Current ahd timeout code does nothing. Add fbsd style timeout

handling which will at least try to recover.

ok marco@.

4 files changed, 94 insertions, 48 deletions

diff --git a/sys/dev/ic/aic79xx.c b/sys/dev/ic/aic79xx.c
index 979cd38b77a..2cc6d0995d4 100644
--- a/sys/dev/ic/aic79xx.c
+++ b/sys/dev/ic/aic79xx.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: aic79xx.c,v 1.24 2004/12/20 20:56:32 krw Exp $	*/
+/*	$OpenBSD: aic79xx.c,v 1.25 2004/12/30 17:29:55 krw Exp $	*/
 
 /*
  * Copyright (c) 2004 Milos Urbanek, Kenneth R. Westerback & Marco Peereboom
@@ -253,11 +253,9 @@ u_int		ahd_resolve_seqaddr(struct ahd_softc *ahd,
 void		ahd_download_instr(struct ahd_softc *ahd,
 					   u_int instrptr, uint8_t *dconsts);
 int		ahd_probe_stack_size(struct ahd_softc *ahd);
-#if 0
 int		ahd_other_scb_timeout(struct ahd_softc *ahd,
 					      struct scb *scb,
 					      struct scb *other_scb);
-#endif
 int		ahd_scb_active_in_fifo(struct ahd_softc *ahd,
 					       struct scb *scb);
 void		ahd_run_data_fifo(struct ahd_softc *ahd,
@@ -8232,9 +8230,14 @@ ahd_handle_scsi_status(struct ahd_softc *ahd, struct scb *scb)
 		ahd_queue_scb(ahd, scb);
 		/*
 		 * Ensure we have enough time to actually
-		 * retrieve the sense.
+		 * retrieve the sense, but only schedule
+		 * the timer if we are not in recovery or
+		 * this is a recovery SCB that is allowed
+		 * to have an active timer.
 		 */
-		aic_scb_timer_reset(scb, 5 * 1000);
+		if (ahd->scb_data.recovery_scbs == 0
+		 || (scb->flags & SCB_RECOVERY_SCB) != 0)
+			aic_scb_timer_reset(scb, 5 * 1000);
 		break;
 	}
 	case SCSI_STATUS_OK:
@@ -9143,21 +9146,37 @@ ahd_dump_scbs(struct ahd_softc *ahd)
 void
 ahd_timeout(void *arg)
 {
-	struct  scb *scb;
+	struct scb *scb = (struct scb *)arg;
 	struct ahd_softc *ahd;
-	int s;
-#if 0
-	int found;
-	u_int	last_phase;
-	int target;
-	int lun;
-	int i;
-	int channel;
-#endif
-	int was_paused;
 
-	scb = (struct scb *)arg;
-	ahd = (struct ahd_softc *)scb->xs->sc_link->adapter_softc;
+	ahd = scb->ahd_softc;
+	if ((scb->flags & SCB_ACTIVE) != 0) {
+		if ((scb->flags & SCB_TIMEDOUT) == 0) {
+			LIST_INSERT_HEAD(&ahd->timedout_scbs, scb,
+					 timedout_links);
+			scb->flags |= SCB_TIMEDOUT;
+		}
+		ahd_recover_commands(ahd);
+	}
+}
+
+/*
+ * ahd_recover_commands determines if any of the commands that have currently
+ * timedout are the root cause for this timeout.  Innocent commands are given
+ * a new timeout while we wait for the command executing on the bus to timeout.
+ * This routine is invoked from a thread context so we are allowed to sleep.
+ * Our lock is not held on entry.
+ */
+void
+ahd_recover_commands(struct ahd_softc *ahd)
+{
+	struct	scb *scb;
+	struct	scb *active_scb;
+	long	s;
+	int	found;
+	int	was_paused;
+	u_int	active_scbptr;
+	u_int	last_phase;
 
 	ahd_lock(ahd, &s);
 
@@ -9174,23 +9193,22 @@ ahd_timeout(void *arg)
 
 	ahd_pause_and_flushwork(ahd);
 
