[SCSI] lpfc: Removed obsolete fcp_eq_count and fcp_wq_count driver attributes
[deliverable/linux.git] / drivers / scsi / scsi_error.c
CommitLineData
1da177e4
LT
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
0bf8c869 6 * Leonard Zubkoff and David Miller at Linux Expo,
1da177e4
LT
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
0bf8c869 13 * minor cleanups.
1da177e4
LT
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
5a0e3ad6 19#include <linux/gfp.h>
1da177e4
LT
20#include <linux/timer.h>
21#include <linux/string.h>
1da177e4 22#include <linux/kernel.h>
83144186 23#include <linux/freezer.h>
c5478def 24#include <linux/kthread.h>
1da177e4
LT
25#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
fc73648a 28#include <linux/jiffies.h>
1da177e4
LT
29
30#include <scsi/scsi.h>
beb40487 31#include <scsi/scsi_cmnd.h>
1da177e4
LT
32#include <scsi/scsi_dbg.h>
33#include <scsi/scsi_device.h>
18a4d0a2 34#include <scsi/scsi_driver.h>
1da177e4 35#include <scsi/scsi_eh.h>
c829c394 36#include <scsi/scsi_transport.h>
1da177e4
LT
37#include <scsi/scsi_host.h>
38#include <scsi/scsi_ioctl.h>
1da177e4
LT
39
40#include "scsi_priv.h"
41#include "scsi_logging.h"
79ee8304 42#include "scsi_transport_api.h"
1da177e4 43
bf816235
KT
44#include <trace/events/scsi.h>
45
14216561
JB
46static void scsi_eh_done(struct scsi_cmnd *scmd);
47
1da177e4
LT
48/*
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
51 */
52#define BUS_RESET_SETTLE_TIME (10)
53#define HOST_RESET_SETTLE_TIME (10)
54
3eef6257
DJ
55static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
56
1da177e4
LT
57/* called with shost->host_lock held */
58void scsi_eh_wakeup(struct Scsi_Host *shost)
59{
60 if (shost->host_busy == shost->host_failed) {
bf816235 61 trace_scsi_eh_wakeup(shost);
3ed7a470 62 wake_up_process(shost->ehandler);
1da177e4
LT
63 SCSI_LOG_ERROR_RECOVERY(5,
64 printk("Waking error handler thread\n"));
65 }
66}
f8bbfc24
TH
67
68/**
69 * scsi_schedule_eh - schedule EH for SCSI host
70 * @shost: SCSI host to invoke error handling on.
71 *
72 * Schedule SCSI EH without scmd.
dc8875e1 73 */
f8bbfc24
TH
74void scsi_schedule_eh(struct Scsi_Host *shost)
75{
76 unsigned long flags;
77
78 spin_lock_irqsave(shost->host_lock, flags);
79
80 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
81 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
82 shost->host_eh_scheduled++;
83 scsi_eh_wakeup(shost);
84 }
85
86 spin_unlock_irqrestore(shost->host_lock, flags);
87}
88EXPORT_SYMBOL_GPL(scsi_schedule_eh);
1da177e4
LT
89
90/**
91 * scsi_eh_scmd_add - add scsi cmd to error handling.
92 * @scmd: scmd to run eh on.
93 * @eh_flag: optional SCSI_EH flag.
94 *
95 * Return value:
96 * 0 on failure.
dc8875e1 97 */
1da177e4
LT
98int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
99{
100 struct Scsi_Host *shost = scmd->device->host;
101 unsigned long flags;
939647ee 102 int ret = 0;
1da177e4 103
3ed7a470 104 if (!shost->ehandler)
1da177e4
LT
105 return 0;
106
107 spin_lock_irqsave(shost->host_lock, flags);
939647ee
JB
108 if (scsi_host_set_state(shost, SHOST_RECOVERY))
109 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
110 goto out_unlock;
1da177e4 111
939647ee 112 ret = 1;
3111b0d1 113 scmd->eh_eflags |= eh_flag;
1da177e4 114 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
1da177e4
LT
115 shost->host_failed++;
116 scsi_eh_wakeup(shost);
939647ee 117 out_unlock:
1da177e4 118 spin_unlock_irqrestore(shost->host_lock, flags);
939647ee 119 return ret;
1da177e4
LT
120}
121
1da177e4
LT
122/**
123 * scsi_times_out - Timeout function for normal scsi commands.
242f9dcb 124 * @req: request that is timing out.
1da177e4
LT
125 *
126 * Notes:
127 * We do not need to lock this. There is the potential for a race
128 * only in that the normal completion handling might run, but if the
129 * normal completion function determines that the timer has already
130 * fired, then it mustn't do anything.
dc8875e1 131 */
242f9dcb 132enum blk_eh_timer_return scsi_times_out(struct request *req)
1da177e4 133{
242f9dcb 134 struct scsi_cmnd *scmd = req->special;
242f9dcb 135 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
0bf8c869 136 struct Scsi_Host *host = scmd->device->host;
6c5f8ce1 137
bf816235 138 trace_scsi_dispatch_cmd_timeout(scmd);
1da177e4
LT
139 scsi_log_completion(scmd, TIMEOUT_ERROR);
140
0bf8c869
JJ
141 if (host->transportt->eh_timed_out)
142 rtn = host->transportt->eh_timed_out(scmd);
143 else if (host->hostt->eh_timed_out)
144 rtn = host->hostt->eh_timed_out(scmd);
1da177e4 145
18a4d0a2
MP
146 scmd->result |= DID_TIME_OUT << 16;
147
fa990781 148 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
18a4d0a2 149 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
fa990781 150 rtn = BLK_EH_HANDLED;
242f9dcb 151
fa990781 152 return rtn;
1da177e4
LT
153}
154
155/**
156 * scsi_block_when_processing_errors - Prevent cmds from being queued.
157 * @sdev: Device on which we are performing recovery.
158 *
159 * Description:
160 * We block until the host is out of error recovery, and then check to
161 * see whether the host or the device is offline.
162 *
163 * Return value:
164 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
dc8875e1 165 */
1da177e4
LT
166int scsi_block_when_processing_errors(struct scsi_device *sdev)
167{
168 int online;
169
939647ee 170 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
1da177e4
LT
171
172 online = scsi_device_online(sdev);
173
cadbd4a5 174 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
1da177e4
LT
175 online));
176
177 return online;
178}
179EXPORT_SYMBOL(scsi_block_when_processing_errors);
180
181#ifdef CONFIG_SCSI_LOGGING
182/**
183 * scsi_eh_prt_fail_stats - Log info on failures.
184 * @shost: scsi host being recovered.
185 * @work_q: Queue of scsi cmds to process.
dc8875e1 186 */
1da177e4
LT
187static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
188 struct list_head *work_q)
189{
190 struct scsi_cmnd *scmd;
191 struct scsi_device *sdev;
192 int total_failures = 0;
193 int cmd_failed = 0;
194 int cmd_cancel = 0;
195 int devices_failed = 0;
196
197 shost_for_each_device(sdev, shost) {
198 list_for_each_entry(scmd, work_q, eh_entry) {
199 if (scmd->device == sdev) {
200 ++total_failures;
3111b0d1 201 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
1da177e4 202 ++cmd_cancel;
0bf8c869 203 else
1da177e4
LT
204 ++cmd_failed;
205 }
206 }
207
208 if (cmd_cancel || cmd_failed) {
209 SCSI_LOG_ERROR_RECOVERY(3,
9ccfc756
JB
210 sdev_printk(KERN_INFO, sdev,
211 "%s: cmds failed: %d, cancel: %d\n",
cadbd4a5 212 __func__, cmd_failed,
9ccfc756 213 cmd_cancel));
1da177e4
LT
214 cmd_cancel = 0;
215 cmd_failed = 0;
216 ++devices_failed;
217 }
218 }
219
220 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
221 " devices require eh work\n",
0bf8c869 222 total_failures, devices_failed));
1da177e4
LT
223}
224#endif
225
226/**
227 * scsi_check_sense - Examine scsi cmd sense
228 * @scmd: Cmd to have sense checked.
229 *
230 * Return value:
87f14e65 231 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1da177e4
LT
232 *
233 * Notes:
234 * When a deferred error is detected the current command has
235 * not been executed and needs retrying.
dc8875e1 236 */
1da177e4
LT
237static int scsi_check_sense(struct scsi_cmnd *scmd)
238{
a6a8d9f8 239 struct scsi_device *sdev = scmd->device;
1da177e4
LT
240 struct scsi_sense_hdr sshdr;
241
242 if (! scsi_command_normalize_sense(scmd, &sshdr))
243 return FAILED; /* no valid sense data */
244
14216561
JB
245 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
246 /*
247 * nasty: for mid-layer issued TURs, we need to return the
248 * actual sense data without any recovery attempt. For eh
249 * issued ones, we need to try to recover and interpret
250 */
251 return SUCCESS;
252
1da177e4
LT
253 if (scsi_sense_is_deferred(&sshdr))
254 return NEEDS_RETRY;
255
a6a8d9f8
CS
256 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
257 sdev->scsi_dh_data->scsi_dh->check_sense) {
258 int rc;
259
260 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
261 if (rc != SCSI_RETURN_NOT_HANDLED)
262 return rc;
263 /* handler does not care. Drop down to default handling */
264 }
265
1da177e4
LT
266 /*
267 * Previous logic looked for FILEMARK, EOM or ILI which are
268 * mainly associated with tapes and returned SUCCESS.
269 */
270 if (sshdr.response_code == 0x70) {
271 /* fixed format */
272 if (scmd->sense_buffer[2] & 0xe0)
273 return SUCCESS;
274 } else {
275 /*
276 * descriptor format: look for "stream commands sense data
277 * descriptor" (see SSC-3). Assume single sense data
278 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
279 */
280 if ((sshdr.additional_length > 3) &&
281 (scmd->sense_buffer[8] == 0x4) &&
282 (scmd->sense_buffer[11] & 0xe0))
283 return SUCCESS;
284 }
285
286 switch (sshdr.sense_key) {
287 case NO_SENSE:
288 return SUCCESS;
289 case RECOVERED_ERROR:
290 return /* soft_error */ SUCCESS;
291
292 case ABORTED_COMMAND:
511e44f4
MP
293 if (sshdr.asc == 0x10) /* DIF */
294 return SUCCESS;
295
1da177e4
LT
296 return NEEDS_RETRY;
297 case NOT_READY:
298 case UNIT_ATTENTION:
299 /*
300 * if we are expecting a cc/ua because of a bus reset that we
301 * performed, treat this just as a retry. otherwise this is
302 * information that we should pass up to the upper-level driver
303 * so that we can deal with it there.
