[SCSI] ipr: Cleanup error structures
[linux-2.6] / drivers / scsi / scsi_error.c
index ceb4e0c..18c5d25 100644 (file)
@@ -20,6 +20,7 @@
 #include <linux/string.h>
 #include <linux/slab.h>
 #include <linux/kernel.h>
+#include <linux/kthread.h>
 #include <linux/interrupt.h>
 #include <linux/blkdev.h>
 #include <linux/delay.h>
@@ -49,7 +50,7 @@
 void scsi_eh_wakeup(struct Scsi_Host *shost)
 {
        if (shost->host_busy == shost->host_failed) {
-               up(shost->eh_wait);
+               wake_up_process(shost->ehandler);
                SCSI_LOG_ERROR_RECOVERY(5,
                                printk("Waking error handler thread\n"));
        }
@@ -67,24 +68,24 @@ int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
 {
        struct Scsi_Host *shost = scmd->device->host;
        unsigned long flags;
+       int ret = 0;
 
-       if (shost->eh_wait == NULL)
+       if (!shost->ehandler)
                return 0;
 
        spin_lock_irqsave(shost->host_lock, flags);
+       if (scsi_host_set_state(shost, SHOST_RECOVERY))
+               if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
+                       goto out_unlock;
 
-       scsi_eh_eflags_set(scmd, eh_flag);
-       /*
-        * FIXME: Can we stop setting owner and state.
-        */
-       scmd->owner = SCSI_OWNER_ERROR_HANDLER;
-       scmd->state = SCSI_STATE_FAILED;
+       ret = 1;
+       scmd->eh_eflags |= eh_flag;
        list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
-       set_bit(SHOST_RECOVERY, &shost->shost_state);
        shost->host_failed++;
        scsi_eh_wakeup(shost);
+ out_unlock:
        spin_unlock_irqrestore(shost->host_lock, flags);
-       return 1;
+       return ret;
 }
 
 /**
@@ -120,7 +121,6 @@ void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
 
        add_timer(&scmd->eh_timeout);
 }
-EXPORT_SYMBOL(scsi_add_timer);
 
 /**
  * scsi_delete_timer - Delete/cancel timer for a given function.
@@ -148,7 +148,6 @@ int scsi_delete_timer(struct scsi_cmnd *scmd)
 
        return rtn;
 }
-EXPORT_SYMBOL(scsi_delete_timer);
 
 /**
  * scsi_times_out - Timeout function for normal scsi commands.
@@ -182,8 +181,8 @@ void scsi_times_out(struct scsi_cmnd *scmd)
                }
 
        if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
-               panic("Error handler thread not present at %p %p %s %d",
-                     scmd, scmd->device->host, __FILE__, __LINE__);
+               scmd->result |= DID_TIME_OUT << 16;
+               __scsi_done(scmd);
        }
 }
 
@@ -202,7 +201,7 @@ int scsi_block_when_processing_errors(struct scsi_device *sdev)
 {
        int online;
 
-       wait_event(sdev->host->host_wait, (!test_bit(SHOST_RECOVERY, &sdev->host->shost_state)));
+       wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 
        online = scsi_device_online(sdev);
 
@@ -233,8 +232,7 @@ static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
                list_for_each_entry(scmd, work_q, eh_entry) {
                        if (scmd->device == sdev) {
                                ++total_failures;
-                               if (scsi_eh_eflags_chk(scmd,
-                                                      SCSI_EH_CANCEL_CMD))
+                               if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
                                        ++cmd_cancel;
                                else 
                                        ++cmd_failed;
@@ -243,11 +241,10 @@ static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 
                if (cmd_cancel || cmd_failed) {
                        SCSI_LOG_ERROR_RECOVERY(3,
-                               printk("%s: %d:%d:%d:%d cmds failed: %d,"
-                                      " cancel: %d\n",
-                                      __FUNCTION__, shost->host_no,
-                                      sdev->channel, sdev->id, sdev->lun,
-                                      cmd_failed, cmd_cancel));
+                               sdev_printk(KERN_INFO, sdev,
+                                           "%s: cmds failed: %d, cancel: %d\n",
+                                           __FUNCTION__, cmd_failed,
+                                           cmd_cancel));
                        cmd_cancel = 0;
                        cmd_failed = 0;
                        ++devices_failed;
@@ -419,45 +416,16 @@ 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)
-{
-       scsi_eh_eflags_set(scmd, 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;
-               scmd->owner = SCSI_OWNER_ERROR_HANDLER;
-
-               SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
-                                          __FUNCTION__, scmd, scmd->result));
-
-               up(scmd->device->host->eh_action);
-       }
+       SCSI_LOG_ERROR_RECOVERY(3,
+               printk("%s scmd: %p result: %x\n",
+                       __FUNCTION__, scmd, scmd->result));
+       complete(scmd->device->host->eh_action);
 }
 
 /**
@@ -465,10 +433,6 @@ 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
  **/
@@ -476,26 +440,16 @@ 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;
-
-       /*
-        * we will use a queued command if possible, otherwise we will
-        * emulate the queuing and calling of completion function ourselves.
-        */
-       scmd->owner = SCSI_OWNER_LOWLEVEL;
+       int rtn;
 
        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);
@@ -503,50 +457,29 @@ 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 (scsi_eh_eflags_chk(scmd, SCSI_EH_REC_TIMEOUT)) {
-               scsi_eh_eflags_clr(scmd,  SCSI_EH_REC_TIMEOUT);
-               scmd->owner = SCSI_OWNER_LOWLEVEL;
-
-               /*
-                * 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;
-               scmd->owner = SCSI_OWNER_ERROR_HANDLER;
-                       
-               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:
@@ -556,6 +489,15 @@ 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;
@@ -641,9 +583,7 @@ static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd,
                               struct list_head *done_q)
 {
        scmd->device->host->host_failed--;
-       scmd->state = SCSI_STATE_BHQUEUE;
-
-       scsi_eh_eflags_clr_all(scmd);
+       scmd->eh_eflags = 0;
 
        /*
         * set this back so that the upper level can correctly free up
@@ -676,20 +616,17 @@ static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd,
 static int scsi_eh_get_sense(struct list_head *work_q,
                             struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd;
+       struct scsi_cmnd *scmd, *next;
        int rtn;
 
-       list_for_each_safe(lh, lh_sf, work_q) {
-               scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
-               if (scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD) ||
+       list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
+               if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
                    SCSI_SENSE_VALID(scmd))
                        continue;
 
-               SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
-                                                 " for id: %d\n",
-                                                 current->comm,
-                                                 scmd->device->id));
+               SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
+                                                 "%s: requesting sense\n",
+                                                 current->comm));
                rtn = scsi_request_sense(scmd);
                if (rtn != SUCCESS)
                        continue;
@@ -742,9 +679,6 @@ static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
         */
        if (scmd->serial_number == 0)
                return SUCCESS;
-
-       scmd->owner = SCSI_OWNER_LOWLEVEL;
-
        return scmd->device->host->hostt->eh_abort_handler(scmd);
 }
 
@@ -794,9 +728,11 @@ retry_tur:
                __FUNCTION__, scmd, rtn));
        if (rtn == SUCCESS)
                return 0;
-       else if (rtn == NEEDS_RETRY)
+       else if (rtn == NEEDS_RETRY) {
                if (retry_cnt--)
                        goto retry_tur;
+               return 0;
+       }
        return 1;
 }
 
@@ -815,20 +751,18 @@ retry_tur:
 static int scsi_eh_abort_cmds(struct list_head *work_q,
                              struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd;
+       struct scsi_cmnd *scmd, *next;
        int rtn;
 
-       list_for_each_safe(lh, lh_sf, work_q) {
-               scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
-               if (!scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD))
+       list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
+               if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
                        continue;
                SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
                                                  "0x%p\n", current->comm,
                                                  scmd));
                rtn = scsi_try_to_abort_cmd(scmd);
                if (rtn == SUCCESS) {
-                       scsi_eh_eflags_clr(scmd,  SCSI_EH_CANCEL_CMD);
+                       scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
                        if (!scsi_device_online(scmd->device) ||
                            !scsi_eh_tur(scmd)) {
                                scsi_eh_finish_cmd(scmd, done_q);
@@ -862,10 +796,7 @@ static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
        if (!scmd->device->host->hostt->eh_device_reset_handler)
                return FAILED;
 
-       scmd->owner = SCSI_OWNER_LOWLEVEL;
-
        rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
-
        if (rtn == SUCCESS) {
                scmd->device->was_reset = 1;
                scmd->device->expecting_cc_ua = 1;
@@ -938,8 +869,7 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
                              struct list_head *work_q,
                              struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd, *stu_scmd;
+       struct scsi_cmnd *scmd, *stu_scmd, *next;
        struct scsi_device *sdev;
 
        shost_for_each_device(sdev, shost) {
@@ -960,8 +890,8 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
                if (!scsi_eh_try_stu(stu_scmd)) {
                        if (!scsi_device_online(sdev) ||
                            !scsi_eh_tur(stu_scmd)) {
-                               list_for_each_safe(lh, lh_sf, work_q) {
-                                       scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
+                               list_for_each_entry_safe(scmd, next,
+                                                         work_q, eh_entry) {
                                        if (scmd->device == sdev)
                                                scsi_eh_finish_cmd(scmd, done_q);
                                }
@@ -992,8 +922,7 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
                                    struct list_head *work_q,
                                    struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd, *bdr_scmd;
+       struct scsi_cmnd *scmd, *bdr_scmd, *next;
        struct scsi_device *sdev;
        int rtn;
 
@@ -1015,11 +944,8 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
                if (rtn == SUCCESS) {
                        if (!scsi_device_online(sdev) ||
                            !scsi_eh_tur(bdr_scmd)) {
-                               list_for_each_safe(lh, lh_sf,
-                                                  work_q) {
-                                       scmd = list_entry(lh, struct
-                                                         scsi_cmnd,
-                                                         eh_entry);
+                               list_for_each_entry_safe(scmd, next,
+                                                        work_q, eh_entry) {
                                        if (scmd->device == sdev)
                                                scsi_eh_finish_cmd(scmd,
                                                                   done_q);
@@ -1048,7 +974,6 @@ static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
 
        SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
                                          __FUNCTION__));
-       scmd->owner = SCSI_OWNER_LOWLEVEL;
 
        if (!scmd->device->host->hostt->eh_bus_reset_handler)
                return FAILED;
@@ -1059,7 +984,8 @@ static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
                if (!scmd->device->host->hostt->skip_settle_delay)
                        ssleep(BUS_RESET_SETTLE_TIME);
                spin_lock_irqsave(scmd->device->host->host_lock, flags);
-               scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
+               scsi_report_bus_reset(scmd->device->host,
+                                     scmd_channel(scmd));
                spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
        }
 
@@ -1077,7 +1003,6 @@ static int scsi_try_host_reset(struct scsi_cmnd *scmd)
 
        SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
                                          __FUNCTION__));
-       scmd->owner = SCSI_OWNER_LOWLEVEL;
 
        if (!scmd->device->host->hostt->eh_host_reset_handler)
                return FAILED;
@@ -1088,7 +1013,8 @@ static int scsi_try_host_reset(struct scsi_cmnd *scmd)
                if (!scmd->device->host->hostt->skip_settle_delay)
                        ssleep(HOST_RESET_SETTLE_TIME);
                spin_lock_irqsave(scmd->device->host->host_lock, flags);
-               scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
+               scsi_report_bus_reset(scmd->device->host,
+                                     scmd_channel(scmd));
                spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
        }
 
@@ -1104,9 +1030,7 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
                             struct list_head *work_q,
                             struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd;
-       struct scsi_cmnd *chan_scmd;
+       struct scsi_cmnd *scmd, *chan_scmd, *next;
        unsigned int channel;
        int rtn;
 
@@ -1120,7 +1044,7 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
        for (channel = 0; channel <= shost->max_channel; channel++) {
                chan_scmd = NULL;
                list_for_each_entry(scmd, work_q, eh_entry) {
-                       if (channel == scmd->device->channel) {
+                       if (channel == scmd_channel(scmd)) {
                                chan_scmd = scmd;
                                break;
                                /*
@@ -1137,10 +1061,8 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
                                                  channel));
                rtn = scsi_try_bus_reset(chan_scmd);
                if (rtn == SUCCESS) {
-                       list_for_each_safe(lh, lh_sf, work_q) {
-                               scmd = list_entry(lh, struct scsi_cmnd,
-                                                 eh_entry);
-                               if (channel == scmd->device->channel)
+                       list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
+                               if (channel == scmd_channel(scmd))
                                        if (!scsi_device_online(scmd->device) ||
                                            !scsi_eh_tur(scmd))
                                                scsi_eh_finish_cmd(scmd,
@@ -1164,9 +1086,8 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
 static int scsi_eh_host_reset(struct list_head *work_q,
                              struct list_head *done_q)
 {
+       struct scsi_cmnd *scmd, *next;
        int rtn;
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd;
 
