]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
scsi: fix eh wakeup (scsi_schedule_eh vs scsi_restart_operations)
authorDan Williams <dan.j.williams@intel.com>
Fri, 6 Apr 2012 23:35:36 +0000 (16:35 -0700)
committerDan Williams <dan.j.williams@intel.com>
Wed, 11 Apr 2012 02:15:23 +0000 (19:15 -0700)
Rapid ata hotplug on a libsas controller results in cases where libsas
is waiting indefinitely on eh to perform an ata probe.

A race exists between scsi_schedule_eh() and scsi_restart_operations()
in the case when scsi_restart_operations() issues i/o to other devices
in the sas domain.  When this happens the host state transitions from
SHOST_RECOVERY (set by scsi_schedule_eh) back to SHOST_RUNNING and
->host_busy is non-zero so we put the eh thread to sleep even though
->host_eh_scheduled is active.

Before putting the error handler to sleep we need to check if the
host_state needs to return to SHOST_RECOVERY for another trip through
eh.

Cc: Tejun Heo <tj@kernel.org>
Reported-by: Tom Jackson <thomas.p.jackson@intel.com>
Tested-by: Tom Jackson <thomas.p.jackson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/scsi/scsi_error.c

index 2cfcbffa41fd559b075e7c75eae57e2c7f637488..0945d47907cebb4a3fd5e665e7d53e56e8ebf127 100644 (file)
@@ -1687,6 +1687,20 @@ static void scsi_restart_operations(struct Scsi_Host *shost)
         * requests are started.
         */
        scsi_run_host_queues(shost);
+
+       /*
+        * if eh is active and host_eh_scheduled is pending we need to re-run
+        * recovery.  we do this check after scsi_run_host_queues() to allow
+        * everything pent up since the last eh run a chance to make forward
+        * progress before we sync again.  Either we'll immediately re-run
+        * recovery or scsi_device_unbusy() will wake us again when these
+        * pending commands complete.
+        */
+       spin_lock_irqsave(shost->host_lock, flags);
+       if (shost->host_eh_scheduled)
+               if (scsi_host_set_state(shost, SHOST_RECOVERY))
+                       WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
+       spin_unlock_irqrestore(shost->host_lock, flags);
 }
 
 /**