]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - drivers/scsi/scsi_error.c
[SCSI] Return NEEDS_RETRY for eh commands with status BUSY
[karo-tx-linux.git] / drivers / scsi / scsi_error.c
1 /*
2  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
3  *
4  *  SCSI error/timeout handling
5  *      Initial versions: Eric Youngdale.  Based upon conversations with
6  *                        Leonard Zubkoff and David Miller at Linux Expo, 
7  *                        ideas originating from all over the place.
8  *
9  *      Restructured scsi_unjam_host and associated functions.
10  *      September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11  *
12  *      Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13  *      minor  cleanups.
14  *      September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15  */
16
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
28
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
37
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
41
42 #include <trace/events/scsi.h>
43
44 #define SENSE_TIMEOUT           (10*HZ)
45
46 /*
47  * These should *probably* be handled by the host itself.
48  * Since it is allowed to sleep, it probably should.
49  */
50 #define BUS_RESET_SETTLE_TIME   (10)
51 #define HOST_RESET_SETTLE_TIME  (10)
52
53 /* called with shost->host_lock held */
54 void scsi_eh_wakeup(struct Scsi_Host *shost)
55 {
56         if (shost->host_busy == shost->host_failed) {
57                 trace_scsi_eh_wakeup(shost);
58                 wake_up_process(shost->ehandler);
59                 SCSI_LOG_ERROR_RECOVERY(5,
60                                 printk("Waking error handler thread\n"));
61         }
62 }
63
64 /**
65  * scsi_schedule_eh - schedule EH for SCSI host
66  * @shost:      SCSI host to invoke error handling on.
67  *
68  * Schedule SCSI EH without scmd.
69  */
70 void scsi_schedule_eh(struct Scsi_Host *shost)
71 {
72         unsigned long flags;
73
74         spin_lock_irqsave(shost->host_lock, flags);
75
76         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
77             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
78                 shost->host_eh_scheduled++;
79                 scsi_eh_wakeup(shost);
80         }
81
82         spin_unlock_irqrestore(shost->host_lock, flags);
83 }
84 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
85
86 /**
87  * scsi_eh_scmd_add - add scsi cmd to error handling.
88  * @scmd:       scmd to run eh on.
89  * @eh_flag:    optional SCSI_EH flag.
90  *
91  * Return value:
92  *      0 on failure.
93  */
94 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
95 {
96         struct Scsi_Host *shost = scmd->device->host;
97         unsigned long flags;
98         int ret = 0;
99
100         if (!shost->ehandler)
101                 return 0;
102
103         spin_lock_irqsave(shost->host_lock, flags);
104         if (scsi_host_set_state(shost, SHOST_RECOVERY))
105                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
106                         goto out_unlock;
107
108         ret = 1;
109         scmd->eh_eflags |= eh_flag;
110         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
111         shost->host_failed++;
112         scsi_eh_wakeup(shost);
113  out_unlock:
114         spin_unlock_irqrestore(shost->host_lock, flags);
115         return ret;
116 }
117
118 /**
119  * scsi_times_out - Timeout function for normal scsi commands.
120  * @req:        request that is timing out.
121  *
122  * Notes:
123  *     We do not need to lock this.  There is the potential for a race
124  *     only in that the normal completion handling might run, but if the
125  *     normal completion function determines that the timer has already
126  *     fired, then it mustn't do anything.
127  */
128 enum blk_eh_timer_return scsi_times_out(struct request *req)
129 {
130         struct scsi_cmnd *scmd = req->special;
131         enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
132
133         trace_scsi_dispatch_cmd_timeout(scmd);
134         scsi_log_completion(scmd, TIMEOUT_ERROR);
135
136         if (scmd->device->host->transportt->eh_timed_out)
137                 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
138         else if (scmd->device->host->hostt->eh_timed_out)
139                 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
140
141         if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
142                      !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
143                 scmd->result |= DID_TIME_OUT << 16;
144                 rtn = BLK_EH_HANDLED;
145         }
146
147         return rtn;
148 }
149
150 /**
151  * scsi_block_when_processing_errors - Prevent cmds from being queued.
152  * @sdev:       Device on which we are performing recovery.
153  *
154  * Description:
155  *     We block until the host is out of error recovery, and then check to
156  *     see whether the host or the device is offline.
157  *
158  * Return value:
159  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
160  */
161 int scsi_block_when_processing_errors(struct scsi_device *sdev)
162 {
163         int online;
164
165         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
166
167         online = scsi_device_online(sdev);
168
169         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
170                                           online));
171
172         return online;
173 }
174 EXPORT_SYMBOL(scsi_block_when_processing_errors);
175
176 #ifdef CONFIG_SCSI_LOGGING
177 /**
178  * scsi_eh_prt_fail_stats - Log info on failures.
179  * @shost:      scsi host being recovered.
180  * @work_q:     Queue of scsi cmds to process.
181  */
182 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
183                                           struct list_head *work_q)
184 {
185         struct scsi_cmnd *scmd;
186         struct scsi_device *sdev;
187         int total_failures = 0;
188         int cmd_failed = 0;
189         int cmd_cancel = 0;
190         int devices_failed = 0;
191
192         shost_for_each_device(sdev, shost) {
193                 list_for_each_entry(scmd, work_q, eh_entry) {
194                         if (scmd->device == sdev) {
195                                 ++total_failures;
196                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
197                                         ++cmd_cancel;
198                                 else 
199                                         ++cmd_failed;
200                         }
201                 }
202
203                 if (cmd_cancel || cmd_failed) {
204                         SCSI_LOG_ERROR_RECOVERY(3,
205                                 sdev_printk(KERN_INFO, sdev,
206                                             "%s: cmds failed: %d, cancel: %d\n",
207                                             __func__, cmd_failed,
208                                             cmd_cancel));
209                         cmd_cancel = 0;
210                         cmd_failed = 0;
211                         ++devices_failed;
212                 }
213         }
214
215         SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
216                                           " devices require eh work\n",
217                                   total_failures, devices_failed));
218 }
219 #endif
220
221 /**
222  * scsi_check_sense - Examine scsi cmd sense
223  * @scmd:       Cmd to have sense checked.
224  *
225  * Return value:
226  *      SUCCESS or FAILED or NEEDS_RETRY
227  *
228  * Notes:
229  *      When a deferred error is detected the current command has
230  *      not been executed and needs retrying.
231  */
232 static int scsi_check_sense(struct scsi_cmnd *scmd)
233 {
234         struct scsi_device *sdev = scmd->device;
235         struct scsi_sense_hdr sshdr;
236
237         if (! scsi_command_normalize_sense(scmd, &sshdr))
238                 return FAILED;  /* no valid sense data */
239
240         if (scsi_sense_is_deferred(&sshdr))
241                 return NEEDS_RETRY;
242
243         if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
244                         sdev->scsi_dh_data->scsi_dh->check_sense) {
245                 int rc;
246
247                 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
248                 if (rc != SCSI_RETURN_NOT_HANDLED)
249                         return rc;
250                 /* handler does not care. Drop down to default handling */
251         }
252
253         /*
254          * Previous logic looked for FILEMARK, EOM or ILI which are
255          * mainly associated with tapes and returned SUCCESS.
256          */
257         if (sshdr.response_code == 0x70) {
258                 /* fixed format */
259                 if (scmd->sense_buffer[2] & 0xe0)
260                         return SUCCESS;
261         } else {
262                 /*
263                  * descriptor format: look for "stream commands sense data
264                  * descriptor" (see SSC-3). Assume single sense data
265                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
266                  */
267                 if ((sshdr.additional_length > 3) &&
268                     (scmd->sense_buffer[8] == 0x4) &&
269                     (scmd->sense_buffer[11] & 0xe0))
270                         return SUCCESS;
271         }
272
273         switch (sshdr.sense_key) {
274         case NO_SENSE:
275                 return SUCCESS;
276         case RECOVERED_ERROR:
277                 return /* soft_error */ SUCCESS;
278
279         case ABORTED_COMMAND:
280                 if (sshdr.asc == 0x10) /* DIF */
281                         return SUCCESS;
282
283                 return NEEDS_RETRY;
284         case NOT_READY:
285         case UNIT_ATTENTION:
286                 /*
287                  * if we are expecting a cc/ua because of a bus reset that we
288                  * performed, treat this just as a retry.  otherwise this is
289                  * information that we should pass up to the upper-level driver
290                  * so that we can deal with it there.
291                  */
292                 if (scmd->device->expecting_cc_ua) {
293                         scmd->device->expecting_cc_ua = 0;
294                         return NEEDS_RETRY;
295                 }
296                 /*
297                  * if the device is in the process of becoming ready, we 
298                  * should retry.
299                  */
300                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
301                         return NEEDS_RETRY;
302                 /*
303                  * if the device is not started, we need to wake
304                  * the error handler to start the motor
305                  */
306                 if (scmd->device->allow_restart &&
307                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
308                         return FAILED;
309
310                 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
311                         scmd_printk(KERN_WARNING, scmd,
312                                     "Warning! Received an indication that the "
313                                     "LUN assignments on this target have "
314                                     "changed. The Linux SCSI layer does not "
315                                     "automatically remap LUN assignments.\n");
316                 else if (sshdr.asc == 0x3f)
317                         scmd_printk(KERN_WARNING, scmd,
318                                     "Warning! Received an indication that the "
319                                     "operating parameters on this target have "
320                                     "changed. The Linux SCSI layer does not "
321                                     "automatically adjust these parameters.\n");
322
323                 if (blk_barrier_rq(scmd->request))
324                         /*
325                          * barrier requests should always retry on UA
326                          * otherwise block will get a spurious error
327                          */
328                         return NEEDS_RETRY;
329                 else
330                         /*
331                          * for normal (non barrier) commands, pass the
332                          * UA upwards for a determination in the
333                          * completion functions
334                          */
335                         return SUCCESS;
336
337                 /* these three are not supported */
338         case COPY_ABORTED:
339         case VOLUME_OVERFLOW:
340         case MISCOMPARE:
341                 return SUCCESS;
342
343         case MEDIUM_ERROR:
344                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
345                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
346                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
347                         return SUCCESS;
348                 }
349                 return NEEDS_RETRY;
350
351         case HARDWARE_ERROR:
352                 if (scmd->device->retry_hwerror)
353                         return ADD_TO_MLQUEUE;
354                 else
355                         return SUCCESS;
356
357         case ILLEGAL_REQUEST:
358         case BLANK_CHECK:
359         case DATA_PROTECT:
360         default:
361                 return SUCCESS;
362         }
363 }
364
365 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
366 {
367         struct scsi_host_template *sht = sdev->host->hostt;
368         struct scsi_device *tmp_sdev;
369
370         if (!sht->change_queue_depth ||
371             sdev->queue_depth >= sdev->max_queue_depth)
372                 return;
373
374         if (time_before(jiffies,
375             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
376                 return;
377
378         if (time_before(jiffies,
379             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
380                 return;
381
382         /*
383          * Walk all devices of a target and do
384          * ramp up on them.
385          */
386         shost_for_each_device(tmp_sdev, sdev->host) {
387                 if (tmp_sdev->channel != sdev->channel ||
388                     tmp_sdev->id != sdev->id ||
389                     tmp_sdev->queue_depth == sdev->max_queue_depth)
390                         continue;
391                 /*
392                  * call back into LLD to increase queue_depth by one
393                  * with ramp up reason code.
394                  */
395                 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
396                                         SCSI_QDEPTH_RAMP_UP);
397                 sdev->last_queue_ramp_up = jiffies;
398         }
399 }
400
401 static void scsi_handle_queue_full(struct scsi_device *sdev)
402 {
403         struct scsi_host_template *sht = sdev->host->hostt;
404         struct scsi_device *tmp_sdev;
405
406         if (!sht->change_queue_depth)
407                 return;
408
409         shost_for_each_device(tmp_sdev, sdev->host) {
410                 if (tmp_sdev->channel != sdev->channel ||
411                     tmp_sdev->id != sdev->id)
412                         continue;
413                 /*
414                  * We do not know the number of commands that were at
415                  * the device when we got the queue full so we start
416                  * from the highest possible value and work our way down.
417                  */
418                 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
419                                         SCSI_QDEPTH_QFULL);
420         }
421 }
422
423 /**
424  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
425  * @scmd:       SCSI cmd to examine.
426  *
427  * Notes:
428  *    This is *only* called when we are examining the status of commands
429  *    queued during error recovery.  the main difference here is that we
430  *    don't allow for the possibility of retries here, and we are a lot
431  *    more restrictive about what we consider acceptable.
432  */
433 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
434 {
435         /*
436          * first check the host byte, to see if there is anything in there
437          * that would indicate what we need to do.
438          */
439         if (host_byte(scmd->result) == DID_RESET) {
440                 /*
441                  * rats.  we are already in the error handler, so we now
442                  * get to try and figure out what to do next.  if the sense
443                  * is valid, we have a pretty good idea of what to do.
444                  * if not, we mark it as FAILED.
445                  */
446                 return scsi_check_sense(scmd);
447         }
448         if (host_byte(scmd->result) != DID_OK)
449                 return FAILED;
450
451         /*
452          * next, check the message byte.
453          */
454         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
455                 return FAILED;
456
457         /*
458          * now, check the status byte to see if this indicates
459          * anything special.
460          */
461         switch (status_byte(scmd->result)) {
462         case GOOD:
463                 scsi_handle_queue_ramp_up(scmd->device);
464         case COMMAND_TERMINATED:
465                 return SUCCESS;
466         case CHECK_CONDITION:
467                 return scsi_check_sense(scmd);
468         case CONDITION_GOOD:
469         case INTERMEDIATE_GOOD:
470         case INTERMEDIATE_C_GOOD:
471                 /*
472                  * who knows?  FIXME(eric)
473                  */
474                 return SUCCESS;
475         case RESERVATION_CONFLICT:
476                 /*
477                  * let issuer deal with this, it could be just fine
478                  */
479                 return SUCCESS;
480         case QUEUE_FULL:
481                 scsi_handle_queue_full(scmd->device);
482                 /* fall through */
483         case BUSY:
484                 return NEEDS_RETRY;
485         default:
486                 return FAILED;
487         }
488         return FAILED;
489 }
490
491 /**
492  * scsi_eh_done - Completion function for error handling.
493  * @scmd:       Cmd that is done.
494  */
495 static void scsi_eh_done(struct scsi_cmnd *scmd)
496 {
497         struct completion     *eh_action;
498
499         SCSI_LOG_ERROR_RECOVERY(3,
500                 printk("%s scmd: %p result: %x\n",
501                         __func__, scmd, scmd->result));
502
503         eh_action = scmd->device->host->eh_action;
504         if (eh_action)
505                 complete(eh_action);
506 }
507
508 /**
509  * scsi_try_host_reset - ask host adapter to reset itself
510  * @scmd:       SCSI cmd to send hsot reset.
511  */
512 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
513 {
514         unsigned long flags;
515         int rtn;
516
517         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
518                                           __func__));
519
520         if (!scmd->device->host->hostt->eh_host_reset_handler)
521                 return FAILED;
522
523         rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
524
525         if (rtn == SUCCESS) {
526                 if (!scmd->device->host->hostt->skip_settle_delay)
527                         ssleep(HOST_RESET_SETTLE_TIME);
528                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
529                 scsi_report_bus_reset(scmd->device->host,
530                                       scmd_channel(scmd));
531                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
532         }
533
534         return rtn;
535 }
536
537 /**
538  * scsi_try_bus_reset - ask host to perform a bus reset
539  * @scmd:       SCSI cmd to send bus reset.
540  */
541 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
542 {
543         unsigned long flags;
544         int rtn;
545
546         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
547                                           __func__));
548
549         if (!scmd->device->host->hostt->eh_bus_reset_handler)
550                 return FAILED;
551
552         rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
553
554         if (rtn == SUCCESS) {
555                 if (!scmd->device->host->hostt->skip_settle_delay)
556                         ssleep(BUS_RESET_SETTLE_TIME);
557                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
558                 scsi_report_bus_reset(scmd->device->host,
559                                       scmd_channel(scmd));
560                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
561         }
562
563         return rtn;
564 }
565
566 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
567 {
568         sdev->was_reset = 1;
569         sdev->expecting_cc_ua = 1;
570 }
571
572 /**
573  * scsi_try_target_reset - Ask host to perform a target reset
574  * @scmd:       SCSI cmd used to send a target reset
575  *
576  * Notes:
577  *    There is no timeout for this operation.  if this operation is
578  *    unreliable for a given host, then the host itself needs to put a
579  *    timer on it, and set the host back to a consistent state prior to
580  *    returning.
581  */
582 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
583 {
584         unsigned long flags;
585         int rtn;
586
587         if (!scmd->device->host->hostt->eh_target_reset_handler)
588                 return FAILED;
589
590         rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
591         if (rtn == SUCCESS) {
592                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
593                 __starget_for_each_device(scsi_target(scmd->device), NULL,
594                                           __scsi_report_device_reset);
595                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
596         }
597
598         return rtn;
599 }
600
601 /**
602  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
603  * @scmd:       SCSI cmd used to send BDR
604  *
605  * Notes:
606  *    There is no timeout for this operation.  if this operation is
607  *    unreliable for a given host, then the host itself needs to put a
608  *    timer on it, and set the host back to a consistent state prior to
609  *    returning.
610  */
611 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
612 {
613         int rtn;
614
615         if (!scmd->device->host->hostt->eh_device_reset_handler)
616                 return FAILED;
617
618         rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
619         if (rtn == SUCCESS)
620                 __scsi_report_device_reset(scmd->device, NULL);
621         return rtn;
622 }
623
624 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
625 {
626         if (!scmd->device->host->hostt->eh_abort_handler)
627                 return FAILED;
628
629         return scmd->device->host->hostt->eh_abort_handler(scmd);
630 }
631
632 /**
633  * scsi_try_to_abort_cmd - Ask host to abort a running command.
634  * @scmd:       SCSI cmd to abort from Lower Level.
635  *
636  * Notes:
637  *    This function will not return until the user's completion function
638  *    has been called.  there is no timeout on this operation.  if the
639  *    author of the low-level driver wishes this operation to be timed,
640  *    they can provide this facility themselves.  helper functions in
641  *    scsi_error.c can be supplied to make this easier to do.
642  */
643 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
644 {
645         /*
646          * scsi_done was called just after the command timed out and before
647          * we had a chance to process it. (db)
648          */
649         if (scmd->serial_number == 0)
650                 return SUCCESS;
651         return __scsi_try_to_abort_cmd(scmd);
652 }
653
654 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
655 {
656         if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
657                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
658                         if (scsi_try_target_reset(scmd) != SUCCESS)
659                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
660                                         scsi_try_host_reset(scmd);
661 }
662
663 /**
664  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recory
665  * @scmd:       SCSI command structure to hijack
666  * @ses:        structure to save restore information
667  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
668  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
669  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
670  *
671  * This function is used to save a scsi command information before re-execution
672  * as part of the error recovery process.  If @sense_bytes is 0 the command
673  * sent must be one that does not transfer any data.  If @sense_bytes != 0
674  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
675  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
676  */
677 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
678                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
679 {
680         struct scsi_device *sdev = scmd->device;
681
682         /*
683          * We need saved copies of a number of fields - this is because
684          * error handling may need to overwrite these with different values
685          * to run different commands, and once error handling is complete,
686          * we will need to restore these values prior to running the actual
687          * command.
688          */
689         ses->cmd_len = scmd->cmd_len;
690         ses->cmnd = scmd->cmnd;
691         ses->data_direction = scmd->sc_data_direction;
692         ses->sdb = scmd->sdb;
693         ses->next_rq = scmd->request->next_rq;
694         ses->result = scmd->result;
695         ses->underflow = scmd->underflow;
696         ses->prot_op = scmd->prot_op;
697
698         scmd->prot_op = SCSI_PROT_NORMAL;
699         scmd->cmnd = ses->eh_cmnd;
700         memset(scmd->cmnd, 0, BLK_MAX_CDB);
701         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
702         scmd->request->next_rq = NULL;
703
704         if (sense_bytes) {
705                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
706                                          sense_bytes);
707                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
708                             scmd->sdb.length);
709                 scmd->sdb.table.sgl = &ses->sense_sgl;
710                 scmd->sc_data_direction = DMA_FROM_DEVICE;
711                 scmd->sdb.table.nents = 1;
712                 scmd->cmnd[0] = REQUEST_SENSE;
713                 scmd->cmnd[4] = scmd->sdb.length;
714                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
715         } else {
716                 scmd->sc_data_direction = DMA_NONE;
717                 if (cmnd) {
718                         BUG_ON(cmnd_size > BLK_MAX_CDB);
719                         memcpy(scmd->cmnd, cmnd, cmnd_size);
720                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
721                 }
722         }
723
724         scmd->underflow = 0;
725
726         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
727                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
728                         (sdev->lun << 5 & 0xe0);
729
730         /*
731          * Zero the sense buffer.  The scsi spec mandates that any
732          * untransferred sense data should be interpreted as being zero.
733          */
734         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
735 }
736 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
737
738 /**
739  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recory
740  * @scmd:       SCSI command structure to restore
741  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
742  *
743  * Undo any damage done by above scsi_eh_prep_cmnd().
744  */
745 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
746 {
747         /*
748          * Restore original data
749          */
750         scmd->cmd_len = ses->cmd_len;
751         scmd->cmnd = ses->cmnd;
752         scmd->sc_data_direction = ses->data_direction;
753         scmd->sdb = ses->sdb;
754         scmd->request->next_rq = ses->next_rq;
755         scmd->result = ses->result;
756         scmd->underflow = ses->underflow;
757         scmd->prot_op = ses->prot_op;
758 }
759 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
760
761 /**
762  * scsi_send_eh_cmnd  - submit a scsi command as part of error recory
763  * @scmd:       SCSI command structure to hijack
764  * @cmnd:       CDB to send
765  * @cmnd_size:  size in bytes of @cmnd
766  * @timeout:    timeout for this request
767  * @sense_bytes: size of sense data to copy or 0
768  *
769  * This function is used to send a scsi command down to a target device
770  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
771  *
772  * Return value:
773  *    SUCCESS or FAILED or NEEDS_RETRY
774  */
775 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
776                              int cmnd_size, int timeout, unsigned sense_bytes)
777 {
778         struct scsi_device *sdev = scmd->device;
779         struct Scsi_Host *shost = sdev->host;
780         DECLARE_COMPLETION_ONSTACK(done);
781         unsigned long timeleft;
782         unsigned long flags;
783         struct scsi_eh_save ses;
784         int rtn;
785
786         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
787         shost->eh_action = &done;
788
789         spin_lock_irqsave(shost->host_lock, flags);
790         scsi_log_send(scmd);
791         shost->hostt->queuecommand(scmd, scsi_eh_done);
792         spin_unlock_irqrestore(shost->host_lock, flags);
793
794         timeleft = wait_for_completion_timeout(&done, timeout);
795
796         shost->eh_action = NULL;
797
798         scsi_log_completion(scmd, SUCCESS);
799
800         SCSI_LOG_ERROR_RECOVERY(3,
801                 printk("%s: scmd: %p, timeleft: %ld\n",
802                         __func__, scmd, timeleft));
803
804         /*
805          * If there is time left scsi_eh_done got called, and we will
806          * examine the actual status codes to see whether the command
807          * actually did complete normally, else tell the host to forget
808          * about this command.
809          */
810         if (timeleft) {
811                 rtn = scsi_eh_completed_normally(scmd);
812                 SCSI_LOG_ERROR_RECOVERY(3,
813                         printk("%s: scsi_eh_completed_normally %x\n",
814                                __func__, rtn));
815
816                 switch (rtn) {
817                 case SUCCESS:
818                 case NEEDS_RETRY:
819                 case FAILED:
820                         break;
821                 case ADD_TO_MLQUEUE:
822                         rtn = NEEDS_RETRY;
823                         break;
824                 default:
825                         rtn = FAILED;
826                         break;
827                 }
828         } else {
829                 scsi_abort_eh_cmnd(scmd);
830                 rtn = FAILED;
831         }
832
833         scsi_eh_restore_cmnd(scmd, &ses);
834         return rtn;
835 }
836
837 /**
838  * scsi_request_sense - Request sense data from a particular target.
839  * @scmd:       SCSI cmd for request sense.
840  *
841  * Notes:
842  *    Some hosts automatically obtain this information, others require
843  *    that we obtain it on our own. This function will *not* return until
844  *    the command either times out, or it completes.
845  */
846 static int scsi_request_sense(struct scsi_cmnd *scmd)
847 {
848         return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
849 }
850
851 /**
852  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
853  * @scmd:       Original SCSI cmd that eh has finished.
854  * @done_q:     Queue for processed commands.
855  *
856  * Notes:
857  *    We don't want to use the normal command completion while we are are
858  *    still handling errors - it may cause other commands to be queued,
859  *    and that would disturb what we are doing.  Thus we really want to
860  *    keep a list of pending commands for final completion, and once we
861  *    are ready to leave error handling we handle completion for real.
862  */
863 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
864 {
865         scmd->device->host->host_failed--;
866         scmd->eh_eflags = 0;
867         list_move_tail(&scmd->eh_entry, done_q);
868 }
869 EXPORT_SYMBOL(scsi_eh_finish_cmd);
870
871 /**
872  * scsi_eh_get_sense - Get device sense data.
873  * @work_q:     Queue of commands to process.
874  * @done_q:     Queue of processed commands.
875  *
876  * Description:
877  *    See if we need to request sense information.  if so, then get it
878  *    now, so we have a better idea of what to do.  
879  *
880  * Notes:
881  *    This has the unfortunate side effect that if a shost adapter does
882  *    not automatically request sense information, we end up shutting
883  *    it down before we request it.
884  *
885  *    All drivers should request sense information internally these days,
886  *    so for now all I have to say is tough noogies if you end up in here.
887  *
888  *    XXX: Long term this code should go away, but that needs an audit of
889  *         all LLDDs first.
890  */
891 int scsi_eh_get_sense(struct list_head *work_q,
892                       struct list_head *done_q)
893 {
894         struct scsi_cmnd *scmd, *next;
895         int rtn;
896
897         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
898                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
899                     SCSI_SENSE_VALID(scmd))
900                         continue;
901
902                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
903                                                   "%s: requesting sense\n",
904                                                   current->comm));
905                 rtn = scsi_request_sense(scmd);
906                 if (rtn != SUCCESS)
907                         continue;
908
909                 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
910                                                   " result %x\n", scmd,
911                                                   scmd->result));
912                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
913
914                 rtn = scsi_decide_disposition(scmd);
915
916                 /*
917                  * if the result was normal, then just pass it along to the
918                  * upper level.
919                  */
920                 if (rtn == SUCCESS)
921                         /* we don't want this command reissued, just
922                          * finished with the sense data, so set
923                          * retries to the max allowed to ensure it
924                          * won't get reissued */
925                         scmd->retries = scmd->allowed;
926                 else if (rtn != NEEDS_RETRY)
927                         continue;
928
929                 scsi_eh_finish_cmd(scmd, done_q);
930         }
931
932         return list_empty(work_q);
933 }
934 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
935
936 /**
937  * scsi_eh_tur - Send TUR to device.
938  * @scmd:       &scsi_cmnd to send TUR
939  *
940  * Return value:
941  *    0 - Device is ready. 1 - Device NOT ready.
942  */
943 static int scsi_eh_tur(struct scsi_cmnd *scmd)
944 {
945         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
946         int retry_cnt = 1, rtn;
947
948 retry_tur:
949         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
950
951         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
952                 __func__, scmd, rtn));
953
954         switch (rtn) {
955         case NEEDS_RETRY:
956                 if (retry_cnt--)
957                         goto retry_tur;
958                 /*FALLTHRU*/
959         case SUCCESS:
960                 return 0;
961         default:
962                 return 1;
963         }
964 }
965
966 /**
967  * scsi_eh_abort_cmds - abort pending commands.
968  * @work_q:     &list_head for pending commands.
969  * @done_q:     &list_head for processed commands.
970  *
971  * Decription:
972  *    Try and see whether or not it makes sense to try and abort the
973  *    running command.  This only works out to be the case if we have one
974  *    command that has timed out.  If the command simply failed, it makes
975  *    no sense to try and abort the command, since as far as the shost
976  *    adapter is concerned, it isn't running.
977  */
978 static int scsi_eh_abort_cmds(struct list_head *work_q,
979                               struct list_head *done_q)
980 {
981         struct scsi_cmnd *scmd, *next;
982         int rtn;
983
984         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
985                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
986                         continue;
987                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
988                                                   "0x%p\n", current->comm,
989                                                   scmd));
990                 rtn = scsi_try_to_abort_cmd(scmd);
991                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
992                         scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
993                         if (!scsi_device_online(scmd->device) ||
994                             rtn == FAST_IO_FAIL ||
995                             !scsi_eh_tur(scmd)) {
996                                 scsi_eh_finish_cmd(scmd, done_q);
997                         }
998                                 
999                 } else
1000                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1001                                                           " cmd failed:"
1002                                                           "0x%p\n",
1003                                                           current->comm,
1004                                                           scmd));
1005         }
1006
1007         return list_empty(work_q);
1008 }
1009
1010 /**
1011  * scsi_eh_try_stu - Send START_UNIT to device.
1012  * @scmd:       &scsi_cmnd to send START_UNIT
1013  *
1014  * Return value:
1015  *    0 - Device is ready. 1 - Device NOT ready.
1016  */
1017 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1018 {
1019         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1020
1021         if (scmd->device->allow_restart) {
1022                 int i, rtn = NEEDS_RETRY;
1023
1024                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1025                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1026
1027                 if (rtn == SUCCESS)
1028                         return 0;
1029         }
1030
1031         return 1;
1032 }
1033
1034  /**
1035  * scsi_eh_stu - send START_UNIT if needed
1036  * @shost:      &scsi host being recovered.
1037  * @work_q:     &list_head for pending commands.
1038  * @done_q:     &list_head for processed commands.
1039  *
1040  * Notes:
1041  *    If commands are failing due to not ready, initializing command required,
1042  *      try revalidating the device, which will end up sending a start unit. 
1043  */
1044 static int scsi_eh_stu(struct Scsi_Host *shost,
1045                               struct list_head *work_q,
1046                               struct list_head *done_q)
1047 {
1048         struct scsi_cmnd *scmd, *stu_scmd, *next;
1049         struct scsi_device *sdev;
1050
1051         shost_for_each_device(sdev, shost) {
1052                 stu_scmd = NULL;
1053                 list_for_each_entry(scmd, work_q, eh_entry)
1054                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1055                             scsi_check_sense(scmd) == FAILED ) {
1056                                 stu_scmd = scmd;
1057                                 break;
1058                         }
1059
1060                 if (!stu_scmd)
1061                         continue;
1062
1063                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1064                                                   " 0x%p\n", current->comm, sdev));
1065
1066                 if (!scsi_eh_try_stu(stu_scmd)) {
1067                         if (!scsi_device_online(sdev) ||
1068                             !scsi_eh_tur(stu_scmd)) {
1069                                 list_for_each_entry_safe(scmd, next,
1070                                                           work_q, eh_entry) {
1071                                         if (scmd->device == sdev)
1072                                                 scsi_eh_finish_cmd(scmd, done_q);
1073                                 }
1074                         }
1075                 } else {
1076                         SCSI_LOG_ERROR_RECOVERY(3,
1077                                                 printk("%s: START_UNIT failed to sdev:"
1078                                                        " 0x%p\n", current->comm, sdev));
1079                 }
1080         }
1081
1082         return list_empty(work_q);
1083 }
1084
1085
1086 /**
1087  * scsi_eh_bus_device_reset - send bdr if needed
1088  * @shost:      scsi host being recovered.
1089  * @work_q:     &list_head for pending commands.
1090  * @done_q:     &list_head for processed commands.
1091  *
1092  * Notes:
1093  *    Try a bus device reset.  Still, look to see whether we have multiple
1094  *    devices that are jammed or not - if we have multiple devices, it
1095  *    makes no sense to try bus_device_reset - we really would need to try
1096  *    a bus_reset instead. 
1097  */
1098 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1099                                     struct list_head *work_q,
1100                                     struct list_head *done_q)
1101 {
1102         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1103         struct scsi_device *sdev;
1104         int rtn;
1105
1106         shost_for_each_device(sdev, shost) {
1107                 bdr_scmd = NULL;
1108                 list_for_each_entry(scmd, work_q, eh_entry)
1109                         if (scmd->device == sdev) {
1110                                 bdr_scmd = scmd;
1111                                 break;
1112                         }
1113
1114                 if (!bdr_scmd)
1115                         continue;
1116
1117                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1118                                                   " 0x%p\n", current->comm,
1119                                                   sdev));
1120                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1121                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1122                         if (!scsi_device_online(sdev) ||
1123                             rtn == FAST_IO_FAIL ||
1124                             !scsi_eh_tur(bdr_scmd)) {
1125                                 list_for_each_entry_safe(scmd, next,
1126                                                          work_q, eh_entry) {
1127                                         if (scmd->device == sdev)
1128                                                 scsi_eh_finish_cmd(scmd,
1129                                                                    done_q);
1130                                 }
1131                         }
1132                 } else {
1133                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1134                                                           " failed sdev:"
1135                                                           "0x%p\n",
1136                                                           current->comm,
1137                                                            sdev));
1138                 }
1139         }
1140
1141         return list_empty(work_q);
1142 }
1143
1144 /**
1145  * scsi_eh_target_reset - send target reset if needed
1146  * @shost:      scsi host being recovered.
