Re: [PATCH 09/10 v2] sched/fair: Fix sched_can_stop_tick() for fair tasks

From: K Prateek Nayak
Date: Fri Nov 29 2024 - 13:26:39 EST


(+ Tejun, David)

Hello Vincent,

On 11/29/2024 9:47 PM, Vincent Guittot wrote:
We can't stop the tick of a rq if there are at least 2 tasks enqueued in
the whole hierarchy and not only at the root cfs rq.

rq->cfs.nr_queued tracks the number of sched_entity at one level
whereas rq->cfs.h_nr_enqueued tracks all enqueued tasks in the
hierarchy.

Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
---
kernel/sched/core.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 3571f91d4b0d..866a1605656c 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -1341,7 +1341,7 @@ bool sched_can_stop_tick(struct rq *rq)
if (scx_enabled() && !scx_can_stop_tick(rq))
return false;
- if (rq->cfs.nr_queued > 1)
+ if (rq->cfs.h_nr_queued > 1)

Perhaps we can move this fix to the beginning of the series and add:

Fixes: 11cc374f4643b ("sched_ext: Simplify scx_can_stop_tick() invocation in sched_can_stop_tick()")

before converting the h_nr_running to h_nr_queued since prior to that
commit, sched_can_stop_tick() used to check "rq->nr_running" and since
we check the count of DL, RR, and FIFO tasks up above, it would have
captured number of fair tasks running before sched-ext. That way the fix
can be backported easily to LTS too. Thoughts?

return false;
/*

--
Thanks and Regards,
Prateek