Re: [PATCH 1/4] sched/fair: Propagate load for throttled cfs_rq

[Aaron Lu]

On Tue, Sep 23, 2025 at 03:05:29PM +0200, Matteo Martelli wrote:
> Hi Aaron,
> 
> On Wed, 10 Sep 2025 17:50:41 +0800, Aaron Lu <ziqianlu@xxxxxxxxxxxxx> wrote:
> > Before task based throttle model, propagating load will stop at a
> > throttled cfs_rq and that propagate will happen on unthrottle time by
> > update_load_avg().
> > 
> > Now that there is no update_load_avg() on unthrottle for throttled
> > cfs_rq and all load tracking is done by task related operations, let the
> > propagate happen immediately.
> > 
> > While at it, add a comment to explain why cfs_rqs that are not affected
> > by throttle have to be added to leaf cfs_rq list in
> > propagate_entity_cfs_rq() per my understanding of commit 0258bdfaff5b
> > ("sched/fair: Fix unfairness caused by missing load decay").
> > 
> > Signed-off-by: Aaron Lu <ziqianlu@xxxxxxxxxxxxx>
> > ---
> 
> I have been testing again the patch set "[PATCH v4 0/5] Defer throttle
> when task exits to user" [1] together with these follow up patches. I
> found out that with this patch the kernel sometimes produces the warning
> WARN_ON_ONCE(rq->tmp_alone_branch != &rq->leaf_cfs_rq_list); in
> assert_list_leaf_cfs_rq() called by enqueue_task_fair(). I could
> reproduce this systematically by applying both [1] and this patch on top
> of tag v6.17-rc6 and also by directly testing at commit fe8d238e646e
> from sched/core branch of tip tree. I couldn't reproduce the warning by
> testing at commmit 5b726e9bf954 ("sched/fair: Get rid of
> throttled_lb_pair()").

Just a note that while trying to reproduce this problem, I noticed a
warn triggered in tg_throttle_down() with one setup. It's a different
problem and I've sent a patch to address that:
https://lore.kernel.org/lkml/20250929074645.416-1-ziqianlu@xxxxxxxxxxxxx/