1147  * @work_q:     &list_head for pending commands.
1148  * @done_q:     &list_head for processed commands.
1149  *
1150  * Notes:
1151  *    Try a target reset.
1152  */
1153 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1154                                 struct list_head *work_q,
1155                                 struct list_head *done_q)
1156 {
1157         struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1158         unsigned int id = 0;
1159         int rtn;
1160
1161         do {
1162                 tgtr_scmd = NULL;
1163                 list_for_each_entry(scmd, work_q, eh_entry) {
1164                         if (id == scmd_id(scmd)) {
1165                                 tgtr_scmd = scmd;
1166                                 break;
1167                         }
1168                 }
1169                 if (!tgtr_scmd) {
1170                         /* not one exactly equal; find the next highest */
1171                         list_for_each_entry(scmd, work_q, eh_entry) {
1172                                 if (scmd_id(scmd) > id &&
1173                                     (!tgtr_scmd ||
1174                                      scmd_id(tgtr_scmd) > scmd_id(scmd)))
1175                                                 tgtr_scmd = scmd;
1176                         }
1177                 }
1178                 if (!tgtr_scmd)
1179                         /* no more commands, that's it */
1180                         break;
1181
1182                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1183                                                   "to target %d\n",
1184                                                   current->comm, id));
1185                 rtn = scsi_try_target_reset(tgtr_scmd);
1186                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1187                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1188                                 if (id == scmd_id(scmd))
1189                                         if (!scsi_device_online(scmd->device) ||
1190                                             rtn == FAST_IO_FAIL ||
1191                                             !scsi_eh_tur(tgtr_scmd))
1192                                                 scsi_eh_finish_cmd(scmd,
1193                                                                    done_q);
1194                         }
1195                 } else
1196                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1197                                                           " failed target: "
1198                                                           "%d\n",
1199                                                           current->comm, id));
1200                 id++;
1201         } while(id != 0);
1202
1203         return list_empty(work_q);
1204 }
1205
1206 /**
1207  * scsi_eh_bus_reset - send a bus reset 
1208  * @shost:      &scsi host being recovered.
1209  * @work_q:     &list_head for pending commands.
1210  * @done_q:     &list_head for processed commands.
1211  */
1212 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1213                              struct list_head *work_q,
1214                              struct list_head *done_q)
1215 {
1216         struct scsi_cmnd *scmd, *chan_scmd, *next;
1217         unsigned int channel;
1218         int rtn;
1219
1220         /*
1221          * we really want to loop over the various channels, and do this on
1222          * a channel by channel basis.  we should also check to see if any
1223          * of the failed commands are on soft_reset devices, and if so, skip
1224          * the reset.  
1225          */
1226
1227         for (channel = 0; channel <= shost->max_channel; channel++) {
1228                 chan_scmd = NULL;
1229                 list_for_each_entry(scmd, work_q, eh_entry) {
1230                         if (channel == scmd_channel(scmd)) {
1231                                 chan_scmd = scmd;
1232                                 break;
1233                                 /*
1234                                  * FIXME add back in some support for
1235                                  * soft_reset devices.
1236                                  */
1237                         }
1238                 }
1239
1240                 if (!chan_scmd)
1241                         continue;
1242                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1243                                                   " %d\n", current->comm,
1244                                                   channel));
1245                 rtn = scsi_try_bus_reset(chan_scmd);
1246                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1247                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1248                                 if (channel == scmd_channel(scmd))
1249                                         if (!scsi_device_online(scmd->device) ||
1250                                             rtn == FAST_IO_FAIL ||
1251                                             !scsi_eh_tur(scmd))
1252                                                 scsi_eh_finish_cmd(scmd,
1253                                                                    done_q);
1254                         }
1255                 } else {
1256                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1257                                                           " failed chan: %d\n",
1258                                                           current->comm,
1259                                                           channel));
1260                 }
1261         }
1262         return list_empty(work_q);
1263 }
1264
1265 /**
1266  * scsi_eh_host_reset - send a host reset 
1267  * @work_q:     list_head for processed commands.
1268  * @done_q:     list_head for processed commands.
1269  */
1270 static int scsi_eh_host_reset(struct list_head *work_q,
1271                               struct list_head *done_q)
1272 {
1273         struct scsi_cmnd *scmd, *next;
1274         int rtn;
1275
1276         if (!list_empty(work_q)) {
1277                 scmd = list_entry(work_q->next,
1278                                   struct scsi_cmnd, eh_entry);
1279
1280                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1281                                                   , current->comm));
1282
1283                 rtn = scsi_try_host_reset(scmd);
1284                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1285                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1286                                 if (!scsi_device_online(scmd->device) ||
1287                                     rtn == FAST_IO_FAIL ||
1288                                     (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1289                                     !scsi_eh_tur(scmd))
1290                                         scsi_eh_finish_cmd(scmd, done_q);
1291                         }
1292                 } else {
1293                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1294                                                           " failed\n",
1295                                                           current->comm));
1296                 }
1297         }
1298         return list_empty(work_q);
1299 }
1300
1301 /**
1302  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1303  * @work_q:     list_head for processed commands.
1304  * @done_q:     list_head for processed commands.
1305  */
1306 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1307                                   struct list_head *done_q)
1308 {
1309         struct scsi_cmnd *scmd, *next;
1310
1311         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1312                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1313                             "not ready after error recovery\n");
1314                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1315                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1316                         /*
1317                          * FIXME: Handle lost cmds.
1318                          */
1319                 }
1320                 scsi_eh_finish_cmd(scmd, done_q);
1321         }
1322         return;
1323 }
1324
1325 /**
1326  * scsi_noretry_cmd - determinte if command should be failed fast
1327  * @scmd:       SCSI cmd to examine.
1328  */
1329 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1330 {
1331         switch (host_byte(scmd->result)) {
1332         case DID_OK:
1333                 break;
1334         case DID_BUS_BUSY:
1335                 return blk_failfast_transport(scmd->request);
1336         case DID_PARITY:
1337                 return blk_failfast_dev(scmd->request);
1338         case DID_ERROR:
1339                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1340                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1341                         return 0;
1342                 /* fall through */
1343         case DID_SOFT_ERROR:
1344                 return blk_failfast_driver(scmd->request);
1345         }
1346
1347         switch (status_byte(scmd->result)) {
1348         case CHECK_CONDITION:
1349                 /*
1350                  * assume caller has checked sense and determinted
1351                  * the check condition was retryable.
1352                  */
1353                 return blk_failfast_dev(scmd->request);
1354         }
1355
1356         return 0;
1357 }
1358
1359 /**
1360  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1361  * @scmd:       SCSI cmd to examine.
1362  *
1363  * Notes:
1364  *    This is *only* called when we are examining the status after sending
1365  *    out the actual data command.  any commands that are queued for error
1366  *    recovery (e.g. test_unit_ready) do *not* come through here.
1367  *
1368  *    When this routine returns failed, it means the error handler thread
1369  *    is woken.  In cases where the error code indicates an error that
1370  *    doesn't require the error handler read (i.e. we don't need to
1371  *    abort/reset), this function should return SUCCESS.
1372  */
1373 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1374 {
1375         int rtn;
1376
1377         /*
1378          * if the device is offline, then we clearly just pass the result back
1379          * up to the top level.
1380          */
1381         if (!scsi_device_online(scmd->device)) {
1382                 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1383                                                   " as SUCCESS\n",
1384                                                   __func__));
1385                 return SUCCESS;
1386         }
1387
1388         /*
1389          * first check the host byte, to see if there is anything in there
1390          * that would indicate what we need to do.
1391          */
1392         switch (host_byte(scmd->result)) {
1393         case DID_PASSTHROUGH:
1394                 /*
1395                  * no matter what, pass this through to the upper layer.
1396                  * nuke this special code so that it looks like we are saying
1397                  * did_ok.
1398                  */
1399                 scmd->result &= 0xff00ffff;
1400                 return SUCCESS;
1401         case DID_OK:
1402                 /*
1403                  * looks good.  drop through, and check the next byte.
