Fix the mly driver! If card resources became unavailable, the driver would

correctly tell CAM to requeue the command and then freeze it's queue.  The
problem was that when resources became available again, it wouldn't tell
CAM to unfreeze it's queue, so no more commands would ever be delivered.
This is simialr to the bug that was fixed in the cciss driver last year.
This is a bug in 4-STABLE also, but is probably masked by the OS being
fast enough to drain the completion queue before it fills up.
Also add some diagnostics avaialble when compiled with MLY_DEBUG.
Thanks very much to LSI Corp for donating equipment to track this down,
and Vaidus Damosevicius for pestering me long enough to get it fixed.

1 files changed, 46 insertions, 0 deletions

diff --git a/sys/dev/mly/mly.c b/sys/dev/mly/mly.c
index 50b81ac..a91b295 100644
--- a/sys/dev/mly/mly.c
+++ b/sys/dev/mly/mly.c
@@ -121,6 +121,7 @@ static void	mly_print_packet(struct mly_command *mc);
 static void	mly_panic(struct mly_softc *sc, char *reason);
 #endif
 void		mly_print_controller(int controller);
+static int	mly_timeout(struct mly_softc *sc);
 
 
 static d_open_t		mly_user_open;
@@ -129,6 +130,7 @@ static d_ioctl_t	mly_user_ioctl;
 static int	mly_user_command(struct mly_softc *sc, struct mly_user_command *uc);
 static int	mly_user_health(struct mly_softc *sc, struct mly_user_health *uh);
 
+#define MLY_CMD_TIMEOUT		20
 
 static device_method_t mly_methods[] = {
     /* Device interface */
@@ -330,6 +332,10 @@ mly_attach(device_t dev)
     /* enable interrupts now */
     MLY_UNMASK_INTERRUPTS(sc);
 
+#ifdef MLY_DEBUG
+    timeout((timeout_t *)mly_timeout, sc, MLY_CMD_TIMEOUT * hz);
+#endif
+
  out:
     if (error != 0)
 	mly_free(sc);
@@ -1473,6 +1479,10 @@ mly_start(struct mly_command *mc)
     mly_map_command(mc);
     mc->mc_packet->generic.command_id = mc->mc_slot;
 
+#ifdef MLY_DEBUG
+    mc->mc_timestamp = time_second;
+#endif
+
     s = splcam();
 
     /*
@@ -2173,6 +2183,7 @@ mly_cam_action_io(struct cam_sim *sim, struct ccb_scsiio *csio)
     struct mly_command_scsi_small	*ss;
     int					bus, target;
     int					error;
+    int					s;
 
     bus = cam_sim_bus(sim);
     target = csio->ccb_h.target_id;
@@ -2231,8 +2242,11 @@ mly_cam_action_io(struct cam_sim *sim, struct ccb_scsiio *csio)
      * Get a command, or push the ccb back to CAM and freeze the queue.
      */
     if ((error = mly_alloc_command(sc, &mc))) {
+	s = splcam();
 	xpt_freeze_simq(sim, 1);
 	csio->ccb_h.status |= CAM_REQUEUE_REQ;
+	sc->mly_qfrzn_cnt++;
+	splx(s);
 	return(error);
     }
     
@@ -2276,8 +2290,11 @@ mly_cam_action_io(struct cam_sim *sim, struct ccb_scsiio *csio)
 
     /* give the command to the controller */
     if ((error = mly_start(mc))) {
+	s = splcam();
 	xpt_freeze_simq(sim, 1);
 	csio->ccb_h.status |= CAM_REQUEUE_REQ;
+	sc->mly_qfrzn_cnt++;
+	splx(s);
 	return(error);
     }
 
@@ -2309,6 +2326,7 @@ mly_cam_complete(struct mly_command *mc)
     struct mly_btl		*btl;
     u_int8_t			cmd;
     int				bus, target;
+    int				s;
 
     debug_called(2);
 
@@ -2360,6 +2378,14 @@ mly_cam_complete(struct mly_command *mc)
 	csio->ccb_h.status = CAM_REQ_CMP_ERR;
 	break;
     }
+
+    s = splcam();
+    if (sc->mly_qfrzn_cnt) {
+	csio->ccb_h.status |= CAM_RELEASE_SIMQ;
+	sc->mly_qfrzn_cnt--;
+    }
+    splx(s);
+
     xpt_done((union ccb *)csio);
     mly_release_command(mc);
 }
@@ -2945,3 +2971,23 @@ mly_user_health(struct mly_softc *sc, struct mly_user_health *uh)
 		    sizeof(uh->HealthStatusBuffer));
     return(error);
 }
+
+static int
+mly_timeout(struct mly_softc *sc)
+{
+	struct mly_command *mc;
+	int deadline;
+
+	deadline = time_second - MLY_CMD_TIMEOUT;
+	TAILQ_FOREACH(mc, &sc->mly_busy, mc_link) {
+		if ((mc->mc_timestamp < deadline)) {
+			device_printf(sc->mly_dev,
+			    "COMMAND %p TIMEOUT AFTER %d SECONDS\n", mc,
+			    (int)(time_second - mc->mc_timestamp));
+		}
+	}
+
+	timeout((timeout_t *)mly_timeout, sc, MLY_CMD_TIMEOUT * hz);
+
+	return (0);
+}