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

[Michal Koutný]

On Tue, Sep 21, 2021 at 08:21:50PM +0100, Odin Ugedal <odin@xxxxxxx> wrote:
> With the changes in PATCH 1 of this series [0],

That patch is an independent fix of the described bug.

> I think this logic needs to be updated as well (or correct me if I am
> wrong).

I might be dense, what update are you referring to? (The adding to the
list in tg_unthrottle_up() is already guarded by
cfs_rq->throttle_count.)

> Did a quick look now, and it looks like there are some conflicts with
> the linus' tree when applying the series as well, but didn't look that
> deep into what caused it (for ref I tested on v5.15-rc2).

This v2 was based on v5.14-rc6 back then and yes when I was rebasing
(locally), I had to resolve conflicts because of SCHED_IDLE for cgroups.

> Not sure how you want to structure this patch series as all the
> patches kinda depend on each other, since you sent the updated one
> separately (and I am fairly new to kernel development, so I have no
> idea), while patch 1 is fixing a "real" issue that we probably want to
> get fixed as soon as possible.

I put the patch first in the series to be backport friendly but then I
decided to send v3 of just that one patch, exactly for the reason of
making the fix earlier.
I may get down to addressing feedback of the remaining patches (v2, 2--5)
only later. Thanks for your comments and patience :-)

Michal