304 */
305 if (scmd->device->expecting_cc_ua) {
dfcf7775
TH
306 /*
307 * Because some device does not queue unit
308 * attentions correctly, we carefully check
309 * additional sense code and qualifier so as
310 * not to squash media change unit attention.
311 */
312 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
313 scmd->device->expecting_cc_ua = 0;
314 return NEEDS_RETRY;
315 }
1da177e4
LT
316 }
317 /*
0bf8c869 318 * if the device is in the process of becoming ready, we
1da177e4
LT
319 * should retry.
320 */
321 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
322 return NEEDS_RETRY;
323 /*
324 * if the device is not started, we need to wake
325 * the error handler to start the motor
326 */
327 if (scmd->device->allow_restart &&
328 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
329 return FAILED;
77a42297 330
6e49949c
MC
331 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
332 scmd_printk(KERN_WARNING, scmd,
333 "Warning! Received an indication that the "
334 "LUN assignments on this target have "
335 "changed. The Linux SCSI layer does not "
336 "automatically remap LUN assignments.\n");
337 else if (sshdr.asc == 0x3f)
338 scmd_printk(KERN_WARNING, scmd,
339 "Warning! Received an indication that the "
340 "operating parameters on this target have "
341 "changed. The Linux SCSI layer does not "
342 "automatically adjust these parameters.\n");
343
deb1cb63
SI
344 if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
345 scmd_printk(KERN_WARNING, scmd,
346 "Warning! Received an indication that the "
347 "LUN reached a thin provisioning soft "
348 "threshold.\n");
349
02e031cb
CH
350 /*
351 * Pass the UA upwards for a determination in the completion
352 * functions.
353 */
354 return SUCCESS;
1da177e4 355
63583cca 356 /* these are not supported */
a9d6ceb8
HR
357 case DATA_PROTECT:
358 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
359 /* Thin provisioning hard threshold reached */
360 set_host_byte(scmd, DID_ALLOC_FAILURE);
361 return SUCCESS;
362 }
1da177e4
LT
363 case COPY_ABORTED:
364 case VOLUME_OVERFLOW:
365 case MISCOMPARE:
63583cca 366 case BLANK_CHECK:
87f14e65
HR
367 set_host_byte(scmd, DID_TARGET_FAILURE);
368 return SUCCESS;
1da177e4
LT
369
370 case MEDIUM_ERROR:
fd1b494d
LT
371 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
372 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
373 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
7e782af5 374 set_host_byte(scmd, DID_MEDIUM_ERROR);
87f14e65 375 return SUCCESS;
fd1b494d 376 }
1da177e4
LT
377 return NEEDS_RETRY;
378
379 case HARDWARE_ERROR:
380 if (scmd->device->retry_hwerror)
bb0003c1 381 return ADD_TO_MLQUEUE;
1da177e4 382 else
87f14e65 383 set_host_byte(scmd, DID_TARGET_FAILURE);
1da177e4
LT
384
385 case ILLEGAL_REQUEST:
47ac56db
MS
386 if (sshdr.asc == 0x20 || /* Invalid command operation code */
387 sshdr.asc == 0x21 || /* Logical block address out of range */
388 sshdr.asc == 0x24 || /* Invalid field in cdb */
389 sshdr.asc == 0x26) { /* Parameter value invalid */
87f14e65 390 set_host_byte(scmd, DID_TARGET_FAILURE);
47ac56db
MS
391 }
392 return SUCCESS;
393
1da177e4
LT
394 default:
395 return SUCCESS;
396 }
397}
398
4a84067d
VD
399static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
400{
401 struct scsi_host_template *sht = sdev->host->hostt;
402 struct scsi_device *tmp_sdev;
403
404 if (!sht->change_queue_depth ||
405 sdev->queue_depth >= sdev->max_queue_depth)
406 return;
407
408 if (time_before(jiffies,
409 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
410 return;
411
412 if (time_before(jiffies,
413 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
414 return;
415
416 /*
417 * Walk all devices of a target and do
418 * ramp up on them.
419 */
420 shost_for_each_device(tmp_sdev, sdev->host) {
421 if (tmp_sdev->channel != sdev->channel ||
422 tmp_sdev->id != sdev->id ||
423 tmp_sdev->queue_depth == sdev->max_queue_depth)
424 continue;
425 /*
426 * call back into LLD to increase queue_depth by one
427 * with ramp up reason code.
428 */
429 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
430 SCSI_QDEPTH_RAMP_UP);
431 sdev->last_queue_ramp_up = jiffies;
432 }
433}
434
42a6a918
MC
435static void scsi_handle_queue_full(struct scsi_device *sdev)
436{
437 struct scsi_host_template *sht = sdev->host->hostt;
438 struct scsi_device *tmp_sdev;
439
440 if (!sht->change_queue_depth)
441 return;
442
443 shost_for_each_device(tmp_sdev, sdev->host) {
444 if (tmp_sdev->channel != sdev->channel ||
445 tmp_sdev->id != sdev->id)
446 continue;
447 /*
448 * We do not know the number of commands that were at
449 * the device when we got the queue full so we start
450 * from the highest possible value and work our way down.
451 */
452 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
453 SCSI_QDEPTH_QFULL);
454 }
455}
456
1da177e4
LT
457/**
458 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
459 * @scmd: SCSI cmd to examine.
460 *
461 * Notes:
462 * This is *only* called when we are examining the status of commands
463 * queued during error recovery. the main difference here is that we
464 * don't allow for the possibility of retries here, and we are a lot
465 * more restrictive about what we consider acceptable.
dc8875e1 466 */
1da177e4
LT
467static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
468{
469 /*
470 * first check the host byte, to see if there is anything in there
471 * that would indicate what we need to do.
472 */
473 if (host_byte(scmd->result) == DID_RESET) {
474 /*
475 * rats. we are already in the error handler, so we now
476 * get to try and figure out what to do next. if the sense
477 * is valid, we have a pretty good idea of what to do.
478 * if not, we mark it as FAILED.
479 */
480 return scsi_check_sense(scmd);
481 }
482 if (host_byte(scmd->result) != DID_OK)
483 return FAILED;
484
485 /*
486 * next, check the message byte.
487 */
488 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
489 return FAILED;
490
491 /*
492 * now, check the status byte to see if this indicates
493 * anything special.
494 */
495 switch (status_byte(scmd->result)) {
496 case GOOD:
4a84067d 497 scsi_handle_queue_ramp_up(scmd->device);
1da177e4
LT
498 case COMMAND_TERMINATED:
499 return SUCCESS;
500 case CHECK_CONDITION:
501 return scsi_check_sense(scmd);
502 case CONDITION_GOOD:
503 case INTERMEDIATE_GOOD:
504 case INTERMEDIATE_C_GOOD:
505 /*
506 * who knows? FIXME(eric)
507 */
508 return SUCCESS;
5f91bb05 509 case RESERVATION_CONFLICT:
67110dfd
JB
510 if (scmd->cmnd[0] == TEST_UNIT_READY)
511 /* it is a success, we probed the device and
512 * found it */
513 return SUCCESS;
514 /* otherwise, we failed to send the command */
515 return FAILED;
1da177e4 516 case QUEUE_FULL:
42a6a918
MC
517 scsi_handle_queue_full(scmd->device);
518 /* fall through */
519 case BUSY:
3eb3a928 520 return NEEDS_RETRY;
1da177e4
LT
521 default:
522 return FAILED;
523 }
524 return FAILED;
525}
526
1da177e4
LT
527/**
528 * scsi_eh_done - Completion function for error handling.
529 * @scmd: Cmd that is done.
dc8875e1 530 */
1da177e4
LT
531static void scsi_eh_done(struct scsi_cmnd *scmd)
532{
0bf8c869 533 struct completion *eh_action;
85631672 534
7dfdc9a5
CH
535 SCSI_LOG_ERROR_RECOVERY(3,
536 printk("%s scmd: %p result: %x\n",
cadbd4a5 537 __func__, scmd, scmd->result));
85631672
MR
538
539 eh_action = scmd->device->host->eh_action;
540 if (eh_action)
541 complete(eh_action);
1da177e4
LT
542}
543
292148f8
BK
544/**
545 * scsi_try_host_reset - ask host adapter to reset itself
c2b3ebd0 546 * @scmd: SCSI cmd to send host reset.
dc8875e1 547 */
292148f8
BK
548static int scsi_try_host_reset(struct scsi_cmnd *scmd)
549{
550 unsigned long flags;
551 int rtn;
0bf8c869
JJ
552 struct Scsi_Host *host = scmd->device->host;
553 struct scsi_host_template *hostt = host->hostt;
292148f8
BK
554
555 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
cadbd4a5 556 __func__));
292148f8 557
0bf8c869 558 if (!hostt->eh_host_reset_handler)
292148f8
BK
559 return FAILED;
560
0bf8c869 561 rtn = hostt->eh_host_reset_handler(scmd);
292148f8
BK
562
563 if (rtn == SUCCESS) {
0bf8c869 564 if (!hostt->skip_settle_delay)
292148f8 565 ssleep(HOST_RESET_SETTLE_TIME);
0bf8c869
JJ
566 spin_lock_irqsave(host->host_lock, flags);
567 scsi_report_bus_reset(host, scmd_channel(scmd));
568 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
569 }
570
571 return rtn;
572}
573
574/**
575 * scsi_try_bus_reset - ask host to perform a bus reset
576 * @scmd: SCSI cmd to send bus reset.
dc8875e1 577 */
292148f8
BK
578static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
579{
580 unsigned long flags;
581 int rtn;
0bf8c869
JJ
582 struct Scsi_Host *host = scmd->device->host;
583 struct scsi_host_template *hostt = host->hostt;
292148f8
BK
584
585 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
cadbd4a5 586 __func__));
292148f8 587
0bf8c869 588 if (!hostt->eh_bus_reset_handler)
292148f8
BK
589 return FAILED;
590
0bf8c869 591 rtn = hostt->eh_bus_reset_handler(scmd);
292148f8
BK
592
593 if (rtn == SUCCESS) {
0bf8c869 594 if (!hostt->skip_settle_delay)
292148f8 595 ssleep(BUS_RESET_SETTLE_TIME);
0bf8c869
JJ
596 spin_lock_irqsave(host->host_lock, flags);
597 scsi_report_bus_reset(host, scmd_channel(scmd));
598 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
599 }
600
601 return rtn;
602}
603
30bd7df8
MC
604static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
605{
606 sdev->was_reset = 1;
607 sdev->expecting_cc_ua = 1;
608}
609
610/**
611 * scsi_try_target_reset - Ask host to perform a target reset
612 * @scmd: SCSI cmd used to send a target reset
613 *
614 * Notes:
615 * There is no timeout for this operation. if this operation is
616 * unreliable for a given host, then the host itself needs to put a
617 * timer on it, and set the host back to a consistent state prior to
618 * returning.
