Re: [PATCH 4/4] sched,fair: Fix PELT integrity for new tasks

From: Peter Zijlstra
Date: Mon Jun 20 2016 - 05:52:38 EST


On Mon, Jun 20, 2016 at 11:23:39AM +0200, Vincent Guittot wrote:

> > @@ -738,7 +739,20 @@ void post_init_entity_util_avg(struct sc
> > sa->util_sum = sa->util_avg * LOAD_AVG_MAX;
> > }
> >
> > - update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, false);
> > + if (entity_is_task(se)) {
>
> Why only task ?

Only tasks can be of another class, the cgroup se's must be fair.

> > + struct task_struct *p = task_of(se);
> > + if (p->sched_class != &fair_sched_class) {
> > + /*
> > + * For !fair tasks do attach_entity_load_avg()
> > + * followed by detach_entity_load_avg() as per
> > + * switched_from_fair().
> > + */
> > + se->avg.last_update_time = now;
> > + return;
> > + }
> > + }
> > +
> > + update_cfs_rq_load_avg(now, cfs_rq, false);
> > attach_entity_load_avg(cfs_rq, se);
>
> Don't we have to do a complete attach with attach_task_cfs_rq instead
> of just the load_avg ? to set also depth ?

init_tg_cfs_entity() sets depth for cgroup entities,
attach_task_cfs_rq() reset depth on switched_to().

So that should be fine.

> ---
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -723,6 +723,7 @@ void post_init_entity_util_avg(struct sched_entity *se)
> struct cfs_rq *cfs_rq = cfs_rq_of(se);
> struct sched_avg *sa = &se->avg;
> long cap = (long)(SCHED_CAPACITY_SCALE - cfs_rq->avg.util_avg) / 2;
> + u64 now = cfs_rq_clock_task(cfs_rq);
>
> if (cap > 0) {
> if (cfs_rq->avg.util_avg != 0) {
> @@ -737,8 +738,18 @@ void post_init_entity_util_avg(struct sched_entity *se)
> sa->util_sum = sa->util_avg * LOAD_AVG_MAX;
> }
>
> - update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, false);
> - attach_entity_load_avg(cfs_rq, se);

There is no p here...

> + if (p->sched_class == &fair_sched_class) {
> + /* fair entity must be attached to cfs_rq */
> + attach_task_cfs_rq(se);
> + } else {
> + /*
> + * For !fair tasks do attach_entity_load_avg()
> + * followed by detach_entity_load_avg() as per
> + * switched_from_fair().
> + */
> + se->avg.last_update_time = now;
> + }
> +
> }

And you're now not attaching new cgroup thingies.