scsi: make asynchronous aborts mandatory
authorHannes Reinecke <hare@suse.de>
Thu, 6 Apr 2017 13:36:35 +0000 (15:36 +0200)
committerMartin K. Petersen <martin.petersen@oracle.com>
Thu, 6 Apr 2017 17:07:33 +0000 (13:07 -0400)
There hasn't been any reports for HBAs where asynchronous abort
would not work, so we should make it mandatory and remove
the fallback.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Documentation/scsi/scsi_eh.txt
drivers/scsi/scsi_error.c
drivers/scsi/scsi_lib.c
drivers/scsi/scsi_priv.h
include/scsi/scsi_host.h

index 4edb9c1cbef52a2692fccf120a58e888f72d8d6e..11e447bdb3a58e6732f4e1f9ce5311c8bd534d6e 100644 (file)
@@ -70,7 +70,7 @@ with the command.
        scmd is requeued to blk queue.
 
  - otherwise
-       scsi_eh_scmd_add(scmd, 0) is invoked for the command.  See
+       scsi_eh_scmd_add(scmd) is invoked for the command.  See
        [1-3] for details of this function.
 
 
@@ -103,9 +103,7 @@ function
         eh_timed_out() callback did not handle the command.
        Step #2 is taken.
 
- 2. If the host supports asynchronous completion (as indicated by the
-    no_async_abort setting in the host template) scsi_abort_command()
-    is invoked to schedule an asynchrous abort.
+ 2. scsi_abort_command() is invoked to schedule an asynchrous abort.
     Asynchronous abort are not invoked for commands which the
     SCSI_EH_ABORT_SCHEDULED flag is set (this indicates that the command
     already had been aborted once, and this is a retry which failed),
@@ -127,16 +125,13 @@ function
 
  scmds enter EH via scsi_eh_scmd_add(), which does the following.
 
- 1. Turns on scmd->eh_eflags as requested.  It's 0 for error
-    completions and SCSI_EH_CANCEL_CMD for timeouts.
+ 1. Links scmd->eh_entry to shost->eh_cmd_q
 
- 2. Links scmd->eh_entry to shost->eh_cmd_q
+ 2. Sets SHOST_RECOVERY bit in shost->shost_state
 
- 3. Sets SHOST_RECOVERY bit in shost->shost_state
+ 3. Increments shost->host_failed
 
- 4. Increments shost->host_failed
-
- 5. Wakes up SCSI EH thread if shost->host_busy == shost->host_failed
+ 4. Wakes up SCSI EH thread if shost->host_busy == shost->host_failed
 
  As can be seen above, once any scmd is added to shost->eh_cmd_q,
 SHOST_RECOVERY shost_state bit is turned on.  This prevents any new
@@ -252,7 +247,6 @@ scmd->allowed.
 
  1. Error completion / time out
     ACTION: scsi_eh_scmd_add() is invoked for scmd
-       - set scmd->eh_eflags
        - add scmd to shost->eh_cmd_q
        - set SHOST_RECOVERY
        - shost->host_failed++
index bbc431897606060d46c3cc1bc6b59a901f4f5d91..53e334356f31bc3eab742ad65da3f1953c12e319 100644 (file)
@@ -162,7 +162,7 @@ scmd_eh_abort_handler(struct work_struct *work)
                }
        }
 
-       scsi_eh_scmd_add(scmd, 0);
+       scsi_eh_scmd_add(scmd);
 }
 
 /**
@@ -221,9 +221,8 @@ static void scsi_eh_reset(struct scsi_cmnd *scmd)
 /**
  * scsi_eh_scmd_add - add scsi cmd to error handling.
  * @scmd:      scmd to run eh on.
- * @eh_flag:   optional SCSI_EH flag.
  */
-void scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
+void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
 {
        struct Scsi_Host *shost = scmd->device->host;
        unsigned long flags;
@@ -239,9 +238,6 @@ void scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
        if (shost->eh_deadline != -1 && !shost->last_reset)
                shost->last_reset = jiffies;
 
-       if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
-               eh_flag &= ~SCSI_EH_CANCEL_CMD;
-       scmd->eh_eflags |= eh_flag;
        scsi_eh_reset(scmd);
        list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
        shost->host_failed++;
@@ -275,10 +271,9 @@ enum blk_eh_timer_return scsi_times_out(struct request *req)
                rtn = host->hostt->eh_timed_out(scmd);
 
        if (rtn == BLK_EH_NOT_HANDLED) {
-               if (host->hostt->no_async_abort ||
-                   scsi_abort_command(scmd) != SUCCESS) {
+               if (scsi_abort_command(scmd) != SUCCESS) {
                        set_host_byte(scmd, DID_TIME_OUT);
-                       scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD);
+                       scsi_eh_scmd_add(scmd);
                }
        }
 
@@ -331,7 +326,7 @@ 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 (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
+                               if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
                                        ++cmd_cancel;
                                else
                                        ++cmd_failed;
@@ -1154,8 +1149,7 @@ int scsi_eh_get_sense(struct list_head *work_q,
         * should not get sense.
         */
        list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
-               if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
-                   (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
+               if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
                    SCSI_SENSE_VALID(scmd))
                        continue;
 
