[SCSI] qla4xxx: Fix cmd check in qla4xxx_cmd_wait

If the command has timedout then the block layer has called
blk_mark_rq_complete. If qla4xxx_cmd_wait is then called
from qla4xxx_eh_host_reset, we will always fail, because if
the driver calls scsi_done then the the block layer will fail
at blk_complete_request's blk_mark_rq_complete call instead of
calling the normal completion path including the function,
blk_queue_end_tag, which releases the tag.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Vikas Chaudhary <vikas.chaudhary@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>

1 files changed, 7 insertions, 1 deletions

diff --git a/drivers/scsi/qla4xxx/ql4_os.c b/drivers/scsi/qla4xxx/ql4_os.c
index f1c58eb26ba2..f4cd846abf6d 100644
--- a/drivers/scsi/qla4xxx/ql4_os.c
+++ b/drivers/scsi/qla4xxx/ql4_os.c
@@ -881,7 +881,13 @@ static int qla4xxx_cmd_wait(struct scsi_qla_host *ha)
                 /* Find a command that hasn't completed. */
                 for (index = 0; index < ha->host->can_queue; index++) {
                         cmd = scsi_host_find_tag(ha->host, index);
-                        if (cmd != NULL)
+                        /*
+                         * We cannot just check if the index is valid,
+                         * becase if we are run from the scsi eh, then
+                         * the scsi/block layer is going to prevent
+                         * the tag from being released.
+                         */
+                        if (cmd != NULL && CMD_SP(cmd))
                                 break;
                 }
                 spin_unlock_irqrestore(&ha->hardware_lock, flags);