1404                  */
1405                 break;
1406         case DID_NO_CONNECT:
1407         case DID_BAD_TARGET:
1408         case DID_ABORT:
1409                 /*
1410                  * note - this means that we just report the status back
1411                  * to the top level driver, not that we actually think
1412                  * that it indicates SUCCESS.
1413                  */
1414                 return SUCCESS;
1415                 /*
1416                  * when the low level driver returns did_soft_error,
1417                  * it is responsible for keeping an internal retry counter 
1418                  * in order to avoid endless loops (db)
1419                  *
1420                  * actually this is a bug in this function here.  we should
1421                  * be mindful of the maximum number of retries specified
1422                  * and not get stuck in a loop.
1423                  */
1424         case DID_SOFT_ERROR:
1425                 goto maybe_retry;
1426         case DID_IMM_RETRY:
1427                 return NEEDS_RETRY;
1428
1429         case DID_REQUEUE:
1430                 return ADD_TO_MLQUEUE;
1431         case DID_TRANSPORT_DISRUPTED:
1432                 /*
1433                  * LLD/transport was disrupted during processing of the IO.
1434                  * The transport class is now blocked/blocking,
1435                  * and the transport will decide what to do with the IO
1436                  * based on its timers and recovery capablilities if
1437                  * there are enough retries.
1438                  */
1439                 goto maybe_retry;
1440         case DID_TRANSPORT_FAILFAST:
1441                 /*
1442                  * The transport decided to failfast the IO (most likely
1443                  * the fast io fail tmo fired), so send IO directly upwards.
1444                  */
1445                 return SUCCESS;
1446         case DID_ERROR:
1447                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1448                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1449                         /*
1450                          * execute reservation conflict processing code
1451                          * lower down
1452                          */
1453                         break;
1454                 /* fallthrough */
1455
1456         case DID_BUS_BUSY:
1457         case DID_PARITY:
1458                 goto maybe_retry;
1459         case DID_TIME_OUT:
1460                 /*
1461                  * when we scan the bus, we get timeout messages for
1462                  * these commands if there is no device available.
1463                  * other hosts report did_no_connect for the same thing.
1464                  */
1465                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1466                      scmd->cmnd[0] == INQUIRY)) {
1467                         return SUCCESS;
1468                 } else {
1469                         return FAILED;
1470                 }
1471         case DID_RESET:
1472                 return SUCCESS;
1473         default:
1474                 return FAILED;
1475         }
1476
1477         /*
1478          * next, check the message byte.
1479          */
1480         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1481                 return FAILED;
1482
1483         /*
1484          * check the status byte to see if this indicates anything special.
1485          */
1486         switch (status_byte(scmd->result)) {
1487         case QUEUE_FULL:
1488                 scsi_handle_queue_full(scmd->device);
1489                 /*
1490                  * the case of trying to send too many commands to a
1491                  * tagged queueing device.
1492                  */
1493         case BUSY:
1494                 /*
1495                  * device can't talk to us at the moment.  Should only
1496                  * occur (SAM-3) when the task queue is empty, so will cause
1497                  * the empty queue handling to trigger a stall in the
1498                  * device.
1499                  */
1500                 return ADD_TO_MLQUEUE;
1501         case GOOD:
1502                 scsi_handle_queue_ramp_up(scmd->device);
1503         case COMMAND_TERMINATED:
1504                 return SUCCESS;
1505         case TASK_ABORTED:
1506                 goto maybe_retry;
1507         case CHECK_CONDITION:
1508                 rtn = scsi_check_sense(scmd);
1509                 if (rtn == NEEDS_RETRY)
1510                         goto maybe_retry;
1511                 /* if rtn == FAILED, we have no sense information;
1512                  * returning FAILED will wake the error handler thread
1513                  * to collect the sense and redo the decide
1514                  * disposition */
1515                 return rtn;
1516         case CONDITION_GOOD:
1517         case INTERMEDIATE_GOOD:
1518         case INTERMEDIATE_C_GOOD:
1519         case ACA_ACTIVE:
1520                 /*
1521                  * who knows?  FIXME(eric)
1522                  */
1523                 return SUCCESS;
1524
1525         case RESERVATION_CONFLICT:
1526                 sdev_printk(KERN_INFO, scmd->device,
1527                             "reservation conflict\n");
1528                 return SUCCESS; /* causes immediate i/o error */
1529         default:
1530                 return FAILED;
1531         }
1532         return FAILED;
1533
1534       maybe_retry:
1535
1536         /* we requeue for retry because the error was retryable, and
1537          * the request was not marked fast fail.  Note that above,
1538          * even if the request is marked fast fail, we still requeue
1539          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1540         if ((++scmd->retries) <= scmd->allowed
1541             && !scsi_noretry_cmd(scmd)) {
1542                 return NEEDS_RETRY;
1543         } else {
1544                 /*
1545                  * no more retries - report this one back to upper level.
1546                  */
1547                 return SUCCESS;
1548         }
1549 }
1550
1551 static void eh_lock_door_done(struct request *req, int uptodate)
1552 {
1553         __blk_put_request(req->q, req);
1554 }
1555
1556 /**
1557  * scsi_eh_lock_door - Prevent medium removal for the specified device
1558  * @sdev:       SCSI device to prevent medium removal
1559  *
1560  * Locking:
1561  *      We must be called from process context.
1562  *
1563  * Notes:
1564  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1565  *      head of the devices request queue, and continue.
1566  */
1567 static void scsi_eh_lock_door(struct scsi_device *sdev)
1568 {
1569         struct request *req;
1570
1571         /*
1572          * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1573          * request becomes available
1574          */
1575         req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1576
1577         req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1578         req->cmd[1] = 0;
1579         req->cmd[2] = 0;
1580         req->cmd[3] = 0;
1581         req->cmd[4] = SCSI_REMOVAL_PREVENT;
1582         req->cmd[5] = 0;
1583
1584         req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1585
1586         req->cmd_type = REQ_TYPE_BLOCK_PC;
1587         req->cmd_flags |= REQ_QUIET;
1588         req->timeout = 10 * HZ;
1589         req->retries = 5;
1590
1591         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1592 }
1593
1594 /**
1595  * scsi_restart_operations - restart io operations to the specified host.
1596  * @shost:      Host we are restarting.
1597  *
1598  * Notes:
1599  *    When we entered the error handler, we blocked all further i/o to
1600  *    this device.  we need to 'reverse' this process.
1601  */
1602 static void scsi_restart_operations(struct Scsi_Host *shost)
1603 {
1604         struct scsi_device *sdev;
1605         unsigned long flags;
1606
1607         /*
1608          * If the door was locked, we need to insert a door lock request
1609          * onto the head of the SCSI request queue for the device.  There
1610          * is no point trying to lock the door of an off-line device.
1611          */
1612         shost_for_each_device(sdev, shost) {
1613                 if (scsi_device_online(sdev) && sdev->locked)
1614                         scsi_eh_lock_door(sdev);
1615         }
1616
1617         /*
1618          * next free up anything directly waiting upon the host.  this
1619          * will be requests for character device operations, and also for
1620          * ioctls to queued block devices.
1621          */
1622         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1623                                           __func__));
1624
1625         spin_lock_irqsave(shost->host_lock, flags);
1626         if (scsi_host_set_state(shost, SHOST_RUNNING))
1627                 if (scsi_host_set_state(shost, SHOST_CANCEL))
1628                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1629         spin_unlock_irqrestore(shost->host_lock, flags);
1630
1631         wake_up(&shost->host_wait);
1632
1633         /*
1634          * finally we need to re-initiate requests that may be pending.  we will
1635          * have had everything blocked while error handling is taking place, and
1636          * now that error recovery is done, we will need to ensure that these
1637          * requests are started.
1638          */
1639         scsi_run_host_queues(shost);
1640 }
1641
1642 /**
1643  * scsi_eh_ready_devs - check device ready state and recover if not.
1644  * @shost:      host to be recovered.
1645  * @work_q:     &list_head for pending commands.
1646  * @done_q:     &list_head for processed commands.
1647  */
1648 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1649                         struct list_head *work_q,
1650                         struct list_head *done_q)
1651 {
1652         if (!scsi_eh_stu(shost, work_q, done_q))
1653                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1654                         if (!scsi_eh_target_reset(shost, work_q, done_q))
1655                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1656                                         if (!scsi_eh_host_reset(work_q, done_q))
1657                                                 scsi_eh_offline_sdevs(work_q,
1658                                                                       done_q);
1659 }
1660 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1661
1662 /**
1663  * scsi_eh_flush_done_q - finish processed commands or retry them.
