sched/walt: Accounting for number of irqs pending on each core

Schedules on a core whose irq count is less than a threshold.
Improves I/O performance of EAS.

Change-Id: I08ff7dd0d22502a0106fc636b1af2e6fe9e758b5

7 files changed, 107 insertions, 1 deletions

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 4c981dfc34ee..67abbbd3965b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7754,6 +7754,11 @@ void __init sched_init(void)
 		rq->idle_stamp = 0;
 		rq->avg_idle = 2*sysctl_sched_migration_cost;
 		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
+#ifdef CONFIG_SCHED_WALT
+		rq->cur_irqload = 0;
+		rq->avg_irqload = 0;
+		rq->irqload_ts = 0;
+#endif
 
 		INIT_LIST_HEAD(&rq->cfs_tasks);
 
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index f74ea89e77a8..3f232c8b2bdd 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -5,6 +5,7 @@
 #include <linux/static_key.h>
 #include <linux/context_tracking.h>
 #include "sched.h"
+#include "walt.h"
 
 
 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
@@ -49,6 +50,10 @@ void irqtime_account_irq(struct task_struct *curr)
 	unsigned long flags;
 	s64 delta;
 	int cpu;
+#ifdef CONFIG_SCHED_WALT
+	u64 wallclock;
+	bool account = true;
+#endif
 
 	if (!sched_clock_irqtime)
 		return;
@@ -56,6 +61,9 @@ void irqtime_account_irq(struct task_struct *curr)
 	local_irq_save(flags);
 
 	cpu = smp_processor_id();
+#ifdef CONFIG_SCHED_WALT
+	wallclock = sched_clock_cpu(cpu);
+#endif
 	delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
 	__this_cpu_add(irq_start_time, delta);
 
@@ -70,8 +78,16 @@ void irqtime_account_irq(struct task_struct *curr)
 		__this_cpu_add(cpu_hardirq_time, delta);
 	else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
 		__this_cpu_add(cpu_softirq_time, delta);
+#ifdef CONFIG_SCHED_WALT
+	else
+		account = false;
+#endif
 
 	irq_time_write_end();
+#ifdef CONFIG_SCHED_WALT
+	if (account)
+		walt_account_irqtime(cpu, curr, delta, wallclock);
+#endif
 	local_irq_restore(flags);
 }
 EXPORT_SYMBOL_GPL(irqtime_account_irq);
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 15b8a8f34bd9..8560a5530035 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -61,6 +61,8 @@ unsigned int sysctl_sched_cstate_aware = 1;
 #ifdef CONFIG_SCHED_WALT
 unsigned int sysctl_sched_use_walt_cpu_util = 1;
 unsigned int sysctl_sched_use_walt_task_util = 1;
+__read_mostly unsigned int sysctl_sched_walt_cpu_high_irqload =
+    (10 * NSEC_PER_MSEC);
 #endif
 /*
  * The initial- and re-scaling of tunables is configurable
@@ -4274,7 +4276,6 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 	schedtune_enqueue_task(p, cpu_of(rq));
 
 #endif /* CONFIG_SMP */
-
 	hrtick_update(rq);
 }
 
@@ -5648,6 +5649,10 @@ static inline int find_best_target(struct task_struct *p, bool boosted)
 		if (new_util > capacity_orig_of(i))
 			continue;
 
+#ifdef CONFIG_SCHED_WALT
+		if (walt_cpu_high_irqload(i))
+			continue;
+#endif
 		/*
 		 * For boosted tasks we favor idle cpus unconditionally to
 		 * improve latency.
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index f48fb371913a..51c632bc94b6 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -685,6 +685,9 @@ struct rq {
 	u64 prev_runnable_sum;
 	u64 nt_curr_runnable_sum;
 	u64 nt_prev_runnable_sum;
+	u64 cur_irqload;
+	u64 avg_irqload;
+	u64 irqload_ts;
 #endif /* CONFIG_SCHED_WALT */
 
 
diff --git a/kernel/sched/walt.c b/kernel/sched/walt.c
index 1dff3d2e2358..b9ae8d5c4393 100644
--- a/kernel/sched/walt.c
+++ b/kernel/sched/walt.c
@@ -221,6 +221,71 @@ static int cpu_is_waiting_on_io(struct rq *rq)
 	return atomic_read(&rq->nr_iowait);
 }
 
