KERNEL); - if (!entry->data) { - hibernate_nvs_free(); - return -ENOMEM; - } - } - return 0; -} - -/** - * hibernate_nvs_save - save NVS memory regions - */ -void hibernate_nvs_save(void) -{ - struct nvs_page *entry; - - printk(KERN_INFO "PM: Saving platform NVS memory\n"); - - list_for_each_entry(entry, &nvs_list, node) - if (entry->data) { - entry->kaddr = ioremap(entry->phys_start, entry->size); - memcpy(entry->data, entry->kaddr, entry->size); - } -} - -/** - * hibernate_nvs_restore - restore NVS memory regions - * - * This function is going to be called with interrupts disabled, so it - * cannot iounmap the virtual addresses used to access the NVS region. - */ -void hibernate_nvs_restore(void) -{ - struct nvs_page *entry; - - printk(KERN_INFO "PM: Restoring platform NVS memory\n"); - - list_for_each_entry(entry, &nvs_list, node) - if (entry->data) - memcpy(entry->kaddr, entry->data, entry->size); -} -- cgit v1.2.3 From 7c692cbade8b8884f1c20500393bcc7cd6d24ef8 Mon Sep 17 00:00:00 2001 From: Vegard Nossum Date: Wed, 21 May 2008 22:53:13 +0200 Subject: tasklets: new tasklet scheduling function Rationale: kmemcheck needs to be able to schedule a tasklet without touching any dynamically allocated memory _at_ _all_ (since that would lead to a recursive page fault). This tasklet is used for writing the error reports to the kernel log. The new scheduling function avoids touching any other tasklets by inserting the new tasklist as the head of the "tasklet_hi" list instead of on the tail. Also don't wake up the softirq thread lest the scheduler access some tracked memory and we go down with a recursive page fault. In this case, we'd better just wait for the maximum time of 1/HZ for the message to appear. Signed-off-by: Vegard Nossum --- kernel/softirq.c | 11 +++++++++++ 1 file changed, 11 insertions(+) (limited to 'kernel') diff --git a/kernel/softirq.c b/kernel/softirq.c index 258885a543db..b41fb710e114 100644 --- a/kernel/softirq.c +++ b/kernel/softirq.c @@ -382,6 +382,17 @@ void __tasklet_hi_schedule(struct tasklet_struct *t) EXPORT_SYMBOL(__tasklet_hi_schedule); +void __tasklet_hi_schedule_first(struct tasklet_struct *t) +{ + BUG_ON(!irqs_disabled()); + + t->next = __get_cpu_var(tasklet_hi_vec).head; + __get_cpu_var(tasklet_hi_vec).head = t; + __raise_softirq_irqoff(HI_SOFTIRQ); +} + +EXPORT_SYMBOL(__tasklet_hi_schedule_first); + static void tasklet_action(struct softirq_action *a) { struct tasklet_struct *list; -- cgit v1.2.3 From cd6d95d8449b7c9f415f26041e9ae173d387b6bd Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 12 Jun 2009 11:29:27 +0200 Subject: clocksource: prevent selection of low resolution clocksourse also for nohz=on commit 3f68535adad (clocksource: sanity check sysfs clocksource changes) prevents selection of non high resolution capable clocksources when high resolution mode is active, but did not take into account that the same rules apply for highres=off nohz=on. Check the tick device mode instead of hrtimer_hres_active() to verify whether the system needs to be protected from a switch to jiffies or other non highres capable clock sources. Reported-by: Luming Yu Signed-off-by: Thomas Gleixner --- kernel/hrtimer.c | 4 ++-- kernel/time/clocksource.c | 18 ++++++++++-------- kernel/time/tick-oneshot.c | 17 +++++++++++++++++ 3 files changed, 29 insertions(+), 10 deletions(-) (limited to 'kernel') diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c index 1a70c18cdffe..cb8a15c19583 100644 --- a/kernel/hrtimer.c +++ b/kernel/hrtimer.c @@ -476,7 +476,7 @@ static inline int hrtimer_is_hres_enabled(void) /* * Is the high resolution mode active ? */ -int hrtimer_hres_active(void) +static inline int hrtimer_hres_active(void) { return __get_cpu_var(hrtimer_bases).hres_active; } @@ -704,7 +704,7 @@ static int hrtimer_switch_to_hres(void) #else -int hrtimer_hres_active(void) { return 0; } +static inline int hrtimer_hres_active(void) { return 0; } static inline int hrtimer_is_hres_enabled(void) { return 0; } static inline int hrtimer_switch_to_hres(void) { return 0; } static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c index 18b9f5da4ee9..592bf584d1d2 100644 --- a/kernel/time/clocksource.c +++ b/kernel/time/clocksource.c @@ -30,7 +30,6 @@ #include #include /* for spin_unlock_irq() using preempt_count() m68k */ #include -#include void timecounter_init(struct timecounter *tc, const struct cyclecounter *cc, @@ -511,13 +510,13 @@ static ssize_t sysfs_override_clocksource(struct sys_device *dev, } /* - * Check to make sure we don't switch to a non-HRT usable - * clocksource if HRT is enabled and running + * Check to make sure we don't switch to a non-highres capable + * clocksource if the tick code is in oneshot mode (highres or nohz) */ - if (hrtimer_hres_active() && + if (tick_oneshot_mode_active() && !(ovr->flags & CLOCK_SOURCE_VALID_FOR_HRES)) { printk(KERN_WARNING "%s clocksource is not HRT compatible. " - "Cannot switch while in HRT mode\n", ovr->name); + "Cannot switch while in HRT/NOHZ mode\n", ovr->name); ovr = NULL; override_name[0] = 0; } @@ -550,9 +549,12 @@ sysfs_show_available_clocksources(struct sys_device *dev, spin_lock_irq(&clocksource_lock); list_for_each_entry(src, &clocksource_list, list) { - /* Don't show non-HRES clocksource if HRES is enabled */ - if (!hrtimer_hres_active() || - (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) + /* + * Don't show non-HRES clocksource if the tick code is + * in one shot mode (highres=on or nohz=on) + */ + if (!tick_oneshot_mode_active() || + (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) count += snprintf(buf + count, max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "%s ", src->name); diff --git a/kernel/time/tick-oneshot.c b/kernel/time/tick-oneshot.c index 2e8de678e767..a96c0e2b89cf 100644 --- a/kernel/time/tick-oneshot.c +++ b/kernel/time/tick-oneshot.c @@ -128,6 +128,23 @@ int tick_switch_to_oneshot(void (*handler)(struct clock_event_device *)) return 0; } +/** + * tick_check_oneshot_mode - check whether the system is in oneshot mode + * + * returns 1 when either nohz or highres are enabled. otherwise 0. + */ +int tick_oneshot_mode_active(void) +{ + unsigned long flags; + int ret; + + local_irq_save(flags); + ret = __get_cpu_var(tick_cpu_device).mode == TICKDEV_MODE_ONESHOT; + local_irq_restore(flags); + + return ret; +} + #ifdef CONFIG_HIGH_RES_TIMERS /** * tick_init_highres - switch to high resolution mode -- cgit v1.2.3 From dfec072ecd35ba6ecad2d51dde325253ac9a2936 Mon Sep 17 00:00:00 2001 From: Vegard Nossum Date: Fri, 4 Apr 2008 00:51:41 +0200 Subject: kmemcheck: add the kmemcheck core General description: kmemcheck is a patch to the linux kernel that detects use of uninitialized memory. It does this by trapping every read and write to memory that was allocated dynamically (e.g. using kmalloc()). If a memory address is read that has not previously been written to, a message is printed to the kernel log. Thanks to Andi Kleen for the set_memory_4k() solution. Andrew Morton suggested documenting the shadow member of struct page. Signed-off-by: Vegard Nossum Signed-off-by: Pekka Enberg [export kmemcheck_mark_initialized] [build fix for setup_max_cpus] Signed-off-by: Ingo Molnar [rebased for mainline inclusion] Signed-off-by: Vegard Nossum --- kernel/sysctl.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'kernel') diff --git a/kernel/sysctl.c b/kernel/sysctl.c index ce664f98e3fb..9ef80bba3509 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -27,6 +27,7 @@ #include #include #include +#include #include #include #include @@ -959,6 +960,17 @@ static struct ctl_table kern_table[] = { .proc_handler = &proc_dointvec, }, #endif +#ifdef CONFIG_KMEMCHECK + { + .ctl_name = CTL_UNNUMBERED, + .procname = "kmemcheck", + .data = &kmemcheck_enabled, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, +#endif + /* * NOTE: do not add new entries to this table unless you have read * Documentation/sysctl/ctl_unnumbered.txt -- cgit v1.2.3 From 2dff440525f8faba8836e9f05297b76f23b4af30 Mon Sep 17 00:00:00 2001 From: Vegard Nossum Date: Sat, 31 May 2008 15:56:17 +0200 Subject: kmemcheck: add mm functions With kmemcheck enabled, the slab allocator needs to do this: 1. Tell kmemcheck to allocate the shadow memory which stores the status of each byte in the allocation proper, e.g. whether it is initialized or uninitialized. 2. Tell kmemcheck which parts of memory that should be marked uninitialized. There are actually a few more states, such as "not yet allocated" and "recently freed". If a slab cache is set up using the SLAB_NOTRACK flag, it will never return memory that can take page faults because of kmemcheck. If a slab cache is NOT set up using the SLAB_NOTRACK flag, callers can still request memory with the __GFP_NOTRACK flag. This does not prevent the page faults from occuring, however, but marks the object in question as being initialized so that no warnings will ever be produced for this object. In addition to (and in contrast to) __GFP_NOTRACK, the __GFP_NOTRACK_FALSE_POSITIVE flag indicates that the allocation should not be tracked _because_ it would produce a false positive. Their values are identical, but need not be so in the future (for example, we could now enable/disable false positives with a config option). Parts of this patch were contributed by Pekka Enberg but merged for atomicity. Signed-off-by: Vegard Nossum Signed-off-by: Pekka Enberg Signed-off-by: Ingo Molnar [rebased for mainline inclusion] Signed-off-by: Vegard Nossum --- kernel/fork.