diff --git a/sys/dev/nvme/nvme_qpair.c b/sys/dev/nvme/nvme_qpair.c
--- a/sys/dev/nvme/nvme_qpair.c
+++ b/sys/dev/nvme/nvme_qpair.c
@@ -956,36 +956,44 @@
 	struct nvme_tracker	*tr;
 	sbintime_t		now;
 	bool			idle;
+	bool			expired;
 	uint32_t		csts;
 	uint8_t			cfs;
 
 	mtx_lock(&qpair->lock);
 	idle = TAILQ_EMPTY(&qpair->outstanding_tr);
+
 again:
 	switch (qpair->recovery_state) {
 	case RECOVERY_NONE:
-		if (idle)
-			break;
+		/*
+		 * Check to see if we need to timeout any commands. If we do, then
+		 * we also enter a recovery phase.
+		 */
 		now = getsbinuptime();
-		idle = true;
+		expired = false;
 		TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq) {
 			if (tr->deadline == SBT_MAX)
 				continue;
 			idle = false;
 			if (now > tr->deadline) {
-				/*
-				 * We're now passed our earliest deadline. We
-				 * need to do expensive things to cope, but next
-				 * time. Flag that and close the door to any
-				 * further processing.
-				 */
-				qpair->recovery_state = RECOVERY_START;
-				nvme_printf(ctrlr, "RECOVERY_START %jd vs %jd\n",
-				    (uintmax_t)now, (uintmax_t)tr->deadline);
-				break;
+				expired = true;
+				nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id,
+				    nvme_abort_complete, tr);
 			}
 		}
-		break;
+		if (!expired)
+			break;
+
+		/*
+		 * We're now passed our earliest deadline. We need to do
+		 * expensive things to cope, but next time. Flag that
+		 * and close the door to any further processing.
+		 */
+		qpair->recovery_state = RECOVERY_START;
+		nvme_printf(ctrlr, "RECOVERY_START %jd vs %jd\n",
+		    (uintmax_t)now, (uintmax_t)tr->deadline);
+		/* FALLTHROUGH */
 	case RECOVERY_START:
 		/*
 		 * Read csts to get value of cfs - controller fatal status.