+void walt_account_irqtime(int cpu, struct task_struct *curr,
+				 u64 delta, u64 wallclock)
+{
+	struct rq *rq = cpu_rq(cpu);
+	unsigned long flags, nr_windows;
+	u64 cur_jiffies_ts;
+
+	raw_spin_lock_irqsave(&rq->lock, flags);
+
+	/*
+	 * cputime (wallclock) uses sched_clock so use the same here for
+	 * consistency.
+	 */
+	delta += sched_clock() - wallclock;
+	cur_jiffies_ts = get_jiffies_64();
+
+	if (is_idle_task(curr))
+		walt_update_task_ravg(curr, rq, IRQ_UPDATE, walt_ktime_clock(),
+				 delta);
+
+	nr_windows = cur_jiffies_ts - rq->irqload_ts;
+
+	if (nr_windows) {
+		if (nr_windows < 10) {
+			/* Decay CPU's irqload by 3/4 for each window. */
+			rq->avg_irqload *= (3 * nr_windows);
+			rq->avg_irqload = div64_u64(rq->avg_irqload,
+						    4 * nr_windows);
+		} else {
+			rq->avg_irqload = 0;
+		}
+		rq->avg_irqload += rq->cur_irqload;
+		rq->cur_irqload = 0;
+	}
+
+	rq->cur_irqload += delta;
+	rq->irqload_ts = cur_jiffies_ts;
+	raw_spin_unlock_irqrestore(&rq->lock, flags);
+}
+
+
+#define WALT_HIGH_IRQ_TIMEOUT 3
+
+u64 walt_irqload(int cpu) {
+	struct rq *rq = cpu_rq(cpu);
+	s64 delta;
+	delta = get_jiffies_64() - rq->irqload_ts;
+
+        /*
+	 * Current context can be preempted by irq and rq->irqload_ts can be
+	 * updated by irq context so that delta can be negative.
+	 * But this is okay and we can safely return as this means there
+	 * was recent irq occurrence.
+	 */
+
+        if (delta < WALT_HIGH_IRQ_TIMEOUT)
+		return rq->avg_irqload;
+        else
+		return 0;
+}
+
+int walt_cpu_high_irqload(int cpu) {
+	return walt_irqload(cpu) >= sysctl_sched_walt_cpu_high_irqload;
+}
+
 static int account_busy_for_cpu_time(struct rq *rq, struct task_struct *p,
 				     u64 irqtime, int event)
 {
diff --git a/kernel/sched/walt.h b/kernel/sched/walt.h
index cabc193a683d..e181c87a928d 100644
--- a/kernel/sched/walt.h
+++ b/kernel/sched/walt.h
@@ -31,6 +31,11 @@ void walt_set_window_start(struct rq *rq);
 void walt_migrate_sync_cpu(int cpu);
 void walt_init_cpu_efficiency(void);
 u64 walt_ktime_clock(void);
+void walt_account_irqtime(int cpu, struct task_struct *curr, u64 delta,
+                                  u64 wallclock);
+
+u64 walt_irqload(int cpu);
+int walt_cpu_high_irqload(int cpu);
 
 #else /* CONFIG_SCHED_WALT */
 
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index e2d9953822be..d964422eb601 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -333,6 +333,13 @@ static struct ctl_table kern_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec,
 	},
+	{
+		.procname	= "sched_walt_cpu_high_irqload",
+		.data		= &sysctl_sched_walt_cpu_high_irqload,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
 #endif
 	{
 		.procname	= "sched_sync_hint_enable",