1664  * @done_q:     list_head of processed commands.
1665  */
1666 void scsi_eh_flush_done_q(struct list_head *done_q)
1667 {
1668         struct scsi_cmnd *scmd, *next;
1669
1670         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1671                 list_del_init(&scmd->eh_entry);
1672                 if (scsi_device_online(scmd->device) &&
1673                     !scsi_noretry_cmd(scmd) &&
1674                     (++scmd->retries <= scmd->allowed)) {
1675                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1676                                                           " retry cmd: %p\n",
1677                                                           current->comm,
1678                                                           scmd));
1679                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1680                 } else {
1681                         /*
1682                          * If just we got sense for the device (called
1683                          * scsi_eh_get_sense), scmd->result is already
1684                          * set, do not set DRIVER_TIMEOUT.
1685                          */
1686                         if (!scmd->result)
1687                                 scmd->result |= (DRIVER_TIMEOUT << 24);
1688                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1689                                                         " cmd: %p\n",
1690                                                         current->comm, scmd));
1691                         scsi_finish_command(scmd);
1692                 }
1693         }
1694 }
1695 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1696
1697 /**
1698  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1699  * @shost:      Host to unjam.
1700  *
1701  * Notes:
1702  *    When we come in here, we *know* that all commands on the bus have
1703  *    either completed, failed or timed out.  we also know that no further
1704  *    commands are being sent to the host, so things are relatively quiet
1705  *    and we have freedom to fiddle with things as we wish.
1706  *
1707  *    This is only the *default* implementation.  it is possible for
1708  *    individual drivers to supply their own version of this function, and
1709  *    if the maintainer wishes to do this, it is strongly suggested that
1710  *    this function be taken as a template and modified.  this function
1711  *    was designed to correctly handle problems for about 95% of the
1712  *    different cases out there, and it should always provide at least a
1713  *    reasonable amount of error recovery.
1714  *
1715  *    Any command marked 'failed' or 'timeout' must eventually have
1716  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1717  *    here, so when we restart the host after we return it should have an
1718  *    empty queue.
1719  */
1720 static void scsi_unjam_host(struct Scsi_Host *shost)
1721 {
1722         unsigned long flags;
1723         LIST_HEAD(eh_work_q);
1724         LIST_HEAD(eh_done_q);
1725
1726         spin_lock_irqsave(shost->host_lock, flags);
1727         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1728         spin_unlock_irqrestore(shost->host_lock, flags);
1729
1730         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1731
1732         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1733                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1734                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1735
1736         scsi_eh_flush_done_q(&eh_done_q);
1737 }
1738
1739 /**
1740  * scsi_error_handler - SCSI error handler thread
1741  * @data:       Host for which we are running.
1742  *
1743  * Notes:
1744  *    This is the main error handling loop.  This is run as a kernel thread
1745  *    for every SCSI host and handles all error handling activity.
1746  */
1747 int scsi_error_handler(void *data)
1748 {
1749         struct Scsi_Host *shost = data;
1750
1751         /*
1752          * We use TASK_INTERRUPTIBLE so that the thread is not
1753          * counted against the load average as a running process.
1754          * We never actually get interrupted because kthread_run
1755          * disables signal delivery for the created thread.
1756          */
1757         set_current_state(TASK_INTERRUPTIBLE);
1758         while (!kthread_should_stop()) {
1759                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1760                     shost->host_failed != shost->host_busy) {
1761                         SCSI_LOG_ERROR_RECOVERY(1,
1762                                 printk("Error handler scsi_eh_%d sleeping\n",
1763                                         shost->host_no));
1764                         schedule();
1765                         set_current_state(TASK_INTERRUPTIBLE);
1766                         continue;
1767                 }
1768
1769                 __set_current_state(TASK_RUNNING);
1770                 SCSI_LOG_ERROR_RECOVERY(1,
1771                         printk("Error handler scsi_eh_%d waking up\n",
1772                                 shost->host_no));
1773
1774                 /*
1775                  * We have a host that is failing for some reason.  Figure out
1776                  * what we need to do to get it up and online again (if we can).
1777                  * If we fail, we end up taking the thing offline.
1778                  */
1779                 if (scsi_autopm_get_host(shost) != 0) {
1780                         SCSI_LOG_ERROR_RECOVERY(1,
1781                                 printk(KERN_ERR "Error handler scsi_eh_%d "
1782                                                 "unable to autoresume\n",
1783                                                 shost->host_no));
1784                         continue;
1785                 }
1786
1787                 if (shost->transportt->eh_strategy_handler)
1788                         shost->transportt->eh_strategy_handler(shost);
1789                 else
1790                         scsi_unjam_host(shost);
1791
1792                 /*
1793                  * Note - if the above fails completely, the action is to take
1794                  * individual devices offline and flush the queue of any
1795                  * outstanding requests that may have been pending.  When we
1796                  * restart, we restart any I/O to any other devices on the bus
1797                  * which are still online.
1798                  */
1799                 scsi_restart_operations(shost);
1800                 scsi_autopm_put_host(shost);
1801                 set_current_state(TASK_INTERRUPTIBLE);
1802         }
1803         __set_current_state(TASK_RUNNING);
1804
1805         SCSI_LOG_ERROR_RECOVERY(1,
1806                 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1807         shost->ehandler = NULL;
1808         return 0;
1809 }
1810
1811 /*
1812  * Function:    scsi_report_bus_reset()
1813  *
1814  * Purpose:     Utility function used by low-level drivers to report that
1815  *              they have observed a bus reset on the bus being handled.
1816  *
1817  * Arguments:   shost       - Host in question
1818  *              channel     - channel on which reset was observed.
1819  *
1820  * Returns:     Nothing
1821  *
1822  * Lock status: Host lock must be held.
1823  *
1824  * Notes:       This only needs to be called if the reset is one which
1825  *              originates from an unknown location.  Resets originated
1826  *              by the mid-level itself don't need to call this, but there
1827  *              should be no harm.
1828  *
1829  *              The main purpose of this is to make sure that a CHECK_CONDITION
1830  *              is properly treated.
1831  */
1832 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1833 {
1834         struct scsi_device *sdev;
1835
1836         __shost_for_each_device(sdev, shost) {
1837                 if (channel == sdev_channel(sdev))
1838                         __scsi_report_device_reset(sdev, NULL);
1839         }
1840 }
1841 EXPORT_SYMBOL(scsi_report_bus_reset);
1842
1843 /*
1844  * Function:    scsi_report_device_reset()
1845  *
1846  * Purpose:     Utility function used by low-level drivers to report that
1847  *              they have observed a device reset on the device being handled.
1848  *
1849  * Arguments:   shost       - Host in question
1850  *              channel     - channel on which reset was observed
1851  *              target      - target on which reset was observed
1852  *
1853  * Returns:     Nothing
1854  *
1855  * Lock status: Host lock must be held
1856  *
1857  * Notes:       This only needs to be called if the reset is one which
1858  *              originates from an unknown location.  Resets originated
1859  *              by the mid-level itself don't need to call this, but there
1860  *              should be no harm.
1861  *
1862  *              The main purpose of this is to make sure that a CHECK_CONDITION
1863  *              is properly treated.
1864  */
1865 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1866 {
1867         struct scsi_device *sdev;
1868
1869         __shost_for_each_device(sdev, shost) {
1870                 if (channel == sdev_channel(sdev) &&
1871                     target == sdev_id(sdev))
1872                         __scsi_report_device_reset(sdev, NULL);
1873         }
1874 }
1875 EXPORT_SYMBOL(scsi_report_device_reset);
1876
1877 static void
1878 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1879 {
1880 }
1881
1882 /*
1883  * Function:    scsi_reset_provider
1884  *
1885  * Purpose:     Send requested reset to a bus or device at any phase.
1886  *
1887  * Arguments:   device  - device to send reset to
1888  *              flag - reset type (see scsi.h)
1889  *
1890  * Returns:     SUCCESS/FAILURE.
1891  *
1892  * Notes:       This is used by the SCSI Generic driver to provide
1893  *              Bus/Device reset capability.
