]> www.infradead.org Git - users/jedix/linux-maple.git/commit
sched/fair: Properly deactivate sched_delayed task upon class change
authorValentin Schneider <vschneid@redhat.com>
Thu, 29 Aug 2024 13:53:53 +0000 (15:53 +0200)
committerPeter Zijlstra <peterz@infradead.org>
Tue, 3 Sep 2024 13:26:30 +0000 (15:26 +0200)
commit75b6499024a6c1a4ef0288f280534a5c54269076
tree5ff6cd1eb6bc213ec76c3d6e8807868bf0a482f9
parent9c602adb799e72ee537c0c7ca7e828c3fe2acad6
sched/fair: Properly deactivate sched_delayed task upon class change

__sched_setscheduler() goes through an enqueue/dequeue cycle like so:

  flags := DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
  prev_class->dequeue_task(rq, p, flags);
  new_class->enqueue_task(rq, p, flags);

when prev_class := fair_sched_class, this is followed by:

  dequeue_task(rq, p, DEQUEUE_NOCLOCK | DEQUEUE_SLEEP);

the idea being that since the task has switched classes, we need to drop
the sched_delayed logic and have that task be deactivated per its previous
dequeue_task(..., DEQUEUE_SLEEP).

Unfortunately, this leaves the task on_rq. This is missing the tail end of
dequeue_entities() that issues __block_task(), which __sched_setscheduler()
won't have done due to not using DEQUEUE_DELAYED - not that it should, as
it is pretty much a fair_sched_class specific thing.

Make switched_from_fair() properly deactivate sched_delayed tasks upon
class changes via __block_task(), as if a
  dequeue_task(..., DEQUEUE_DELAYED)
had been issued.

Fixes: 2e0199df252a ("sched/fair: Prepare exit/cleanup paths for delayed_dequeue")
Reported-by: "Paul E. McKenney" <paulmck@kernel.org>
Reported-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20240829135353.1524260-1-vschneid@redhat.com
kernel/sched/fair.c