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) (limited to 'kernel') diff --git a/kernel/fork.c b/kernel/fork.c index 4430eb1376f2..be022c200da6 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -178,7 +178,7 @@ void __init fork_init(unsigned long mempages) /* create a slab on which task_structs can be allocated */ task_struct_cachep = kmem_cache_create("task_struct", sizeof(struct task_struct), - ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL); + ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL); #endif /* do the arch specific task caches init */ @@ -1470,20 +1470,20 @@ void __init proc_caches_init(void) { sighand_cachep = kmem_cache_create("sighand_cache", sizeof(struct sighand_struct), 0, - SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU, - sighand_ctor); + SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU| + SLAB_NOTRACK, sighand_ctor); signal_cachep = kmem_cache_create("signal_cache", sizeof(struct signal_struct), 0, - SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); + SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); files_cachep = kmem_cache_create("files_cache", sizeof(struct files_struct), 0, - SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); + SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); fs_cachep = kmem_cache_create("fs_cache", sizeof(struct fs_struct), 0, - SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); + SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); mm_cachep = kmem_cache_create("mm_struct", sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, - SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); + SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL); vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC); mmap_init(); } -- cgit v1.2.3 From 1744a21d57d9c60136461adb6afa85e51b3e94d9 Mon Sep 17 00:00:00 2001 From: Vegard Nossum Date: Sat, 28 Feb 2009 08:29:44 +0100 Subject: trace: annotate bitfields in struct ring_buffer_event This gets rid of a heap of false-positive warnings from the tracer code due to the use of bitfields. [rebased for mainline inclusion] Signed-off-by: Vegard Nossum --- kernel/trace/ring_buffer.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'kernel') diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 2e642b2b7253..dc4dc70171ce 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -10,6 +10,7 @@ #include #include #include +#include #include #include #include @@ -1270,6 +1271,7 @@ rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, if (tail < BUF_PAGE_SIZE) { /* Mark the rest of the page with padding */ event = __rb_page_index(tail_page, tail); + kmemcheck_annotate_bitfield(event, bitfield); rb_event_set_padding(event); } @@ -1327,6 +1329,7 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, return NULL; event = __rb_page_index(tail_page, tail); + kmemcheck_annotate_bitfield(event, bitfield); rb_update_event(event, type, length); /* The passed in type is zero for DATA */ -- cgit v1.2.3 From 7a0aeb14e18ad59394bd9bbc6e57fb345819e748 Mon Sep 17 00:00:00 2001 From: Vegard Nossum Date: Sat, 16 May 2009 11:28:33 +0200 Subject: signal: fix __send_signal() false positive kmemcheck warning This false positive is due to field padding in struct sigqueue. When this dynamically allocated structure is copied to the stack (in arch- specific delivery code), kmemcheck sees a read from the padding, which is, naturally, uninitialized. Hide the false positive using the __GFP_NOTRACK_FALSE_POSITIVE flag. Also made the rlimit override code a bit clearer by introducing a new variable. Cc: Oleg Nesterov Signed-off-by: Vegard Nossum --- kernel/signal.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'kernel') diff --git a/kernel/signal.c b/kernel/signal.c index 809a228019ad..d81f4952eebb 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -832,6 +832,7 @@ static int __send_signal(int sig, struct siginfo *info, struct task_struct *t, { struct sigpending *pending; struct sigqueue *q; + int override_rlimit; trace_sched_signal_send(sig, t); @@ -863,9 +864,13 @@ static int __send_signal(int sig, struct siginfo *info, struct task_struct *t, make sure at least one signal gets delivered and don't pass on the info struct. */ - q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN && - (is_si_special(info) || - info->si_code >= 0))); + if (sig < SIGRTMIN) + override_rlimit = (is_si_special(info) || info->si_code >= 0); + else + override_rlimit = 0; + + q = __sigqueue_alloc(t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE, + override_rlimit); if (q) { list_add_tail(&q->list, &pending->list); switch ((unsigned long) info) { -- cgit v1.2.3 From 3959214f971417f4162926ac52ad4cd042958caa Mon Sep 17 00:00:00 2001 From: Kay Sievers Date: Tue, 24 Mar 2009 15:43:30 +0100 Subject: sched: delayed cleanup of user_struct During bootup performance tracing we see repeated occurrences of /sys/kernel/uid/* events for the same uid, leading to a, in this case, rather pointless userspace processing for the same uid over and over. This is usually caused by tools which change their uid to "nobody", to run without privileges to read data supplied by untrusted users. This change delays the execution of the (already existing) scheduled work, to cleanup the uid after one second, so the allocated and announced uid can possibly be re-used by another process. This is the current behavior, where almost every invocation of a binary, which changes the uid, creates two events: $ read START < /sys/kernel/uevent_seqnum; \ for i in `seq 100`; do su --shell=/bin/true bin; done; \ read END < /sys/kernel/uevent_seqnum; \ echo $(($END - $START)) 178 With the delayed cleanup, we get only two events, and userspace finishes a bit faster too: $ read START < /sys/kernel/uevent_seqnum; \ for i in `seq 100`; do su --shell=/bin/true bin; done; \ read END < /sys/kernel/uevent_seqnum; \ echo $(($END - $START)) 1 Acked-by: Dhaval Giani Signed-off-by: Kay Sievers Signed-off-by: Greg Kroah-Hartman --- kernel/user.c | 67 ++++++++++++++++++++++++++++++++++------------------------- 1 file changed, 39 insertions(+), 28 deletions(-) (limited to 'kernel') diff --git a/kernel/user.c b/kernel/user.c index 850e0ba41c1e..2c000e7132ac 100644 --- a/kernel/user.c +++ b/kernel/user.c @@ -75,21 +75,6 @@ static void uid_hash_remove(struct user_struct *up) put_user_ns(up->user_ns); } -static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent) -{ - struct user_struct *user; - struct hlist_node *h; - - hlist_for_each_entry(user, h, hashent, uidhash_node) { - if (user->uid == uid) { - atomic_inc(&user->__count); - return user; - } - } - - return NULL; -} - #ifdef CONFIG_USER_SCHED static void sched_destroy_user(struct user_struct *up) @@ -119,6 +104,23 @@ static int sched_create_user(struct user_struct *up) { return 0; } #if defined(CONFIG_USER_SCHED) && defined(CONFIG_SYSFS) +static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent) +{ + struct user_struct *user; + struct hlist_node *h; + + hlist_for_each_entry(user, h, hashent, uidhash_node) { + if (user->uid == uid) { + /* possibly resurrect an "almost deleted" object */ + if (atomic_inc_return(&user->__count) == 1) + cancel_delayed_work(&user->work); + return user; + } + } + + return NULL; +} + static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */ static DEFINE_MUTEX(uids_mutex); @@ -283,12 +285,12 @@ int __init uids_sysfs_init(void) return uids_user_create(&root_user); } -/* work function to remove sysfs directory for a user and free up +/* delayed work function to remove sysfs directory for a user and free up * corresponding structures. */ static void cleanup_user_struct(struct work_struct *w) { - struct user_struct *up = container_of(w, struct user_struct, work); + struct user_struct *up = container_of(w, struct user_struct, work.work); unsigned long flags; int remove_user = 0; @@ -297,15 +299,12 @@ static void cleanup_user_struct(struct work_struct *w) */ uids_mutex_lock(); - local_irq_save(flags); - - if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) { + spin_lock_irqsave(&uidhash_lock, flags); + if (atomic_read(&up->__count) == 0) { uid_hash_remove(up); remove_user = 1; - spin_unlock_irqrestore(&uidhash_lock, flags); - } else { - local_irq_restore(flags); } + spin_unlock_irqrestore(&uidhash_lock, flags); if (!remove_user) goto done; @@ -331,16 +330,28 @@ done: */ static void free_user(struct user_struct *up, unsigned long flags) { - /* restore back the count */ - atomic_inc(&up->__count); spin_unlock_irqrestore(&uidhash_lock, flags); - - INIT_WORK(&up->work, cleanup_user_struct); - schedule_work(&up->work); + INIT_DELAYED_WORK(&up->work, cleanup_user_struct); + schedule_delayed_work(&up->work, msecs_to_jiffies(1000)); } #else /* CONFIG_USER_SCHED && CONFIG_SYSFS */ +static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent) +{ + struct user_struct *user; + struct hlist_node *h; + + hlist_for_each_entry(user, h, hashent, uidhash_node) { + if (user->uid == uid) { + atomic_inc(&user->__count); + return user; + } + } + + return NULL; +} + int uids_sysfs_init(void) { return 0; } static inline int uids_user_create(struct user_struct *up) { return 0; } static inline void uids_mutex_lock(void) { } -- cgit v1.2.3 From 156f5a7801195fa2ce44aeeb62d6cf8468f3332a Mon Sep 17 00:00:00 2001 From: GeunSik Lim Date: Tue, 2 Jun 2009 15:01:37 +0900 Subject: debugfs: Fix terminology inconsistency of dir name to mount debugfs filesystem. Many developers use "/debug/" or "/debugfs/" or "/sys/kernel/debug/" directory name to mount debugfs filesystem for ftrace according to ./Documentation/tracers/ftrace.txt file. And, three directory names(ex:/debug/, /debugfs/, /sys/kernel/debug/) is existed in kernel source like ftrace, DRM, Wireless, Documentation, Network[sky2]files to mount debugfs filesystem. debugfs means debug filesystem for debugging easy to use by greg kroah hartman. "/sys/kernel/debug/" name is suitable as directory name of debugfs filesystem. - debugfs related reference: http://lwn.net/Articles/334546/ Fix inconsistency of directory name to mount debugfs filesystem. * From Steven Rostedt - find_debugfs() and tracing_files() in this patch. Signed-off-by: GeunSik Lim Acked-by : Inaky Perez-Gonzalez Reviewed-by : Steven Rostedt Reviewed-by : James Smart CC: Jiri Kosina CC: David Airlie CC: Peter Osterlund CC: Ananth N Mavinakayanahalli CC: Anil S Keshavamurthy CC: Masami Hiramatsu Signed-off-by: Greg Kroah-Hartman --- kernel/trace/Kconfig | 10 +++++----- kernel/trace/trace.c | 23 +++++++++++------------ 2 files changed, 16 insertions(+), 17 deletions(-) (limited to 'kernel') diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index 4a13e5a01ce3..61071fecc82e 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -147,7 +147,7 @@ config IRQSOFF_TRACER disabled by default and can be runtime (re-)started via: - echo 0 > /debugfs/tracing/tracing_max_latency + echo 0 > /sys/kernel/debug/tracing/tracing_max_latency (Note that kernel size and overhead increases with this option enabled. This option and the preempt-off timing option can be @@ -168,7 +168,7 @@ config PREEMPT_TRACER disabled by default and can be runtime (re-)started via: - echo 0 > /debugfs/tracing/tracing_max_latency + echo 0 > /sys/kernel/debug/tracing/tracing_max_latency (Note that kernel size and overhead increases with this option enabled. This option and the irqs-off timing option can be @@ -261,7 +261,7 @@ config PROFILE_ANNOTATED_BRANCHES This tracer profiles all the the likely and unlikely macros in the kernel. It will display the results in: - /debugfs/tracing/profile_annotated_branch + /sys/kernel/debug/tracing/profile_annotated_branch Note: this will add a significant overhead, only turn this on if you need to profile the system's use of these macros. @@ -274,7 +274,7 @@ config PROFILE_ALL_BRANCHES taken in the kernel is recorded whether it hit or miss. The results will be displayed in: - /debugfs/tracing/profile_branch + /sys/kernel/debug/tracing/profile_branch This option also enables the likely/unlikely profiler. @@ -323,7 +323,7 @@ config STACK_TRACER select KALLSYMS help This special tracer records the maximum stack footprint of the - kernel and displays it in debugfs/tracing/stack_trace. + kernel and displays it in /sys/kernel/debug/tracing/stack_trace. This tracer works by hooking into every function call that the kernel executes, and keeping a maximum stack depth value and diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 8acd9b81a5d7..c1878bfb2e1e 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -344,7 +344,7 @@ static raw_spinlock_t ftrace_max_lock = /* * Copy the new maximum trace into the separate maximum-trace * structure. (this way the maximum trace is permanently saved, - * for later retrieval via /debugfs/tracing/latency_trace) + * for later retrieval via /sys/kernel/debug/tracing/latency_trace) */ static void __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) @@ -2414,21 +2414,20 @@ static const struct file_operations tracing_iter_fops = { static const char readme_msg[] = "tracing mini-HOWTO:\n\n" - "# mkdir /debug\n" - "# mount -t debugfs nodev /debug\n\n" - "# cat /debug/tracing/available_tracers\n" + "# mount -t debugfs nodev /sys/kernel/debug\n\n" + "# cat /sys/kernel/debug/tracing/available_tracers\n" "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n" - "# cat /debug/tracing/current_tracer\n" + "# cat /sys/kernel/debug/tracing/current_tracer\n" "nop\n" - "# echo sched_switch > /debug/tracing/current_tracer\n" - "# cat /debug/tracing/current_tracer\n" + "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n" + "# cat /sys/kernel/debug/tracing/current_tracer\n" "sched_switch\n" - "# cat /debug/tracing/trace_options\n" + "# cat /sys/kernel/debug/tracing/trace_options\n" "noprint-parent nosym-offset nosym-addr noverbose\n" - "# echo print-parent > /debug/tracing/trace_options\n" - "# echo 1 > /debug/tracing/tracing_enabled\n" - "# cat /debug/tracing/trace > /tmp/trace.txt\n" - "# echo 0 > /debug/tracing/tracing_enabled\n" + "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n" + "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n" + "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n" + "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n" ; static ssize_t -- cgit v1.2.3 From 5fd29d6ccbc98884569d6f3105aeca70858b3e0f Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Tue, 16 Jun 2009 10:57:02 -0700 Subject: printk: clean up handling of log-levels and newlines It used to be that we would only look at the log-level in a printk() after explicit newlines, which can cause annoying problems when the previous printk() did not end with a '\n'. In that case, the log-level marker would be just printed out in the middle of the line, and be seen as just noise rather than change the logging level. This changes things to always look at the log-level in the first bytes of the printout. If a log level marker is found, it is always used as the log-level. Additionally, if no newline existed, one is added (unless the log-level is the explicit KERN_CONT marker, to explicitly show that it's a continuation of a previous line). Acked-by: Arjan van de Ven Signed-off-by: Linus Torvalds --- kernel/printk.c | 31 ++++++++++++++++++++++--------- 1 file changed, 22 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/printk.c b/kernel/printk.c index 5052b5497c67..a87770ce73a2 100644 --- a/kernel/printk.c +++ b/kernel/printk.c @@ -687,20 +687,33 @@ asmlinkage int vprintk(const char *fmt, va_list args) sizeof(printk_buf) - printed_len, fmt, args); + p = printk_buf; + + /* Do we have a loglevel in the string? */ + if (p[0] == '<') { + unsigned char c = p[1]; + if (c && p[2] == '>') { + switch (c) { + case '0' ... '7': /* loglevel */ + current_log_level = c - '0'; + if (!new_text_line) { + emit_log_char('\n'); + new_text_line = 1; + } + /* Fallthrough - skip the loglevel */ + case 'c': /* KERN_CONT */ + p += 3; + break; + } + } + } + /* * Copy the output into log_buf. If the caller didn't provide * appropriate log level tags, we insert them here */ - for (p = printk_buf; *p; p++) { + for ( ; *p; p++) { if (new_text_line) { - /* If a token, set current_log_level and skip over */ - if (p[0] == '<' && p[1] >= '0' && p[1] <= '7' && - p[2] == '>') { - current_log_level = p[1] - '0'; - p += 3; - printed_len -= 3; - } - /* Always output the token */ emit_log_char('<'); emit_log_char(current_log_level + '0'); -- cgit v1.2.3 From e28d713704117bca0820c732210df6075b09f13b Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Tue, 16 Jun 2009 11:02:28 -0700 Subject: printk: Add KERN_DEFAULT printk log-level This adds a KERN_DEFAULT loglevel marker, for when you cannot decide which loglevel you want, and just want to keep an existing printk with the default loglevel. The difference between having KERN_DEFAULT and having no log-level marker at all is two-fold: - having the log-level marker will now force a new-line if the previous printout had not added one (perhaps because it forgot, but perhaps because it expected a continuation) - having a log-level marker is required if you are printing out a message that otherwise itself could perhaps otherwise be mistaken for a log-level. Signed-of-by: Linus Torvalds --- kernel/printk.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'kernel') diff --git a/kernel/printk.c b/kernel/printk.c index a87770ce73a2..b4d97b54c1ec 100644 --- a/kernel/printk.c +++ b/kernel/printk.c @@ -696,6 +696,8 @@ asmlinkage int vprintk(const char *fmt, va_list args) switch (c) { case '0' ... '7': /* loglevel */ current_log_level = c - '0'; + /* Fallthrough - make sure we're on a new line */ + case 'd': /* KERN_DEFAULT */ if (!new_text_line) { emit_log_char('\n'); new_text_line = 1; -- cgit v1.2.3 From b231125af7811a2f68c455d3bda95ac170ee4fa6 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Tue, 16 Jun 2009 11:07:14 -0700 Subject: printk: add KERN_DEFAULT loglevel to print_modules() Several WARN_ON() messages omit the '\n' at the end of the string, which is a simple (and understandable) error. The next line printed after that warning line is usually the current module list, and that printk does not have a log-level marker - resulting in one long mixed-up line. Adding this loglevel marker will now avoid this unreadable mess. Signed-off-by: Linus Torvalds --- kernel/module.