Re: [tip:sched/core] [sched/fair] 089d84203a: pts.schbench.32.usec,_99.9th_latency_percentile 52.4% regression

[Vincent Guittot]

On Thu, 18 Dec 2025 at 11:20, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Thu, Dec 18, 2025 at 03:41:55PM +0530, Shrikanth Hegde wrote:
> > On 12/18/25 2:07 PM, Peter Zijlstra wrote:
> > > On Thu, Dec 18, 2025 at 12:59:53PM +0800, kernel test robot wrote:
> > > >
> > > >
> > > > Hello,
> > > >
> > > > kernel test robot noticed a 52.4% regression of pts.schbench.32.usec,_99.9th_latency_percentile on:
> > > >
> > > >
> > > > commit: 089d84203ad42bc8fd6dbf41683e162ac6e848cd ("sched/fair: Fold the sched_avg update")
> > > > https://git.kernel.org/cgit/linux/kernel/git/tip/tip.git sched/core
> > >
> > > Well, that obviously wasn't the intention. Let me pull that patch :/
> >
> > Is it possible because it missed scaling by se_weight(se) ??
>
> >  static inline void
> >  enqueue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
> >  {
> > -       cfs_rq->avg.load_avg += se->avg.load_avg;
> > -       cfs_rq->avg.load_sum += se_weight(se) * se->avg.load_sum;
> > +       __update_sa(&cfs_rq->avg, load, se->avg.load_avg, se->avg.load_sum);
> >  }
>
> Ah, indeed, something like so then? Can the robot (Oliver/Philip)
> verify?

yes, rq tracks the weighted sum whereas se tracks the unweighted sum

>
> (I was going to shelf it and look at it after the holidays, but if this
> is it, we can get it fixed before I dissapear).
>
> ---
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 76f5e4b78b30..7377f9117501 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -3775,13 +3775,15 @@ account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  static inline void
>  enqueue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  {
> -       __update_sa(&cfs_rq->avg, load, se->avg.load_avg, se->avg.load_sum);
> +       __update_sa(&cfs_rq->avg, load, se->avg.load_avg,
> +                   se_weight(se) * se->avg.load_sum);
>  }
>
>  static inline void
>  dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  {
> -       __update_sa(&cfs_rq->avg, load, -se->avg.load_avg, -se->avg.load_sum);
> +       __update_sa(&cfs_rq->avg, load, -se->avg.load_avg,
> +                   se_weight(se) * -se->avg.load_sum);
>  }
>
>  static void place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags);