Merge with /pub/scm/linux/kernel/git/torvalds/linux-2.6.git
[pandora-kernel.git] / drivers / scsi / scsi_error.c
index 0c5b02d..a2333d2 100644 (file)
@@ -416,44 +416,21 @@ static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
        return FAILED;
 }
 
-/**
- * scsi_eh_times_out - timeout function for error handling.
- * @scmd:      Cmd that is timing out.
- *
- * Notes:
- *    During error handling, the kernel thread will be sleeping waiting
- *    for some action to complete on the device.  our only job is to
- *    record that it timed out, and to wake up the thread.
- **/
-static void scsi_eh_times_out(struct scsi_cmnd *scmd)
-{
-       scmd->eh_eflags |= SCSI_EH_REC_TIMEOUT;
-       SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__,
-                                         scmd));
-
-       up(scmd->device->host->eh_action);
-}
-
 /**
  * scsi_eh_done - Completion function for error handling.
  * @scmd:      Cmd that is done.
  **/
 static void scsi_eh_done(struct scsi_cmnd *scmd)
 {
-       /*
-        * if the timeout handler is already running, then just set the
-        * flag which says we finished late, and return.  we have no
-        * way of stopping the timeout handler from running, so we must
-        * always defer to it.
-        */
-       if (del_timer(&scmd->eh_timeout)) {
-               scmd->request->rq_status = RQ_SCSI_DONE;
+       struct completion     *eh_action;
 
-               SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
-                                          __FUNCTION__, scmd, scmd->result));
+       SCSI_LOG_ERROR_RECOVERY(3,
+               printk("%s scmd: %p result: %x\n",
+                       __FUNCTION__, scmd, scmd->result));
 
-               up(scmd->device->host->eh_action);
-       }
+       eh_action = scmd->device->host->eh_action;
+       if (eh_action)
+               complete(eh_action);
 }
 
 /**
@@ -461,10 +438,6 @@ static void scsi_eh_done(struct scsi_cmnd *scmd)
  * @scmd:      SCSI Cmd to send.
  * @timeout:   Timeout for cmd.
  *
- * Notes:
- *    The initialization of the structures is quite a bit different in
- *    this case, and furthermore, there is a different completion handler
- *    vs scsi_dispatch_cmd.
  * Return value:
  *    SUCCESS or FAILED or NEEDS_RETRY
  **/
@@ -472,24 +445,16 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
 {
        struct scsi_device *sdev = scmd->device;
        struct Scsi_Host *shost = sdev->host;
-       DECLARE_MUTEX_LOCKED(sem);
+       DECLARE_COMPLETION(done);
+       unsigned long timeleft;
        unsigned long flags;
-       int rtn = SUCCESS;
+       int rtn;
 
-       /*
-        * we will use a queued command if possible, otherwise we will
-        * emulate the queuing and calling of completion function ourselves.
-        */
        if (sdev->scsi_level <= SCSI_2)
                scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
                        (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.
-        */
-       shost->eh_action = &sem;
+       shost->eh_action = &done;
        scmd->request->rq_status = RQ_SCSI_BUSY;
 
        spin_lock_irqsave(shost->host_lock, flags);
@@ -497,47 +462,29 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
        shost->hostt->queuecommand(scmd, scsi_eh_done);
        spin_unlock_irqrestore(shost->host_lock, flags);
 
-       down(&sem);
-       scsi_log_completion(scmd, SUCCESS);
+       timeleft = wait_for_completion_timeout(&done, timeout);
 
+       scmd->request->rq_status = RQ_SCSI_DONE;
        shost->eh_action = NULL;
 
-       /*
-        * see if timeout.  if so, tell the host to forget about it.
-        * in other words, we don't want a callback any more.
-        */
-       if (scmd->eh_eflags & SCSI_EH_REC_TIMEOUT) {
-               scmd->eh_eflags &= ~SCSI_EH_REC_TIMEOUT;
-
-               /*
-                * as far as the low level driver is
-                * concerned, this command is still active, so
-                * we must give the low level driver a chance
-                * to abort it. (db) 
-                *
-                * FIXME(eric) - we are not tracking whether we could
-                * abort a timed out command or not.  not sure how
-                * we should treat them differently anyways.
-                */
-               if (shost->hostt->eh_abort_handler)
-                       shost->hostt->eh_abort_handler(scmd);
-                       
-               scmd->request->rq_status = RQ_SCSI_DONE;
-               rtn = FAILED;
-       }
+       scsi_log_completion(scmd, SUCCESS);
 
-       SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
-                                         __FUNCTION__, scmd, rtn));
+       SCSI_LOG_ERROR_RECOVERY(3,
+               printk("%s: scmd: %p, timeleft: %ld\n",
+                       __FUNCTION__, scmd, timeleft));
 
        /*
-        * now examine the actual status codes to see whether the command
-        * actually did complete normally.
+        * If there is time left scsi_eh_done got called, and we will
+        * examine the actual status codes to see whether the command
+        * actually did complete normally, else tell the host to forget
+        * about this command.
         */
-       if (rtn == SUCCESS) {
+       if (timeleft) {
                rtn = scsi_eh_completed_normally(scmd);
                SCSI_LOG_ERROR_RECOVERY(3,
                        printk("%s: scsi_eh_completed_normally %x\n",
                               __FUNCTION__, rtn));
+
                switch (rtn) {
                case SUCCESS:
                case NEEDS_RETRY:
@@ -547,6 +494,15 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
                        rtn = FAILED;
                        break;
                }
+       } else {
+               /*
+                * FIXME(eric) - we are not tracking whether we could
+                * abort a timed out command or not.  not sure how
+                * we should treat them differently anyways.
+                */
+               if (shost->hostt->eh_abort_handler)
+                       shost->hostt->eh_abort_handler(scmd);
+               rtn = FAILED;
        }
 
        return rtn;
@@ -1363,23 +1319,6 @@ int scsi_decide_disposition(struct scsi_cmnd *scmd)
        }
 }
 
-/**
- * scsi_eh_lock_done - done function for eh door lock request
- * @scmd:      SCSI command block for the door lock request
- *
- * Notes:
- *     We completed the asynchronous door lock request, and it has either
- *     locked the door or failed.  We must free the command structures
- *     associated with this request.
- **/
-static void scsi_eh_lock_done(struct scsi_cmnd *scmd)
-{
-       struct scsi_request *sreq = scmd->sc_request;
-
-       scsi_release_request(sreq);
-}
-
-
 /**
  * scsi_eh_lock_door - Prevent medium removal for the specified device
  * @sdev:      SCSI device to prevent medium removal
@@ -1402,29 +1341,17 @@ static void scsi_eh_lock_done(struct scsi_cmnd *scmd)
  **/
 static void scsi_eh_lock_door(struct scsi_device *sdev)
 {
-       struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL);
+       unsigned char cmnd[MAX_COMMAND_SIZE];
 
-       if (unlikely(!sreq)) {
-               printk(KERN_ERR "%s: request allocate failed,"
-                      "prevent media removal cmd not sent\n", __FUNCTION__);
-               return;
-       }
+       cmnd[0] = ALLOW_MEDIUM_REMOVAL;
+       cmnd[1] = 0;
+       cmnd[2] = 0;
+       cmnd[3] = 0;
+       cmnd[4] = SCSI_REMOVAL_PREVENT;
+       cmnd[5] = 0;
 
