Re: [PATCH] sched: fix lose fair sleeper bonus in switch_to_fair()

[Wanpeng Li]

On 9/8/15 3:57 PM, Byungchul Park wrote:
> On Tue, Sep 08, 2015 at 03:30:08PM +0800, Wanpeng Li wrote:
> > On 9/8/15 3:11 PM, Byungchul Park wrote:
> > > On Tue, Sep 08, 2015 at 02:42:52PM +0800, Wanpeng Li wrote:
> > > > On 9/8/15 2:32 PM, Byungchul Park wrote:
> > > > > On Tue, Sep 08, 2015 at 03:14:26PM +0900, Byungchul Park wrote:
> > > > > > On Tue, Sep 08, 2015 at 01:38:08PM +0800, Wanpeng Li wrote:
> > > > > > > On 9/8/15 1:28 PM, Byungchul Park wrote:
> > > > > > > > On Tue, Sep 08, 2015 at 11:46:01AM +0800, Wanpeng Li wrote:
> > > > > > > > > On 9/7/15 10:02 PM, Peter Zijlstra wrote:
> > > > > > > > > > Please always Cc at least the person who wrote the lines you modify.
> > > > > > > > > >
> > > > > > > > > > On Mon, Sep 07, 2015 at 05:45:20PM +0800, Wanpeng Li wrote:
> > > > > > > > > > > The sleeper task will be normalized when moved from fair_sched_class, in
> > > > > > > > > > > order that vruntime will be adjusted either the task is running or sleeping
> > > > > > > > > > > when moved back. The nomalization in switch_to_fair for sleep task will
> > > > > > > > > > > result in lose fair sleeper bonus in place_entity() once the vruntime -
> > > > > > > > > > > cfs_rq->min_vruntime is big when moved from fair_sched_class.
> > > > > > > > > > >
> > > > > > > > > > > This patch fix it by adjusting vruntime just during migrating as original
> > > > > > > > > > > codes since the vruntime of the task has usually NOT been normalized in
> > > > > > > > > > > this case.
> > > > > > > > > > Sorry, I cannot follow that at all. Maybe its me being sleep deprived,
> > > > > > > > > > but could you try that again?
> > > > > > > > > When changing away from the fair class while sleeping, relative
> > > > > > > > > vruntime is calculated to handle the case sleep when moved from
> > > > > > > > > fair_sched_class and running when moved to fair_sched_class. The
> > > > > > > > i don't think relative vruntime is calculated to handle the special case
> > > > > > > > you mentioned. i think the calculation is necessary for all cases detaching
> > > > > > > Please refer why the relative vruntime caculation is introduced to
> > > > > > > switched_from_fair(): https://lkml.org/lkml/2011/1/17/129
> > > > > > hello,
> > > > > >
> > > > > > it is just a bug caused by not calculating a relative vruntime when
> > > > > > detached a task from cfs_rq, which is necessary though.
> > > > > >
> > > > > > > > a task from a cfs_rq.
> > > > > > > >
> > > > > > > > > absolute vruntime will be calculated in enqueue_entity() either the
> > > > > > > > > task is running or sleeping when moved back. The fair sleeper bonus
> > > > > > > > i think absolute vruntime is calculated in enqueue_entuty() only when the
> > > > > > > > task is on rq. therefore in the case that the task is not on rq,
> > > > > > > > switched_to_fair() has to calculate the absolute vruntime instread.
> > > > > > > Absolute vruntime is caculated in place_entity() which is called by
> > > > > > > enqueue_entity() for DEQUEUE_SLEEP task.
> > > > > > as you may know, place_entity() is not for calculating an absolute
> > > > > > vruntime though.. anyway the important thing here is that, when a
> > > > > > sleeping task is moved back to fair class, enqueue_entity() for
> > > > > > DEQUEUE_SLEEP task won't be called.
> > > > > you may talk about calling enqueue_entity() when the task is woken up,
> > > > > not just when it is moved back. right?
> > > > Exactly.
> > > >
> > > > > even if yes, i think place_entity() should not be used directly for
> > > > > calculating an absolute vruntime. it should be called after non/normalizing
> > > > > operations.
> > > > The se->vruntime += cfs_rq->min_vruntime(in your switched_to_fair())
> > > > which means that se->vruntime is bigger than cfs_rq->min_vruntime,
> > > it is not always true since se->vruntime can have a negative value (even
> > > though it is a unsigned type.. i think it can be another problem) by
> > > se->vruntime -= cfs_rq->min_vruntime in detach_task_cfs_rq().
> > Yeah, it can be negative.
> >
> > > > however, fair sleeper bonus is min_vuntime - sysctl_sched_latency/2,
> > > > which means that max_vruntime() will select the absolute vruntime
> > > > which is caculated in your switched_to_fair() as the se->vruntime,
> > > since se->vruntime can have a negative value, max_vruntime() may select
> > > the fair sleeper bonused value.
> > >
> > > by the way, this logic is unchanged by my patch. which part of my patch
> > > changed this kind of logic?
> > However, if se->vruntime -= cfs_rq->min_vruntime is positive, the
> > behavior is different after your patch. e.g. se->vruntime(the
> > relative vruntime in switched_to_fair()) < min_vruntime -
> > sysctl_sched_latency/2
> >
> > before your patch:
> >
> > se->vruntime = min_vruntime - sysctl_sched_latency/2 (place_entity())
> my patch is based on ff277d4 commit at tip/sched/core.
>
> there's no change between before and after.
>
> check it please.
>
> and this logic seems to be no problem to me. :(

Your logic will lose fair sleeper bonus in the scenario which I pointed out.

> > after your patch:
> >
> > se->vruntime += cfs->min_vruntime  (switched_to_fair())
> > se->vruntime = se->vruntime  (place_entity())
> >
> >
> > Regards,
> > Wanpeng Li
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/