619 */
620static int scsi_try_target_reset(struct scsi_cmnd *scmd)
621{
622 unsigned long flags;
623 int rtn;
0bf8c869
JJ
624 struct Scsi_Host *host = scmd->device->host;
625 struct scsi_host_template *hostt = host->hostt;
30bd7df8 626
0bf8c869 627 if (!hostt->eh_target_reset_handler)
30bd7df8
MC
628 return FAILED;
629
0bf8c869 630 rtn = hostt->eh_target_reset_handler(scmd);
30bd7df8 631 if (rtn == SUCCESS) {
0bf8c869 632 spin_lock_irqsave(host->host_lock, flags);
30bd7df8
MC
633 __starget_for_each_device(scsi_target(scmd->device), NULL,
634 __scsi_report_device_reset);
0bf8c869 635 spin_unlock_irqrestore(host->host_lock, flags);
30bd7df8
MC
636 }
637
638 return rtn;
639}
640
292148f8
BK
641/**
642 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
643 * @scmd: SCSI cmd used to send BDR
644 *
645 * Notes:
646 * There is no timeout for this operation. if this operation is
647 * unreliable for a given host, then the host itself needs to put a
648 * timer on it, and set the host back to a consistent state prior to
649 * returning.
dc8875e1 650 */
292148f8
BK
651static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
652{
653 int rtn;
0bf8c869 654 struct scsi_host_template *hostt = scmd->device->host->hostt;
292148f8 655
0bf8c869 656 if (!hostt->eh_device_reset_handler)
292148f8
BK
657 return FAILED;
658
0bf8c869 659 rtn = hostt->eh_device_reset_handler(scmd);
30bd7df8
MC
660 if (rtn == SUCCESS)
661 __scsi_report_device_reset(scmd->device, NULL);
292148f8
BK
662 return rtn;
663}
664
0bf8c869 665static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
292148f8 666{
0bf8c869 667 if (!hostt->eh_abort_handler)
292148f8
BK
668 return FAILED;
669
0bf8c869 670 return hostt->eh_abort_handler(scmd);
292148f8
BK
671}
672
292148f8
BK
673static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
674{
0bf8c869 675 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
292148f8 676 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
30bd7df8
MC
677 if (scsi_try_target_reset(scmd) != SUCCESS)
678 if (scsi_try_bus_reset(scmd) != SUCCESS)
679 scsi_try_host_reset(scmd);
292148f8
BK
680}
681
1da177e4 682/**
3b729f76 683 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
2dc611de 684 * @scmd: SCSI command structure to hijack
e1c23468 685 * @ses: structure to save restore information
55db6c1b 686 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
64a87b24 687 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
55db6c1b 688 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
2dc611de 689 *
e1c23468 690 * This function is used to save a scsi command information before re-execution
55db6c1b
BH
691 * as part of the error recovery process. If @sense_bytes is 0 the command
692 * sent must be one that does not transfer any data. If @sense_bytes != 0
693 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
694 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
dc8875e1 695 */
e1c23468
BH
696void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
697 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1da177e4 698{
f59114b7 699 struct scsi_device *sdev = scmd->device;
1da177e4 700
631c228c
CH
701 /*
702 * We need saved copies of a number of fields - this is because
703 * error handling may need to overwrite these with different values
704 * to run different commands, and once error handling is complete,
705 * we will need to restore these values prior to running the actual
706 * command.
707 */
e1c23468 708 ses->cmd_len = scmd->cmd_len;
64a87b24 709 ses->cmnd = scmd->cmnd;
e1c23468 710 ses->data_direction = scmd->sc_data_direction;
30b0c37b 711 ses->sdb = scmd->sdb;
6f9a35e2 712 ses->next_rq = scmd->request->next_rq;
e1c23468 713 ses->result = scmd->result;
12265709 714 ses->underflow = scmd->underflow;
db007fc5 715 ses->prot_op = scmd->prot_op;
631c228c 716
db007fc5 717 scmd->prot_op = SCSI_PROT_NORMAL;
64a87b24
BH
718 scmd->cmnd = ses->eh_cmnd;
719 memset(scmd->cmnd, 0, BLK_MAX_CDB);
30b0c37b 720 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
6f9a35e2 721 scmd->request->next_rq = NULL;
30b0c37b 722
55db6c1b 723 if (sense_bytes) {
30b0c37b
BH
724 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
725 sense_bytes);
e1c23468 726 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
30b0c37b
BH
727 scmd->sdb.length);
728 scmd->sdb.table.sgl = &ses->sense_sgl;
55db6c1b 729 scmd->sc_data_direction = DMA_FROM_DEVICE;
30b0c37b 730 scmd->sdb.table.nents = 1;
55db6c1b 731 scmd->cmnd[0] = REQUEST_SENSE;
30b0c37b 732 scmd->cmnd[4] = scmd->sdb.length;
55db6c1b 733 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
631c228c 734 } else {
631c228c 735 scmd->sc_data_direction = DMA_NONE;
55db6c1b 736 if (cmnd) {
64a87b24 737 BUG_ON(cmnd_size > BLK_MAX_CDB);
55db6c1b
BH
738 memcpy(scmd->cmnd, cmnd, cmnd_size);
739 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
740 }
631c228c
CH
741 }
742
743 scmd->underflow = 0;
631c228c 744
55db6c1b 745 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1da177e4 746 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b7 747 (sdev->lun << 5 & 0xe0);
1da177e4 748
631c228c
CH
749 /*
750 * Zero the sense buffer. The scsi spec mandates that any
751 * untransferred sense data should be interpreted as being zero.
752 */
b80ca4f7 753 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
e1c23468
BH
754}
755EXPORT_SYMBOL(scsi_eh_prep_cmnd);
756
757/**
3b729f76 758 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
e1c23468 759 * @scmd: SCSI command structure to restore
477e608c 760 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
e1c23468 761 *
477e608c 762 * Undo any damage done by above scsi_eh_prep_cmnd().
dc8875e1 763 */
e1c23468
BH
764void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
765{
766 /*
767 * Restore original data
768 */
769 scmd->cmd_len = ses->cmd_len;
64a87b24 770 scmd->cmnd = ses->cmnd;
e1c23468 771 scmd->sc_data_direction = ses->data_direction;
30b0c37b 772 scmd->sdb = ses->sdb;
6f9a35e2 773 scmd->request->next_rq = ses->next_rq;
e1c23468 774 scmd->result = ses->result;
12265709 775 scmd->underflow = ses->underflow;
db007fc5 776 scmd->prot_op = ses->prot_op;
e1c23468
BH
777}
778EXPORT_SYMBOL(scsi_eh_restore_cmnd);
631c228c 779
e1c23468 780/**
3b729f76 781 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
e1c23468
BH
782 * @scmd: SCSI command structure to hijack
783 * @cmnd: CDB to send
784 * @cmnd_size: size in bytes of @cmnd
785 * @timeout: timeout for this request
786 * @sense_bytes: size of sense data to copy or 0
787 *
788 * This function is used to send a scsi command down to a target device
789 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
790 *
791 * Return value:
792 * SUCCESS or FAILED or NEEDS_RETRY
dc8875e1 793 */
e1c23468
BH
794static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
795 int cmnd_size, int timeout, unsigned sense_bytes)
796{
797 struct scsi_device *sdev = scmd->device;
798 struct Scsi_Host *shost = sdev->host;
799 DECLARE_COMPLETION_ONSTACK(done);
fc73648a 800 unsigned long timeleft = timeout;
e1c23468 801 struct scsi_eh_save ses;
fc73648a 802 const unsigned long stall_for = msecs_to_jiffies(100);
e1c23468
BH
803 int rtn;
804
fc73648a 805retry:
e1c23468 806 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
7dfdc9a5 807 shost->eh_action = &done;
1da177e4 808
1da177e4 809 scsi_log_send(scmd);
f281233d 810 scmd->scsi_done = scsi_eh_done;
fc73648a
HR
811 rtn = shost->hostt->queuecommand(shost, scmd);
812 if (rtn) {
813 if (timeleft > stall_for) {
814 scsi_eh_restore_cmnd(scmd, &ses);
815 timeleft -= stall_for;
816 msleep(jiffies_to_msecs(stall_for));
817 goto retry;
818 }
819 /* signal not to enter either branch of the if () below */
820 timeleft = 0;
821 rtn = NEEDS_RETRY;
822 } else {
823 timeleft = wait_for_completion_timeout(&done, timeout);
824 }
1da177e4 825
f59114b7 826 shost->eh_action = NULL;
1da177e4 827
fc73648a 828 scsi_log_completion(scmd, rtn);
1da177e4 829
7dfdc9a5
CH
830 SCSI_LOG_ERROR_RECOVERY(3,
831 printk("%s: scmd: %p, timeleft: %ld\n",
cadbd4a5 832 __func__, scmd, timeleft));
1da177e4
LT
833
834 /*
fc73648a
HR
835 * If there is time left scsi_eh_done got called, and we will examine
836 * the actual status codes to see whether the command actually did
837 * complete normally, else if we have a zero return and no time left,
838 * the command must still be pending, so abort it and return FAILED.
839 * If we never actually managed to issue the command, because
840 * ->queuecommand() kept returning non zero, use the rtn = FAILED
841 * value above (so don't execute either branch of the if)
1da177e4 842 */
7dfdc9a5 843 if (timeleft) {
1da177e4
LT
844 rtn = scsi_eh_completed_normally(scmd);
845 SCSI_LOG_ERROR_RECOVERY(3,
846 printk("%s: scsi_eh_completed_normally %x\n",
cadbd4a5 847 __func__, rtn));
7dfdc9a5 848
1da177e4
LT
849 switch (rtn) {
850 case SUCCESS:
851 case NEEDS_RETRY:
852 case FAILED:
853 break;
6e883b0e
HR
854 case ADD_TO_MLQUEUE:
855 rtn = NEEDS_RETRY;
856 break;
1da177e4
LT
857 default:
858 rtn = FAILED;
859 break;
860 }
fc73648a 861 } else if (!rtn) {
292148f8 862 scsi_abort_eh_cmnd(scmd);
7dfdc9a5 863 rtn = FAILED;
1da177e4
LT
864 }
865
e1c23468 866 scsi_eh_restore_cmnd(scmd, &ses);
18a4d0a2 867
329a402c
LZ
868 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
869 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
870 if (sdrv->eh_action)
871 rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
872 }
18a4d0a2 873
1da177e4
LT
874 return rtn;
875}
876
877/**
878 * scsi_request_sense - Request sense data from a particular target.
879 * @scmd: SCSI cmd for request sense.
880 *
881 * Notes:
882 * Some hosts automatically obtain this information, others require
883 * that we obtain it on our own. This function will *not* return until
884 * the command either times out, or it completes.
dc8875e1 885 */
1da177e4
LT
886static int scsi_request_sense(struct scsi_cmnd *scmd)
887{
0816c925 888 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1da177e4
LT
889}
890
891/**
892 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
893 * @scmd: Original SCSI cmd that eh has finished.
894 * @done_q: Queue for processed commands.
895 *
896 * Notes:
897 * We don't want to use the normal command completion while we are are
898 * still handling errors - it may cause other commands to be queued,
eb44820c 899 * and that would disturb what we are doing. Thus we really want to
1da177e4
LT
900 * keep a list of pending commands for final completion, and once we
901 * are ready to leave error handling we handle completion for real.
dc8875e1 902 */
041c5fc3 903void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1da177e4
LT
904{
905 scmd->device->host->host_failed--;
3111b0d1 906 scmd->eh_eflags = 0;
1da177e4
LT
907 list_move_tail(&scmd->eh_entry, done_q);
908}
041c5fc3 909EXPORT_SYMBOL(scsi_eh_finish_cmd);
1da177e4
LT
910
911/**
912 * scsi_eh_get_sense - Get device sense data.
913 * @work_q: Queue of commands to process.
eb44820c 914 * @done_q: Queue of processed commands.
1da177e4
LT
915 *
916 * Description:
917 * See if we need to request sense information. if so, then get it
0bf8c869 918 * now, so we have a better idea of what to do.
1da177e4
LT
919 *
920 * Notes:
921 * This has the unfortunate side effect that if a shost adapter does
eb44820c 922 * not automatically request sense information, we end up shutting
1da177e4
LT
923 * it down before we request it.
924 *
925 * All drivers should request sense information internally these days,
926 * so for now all I have to say is tough noogies if you end up in here.
927 *
928 * XXX: Long term this code should go away, but that needs an audit of
929 * all LLDDs first.
dc8875e1 930 */
dca84e46
DW
931int scsi_eh_get_sense(struct list_head *work_q,
932 struct list_head *done_q)
1da177e4 933{
937abeaa 934 struct scsi_cmnd *scmd, *next;
1da177e4
LT
935 int rtn;
936
937abeaa 937 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3111b0d1 938 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1da177e4
LT
939 SCSI_SENSE_VALID(scmd))
940 continue;
941
3bf743e7
JG
942 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
943 "%s: requesting sense\n",
944 current->comm));
1da177e4
LT
945 rtn = scsi_request_sense(scmd);
946 if (rtn != SUCCESS)
947 continue;
948
949 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
950 " result %x\n", scmd,
951 scmd->result));
952 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
953
954 rtn = scsi_decide_disposition(scmd);
955
956 /*
957 * if the result was normal, then just pass it along to the
958 * upper level.
959 */
960 if (rtn == SUCCESS)
961 /* we don't want this command reissued, just
962 * finished with the sense data, so set
963 * retries to the max allowed to ensure it
964 * won't get reissued */
965 scmd->retries = scmd->allowed;
966 else if (rtn != NEEDS_RETRY)
967 continue;
968
969 scsi_eh_finish_cmd(scmd, done_q);
970 }
971
972 return list_empty(work_q);
973}
dca84e46 974EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1da177e4 975
1da177e4
LT
976/**
977 * scsi_eh_tur - Send TUR to device.
eb44820c 978 * @scmd: &scsi_cmnd to send TUR
1da177e4
LT
979 *
980 * Return value:
981 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 982 */
1da177e4
LT
983static int scsi_eh_tur(struct scsi_cmnd *scmd)
984{
985 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
986 int retry_cnt = 1, rtn;
987
988retry_tur:
0816c925
MP
989 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
990 scmd->device->eh_timeout, 0);
1da177e4 991
1da177e4 992 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
cadbd4a5 993 __func__, scmd, rtn));
631c228c
CH
994
995 switch (rtn) {
996 case NEEDS_RETRY:
1da177e4
LT
997 if (retry_cnt--)
998 goto retry_tur;
631c228c
CH
999 /*FALLTHRU*/
1000 case SUCCESS:
e47373ec 1001 return 0;
631c228c
CH
1002 default:
1003 return 1;
e47373ec 1004 }
1da177e4
LT
1005}
1006
3eef6257
DJ
1007/**
1008 * scsi_eh_test_devices - check if devices are responding from error recovery.
1009 * @cmd_list: scsi commands in error recovery.
1010 * @work_q: queue for commands which still need more error recovery
1011 * @done_q: queue for commands which are finished
1012 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1013 *
1014 * Decription:
1015 * Tests if devices are in a working state. Commands to devices now in
1016 * a working state are sent to the done_q while commands to devices which
1017 * are still failing to respond are returned to the work_q for more
1018 * processing.
1019 **/
1020static int scsi_eh_test_devices(struct list_head *cmd_list,
1021 struct list_head *work_q,
1022 struct list_head *done_q, int try_stu)
1023{
1024 struct scsi_cmnd *scmd, *next;
1025 struct scsi_device *sdev;
1026 int finish_cmds;
1027
1028 while (!list_empty(cmd_list)) {
1029 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1030 sdev = scmd->device;
1031
1032 finish_cmds = !scsi_device_online(scmd->device) ||
1033 (try_stu && !scsi_eh_try_stu(scmd) &&
1034 !scsi_eh_tur(scmd)) ||
1035 !scsi_eh_tur(scmd);
1036
1037 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1038 if (scmd->device == sdev) {
1039 if (finish_cmds)
1040 scsi_eh_finish_cmd(scmd, done_q);
1041 else
1042 list_move_tail(&scmd->eh_entry, work_q);
1043 }
1044 }
1045 return list_empty(work_q);
1046}
1047
1048
1da177e4 1049/**
eb44820c
RL
1050 * scsi_eh_abort_cmds - abort pending commands.
1051 * @work_q: &list_head for pending commands.
1052 * @done_q: &list_head for processed commands.
1da177e4
LT
1053 *
1054 * Decription:
1055 * Try and see whether or not it makes sense to try and abort the
eb44820c
RL
1056 * running command. This only works out to be the case if we have one
1057 * command that has timed out. If the command simply failed, it makes
1da177e4
LT
1058 * no sense to try and abort the command, since as far as the shost
1059 * adapter is concerned, it isn't running.
dc8875e1 1060 */
1da177e4
LT
1061static int scsi_eh_abort_cmds(struct list_head *work_q,
1062 struct list_head *done_q)
1063{
937abeaa 1064 struct scsi_cmnd *scmd, *next;
3eef6257 1065 LIST_HEAD(check_list);
1da177e4
LT
1066 int rtn;
1067
937abeaa 1068 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3111b0d1 1069 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1da177e4
LT
1070 continue;
1071 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1072 "0x%p\n", current->comm,
1073 scmd));
0bf8c869 1074 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
2f2eb587 1075 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
3111b0d1 1076 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
3eef6257 1077 if (rtn == FAST_IO_FAIL)
1da177e4 1078 scsi_eh_finish_cmd(scmd, done_q);
3eef6257
DJ
1079 else
1080 list_move_tail(&scmd->eh_entry, &check_list);
1da177e4
LT
1081 } else
1082 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1083 " cmd failed:"
1084 "0x%p\n",
1085 current->comm,
1086 scmd));
1087 }
1088
3eef6257 1089 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1090}
1091
1da177e4
LT
1092/**
1093 * scsi_eh_try_stu - Send START_UNIT to device.
eb44820c 1094 * @scmd: &scsi_cmnd to send START_UNIT
1da177e4
LT
1095 *
1096 * Return value:
1097 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1098 */
1da177e4
LT
1099static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1100{
1101 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1da177e4 1102
631c228c 1103 if (scmd->device->allow_restart) {
ed773e66
BK
1104 int i, rtn = NEEDS_RETRY;
1105
1106 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
9728c081 1107 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1da177e4 1108
631c228c
CH
1109 if (rtn == SUCCESS)
1110 return 0;
1111 }
1da177e4 1112
1da177e4
LT
1113 return 1;
1114}
1115
1116 /**
1117 * scsi_eh_stu - send START_UNIT if needed
eb44820c
RL
1118 * @shost: &scsi host being recovered.
