sched/rt: Plug rt_mutex_setprio() vs push_rt_task() race
authorValentin Schneider <valentin.schneider@arm.com>
Thu, 27 Jan 2022 15:40:59 +0000 (15:40 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 8 Apr 2022 12:23:11 +0000 (14:23 +0200)
commit1e0e63ad6243f378e8476b42412f83c1be6ebbed
tree2f8e42cb5a2fe228f2ea5eab27f8adcab5709194
parentb7aec0843e2883b4bd5bed46f70b0febb046f9cc
sched/rt: Plug rt_mutex_setprio() vs push_rt_task() race

[ Upstream commit 49bef33e4b87b743495627a529029156c6e09530 ]

John reported that push_rt_task() can end up invoking
find_lowest_rq(rq->curr) when curr is not an RT task (in this case a CFS
one), which causes mayhem down convert_prio().

This can happen when current gets demoted to e.g. CFS when releasing an
rt_mutex, and the local CPU gets hit with an rto_push_work irqwork before
getting the chance to reschedule. Exactly who triggers this work isn't
entirely clear to me - switched_from_rt() only invokes rt_queue_pull_task()
if there are no RT tasks on the local RQ, which means the local CPU can't
be in the rto_mask.

My current suspected sequence is something along the lines of the below,
with the demoted task being current.

  mark_wakeup_next_waiter()
    rt_mutex_adjust_prio()
      rt_mutex_setprio() // deboost originally-CFS task
check_class_changed()
  switched_from_rt() // Only rt_queue_pull_task() if !rq->rt.rt_nr_running
  switched_to_fair() // Sets need_resched
      __balance_callbacks() // if pull_rt_task(), tell_cpu_to_push() can't select local CPU per the above
      raw_spin_rq_unlock(rq)

       // need_resched is set, so task_woken_rt() can't
       // invoke push_rt_tasks(). Best I can come up with is
       // local CPU has rt_nr_migratory >= 2 after the demotion, so stays
       // in the rto_mask, and then:

       <some other CPU running rto_push_irq_work_func() queues rto_push_work on this CPU>
 push_rt_task()
   // breakage follows here as rq->curr is CFS

Move an existing check to check rq->curr vs the next pushable task's
priority before getting anywhere near find_lowest_rq(). While at it, add an
explicit sched_class of rq->curr check prior to invoking
find_lowest_rq(rq->curr). Align the DL logic to also reschedule regardless
of next_task's migratability.

Fixes: a7c81556ec4d ("sched: Fix migrate_disable() vs rt/dl balancing")
Reported-by: John Keeping <john@metanate.com>
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Tested-by: John Keeping <john@metanate.com>
Link: https://lore.kernel.org/r/20220127154059.974729-1-valentin.schneider@arm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
kernel/sched/deadline.c
kernel/sched/rt.c