[SCSI] aic94xx: fix REQ_TASK_ABORT and REQ_DEVICE_RESET
James Bottomley [Fri, 15 Feb 2008 15:28:43 +0000 (09:28 -0600)]
This driver has been failing under heavy load with

aic94xx: escb_tasklet_complete: REQ_TASK_ABORT, reason=0x6
aic94xx: escb_tasklet_complete: Can't find task (tc=4) to abort!

The second message is because the driver fails to identify the task
it's being asked to abort.  On closer inpection, there's a thinko in
the for each task loop over pending tasks in both the REQ_TASK_ABORT
and REQ_DEVICE_RESET cases where it doesn't look at the task on the
pending list but at the one on the ESCB (which is always NULL).

Fix by looking at the right task.  Also add a print for the case where
the pending SCB doesn't have a task attached.

Not sure if this will fix all the problems, but it's a definite first
step.

Cc: Stable Tree <stable@kernel.org>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

drivers/scsi/aic94xx/aic94xx_scb.c

index 0febad4..ab35050 100644 (file)
@@ -458,13 +458,19 @@ static void escb_tasklet_complete(struct asd_ascb *ascb,
                tc_abort = le16_to_cpu(tc_abort);
 
                list_for_each_entry_safe(a, b, &asd_ha->seq.pend_q, list) {
-                       struct sas_task *task = ascb->uldd_task;
+                       struct sas_task *task = a->uldd_task;
+
+                       if (a->tc_index != tc_abort)
+                               continue;
 
-                       if (task && a->tc_index == tc_abort) {
+                       if (task) {
                                failed_dev = task->dev;
                                sas_task_abort(task);
-                               break;
+                       } else {
+                               ASD_DPRINTK("R_T_A for non TASK scb 0x%x\n",
+                                           a->scb->header.opcode);
                        }
+                       break;
                }
 
                if (!failed_dev) {
@@ -478,7 +484,7 @@ static void escb_tasklet_complete(struct asd_ascb *ascb,
                 * that the EH will wake up and do something.
                 */
                list_for_each_entry_safe(a, b, &asd_ha->seq.pend_q, list) {
-                       struct sas_task *task = ascb->uldd_task;
+                       struct sas_task *task = a->uldd_task;
 
                        if (task &&
                            task->dev == failed_dev &&