1119 * @work_q: &list_head for pending commands.
1120 * @done_q: &list_head for processed commands.
1da177e4
LT
1121 *
1122 * Notes:
1123 * If commands are failing due to not ready, initializing command required,
0bf8c869 1124 * try revalidating the device, which will end up sending a start unit.
dc8875e1 1125 */
1da177e4
LT
1126static int scsi_eh_stu(struct Scsi_Host *shost,
1127 struct list_head *work_q,
1128 struct list_head *done_q)
1129{
937abeaa 1130 struct scsi_cmnd *scmd, *stu_scmd, *next;
1da177e4
LT
1131 struct scsi_device *sdev;
1132
1133 shost_for_each_device(sdev, shost) {
1134 stu_scmd = NULL;
1135 list_for_each_entry(scmd, work_q, eh_entry)
1136 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1137 scsi_check_sense(scmd) == FAILED ) {
1138 stu_scmd = scmd;
1139 break;
1140 }
1141
1142 if (!stu_scmd)
1143 continue;
1144
1145 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1146 " 0x%p\n", current->comm, sdev));
1147
1148 if (!scsi_eh_try_stu(stu_scmd)) {
1149 if (!scsi_device_online(sdev) ||
1150 !scsi_eh_tur(stu_scmd)) {
937abeaa
CH
1151 list_for_each_entry_safe(scmd, next,
1152 work_q, eh_entry) {
1da177e4
LT
1153 if (scmd->device == sdev)
1154 scsi_eh_finish_cmd(scmd, done_q);
1155 }
1156 }
1157 } else {
1158 SCSI_LOG_ERROR_RECOVERY(3,
1159 printk("%s: START_UNIT failed to sdev:"
1160 " 0x%p\n", current->comm, sdev));
1161 }
1162 }
1163
1164 return list_empty(work_q);
1165}
1166
1167
1168/**
1169 * scsi_eh_bus_device_reset - send bdr if needed
1170 * @shost: scsi host being recovered.
eb44820c
RL
1171 * @work_q: &list_head for pending commands.
1172 * @done_q: &list_head for processed commands.
1da177e4
LT
1173 *
1174 * Notes:
eb44820c 1175 * Try a bus device reset. Still, look to see whether we have multiple
1da177e4
LT
1176 * devices that are jammed or not - if we have multiple devices, it
1177 * makes no sense to try bus_device_reset - we really would need to try
0bf8c869 1178 * a bus_reset instead.
dc8875e1 1179 */
1da177e4
LT
1180static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1181 struct list_head *work_q,
1182 struct list_head *done_q)
1183{
937abeaa 1184 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1da177e4
LT
1185 struct scsi_device *sdev;
1186 int rtn;
1187
1188 shost_for_each_device(sdev, shost) {
1189 bdr_scmd = NULL;
1190 list_for_each_entry(scmd, work_q, eh_entry)
1191 if (scmd->device == sdev) {
1192 bdr_scmd = scmd;
1193 break;
1194 }
1195
1196 if (!bdr_scmd)
1197 continue;
1198
1199 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1200 " 0x%p\n", current->comm,
1201 sdev));
1202 rtn = scsi_try_bus_device_reset(bdr_scmd);
2f2eb587 1203 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1da177e4 1204 if (!scsi_device_online(sdev) ||
2f2eb587 1205 rtn == FAST_IO_FAIL ||
1da177e4 1206 !scsi_eh_tur(bdr_scmd)) {
937abeaa
CH
1207 list_for_each_entry_safe(scmd, next,
1208 work_q, eh_entry) {
1da177e4
LT
1209 if (scmd->device == sdev)
1210 scsi_eh_finish_cmd(scmd,
1211 done_q);
1212 }
1213 }
1214 } else {
1215 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1216 " failed sdev:"
1217 "0x%p\n",
1218 current->comm,
1219 sdev));
1220 }
1221 }
1222
1223 return list_empty(work_q);
1224}
1225
30bd7df8
MC
1226/**
1227 * scsi_eh_target_reset - send target reset if needed
1228 * @shost: scsi host being recovered.
1229 * @work_q: &list_head for pending commands.
1230 * @done_q: &list_head for processed commands.
1231 *
1232 * Notes:
1233 * Try a target reset.
1234 */
1235static int scsi_eh_target_reset(struct Scsi_Host *shost,
1236 struct list_head *work_q,
1237 struct list_head *done_q)
1238{
98db5195 1239 LIST_HEAD(tmp_list);
3eef6257 1240 LIST_HEAD(check_list);
30bd7df8 1241
98db5195
JB
1242 list_splice_init(work_q, &tmp_list);
1243
1244 while (!list_empty(&tmp_list)) {
1245 struct scsi_cmnd *next, *scmd;
1246 int rtn;
1247 unsigned int id;
1248
1249 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1250 id = scmd_id(scmd);
30bd7df8
MC
1251
1252 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1253 "to target %d\n",
1254 current->comm, id));
98db5195
JB
1255 rtn = scsi_try_target_reset(scmd);
1256 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
30bd7df8
MC
1257 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1258 " failed target: "
1259 "%d\n",
1260 current->comm, id));
98db5195
JB
1261 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1262 if (scmd_id(scmd) != id)
1263 continue;
1264
3eef6257
DJ
1265 if (rtn == SUCCESS)
1266 list_move_tail(&scmd->eh_entry, &check_list);
1267 else if (rtn == FAST_IO_FAIL)
98db5195
JB
1268 scsi_eh_finish_cmd(scmd, done_q);
1269 else
1270 /* push back on work queue for further processing */
1271 list_move(&scmd->eh_entry, work_q);
1272 }
1273 }
30bd7df8 1274
3eef6257 1275 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
30bd7df8
MC
1276}
1277
1da177e4 1278/**
0bf8c869 1279 * scsi_eh_bus_reset - send a bus reset
eb44820c
RL
1280 * @shost: &scsi host being recovered.
1281 * @work_q: &list_head for pending commands.
1282 * @done_q: &list_head for processed commands.
dc8875e1 1283 */
1da177e4
LT
1284static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1285 struct list_head *work_q,
1286 struct list_head *done_q)
1287{
937abeaa 1288 struct scsi_cmnd *scmd, *chan_scmd, *next;
3eef6257 1289 LIST_HEAD(check_list);
1da177e4
LT
1290 unsigned int channel;
1291 int rtn;
1292
1293 /*
1294 * we really want to loop over the various channels, and do this on
1295 * a channel by channel basis. we should also check to see if any
1296 * of the failed commands are on soft_reset devices, and if so, skip
0bf8c869 1297 * the reset.
1da177e4
LT
1298 */
1299
1300 for (channel = 0; channel <= shost->max_channel; channel++) {
1301 chan_scmd = NULL;
1302 list_for_each_entry(scmd, work_q, eh_entry) {
422c0d61 1303 if (channel == scmd_channel(scmd)) {
1da177e4
LT
1304 chan_scmd = scmd;
1305 break;
1306 /*
1307 * FIXME add back in some support for
1308 * soft_reset devices.
1309 */
1310 }
1311 }
1312
1313 if (!chan_scmd)
1314 continue;
1315 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1316 " %d\n", current->comm,
1317 channel));
1318 rtn = scsi_try_bus_reset(chan_scmd);
2f2eb587 1319 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
937abeaa 1320 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3eef6257
DJ
1321 if (channel == scmd_channel(scmd)) {
1322 if (rtn == FAST_IO_FAIL)
1da177e4
LT
1323 scsi_eh_finish_cmd(scmd,
1324 done_q);
3eef6257
DJ
1325 else
1326 list_move_tail(&scmd->eh_entry,
1327 &check_list);
1328 }
1da177e4
LT
1329 }
1330 } else {
1331 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1332 " failed chan: %d\n",
1333 current->comm,
1334 channel));
1335 }
1336 }
3eef6257 1337 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1338}
1339
1340/**
0bf8c869 1341 * scsi_eh_host_reset - send a host reset
1da177e4
LT
1342 * @work_q: list_head for processed commands.
1343 * @done_q: list_head for processed commands.
dc8875e1 1344 */
1da177e4
LT
1345static int scsi_eh_host_reset(struct list_head *work_q,
1346 struct list_head *done_q)
1347{
937abeaa 1348 struct scsi_cmnd *scmd, *next;
3eef6257 1349 LIST_HEAD(check_list);
1da177e4 1350 int rtn;
1da177e4
LT
1351
1352 if (!list_empty(work_q)) {
1353 scmd = list_entry(work_q->next,
1354 struct scsi_cmnd, eh_entry);
1355
1356 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1357 , current->comm));
1358
1359 rtn = scsi_try_host_reset(scmd);
3eef6257
DJ
1360 if (rtn == SUCCESS) {
1361 list_splice_init(work_q, &check_list);
1362 } else if (rtn == FAST_IO_FAIL) {
937abeaa 1363 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1da177e4
LT
1364 scsi_eh_finish_cmd(scmd, done_q);
1365 }
1366 } else {
1367 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1368 " failed\n",
1369 current->comm));
1370 }
1371 }
3eef6257 1372 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1da177e4
LT
1373}
1374
1375/**
1376 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1377 * @work_q: list_head for processed commands.
1378 * @done_q: list_head for processed commands.
dc8875e1 1379 */
1da177e4
LT
1380static void scsi_eh_offline_sdevs(struct list_head *work_q,
1381 struct list_head *done_q)
1382{
937abeaa 1383 struct scsi_cmnd *scmd, *next;
1da177e4 1384
937abeaa 1385 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
31765d7d
MW
1386 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1387 "not ready after error recovery\n");
1da177e4 1388 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
3111b0d1 1389 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1da177e4
LT
1390 /*
1391 * FIXME: Handle lost cmds.
1392 */
1393 }
1394 scsi_eh_finish_cmd(scmd, done_q);
1395 }
1396 return;
1397}
1398
4a27446f
MC
1399/**
1400 * scsi_noretry_cmd - determinte if command should be failed fast
1401 * @scmd: SCSI cmd to examine.
