[RFC/PATCH] sched: Honor sync wake up in select_idle_sibling

[RFC/PATCH] sched: Honor sync wake up in select_idle_sibling

[Namhyung Kim]

From: Namhyung Kim <namhyung.kim@lge.com>

When sync wakeup happens and there's the waker task running alone,
select the target cpu as if it's already idle.

Cc: Mike Galbraith <efault@gmx.de>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 kernel/sched/fair.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index d0cc03b3e70b..64b68bc82b52 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2632,7 +2632,7 @@ find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
 /*
  * Try and locate an idle CPU in the sched_domain.
  */
-static int select_idle_sibling(struct task_struct *p, int target)
+static int select_idle_sibling(struct task_struct *p, int target, int sync)
 {
 	int cpu = smp_processor_id();
 	int prev_cpu = task_cpu(p);
@@ -2646,6 +2646,13 @@ static int select_idle_sibling(struct task_struct *p, int target)
 		return cpu;
 
 	/*
+	 * If the task is going to be woken-up on this cpu and there's only
+	 * the sync-waker running, then it is the right target.
+	 */
+	if (target == cpu && sync && this_rq()->nr_running == 1)
+		return cpu;
+
+	/*
 	 * If the task is going to be woken-up on the cpu where it previously
 	 * ran and if it is currently idle, then it the right target.
 	 */
@@ -2748,7 +2755,7 @@ select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
 		if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
 			prev_cpu = cpu;
 
-		new_cpu = select_idle_sibling(p, prev_cpu);
+		new_cpu = select_idle_sibling(p, prev_cpu, sync);
 		goto unlock;
 	}
 
-- 
1.7.11.2

Re: [RFC/PATCH] sched: Honor sync wake up in select_idle_sibling

[Mike Galbraith]

On Fri, 2012-08-10 at 16:25 +0900, Namhyung Kim wrote: 
> From: Namhyung Kim <namhyung.kim@lge.com>
> 
> When sync wakeup happens and there's the waker task running alone,
> select the target cpu as if it's already idle.

Ouch.  That defeats the purpose of select_idle_sibling().  Just because
we're doing a sync wakeup does not mean there's no convertible overlap,
nor that the waker is really really going to take a nap immediately.  A
lot of the places that benefit up to and including hugely from looking
for an idle shared cache to wake to do sync wakeups.

> Cc: Mike Galbraith <efault@gmx.de>
> Cc: Suresh Siddha <suresh.b.siddha@intel.com>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> ---
>  kernel/sched/fair.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index d0cc03b3e70b..64b68bc82b52 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -2632,7 +2632,7 @@ find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
>  /*
>   * Try and locate an idle CPU in the sched_domain.
>   */
> -static int select_idle_sibling(struct task_struct *p, int target)
> +static int select_idle_sibling(struct task_struct *p, int target, int sync)
>  {
>  	int cpu = smp_processor_id();
>  	int prev_cpu = task_cpu(p);
> @@ -2646,6 +2646,13 @@ static int select_idle_sibling(struct task_struct *p, int target)
>  		return cpu;
>  
>  	/*
> +	 * If the task is going to be woken-up on this cpu and there's only
> +	 * the sync-waker running, then it is the right target.
> +	 */
> +	if (target == cpu && sync && this_rq()->nr_running == 1)
> +		return cpu;
> +
> +	/*
>  	 * If the task is going to be woken-up on the cpu where it previously
>  	 * ran and if it is currently idle, then it the right target.
>  	 */
> @@ -2748,7 +2755,7 @@ select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
>  		if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
>  			prev_cpu = cpu;
>  
> -		new_cpu = select_idle_sibling(p, prev_cpu);
> +		new_cpu = select_idle_sibling(p, prev_cpu, sync);
>  		goto unlock;
>  	}
>  

Re: [RFC/PATCH] sched: Honor sync wake up in select_idle_sibling

[Namhyung Kim]

Hi, Mike

On Fri, 10 Aug 2012 10:14:44 +0200, Mike Galbraith wrote:
> On Fri, 2012-08-10 at 16:25 +0900, Namhyung Kim wrote: 
>> From: Namhyung Kim <namhyung.kim@lge.com>
>> 
>> When sync wakeup happens and there's the waker task running alone,
>> select the target cpu as if it's already idle.
>
> Ouch.  That defeats the purpose of select_idle_sibling().  Just because
> we're doing a sync wakeup does not mean there's no convertible overlap,
> nor that the waker is really really going to take a nap immediately.  A
> lot of the places that benefit up to and including hugely from looking
> for an idle shared cache to wake to do sync wakeups.

Thanks for the explanation! I misunderstood the function ;)

Thanks,
Namhyung