Re: [PATCH v2] sched/fair: Fix enqueue_task_fair warning some more

[Vincent Guittot]

On Mon, 11 May 2020 at 10:40, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
>
> On 08/05/2020 19:02, Tao Zhou wrote:
> > On Fri, May 08, 2020 at 05:27:44PM +0200, Vincent Guittot wrote:
> >> On Fri, 8 May 2020 at 17:12, Tao Zhou <zohooouoto@xxxxxxxxxxx> wrote:
> >>>
> >>> Hi Phil,
> >>>
> >>> On Thu, May 07, 2020 at 04:36:12PM -0400, Phil Auld wrote:
> >>>> sched/fair: Fix enqueue_task_fair warning some more
>
> [...]
>
> >>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> >>>> index 02f323b85b6d..c6d57c334d51 100644
> >>>> --- a/kernel/sched/fair.c
> >>>> +++ b/kernel/sched/fair.c
> >>>> @@ -5479,6 +5479,13 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
> >>>>               /* end evaluation on encountering a throttled cfs_rq */
> >>>>               if (cfs_rq_throttled(cfs_rq))
> >>>>                       goto enqueue_throttle;
> >>>> +
> >>>> +               /*
> >>>> +                * One parent has been throttled and cfs_rq removed from the
> >>>> +                * list. Add it back to not break the leaf list.
> >>>> +                */
> >>>> +               if (throttled_hierarchy(cfs_rq))
> >>>> +                       list_add_leaf_cfs_rq(cfs_rq);
> >>>>       }
> >>>
> >>> I was confused by why the throttled cfs rq can be on list.
> >>> It is possible when enqueue a task and thanks to the 'threads'.
> >>> But I think the above comment does not truely put the right
> >>> intention, right ?
> >>> If throttled parent is onlist, the child cfs_rq is ignored
> >>> to be added to the leaf cfs_rq list me think.
> >>>
> >>> unthrottle_cfs_rq() follows the same logic if i am not wrong.
> >>> Is it necessary to add the above to it ?
> >>
> >> When a cfs_rq is throttled, its sched group is dequeued and all child
> >> cfs_rq are removed from  leaf_cfs_rq list. But the sched group of the
> >> child cfs_rq stay enqueued in the throttled cfs_rq so child sched
> >> group->on_rq might be still set.
> >
> > If there is a throttle of throttle, and unthrottle the child throttled
> > cfs_rq(ugly):
> >                                ...
> >                                 |
> >                       cfs_rq throttled (parent A)
> >                                 |
> >                                 |
> >                       cfs_rq in hierarchy (B)
> >                                 |
> >                                 |
> >                       cfs_rq throttled (C)
> >                                 |
> >                                ...
> >
> > Then unthrottle the child throttled cfs_rq C, now the A is on the
> > leaf_cfs_rq list. sched_group entity of C is enqueued to B, and
> > sched_group entity of B is on_rq and is ignored by enqueue but in
> > the throttled hierarchy and not add to leaf_cfs_rq list.
> > The above may be absolutely wrong that I miss something.
> >
> > Another thing :
> > In enqueue_task_fair():
> >
> >       for_each_sched_entity(se) {
> >               cfs_rq = cfs_rq_of(se);
> >
> >               if (list_add_leaf_cfs_rq(cfs_rq))
> >                       break;
> >       }
> >
> > In unthrottle_cfs_rq():
> >
> >       for_each_sched_entity(se) {
> >               cfs_rq = cfs_rq_of(se);
> >
> >               list_add_leaf_cfs_rq(cfs_rq);
> >       }
> >
> > The difference between them is that if condition, add if
> > condition to unthrottle_cfs_rq() may be an optimization and
> > keep the same.
> >
>
> I'm not 100% sure if this is exactly what Tao pointed out here but I
> also had difficulties understanding understanding how this patch works:
>
>                        p.se
>                         |
>       __________________|
>       |
>       V
>      cfs_c -> tg_c ->  se_c (se->on_rq = 1)
>                         |
>       __________________|
>       |
>       v
>      cfs_b -> tg_b ->  se_b
>                         |
>       __________________|
>       |
>       V
>      cfs_a -> tg_a ->  se_a
>                         |
>       __________________|
>       |
>       V
>      cfs_r -> tg_r
>       |
>       V
>       rq
>

In your example, which cfs_ rq has been throttled ? cfs_a ?

> (1) The incomplete update happens with cfs_c at the end of
>     enqueue_entity() in the first loop because of 'if ( .... ||
>     cfs_bandwidth_used())' (cfs_b->on_list=0 since cfs_a is throttled)

so cfs_c is added with the 1st loop

>
> (2) se_c breaks out of the first loop (se_c->on_rq = 1)
>
> (3) With the patch cfs_b is added back to the list.
>     But only because cfs_a->on_list=1.

hmm I don't understand the link between cfs_b been added and cfs_a->on_list=1

cfs_b is added with 2nd loop because its throttle_count > 0 due to
cfs_a been throttled (purpose of this patch)

>
> But since cfs_a is throttled it should be cfs_a->on_list=0 as well.

So 2nd loop breaks because cfs_a is throttled

The 3rd loop will add cfs_a

> throttle_cfs_rq()->walk_tg_tree_from(..., tg_throttle_down, ...) should
> include cfs_a when calling list_del_leaf_cfs_rq().
>
> IMHO, throttle_cfs_rq() calls tg_throttle_down() for the throttled
> cfs_rq too.
>
>
> Another thing: Why don't we use throttled_hierarchy(cfs_rq) instead of
> cfs_bandwidth_used() in enqueue_entity() as well?

Mainly to be conservative because as this patch demonstrates, there
are a lot of possible use cases and combinations and I can't ensure
that it is always safe to use the throttled_hierarchy.