Re: [PATCH 5/7 v4] sched: propagate asynchrous detach

From: Dietmar Eggemann
Date: Wed Oct 12 2016 - 12:26:48 EST


On 12/10/16 16:45, Vincent Guittot wrote:
> On 12 October 2016 at 17:03, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
>> On 26/09/16 13:19, Vincent Guittot wrote:

[...]

>>> @@ -6607,6 +6609,10 @@ static void update_blocked_averages(int cpu)
>>>
>>> if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, true))
>>> update_tg_load_avg(cfs_rq, 0);
>>> +
>>> + /* Propagate pending load changes to the parent */
>>> + if (cfs_rq->tg->se[cpu])
>>> + update_load_avg(cfs_rq->tg->se[cpu], 0);
>>
>> In my test (1 task (run/period: 8ms/16ms) in tg_root->tg_x->tg_y->*tg_z*
>> and oscillating between cpu1 and cpu2) the cfs_rq related signals are
>> nicely going down to 0 after the task has left the cpu but it doesn't
>> seem to be the case for the corresponding se (cfs_rq->tg->se[cpu])?
>
> strange because such use case is part of the functional tests that I
> run and it was working fine according to last test that I did
>
>>
>> It should actually work correctly because of the
>> update_tg_cfs_util/load() calls in update_load_avg(cfs_rq->tg->se[cpu],
>> 0)->propagate_entity_load_avg()
>
> Furthermore, the update of the parent cfs_rq tg_x->cfs_rq[cpu] uses
> the delta between previous and new value for the child tg_y->se[cpu].
> So it means that if tg_x->cfs_rq[cpu]->avg.load_avg goes down to 0,
> tg_y->se[cpu]->avg.load_avg has at least changed and most probably
> goes down to 0 too

Makes sense.

>
> Can't it be a misplaced trace point ?

Yes, you're right, it was a missing tracepoint. I only had se and cfs_rq
pelt tracepoints in __update_load_avg() and
attach/detach_entity_load_avg(). I've added them as well to
propagate_entity_load_avg() after the update_tg_cfs_load() call and now
it makes sense. Thanks!

[...]