Re: [PATCH] sched: Skip double execution of pick_next_task_fair

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Apr 23, 2014 at 03:31:57PM -0700, Tim Chen wrote:
> The current code will call pick_next_task_fair a second time
> in the slow path if we did not pull any task in our first try.
> This is really unnecessary as we already know no task can
> be pulled and it doubles the delay for the cpu to enter idle.
> 
> We instrumented some network workloads and that saw that
> pick_next_task_fair is frequently called twice before a cpu enters idle.
> The call to pick_next_task_fair can add
> non trivial latency as it calls load_balance which runs find_busiest_group
> on an hierarchy of sched domains spanning the cpus for a large system.  For
> some 4 socket systems, we saw almost 0.25 msec spent per call
> of pick_next_task_fair before a cpu can be idled.
> 
> This patch skips pick_next_task_fair in the slow path if it
> has already been invoked.

How about something like so?

Its a little more contained.

--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2636,8 +2636,14 @@ pick_next_task(struct rq *rq, struct tas
 	if (likely(prev->sched_class == class &&
 		   rq->nr_running == rq->cfs.h_nr_running)) {
 		p = fair_sched_class.pick_next_task(rq, prev);
-		if (likely(p && p != RETRY_TASK))
-			return p;
+		if (unlikely(p == RETRY_TASK))
+			goto again;
+
+		/* assumes fair_sched_class->next == idle_sched_class */
+		if (unlikely(!p))
+			p = pick_next_task_idle(rq, prev);
+
+		return p;
 	}
 
 again: