block,scsi: fixup blk_get_request dead queue scenarios
[GitHub/exynos8895/android_kernel_samsung_universal8895.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 55static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
e494f6a7
HR
56static int scsi_try_to_abort_cmd(struct scsi_host_template *,
57 struct scsi_cmnd *);
3eef6257 58
1da177e4
LT
59/* called with shost->host_lock held */
60void scsi_eh_wakeup(struct Scsi_Host *shost)
61{
74665016 62 if (atomic_read(&shost->host_busy) == shost->host_failed) {
bf816235 63 trace_scsi_eh_wakeup(shost);
3ed7a470 64 wake_up_process(shost->ehandler);
91921e01
HR
65 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
66 "Waking error handler thread\n"));
1da177e4
LT
67 }
68}
f8bbfc24
TH
69
70/**
71 * scsi_schedule_eh - schedule EH for SCSI host
72 * @shost: SCSI host to invoke error handling on.
73 *
74 * Schedule SCSI EH without scmd.
dc8875e1 75 */
f8bbfc24
TH
76void scsi_schedule_eh(struct Scsi_Host *shost)
77{
78 unsigned long flags;
79
80 spin_lock_irqsave(shost->host_lock, flags);
81
82 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
83 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
84 shost->host_eh_scheduled++;
85 scsi_eh_wakeup(shost);
86 }
87
88 spin_unlock_irqrestore(shost->host_lock, flags);
89}
90EXPORT_SYMBOL_GPL(scsi_schedule_eh);
1da177e4 91
b4562022
HR
92static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
93{
bb3b621a 94 if (!shost->last_reset || shost->eh_deadline == -1)
b4562022
HR
95 return 0;
96
76ad3e59
HR
97 /*
98 * 32bit accesses are guaranteed to be atomic
99 * (on all supported architectures), so instead
100 * of using a spinlock we can as well double check
bb3b621a 101 * if eh_deadline has been set to 'off' during the
76ad3e59
HR
102 * time_before call.
103 */
104 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
bb3b621a 105 shost->eh_deadline > -1)
b4562022
HR
106 return 0;
107
108 return 1;
109}
110
e494f6a7
HR
111/**
112 * scmd_eh_abort_handler - Handle command aborts
113 * @work: command to be aborted.
114 */
115void
116scmd_eh_abort_handler(struct work_struct *work)
117{
118 struct scsi_cmnd *scmd =
119 container_of(work, struct scsi_cmnd, abort_work.work);
120 struct scsi_device *sdev = scmd->device;
e494f6a7
HR
121 int rtn;
122
e494f6a7 123 if (scsi_host_eh_past_deadline(sdev->host)) {
e494f6a7
HR
124 SCSI_LOG_ERROR_RECOVERY(3,
125 scmd_printk(KERN_INFO, scmd,
126 "scmd %p eh timeout, not aborting\n",
127 scmd));
128 } else {
e494f6a7
HR
129 SCSI_LOG_ERROR_RECOVERY(3,
130 scmd_printk(KERN_INFO, scmd,
131 "aborting command %p\n", scmd));
132 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
133 if (rtn == SUCCESS) {
8922a908 134 set_host_byte(scmd, DID_TIME_OUT);
bb3b621a
RM
135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "scmd %p eh timeout, "
139 "not retrying aborted "
140 "command\n", scmd));
141 } else if (!scsi_noretry_cmd(scmd) &&
e494f6a7
HR
142 (++scmd->retries <= scmd->allowed)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING, scmd,
145 "scmd %p retry "
146 "aborted command\n", scmd));
147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
bb3b621a 148 return;
e494f6a7
HR
149 } else {
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING, scmd,
152 "scmd %p finish "
153 "aborted command\n", scmd));
154 scsi_finish_command(scmd);
bb3b621a 155 return;
e494f6a7 156 }
bb3b621a
RM
157 } else {
158 SCSI_LOG_ERROR_RECOVERY(3,
159 scmd_printk(KERN_INFO, scmd,
160 "scmd %p abort failed, rtn %d\n",
161 scmd, rtn));
e494f6a7 162 }
e494f6a7
HR
163 }
164
165 if (!scsi_eh_scmd_add(scmd, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING, scmd,
168 "scmd %p terminate "
169 "aborted command\n", scmd));
8922a908 170 set_host_byte(scmd, DID_TIME_OUT);
e494f6a7
HR
171 scsi_finish_command(scmd);
172 }
173}
174
175/**
176 * scsi_abort_command - schedule a command abort
177 * @scmd: scmd to abort.
178 *
179 * We only need to abort commands after a command timeout
180 */
181static int
182scsi_abort_command(struct scsi_cmnd *scmd)
183{
184 struct scsi_device *sdev = scmd->device;
185 struct Scsi_Host *shost = sdev->host;
186 unsigned long flags;
187
188 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
189 /*
190 * Retry after abort failed, escalate to next level.
191 */
7daf4804 192 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
e494f6a7
HR
193 SCSI_LOG_ERROR_RECOVERY(3,
194 scmd_printk(KERN_INFO, scmd,
195 "scmd %p previous abort failed\n", scmd));
fcc95a76 196 BUG_ON(delayed_work_pending(&scmd->abort_work));
e494f6a7
HR
197 return FAILED;
198 }
199
200 /*
201 * Do not try a command abort if
202 * SCSI EH has already started.
203 */
204 spin_lock_irqsave(shost->host_lock, flags);
205 if (scsi_host_in_recovery(shost)) {
206 spin_unlock_irqrestore(shost->host_lock, flags);
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd,
209 "scmd %p not aborting, host in recovery\n",
210 scmd));
211 return FAILED;
212 }
213
bb3b621a 214 if (shost->eh_deadline != -1 && !shost->last_reset)
e494f6a7
HR
215 shost->last_reset = jiffies;
216 spin_unlock_irqrestore(shost->host_lock, flags);
217
218 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
219 SCSI_LOG_ERROR_RECOVERY(3,
220 scmd_printk(KERN_INFO, scmd,
221 "scmd %p abort scheduled\n", scmd));
222 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
223 return SUCCESS;
224}
225
1da177e4
LT
226/**
227 * scsi_eh_scmd_add - add scsi cmd to error handling.
228 * @scmd: scmd to run eh on.
229 * @eh_flag: optional SCSI_EH flag.
230 *
231 * Return value:
232 * 0 on failure.
dc8875e1 233 */
1da177e4
LT
234int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
235{
236 struct Scsi_Host *shost = scmd->device->host;
237 unsigned long flags;
939647ee 238 int ret = 0;
1da177e4 239
3ed7a470 240 if (!shost->ehandler)
1da177e4
LT
241 return 0;
242
243 spin_lock_irqsave(shost->host_lock, flags);
939647ee
JB
244 if (scsi_host_set_state(shost, SHOST_RECOVERY))
245 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
246 goto out_unlock;
1da177e4 247
bb3b621a 248 if (shost->eh_deadline != -1 && !shost->last_reset)
b4562022
HR
249 shost->last_reset = jiffies;
250
939647ee 251 ret = 1;
e494f6a7
HR
252 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
253 eh_flag &= ~SCSI_EH_CANCEL_CMD;
3111b0d1 254 scmd->eh_eflags |= eh_flag;
1da177e4 255 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
1da177e4
LT
256 shost->host_failed++;
257 scsi_eh_wakeup(shost);
939647ee 258 out_unlock:
1da177e4 259 spin_unlock_irqrestore(shost->host_lock, flags);
939647ee 260 return ret;
1da177e4
LT
261}
262
1da177e4
LT
263/**
264 * scsi_times_out - Timeout function for normal scsi commands.
242f9dcb 265 * @req: request that is timing out.
1da177e4
LT
266 *
267 * Notes:
268 * We do not need to lock this. There is the potential for a race
269 * only in that the normal completion handling might run, but if the
270 * normal completion function determines that the timer has already
271 * fired, then it mustn't do anything.
dc8875e1 272 */
242f9dcb 273enum blk_eh_timer_return scsi_times_out(struct request *req)
1da177e4 274{
242f9dcb 275 struct scsi_cmnd *scmd = req->special;
242f9dcb 276 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
0bf8c869 277 struct Scsi_Host *host = scmd->device->host;
6c5f8ce1 278
bf816235 279 trace_scsi_dispatch_cmd_timeout(scmd);
1da177e4
LT
280 scsi_log_completion(scmd, TIMEOUT_ERROR);
281
bb3b621a 282 if (host->eh_deadline != -1 && !host->last_reset)
b4562022
HR
283 host->last_reset = jiffies;
284
0bf8c869
JJ
285 if (host->transportt->eh_timed_out)
286 rtn = host->transportt->eh_timed_out(scmd);
287 else if (host->hostt->eh_timed_out)
288 rtn = host->hostt->eh_timed_out(scmd);
1da177e4 289
a33c070b
HR
290 if (rtn == BLK_EH_NOT_HANDLED) {
291 if (!host->hostt->no_async_abort &&
292 scsi_abort_command(scmd) == SUCCESS)
e494f6a7
HR
293 return BLK_EH_NOT_HANDLED;
294
a33c070b
HR
295 set_host_byte(scmd, DID_TIME_OUT);
296 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
297 rtn = BLK_EH_HANDLED;
298 }
242f9dcb 299
fa990781 300 return rtn;
1da177e4
LT
301}
302
303/**
304 * scsi_block_when_processing_errors - Prevent cmds from being queued.
305 * @sdev: Device on which we are performing recovery.
306 *
307 * Description:
308 * We block until the host is out of error recovery, and then check to
309 * see whether the host or the device is offline.
310 *
311 * Return value:
312 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
dc8875e1 313 */
1da177e4
LT
314int scsi_block_when_processing_errors(struct scsi_device *sdev)
315{
316 int online;
317
939647ee 318 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
1da177e4
LT
319
320 online = scsi_device_online(sdev);
321
91921e01
HR
322 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
323 "%s: rtn: %d\n", __func__, online));
1da177e4
LT
324
325 return online;
326}
327EXPORT_SYMBOL(scsi_block_when_processing_errors);
328
329#ifdef CONFIG_SCSI_LOGGING
330/**
331 * scsi_eh_prt_fail_stats - Log info on failures.
332 * @shost: scsi host being recovered.
333 * @work_q: Queue of scsi cmds to process.
dc8875e1 334 */
1da177e4
LT
335static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
336 struct list_head *work_q)
337{
338 struct scsi_cmnd *scmd;
339 struct scsi_device *sdev;
340 int total_failures = 0;
341 int cmd_failed = 0;
342 int cmd_cancel = 0;
343 int devices_failed = 0;
344
345 shost_for_each_device(sdev, shost) {
346 list_for_each_entry(scmd, work_q, eh_entry) {
347 if (scmd->device == sdev) {
348 ++total_failures;
3111b0d1 349 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
1da177e4 350 ++cmd_cancel;
0bf8c869 351 else
1da177e4
LT
352 ++cmd_failed;
353 }
354 }
355
356 if (cmd_cancel || cmd_failed) {
357 SCSI_LOG_ERROR_RECOVERY(3,
9ccfc756
JB
358 sdev_printk(KERN_INFO, sdev,
359 "%s: cmds failed: %d, cancel: %d\n",
cadbd4a5 360 __func__, cmd_failed,
9ccfc756 361 cmd_cancel));
1da177e4
LT
362 cmd_cancel = 0;
363 cmd_failed = 0;
364 ++devices_failed;
365 }
366 }
367
91921e01
HR
368 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
369 "Total of %d commands on %d"
370 " devices require eh work\n",
0bf8c869 371 total_failures, devices_failed));
1da177e4
LT
372}
373#endif
374
279afdfe
EM
375 /**
376 * scsi_report_lun_change - Set flag on all *other* devices on the same target
377 * to indicate that a UNIT ATTENTION is expected.
378 * @sdev: Device reporting the UNIT ATTENTION
379 */
380static void scsi_report_lun_change(struct scsi_device *sdev)
381{
382 sdev->sdev_target->expecting_lun_change = 1;
383}
384
385/**
386 * scsi_report_sense - Examine scsi sense information and log messages for
387 * certain conditions, also issue uevents for some of them.
388 * @sdev: Device reporting the sense code
389 * @sshdr: sshdr to be examined
390 */
391static void scsi_report_sense(struct scsi_device *sdev,
392 struct scsi_sense_hdr *sshdr)
393{
394 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
395
396 if (sshdr->sense_key == UNIT_ATTENTION) {
397 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
398 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
399 sdev_printk(KERN_WARNING, sdev,
400 "Inquiry data has changed");
401 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
402 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
403 scsi_report_lun_change(sdev);
404 sdev_printk(KERN_WARNING, sdev,
405 "Warning! Received an indication that the "
406 "LUN assignments on this target have "
407 "changed. The Linux SCSI layer does not "
408 "automatically remap LUN assignments.\n");
409 } else if (sshdr->asc == 0x3f)
410 sdev_printk(KERN_WARNING, sdev,
411 "Warning! Received an indication that the "
412 "operating parameters on this target have "
413 "changed. The Linux SCSI layer does not "
414 "automatically adjust these parameters.\n");
415
416 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
417 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
418 sdev_printk(KERN_WARNING, sdev,
419 "Warning! Received an indication that the "
420 "LUN reached a thin provisioning soft "
421 "threshold.\n");
422 }
423
424 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
425 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
426 sdev_printk(KERN_WARNING, sdev,
427 "Mode parameters changed");
428 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
429 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
430 sdev_printk(KERN_WARNING, sdev,
431 "Capacity data has changed");
432 } else if (sshdr->asc == 0x2a)
433 sdev_printk(KERN_WARNING, sdev,
434 "Parameters changed");
435 }
436
437 if (evt_type != SDEV_EVT_MAXBITS) {
438 set_bit(evt_type, sdev->pending_events);
439 schedule_work(&sdev->event_work);
440 }
441}
442
1da177e4
LT
443/**
444 * scsi_check_sense - Examine scsi cmd sense
445 * @scmd: Cmd to have sense checked.
446 *
447 * Return value:
87f14e65 448 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1da177e4
LT
449 *
450 * Notes:
451 * When a deferred error is detected the current command has
452 * not been executed and needs retrying.
dc8875e1 453 */
1da177e4
LT
454static int scsi_check_sense(struct scsi_cmnd *scmd)
455{
a6a8d9f8 456 struct scsi_device *sdev = scmd->device;
1da177e4
LT
457 struct scsi_sense_hdr sshdr;
458
459 if (! scsi_command_normalize_sense(scmd, &sshdr))
460 return FAILED; /* no valid sense data */
461
14216561
JB
462 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
463 /*
464 * nasty: for mid-layer issued TURs, we need to return the
465 * actual sense data without any recovery attempt. For eh
466 * issued ones, we need to try to recover and interpret
467 */
468 return SUCCESS;
469
279afdfe
EM
470 scsi_report_sense(sdev, &sshdr);
471
1da177e4
LT
472 if (scsi_sense_is_deferred(&sshdr))
473 return NEEDS_RETRY;
474
a6a8d9f8
CS
475 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
476 sdev->scsi_dh_data->scsi_dh->check_sense) {
477 int rc;
478
479 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
480 if (rc != SCSI_RETURN_NOT_HANDLED)
481 return rc;
482 /* handler does not care. Drop down to default handling */
483 }
484
1da177e4
LT
485 /*
486 * Previous logic looked for FILEMARK, EOM or ILI which are
487 * mainly associated with tapes and returned SUCCESS.
488 */
489 if (sshdr.response_code == 0x70) {
490 /* fixed format */
491 if (scmd->sense_buffer[2] & 0xe0)
492 return SUCCESS;
493 } else {
494 /*
495 * descriptor format: look for "stream commands sense data
496 * descriptor" (see SSC-3). Assume single sense data
497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
498 */
499 if ((sshdr.additional_length > 3) &&
500 (scmd->sense_buffer[8] == 0x4) &&
501 (scmd->sense_buffer[11] & 0xe0))
502 return SUCCESS;
503 }
504
505 switch (sshdr.sense_key) {
506 case NO_SENSE:
507 return SUCCESS;
508 case RECOVERED_ERROR:
509 return /* soft_error */ SUCCESS;
510
511 case ABORTED_COMMAND:
511e44f4
MP
512 if (sshdr.asc == 0x10) /* DIF */
513 return SUCCESS;
514
1da177e4
LT
515 return NEEDS_RETRY;
516 case NOT_READY:
517 case UNIT_ATTENTION:
518 /*
519 * if we are expecting a cc/ua because of a bus reset that we
520 * performed, treat this just as a retry. otherwise this is
521 * information that we should pass up to the upper-level driver
522 * so that we can deal with it there.
523 */
524 if (scmd->device->expecting_cc_ua) {
dfcf7775
TH
525 /*
526 * Because some device does not queue unit
527 * attentions correctly, we carefully check
528 * additional sense code and qualifier so as
529 * not to squash media change unit attention.
530 */
531 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
532 scmd->device->expecting_cc_ua = 0;
533 return NEEDS_RETRY;
534 }
1da177e4 535 }
279afdfe
EM
536 /*
537 * we might also expect a cc/ua if another LUN on the target
538 * reported a UA with an ASC/ASCQ of 3F 0E -
539 * REPORTED LUNS DATA HAS CHANGED.
540 */
541 if (scmd->device->sdev_target->expecting_lun_change &&
542 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
543 return NEEDS_RETRY;
1da177e4 544 /*
0bf8c869 545 * if the device is in the process of becoming ready, we
1da177e4
LT
546 * should retry.
547 */
548 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
549 return NEEDS_RETRY;
550 /*
551 * if the device is not started, we need to wake
552 * the error handler to start the motor
553 */
554 if (scmd->device->allow_restart &&
555 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
556 return FAILED;
02e031cb
CH
557 /*
558 * Pass the UA upwards for a determination in the completion
559 * functions.
560 */
561 return SUCCESS;
1da177e4 562
63583cca 563 /* these are not supported */
a9d6ceb8
HR
564 case DATA_PROTECT:
565 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
566 /* Thin provisioning hard threshold reached */
567 set_host_byte(scmd, DID_ALLOC_FAILURE);
568 return SUCCESS;
569 }
1da177e4
LT
570 case COPY_ABORTED:
571 case VOLUME_OVERFLOW:
572 case MISCOMPARE:
63583cca 573 case BLANK_CHECK:
87f14e65
HR
574 set_host_byte(scmd, DID_TARGET_FAILURE);
575 return SUCCESS;
1da177e4
LT
576
577 case MEDIUM_ERROR:
fd1b494d
LT
578 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
579 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
580 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
7e782af5 581 set_host_byte(scmd, DID_MEDIUM_ERROR);
87f14e65 582 return SUCCESS;
fd1b494d 583 }
1da177e4
LT
584 return NEEDS_RETRY;
585
586 case HARDWARE_ERROR:
587 if (scmd->device->retry_hwerror)
bb0003c1 588 return ADD_TO_MLQUEUE;
1da177e4 589 else
87f14e65 590 set_host_byte(scmd, DID_TARGET_FAILURE);
1da177e4
LT
591
592 case ILLEGAL_REQUEST:
47ac56db
MS
593 if (sshdr.asc == 0x20 || /* Invalid command operation code */
594 sshdr.asc == 0x21 || /* Logical block address out of range */
595 sshdr.asc == 0x24 || /* Invalid field in cdb */
596 sshdr.asc == 0x26) { /* Parameter value invalid */
87f14e65 597 set_host_byte(scmd, DID_TARGET_FAILURE);
47ac56db
MS
598 }
599 return SUCCESS;
600
1da177e4
LT
601 default:
602 return SUCCESS;
603 }
604}
605
4a84067d
VD
606static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
607{
608 struct scsi_host_template *sht = sdev->host->hostt;
609 struct scsi_device *tmp_sdev;
610
611 if (!sht->change_queue_depth ||
612 sdev->queue_depth >= sdev->max_queue_depth)
613 return;
614
615 if (time_before(jiffies,
616 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
617 return;
618
619 if (time_before(jiffies,
620 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
621 return;
622
623 /*
624 * Walk all devices of a target and do
625 * ramp up on them.
626 */
627 shost_for_each_device(tmp_sdev, sdev->host) {
628 if (tmp_sdev->channel != sdev->channel ||
629 tmp_sdev->id != sdev->id ||
630 tmp_sdev->queue_depth == sdev->max_queue_depth)
631 continue;
632 /*
633 * call back into LLD to increase queue_depth by one
634 * with ramp up reason code.
635 */
636 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
637 SCSI_QDEPTH_RAMP_UP);
638 sdev->last_queue_ramp_up = jiffies;
639 }
640}
641
42a6a918
MC
642static void scsi_handle_queue_full(struct scsi_device *sdev)
643{
644 struct scsi_host_template *sht = sdev->host->hostt;
645 struct scsi_device *tmp_sdev;
646
647 if (!sht->change_queue_depth)
648 return;
649
650 shost_for_each_device(tmp_sdev, sdev->host) {
651 if (tmp_sdev->channel != sdev->channel ||
652 tmp_sdev->id != sdev->id)
653 continue;
654 /*
655 * We do not know the number of commands that were at
656 * the device when we got the queue full so we start
657 * from the highest possible value and work our way down.
658 */
659 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
660 SCSI_QDEPTH_QFULL);
661 }
662}
663
1da177e4
LT
664/**
665 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
666 * @scmd: SCSI cmd to examine.
667 *
668 * Notes:
669 * This is *only* called when we are examining the status of commands
670 * queued during error recovery. the main difference here is that we
671 * don't allow for the possibility of retries here, and we are a lot
672 * more restrictive about what we consider acceptable.
dc8875e1 673 */
1da177e4
LT
674static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
675{
676 /*
677 * first check the host byte, to see if there is anything in there
678 * that would indicate what we need to do.
679 */
680 if (host_byte(scmd->result) == DID_RESET) {
681 /*
682 * rats. we are already in the error handler, so we now
683 * get to try and figure out what to do next. if the sense
684 * is valid, we have a pretty good idea of what to do.
685 * if not, we mark it as FAILED.
686 */
687 return scsi_check_sense(scmd);
688 }
689 if (host_byte(scmd->result) != DID_OK)
690 return FAILED;
691
692 /*
693 * next, check the message byte.
694 */
695 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
696 return FAILED;
697
698 /*
699 * now, check the status byte to see if this indicates
700 * anything special.
701 */
702 switch (status_byte(scmd->result)) {
703 case GOOD:
4a84067d 704 scsi_handle_queue_ramp_up(scmd->device);
1da177e4
LT
705 case COMMAND_TERMINATED:
706 return SUCCESS;
707 case CHECK_CONDITION:
708 return scsi_check_sense(scmd);
709 case CONDITION_GOOD:
710 case INTERMEDIATE_GOOD:
711 case INTERMEDIATE_C_GOOD:
712 /*
713 * who knows? FIXME(eric)
714 */
715 return SUCCESS;
5f91bb05 716 case RESERVATION_CONFLICT:
67110dfd
JB
717 if (scmd->cmnd[0] == TEST_UNIT_READY)
718 /* it is a success, we probed the device and
719 * found it */
720 return SUCCESS;
721 /* otherwise, we failed to send the command */
722 return FAILED;
1da177e4 723 case QUEUE_FULL:
42a6a918
MC
724 scsi_handle_queue_full(scmd->device);
725 /* fall through */
726 case BUSY:
3eb3a928 727 return NEEDS_RETRY;
1da177e4
LT
728 default:
729 return FAILED;
730 }
731 return FAILED;
732}
733
1da177e4
LT
734/**
735 * scsi_eh_done - Completion function for error handling.
736 * @scmd: Cmd that is done.
dc8875e1 737 */
1da177e4
LT
738static void scsi_eh_done(struct scsi_cmnd *scmd)
739{
0bf8c869 740 struct completion *eh_action;
85631672 741
91921e01
HR
742 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
743 "%s scmd: %p result: %x\n",
cadbd4a5 744 __func__, scmd, scmd->result));
85631672
MR
745
746 eh_action = scmd->device->host->eh_action;
747 if (eh_action)
748 complete(eh_action);
1da177e4
LT
749}
750
292148f8
BK
751/**
752 * scsi_try_host_reset - ask host adapter to reset itself
c2b3ebd0 753 * @scmd: SCSI cmd to send host reset.
dc8875e1 754 */
292148f8
BK
755static int scsi_try_host_reset(struct scsi_cmnd *scmd)
756{
757 unsigned long flags;
758 int rtn;
0bf8c869
JJ
759 struct Scsi_Host *host = scmd->device->host;
760 struct scsi_host_template *hostt = host->hostt;
292148f8 761
91921e01
HR
762 SCSI_LOG_ERROR_RECOVERY(3,
763 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
292148f8 764
0bf8c869 765 if (!hostt->eh_host_reset_handler)
292148f8
BK
766 return FAILED;
767
0bf8c869 768 rtn = hostt->eh_host_reset_handler(scmd);
292148f8
BK
769
770 if (rtn == SUCCESS) {
0bf8c869 771 if (!hostt->skip_settle_delay)
292148f8 772 ssleep(HOST_RESET_SETTLE_TIME);
0bf8c869
JJ
773 spin_lock_irqsave(host->host_lock, flags);
774 scsi_report_bus_reset(host, scmd_channel(scmd));
775 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
776 }
777
778 return rtn;
779}
780
781/**
782 * scsi_try_bus_reset - ask host to perform a bus reset
783 * @scmd: SCSI cmd to send bus reset.
dc8875e1 784 */
292148f8
BK
785static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
786{
787 unsigned long flags;
788 int rtn;
0bf8c869
JJ
789 struct Scsi_Host *host = scmd->device->host;
790 struct scsi_host_template *hostt = host->hostt;
292148f8 791
91921e01
HR
792 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
793 "%s: Snd Bus RST\n", __func__));
292148f8 794
0bf8c869 795 if (!hostt->eh_bus_reset_handler)
292148f8
BK
796 return FAILED;
797
0bf8c869 798 rtn = hostt->eh_bus_reset_handler(scmd);
292148f8
BK
799
800 if (rtn == SUCCESS) {
0bf8c869 801 if (!hostt->skip_settle_delay)
292148f8 802 ssleep(BUS_RESET_SETTLE_TIME);
0bf8c869
JJ
803 spin_lock_irqsave(host->host_lock, flags);
804 scsi_report_bus_reset(host, scmd_channel(scmd));
805 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
806 }
807
808 return rtn;
809}
810
30bd7df8
MC
811static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
812{
813 sdev->was_reset = 1;
814 sdev->expecting_cc_ua = 1;
815}
816
817/**
818 * scsi_try_target_reset - Ask host to perform a target reset
819 * @scmd: SCSI cmd used to send a target reset
820 *
821 * Notes:
822 * There is no timeout for this operation. if this operation is
823 * unreliable for a given host, then the host itself needs to put a
824 * timer on it, and set the host back to a consistent state prior to
825 * returning.
826 */
827static int scsi_try_target_reset(struct scsi_cmnd *scmd)
828{
829 unsigned long flags;
830 int rtn;
0bf8c869
JJ
831 struct Scsi_Host *host = scmd->device->host;
832 struct scsi_host_template *hostt = host->hostt;
30bd7df8 833
0bf8c869 834 if (!hostt->eh_target_reset_handler)
30bd7df8
MC
835 return FAILED;
836
0bf8c869 837 rtn = hostt->eh_target_reset_handler(scmd);
30bd7df8 838 if (rtn == SUCCESS) {
0bf8c869 839 spin_lock_irqsave(host->host_lock, flags);
30bd7df8
MC
840 __starget_for_each_device(scsi_target(scmd->device), NULL,
841 __scsi_report_device_reset);
0bf8c869 842 spin_unlock_irqrestore(host->host_lock, flags);
30bd7df8
MC
843 }
844
845 return rtn;
846}
847
292148f8
BK
848/**
849 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
850 * @scmd: SCSI cmd used to send BDR
851 *
852 * Notes:
853 * There is no timeout for this operation. if this operation is
854 * unreliable for a given host, then the host itself needs to put a
855 * timer on it, and set the host back to a consistent state prior to
856 * returning.
dc8875e1 857 */
292148f8
BK
858static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
859{
860 int rtn;
0bf8c869 861 struct scsi_host_template *hostt = scmd->device->host->hostt;
292148f8 862
0bf8c869 863 if (!hostt->eh_device_reset_handler)
292148f8
BK
864 return FAILED;
865
0bf8c869 866 rtn = hostt->eh_device_reset_handler(scmd);
30bd7df8
MC
867 if (rtn == SUCCESS)
868 __scsi_report_device_reset(scmd->device, NULL);
292148f8
BK
869 return rtn;
870}
871
0bf8c869 872static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
292148f8 873{
0bf8c869 874 if (!hostt->eh_abort_handler)
292148f8
BK
875 return FAILED;
876
0bf8c869 877 return hostt->eh_abort_handler(scmd);
292148f8
BK
878}
879
292148f8
BK
880static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
881{
0bf8c869 882 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
292148f8 883 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
30bd7df8
MC
884 if (scsi_try_target_reset(scmd) != SUCCESS)
885 if (scsi_try_bus_reset(scmd) != SUCCESS)
886 scsi_try_host_reset(scmd);
292148f8
BK
887}
888
1da177e4 889/**
3b729f76 890 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
2dc611de 891 * @scmd: SCSI command structure to hijack
e1c23468 892 * @ses: structure to save restore information
55db6c1b 893 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
64a87b24 894 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
55db6c1b 895 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
2dc611de 896 *
e1c23468 897 * This function is used to save a scsi command information before re-execution
55db6c1b
BH
898 * as part of the error recovery process. If @sense_bytes is 0 the command
899 * sent must be one that does not transfer any data. If @sense_bytes != 0
900 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
901 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
dc8875e1 902 */
e1c23468
BH
903void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
904 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1da177e4 905{
f59114b7 906 struct scsi_device *sdev = scmd->device;
1da177e4 907
631c228c
CH
908 /*
909 * We need saved copies of a number of fields - this is because
910 * error handling may need to overwrite these with different values
911 * to run different commands, and once error handling is complete,
912 * we will need to restore these values prior to running the actual
913 * command.
914 */
e1c23468 915 ses->cmd_len = scmd->cmd_len;
64a87b24 916 ses->cmnd = scmd->cmnd;
e1c23468 917 ses->data_direction = scmd->sc_data_direction;
30b0c37b 918 ses->sdb = scmd->sdb;
6f9a35e2 919 ses->next_rq = scmd->request->next_rq;
e1c23468 920 ses->result = scmd->result;
12265709 921 ses->underflow = scmd->underflow;
db007fc5 922 ses->prot_op = scmd->prot_op;
631c228c 923
db007fc5 924 scmd->prot_op = SCSI_PROT_NORMAL;
c69e6f81 925 scmd->eh_eflags = 0;
64a87b24
BH
926 scmd->cmnd = ses->eh_cmnd;
927 memset(scmd->cmnd, 0, BLK_MAX_CDB);
30b0c37b 928 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
6f9a35e2 929 scmd->request->next_rq = NULL;
644373a4 930 scmd->result = 0;
30b0c37b 931
55db6c1b 932 if (sense_bytes) {
30b0c37b
BH
933 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
934 sense_bytes);
e1c23468 935 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
30b0c37b
BH
936 scmd->sdb.length);
937 scmd->sdb.table.sgl = &ses->sense_sgl;
55db6c1b 938 scmd->sc_data_direction = DMA_FROM_DEVICE;
30b0c37b 939 scmd->sdb.table.nents = 1;
55db6c1b 940 scmd->cmnd[0] = REQUEST_SENSE;
30b0c37b 941 scmd->cmnd[4] = scmd->sdb.length;
55db6c1b 942 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
631c228c 943 } else {
631c228c 944 scmd->sc_data_direction = DMA_NONE;
55db6c1b 945 if (cmnd) {
64a87b24 946 BUG_ON(cmnd_size > BLK_MAX_CDB);
55db6c1b
BH
947 memcpy(scmd->cmnd, cmnd, cmnd_size);
948 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
949 }
631c228c
CH
950 }
951
952 scmd->underflow = 0;
631c228c 953
55db6c1b 954 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1da177e4 955 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b7 956 (sdev->lun << 5 & 0xe0);
1da177e4 957
631c228c
CH
958 /*
959 * Zero the sense buffer. The scsi spec mandates that any
960 * untransferred sense data should be interpreted as being zero.
961 */
b80ca4f7 962 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
e1c23468
BH
963}
964EXPORT_SYMBOL(scsi_eh_prep_cmnd);
965
966/**
3b729f76 967 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
e1c23468 968 * @scmd: SCSI command structure to restore
477e608c 969 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
e1c23468 970 *
477e608c 971 * Undo any damage done by above scsi_eh_prep_cmnd().
dc8875e1 972 */
e1c23468
BH
973void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
974{
975 /*
976 * Restore original data
977 */
978 scmd->cmd_len = ses->cmd_len;
64a87b24 979 scmd->cmnd = ses->cmnd;
e1c23468 980 scmd->sc_data_direction = ses->data_direction;
30b0c37b 981 scmd->sdb = ses->sdb;
6f9a35e2 982 scmd->request->next_rq = ses->next_rq;
e1c23468 983 scmd->result = ses->result;
12265709 984 scmd->underflow = ses->underflow;
db007fc5 985 scmd->prot_op = ses->prot_op;
e1c23468
BH
986}
987EXPORT_SYMBOL(scsi_eh_restore_cmnd);
631c228c 988
e1c23468 989/**
3b729f76 990 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
e1c23468
BH
991 * @scmd: SCSI command structure to hijack
992 * @cmnd: CDB to send
993 * @cmnd_size: size in bytes of @cmnd
994 * @timeout: timeout for this request
995 * @sense_bytes: size of sense data to copy or 0
996 *
997 * This function is used to send a scsi command down to a target device
998 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
999 *
1000 * Return value:
1001 * SUCCESS or FAILED or NEEDS_RETRY
dc8875e1 1002 */
e1c23468
BH
1003static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1004 int cmnd_size, int timeout, unsigned sense_bytes)
1005{
1006 struct scsi_device *sdev = scmd->device;
1007 struct Scsi_Host *shost = sdev->host;
1008 DECLARE_COMPLETION_ONSTACK(done);
fc73648a 1009 unsigned long timeleft = timeout;
e1c23468 1010 struct scsi_eh_save ses;
fc73648a 1011 const unsigned long stall_for = msecs_to_jiffies(100);
e1c23468
BH
1012 int rtn;
1013
fc73648a 1014retry:
e1c23468 1015 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
7dfdc9a5 1016 shost->eh_action = &done;
1da177e4 1017
1da177e4 1018 scsi_log_send(scmd);
f281233d 1019 scmd->scsi_done = scsi_eh_done;
fc73648a
HR
1020 rtn = shost->hostt->queuecommand(shost, scmd);
1021 if (rtn) {
1022 if (timeleft > stall_for) {
1023 scsi_eh_restore_cmnd(scmd, &ses);
1024 timeleft -= stall_for;
1025 msleep(jiffies_to_msecs(stall_for));
1026 goto retry;
1027 }
1028 /* signal not to enter either branch of the if () below */
1029 timeleft = 0;
1030 rtn = NEEDS_RETRY;
1031 } else {
1032 timeleft = wait_for_completion_timeout(&done, timeout);
ac61d195 1033 rtn = SUCCESS;
fc73648a 1034 }
1da177e4 1035
f59114b7 1036 shost->eh_action = NULL;
1da177e4 1037
fc73648a 1038 scsi_log_completion(scmd, rtn);
1da177e4 1039
91921e01
HR
1040 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1041 "%s: scmd: %p, timeleft: %ld\n",
cadbd4a5 1042 __func__, scmd, timeleft));
1da177e4
LT
1043
1044 /*
fc73648a
HR
1045 * If there is time left scsi_eh_done got called, and we will examine
1046 * the actual status codes to see whether the command actually did
1047 * complete normally, else if we have a zero return and no time left,
1048 * the command must still be pending, so abort it and return FAILED.
1049 * If we never actually managed to issue the command, because
1050 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1051 * value above (so don't execute either branch of the if)
1da177e4 1052 */
7dfdc9a5 1053 if (timeleft) {
1da177e4 1054 rtn = scsi_eh_completed_normally(scmd);
91921e01
HR
1055 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1056 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
7dfdc9a5 1057
1da177e4
LT
1058 switch (rtn) {
1059 case SUCCESS:
1060 case NEEDS_RETRY:
1061 case FAILED:
1062 break;
6e883b0e
HR
1063 case ADD_TO_MLQUEUE:
1064 rtn = NEEDS_RETRY;
1065 break;
1da177e4
LT
1066 default:
1067 rtn = FAILED;
1068 break;
1069 }
fc73648a 1070 } else if (!rtn) {
292148f8 1071 scsi_abort_eh_cmnd(scmd);
7dfdc9a5 1072 rtn = FAILED;
1da177e4
LT
1073 }
1074
e1c23468 1075 scsi_eh_restore_cmnd(scmd, &ses);
18a4d0a2 1076
1da177e4
LT
1077 return rtn;
1078}
1079
1080/**
1081 * scsi_request_sense - Request sense data from a particular target.
1082 * @scmd: SCSI cmd for request sense.
1083 *
1084 * Notes:
1085 * Some hosts automatically obtain this information, others require
1086 * that we obtain it on our own. This function will *not* return until
1087 * the command either times out, or it completes.
dc8875e1 1088 */
1da177e4
LT
1089static int scsi_request_sense(struct scsi_cmnd *scmd)
1090{
0816c925 1091 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1da177e4
LT
1092}
1093
2451079b
JB
1094static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1095{
1096 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1097 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1098 if (sdrv->eh_action)
1099 rtn = sdrv->eh_action(scmd, rtn);
1100 }
1101 return rtn;
1102}
1103
1da177e4
LT
1104/**
1105 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1106 * @scmd: Original SCSI cmd that eh has finished.
1107 * @done_q: Queue for processed commands.
1108 *
1109 * Notes:
1110 * We don't want to use the normal command completion while we are are
1111 * still handling errors - it may cause other commands to be queued,
eb44820c 1112 * and that would disturb what we are doing. Thus we really want to
1da177e4
LT
1113 * keep a list of pending commands for final completion, and once we
1114 * are ready to leave error handling we handle completion for real.
dc8875e1 1115 */
041c5fc3 1116void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1da177e4
LT
1117{
1118 scmd->device->host->host_failed--;
3111b0d1 1119 scmd->eh_eflags = 0;
1da177e4
LT
1120 list_move_tail(&scmd->eh_entry, done_q);
1121}
041c5fc3 1122EXPORT_SYMBOL(scsi_eh_finish_cmd);
1da177e4
LT
1123
1124/**
1125 * scsi_eh_get_sense - Get device sense data.
1126 * @work_q: Queue of commands to process.
eb44820c 1127 * @done_q: Queue of processed commands.
1da177e4
LT
1128 *
1129 * Description:
1130 * See if we need to request sense information. if so, then get it
0bf8c869 1131 * now, so we have a better idea of what to do.
1da177e4
LT
1132 *
1133 * Notes:
1134 * This has the unfortunate side effect that if a shost adapter does
eb44820c 1135 * not automatically request sense information, we end up shutting
1da177e4
LT
1136 * it down before we request it.
1137 *
1138 * All drivers should request sense information internally these days,
1139 * so for now all I have to say is tough noogies if you end up in here.
1140 *
1141 * XXX: Long term this code should go away, but that needs an audit of
1142 * all LLDDs first.
dc8875e1 1143 */
dca84e46
DW
1144int scsi_eh_get_sense(struct list_head *work_q,
1145 struct list_head *done_q)
1da177e4 1146{
937abeaa 1147 struct scsi_cmnd *scmd, *next;
b4562022 1148 struct Scsi_Host *shost;
1da177e4
LT
1149 int rtn;
1150
937abeaa 1151 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3111b0d1 1152 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1da177e4
LT
1153 SCSI_SENSE_VALID(scmd))
1154 continue;
1155
b4562022 1156 shost = scmd->device->host;
b4562022 1157 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1158 SCSI_LOG_ERROR_RECOVERY(3,
1159 shost_printk(KERN_INFO, shost,
1160 "skip %s, past eh deadline\n",
1161 __func__));
1162 break;
1163 }
d555a2ab
JB
1164 if (status_byte(scmd->result) != CHECK_CONDITION)
1165 /*
1166 * don't request sense if there's no check condition
1167 * status because the error we're processing isn't one
1168 * that has a sense code (and some devices get
1169 * confused by sense requests out of the blue)
1170 */
1171 continue;
1172
3bf743e7
JG
1173 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1174 "%s: requesting sense\n",
1175 current->comm));
1da177e4
LT
1176 rtn = scsi_request_sense(scmd);
1177 if (rtn != SUCCESS)
1178 continue;
1179
91921e01
HR
1180 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1181 "sense requested for %p result %x\n",
1182 scmd, scmd->result));
1da177e4
LT
1183 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1184
1185 rtn = scsi_decide_disposition(scmd);
1186
1187 /*
1188 * if the result was normal, then just pass it along to the
1189 * upper level.