1894  */
1895 int
1896 scsi_reset_provider(struct scsi_device *dev, int flag)
1897 {
1898         struct scsi_cmnd *scmd;
1899         struct Scsi_Host *shost = dev->host;
1900         struct request req;
1901         unsigned long flags;
1902         int rtn;
1903
1904         if (scsi_autopm_get_host(shost) < 0)
1905                 return FAILED;
1906
1907         scmd = scsi_get_command(dev, GFP_KERNEL);
1908         blk_rq_init(NULL, &req);
1909         scmd->request = &req;
1910
1911         scmd->cmnd = req.cmd;
1912
1913         scmd->scsi_done         = scsi_reset_provider_done_command;
1914         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1915
1916         scmd->cmd_len                   = 0;
1917
1918         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
1919
1920         spin_lock_irqsave(shost->host_lock, flags);
1921         shost->tmf_in_progress = 1;
1922         spin_unlock_irqrestore(shost->host_lock, flags);
1923
1924         switch (flag) {
1925         case SCSI_TRY_RESET_DEVICE:
1926                 rtn = scsi_try_bus_device_reset(scmd);
1927                 if (rtn == SUCCESS)
1928                         break;
1929                 /* FALLTHROUGH */
1930         case SCSI_TRY_RESET_TARGET:
1931                 rtn = scsi_try_target_reset(scmd);
1932                 if (rtn == SUCCESS)
1933                         break;
1934                 /* FALLTHROUGH */
1935         case SCSI_TRY_RESET_BUS:
1936                 rtn = scsi_try_bus_reset(scmd);
1937                 if (rtn == SUCCESS)
1938                         break;
1939                 /* FALLTHROUGH */
1940         case SCSI_TRY_RESET_HOST:
1941                 rtn = scsi_try_host_reset(scmd);
1942                 break;
1943         default:
1944                 rtn = FAILED;
1945         }
1946
1947         spin_lock_irqsave(shost->host_lock, flags);
1948         shost->tmf_in_progress = 0;
1949         spin_unlock_irqrestore(shost->host_lock, flags);
1950
1951         /*
1952          * be sure to wake up anyone who was sleeping or had their queue
1953          * suspended while we performed the TMF.
1954          */
1955         SCSI_LOG_ERROR_RECOVERY(3,
1956                 printk("%s: waking up host to restart after TMF\n",
1957                 __func__));
1958
1959         wake_up(&shost->host_wait);
1960
1961         scsi_run_host_queues(shost);
1962
1963         scsi_next_command(scmd);
1964         scsi_autopm_put_host(shost);
1965         return rtn;
1966 }
1967 EXPORT_SYMBOL(scsi_reset_provider);
1968
1969 /**
1970  * scsi_normalize_sense - normalize main elements from either fixed or
1971  *                      descriptor sense data format into a common format.
1972  *
1973  * @sense_buffer:       byte array containing sense data returned by device
1974  * @sb_len:             number of valid bytes in sense_buffer
1975  * @sshdr:              pointer to instance of structure that common
1976  *                      elements are written to.
1977  *
1978  * Notes:
1979  *      The "main elements" from sense data are: response_code, sense_key,
1980  *      asc, ascq and additional_length (only for descriptor format).
1981  *
1982  *      Typically this function can be called after a device has
1983  *      responded to a SCSI command with the CHECK_CONDITION status.
1984  *
1985  * Return value:
1986  *      1 if valid sense data information found, else 0;
1987  */
1988 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1989                          struct scsi_sense_hdr *sshdr)
1990 {
1991         if (!sense_buffer || !sb_len)
1992                 return 0;
1993
1994         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1995
1996         sshdr->response_code = (sense_buffer[0] & 0x7f);
1997
1998         if (!scsi_sense_valid(sshdr))
1999                 return 0;
2000
2001         if (sshdr->response_code >= 0x72) {
2002                 /*
2003                  * descriptor format
2004                  */
2005                 if (sb_len > 1)
2006                         sshdr->sense_key = (sense_buffer[1] & 0xf);
2007                 if (sb_len > 2)
2008                         sshdr->asc = sense_buffer[2];
2009                 if (sb_len > 3)
2010                         sshdr->ascq = sense_buffer[3];
2011                 if (sb_len > 7)
2012                         sshdr->additional_length = sense_buffer[7];
2013         } else {
2014                 /* 
2015                  * fixed format
2016                  */
2017                 if (sb_len > 2)
2018                         sshdr->sense_key = (sense_buffer[2] & 0xf);
2019                 if (sb_len > 7) {
2020                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2021                                          sb_len : (sense_buffer[7] + 8);
2022                         if (sb_len > 12)
2023                                 sshdr->asc = sense_buffer[12];
2024                         if (sb_len > 13)
2025                                 sshdr->ascq = sense_buffer[13];
2026                 }
2027         }
2028
2029         return 1;
2030 }
2031 EXPORT_SYMBOL(scsi_normalize_sense);
2032
2033 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2034                                  struct scsi_sense_hdr *sshdr)
2035 {
2036         return scsi_normalize_sense(cmd->sense_buffer,
2037                         SCSI_SENSE_BUFFERSIZE, sshdr);
2038 }
2039 EXPORT_SYMBOL(scsi_command_normalize_sense);
2040
2041 /**
2042  * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2043  * @sense_buffer:       byte array of descriptor format sense data
2044  * @sb_len:             number of valid bytes in sense_buffer
2045  * @desc_type:          value of descriptor type to find
2046  *                      (e.g. 0 -> information)
2047  *
2048  * Notes:
2049  *      only valid when sense data is in descriptor format
2050  *
2051  * Return value:
2052  *      pointer to start of (first) descriptor if found else NULL
2053  */
2054 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2055                                 int desc_type)
2056 {
2057         int add_sen_len, add_len, desc_len, k;
2058         const u8 * descp;
2059
2060         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2061                 return NULL;
2062         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2063                 return NULL;
2064         add_sen_len = (add_sen_len < (sb_len - 8)) ?
2065                         add_sen_len : (sb_len - 8);
2066         descp = &sense_buffer[8];
2067         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2068                 descp += desc_len;
2069                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2070                 desc_len = add_len + 2;
2071                 if (descp[0] == desc_type)
2072                         return descp;
2073                 if (add_len < 0) // short descriptor ??
2074                         break;
2075         }
2076         return NULL;
2077 }
2078 EXPORT_SYMBOL(scsi_sense_desc_find);
2079
2080 /**
2081  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2082  * @sense_buffer:       byte array of sense data
2083  * @sb_len:             number of valid bytes in sense_buffer
2084  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2085  *                      field will be placed if found.
2086  *
2087  * Return value:
2088  *      1 if information field found, 0 if not found.
2089  */
2090 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2091                             u64 * info_out)
2092 {
2093         int j;
2094         const u8 * ucp;
2095         u64 ull;
2096
2097         if (sb_len < 7)
2098                 return 0;
2099         switch (sense_buffer[0] & 0x7f) {
2100         case 0x70:
2101         case 0x71:
2102                 if (sense_buffer[0] & 0x80) {
2103                         *info_out = (sense_buffer[3] << 24) +
2104                                     (sense_buffer[4] << 16) +
2105                                     (sense_buffer[5] << 8) + sense_buffer[6];
2106                         return 1;
2107                 } else
2108                         return 0;
2109         case 0x72:
2110         case 0x73:
2111                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2112                                            0 /* info desc */);
2113                 if (ucp && (0xa == ucp[1])) {
2114                         ull = 0;
2115                         for (j = 0; j < 8; ++j) {
2116                                 if (j > 0)
2117                                         ull <<= 8;
2118                                 ull |= ucp[4 + j];
2119                         }
2120                         *info_out = ull;
2121                         return 1;
2122                 } else
2123                         return 0;
2124         default:
2125                 return 0;
2126         }
2127 }
2128 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2129
2130 /**
2131  * scsi_build_sense_buffer - build sense data in a buffer
2132  * @desc:       Sense format (non zero == descriptor format,
2133  *              0 == fixed format)
2134  * @buf:        Where to build sense data
2135  * @key:        Sense key
2136  * @asc:        Additional sense code
2137  * @ascq:       Additional sense code qualifier
2138  *
2139  **/
2140 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2141 {
2142         if (desc) {
2143                 buf[0] = 0x72;  /* descriptor, current */
2144                 buf[1] = key;
2145                 buf[2] = asc;
2146                 buf[3] = ascq;
2147                 buf[7] = 0;
2148         } else {
2149                 buf[0] = 0x70;  /* fixed, current */
2150                 buf[2] = key;
2151                 buf[7] = 0xa;
2152                 buf[12] = asc;
2153                 buf[13] = ascq;
2154         }
2155 }
2156 EXPORT_SYMBOL(scsi_build_sense_buffer);