[SCSI] libsas: Don't give scsi_cmnds to the EH if they never made it to the SAS LLDD...

On a system with many SAS targets, it appears possible that a scsi_cmnd
can time out without ever making it to the SAS LLDD or at the same time
that a completion is occurring.  In both of these cases, telling the
LLDD to abort the sas_task makes no sense because the LLDD won't know
about the sas_task; what we really want to do is to increase the timer.
Note that this involves creating another sas_task bit to indicate
whether or not the task has been sent to the LLDD; I could have
implemented this by slightly redefining SAS_TASK_STATE_PENDING, but
this way seems cleaner.

This second version amends the aic94xx portion to set the
TASK_AT_INITIATOR flag for all sas_tasks that were passed to
lldd_execute_task.

Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>

diff --git a/drivers/scsi/aic94xx/aic94xx_task.c b/drivers/scsi/aic94xx/aic94xx_task.c
index d202ed5..e2ad5be 100644 (file)
--- a/drivers/scsi/aic94xx/aic94xx_task.c
+++ b/drivers/scsi/aic94xx/aic94xx_task.c
@@ -349,6 +349,7 @@ Again:
 
        spin_lock_irqsave(&task->task_state_lock, flags);
        task->task_state_flags &= ~SAS_TASK_STATE_PENDING;
+       task->task_state_flags &= ~SAS_TASK_AT_INITIATOR;
        task->task_state_flags |= SAS_TASK_STATE_DONE;
        if (unlikely((task->task_state_flags & SAS_TASK_STATE_ABORTED))) {
                spin_unlock_irqrestore(&task->task_state_lock, flags);
@@ -557,6 +558,7 @@ int asd_execute_task(struct sas_task *task, const int num,
        struct sas_task *t = task;
        struct asd_ascb *ascb = NULL, *a;
        struct asd_ha_struct *asd_ha = task->dev->port->ha->lldd_ha;
+       unsigned long flags;
 
        res = asd_can_queue(asd_ha, num);
        if (res)
@@ -599,6 +601,10 @@ int asd_execute_task(struct sas_task *task, const int num,
                }
                if (res)
                        goto out_err_unmap;
+
+               spin_lock_irqsave(&t->task_state_lock, flags);
+               t->task_state_flags |= SAS_TASK_AT_INITIATOR;
+               spin_unlock_irqrestore(&t->task_state_lock, flags);
        }
        list_del_init(&alist);
 
@@ -617,6 +623,9 @@ out_err_unmap:
                        if (a == b)
                                break;
                        t = a->uldd_task;
+                       spin_lock_irqsave(&t->task_state_lock, flags);
+                       t->task_state_flags &= ~SAS_TASK_AT_INITIATOR;
+                       spin_unlock_irqrestore(&t->task_state_lock, flags);
                        switch (t->task_proto) {
                        case SATA_PROTO:
                        case SAS_PROTO_STP:

diff --git a/drivers/scsi/libsas/sas_scsi_host.c b/drivers/scsi/libsas/sas_scsi_host.c
index 22672d5..21bd247 100644 (file)
--- a/drivers/scsi/libsas/sas_scsi_host.c
+++ b/drivers/scsi/libsas/sas_scsi_host.c
@@ -542,6 +542,13 @@ enum scsi_eh_timer_return sas_scsi_timed_out(struct scsi_cmnd *cmd)
                            cmd, task);
                return EH_HANDLED;
        }
+       if (!(task->task_state_flags & SAS_TASK_AT_INITIATOR)) {
+               spin_unlock_irqrestore(&task->task_state_lock, flags);
+               SAS_DPRINTK("command 0x%p, task 0x%p, not at initiator: "
+                           "EH_RESET_TIMER\n",
+                           cmd, task);
+               return EH_RESET_TIMER;
+       }
        task->task_state_flags |= SAS_TASK_STATE_ABORTED;
        spin_unlock_irqrestore(&task->task_state_lock, flags);
 

diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h
index 0c775fc..da96bcf 100644 (file)
--- a/include/scsi/libsas.h
+++ b/include/scsi/libsas.h
@@ -554,6 +554,7 @@ struct sas_task {
 #define SAS_TASK_STATE_DONE         2
 #define SAS_TASK_STATE_ABORTED      4
 #define SAS_TASK_INITIATOR_ABORTED  8
+#define SAS_TASK_AT_INITIATOR       16
 
 static inline struct sas_task *sas_alloc_task(gfp_t flags)
 {