Re: net/bluetooth: workqueue destruction WARNING in hci_unregister_dev

[Tejun Heo]

Hello,

Can you please do the followings?

1. Remove WQ_MEM_RECLAIM from the affected workqueue and see whether
   the problem is reproducible.  WQ_MEM_RECLAIM on anything bluetooth
   doesn't make sense btw.  Why is it there?

2. If WQ_MEM_RECLAIM makes the issue go away, see whether the attached
   patch works too.

Thanks.

-- 
tejun
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 7ff5dc7..9824d4f 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -4012,7 +4012,7 @@ void destroy_workqueue(struct workqueue_struct *wq)
 	/* drain it before proceeding with destruction */
 	drain_workqueue(wq);
 
-	/* sanity checks */
+	/* nothing should be in flight */
 	mutex_lock(&wq->mutex);
 	for_each_pwq(pwq, wq) {
 		int i;
@@ -4024,8 +4024,7 @@ void destroy_workqueue(struct workqueue_struct *wq)
 			}
 		}
 
-		if (WARN_ON((pwq != wq->dfl_pwq) && (pwq->refcnt > 1)) ||
-		    WARN_ON(pwq->nr_active) ||
+		if (WARN_ON(pwq->nr_active) ||
 		    WARN_ON(!list_empty(&pwq->delayed_works))) {
 			mutex_unlock(&wq->mutex);
 			return;
@@ -4046,6 +4045,13 @@ void destroy_workqueue(struct workqueue_struct *wq)
 	if (wq->rescuer)
 		kthread_stop(wq->rescuer->task);
 
+	/* rescuer is gone, everything should be quiescent now */
+	WARN_ON(!list_empty(&wq->maydays));
+	mutex_lock(&wq->mutex);
+	for_each_pwq(pwq, wq)
+		WARN_ON((pwq != wq->dfl_pwq) && (pwq->refcnt > 1));
+	mutex_unlock(&wq->mutex);
+
 	if (!(wq->flags & WQ_UNBOUND)) {
 		/*
 		 * The base ref is never dropped on per-cpu pwqs.  Directly