Re: [PATCH] sched/uclamp: Let each sched_class handle uclamp

[Hongyan Xia]

On 06/03/2025 13:48, Dietmar Eggemann wrote:
> On 06/03/2025 11:53, Hongyan Xia wrote:
> > On 05/03/2025 18:22, Dietmar Eggemann wrote:
> > > On 27/02/2025 14:54, Hongyan Xia wrote:
> > >
> > > [...]
> > >
> > > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > > index 857808da23d8..7e5a653811ad 100644
> > > > --- a/kernel/sched/fair.c
> > > > +++ b/kernel/sched/fair.c
> > > > @@ -6941,8 +6941,10 @@ enqueue_task_fair(struct rq *rq, struct
> > > > task_struct *p, int flags)
> > > >         * Let's add the task's estimated utilization to the cfs_rq's
> > > >         * estimated utilization, before we update schedutil.
> > > >         */
> > > > -    if (!(p->se.sched_delayed && (task_on_rq_migrating(p) || (flags
> > > > & ENQUEUE_RESTORE))))
> > > > +    if (!(p->se.sched_delayed && (task_on_rq_migrating(p) || (flags
> > > > & ENQUEUE_RESTORE)))) {
> > > > +        uclamp_rq_inc(rq, p);
> > > >            util_est_enqueue(&rq->cfs, p);
> > > > +    }
> > >
> > > So you want to have p uclamp-enqueued so that its uclamp_min value
> > > counts for the cpufreq_update_util()/cfs_rq_util_change() calls later in
> > > enqueue_task_fair?
> > >
> > >     if (p->in_iowait)
> > >       cpufreq_update_util(rq, SCHED_CPUFREQ_IOWAIT);
> > >
> > >     enqueue_entity() -> update_load_avg() -> cfs_rq_util_change() ->
> > >     cpufreq_update_util()
> > >
> > > But if you do this before requeue_delayed_entity() (1) you will not
> > > uclamp-enqueue p which got his ->sched_delayed just cleared in (1)?
> >
> > Sorry I'm not sure I'm following. The only condition of the
> > uclamp_rq_inc() here should be
> >
> >      if (!(p->se.sched_delayed && (task_on_rq_migrating(p) || (flags &
> > ENQUEUE_RESTORE))))
> >
> > Could you elaborate why it doesn't get enqueued?
>
> Let's say 'p->se.sched_delayed = 1' and we are in
>
> enqueue_task()
>
>    enqueue_task_fair()
>
>      if (!(p->se.sched_delayed && ...)
>
>        uclamp_rq_inc(rq, p);
>
> So p wouldn't be included here.
>
> But then p would be requeued in
>
>        requeue_delayed_entity(se)
>
> since you removed the uclamp_rq_inc() from enqueue_task() (after the
> p->sched_class->enqueue_task) p will not be considered for uclamp.

I doubt this would be a concern as there are other conditions after 
checking p->se.sched_delayed. You would only skip the uclamp inc if you 
are both sched_delayed and meet the second part of the if.

Another reason is that, I think whatever we do should be consistent with 
what we did for util_est. If util_est also affects cpufreq then I doubt 
uclamp should be enqueue/dequeued differently, as it would be difficult 
to argue why sometimes util_est affects cpufreq while uclamp doesn't and 
why sometimes uclamp does and util_est doesn't.