psi: get poll_work to run when calling poll syscall next time
authorJason Xing <kerneljasonxing@linux.alibaba.com>
Sun, 25 Aug 2019 00:54:53 +0000 (17:54 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sun, 25 Aug 2019 02:48:42 +0000 (19:48 -0700)
Only when calling the poll syscall the first time can user receive
POLLPRI correctly.  After that, user always fails to acquire the event
signal.

Reproduce case:
 1. Get the monitor code in Documentation/accounting/psi.txt
 2. Run it, and wait for the event triggered.
 3. Kill and restart the process.

The question is why we can end up with poll_scheduled = 1 but the work
not running (which would reset it to 0).  And the answer is because the
scheduling side sees group->poll_kworker under RCU protection and then
schedules it, but here we cancel the work and destroy the worker.  The
cancel needs to pair with resetting the poll_scheduled flag.

Link: http://lkml.kernel.org/r/1566357985-97781-1-git-send-email-joseph.qi@linux.alibaba.com
Signed-off-by: Jason Xing <kerneljasonxing@linux.alibaba.com>
Signed-off-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Reviewed-by: Caspar Zhang <caspar@linux.alibaba.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/sched/psi.c

index 23fbbcc414d5d739507ac1f8ecb5757c87a9ce55..6e52b67b420e7a3312f463f8d3bb6baad6576041 100644 (file)
@@ -1131,7 +1131,15 @@ static void psi_trigger_destroy(struct kref *ref)
         * deadlock while waiting for psi_poll_work to acquire trigger_lock
         */
        if (kworker_to_destroy) {
         * deadlock while waiting for psi_poll_work to acquire trigger_lock
         */
        if (kworker_to_destroy) {
+               /*
+                * After the RCU grace period has expired, the worker
+                * can no longer be found through group->poll_kworker.
+                * But it might have been already scheduled before
+                * that - deschedule it cleanly before destroying it.
+                */
                kthread_cancel_delayed_work_sync(&group->poll_work);
                kthread_cancel_delayed_work_sync(&group->poll_work);
+               atomic_set(&group->poll_scheduled, 0);
+
                kthread_destroy_worker(kworker_to_destroy);
        }
        kfree(t);
                kthread_destroy_worker(kworker_to_destroy);
        }
        kfree(t);