]> err.no Git - linux-2.6/commitdiff
[PATCH] scsi: scsi_send_eh_cmnd() cleanup
author <htejun@gmail.com> <>
Sun, 17 Apr 2005 20:00:23 +0000 (15:00 -0500)
committerJames Bottomley <jejb@titanic>
Mon, 18 Apr 2005 17:35:56 +0000 (12:35 -0500)
This patch makes scsi_send_eh_cmnd() use sdev and shost instead of
referencing them through scmd-> everytime.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
drivers/scsi/scsi_error.c

index 203a0812508a1962c2c9d58020d3a6a9bc61aaf5..1a135f38e78d8965d9a9fab95ca58614e7db61a3 100644 (file)
@@ -476,7 +476,8 @@ static void scsi_eh_done(struct scsi_cmnd *scmd)
  **/
 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
 {
-       struct Scsi_Host *host = scmd->device->host;
+       struct scsi_device *sdev = scmd->device;
+       struct Scsi_Host *shost = sdev->host;
        DECLARE_MUTEX_LOCKED(sem);
        unsigned long flags;
        int rtn = SUCCESS;
@@ -487,27 +488,27 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
         */
        scmd->owner = SCSI_OWNER_LOWLEVEL;
 
-       if (scmd->device->scsi_level <= SCSI_2)
+       if (sdev->scsi_level <= SCSI_2)
                scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
-                       (scmd->device->lun << 5 & 0xe0);
+                       (sdev->lun << 5 & 0xe0);
 
        scsi_add_timer(scmd, timeout, scsi_eh_times_out);
 
        /*
         * set up the semaphore so we wait for the command to complete.
         */
-       scmd->device->host->eh_action = &sem;
+       shost->eh_action = &sem;
        scmd->request->rq_status = RQ_SCSI_BUSY;
 
-       spin_lock_irqsave(scmd->device->host->host_lock, flags);
+       spin_lock_irqsave(shost->host_lock, flags);
        scsi_log_send(scmd);
-       host->hostt->queuecommand(scmd, scsi_eh_done);
-       spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
+       shost->hostt->queuecommand(scmd, scsi_eh_done);
+       spin_unlock_irqrestore(shost->host_lock, flags);
 
        down(&sem);
        scsi_log_completion(scmd, SUCCESS);
 
-       scmd->device->host->eh_action = NULL;
+       shost->eh_action = NULL;
 
        /*
         * see if timeout.  if so, tell the host to forget about it.
@@ -527,10 +528,10 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
                 * abort a timed out command or not.  not sure how
                 * we should treat them differently anyways.
                 */
-               spin_lock_irqsave(scmd->device->host->host_lock, flags);
-               if (scmd->device->host->hostt->eh_abort_handler)
-                       scmd->device->host->hostt->eh_abort_handler(scmd);
-               spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
+               spin_lock_irqsave(shost->host_lock, flags);
+               if (shost->hostt->eh_abort_handler)
+                       shost->hostt->eh_abort_handler(scmd);
+               spin_unlock_irqrestore(shost->host_lock, flags);
                        
                scmd->request->rq_status = RQ_SCSI_DONE;
                scmd->owner = SCSI_OWNER_ERROR_HANDLER;