@@ -1295,61 +1289,6 @@ static int scsi_eh_test_devices(struct list_head *cmd_list,
        return list_empty(work_q);
 }
 
-
-/**
- * scsi_eh_abort_cmds - abort pending commands.
- * @work_q:    &list_head for pending commands.
- * @done_q:    &list_head for processed commands.
- *
- * Decription:
- *    Try and see whether or not it makes sense to try and abort the
- *    running command.  This only works out to be the case if we have one
- *    command that has timed out.  If the command simply failed, it makes
- *    no sense to try and abort the command, since as far as the shost
- *    adapter is concerned, it isn't running.
- */
-static int scsi_eh_abort_cmds(struct list_head *work_q,
-                             struct list_head *done_q)
-{
-       struct scsi_cmnd *scmd, *next;
-       LIST_HEAD(check_list);
-       int rtn;
-       struct Scsi_Host *shost;
-
-       list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
-               if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
-                       continue;
-               shost = scmd->device->host;
-               if (scsi_host_eh_past_deadline(shost)) {
-                       list_splice_init(&check_list, work_q);
-                       SCSI_LOG_ERROR_RECOVERY(3,
-                               scmd_printk(KERN_INFO, scmd,
-                                           "%s: skip aborting cmd, past eh deadline\n",
-                                           current->comm));
-                       return list_empty(work_q);
-               }
-               SCSI_LOG_ERROR_RECOVERY(3,
-                       scmd_printk(KERN_INFO, scmd,
-                                    "%s: aborting cmd\n", current->comm));
-               rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
-               if (rtn == FAILED) {
-                       SCSI_LOG_ERROR_RECOVERY(3,
-                               scmd_printk(KERN_INFO, scmd,
-                                           "%s: aborting cmd failed\n",
-                                            current->comm));
-                       list_splice_init(&check_list, work_q);
-                       return list_empty(work_q);
-               }
-               scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
-               if (rtn == FAST_IO_FAIL)
-                       scsi_eh_finish_cmd(scmd, done_q);
-               else
-                       list_move_tail(&scmd->eh_entry, &check_list);
-       }
-
-       return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
-}
-
 /**
  * scsi_eh_try_stu - Send START_UNIT to device.
  * @scmd:      &scsi_cmnd to send START_UNIT
@@ -1692,11 +1631,6 @@ static void scsi_eh_offline_sdevs(struct list_head *work_q,
                sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
                            "not ready after error recovery\n");
                scsi_device_set_state(scmd->device, SDEV_OFFLINE);
-               if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
-                       /*
-                        * FIXME: Handle lost cmds.
-                        */
-               }
                scsi_eh_finish_cmd(scmd, done_q);
        }
        return;
@@ -2140,8 +2074,7 @@ static void scsi_unjam_host(struct Scsi_Host *shost)
        SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
 
        if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
-               if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
-                       scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
+               scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
 
        spin_lock_irqsave(shost->host_lock, flags);
        if (shost->eh_deadline != -1)
index ba8da904e774dc59f50e9cd6a68ae3169b8c713e..9822fdeed0ced52c852e148625d3f5382c12ecbe 100644 (file)
@@ -1593,7 +1593,7 @@ static void scsi_softirq_done(struct request *rq)
                        scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
                        break;
                default:
-                       scsi_eh_scmd_add(cmd, 0);
+                       scsi_eh_scmd_add(cmd);
                        break;
        }
 }
index 5be6cbf69df64d49f9d7a8de4fe1659b097fbbf9..e20ab10623fbc9e54e7532978c81ac76be384975 100644 (file)
@@ -18,7 +18,6 @@ struct scsi_nl_hdr;
 /*
  * Scsi Error Handler Flags
  */
-#define SCSI_EH_CANCEL_CMD     0x0001  /* Cancel this cmd */
 #define SCSI_EH_ABORT_SCHEDULED        0x0002  /* Abort has been scheduled */
 
 #define SCSI_SENSE_VALID(scmd) \
@@ -72,7 +71,7 @@ extern enum blk_eh_timer_return scsi_times_out(struct request *req);
 extern int scsi_error_handler(void *host);
 extern int scsi_decide_disposition(struct scsi_cmnd *cmd);
 extern void scsi_eh_wakeup(struct Scsi_Host *shost);
-extern void scsi_eh_scmd_add(struct scsi_cmnd *, int);
+extern void scsi_eh_scmd_add(struct scsi_cmnd *);
 void scsi_eh_ready_devs(struct Scsi_Host *shost,
                        struct list_head *work_q,
                        struct list_head *done_q);
index 3cd8c3bec6384e02328bbb31aafb9aec84f78c37..afb04811b7b9c578f24d8360ace4f85efa07632c 100644 (file)
@@ -451,11 +451,6 @@ struct scsi_host_template {
        /* True if the controller does not support WRITE SAME */
        unsigned no_write_same:1;
 
-       /*
-        * True if asynchronous aborts are not supported
-        */
-       unsigned no_async_abort:1;
-
        /*
         * Countdown for host blocking with no commands outstanding.
         */