1402 */
1403int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1404{
1405 switch (host_byte(scmd->result)) {
1406 case DID_OK:
1407 break;
1408 case DID_BUS_BUSY:
33659ebb 1409 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
4a27446f 1410 case DID_PARITY:
33659ebb 1411 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
4a27446f
MC
1412 case DID_ERROR:
1413 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1414 status_byte(scmd->result) == RESERVATION_CONFLICT)
1415 return 0;
1416 /* fall through */
1417 case DID_SOFT_ERROR:
33659ebb 1418 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
4a27446f
MC
1419 }
1420
1421 switch (status_byte(scmd->result)) {
1422 case CHECK_CONDITION:
1423 /*
1424 * assume caller has checked sense and determinted
1425 * the check condition was retryable.
1426 */
e96f6abe
FT
1427 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1428 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1429 return 1;
4a27446f
MC
1430 }
1431
1432 return 0;
1433}
1434
1da177e4
LT
1435/**
1436 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1437 * @scmd: SCSI cmd to examine.
1438 *
1439 * Notes:
1440 * This is *only* called when we are examining the status after sending
1441 * out the actual data command. any commands that are queued for error
1442 * recovery (e.g. test_unit_ready) do *not* come through here.
1443 *
1444 * When this routine returns failed, it means the error handler thread
1445 * is woken. In cases where the error code indicates an error that
1446 * doesn't require the error handler read (i.e. we don't need to
1447 * abort/reset), this function should return SUCCESS.
dc8875e1 1448 */
1da177e4
LT
1449int scsi_decide_disposition(struct scsi_cmnd *scmd)
1450{
1451 int rtn;
1452
1453 /*
1454 * if the device is offline, then we clearly just pass the result back
1455 * up to the top level.
1456 */
1457 if (!scsi_device_online(scmd->device)) {
1458 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1459 " as SUCCESS\n",
cadbd4a5 1460 __func__));
1da177e4
LT
1461 return SUCCESS;
1462 }
1463
1464 /*
1465 * first check the host byte, to see if there is anything in there
1466 * that would indicate what we need to do.
1467 */
1468 switch (host_byte(scmd->result)) {
1469 case DID_PASSTHROUGH:
1470 /*
1471 * no matter what, pass this through to the upper layer.
1472 * nuke this special code so that it looks like we are saying
1473 * did_ok.
1474 */
1475 scmd->result &= 0xff00ffff;
1476 return SUCCESS;
1477 case DID_OK:
1478 /*
1479 * looks good. drop through, and check the next byte.
1480 */
1481 break;
1482 case DID_NO_CONNECT:
1483 case DID_BAD_TARGET:
1484 case DID_ABORT:
1485 /*
1486 * note - this means that we just report the status back
1487 * to the top level driver, not that we actually think
1488 * that it indicates SUCCESS.
1489 */
1490 return SUCCESS;
1491 /*
1492 * when the low level driver returns did_soft_error,
0bf8c869 1493 * it is responsible for keeping an internal retry counter
1da177e4
LT
1494 * in order to avoid endless loops (db)
1495 *
1496 * actually this is a bug in this function here. we should
1497 * be mindful of the maximum number of retries specified
1498 * and not get stuck in a loop.
1499 */
1500 case DID_SOFT_ERROR:
1501 goto maybe_retry;
1502 case DID_IMM_RETRY:
1503 return NEEDS_RETRY;
1504
bf341919 1505 case DID_REQUEUE:
1506 return ADD_TO_MLQUEUE;
a4dfaa6f
MC
1507 case DID_TRANSPORT_DISRUPTED:
1508 /*
1509 * LLD/transport was disrupted during processing of the IO.
1510 * The transport class is now blocked/blocking,
1511 * and the transport will decide what to do with the IO
939c2288
MC
1512 * based on its timers and recovery capablilities if
1513 * there are enough retries.
a4dfaa6f 1514 */
939c2288 1515 goto maybe_retry;
a4dfaa6f
MC
1516 case DID_TRANSPORT_FAILFAST:
1517 /*
1518 * The transport decided to failfast the IO (most likely
1519 * the fast io fail tmo fired), so send IO directly upwards.
1520 */
1521 return SUCCESS;
1da177e4
LT
1522 case DID_ERROR:
1523 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1524 status_byte(scmd->result) == RESERVATION_CONFLICT)
1525 /*
1526 * execute reservation conflict processing code
1527 * lower down
1528 */
1529 break;
1530 /* fallthrough */
1da177e4
LT
1531 case DID_BUS_BUSY:
1532 case DID_PARITY:
1533 goto maybe_retry;
1534 case DID_TIME_OUT:
1535 /*
1536 * when we scan the bus, we get timeout messages for
1537 * these commands if there is no device available.
1538 * other hosts report did_no_connect for the same thing.
1539 */
1540 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1541 scmd->cmnd[0] == INQUIRY)) {
1542 return SUCCESS;
1543 } else {
1544 return FAILED;
1545 }
1546 case DID_RESET:
1547 return SUCCESS;
1548 default:
1549 return FAILED;
1550 }
1551
1552 /*
1553 * next, check the message byte.
1554 */
1555 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1556 return FAILED;
1557
1558 /*
1559 * check the status byte to see if this indicates anything special.
1560 */
1561 switch (status_byte(scmd->result)) {
1562 case QUEUE_FULL:
42a6a918 1563 scsi_handle_queue_full(scmd->device);
1da177e4
LT
1564 /*
1565 * the case of trying to send too many commands to a
1566 * tagged queueing device.
1567 */
1568 case BUSY:
1569 /*
1570 * device can't talk to us at the moment. Should only
1571 * occur (SAM-3) when the task queue is empty, so will cause
1572 * the empty queue handling to trigger a stall in the
1573 * device.
1574 */
1575 return ADD_TO_MLQUEUE;
1576 case GOOD:
4a84067d 1577 scsi_handle_queue_ramp_up(scmd->device);
1da177e4 1578 case COMMAND_TERMINATED:
1da177e4 1579 return SUCCESS;
a9b589d9
VB
1580 case TASK_ABORTED:
1581 goto maybe_retry;
1da177e4
LT
1582 case CHECK_CONDITION:
1583 rtn = scsi_check_sense(scmd);
1584 if (rtn == NEEDS_RETRY)
1585 goto maybe_retry;
1586 /* if rtn == FAILED, we have no sense information;
1587 * returning FAILED will wake the error handler thread
1588 * to collect the sense and redo the decide
1589 * disposition */
1590 return rtn;
1591 case CONDITION_GOOD:
1592 case INTERMEDIATE_GOOD:
1593 case INTERMEDIATE_C_GOOD:
1594 case ACA_ACTIVE:
1595 /*
1596 * who knows? FIXME(eric)
1597 */
1598 return SUCCESS;
1599
1600 case RESERVATION_CONFLICT:
9ccfc756
JB
1601 sdev_printk(KERN_INFO, scmd->device,
1602 "reservation conflict\n");
2082ebc4 1603 set_host_byte(scmd, DID_NEXUS_FAILURE);
1da177e4
LT
1604 return SUCCESS; /* causes immediate i/o error */
1605 default:
1606 return FAILED;
1607 }
1608 return FAILED;
1609
1610 maybe_retry:
1611
1612 /* we requeue for retry because the error was retryable, and
1613 * the request was not marked fast fail. Note that above,
1614 * even if the request is marked fast fail, we still requeue
1615 * for queue congestion conditions (QUEUE_FULL or BUSY) */
8884efab 1616 if ((++scmd->retries) <= scmd->allowed
4a27446f 1617 && !scsi_noretry_cmd(scmd)) {
1da177e4
LT
1618 return NEEDS_RETRY;
1619 } else {
1620 /*
1621 * no more retries - report this one back to upper level.
1622 */
1623 return SUCCESS;
1624 }
1625}
1626
f078727b
FT
1627static void eh_lock_door_done(struct request *req, int uptodate)
1628{
1629 __blk_put_request(req->q, req);
1630}
1631
1da177e4
LT
1632/**
1633 * scsi_eh_lock_door - Prevent medium removal for the specified device
1634 * @sdev: SCSI device to prevent medium removal
1635 *
1636 * Locking:
91bc31fb 1637 * We must be called from process context.
1da177e4
LT
1638 *
1639 * Notes:
1640 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1641 * head of the devices request queue, and continue.
dc8875e1 1642 */
1da177e4
LT
1643static void scsi_eh_lock_door(struct scsi_device *sdev)
1644{
f078727b 1645 struct request *req;
1da177e4 1646
91bc31fb
JB
1647 /*
1648 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1649 * request becomes available
1650 */
f078727b 1651 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1da177e4 1652
f078727b
FT
1653 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1654 req->cmd[1] = 0;
1655 req->cmd[2] = 0;
1656 req->cmd[3] = 0;
1657 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1658 req->cmd[5] = 0;
1da177e4 1659
f078727b
FT
1660 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1661
1662 req->cmd_type = REQ_TYPE_BLOCK_PC;
1663 req->cmd_flags |= REQ_QUIET;
1664 req->timeout = 10 * HZ;
1665 req->retries = 5;
1666
1667 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1668}
1da177e4
LT
1669
1670/**
1671 * scsi_restart_operations - restart io operations to the specified host.
1672 * @shost: Host we are restarting.
1673 *
1674 * Notes:
1675 * When we entered the error handler, we blocked all further i/o to
1676 * this device. we need to 'reverse' this process.
dc8875e1 1677 */
1da177e4
LT
1678static void scsi_restart_operations(struct Scsi_Host *shost)
1679{
1680 struct scsi_device *sdev;
939647ee 1681 unsigned long flags;
1da177e4
LT
1682
1683 /*
1684 * If the door was locked, we need to insert a door lock request
1685 * onto the head of the SCSI request queue for the device. There
1686 * is no point trying to lock the door of an off-line device.
1687 */
1688 shost_for_each_device(sdev, shost) {
1689 if (scsi_device_online(sdev) && sdev->locked)
1690 scsi_eh_lock_door(sdev);
1691 }
1692
1693 /*
1694 * next free up anything directly waiting upon the host. this
1695 * will be requests for character device operations, and also for
1696 * ioctls to queued block devices.