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'kernel') diff --git a/kernel/module.c b/kernel/module.c index e4ab36ce7672..215aaab09e91 100644 --- a/kernel/module.c +++ b/kernel/module.c @@ -2899,7 +2899,7 @@ void print_modules(void) struct module *mod; char buf[8]; - printk("Modules linked in:"); + printk(KERN_DEFAULT "Modules linked in:"); /* Most callers should already have preempt disabled, but make sure */ preempt_disable(); list_for_each_entry_rcu(mod, &modules, list) -- cgit v1.2.3 From f3b39d47ebc51416fc3b690a32dfe030a2035e67 Mon Sep 17 00:00:00 2001 From: Miao Xie Date: Tue, 16 Jun 2009 15:31:46 -0700 Subject: cpusets: restructure the function cpuset_update_task_memory_state() The kernel still allocates the page caches on old node after modifying its cpuset's mems when 'memory_spread_page' was set, or it didn't spread the page cache evenly over all the nodes that faulting task is allowed to usr after memory_spread_page was set. it is caused by the old mem_allowed and flags of the task, the current kernel doesn't updates them unless some function invokes cpuset_update_task_memory_state(), it is too late sometimes.We must update the mem_allowed and the flags of the tasks in time. Slab has the same problem. The following patches fix this bug by updating tasks' mem_allowed and spread flag after its cpuset's mems or spread flag is changed. This patch: Extract a function from cpuset_update_task_memory_state(). It will be used later for update tasks' page/slab spread flags after its cpuset's flag is set Signed-off-by: Miao Xie Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Christoph Lameter Cc: Paul Menage Cc: Nick Piggin Cc: Yasunori Goto Cc: Pekka Enberg Cc: David Rientjes Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/cpuset.c | 27 +++++++++++++++++++-------- 1 file changed, 19 insertions(+), 8 deletions(-) (limited to 'kernel') diff --git a/kernel/cpuset.c b/kernel/cpuset.c index d5a7e17474ee..66b24d9b6638 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -331,6 +331,24 @@ static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); } +/* + * update task's spread flag if cpuset's page/slab spread flag is set + * + * Called with callback_mutex/cgroup_mutex held + */ +static void cpuset_update_task_spread_flag(struct cpuset *cs, + struct task_struct *tsk) +{ + if (is_spread_page(cs)) + tsk->flags |= PF_SPREAD_PAGE; + else + tsk->flags &= ~PF_SPREAD_PAGE; + if (is_spread_slab(cs)) + tsk->flags |= PF_SPREAD_SLAB; + else + tsk->flags &= ~PF_SPREAD_SLAB; +} + /** * cpuset_update_task_memory_state - update task memory placement * @@ -388,14 +406,7 @@ void cpuset_update_task_memory_state(void) cs = task_cs(tsk); /* Maybe changed when task not locked */ guarantee_online_mems(cs, &tsk->mems_allowed); tsk->cpuset_mems_generation = cs->mems_generation; - if (is_spread_page(cs)) - tsk->flags |= PF_SPREAD_PAGE; - else - tsk->flags &= ~PF_SPREAD_PAGE; - if (is_spread_slab(cs)) - tsk->flags |= PF_SPREAD_SLAB; - else - tsk->flags &= ~PF_SPREAD_SLAB; + cpuset_update_task_spread_flag(cs, tsk); task_unlock(tsk); mutex_unlock(&callback_mutex); mpol_rebind_task(tsk, &tsk->mems_allowed); -- cgit v1.2.3 From 950592f7b991f267d707d372b90f508bbe72acbc Mon Sep 17 00:00:00 2001 From: Miao Xie Date: Tue, 16 Jun 2009 15:31:47 -0700 Subject: cpusets: update tasks' page/slab spread flags in time Fix the bug that the kernel didn't spread page cache/slab object evenly over all the allowed nodes when spread flags were set by updating tasks' page/slab spread flags in time. Signed-off-by: Miao Xie Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Christoph Lameter Cc: Paul Menage Cc: Nick Piggin Cc: Yasunori Goto Cc: Pekka Enberg Cc: David Rientjes Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/cpuset.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 55 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/cpuset.c b/kernel/cpuset.c index 66b24d9b6638..af5a83d52187 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -406,7 +406,6 @@ void cpuset_update_task_memory_state(void) cs = task_cs(tsk); /* Maybe changed when task not locked */ guarantee_online_mems(cs, &tsk->mems_allowed); tsk->cpuset_mems_generation = cs->mems_generation; - cpuset_update_task_spread_flag(cs, tsk); task_unlock(tsk); mutex_unlock(&callback_mutex); mpol_rebind_task(tsk, &tsk->mems_allowed); @@ -1203,6 +1202,46 @@ static int update_relax_domain_level(struct cpuset *cs, s64 val) return 0; } +/* + * cpuset_change_flag - make a task's spread flags the same as its cpuset's + * @tsk: task to be updated + * @scan: struct cgroup_scanner containing the cgroup of the task + * + * Called by cgroup_scan_tasks() for each task in a cgroup. + * + * We don't need to re-check for the cgroup/cpuset membership, since we're + * holding cgroup_lock() at this point. + */ +static void cpuset_change_flag(struct task_struct *tsk, + struct cgroup_scanner *scan) +{ + cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk); +} + +/* + * update_tasks_flags - update the spread flags of tasks in the cpuset. + * @cs: the cpuset in which each task's spread flags needs to be changed + * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() + * + * Called with cgroup_mutex held + * + * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, + * calling callback functions for each. + * + * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 + * if @heap != NULL. + */ +static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap) +{ + struct cgroup_scanner scan; + + scan.cg = cs->css.cgroup; + scan.test_task = NULL; + scan.process_task = cpuset_change_flag; + scan.heap = heap; + cgroup_scan_tasks(&scan); +} + /* * update_flag - read a 0 or a 1 in a file and update associated flag * bit: the bit to update (see cpuset_flagbits_t) @@ -1216,8 +1255,10 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, int turning_on) { struct cpuset *trialcs; - int err; int balance_flag_changed; + int spread_flag_changed; + struct ptr_heap heap; + int err; trialcs = alloc_trial_cpuset(cs); if (!trialcs) @@ -1232,9 +1273,16 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, if (err < 0) goto out; + err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); + if (err < 0) + goto out; + balance_flag_changed = (is_sched_load_balance(cs) != is_sched_load_balance(trialcs)); + spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) + || (is_spread_page(cs) != is_spread_page(trialcs))); + mutex_lock(&callback_mutex); cs->flags = trialcs->flags; mutex_unlock(&callback_mutex); @@ -1242,6 +1290,9 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) async_rebuild_sched_domains(); + if (spread_flag_changed) + update_tasks_flags(cs, &heap); + heap_free(&heap); out: free_trial_cpuset(trialcs); return err; @@ -1392,6 +1443,8 @@ static void cpuset_attach(struct cgroup_subsys *ss, if (err) return; + cpuset_update_task_spread_flag(cs, tsk); + from = oldcs->mems_allowed; to = cs->mems_allowed; mm = get_task_mm(tsk); @@ -1453,11 +1506,9 @@ static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) break; case FILE_SPREAD_PAGE: retval = update_flag(CS_SPREAD_PAGE, cs, val); - cs->mems_generation = cpuset_mems_generation++; break; case FILE_SPREAD_SLAB: retval = update_flag(CS_SPREAD_SLAB, cs, val); - cs->mems_generation = cpuset_mems_generation++; break; default: retval = -EINVAL; -- cgit v1.2.3 From 58568d2a8215cb6f55caf2332017d7bdff954e1c Mon Sep 17 00:00:00 2001 From: Miao Xie Date: Tue, 16 Jun 2009 15:31:49 -0700 Subject: cpuset,mm: update tasks' mems_allowed in time Fix allocating page cache/slab object on the unallowed node when memory spread is set by updating tasks' mems_allowed after its cpuset's mems is changed. In order to update tasks' mems_allowed in time, we must modify the code of memory policy. Because the memory policy is applied in the process's context originally. After applying this patch, one task directly manipulates anothers mems_allowed, and we use alloc_lock in the task_struct to protect mems_allowed and memory policy of the task. But in the fast path, we didn't use lock to protect them, because adding a lock may lead to performance regression. But if we don't add a lock,the task might see no nodes when changing cpuset's mems_allowed to some non-overlapping set. In order to avoid it, we set all new allowed nodes, then clear newly disallowed ones. [lee.schermerhorn@hp.com: The rework of mpol_new() to extract the adjusting of the node mask to apply cpuset and mpol flags "context" breaks set_mempolicy() and mbind() with MPOL_PREFERRED and a NULL nodemask--i.e., explicit local allocation. Fix this by adding the check for