diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 3fbaecfc88c2..f98247ec99c2 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2298,9 +2298,13 @@ retry:
 	 * If @work was previously on a different pool, it might still be
 	 * running there, in which case the work needs to be queued on that
 	 * pool to guarantee non-reentrancy.
+	 *
+	 * For ordered workqueue, work items must be queued on the newest pwq
+	 * for accurate order management.  Guaranteed order also guarantees
+	 * non-reentrancy.  See the comments above unplug_oldest_pwq().
 	 */
 	last_pool = get_work_pool(work);
-	if (last_pool && last_pool != pool) {
+	if (last_pool && last_pool != pool && !(wq->flags & __WQ_ORDERED)) {
 		struct worker *worker;
 
 		raw_spin_lock(&last_pool->lock);