1697 */
1698 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
cadbd4a5 1699 __func__));
1da177e4 1700
939647ee
JB
1701 spin_lock_irqsave(shost->host_lock, flags);
1702 if (scsi_host_set_state(shost, SHOST_RUNNING))
1703 if (scsi_host_set_state(shost, SHOST_CANCEL))
1704 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1705 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
1706
1707 wake_up(&shost->host_wait);
1708
1709 /*
1710 * finally we need to re-initiate requests that may be pending. we will
1711 * have had everything blocked while error handling is taking place, and
1712 * now that error recovery is done, we will need to ensure that these
1713 * requests are started.
1714 */
1715 scsi_run_host_queues(shost);
57fc2e33
DW
1716
1717 /*
1718 * if eh is active and host_eh_scheduled is pending we need to re-run
1719 * recovery. we do this check after scsi_run_host_queues() to allow
1720 * everything pent up since the last eh run a chance to make forward
1721 * progress before we sync again. Either we'll immediately re-run
1722 * recovery or scsi_device_unbusy() will wake us again when these
1723 * pending commands complete.
1724 */
1725 spin_lock_irqsave(shost->host_lock, flags);
1726 if (shost->host_eh_scheduled)
1727 if (scsi_host_set_state(shost, SHOST_RECOVERY))
1728 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
1729 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
1730}
1731
1732/**
1733 * scsi_eh_ready_devs - check device ready state and recover if not.
1734 * @shost: host to be recovered.
eb44820c
RL
1735 * @work_q: &list_head for pending commands.
1736 * @done_q: &list_head for processed commands.
dc8875e1 1737 */
dca84e46
DW
1738void scsi_eh_ready_devs(struct Scsi_Host *shost,
1739 struct list_head *work_q,
1740 struct list_head *done_q)
1da177e4
LT
1741{
1742 if (!scsi_eh_stu(shost, work_q, done_q))
1743 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
30bd7df8
MC
1744 if (!scsi_eh_target_reset(shost, work_q, done_q))
1745 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1746 if (!scsi_eh_host_reset(work_q, done_q))
1747 scsi_eh_offline_sdevs(work_q,
1748 done_q);
1da177e4 1749}
dca84e46 1750EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1da177e4
LT
1751
1752/**
1753 * scsi_eh_flush_done_q - finish processed commands or retry them.
1754 * @done_q: list_head of processed commands.
dc8875e1 1755 */
041c5fc3 1756void scsi_eh_flush_done_q(struct list_head *done_q)
1da177e4 1757{
937abeaa 1758 struct scsi_cmnd *scmd, *next;
1da177e4 1759
937abeaa
CH
1760 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1761 list_del_init(&scmd->eh_entry);
1da177e4 1762 if (scsi_device_online(scmd->device) &&
4a27446f 1763 !scsi_noretry_cmd(scmd) &&
8884efab 1764 (++scmd->retries <= scmd->allowed)) {
1da177e4
LT
1765 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1766 " retry cmd: %p\n",
1767 current->comm,
1768 scmd));
1769 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1770 } else {
793698ce
PM
1771 /*
1772 * If just we got sense for the device (called
1773 * scsi_eh_get_sense), scmd->result is already
1774 * set, do not set DRIVER_TIMEOUT.
1775 */
1da177e4
LT
1776 if (!scmd->result)
1777 scmd->result |= (DRIVER_TIMEOUT << 24);
1778 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1779 " cmd: %p\n",
1780 current->comm, scmd));
1781 scsi_finish_command(scmd);
1782 }
1783 }
1784}
041c5fc3 1785EXPORT_SYMBOL(scsi_eh_flush_done_q);
1da177e4
LT
1786
1787/**
1788 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1789 * @shost: Host to unjam.
1790 *
1791 * Notes:
1792 * When we come in here, we *know* that all commands on the bus have
1793 * either completed, failed or timed out. we also know that no further
1794 * commands are being sent to the host, so things are relatively quiet
1795 * and we have freedom to fiddle with things as we wish.
1796 *
1797 * This is only the *default* implementation. it is possible for
1798 * individual drivers to supply their own version of this function, and
1799 * if the maintainer wishes to do this, it is strongly suggested that
1800 * this function be taken as a template and modified. this function
1801 * was designed to correctly handle problems for about 95% of the
1802 * different cases out there, and it should always provide at least a
1803 * reasonable amount of error recovery.
1804 *
1805 * Any command marked 'failed' or 'timeout' must eventually have
1806 * scsi_finish_cmd() called for it. we do all of the retry stuff
1807 * here, so when we restart the host after we return it should have an
1808 * empty queue.
dc8875e1 1809 */
1da177e4
LT
1810static void scsi_unjam_host(struct Scsi_Host *shost)
1811{
1812 unsigned long flags;
1813 LIST_HEAD(eh_work_q);
1814 LIST_HEAD(eh_done_q);
1815
1816 spin_lock_irqsave(shost->host_lock, flags);
1817 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1818 spin_unlock_irqrestore(shost->host_lock, flags);
1819
1820 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1821
1822 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1823 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1824 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1825
1826 scsi_eh_flush_done_q(&eh_done_q);
1827}
1828
1829/**
ad42eb1b 1830 * scsi_error_handler - SCSI error handler thread
1da177e4
LT
1831 * @data: Host for which we are running.
1832 *
1833 * Notes:
ad42eb1b
CH
1834 * This is the main error handling loop. This is run as a kernel thread
1835 * for every SCSI host and handles all error handling activity.
dc8875e1 1836 */
1da177e4
LT
1837int scsi_error_handler(void *data)
1838{
ad42eb1b 1839 struct Scsi_Host *shost = data;
1da177e4 1840
1da177e4 1841 /*
ad42eb1b
CH
1842 * We use TASK_INTERRUPTIBLE so that the thread is not
1843 * counted against the load average as a running process.
1844 * We never actually get interrupted because kthread_run
c03264a7 1845 * disables signal delivery for the created thread.
1da177e4 1846 */
3ed7a470 1847 while (!kthread_should_stop()) {
b9d5c6b7 1848 set_current_state(TASK_INTERRUPTIBLE);
ee7863bc 1849 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
3ed7a470 1850 shost->host_failed != shost->host_busy) {
ad42eb1b
CH
1851 SCSI_LOG_ERROR_RECOVERY(1,
1852 printk("Error handler scsi_eh_%d sleeping\n",
1853 shost->host_no));
3ed7a470 1854 schedule();
3ed7a470
JB
1855 continue;
1856 }
1da177e4 1857
3ed7a470 1858 __set_current_state(TASK_RUNNING);
ad42eb1b
CH
1859 SCSI_LOG_ERROR_RECOVERY(1,
1860 printk("Error handler scsi_eh_%d waking up\n",
1861 shost->host_no));
1da177e4 1862
1da177e4
LT
1863 /*
1864 * We have a host that is failing for some reason. Figure out
1865 * what we need to do to get it up and online again (if we can).
1866 * If we fail, we end up taking the thing offline.
1867 */
ae0751ff 1868 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
bc4f2401
AS
1869 SCSI_LOG_ERROR_RECOVERY(1,
1870 printk(KERN_ERR "Error handler scsi_eh_%d "
1871 "unable to autoresume\n",
1872 shost->host_no));
1873 continue;
1874 }
1875
9227c33d
CH
1876 if (shost->transportt->eh_strategy_handler)
1877 shost->transportt->eh_strategy_handler(shost);
1da177e4
LT
1878 else
1879 scsi_unjam_host(shost);
1880
1da177e4
LT
1881 /*
1882 * Note - if the above fails completely, the action is to take
1883 * individual devices offline and flush the queue of any
1884 * outstanding requests that may have been pending. When we
1885 * restart, we restart any I/O to any other devices on the bus
1886 * which are still online.
1887 */
1888 scsi_restart_operations(shost);
ae0751ff
LM
1889 if (!shost->eh_noresume)
1890 scsi_autopm_put_host(shost);
1da177e4 1891 }
461a0ffb
SR
1892 __set_current_state(TASK_RUNNING);
1893
ad42eb1b
CH
1894 SCSI_LOG_ERROR_RECOVERY(1,
1895 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
3ed7a470 1896 shost->ehandler = NULL;
1da177e4
LT
1897 return 0;
1898}
1899
1900/*
1901 * Function: scsi_report_bus_reset()
1902 *
1903 * Purpose: Utility function used by low-level drivers to report that
1904 * they have observed a bus reset on the bus being handled.
1905 *
1906 * Arguments: shost - Host in question
1907 * channel - channel on which reset was observed.
1908 *
1909 * Returns: Nothing
1910 *
1911 * Lock status: Host lock must be held.
1912 *
1913 * Notes: This only needs to be called if the reset is one which
1914 * originates from an unknown location. Resets originated
1915 * by the mid-level itself don't need to call this, but there
1916 * should be no harm.
1917 *
1918 * The main purpose of this is to make sure that a CHECK_CONDITION
1919 * is properly treated.
1920 */
1921void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1922{
1923 struct scsi_device *sdev;
1924
1925 __shost_for_each_device(sdev, shost) {
30bd7df8
MC
1926 if (channel == sdev_channel(sdev))
1927 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
1928 }
1929}
1930EXPORT_SYMBOL(scsi_report_bus_reset);
1931
1932/*
1933 * Function: scsi_report_device_reset()
1934 *
1935 * Purpose: Utility function used by low-level drivers to report that
1936 * they have observed a device reset on the device being handled.
1937 *
1938 * Arguments: shost - Host in question
1939 * channel - channel on which reset was observed
1940 * target - target on which reset was observed
1941 *
1942 * Returns: Nothing
1943 *
1944 * Lock status: Host lock must be held
1945 *
1946 * Notes: This only needs to be called if the reset is one which
1947 * originates from an unknown location. Resets originated
1948 * by the mid-level itself don't need to call this, but there
1949 * should be no harm.
1950 *
1951 * The main purpose of this is to make sure that a CHECK_CONDITION
1952 * is properly treated.
