Re: [RFC PATCH v2 4/5] sched/fair: Simplify load_cfs_rq_list maintenance

[Vincent Guittot]

On Tue, 14 Sept 2021 at 11:45, Vincent Guittot
<vincent.guittot@xxxxxxxxxx> wrote:
>
> On Tue, 14 Sept 2021 at 11:22, Michal Koutný <mkoutny@xxxxxxxx> wrote:
> >
> > On Fri, Sep 10, 2021 at 04:19:27PM +0200, Vincent Guittot <vincent.guittot@xxxxxxxxxx> wrote:
> > > Your proposal looks interesting but I need more time to make sure that
> > > all cases are covered. We have faced several odd corner cases and
> > > sequences in the past that I need time to check that you don't put
> > > some back
> >
> > Do you have any pointers to the cases that come to your mind? I wonder
> > if those could be reproduced with a simple setup.
>
> I don't have a strict list but several warnings for leaf_list have
> been already reported on lkml in the past and the use cases were quite
> complicated and I want to go through them to make sure they are still
> covered.

The corner cases that I wanted to check, are covered by
a7b359fc6a37 ("sched/fair: Correctly insert cfs_rq's to list on unthrottle")
and
fdaba61ef8a2 ("sched/fair: Ensure that the CFS parent is added after
unthrottling")

This patch looks ok to me.
Also, propagate_entity_cfs_rq() could also get advantage of the same
kind of change

>
> Vincent
>
> > (FTR, I used the LTP test (at b673f49ae) cfs_bandwidth01 to check this change.)
> >
> > Thanks,
> > Michal