MPOL_PREFERRED and empty node mask to mpol_new_mpolicy(). Remove the now unneeded 'nodes = NULL' from mpol_new(). Note that mpol_new_mempolicy() is always called with a non-NULL 'nodes' parameter now that it has been removed from mpol_new(). Therefore, we don't need to test nodes for NULL before testing it for 'empty'. However, just to be extra paranoid, add a VM_BUG_ON() to verify this assumption.] [lee.schermerhorn@hp.com: I don't think the function name 'mpol_new_mempolicy' is descriptive enough to differentiate it from mpol_new(). This function applies cpuset set context, usually constraining nodes to those allowed by the cpuset. However, when the 'RELATIVE_NODES flag is set, it also translates the nodes. So I settled on 'mpol_set_nodemask()', because the comment block for mpol_new() mentions that we need to call this function to "set nodes". Some additional minor line length, whitespace and typo cleanup.] Signed-off-by: Miao Xie Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Christoph Lameter Cc: Paul Menage Cc: Nick Piggin Cc: Yasunori Goto Cc: Pekka Enberg Cc: David Rientjes Signed-off-by: Lee Schermerhorn Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/cpuset.c | 184 ++++++++++++++----------------------------------------- kernel/kthread.c | 2 + 2 files changed, 48 insertions(+), 138 deletions(-) (limited to 'kernel') diff --git a/kernel/cpuset.c b/kernel/cpuset.c index af5a83d52187..7e75a41bd508 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -97,12 +97,6 @@ struct cpuset { struct cpuset *parent; /* my parent */ - /* - * Copy of global cpuset_mems_generation as of the most - * recent time this cpuset changed its mems_allowed. - */ - int mems_generation; - struct fmeter fmeter; /* memory_pressure filter */ /* partition number for rebuild_sched_domains() */ @@ -176,27 +170,6 @@ static inline int is_spread_slab(const struct cpuset *cs) return test_bit(CS_SPREAD_SLAB, &cs->flags); } -/* - * Increment this integer everytime any cpuset changes its - * mems_allowed value. Users of cpusets can track this generation - * number, and avoid having to lock and reload mems_allowed unless - * the cpuset they're using changes generation. - * - * A single, global generation is needed because cpuset_attach_task() could - * reattach a task to a different cpuset, which must not have its - * generation numbers aliased with those of that tasks previous cpuset. - * - * Generations are needed for mems_allowed because one task cannot - * modify another's memory placement. So we must enable every task, - * on every visit to __alloc_pages(), to efficiently check whether - * its current->cpuset->mems_allowed has changed, requiring an update - * of its current->mems_allowed. - * - * Since writes to cpuset_mems_generation are guarded by the cgroup lock - * there is no need to mark it atomic. - */ -static int cpuset_mems_generation; - static struct cpuset top_cpuset = { .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), }; @@ -228,8 +201,9 @@ static struct cpuset top_cpuset = { * If a task is only holding callback_mutex, then it has read-only * access to cpusets. * - * The task_struct fields mems_allowed and mems_generation may only - * be accessed in the context of that task, so require no locks. + * Now, the task_struct fields mems_allowed and mempolicy may be changed + * by other task, we use alloc_lock in the task_struct fields to protect + * them. * * The cpuset_common_file_read() handlers only hold callback_mutex across * small pieces of code, such as when reading out possibly multi-word @@ -349,69 +323,6 @@ static void cpuset_update_task_spread_flag(struct cpuset *cs, tsk->flags &= ~PF_SPREAD_SLAB; } -/** - * cpuset_update_task_memory_state - update task memory placement - * - * If the current tasks cpusets mems_allowed changed behind our - * backs, update current->mems_allowed, mems_generation and task NUMA - * mempolicy to the new value. - * - * Task mempolicy is updated by rebinding it relative to the - * current->cpuset if a task has its memory placement changed. - * Do not call this routine if in_interrupt(). - * - * Call without callback_mutex or task_lock() held. May be - * called with or without cgroup_mutex held. Thanks in part to - * 'the_top_cpuset_hack', the task's cpuset pointer will never - * be NULL. This routine also might acquire callback_mutex during - * call. - * - * Reading current->cpuset->mems_generation doesn't need task_lock - * to guard the current->cpuset derefence, because it is guarded - * from concurrent freeing of current->cpuset using RCU. - * - * The rcu_dereference() is technically probably not needed, - * as I don't actually mind if I see a new cpuset pointer but - * an old value of mems_generation. However this really only - * matters on alpha systems using cpusets heavily. If I dropped - * that rcu_dereference(), it would save them a memory barrier. - * For all other arch's, rcu_dereference is a no-op anyway, and for - * alpha systems not using cpusets, another planned optimization, - * avoiding the rcu critical section for tasks in the root cpuset - * which is statically allocated, so can't vanish, will make this - * irrelevant. Better to use RCU as intended, than to engage in - * some cute trick to save a memory barrier that is impossible to - * test, for alpha systems using cpusets heavily, which might not - * even exist. - * - * This routine is needed to update the per-task mems_allowed data, - * within the tasks context, when it is trying to allocate memory - * (in various mm/mempolicy.c routines) and notices that some other - * task has been modifying its cpuset. - */ - -void cpuset_update_task_memory_state(void) -{ - int my_cpusets_mem_gen; - struct task_struct *tsk = current; - struct cpuset *cs; - - rcu_read_lock(); - my_cpusets_mem_gen = task_cs(tsk)->mems_generation; - rcu_read_unlock(); - - if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) { - mutex_lock(&callback_mutex); - task_lock(tsk); - cs = task_cs(tsk); /* Maybe changed when task not locked */ - guarantee_online_mems(cs, &tsk->mems_allowed); - tsk->cpuset_mems_generation = cs->mems_generation; - task_unlock(tsk); - mutex_unlock(&callback_mutex); - mpol_rebind_task(tsk, &tsk->mems_allowed); - } -} - /* * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? * @@ -1017,14 +928,6 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, * other task, the task_struct mems_allowed that we are hacking * is for our current task, which must allocate new pages for that * migrating memory region. - * - * We call cpuset_update_task_memory_state() before hacking - * our tasks mems_allowed, so that we are assured of being in - * sync with our tasks cpuset, and in particular, callbacks to - * cpuset_update_task_memory_state() from nested page allocations - * won't see any mismatch of our cpuset and task mems_generation - * values, so won't overwrite our hacked tasks mems_allowed - * nodemask. */ static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, @@ -1032,22 +935,37 @@ static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, { struct task_struct *tsk = current; - cpuset_update_task_memory_state(); - - mutex_lock(&callback_mutex); tsk->mems_allowed = *to; - mutex_unlock(&callback_mutex); do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); - mutex_lock(&callback_mutex); guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); - mutex_unlock(&callback_mutex); } /* - * Rebind task's vmas to cpuset's new mems_allowed, and migrate pages to new - * nodes if memory_migrate flag is set. Called with cgroup_mutex held. + * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy + * @tsk: the task to change + * @newmems: new nodes that the task will be set + * + * In order to avoid seeing no nodes if the old and new nodes are disjoint, + * we structure updates as setting all new allowed nodes, then clearing newly + * disallowed ones. + * + * Called with task's alloc_lock held + */ +static void cpuset_change_task_nodemask(struct task_struct *tsk, + nodemask_t *newmems) +{ + nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); + mpol_rebind_task(tsk, &tsk->mems_allowed); + mpol_rebind_task(tsk, newmems); + tsk->mems_allowed = *newmems; +} + +/* + * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy + * of it to cpuset's new mems_allowed, and migrate pages to new nodes if + * memory_migrate flag is set. Called with cgroup_mutex held. */ static void cpuset_change_nodemask(struct task_struct *p, struct cgroup_scanner *scan) @@ -1056,12 +974,19 @@ static void cpuset_change_nodemask(struct task_struct *p, struct cpuset *cs; int migrate; const nodemask_t *oldmem = scan->data; + nodemask_t newmems; + + cs = cgroup_cs(scan->cg); + guarantee_online_mems(cs, &newmems); + + task_lock(p); + cpuset_change_task_nodemask(p, &newmems); + task_unlock(p); mm = get_task_mm(p); if (!mm) return; - cs = cgroup_cs(scan->cg); migrate = is_memory_migrate(cs); mpol_rebind_mm(mm, &cs->mems_allowed); @@ -1114,10 +1039,10 @@ static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem, /* * Handle user request to change the 'mems' memory placement * of a cpuset. Needs