        if (!list_empty(work_q)) {
                scmd = list_entry(work_q->next,
@@ -1177,8 +1098,7 @@ static int scsi_eh_host_reset(struct list_head *work_q,
 
                rtn = scsi_try_host_reset(scmd);
                if (rtn == SUCCESS) {
-                       list_for_each_safe(lh, lh_sf, work_q) {
-                               scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
+                       list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
                                if (!scsi_device_online(scmd->device) ||
                                    (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
                                    !scsi_eh_tur(scmd))
@@ -1202,20 +1122,14 @@ static int scsi_eh_host_reset(struct list_head *work_q,
 static void scsi_eh_offline_sdevs(struct list_head *work_q,
                                  struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd;
+       struct scsi_cmnd *scmd, *next;
 
-       list_for_each_safe(lh, lh_sf, work_q) {
-               scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
-               printk(KERN_INFO "scsi: Device offlined - not"
-                               " ready after error recovery: host"
-                               " %d channel %d id %d lun %d\n",
-                               scmd->device->host->host_no,
-                               scmd->device->channel,
-                               scmd->device->id,
-                               scmd->device->lun);
+       list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
+               sdev_printk(KERN_INFO, scmd->device,
+                           "scsi: Device offlined - not"
+                           " ready after error recovery\n");
                scsi_device_set_state(scmd->device, SDEV_OFFLINE);
-               if (scsi_eh_eflags_chk(scmd, SCSI_EH_CANCEL_CMD)) {
+               if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
                        /*
                         * FIXME: Handle lost cmds.
                         */
@@ -1375,10 +1289,8 @@ int scsi_decide_disposition(struct scsi_cmnd *scmd)
                return SUCCESS;
 
        case RESERVATION_CONFLICT:
-               printk(KERN_INFO "scsi: reservation conflict: host"
-                                " %d channel %d id %d lun %d\n",
-                      scmd->device->host->host_no, scmd->device->channel,
-                      scmd->device->id, scmd->device->lun);
+               sdev_printk(KERN_INFO, scmd->device,
+                           "reservation conflict\n");
                return SUCCESS; /* causes immediate i/o error */
        default:
                return FAILED;
@@ -1478,6 +1390,7 @@ static void scsi_eh_lock_door(struct scsi_device *sdev)
 static void scsi_restart_operations(struct Scsi_Host *shost)
 {
        struct scsi_device *sdev;
+       unsigned long flags;
 
        /*
         * If the door was locked, we need to insert a door lock request
@@ -1497,7 +1410,11 @@ static void scsi_restart_operations(struct Scsi_Host *shost)
        SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
                                          __FUNCTION__));
 
-       clear_bit(SHOST_RECOVERY, &shost->shost_state);
+       spin_lock_irqsave(shost->host_lock, flags);
+       if (scsi_host_set_state(shost, SHOST_RUNNING))
+               if (scsi_host_set_state(shost, SHOST_CANCEL))
+                       BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
+       spin_unlock_irqrestore(shost->host_lock, flags);
 
        wake_up(&shost->host_wait);
 
@@ -1534,12 +1451,10 @@ static void scsi_eh_ready_devs(struct Scsi_Host *shost,
  **/
 static void scsi_eh_flush_done_q(struct list_head *done_q)
 {
-       struct list_head *lh, *lh_sf;
-       struct scsi_cmnd *scmd;
+       struct scsi_cmnd *scmd, *next;
 
-       list_for_each_safe(lh, lh_sf, done_q) {
-               scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
-               list_del_init(lh);
+       list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
+               list_del_init(&scmd->eh_entry);
                if (scsi_device_online(scmd->device) &&
                    !blk_noretry_request(scmd->request) &&
                    (++scmd->retries < scmd->allowed)) {
@@ -1607,69 +1522,41 @@ 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;
-       DECLARE_MUTEX_LOCKED(sem);
-
-       /*
-        *    Flush resources
-        */
-
-       daemonize("scsi_eh_%d", shost->host_no);
+       struct Scsi_Host *shost = data;
 
        current->flags |= PF_NOFREEZE;
 
