Re: [PATCH v2] sched/freq: move call to cpufreq_update_util

[Vincent Guittot]

Le Wednesday 13 Nov 2019 à 18:50:35 (+0100), Vincent Guittot a écrit :
> Le Wednesday 13 Nov 2019 à 15:09:47 (+0100), Dietmar Eggemann a écrit :
> > On 13.11.19 14:30, Vincent Guittot wrote:
> > > On Wed, 13 Nov 2019 at 11:50, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
> > >>
> > >> On 12.11.19 16:05, Vincent Guittot wrote:
> > >>> Le Tuesday 12 Nov 2019 à 15:48:13 (+0100), Vincent Guittot a écrit :
> > 
> > [...]
> > 
> > >>>> @@ -7493,9 +7495,9 @@ static void update_blocked_averages(int cpu)
> > >>>>       * that RT, DL and IRQ signals have been updated before updating CFS.
> > >>>>       */
> > >>>>      curr_class = rq->curr->sched_class;
> > >>>> -    update_rt_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &rt_sched_class);
> > >>>> -    update_dl_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &dl_sched_class);
> > >>>> -    update_irq_load_avg(rq, 0);
> > >>>> +    decayed |= update_rt_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &rt_sched_class);
> > >>>> +    decayed |= update_dl_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &dl_sched_class);
> > >>>> +    decayed |= update_irq_load_avg(rq, 0);
> > >>
> > >> Why not 'decayed  = update_cfs_rq_load_avg()' like in the
> > >> !CONFIG_FAIR_GROUP_SCHED case?
> > > 
> > > Because it is handled by the update_load_avg() in
> > > for_each_leaf_cfs_rq_safe() loop
> > > 
> > > This means that we can have 2 calls to cpufreq_update_util in
> > > update_blocked_average() but at least the values will be up to date in
> > > both calls unlike previously.
> > > 
> > > I'm going to prepare an additional patch to remove this useless call.
> > > I have also seen some possible further optimization that i need to
> > > study a bit more before preparing a patch
> > 
> > I see. The update_load_avg() call for the taskgroup skeleton se
> > (cfs_rq->tg->se[cpu]). But what happens to the cpu which only has the
> > root cfs_rq i the list? It doesn't have a skeleton se.
> 
> you're right. I have to add the following to make sure it will be called
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 2eb1aa8..9fc077c 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -7604,9 +7604,13 @@ static void update_blocked_averages(int cpu)
>                         cpu,
>                         cfs_rq == &rq->cfs ? 0 : (long)cfs_rq->tg );
>  
> -               if (update_cfs_rq_load_avg(cfs_rq_clock_pelt(cfs_rq), cfs_rq))
> +               if (update_cfs_rq_load_avg(cfs_rq_clock_pelt(cfs_rq), cfs_rq)) {
>                         update_tg_load_avg(cfs_rq, 0);
>  
> +                       if (cfs_rq == &rq->cfs)
> +                               decayed = 1;
> +               }
> +
>                 trace_sched_load_contrib_blocked(cpu,
>                         &cfs_rq->avg,
>                         cfs_rq == &rq->cfs ? 0 : (long)cfs_rq->tg );

the proper fix without some debug trace events :-)

@@ -7567,9 +7569,13 @@ static void update_blocked_averages(int cpu)
 	for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos) {
 		struct sched_entity *se;
 
-		if (update_cfs_rq_load_avg(cfs_rq_clock_pelt(cfs_rq), cfs_rq))
+		if (update_cfs_rq_load_avg(cfs_rq_clock_pelt(cfs_rq), cfs_rq)) {
 			update_tg_load_avg(cfs_rq, 0);
 
+			if (cfs_rq == &rq->cfs)
+				decayed = 1;
+		}
+
 		/* Propagate pending load changes to the parent, if any: */
 		se = cfs_rq->tg->se[cpu];
 		if (se && !skip_blocked_update(se))



> 
>