to validate the request, update the - * cpusets mems_allowed and mems_generation, and for each - * task in the cpuset, rebind any vma mempolicies and if - * the cpuset is marked 'memory_migrate', migrate the tasks - * pages to the new memory. + * cpusets mems_allowed, and for each task in the cpuset, + * update mems_allowed and rebind task's mempolicy and any vma + * mempolicies and if the cpuset is marked 'memory_migrate', + * migrate the tasks pages to the new memory. * * Call with cgroup_mutex held. May take callback_mutex during call. * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, @@ -1170,7 +1095,6 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, mutex_lock(&callback_mutex); cs->mems_allowed = trialcs->mems_allowed; - cs->mems_generation = cpuset_mems_generation++; mutex_unlock(&callback_mutex); update_tasks_nodemask(cs, &oldmem, &heap); @@ -1434,15 +1358,18 @@ static void cpuset_attach(struct cgroup_subsys *ss, if (cs == &top_cpuset) { cpumask_copy(cpus_attach, cpu_possible_mask); + to = node_possible_map; } else { - mutex_lock(&callback_mutex); guarantee_online_cpus(cs, cpus_attach); - mutex_unlock(&callback_mutex); + guarantee_online_mems(cs, &to); } err = set_cpus_allowed_ptr(tsk, cpus_attach); if (err) return; + task_lock(tsk); + cpuset_change_task_nodemask(tsk, &to); + task_unlock(tsk); cpuset_update_task_spread_flag(cs, tsk); from = oldcs->mems_allowed; @@ -1848,8 +1775,6 @@ static struct cgroup_subsys_state *cpuset_create( struct cpuset *parent; if (!cont->parent) { - /* This is early initialization for the top cgroup */ - top_cpuset.mems_generation = cpuset_mems_generation++; return &top_cpuset.css; } parent = cgroup_cs(cont->parent); @@ -1861,7 +1786,6 @@ static struct cgroup_subsys_state *cpuset_create( return ERR_PTR(-ENOMEM); } - cpuset_update_task_memory_state(); cs->flags = 0; if (is_spread_page(parent)) set_bit(CS_SPREAD_PAGE, &cs->flags); @@ -1870,7 +1794,6 @@ static struct cgroup_subsys_state *cpuset_create( set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); cpumask_clear(cs->cpus_allowed); nodes_clear(cs->mems_allowed); - cs->mems_generation = cpuset_mems_generation++; fmeter_init(&cs->fmeter); cs->relax_domain_level = -1; @@ -1889,8 +1812,6 @@ static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) { struct cpuset *cs = cgroup_cs(cont); - cpuset_update_task_memory_state(); - if (is_sched_load_balance(cs)) update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); @@ -1911,21 +1832,6 @@ struct cgroup_subsys cpuset_subsys = { .early_init = 1, }; -/* - * cpuset_init_early - just enough so that the calls to - * cpuset_update_task_memory_state() in early init code - * are harmless. - */ - -int __init cpuset_init_early(void) -{ - alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_NOWAIT); - - top_cpuset.mems_generation = cpuset_mems_generation++; - return 0; -} - - /** * cpuset_init - initialize cpusets at system boot * @@ -1936,11 +1842,13 @@ int __init cpuset_init(void) { int err = 0; + if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) + BUG(); + cpumask_setall(top_cpuset.cpus_allowed); nodes_setall(top_cpuset.mems_allowed); fmeter_init(&top_cpuset.fmeter); - top_cpuset.mems_generation = cpuset_mems_generation++; set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); top_cpuset.relax_domain_level = -1; diff --git a/kernel/kthread.c b/kernel/kthread.c index 41c88fe40500..7fa441333529 100644 --- a/kernel/kthread.c +++ b/kernel/kthread.c @@ -9,6 +9,7 @@ #include #include #include +#include #include #include #include @@ -236,6 +237,7 @@ int kthreadd(void *unused) ignore_signals(tsk); set_user_nice(tsk, KTHREAD_NICE_LEVEL); set_cpus_allowed_ptr(tsk, cpu_all_mask); + set_mems_allowed(node_possible_map); current->flags |= PF_NOFREEZE | PF_FREEZER_NOSIG; -- cgit v1.2.3 From 6484eb3e2a81807722c5f28efef94d8338b7b996 Mon Sep 17 00:00:00 2001 From: Mel Gorman Date: Tue, 16 Jun 2009 15:31:54 -0700 Subject: page allocator: do not check NUMA node ID when the caller knows the node is valid Callers of alloc_pages_node() can optionally specify -1 as a node to mean "allocate from the current node". However, a number of the callers in fast paths know for a fact their node is valid. To avoid a comparison and branch, this patch adds alloc_pages_exact_node() that only checks the nid with VM_BUG_ON(). Callers that know their node is valid are then converted. Signed-off-by: Mel Gorman Reviewed-by: Christoph Lameter Reviewed-by: KOSAKI Motohiro Reviewed-by: Pekka Enberg Acked-by: Paul Mundt [for the SLOB NUMA bits] Cc: Peter Zijlstra Cc: Nick Piggin Cc: Dave Hansen Cc: Lee Schermerhorn Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/profile.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel') diff --git a/kernel/profile.c b/kernel/profile.c index 28cf26ad2d24..69911b5745eb 100644 --- a/kernel/profile.c +++ b/kernel/profile.c @@ -365,7 +365,7 @@ static int __cpuinit profile_cpu_callback(struct notifier_block *info, node = cpu_to_node(cpu); per_cpu(cpu_profile_flip, cpu) = 0; if (!per_cpu(cpu_profile_hits, cpu)[1]) { - page = alloc_pages_node(node, + page = alloc_pages_exact_node(node, GFP_KERNEL | __GFP_ZERO, 0); if (!page) @@ -373,7 +373,7 @@ static int __cpuinit profile_cpu_callback(struct notifier_block *info, per_cpu(cpu_profile_hits, cpu)[1] = page_address(page); } if (!per_cpu(cpu_profile_hits, cpu)[0]) { - page = alloc_pages_node(node, + page = alloc_pages_exact_node(node, GFP_KERNEL | __GFP_ZERO, 0); if (!page) @@ -564,14 +564,14 @@ static int create_hash_tables(void) int node = cpu_to_node(cpu); struct page *page; - page = alloc_pages_node(node, + page = alloc_pages_exact_node(node, GFP_KERNEL | __GFP_ZERO | GFP_THISNODE, 0); if (!page) goto out_cleanup; per_cpu(cpu_profile_hits, cpu)[1] = (struct profile_hit *)page_address(page); - page = alloc_pages_node(node, + page = alloc_pages_exact_node(node, GFP_KERNEL | __GFP_ZERO | GFP_THISNODE, 0); if (!page) -- cgit v1.2.3 From 7f33d49a2ed546e01f7b1d0607661810f2421859 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Tue, 16 Jun 2009 15:32:41 -0700 Subject: mm, PM/Freezer: Disable OOM killer when tasks are frozen Currently, the following scenario appears to be possible in theory: * Tasks are frozen for hibernation or suspend. * Free pages are almost exhausted. * Certain piece of code in the suspend code path attempts to allocate some memory using GFP_KERNEL and allocation order less than or equal to PAGE_ALLOC_COSTLY_ORDER. * __alloc_pages_internal() cannot find a free page so it invokes the OOM killer. * The OOM killer attempts to kill a task, but the task is frozen, so it doesn't die immediately. * __alloc_pages_internal() jumps to 'restart', unsuccessfully tries to find a free page and invokes the OOM killer. * No progress can be made. Although it is now hard to trigger during hibernation due to the memory shrinking carried out by the hibernation code, it is theoretically possible to trigger during suspend after the memory shrinking has been removed from that code path. Moreover, since memory allocations are going to be used for the hibernation memory shrinking, it will be even more likely to happen during hibernation. To prevent it from happening, introduce the oom_killer_disabled switch that will cause __alloc_pages_internal() to fail in the situations in which the OOM killer would have been called and make the freezer set this switch after tasks have been successfully frozen. [akpm@linux-foundation.org: be nicer to the namespace] Signed-off-by: Rafael J. Wysocki Cc: Fengguang Wu Cc: David Rientjes Acked-by: Pavel Machek Cc: Mel Gorman Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/power/process.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'kernel') diff --git a/kernel/power/process.c b/kernel/power/process.c index ca634019497a..da2072d73811 100644 --- a/kernel/power/process.c +++ b/kernel/power/process.c @@ -117,9 +117,12 @@ int freeze_processes(void) if (error) goto Exit; printk("done."); + + oom_killer_disable(); Exit: BUG_ON(in_atomic()); printk("\n"); + return error; } @@ -145,6 +148,8 @@ static void thaw_tasks(bool nosig_only) void thaw_processes(void) { + oom_killer_enable(); + printk("Restarting tasks ... "); thaw_tasks(true); thaw_tasks(false); -- cgit v1.2.3 From 6837765963f1723e80ca97b1fae660f3a60d77df Mon Sep 17 00:00:00 2001 From: KOSAKI Motohiro Date: Tue, 16 Jun 2009 15:32:51 -0700 Subject: mm: remove CONFIG_UNEVICTABLE_LRU config option Currently, nobody wants to turn UNEVICTABLE_LRU off. Thus this configurability is unnecessary. Signed-off-by: KOSAKI Motohiro Cc: Johannes Weiner Cc: Andi Kleen Acked-by: Minchan Kim Cc: David Woodhouse Cc: Matt Mackall Cc: Rik van Riel Cc: Lee Schermerhorn Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/sysctl.