-       shost->eh_wait = &sem;
-       shost->ehandler = current;
-
        /*
-        * Wake up the thread that created us.
+        * 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.
         */
-       SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
-                                         " scsi_eh_%d\n",shost->host_no));
-
-       complete(shost->eh_notify);
-
-       while (1) {
-               /*
-                * If we get a signal, it means we are supposed to go
-                * away and die.  This typically happens if the user is
-                * trying to unload a module.
-                */
-               SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
-                                                 " scsi_eh_%d"
-                                                 " sleeping\n",shost->host_no));
-
-               /*
-                * Note - we always use down_interruptible with the semaphore
-                * even if the module was loaded as part of the kernel.  The
-                * reason is that down() will cause this thread to be counted
-                * in the load average as a running process, and down
-                * interruptible doesn't.  Given that we need to allow this
-                * thread to die if the driver was loaded as a module, using
-                * semaphores isn't unreasonable.
-                */
-               down_interruptible(&sem);
-               if (shost->eh_kill)
-                       break;
-
-               SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
-                                                 " scsi_eh_%d waking"
-                                                 " up\n",shost->host_no));
+       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));
+                       schedule();
+                       set_current_state(TASK_INTERRUPTIBLE);
+                       continue;
+               }
 
-               shost->eh_active = 1;
+               __set_current_state(TASK_RUNNING);
+               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
@@ -1677,12 +1564,10 @@ 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
@@ -1691,32 +1576,13 @@ int scsi_error_handler(void *data)
                 * which are still online.
                 */
                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.
-        */
-       shost->eh_wait = NULL;
-
-       /*
-        * Knock this down too.  From this point on, the host is flying
-        * without a pilot.  If this is because the module is being unloaded,
-        * that's fine.  If the user sent a signal to this thing, we are
-        * potentially in real danger.
-        */
-       shost->eh_active = 0;
+       SCSI_LOG_ERROR_RECOVERY(1,
+               printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
        shost->ehandler = NULL;
-
-       /*
-        * If anyone is waiting for us to exit (i.e. someone trying to unload
-        * a driver), then wake up that process to let them know we are on
-        * the way out the door.
-        */
-       complete_and_exit(shost->eh_notify, 0);
        return 0;
 }
 
@@ -1746,7 +1612,7 @@ void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
        struct scsi_device *sdev;
 
        __shost_for_each_device(sdev, shost) {
-               if (channel == sdev->channel) {
+               if (channel == sdev_channel(sdev)) {
                        sdev->was_reset = 1;
                        sdev->expecting_cc_ua = 1;
                }
@@ -1781,8 +1647,8 @@ void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
        struct scsi_device *sdev;
 
        __shost_for_each_device(sdev, shost) {
-               if (channel == sdev->channel &&
-                   target == sdev->id) {
+               if (channel == sdev_channel(sdev) &&
+                   target == sdev_id(sdev)) {
                        sdev->was_reset = 1;
                        sdev->expecting_cc_ua = 1;
                }
@@ -1818,9 +1684,7 @@ scsi_reset_provider(struct scsi_device *dev, int flag)
        scmd->request = &req;
        memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
        scmd->request->rq_status        = RQ_SCSI_BUSY;
-       scmd->state                     = SCSI_STATE_INITIALIZING;
-       scmd->owner                     = SCSI_OWNER_MIDLEVEL;
-    
+
        memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
     
        scmd->scsi_done         = scsi_reset_provider_done_command;
@@ -1829,7 +1693,6 @@ scsi_reset_provider(struct scsi_device *dev, int flag)
        scmd->bufflen                   = 0;
        scmd->request_buffer            = NULL;
        scmd->request_bufflen           = 0;
-       scmd->abort_reason              = DID_ABORT;
 
        scmd->cmd_len                   = 0;
 
@@ -1890,12 +1753,16 @@ EXPORT_SYMBOL(scsi_reset_provider);
 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
                          struct scsi_sense_hdr *sshdr)
 {
-       if (!sense_buffer || !sb_len || (sense_buffer[0] & 0x70) != 0x70)
+       if (!sense_buffer || !sb_len)
                return 0;
 
        memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
 
        sshdr->response_code = (sense_buffer[0] & 0x7f);
+
+       if (!scsi_sense_valid(sshdr))
+               return 0;
+
        if (sshdr->response_code >= 0x72) {
                /*
                 * descriptor format