workqueue: fix possible deadlock in idle worker rebinding
authorTejun Heo <tj@kernel.org>
Wed, 5 Sep 2012 06:16:32 +0000 (23:16 -0700)
committerTejun Heo <tj@kernel.org>
Wed, 5 Sep 2012 23:10:15 +0000 (16:10 -0700)
Currently, rebind_workers() and idle_worker_rebind() are two-way
interlocked.  rebind_workers() waits for idle workers to finish
rebinding and rebound idle workers wait for rebind_workers() to finish
rebinding busy workers before proceeding.

Unfortunately, this isn't enough.  The second wait from idle workers
is implemented as follows.

wait_event(gcwq->rebind_hold, !(worker->flags & WORKER_REBIND));

rebind_workers() clears WORKER_REBIND, wakes up the idle workers and
then returns.  If CPU hotplug cycle happens again before one of the
idle workers finishes the above wait_event(), rebind_workers() will
repeat the first part of the handshake - set WORKER_REBIND again and
wait for the idle worker to finish rebinding - and this leads to
deadlock because the idle worker would be waiting for WORKER_REBIND to
clear.

This is fixed by adding another interlocking step at the end -
rebind_workers() now waits for all the idle workers to finish the
above WORKER_REBIND wait before returning.  This ensures that all
rebinding steps are complete on all idle workers before the next
hotplug cycle can happen.

This problem was diagnosed by Lai Jiangshan who also posted a patch to
fix the issue, upon which this patch is based.

This is the minimal fix and further patches are scheduled for the next
merge window to simplify the CPU hotplug path.

Signed-off-by: Tejun Heo <tj@kernel.org>
Original-patch-by: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <1346516916-1991-3-git-send-email-laijs@cn.fujitsu.com>

kernel/workqueue.c

index d79a18d0c42e198e059f631b21a9365ff5873424..dc7b8458e275f73a38278974676c22d1a5c84d2a 100644 (file)
@@ -1326,6 +1326,15 @@ static void idle_worker_rebind(struct worker *worker)
 
        /* we did our part, wait for rebind_workers() to finish up */
        wait_event(gcwq->rebind_hold, !(worker->flags & WORKER_REBIND));
+
+       /*
+        * rebind_workers() shouldn't finish until all workers passed the
+        * above WORKER_REBIND wait.  Tell it when done.
+        */
+       spin_lock_irq(&worker->pool->gcwq->lock);
+       if (!--worker->idle_rebind->cnt)
+               complete(&worker->idle_rebind->done);
+       spin_unlock_irq(&worker->pool->gcwq->lock);
 }
 
 /*
@@ -1448,12 +1457,28 @@ retry:
         * be cleared inside idle_worker_rebind().  Clear and release.
         * Clearing %WORKER_REBIND from this foreign context is safe
         * because these workers are still guaranteed to be idle.
+        *
+        * We need to make sure all idle workers passed WORKER_REBIND wait
+        * in idle_worker_rebind() before returning; otherwise, workers can
+        * get stuck at the wait if hotplug cycle repeats.
         */
-       for_each_worker_pool(pool, gcwq)
-               list_for_each_entry(worker, &pool->idle_list, entry)
+       idle_rebind.cnt = 1;
+       INIT_COMPLETION(idle_rebind.done);
+
+       for_each_worker_pool(pool, gcwq) {
+               list_for_each_entry(worker, &pool->idle_list, entry) {
                        worker->flags &= ~WORKER_REBIND;
+                       idle_rebind.cnt++;
+               }
+       }
 
        wake_up_all(&gcwq->rebind_hold);
+
+       if (--idle_rebind.cnt) {
+               spin_unlock_irq(&gcwq->lock);
+               wait_for_completion(&idle_rebind.done);
+               spin_lock_irq(&gcwq->lock);
+       }
 }
 
 static struct worker *alloc_worker(void)