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'kernel') diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 0e51a35a4486..2ccee08f92f1 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -1325,7 +1325,6 @@ static struct ctl_table vm_table[] = { .extra2 = &one, }, #endif -#ifdef CONFIG_UNEVICTABLE_LRU { .ctl_name = CTL_UNNUMBERED, .procname = "scan_unevictable_pages", @@ -1334,7 +1333,6 @@ static struct ctl_table vm_table[] = { .mode = 0644, .proc_handler = &scan_unevictable_handler, }, -#endif /* * NOTE: do not add new entries to this table unless you have read * Documentation/sysctl/ctl_unnumbered.txt -- cgit v1.2.3 From b33112d1cc25e658c334125d127a6ae15d5a0ad6 Mon Sep 17 00:00:00 2001 From: "Robert P. J. Day" Date: Tue, 16 Jun 2009 15:33:34 -0700 Subject: kernel/kfifo.c: replace conditional test with is_power_of_2() Signed-off-by: Robert P. J. Day Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/kfifo.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'kernel') diff --git a/kernel/kfifo.c b/kernel/kfifo.c index bc41ad0f24f8..26539e3228e5 100644 --- a/kernel/kfifo.c +++ b/kernel/kfifo.c @@ -72,9 +72,9 @@ struct kfifo *kfifo_alloc(unsigned int size, gfp_t gfp_mask, spinlock_t *lock) /* * round up to the next power of 2, since our 'let the indices - * wrap' tachnique works only in this case. + * wrap' technique works only in this case. */ - if (size & (size - 1)) { + if (!is_power_of_2(size)) { BUG_ON(size > 0x80000000); size = roundup_pow_of_two(size); } -- cgit v1.2.3 From 30639b6af85a92491b22dd14c17b14ca11da60e6 Mon Sep 17 00:00:00 2001 From: Alexey Dobriyan Date: Tue, 16 Jun 2009 15:33:40 -0700 Subject: groups: move code to kernel/groups.c Move supplementary groups implementation to kernel/groups.c . kernel/sys.c already accumulated quite a few random stuff. Do strictly copy/paste + add required headers to compile. Compile-tested on many configs and archs. Signed-off-by: Alexey Dobriyan Cc: Ingo Molnar Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/Makefile | 1 + kernel/groups.c | 288 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++ kernel/sys.c | 283 ------------------------------------------------------- 3 files changed, 289 insertions(+), 283 deletions(-) create mode 100644 kernel/groups.c (limited to 'kernel') diff --git a/kernel/Makefile b/kernel/Makefile index 90b53f6dc226..9df4501cb921 100644 --- a/kernel/Makefile +++ b/kernel/Makefile @@ -11,6 +11,7 @@ obj-y = sched.o fork.o exec_domain.o panic.o printk.o \ hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \ notifier.o ksysfs.o pm_qos_params.o sched_clock.o cred.o \ async.o +obj-y += groups.o ifdef CONFIG_FUNCTION_TRACER # Do not trace debug files and internal ftrace files diff --git a/kernel/groups.c b/kernel/groups.c new file mode 100644 index 000000000000..2b45b2ee3964 --- /dev/null +++ b/kernel/groups.c @@ -0,0 +1,288 @@ +/* + * Supplementary group IDs + */ +#include +#include +#include +#include +#include +#include + +/* init to 2 - one for init_task, one to ensure it is never freed */ +struct group_info init_groups = { .usage = ATOMIC_INIT(2) }; + +struct group_info *groups_alloc(int gidsetsize) +{ + struct group_info *group_info; + int nblocks; + int i; + + nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK; + /* Make sure we always allocate at least one indirect block pointer */ + nblocks = nblocks ? : 1; + group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER); + if (!group_info) + return NULL; + group_info->ngroups = gidsetsize; + group_info->nblocks = nblocks; + atomic_set(&group_info->usage, 1); + + if (gidsetsize <= NGROUPS_SMALL) + group_info->blocks[0] = group_info->small_block; + else { + for (i = 0; i < nblocks; i++) { + gid_t *b; + b = (void *)__get_free_page(GFP_USER); + if (!b) + goto out_undo_partial_alloc; + group_info->blocks[i] = b; + } + } + return group_info; + +out_undo_partial_alloc: + while (--i >= 0) { + free_page((unsigned long)group_info->blocks[i]); + } + kfree(group_info); + return NULL; +} + +EXPORT_SYMBOL(groups_alloc); + +void groups_free(struct group_info *group_info) +{ + if (group_info->blocks[0] != group_info->small_block) { + int i; + for (i = 0; i < group_info->nblocks; i++) + free_page((unsigned long)group_info->blocks[i]); + } + kfree(group_info); +} + +EXPORT_SYMBOL(groups_free); + +/* export the group_info to a user-space array */ +static int groups_to_user(gid_t __user *grouplist, + const struct group_info *group_info) +{ + int i; + unsigned int count = group_info->ngroups; + + for (i = 0; i < group_info->nblocks; i++) { + unsigned int cp_count = min(NGROUPS_PER_BLOCK, count); + unsigned int len = cp_count * sizeof(*grouplist); + + if (copy_to_user(grouplist, group_info->blocks[i], len)) + return -EFAULT; + + grouplist += NGROUPS_PER_BLOCK; + count -= cp_count; + } + return 0; +} + +/* fill a group_info from a user-space array - it must be allocated already */ +static int groups_from_user(struct group_info *group_info, + gid_t __user *grouplist) +{ + int i; + unsigned int count = group_info->ngroups; + + for (i = 0; i < group_info->nblocks; i++) { + unsigned int cp_count = min(NGROUPS_PER_BLOCK, count); + unsigned int len = cp_count * sizeof(*grouplist); + + if (copy_from_user(group_info->blocks[i], grouplist, len)) + return -EFAULT; + + grouplist += NGROUPS_PER_BLOCK; + count -= cp_count; + } + return 0; +} + +/* a simple Shell sort */ +static void groups_sort(struct group_info *group_info) +{ + int base, max, stride; + int gidsetsize = group_info->ngroups; + + for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1) + ; /* nothing */ + stride /= 3; + + while (stride) { + max = gidsetsize - stride; + for (base = 0; base < max; base++) { + int left = base; + int right = left + stride; + gid_t tmp = GROUP_AT(group_info, right); + + while (left >= 0 && GROUP_AT(group_info, left) > tmp) { + GROUP_AT(group_info, right) = + GROUP_AT(group_info, left); + right = left; + left -= stride; + } + GROUP_AT(group_info, right) = tmp; + } + stride /= 3; + } +} + +/* a simple bsearch */ +int groups_search(const struct group_info *group_info, gid_t grp) +{ + unsigned int left, right; + + if (!group_info) + return 0; + + left = 0; + right = group_info->ngroups; + while (left < right) { + unsigned int mid = (left+right)/2; + int cmp = grp - GROUP_AT(group_info, mid); + if (cmp > 0) + left = mid + 1; + else if (cmp < 0) + right = mid; + else + return 1; + } + return 0; +} + +/** + * set_groups - Change a group subscription in a set of credentials + * @new: The newly prepared set of credentials to alter + * @group_info: The group list to install + * + * Validate a group subscription and, if valid, insert it into a set + * of credentials. + */ +int set_groups(struct cred *new, struct group_info *group_info) +{ + int retval; + + retval = security_task_setgroups(group_info); + if (retval) + return retval; + + put_group_info(new->group_info); + groups_sort(group_info); + get_group_info(group_info); + new->group_info = group_info; + return 0; +} + +EXPORT_SYMBOL(set_groups); + +/** + * set_current_groups - Change current's group subscription + * @group_info: The group list to impose + * + * Validate a group subscription and, if valid, impose it upon current's task + * security record. + */ +int set_current_groups(struct group_info *group_info) +{ + struct cred *new; + int ret; + + new = prepare_creds(); + if (!new) + return -ENOMEM; + + ret = set_groups(new, group_info); + if (ret < 0) { + abort_creds(new); + return ret; + } + + return commit_creds(new); +} + +EXPORT_SYMBOL(set_current_groups); + +SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist) +{ + const struct cred *cred = current_cred(); + int i; + + if (gidsetsize < 0) + return -EINVAL; + + /* no need to grab task_lock here; it cannot change */ + i = cred->group_info->ngroups; + if (gidsetsize) { + if (i > gidsetsize) { + i = -EINVAL; + goto out; + } + if (groups_to_user(grouplist, cred->group_info)) { + i = -EFAULT; + goto out; + } + } +out: + return i; +} + +/* + * SMP: Our groups are copy-on-write. We can set them safely + * without another task interfering. + */ + +SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist) +{ + struct group_info *group_info; + int retval; + + if (!capable(CAP_SETGID)) + return -EPERM; + if ((unsigned)gidsetsize > NGROUPS_MAX) + return -EINVAL; + + group_info = groups_alloc(gidsetsize); + if (!group_info) + return -ENOMEM; + retval = groups_from_user(group_info, grouplist); + if (retval) { + put_group_info(group_info); + return retval; + } + + retval = set_current_groups(group_info); + put_group_info(group_info); + + return retval; +} + +/* + * Check whether we're fsgid/egid or in the supplemental group.. + */ +int in_group_p(gid_t grp) +{ + const struct cred *cred = current_cred(); + int retval = 1; + + if (grp != cred->fsgid) + retval = groups_search(cred->group_info, grp); + return retval; +} + +EXPORT_SYMBOL(in_group_p); + +int in_egroup_p(gid_t grp) +{ + const struct cred *cred = current_cred(); + int retval = 1; + + if (grp != cred->egid) + retval = groups_search(cred->group_info, grp); + return retval; +} + +EXPORT_SYMBOL(in_egroup_p); diff --git a/kernel/sys.c b/kernel/sys.c index 438d99a38c87..b3f1097c76fa 100644 --- a/kernel/sys.c +++ b/kernel/sys.c @@ -1113,289 +1113,6 @@ out: return err; } -/* - * Supplementary group IDs - */ - -/* init to 2 - one for init_task, one to ensure it is never freed */ -struct group_info init_groups = { .usage = ATOMIC_INIT(2) }; - -struct