-	if ((scb->flags & SCB_ACTIVE) == 0) {
+	if (LIST_EMPTY(&ahd->timedout_scbs) != 0) {
 		/*
-		 * The timed out commands have already
+		 * The timedout commands have already
 		 * completed.  This typically means
 		 * that either the timeout value was on
 		 * the hairy edge of what the device
 		 * requires or - more likely - interrupts
 		 * are not happening.
 		 */
-		printf("%s: Timed out SCBs already complete. "
+		printf("%s: Timedout SCBs already complete. "
 		       "Interrupts may not be functioning.\n", ahd_name(ahd));
 		ahd_unpause(ahd);
 		ahd_unlock(ahd, &s);
 		return;
 	}
 
-#if 0
 	/*
 	 * Determine identity of SCB acting on the bus.
 	 * This test only catches non-packetized transactions.
@@ -9216,7 +9234,7 @@ ahd_timeout(void *arg)
 		lun = SCB_GET_LUN(scb);
 
 		ahd_print_path(ahd, scb);
-		printf("SCB 0x%x - timed out\n", scb->hscb->tag);
+		printf("SCB %d - timed out\n", SCB_GET_TAG(scb));
 
 		if (scb->flags & (SCB_DEVICE_RESET|SCB_ABORT)) {
 			/*
@@ -9283,7 +9301,7 @@ bus_reset:
 			       "Identify Msg.\n", ahd_name(ahd));
 			goto bus_reset;
 		} else if (ahd_search_qinfifo(ahd, target, channel, lun,
-					      scb->hscb->tag, ROLE_INITIATOR,
+					      SCB_GET_TAG(scb), ROLE_INITIATOR,
 					      /*status*/0, SEARCH_COUNT) > 0) {
 
 			/*
@@ -9375,12 +9393,11 @@ bus_reset:
 		LIST_REMOVE(scb, timedout_links);
 		scb->flags &= ~SCB_TIMEDOUT;
 	}
-#endif
+
 	ahd_unpause(ahd);
 	ahd_unlock(ahd, &s);
 }
 
-#if 0
 /*
  * Re-schedule a timeout for the passed in SCB if we determine that some
  * other SCB is in the process of recovery or an SCB with a longer
@@ -9431,7 +9448,7 @@ ahd_other_scb_timeout(struct ahd_softc *ahd, struct scb *scb,
 
 	return (found != 0);
 }
-#endif
+
 /**************************** Flexport Logic **********************************/
 /*
  * Read count 16bit words from 16bit word address start_addr from the
diff --git a/sys/dev/ic/aic79xx.h b/sys/dev/ic/aic79xx.h
index 7e1f887555d..1a917cf6eda 100644
--- a/sys/dev/ic/aic79xx.h
+++ b/sys/dev/ic/aic79xx.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: aic79xx.h,v 1.17 2004/12/19 06:17:54 krw Exp $	*/
+/*	$OpenBSD: aic79xx.h,v 1.18 2004/12/30 17:29:55 krw Exp $	*/
 
 /*
  * Copyright (c) 2004 Milos Urbanek, Kenneth R. Westerback & Marco Peereboom
@@ -1095,6 +1095,8 @@ struct ahd_completion
 	uint8_t		valid_tag;
 };
 
+#define AIC_SCB_DATA(softc) (&(softc)->scb_data)
+
 struct ahd_softc {
 	struct device		sc_dev;
 	struct scsi_link	sc_channel;
@@ -1588,4 +1590,6 @@ int			ahd_print_register(ahd_reg_parse_entry_t *table,
 					   u_int *cur_column,
 					   u_int wrap_point);
 void			ahd_dump_scbs(struct ahd_softc *ahd);
+void			ahd_set_recoveryscb(struct ahd_softc *, struct scb *);
+
 #endif /* _AIC79XX_H_ */
diff --git a/sys/dev/ic/aic79xx_openbsd.c b/sys/dev/ic/aic79xx_openbsd.c
index 9d00225b80a..9e866230a44 100644
--- a/sys/dev/ic/aic79xx_openbsd.c
+++ b/sys/dev/ic/aic79xx_openbsd.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: aic79xx_openbsd.c,v 1.19 2004/12/28 04:12:24 krw Exp $	*/
+/*	$OpenBSD: aic79xx_openbsd.c,v 1.20 2004/12/30 17:29:55 krw Exp $	*/
 
 /*
  * Copyright (c) 2004 Milos Urbanek, Kenneth R. Westerback & Marco Peereboom
@@ -170,6 +170,7 @@ void
 ahd_done(struct ahd_softc *ahd, struct scb *scb)
 {
 	struct scsi_xfer *xs = scb->xs;
+	struct scb *list_scb;
 	int s;
 
 	/* XXX in ahc there is some bus_dmamap_sync(PREREAD|PREWRITE); */
@@ -195,25 +196,29 @@ ahd_done(struct ahd_softc *ahd, struct scb *scb)
 	 * out of our timeout.
 	 */
 	if ((scb->flags & SCB_RECOVERY_SCB) != 0) {
-		struct	scb *list_scb;
+		ahd->scb_data.recovery_scbs--;
 
-		/*
-		 * We were able to complete the command successfully,
-		 * so reinstate the timeouts for all other pending
-		 * commands.
-		 */
-		LIST_FOREACH(list_scb, &ahd->pending_scbs, pending_links) {
-			struct scsi_xfer *txs = list_scb->xs;
-			if (!(txs->flags & SCSI_POLL))
-				timeout_add(&list_scb->xs->stimeout,
-				    (list_scb->xs->timeout * hz)/1000);
-		}
-
-		if (aic_get_transaction_status(scb) != CAM_REQ_INPROG)
+		if (aic_get_transaction_status(scb) == CAM_BDR_SENT
+		 || aic_get_transaction_status(scb) == CAM_REQ_ABORTED)
 			aic_set_transaction_status(scb, CAM_CMD_TIMEOUT);
-		ahd_print_path(ahd, scb);
-		printf("%s: no longer in timeout, status = %x\n",
-		    ahd_name(ahd), xs->status);
+
+		if (ahd->scb_data.recovery_scbs == 0) {
+			/*
+			 * All recovery actions have completed successfully,
+			 * so reinstate the timeouts for all other pending
+			 * commands.
+			 */
+			LIST_FOREACH(list_scb, &ahd->pending_scbs,
+			    pending_links) {
+				if (!(list_scb->xs->flags & SCSI_POLL))
+					aic_scb_timer_reset(scb,
+					    aic_get_timeout(scb));
+			}
+
+			ahd_print_path(ahd, scb);
+			printf("%s: no longer in timeout, status = %x\n",
+			       ahd_name(ahd), aic_get_transaction_status(scb));
+		}
 	}
 
 	/* Translate the CAM status code to a SCSI error code. */
@@ -819,4 +824,24 @@ ahd_platform_flushwork(struct ahd_softc *ahd)
 {
 }
 
+void
+ahd_set_recoveryscb(struct ahd_softc *ahd, struct scb *scb)
+{
+
+	if ((scb->flags & SCB_RECOVERY_SCB) == 0) {
+		struct scb *list_scb;
+
+		scb->flags |= SCB_RECOVERY_SCB;
+
+		AIC_SCB_DATA(ahd)->recovery_scbs++;
 
+		/*
+		 * Go through all of our pending SCBs and remove
+		 * any scheduled timeouts for them.  We will reschedule
+		 * them after we've successfully fixed this problem.
+		 */
+		LIST_FOREACH(list_scb, &ahd->pending_scbs, pending_links) {
+			timeout_del(&list_scb->xs->stimeout);
+		}
+	}
+}
diff --git a/sys/dev/ic/aic79xx_openbsd.h b/sys/dev/ic/aic79xx_openbsd.h
index 51931af62e1..0b4e60aecca 100644
--- a/sys/dev/ic/aic79xx_openbsd.h
+++ b/sys/dev/ic/aic79xx_openbsd.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: aic79xx_openbsd.h,v 1.14 2004/12/24 22:17:56 krw Exp $	*/
+/*	$OpenBSD: aic79xx_openbsd.h,v 1.15 2004/12/30 17:29:55 krw Exp $	*/
 
 /*
  * Copyright (c) 2004 Milos Urbanek, Kenneth R. Westerback & Marco Peereboom
@@ -162,7 +162,7 @@ void aic_timer_reset(aic_timer_t *, u_int, ahd_callback_t *, void *);
 void aic_scb_timer_reset(struct scb *, u_int);
 
 #define aic_timer_stop timeout_del
-
+#define aic_get_timeout(scb) ((scb)->xs->timeout)
 /*************************** Device Access ************************************/
 #define ahd_inb(ahd, port)					\
 	bus_space_read_1((ahd)->tags[(port) >> 8],		\