[PATCH] fix-flush_workqueue-vs-cpu_dead-race-update

[Oleg Nesterov <oleg@tv-sign.ru>]

( Srivatsa, Gautham, could you please verify my thinking ? )

On top of fix-flush_workqueue-vs-cpu_dead-race.patch

hotplug_sequence is incremented under "case CPU_DEAD:". This was ok
before flush_workqueue() was changed to use preempt_disable() instead
of workqueue_mutex. However preempt_disable() can't garantee that there
is no CPU_DEAD event in progress (it is possible that flush_workqueue()
runs after STOPMACHINE_EXIT), so flush_workqueue() can miss CPU_DEAD
event.

Increment hotplug_sequence earlier, under CPU_DOWN_PREPARE. We can't
miss the event, the task running flush_workqueue() will be re-scheduled
at least once before CPU actually disappears from cpu_online_map.

We may have a false positive but this is very unlikely and only means
we will have one unneeded "goto again".

Note: this patch depends on
handle-cpu_lock_acquire-and-cpu_lock_release-in-workqueue_cpu_callback
but only "textually".

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>

--- mm-6.20-rc3/kernel/workqueue.c~1_down	2007-01-06 16:15:59.000000000 +0300
+++ mm-6.20-rc3/kernel/workqueue.c	2007-01-06 17:52:10.000000000 +0300
@@ -886,12 +886,15 @@ static int __devinit workqueue_cpu_callb
 		}
 		break;
 
+	case CPU_DOWN_PREPARE:
+		hotplug_sequence++;
+		break;
+
 	case CPU_DEAD:
 		list_for_each_entry(wq, &workqueues, list)
 			cleanup_workqueue_thread(wq, hotcpu);
 		list_for_each_entry(wq, &workqueues, list)
 			take_over_work(wq, hotcpu);
-		hotplug_sequence++;
 		break;
 
 	case CPU_LOCK_RELEASE:

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/