1190 */
1191 if (rtn == SUCCESS)
1192 /* we don't want this command reissued, just
1193 * finished with the sense data, so set
1194 * retries to the max allowed to ensure it
1195 * won't get reissued */
1196 scmd->retries = scmd->allowed;
1197 else if (rtn != NEEDS_RETRY)
1198 continue;
1199
1200 scsi_eh_finish_cmd(scmd, done_q);
1201 }
1202
1203 return list_empty(work_q);
1204}
dca84e46 1205EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1da177e4 1206
1da177e4
LT
1207/**
1208 * scsi_eh_tur - Send TUR to device.
eb44820c 1209 * @scmd: &scsi_cmnd to send TUR
1da177e4
LT
1210 *
1211 * Return value:
1212 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1213 */
1da177e4
LT
1214static int scsi_eh_tur(struct scsi_cmnd *scmd)
1215{
1216 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1217 int retry_cnt = 1, rtn;
1218
1219retry_tur:
0816c925
MP
1220 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1221 scmd->device->eh_timeout, 0);
1da177e4 1222
91921e01
HR
1223 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1224 "%s: scmd %p rtn %x\n", __func__, scmd, rtn));
631c228c
CH
1225
1226 switch (rtn) {
1227 case NEEDS_RETRY:
1da177e4
LT
1228 if (retry_cnt--)
1229 goto retry_tur;
631c228c
CH
1230 /*FALLTHRU*/
1231 case SUCCESS:
e47373ec 1232 return 0;
631c228c
CH
1233 default:
1234 return 1;
e47373ec 1235 }
1da177e4
LT
1236}
1237
3eef6257
DJ
1238/**
1239 * scsi_eh_test_devices - check if devices are responding from error recovery.
1240 * @cmd_list: scsi commands in error recovery.
1241 * @work_q: queue for commands which still need more error recovery
1242 * @done_q: queue for commands which are finished
1243 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1244 *
1245 * Decription:
1246 * Tests if devices are in a working state. Commands to devices now in
1247 * a working state are sent to the done_q while commands to devices which
1248 * are still failing to respond are returned to the work_q for more
1249 * processing.
1250 **/
1251static int scsi_eh_test_devices(struct list_head *cmd_list,
1252 struct list_head *work_q,
1253 struct list_head *done_q, int try_stu)
1254{
1255 struct scsi_cmnd *scmd, *next;
1256 struct scsi_device *sdev;
1257 int finish_cmds;
1258
1259 while (!list_empty(cmd_list)) {
1260 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1261 sdev = scmd->device;
1262
b4562022 1263 if (!try_stu) {
b4562022
HR
1264 if (scsi_host_eh_past_deadline(sdev->host)) {
1265 /* Push items back onto work_q */
1266 list_splice_init(cmd_list, work_q);
b4562022
HR
1267 SCSI_LOG_ERROR_RECOVERY(3,
1268 shost_printk(KERN_INFO, sdev->host,
1269 "skip %s, past eh deadline",
1270 __func__));
1271 break;
1272 }
b4562022
HR
1273 }
1274
3eef6257
DJ
1275 finish_cmds = !scsi_device_online(scmd->device) ||
1276 (try_stu && !scsi_eh_try_stu(scmd) &&
1277 !scsi_eh_tur(scmd)) ||
1278 !scsi_eh_tur(scmd);
1279
1280 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1281 if (scmd->device == sdev) {
2451079b
JB
1282 if (finish_cmds &&
1283 (try_stu ||
1284 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
3eef6257
DJ
1285 scsi_eh_finish_cmd(scmd, done_q);
1286 else
1287 list_move_tail(&scmd->eh_entry, work_q);
1288 }
1289 }
1290 return list_empty(work_q);
1291}
1292
1293
1da177e4 1294/**
eb44820c
RL
1295 * scsi_eh_abort_cmds - abort pending commands.
1296 * @work_q: &list_head for pending commands.
1297 * @done_q: &list_head for processed commands.
1da177e4
LT
1298 *
1299 * Decription:
1300 * Try and see whether or not it makes sense to try and abort the
eb44820c
RL
1301 * running command. This only works out to be the case if we have one
1302 * command that has timed out. If the command simply failed, it makes
1da177e4
LT
1303 * no sense to try and abort the command, since as far as the shost
1304 * adapter is concerned, it isn't running.
dc8875e1 1305 */
1da177e4
LT
1306static int scsi_eh_abort_cmds(struct list_head *work_q,
1307 struct list_head *done_q)
1308{
937abeaa 1309 struct scsi_cmnd *scmd, *next;
3eef6257 1310 LIST_HEAD(check_list);
1da177e4 1311 int rtn;
b4562022 1312 struct Scsi_Host *shost;
1da177e4 1313
937abeaa 1314 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3111b0d1 1315 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1da177e4 1316 continue;
b4562022 1317 shost = scmd->device->host;
b4562022 1318 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1319 list_splice_init(&check_list, work_q);
1320 SCSI_LOG_ERROR_RECOVERY(3,
1321 shost_printk(KERN_INFO, shost,
1322 "skip %s, past eh deadline\n",
1323 __func__));
1324 return list_empty(work_q);
1325 }
91921e01
HR
1326 SCSI_LOG_ERROR_RECOVERY(3,
1327 shost_printk(KERN_INFO, shost,
1328 "%s: aborting cmd: 0x%p\n",
1329 current->comm, scmd));
b4562022 1330 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
6fd046f9 1331 if (rtn == FAILED) {
91921e01
HR
1332 SCSI_LOG_ERROR_RECOVERY(3,
1333 shost_printk(KERN_INFO, shost,
1334 "%s: aborting cmd failed: 0x%p\n",
1335 current->comm, scmd));
6fd046f9
HR
1336 list_splice_init(&check_list, work_q);
1337 return list_empty(work_q);
1338 }
1339 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1340 if (rtn == FAST_IO_FAIL)
1341 scsi_eh_finish_cmd(scmd, done_q);
1342 else
1343 list_move_tail(&scmd->eh_entry, &check_list);
1da177e4
LT
1344 }
1345
3eef6257 1346 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1347}
1348
1da177e4
LT
1349/**
1350 * scsi_eh_try_stu - Send START_UNIT to device.
eb44820c 1351 * @scmd: &scsi_cmnd to send START_UNIT
1da177e4
LT
1352 *
1353 * Return value:
1354 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1355 */
1da177e4
LT
1356static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1357{
1358 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1da177e4 1359
631c228c 1360 if (scmd->device->allow_restart) {
ed773e66
BK
1361 int i, rtn = NEEDS_RETRY;
1362
1363 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
9728c081 1364 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1da177e4 1365
631c228c
CH
1366 if (rtn == SUCCESS)
1367 return 0;
1368 }
1da177e4 1369
1da177e4
LT
1370 return 1;
1371}
1372
1373 /**
1374 * scsi_eh_stu - send START_UNIT if needed
eb44820c
RL
1375 * @shost: &scsi host being recovered.
1376 * @work_q: &list_head for pending commands.
1377 * @done_q: &list_head for processed commands.
