[SCSI] isci: Leave requests alone if already terminating.
authorJeff Skirvin <jeffrey.d.skirvin@intel.com>
Sat, 30 Jul 2011 00:17:05 +0000 (17:17 -0700)
committerJames Bottomley <JBottomley@Parallels.com>
Wed, 24 Aug 2011 02:58:44 +0000 (19:58 -0700)
Instead of immediately completing any request that has a second
termination call made on it, wait for the TC done/abort HW event.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
drivers/scsi/isci/request.c

index b4cf998385b3659ca292386c187a37d327b6beac..b5d3a8c4d3297ebf87e72a8087d5277500abc17d 100644 (file)
@@ -732,12 +732,20 @@ sci_io_request_terminate(struct isci_request *ireq)
                sci_change_state(&ireq->sm, SCI_REQ_ABORTING);
                return SCI_SUCCESS;
        case SCI_REQ_TASK_WAIT_TC_RESP:
+               /* The task frame was already confirmed to have been
+                * sent by the SCU HW.  Since the state machine is
+                * now only waiting for the task response itself,
+                * abort the request and complete it immediately
+                * and don't wait for the task response.
+                */
                sci_change_state(&ireq->sm, SCI_REQ_ABORTING);
                sci_change_state(&ireq->sm, SCI_REQ_COMPLETED);
                return SCI_SUCCESS;
        case SCI_REQ_ABORTING:
-               sci_change_state(&ireq->sm, SCI_REQ_COMPLETED);
-               return SCI_SUCCESS;
+               /* If a request has a termination requested twice, return
+                * a failure indication, since HW confirmation of the first
+                * abort is still outstanding.
+                */
        case SCI_REQ_COMPLETED:
        default:
                dev_warn(&ireq->owning_controller->pdev->dev,