[PATCH] sched: fix how RT task is pulled

[PATCH] sched: fix how RT task is pulled

[Hillf Danton]

When pulling RT task for a given runqueue, if it is already overloaded
with RT tasks, the pull operation could be avoided at the moment.

btw, it looks like a typo?

Signed-off-by: Hillf Danton <dhillf@gmail.com>
---
 kernel/sched_rt.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index 19ecb31..14c764b 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -1440,7 +1440,7 @@ static int pull_rt_task(struct rq *this_rq)
 	struct task_struct *p;
 	struct rq *src_rq;

-	if (likely(!rt_overloaded(this_rq)))
+	if (unlikely(rt_overloaded(this_rq)))
 		return 0;

 	for_each_cpu(cpu, this_rq->rd->rto_mask) {

Re: [PATCH] sched: fix how RT task is pulled

[Yong Zhang]

On Sun, May 15, 2011 at 10:50 AM, Hillf Danton <dhillf@gmail.com> wrote:
> When pulling RT task for a given runqueue, if it is already overloaded
> with RT tasks, the pull operation could be avoided at the moment.
>
> btw, it looks like a typo?

No.

Below is how rt_overloaded() is realized:
static inline int rt_overloaded(struct rq *rq)
{
        return atomic_read(&rq->rd->rto_count);
}

You can notice it's about the overload of the very root_domain.

Thanks,
Yong

>
> Signed-off-by: Hillf Danton <dhillf@gmail.com>
> ---
>  kernel/sched_rt.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
> index 19ecb31..14c764b 100644
> --- a/kernel/sched_rt.c
> +++ b/kernel/sched_rt.c
> @@ -1440,7 +1440,7 @@ static int pull_rt_task(struct rq *this_rq)
>        struct task_struct *p;
>        struct rq *src_rq;
>
> -       if (likely(!rt_overloaded(this_rq)))
> +       if (unlikely(rt_overloaded(this_rq)))
>                return 0;
>
>        for_each_cpu(cpu, this_rq->rd->rto_mask) {
>



-- 
Only stand for myself

Re: [PATCH] sched: fix how RT task is pulled

[Hillf Danton]

On Tue, May 17, 2011 at 10:35 AM, Yong Zhang <yong.zhang0@gmail.com> wrote:
> On Sun, May 15, 2011 at 10:50 AM, Hillf Danton <dhillf@gmail.com> wrote:
>> When pulling RT task for a given runqueue, if it is already overloaded
>> with RT tasks, the pull operation could be avoided at the moment.
>>
>> btw, it looks like a typo?
>
> No.
>
> Below is how rt_overloaded() is realized:
> static inline int rt_overloaded(struct rq *rq)
> {
>        return atomic_read(&rq->rd->rto_count);
> }
>
> You can notice it's about the overload of the very root_domain.
>
Well, why is it going no head if not overloaded?

thanks
          Hillf

Re: [PATCH] sched: fix how RT task is pulled

[Mike Galbraith]

On Tue, 2011-05-17 at 22:47 +0800, Hillf Danton wrote:
> On Tue, May 17, 2011 at 10:35 AM, Yong Zhang <yong.zhang0@gmail.com> wrote:
> > On Sun, May 15, 2011 at 10:50 AM, Hillf Danton <dhillf@gmail.com> wrote:
> >> When pulling RT task for a given runqueue, if it is already overloaded
> >> with RT tasks, the pull operation could be avoided at the moment.
> >>
> >> btw, it looks like a typo?
> >
> > No.
> >
> > Below is how rt_overloaded() is realized:
> > static inline int rt_overloaded(struct rq *rq)
> > {
> >        return atomic_read(&rq->rd->rto_count);
> > }
> >
> > You can notice it's about the overload of the very root_domain.
> >
> Well, why is it going no head if not overloaded?

To avoid examining masks (maybe huge) routinely.  Challenge is to
improve oddball case (overload) without injuring the common case.

	-Mike

Re: [PATCH] sched: fix how RT task is pulled

[Steven Rostedt]

On Tue, May 17, 2011 at 08:27:09PM +0200, Mike Galbraith wrote:
> > >
> > Well, why is it going no head if not overloaded?
> 
> To avoid examining masks (maybe huge) routinely.  Challenge is to
> improve oddball case (overload) without injuring the common case.

Correct.

-- Steve