group_info *groups_alloc(int gidsetsize) -{ - struct group_info *group_info; - int nblocks; - int i; - - nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK; - /* Make sure we always allocate at least one indirect block pointer */ - nblocks = nblocks ? : 1; - group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER); - if (!group_info) - return NULL; - group_info->ngroups = gidsetsize; - group_info->nblocks = nblocks; - atomic_set(&group_info->usage, 1); - - if (gidsetsize <= NGROUPS_SMALL) - group_info->blocks[0] = group_info->small_block; - else { - for (i = 0; i < nblocks; i++) { - gid_t *b; - b = (void *)__get_free_page(GFP_USER); - if (!b) - goto out_undo_partial_alloc; - group_info->blocks[i] = b; - } - } - return group_info; - -out_undo_partial_alloc: - while (--i >= 0) { - free_page((unsigned long)group_info->blocks[i]); - } - kfree(group_info); - return NULL; -} - -EXPORT_SYMBOL(groups_alloc); - -void groups_free(struct group_info *group_info) -{ - if (group_info->blocks[0] != group_info->small_block) { - int i; - for (i = 0; i < group_info->nblocks; i++) - free_page((unsigned long)group_info->blocks[i]); - } - kfree(group_info); -} - -EXPORT_SYMBOL(groups_free); - -/* export the group_info to a user-space array */ -static int groups_to_user(gid_t __user *grouplist, - const struct group_info *group_info) -{ - int i; - unsigned int count = group_info->ngroups; - - for (i = 0; i < group_info->nblocks; i++) { - unsigned int cp_count = min(NGROUPS_PER_BLOCK, count); - unsigned int len = cp_count * sizeof(*grouplist); - - if (copy_to_user(grouplist, group_info->blocks[i], len)) - return -EFAULT; - - grouplist += NGROUPS_PER_BLOCK; - count -= cp_count; - } - return 0; -} - -/* fill a group_info from a user-space array - it must be allocated already */ -static int groups_from_user(struct group_info *group_info, - gid_t __user *grouplist) -{ - int i; - unsigned int count = group_info->ngroups; - - for (i = 0; i < group_info->nblocks; i++) { - unsigned int cp_count = min(NGROUPS_PER_BLOCK, count); - unsigned int len = cp_count * sizeof(*grouplist); - - if (copy_from_user(group_info->blocks[i], grouplist, len)) - return -EFAULT; - - grouplist += NGROUPS_PER_BLOCK; - count -= cp_count; - } - return 0; -} - -/* a simple Shell sort */ -static void groups_sort(struct group_info *group_info) -{ - int base, max, stride; - int gidsetsize = group_info->ngroups; - - for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1) - ; /* nothing */ - stride /= 3; - - while (stride) { - max = gidsetsize - stride; - for (base = 0; base < max; base++) { - int left = base; - int right = left + stride; - gid_t tmp = GROUP_AT(group_info, right); - - while (left >= 0 && GROUP_AT(group_info, left) > tmp) { - GROUP_AT(group_info, right) = - GROUP_AT(group_info, left); - right = left; - left -= stride; - } - GROUP_AT(group_info, right) = tmp; - } - stride /= 3; - } -} - -/* a simple bsearch */ -int groups_search(const struct group_info *group_info, gid_t grp) -{ - unsigned int left, right; - - if (!group_info) - return 0; - - left = 0; - right = group_info->ngroups; - while (left < right) { - unsigned int mid = (left+right)/2; - int cmp = grp - GROUP_AT(group_info, mid); - if (cmp > 0) - left = mid + 1; - else if (cmp < 0) - right = mid; - else - return 1; - } - return 0; -} - -/** - * set_groups - Change a group subscription in a set of credentials - * @new: The newly prepared set of credentials to alter - * @group_info: The group list to install - * - * Validate a group subscription and, if valid, insert it into a set - * of credentials. - */ -int set_groups(struct cred *new, struct group_info *group_info) -{ - int retval; - - retval = security_task_setgroups(group_info); - if (retval) - return retval; - - put_group_info(new->group_info); - groups_sort(group_info); - get_group_info(group_info); - new->group_info = group_info; - return 0; -} - -EXPORT_SYMBOL(set_groups); - -/** - * set_current_groups - Change current's group subscription - * @group_info: The group list to impose - * - * Validate a group subscription and, if valid, impose it upon current's task - * security record. - */ -int set_current_groups(struct group_info *group_info) -{ - struct cred *new; - int ret; - - new = prepare_creds(); - if (!new) - return -ENOMEM; - - ret = set_groups(new, group_info); - if (ret < 0) { - abort_creds(new); - return ret; - } - - return commit_creds(new); -} - -EXPORT_SYMBOL(set_current_groups); - -SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist) -{ - const struct cred *cred = current_cred(); - int i; - - if (gidsetsize < 0) - return -EINVAL; - - /* no need to grab task_lock here; it cannot change */ - i = cred->group_info->ngroups; - if (gidsetsize) { - if (i > gidsetsize) { - i = -EINVAL; - goto out; - } - if (groups_to_user(grouplist, cred->group_info)) { - i = -EFAULT; - goto out; - } - } -out: - return i; -} - -/* - * SMP: Our groups are copy-on-write. We can set them safely - * without another task interfering. - */ - -SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist) -{ - struct group_info *group_info; - int retval; - - if (!capable(CAP_SETGID)) - return -EPERM; - if ((unsigned)gidsetsize > NGROUPS_MAX) - return -EINVAL; - - group_info = groups_alloc(gidsetsize); - if (!group_info) - return -ENOMEM; - retval = groups_from_user(group_info, grouplist); - if (retval) { - put_group_info(group_info); - return retval; - } - - retval = set_current_groups(group_info); - put_group_info(group_info); - - return retval; -} - -/* - * Check whether we're fsgid/egid or in the supplemental group.. - */ -int in_group_p(gid_t grp) -{ - const struct cred *cred = current_cred(); - int retval = 1; - - if (grp != cred->fsgid) - retval = groups_search(cred->group_info, grp); - return retval; -} - -EXPORT_SYMBOL(in_group_p); - -int in_egroup_p(gid_t grp) -{ - const struct cred *cred = current_cred(); - int retval = 1; - - if (grp != cred->egid) - retval = groups_search(cred->group_info, grp); - return retval; -} - -EXPORT_SYMBOL(in_egroup_p); - DECLARE_RWSEM(uts_sem); SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name) -- cgit v1.2.3 From 009789f040b71699278e70a6664701c10065e430 Mon Sep 17 00:00:00 2001 From: Chris Peterson Date: Tue, 16 Jun 2009 15:33:43 -0700 Subject: slow-work: use round_jiffies() for thread pool's cull and OOM timers Round the slow work queue's cull and OOM timeouts to whole second boundary with round_jiffies(). The slow work queue uses a pair of timers to cull idle threads and, after OOM, to delay new thread creation. This patch also extracts the mod_timer() logic for the cull timer into a separate helper function. By rounding non-time-critical timers such as these to whole seconds, they will be batched up to fire at the same time rather than being spread out. This allows the CPU wake up less, which saves power. Signed-off-by: Chris Peterson Signed-off-by: David Howells Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/slow-work.c | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-) (limited to 'kernel') diff --git a/kernel/slow-work.c b/kernel/slow-work.c index 521ed2004d63..09d7519557d3 100644 --- a/kernel/slow-work.c +++ b/kernel/slow-work.c @@ -318,6 +318,15 @@ cant_get_ref: } EXPORT_SYMBOL(slow_work_enqueue); +/* + * Schedule a cull of the thread pool at some time in the near future + */ +static void slow_work_schedule_cull(void) +{ + mod_timer(&slow_work_cull_timer, + round_jiffies(jiffies + SLOW_WORK_CULL_TIMEOUT)); +} + /* * Worker thread culling algorithm */ @@ -335,8 +344,7 @@ static bool slow_work_cull_thread(void) list_empty(&vslow_work_queue) && atomic_read(&slow_work_thread_count) > slow_work_min_threads) { - mod_timer(&slow_work_cull_timer, - jiffies + SLOW_WORK_CULL_TIMEOUT); + slow_work_schedule_cull(); do_cull = true; } } @@ -393,8 +401,7 @@ static int slow_work_thread(void *_data) list_empty(&vslow_work_queue) && atomic_read(&slow_work_thread_count) > slow_work_min_threads) - mod_timer(&slow_work_cull_timer, - jiffies + SLOW_WORK_CULL_TIMEOUT); + slow_work_schedule_cull(); continue; } @@ -458,7 +465,7 @@ static void slow_work_new_thread_execute(struct slow_work *work) if (atomic_dec_and_test(&slow_work_thread_count)) BUG(); /* we're running on a slow work thread... */ mod_timer(&slow_work_oom_timer, - jiffies + SLOW_WORK_OOM_TIMEOUT); + round_jiffies(jiffies + SLOW_WORK_OOM_TIMEOUT)); } else { /* ratelimit the starting of new threads */ mod_timer(&slow_work_oom_timer, jiffies + 1); @@ -502,8 +509,7 @@ static int slow_work_min_threads_sysctl(struct ctl_table *table, int write, if (n < 0 && !slow_work_may_not_start_new_thread) slow_work_enqueue(&slow_work_new_thread); else if (n > 0) - mod_timer(&slow_work_cull_timer, - jiffies + SLOW_WORK_CULL_TIMEOUT); + slow_work_schedule_cull(); } mutex_unlock(&slow_work_user_lock); } @@ -529,8 +535,7 @@ static int slow_work_max_threads_sysctl(struct ctl_table *table, int write, atomic_read(&slow_work_thread_count); if (n < 0) - mod_timer(&slow_work_cull_timer, - jiffies + SLOW_WORK_CULL_TIMEOUT); + slow_work_schedule_cull(); } mutex_unlock(&slow_work_user_lock); } -- cgit v1.2.3