1953 */
1954void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1955{
1956 struct scsi_device *sdev;
1957
1958 __shost_for_each_device(sdev, shost) {
422c0d61 1959 if (channel == sdev_channel(sdev) &&
30bd7df8
MC
1960 target == sdev_id(sdev))
1961 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
1962 }
1963}
1964EXPORT_SYMBOL(scsi_report_device_reset);
1965
1966static void
1967scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1968{
1969}
1970
1971/*
1972 * Function: scsi_reset_provider
1973 *
1974 * Purpose: Send requested reset to a bus or device at any phase.
1975 *
1976 * Arguments: device - device to send reset to
1977 * flag - reset type (see scsi.h)
1978 *
1979 * Returns: SUCCESS/FAILURE.
1980 *
1981 * Notes: This is used by the SCSI Generic driver to provide
1982 * Bus/Device reset capability.
1983 */
1984int
1985scsi_reset_provider(struct scsi_device *dev, int flag)
1986{
bc4f2401 1987 struct scsi_cmnd *scmd;
d7a1bb0a 1988 struct Scsi_Host *shost = dev->host;
1da177e4 1989 struct request req;
d7a1bb0a 1990 unsigned long flags;
1da177e4
LT
1991 int rtn;
1992
bc4f2401
AS
1993 if (scsi_autopm_get_host(shost) < 0)
1994 return FAILED;
1995
1996 scmd = scsi_get_command(dev, GFP_KERNEL);
4f54eec8 1997 blk_rq_init(NULL, &req);
1da177e4 1998 scmd->request = &req;
b4edcbca 1999
64a87b24
BH
2000 scmd->cmnd = req.cmd;
2001
1da177e4 2002 scmd->scsi_done = scsi_reset_provider_done_command;
30b0c37b 2003 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1da177e4
LT
2004
2005 scmd->cmd_len = 0;
2006
2007 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1da177e4 2008
d7a1bb0a
JS
2009 spin_lock_irqsave(shost->host_lock, flags);
2010 shost->tmf_in_progress = 1;
2011 spin_unlock_irqrestore(shost->host_lock, flags);
2012
1da177e4
LT
2013 switch (flag) {
2014 case SCSI_TRY_RESET_DEVICE:
2015 rtn = scsi_try_bus_device_reset(scmd);
2016 if (rtn == SUCCESS)
2017 break;
2018 /* FALLTHROUGH */
30bd7df8
MC
2019 case SCSI_TRY_RESET_TARGET:
2020 rtn = scsi_try_target_reset(scmd);
2021 if (rtn == SUCCESS)
2022 break;
2023 /* FALLTHROUGH */
1da177e4
LT
2024 case SCSI_TRY_RESET_BUS:
2025 rtn = scsi_try_bus_reset(scmd);
2026 if (rtn == SUCCESS)
2027 break;
2028 /* FALLTHROUGH */
2029 case SCSI_TRY_RESET_HOST:
2030 rtn = scsi_try_host_reset(scmd);
2031 break;
2032 default:
2033 rtn = FAILED;
2034 }
2035
d7a1bb0a
JS
2036 spin_lock_irqsave(shost->host_lock, flags);
2037 shost->tmf_in_progress = 0;
2038 spin_unlock_irqrestore(shost->host_lock, flags);
2039
2040 /*
2041 * be sure to wake up anyone who was sleeping or had their queue
2042 * suspended while we performed the TMF.
2043 */
2044 SCSI_LOG_ERROR_RECOVERY(3,
2045 printk("%s: waking up host to restart after TMF\n",
cadbd4a5 2046 __func__));
d7a1bb0a
JS
2047
2048 wake_up(&shost->host_wait);
2049
2050 scsi_run_host_queues(shost);
2051
1da177e4 2052 scsi_next_command(scmd);
bc4f2401 2053 scsi_autopm_put_host(shost);
1da177e4
LT
2054 return rtn;
2055}
2056EXPORT_SYMBOL(scsi_reset_provider);
2057
2058/**
2059 * scsi_normalize_sense - normalize main elements from either fixed or
2060 * descriptor sense data format into a common format.
2061 *
2062 * @sense_buffer: byte array containing sense data returned by device
2063 * @sb_len: number of valid bytes in sense_buffer
2064 * @sshdr: pointer to instance of structure that common
2065 * elements are written to.
2066 *
2067 * Notes:
2068 * The "main elements" from sense data are: response_code, sense_key,
2069 * asc, ascq and additional_length (only for descriptor format).
2070 *
2071 * Typically this function can be called after a device has
2072 * responded to a SCSI command with the CHECK_CONDITION status.
2073 *
2074 * Return value:
2075 * 1 if valid sense data information found, else 0;
dc8875e1 2076 */
1da177e4
LT
2077int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2078 struct scsi_sense_hdr *sshdr)
2079{
33aa687d 2080 if (!sense_buffer || !sb_len)
1da177e4
LT
2081 return 0;
2082
2083 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2084
2085 sshdr->response_code = (sense_buffer[0] & 0x7f);
33aa687d
JB
2086
2087 if (!scsi_sense_valid(sshdr))
2088 return 0;
2089
1da177e4
LT
2090 if (sshdr->response_code >= 0x72) {
2091 /*
2092 * descriptor format
2093 */
2094 if (sb_len > 1)
2095 sshdr->sense_key = (sense_buffer[1] & 0xf);
2096 if (sb_len > 2)
2097 sshdr->asc = sense_buffer[2];
2098 if (sb_len > 3)
2099 sshdr->ascq = sense_buffer[3];
2100 if (sb_len > 7)
2101 sshdr->additional_length = sense_buffer[7];
2102 } else {
0bf8c869 2103 /*
1da177e4
LT
2104 * fixed format
2105 */
2106 if (sb_len > 2)
2107 sshdr->sense_key = (sense_buffer[2] & 0xf);
2108 if (sb_len > 7) {
2109 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2110 sb_len : (sense_buffer[7] + 8);
2111 if (sb_len > 12)
2112 sshdr->asc = sense_buffer[12];
2113 if (sb_len > 13)
2114 sshdr->ascq = sense_buffer[13];
2115 }
2116 }
2117
2118 return 1;
2119}
2120EXPORT_SYMBOL(scsi_normalize_sense);
2121
1da177e4
LT
2122int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2123 struct scsi_sense_hdr *sshdr)
2124{
2125 return scsi_normalize_sense(cmd->sense_buffer,
b80ca4f7 2126 SCSI_SENSE_BUFFERSIZE, sshdr);
1da177e4
LT
2127}
2128EXPORT_SYMBOL(scsi_command_normalize_sense);
2129
2130/**
eb44820c 2131 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1da177e4
LT
2132 * @sense_buffer: byte array of descriptor format sense data
2133 * @sb_len: number of valid bytes in sense_buffer
2134 * @desc_type: value of descriptor type to find
2135 * (e.g. 0 -> information)
2136 *
2137 * Notes:
2138 * only valid when sense data is in descriptor format
2139 *
2140 * Return value:
2141 * pointer to start of (first) descriptor if found else NULL
dc8875e1 2142 */
1da177e4
LT
2143const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2144 int desc_type)
2145{
2146 int add_sen_len, add_len, desc_len, k;
2147 const u8 * descp;
2148
2149 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2150 return NULL;
2151 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2152 return NULL;
2153 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2154 add_sen_len : (sb_len - 8);
2155 descp = &sense_buffer[8];
2156 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2157 descp += desc_len;
2158 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2159 desc_len = add_len + 2;
2160 if (descp[0] == desc_type)
2161 return descp;
2162 if (add_len < 0) // short descriptor ??
2163 break;
2164 }
2165 return NULL;
2166}
2167EXPORT_SYMBOL(scsi_sense_desc_find);
2168
2169/**
eb44820c 2170 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1da177e4
LT
2171 * @sense_buffer: byte array of sense data
2172 * @sb_len: number of valid bytes in sense_buffer
2173 * @info_out: pointer to 64 integer where 8 or 4 byte information
2174 * field will be placed if found.
2175 *
2176 * Return value:
2177 * 1 if information field found, 0 if not found.
dc8875e1 2178 */
1da177e4
LT
2179int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2180 u64 * info_out)
2181{
2182 int j;
2183 const u8 * ucp;
2184 u64 ull;
2185
2186 if (sb_len < 7)
2187 return 0;
2188 switch (sense_buffer[0] & 0x7f) {
2189 case 0x70:
2190 case 0x71:
2191 if (sense_buffer[0] & 0x80) {
2192 *info_out = (sense_buffer[3] << 24) +
2193 (sense_buffer[4] << 16) +
2194 (sense_buffer[5] << 8) + sense_buffer[6];
2195 return 1;
2196 } else
2197 return 0;
2198 case 0x72:
2199 case 0x73:
2200 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2201 0 /* info desc */);
2202 if (ucp && (0xa == ucp[1])) {
2203 ull = 0;
2204 for (j = 0; j < 8; ++j) {
2205 if (j > 0)
2206 ull <<= 8;
2207 ull |= ucp[4 + j];
2208 }
2209 *info_out = ull;
2210 return 1;
2211 } else
2212 return 0;
2213 default:
2214 return 0;
2215 }
2216}
2217EXPORT_SYMBOL(scsi_get_sense_info_fld);
3bc6a261
FT
2218
2219/**
2220 * scsi_build_sense_buffer - build sense data in a buffer
2221 * @desc: Sense format (non zero == descriptor format,
2222 * 0 == fixed format)
2223 * @buf: Where to build sense data
2224 * @key: Sense key
2225 * @asc: Additional sense code
2226 * @ascq: Additional sense code qualifier
2227 *
2228 **/
2229void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2230{
2231 if (desc) {
2232 buf[0] = 0x72; /* descriptor, current */
2233 buf[1] = key;
2234 buf[2] = asc;
2235 buf[3] = ascq;
2236 buf[7] = 0;
2237 } else {
2238 buf[0] = 0x70; /* fixed, current */
2239 buf[2] = key;
2240 buf[7] = 0xa;
2241 buf[12] = asc;
2242 buf[13] = ascq;
2243 }
2244}
2245EXPORT_SYMBOL(scsi_build_sense_buffer);
This page took 0.812776 seconds and 5 git commands to generate.