1 files changed, 13 insertions, 4 deletions

diff --git a/sys/dev/mps/mps_sas.c b/sys/dev/mps/mps_sas.c
index a4dd3e1a4c78..3790e8f218b6 100644
--- a/sys/dev/mps/mps_sas.c
+++ b/sys/dev/mps/mps_sas.c
@@ -2408,11 +2408,20 @@ mpssas_scsiio_complete(struct mps_softc *sc, struct mps_command *cm)
 	case MPI2_IOCSTATUS_SCSI_IOC_TERMINATED:
 	case MPI2_IOCSTATUS_SCSI_EXT_TERMINATED:
 		/*
-		 * Since these are generally external (i.e. hopefully
-		 * transient transport-related) errors, retry these without
-		 * decrementing the retry count.
+		 * These can sometimes be transient transport-related
+		 * errors, and sometimes persistent drive-related errors.
+		 * We used to retry these without decrementing the retry
+		 * count by returning CAM_REQUEUE_REQ.  Unfortunately, if
+		 * we hit a persistent drive problem that returns one of
+		 * these error codes, we would retry indefinitely.  So,
+		 * return CAM_REQ_CMP_ERROR so that we decrement the retry
+		 * count and avoid infinite retries.  We're taking the
+		 * potential risk of flagging false failures in the event
+		 * of a topology-related error (e.g. a SAS expander problem
+		 * causes a command addressed to a drive to fail), but
+		 * avoiding getting into an infinite retry loop.
 		 */
-		mpssas_set_ccbstatus(ccb, CAM_REQUEUE_REQ);
+		mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR);
 		mpssas_log_command(cm, MPS_INFO,
 		    "terminated ioc %x scsi %x state %x xfer %u\n",
 		    le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState,