lkml.org 
[lkml]   [2018]   [May]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 3/3] sched/fair: schedutil: explicit update only when required
Hi Joel,
sorry for the late reply, this thread is a bit confusing because we
keep discussing while there was already a v2 posted on list.

However, here are few comments below...

[...]

> > > > > @@ -5456,10 +5443,12 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
> > > > > update_cfs_group(se);
> > > > > }
> > > > >
> > > > > + /* The task is no more visible from the root cfs_rq */
> > > > > if (!se)
> > > > > sub_nr_running(rq, 1);
> > > > >
> > > > > util_est_dequeue(&rq->cfs, p, task_sleep);
> > > > > + cpufreq_update_util(rq, 0);
> > > >
> > > > One question about this change. In enqueue, throttle and unthrottle - you are
> > > > conditionally calling cpufreq_update_util incase the task was
> > > > visible/not-visible in the hierarchy.
> > > >
> > > > But in dequeue you're unconditionally calling it. Seems a bit inconsistent.
> > > > Is this because of util_est or something? Could you add a comment here
> > > > explaining why this is so?
> > >
> > > The big question I have is incase se != NULL, then its still visible at the
> > > root RQ level.
> >
> > My understanding it that you get !se at dequeue time when we are
> > dequeuing a task from a throttled RQ. Isn't it?
>
> I don't think so? !se means the RQ is not throttled.

Yes, I agree, I "just" forgot a "not" in the sentence above... my bad!

However, we are on the same page here.

> > Thus, this means you are dequeuing a throttled task, I guess for
> > example because of a migration.
> > However, the point is that a task dequeue from a throttled RQ _is
> > already_ not visible from the root RQ, because of the sub_nr_running()
> > done by throttle_cfs_rq().
>
> Yes that's what I was wondering, so my point was if its already not visible,
> then why call schedutil. I felt call schedutil only if its visible like you
> were doing for the other paths.

Agree, as discussed in Vincent in v2, we should likely move these
schedutil updates at attach/detach time. This is when exectly we know
that the utilization has changed for a CPU.

... and that's what I'll propose in the upcoming v3 for this patch.

[...]

> I agree with your assessments below and about not calling cpufreq
> when CPU is about to idle.

Cool ;)

--
#include <best/regards.h>

Patrick Bellasi

\
 
 \ /
  Last update: 2018-05-24 15:42    [W:0.315 / U:0.368 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site