sched/wait: Fix rcuwait_wake_up() ordering
authorPrateek Sood <prsood@codeaurora.org>
Fri, 30 Nov 2018 15:10:56 +0000 (20:40 +0530)
committerIngo Molnar <mingo@kernel.org>
Mon, 21 Jan 2019 10:15:36 +0000 (11:15 +0100)
For some peculiar reason rcuwait_wake_up() has the right barrier in
the comment, but not in the code.

This mistake has been observed to cause a deadlock in the following
situation:

    P1 P2

    percpu_up_read() percpu_down_write()
      rcu_sync_is_idle() // false
  rcu_sync_enter()
  ...
      __percpu_up_read()

[S] ,-  __this_cpu_dec(*sem->read_count)
    |   smp_rmb();
[L] |   task = rcu_dereference(w->task) // NULL
    |
    |     [S]     w->task = current
    |     smp_mb();
    |     [L]     readers_active_check() // fail
    `-> <store happens here>

Where the smp_rmb() (obviously) fails to constrain the store.

[ peterz: Added changelog. ]

Signed-off-by: Prateek Sood <prsood@codeaurora.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Andrea Parri <andrea.parri@amarulasolutions.com>
Acked-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 8f95c90ceb54 ("sched/wait, RCU: Introduce rcuwait machinery")
Link: https://lkml.kernel.org/r/1543590656-7157-1-git-send-email-prsood@codeaurora.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/exit.c

index 284f2fe..3fb7be0 100644 (file)
@@ -307,7 +307,7 @@ void rcuwait_wake_up(struct rcuwait *w)
         *        MB (A)              MB (B)
         *    [L] cond            [L] tsk
         */
-       smp_rmb(); /* (B) */
+       smp_mb(); /* (B) */
 
        /*
         * Avoid using task_rcu_dereference() magic as long as we are careful,