diff options
Diffstat (limited to 'kernel/sched/fair.c')
| -rw-r--r-- | kernel/sched/fair.c | 40 |
1 files changed, 38 insertions, 2 deletions
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 2ea3a4337dde..f962ab1eb046 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -2402,7 +2402,8 @@ void task_numa_work(struct callback_head *work) return; - down_read(&mm->mmap_sem); + if (!down_read_trylock(&mm->mmap_sem)) + return; vma = find_vma(mm, start); if (!vma) { reset_ptenuma_scan(p); @@ -2641,6 +2642,7 @@ u32 sched_get_wake_up_idle(struct task_struct *p) return !!enabled; } +EXPORT_SYMBOL(sched_get_wake_up_idle); int sched_set_wake_up_idle(struct task_struct *p, int wake_up_idle) { @@ -2653,6 +2655,7 @@ int sched_set_wake_up_idle(struct task_struct *p, int wake_up_idle) return 0; } +EXPORT_SYMBOL(sched_set_wake_up_idle); static const u32 runnable_avg_yN_inv[] = { 0xffffffff, 0xfa83b2da, 0xf5257d14, 0xefe4b99a, 0xeac0c6e6, 0xe5b906e6, @@ -6596,10 +6599,22 @@ static int sched_group_energy(struct energy_env *eenv) { struct cpumask visit_cpus; u64 total_energy = 0; + int cpu_count; WARN_ON(!eenv->sg_top->sge); cpumask_copy(&visit_cpus, sched_group_cpus(eenv->sg_top)); + /* If a cpu is hotplugged in while we are in this function, + * it does not appear in the existing visit_cpus mask + * which came from the sched_group pointer of the + * sched_domain pointed at by sd_ea for either the prev + * or next cpu and was dereferenced in __energy_diff. + * Since we will dereference sd_scs later as we iterate + * through the CPUs we expect to visit, new CPUs can + * be present which are not in the visit_cpus mask. + * Guard this with cpu_count. + */ + cpu_count = cpumask_weight(&visit_cpus); while (!cpumask_empty(&visit_cpus)) { struct sched_group *sg_shared_cap = NULL; @@ -6609,6 +6624,8 @@ static int sched_group_energy(struct energy_env *eenv) /* * Is the group utilization affected by cpus outside this * sched_group? + * This sd may have groups with cpus which were not present + * when we took visit_cpus. */ sd = rcu_dereference(per_cpu(sd_scs, cpu)); @@ -6658,8 +6675,24 @@ static int sched_group_energy(struct energy_env *eenv) total_energy += sg_busy_energy + sg_idle_energy; - if (!sd->child) + if (!sd->child) { + /* + * cpu_count here is the number of + * cpus we expect to visit in this + * calculation. If we race against + * hotplug, we can have extra cpus + * added to the groups we are + * iterating which do not appear in + * the visit_cpus mask. In that case + * we are not able to calculate energy + * without restarting so we will bail + * out and use prev_cpu this time. + */ + if (!cpu_count) + return -EINVAL; cpumask_xor(&visit_cpus, &visit_cpus, sched_group_cpus(sg)); + cpu_count--; + } if (cpumask_equal(sched_group_cpus(sg), sched_group_cpus(eenv->sg_top))) goto next_cpu; @@ -6671,6 +6704,9 @@ static int sched_group_energy(struct energy_env *eenv) * If we raced with hotplug and got an sd NULL-pointer; * returning a wrong energy estimation is better than * entering an infinite loop. + * Specifically: If a cpu is unplugged after we took + * the visit_cpus mask, it no longer has an sd_scs + * pointer, so when we dereference it, we get NULL. */ if (cpumask_test_cpu(cpu, &visit_cpus)) return -EINVAL; |