workqueue: restore WQ_UNBOUND/max_active==1 to be ordered

The combination of WQ_UNBOUND and max_active == 1 used to imply
ordered execution.  After NUMA affinity 4c16bd3 ("workqueue:
implement NUMA affinity for unbound workqueues"), this is no longer
true due to per-node worker pools.

While the right way to create an ordered workqueue is
alloc_ordered_workqueue(), the documentation has been misleading for a
long time and people do use WQ_UNBOUND and max_active == 1 for ordered
workqueues which can lead to subtle bugs which are very difficult to
trigger.

It's unlikely that we'd see noticeable performance impact by enforcing
ordering on WQ_UNBOUND / max_active == 1 workqueues.  Let's
automatically set __WQ_ORDERED for those workqueues.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Christoph Hellwig <hch@infradead.org>
Reported-by: Alexei Potashnik <alexei@purestorage.com>
Fixes: 4c16bd3 ("workqueue: implement NUMA affinity for unbound workqueues")
Cc: stable@vger.kernel.org # v3.10+

kernel/workqueue.c

@@ -3929,6 +3929,16 @@ struct workqueue_struct *__alloc_workqueue_key(const char *fmt,
 	struct workqueue_struct *wq;
 	struct pool_workqueue *pwq;
 
+	/*
+	 * Unbound && max_active == 1 used to imply ordered, which is no
+	 * longer the case on NUMA machines due to per-node pools.  While
+	 * alloc_ordered_workqueue() is the right way to create an ordered
+	 * workqueue, keep the previous behavior to avoid subtle breakages
+	 * on NUMA.
+	 */
+	if ((flags & WQ_UNBOUND) && max_active == 1)
+		flags |= __WQ_ORDERED;
+
 	/* see the comment above the definition of WQ_POWER_EFFICIENT */
 	if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient)
 		flags |= WQ_UNBOUND;