Re: [PATCH 6/6 v3] sched/eevdf: Speedup short slice task scheduling

[Vincent Guittot]

On Fri, 26 Jun 2026 at 05:57, K Prateek Nayak <kprateek.nayak@xxxxxxx> wrote:
>
> Hello Peter,
>
> On 6/26/2026 3:58 AM, Peter Zijlstra wrote:
> > +static u64 ineligible_vruntime(struct cfs_rq *cfs_rq)
> > +{
> > +     struct sched_entity *curr = cfs_rq->curr;
> > +     long weight = cfs_rq->sum_weight;
> > +     s64 delta = 0;
> > +
> > +     if (curr && !curr->on_rq)
> > +             curr = NULL;
> > +
> > +     /*
> > +      * This is called from set_next_task_fair(.first=true) /
> > +      * set_protect_slice() so curr had better be set and on_rq.
> > +      */
> > +     WARN_ON_ONCE(!curr);
>
> set_protect_slice() is indeed called from set_next_entity(.first=true)
> but it is done after __dequeue_entity() and before "cfs_rq->curr" is
> set (both sched/flat and sched/core have the same pattern).

Yes, I confirm that set_protect_slice() is called before setting curr
in tip/sched/core but I think that Peter added flat hierarchy patches
in his tree

>
> You should be hitting this splat very easily unless you have moved
> set_protect_slice() after the setting of cfs_rq->curr in your tree.
>
> > +
> > +     if (weight) {
> > +             s64 runtime = cfs_rq->sum_w_vruntime;
> > +
> > +             /*
> > +              * Do not add @curr to obtain the effective '- w_j' terms.
> > +              */
> > +
> > +             /* sign flips effective floor / ceiling */
> > +             if (runtime < 0)
> > +                     runtime -= (weight - 1);
> > +
> > +             delta = div64_long(runtime, weight);
> > +     }
> > +
> > +     return cfs_rq->zero_vruntime + delta + 1;
> > +}
> > +
> >  static inline u64 cfs_rq_max_slice(struct cfs_rq *cfs_rq);
> >
> >  /*
> > @@ -1058,8 +1119,14 @@ static inline void set_protect_slice(str
> >               slice = cfs_rq_min_slice(cfs_rq);
> >
> >       slice = min(slice, se->slice);
> > -     if (slice != se->slice)
> > -             vprot = min_vruntime(vprot, se->vruntime + calc_delta_fair(slice, se));
> > +
> > +     /* If there are shorter slices than se's one */
> > +     if (slice != se->slice) {
>
> I guess that condition was protecting a simple boot test but If I run:
>
>   for i in 10000000 20000000 30000000 40000000 50000000;
>   do
>       sudo taskset -c 0 chrt -v -o -T $i 0 loop&     # Simple while(1) loop
>   done
>
>
> I see the splat:
>
>   ------------[ cut here ]------------
>   !curr
>   WARNING: kernel/sched/fair.c:858 at set_next_task_fair+0x1d1/0x870, CPU#0: kworker/0:5/3750
>   ...
>
> > +             if (sched_feat(PREEMPT_SHORT))
> > +                     vprot = min_vruntime(vprot, ineligible_vruntime(cfs_rq));
> > +             else
> > +                     vprot = min_vruntime(vprot, se->vruntime + calc_delta_fair(slice, se));
> > +     }
> >
> >       se->vprot = vprot;
> >  }
>
> --
> Thanks and Regards,
> Prateek
>