Re: [PATCH] sched/fair: prefer prev cpu in asymmetric wakeup path

[Vincent Guittot]

On Fri, 23 Oct 2020 at 19:14, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
>
> On 22/10/2020 17:33, Vincent Guittot wrote:
> > On Thu, 22 Oct 2020 at 16:53, Valentin Schneider
> > <valentin.schneider@xxxxxxx> wrote:
> >>
> >>
> >> Hi Vincent,
> >>
> >> On 22/10/20 14:43, Vincent Guittot wrote:
>
> [...]
>
> >>>  static int
> >>> -select_idle_capacity(struct task_struct *p, struct sched_domain *sd, int target)
> >>> +select_idle_capacity(struct task_struct *p, struct sched_domain *sd, int prev, int target)
> >>>  {
> >>>       unsigned long best_cap = 0;
> >>>       int cpu, best_cpu = -1;
> >>> @@ -6178,9 +6178,22 @@ select_idle_capacity(struct task_struct *p, struct sched_domain *sd, int target)
> >>>
> >>>       sync_entity_load_avg(&p->se);
> >>>
> >>> +     if ((available_idle_cpu(target) || sched_idle_cpu(target)) &&
> >>> +         task_fits_capacity(p, capacity_of(target)))
> >>> +             return target;
> >>> +
> >>
> >> I think we still need to check for CPU affinity here.
> >
> > yes good point
>
> We don't check CPU affinity on target and prev in the symmetric case.

Yes that's what i have noticed while reworking the patch to merge asym
and symmetric
>
> I always thought that since we:
>
> (1) check 'want_affine = ... && cpumask_test_cpu(cpu, p->cpus_ptr);' in
>     select_task_rq_fair() and
>
> (2) we have the select_fallback_rq() in select_task_rq() for prev
>
> that this would be sufficient?
>
> [...]