-       sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL;
-       sreq->sr_cmnd[1] = 0;
-       sreq->sr_cmnd[2] = 0;
-       sreq->sr_cmnd[3] = 0;
-       sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT;
-       sreq->sr_cmnd[5] = 0;
-       sreq->sr_data_direction = DMA_NONE;
-       sreq->sr_bufflen = 0;
-       sreq->sr_buffer = NULL;
-       sreq->sr_allowed = 5;
-       sreq->sr_done = scsi_eh_lock_done;
-       sreq->sr_timeout_per_command = 10 * HZ;
-       sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]);
-
-       scsi_insert_special_req(sreq, 1);
+       scsi_execute_async(sdev, cmnd, DMA_NONE, NULL, 0, 0, 10 * HZ,
+                          5, NULL, NULL, GFP_KERNEL);
 }
 
 
@@ -1571,50 +1498,41 @@ static void scsi_unjam_host(struct Scsi_Host *shost)
 }
 
 /**
- * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
+ * scsi_error_handler - SCSI error handler thread
  * @data:      Host for which we are running.
  *
  * Notes:
- *    This is always run in the context of a kernel thread.  The idea is
- *    that we start this thing up when the kernel starts up (one per host
- *    that we detect), and it immediately goes to sleep and waits for some
- *    event (i.e. failure).  When this takes place, we have the job of
- *    trying to unjam the bus and restarting things.
+ *    This is the main error handling loop.  This is run as a kernel thread
+ *    for every SCSI host and handles all error handling activity.
  **/
 int scsi_error_handler(void *data)
 {
-       struct Scsi_Host *shost = (struct Scsi_Host *) data;
-       int rtn;
+       struct Scsi_Host *shost = data;
 
        current->flags |= PF_NOFREEZE;
 
-       
        /*
-        * Note - we always use TASK_INTERRUPTIBLE even if the module
-        * was loaded as part of the kernel.  The reason is that
-        * UNINTERRUPTIBLE would cause this thread to be counted in
-        * the load average as a running process, and an interruptible
-        * wait doesn't.
+        * We use TASK_INTERRUPTIBLE so that the thread is not
+        * counted against the load average as a running process.
+        * We never actually get interrupted because kthread_run
+        * disables singal delivery for the created thread.
         */
        set_current_state(TASK_INTERRUPTIBLE);
        while (!kthread_should_stop()) {
                if (shost->host_failed == 0 ||
                    shost->host_failed != shost->host_busy) {
-                       SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
-                                                         " scsi_eh_%d"
-                                                         " sleeping\n",
-                                                         shost->host_no));
+                       SCSI_LOG_ERROR_RECOVERY(1,
+                               printk("Error handler scsi_eh_%d sleeping\n",
+                                       shost->host_no));
                        schedule();
                        set_current_state(TASK_INTERRUPTIBLE);
                        continue;
                }
 
                __set_current_state(TASK_RUNNING);
-               SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
-                                                 " scsi_eh_%d waking"
-                                                 " up\n",shost->host_no));
-
-               shost->eh_active = 1;
+               SCSI_LOG_ERROR_RECOVERY(1,
+                       printk("Error handler scsi_eh_%d waking up\n",
+                               shost->host_no));
 
                /*
                 * We have a host that is failing for some reason.  Figure out
@@ -1622,12 +1540,10 @@ int scsi_error_handler(void *data)
                 * If we fail, we end up taking the thing offline.
                 */
                if (shost->hostt->eh_strategy_handler) 
-                       rtn = shost->hostt->eh_strategy_handler(shost);
+                       shost->hostt->eh_strategy_handler(shost);
                else
                        scsi_unjam_host(shost);
 
-               shost->eh_active = 0;
-
                /*
                 * Note - if the above fails completely, the action is to take
                 * individual devices offline and flush the queue of any
@@ -1638,15 +1554,10 @@ int scsi_error_handler(void *data)
                scsi_restart_operations(shost);
                set_current_state(TASK_INTERRUPTIBLE);
        }
-
        __set_current_state(TASK_RUNNING);
 
-       SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
-                                         " exiting\n",shost->host_no));
-
-       /*
-        * Make sure that nobody tries to wake us up again.
-        */
+       SCSI_LOG_ERROR_RECOVERY(1,
+               printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
        shost->ehandler = NULL;
        return 0;
 }