Index: sys/dev/nvme/nvme_qpair.c
===================================================================
--- sys/dev/nvme/nvme_qpair.c
+++ sys/dev/nvme/nvme_qpair.c
@@ -982,14 +982,6 @@
 
 	bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map,
 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
-#if !defined( __powerpc__) && !defined( __aarch64__)  && !defined( __arm__)
-	/*
-	 * powerpc's bus_dmamap_sync() already includes a heavyweight sync, but
-	 * no other archs do.
-	 */
-	wmb();
-#endif
-
 	bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle,
 	    qpair->sq_tdbl_off, qpair->sq_tail);
 	qpair->num_cmds++;
Index: sys/x86/x86/busdma_bounce.c
===================================================================
--- sys/x86/x86/busdma_bounce.c
+++ sys/x86/x86/busdma_bounce.c
@@ -969,7 +969,7 @@
 	bus_size_t datacount1, datacount2;
 
 	if (map == NULL || (bpage = STAILQ_FIRST(&map->bpages)) == NULL)
-		return;
+		goto out;
 
 	/*
 	 * Handle data bouncing.  We might also want to add support for
@@ -1059,6 +1059,8 @@
 		}
 		dmat->bounce_zone->total_bounced++;
 	}
+out:
+	atomic_thread_fence_rel();
 }
 
 static void