1da177e4
LT
1378 *
1379 * Notes:
1380 * If commands are failing due to not ready, initializing command required,
0bf8c869 1381 * try revalidating the device, which will end up sending a start unit.
dc8875e1 1382 */
1da177e4
LT
1383static int scsi_eh_stu(struct Scsi_Host *shost,
1384 struct list_head *work_q,
1385 struct list_head *done_q)
1386{
937abeaa 1387 struct scsi_cmnd *scmd, *stu_scmd, *next;
1da177e4
LT
1388 struct scsi_device *sdev;
1389
1390 shost_for_each_device(sdev, shost) {
b4562022 1391 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1392 SCSI_LOG_ERROR_RECOVERY(3,
1393 shost_printk(KERN_INFO, shost,
1394 "skip %s, past eh deadline\n",
1395 __func__));
1396 break;
1397 }
1da177e4
LT
1398 stu_scmd = NULL;
1399 list_for_each_entry(scmd, work_q, eh_entry)
1400 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1401 scsi_check_sense(scmd) == FAILED ) {
1402 stu_scmd = scmd;
1403 break;
1404 }
1405
1406 if (!stu_scmd)
1407 continue;
1408
91921e01
HR
1409 SCSI_LOG_ERROR_RECOVERY(3,
1410 shost_printk(KERN_INFO, shost,
1411 "%s: Sending START_UNIT to sdev: 0x%p\n",
1412 current->comm, sdev));
1da177e4
LT
1413
1414 if (!scsi_eh_try_stu(stu_scmd)) {
1415 if (!scsi_device_online(sdev) ||
1416 !scsi_eh_tur(stu_scmd)) {
937abeaa
CH
1417 list_for_each_entry_safe(scmd, next,
1418 work_q, eh_entry) {
2451079b
JB
1419 if (scmd->device == sdev &&
1420 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1da177e4
LT
1421 scsi_eh_finish_cmd(scmd, done_q);
1422 }
1423 }
1424 } else {
1425 SCSI_LOG_ERROR_RECOVERY(3,
91921e01
HR
1426 shost_printk(KERN_INFO, shost,
1427 "%s: START_UNIT failed to sdev:"
1428 " 0x%p\n", current->comm, sdev));
1da177e4
LT
1429 }
1430 }
1431
1432 return list_empty(work_q);
1433}
1434
1435
1436/**
1437 * scsi_eh_bus_device_reset - send bdr if needed
1438 * @shost: scsi host being recovered.
eb44820c
RL
1439 * @work_q: &list_head for pending commands.
1440 * @done_q: &list_head for processed commands.
1da177e4
LT
1441 *
1442 * Notes:
eb44820c 1443 * Try a bus device reset. Still, look to see whether we have multiple
1da177e4
LT
1444 * devices that are jammed or not - if we have multiple devices, it
1445 * makes no sense to try bus_device_reset - we really would need to try
0bf8c869 1446 * a bus_reset instead.
dc8875e1 1447 */
1da177e4
LT
1448static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1449 struct list_head *work_q,
1450 struct list_head *done_q)
1451{
937abeaa 1452 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1da177e4
LT
1453 struct scsi_device *sdev;
1454 int rtn;
1455
1456 shost_for_each_device(sdev, shost) {
b4562022 1457 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1458 SCSI_LOG_ERROR_RECOVERY(3,
1459 shost_printk(KERN_INFO, shost,
1460 "skip %s, past eh deadline\n",
1461 __func__));
1462 break;
1463 }
1da177e4
LT
1464 bdr_scmd = NULL;
1465 list_for_each_entry(scmd, work_q, eh_entry)
1466 if (scmd->device == sdev) {
1467 bdr_scmd = scmd;
1468 break;
1469 }
1470
1471 if (!bdr_scmd)
1472 continue;
1473
91921e01
HR
1474 SCSI_LOG_ERROR_RECOVERY(3,
1475 shost_printk(KERN_INFO, shost,
1476 "%s: Sending BDR sdev: 0x%p\n",
1477 current->comm, sdev));
1da177e4 1478 rtn = scsi_try_bus_device_reset(bdr_scmd);
2f2eb587 1479 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1da177e4 1480 if (!scsi_device_online(sdev) ||
2f2eb587 1481 rtn == FAST_IO_FAIL ||
1da177e4 1482 !scsi_eh_tur(bdr_scmd)) {
937abeaa
CH
1483 list_for_each_entry_safe(scmd, next,
1484 work_q, eh_entry) {
2451079b
JB
1485 if (scmd->device == sdev &&
1486 scsi_eh_action(scmd, rtn) != FAILED)
1da177e4
LT
1487 scsi_eh_finish_cmd(scmd,
1488 done_q);
1489 }
1490 }
1491 } else {
91921e01
HR
1492 SCSI_LOG_ERROR_RECOVERY(3,
1493 shost_printk(KERN_INFO, shost,
1494 "%s: BDR failed sdev: 0x%p\n",
1495 current->comm, sdev));
1da177e4
LT
1496 }
1497 }
1498
1499 return list_empty(work_q);
1500}
1501
30bd7df8
MC
1502/**
1503 * scsi_eh_target_reset - send target reset if needed
1504 * @shost: scsi host being recovered.
1505 * @work_q: &list_head for pending commands.
1506 * @done_q: &list_head for processed commands.
1507 *
1508 * Notes:
1509 * Try a target reset.
1510 */
1511static int scsi_eh_target_reset(struct Scsi_Host *shost,
1512 struct list_head *work_q,
1513 struct list_head *done_q)
1514{
98db5195 1515 LIST_HEAD(tmp_list);
3eef6257 1516 LIST_HEAD(check_list);
30bd7df8 1517
98db5195
JB
1518 list_splice_init(work_q, &tmp_list);
1519
1520 while (!list_empty(&tmp_list)) {
1521 struct scsi_cmnd *next, *scmd;
1522 int rtn;
1523 unsigned int id;
b4562022 1524
b4562022 1525 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1526 /* push back on work queue for further processing */
1527 list_splice_init(&check_list, work_q);
1528 list_splice_init(&tmp_list, work_q);
1529 SCSI_LOG_ERROR_RECOVERY(3,
1530 shost_printk(KERN_INFO, shost,
1531 "skip %s, past eh deadline\n",
1532 __func__));
1533 return list_empty(work_q);
1534 }
98db5195
JB
1535
1536 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1537 id = scmd_id(scmd);
30bd7df8 1538
91921e01
HR
1539 SCSI_LOG_ERROR_RECOVERY(3,
1540 shost_printk(KERN_INFO, shost,
1541 "%s: Sending target reset to target %d\n",
1542 current->comm, id));
98db5195
JB
1543 rtn = scsi_try_target_reset(scmd);
1544 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
91921e01
HR
1545 SCSI_LOG_ERROR_RECOVERY(3,
1546 shost_printk(KERN_INFO, shost,
1547 "%s: Target reset failed"
1548 " target: %d\n",
1549 current->comm, id));
98db5195
JB
1550 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1551 if (scmd_id(scmd) != id)
1552 continue;
1553
3eef6257
DJ
1554 if (rtn == SUCCESS)
1555 list_move_tail(&scmd->eh_entry, &check_list);
1556 else if (rtn == FAST_IO_FAIL)
98db5195
JB
1557 scsi_eh_finish_cmd(scmd, done_q);
1558 else
1559 /* push back on work queue for further processing */
1560 list_move(&scmd->eh_entry, work_q);
1561 }
1562 }
30bd7df8 1563
3eef6257 1564 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
30bd7df8
MC
1565}
1566
1da177e4 1567/**
0bf8c869 1568 * scsi_eh_bus_reset - send a bus reset
eb44820c
RL
1569 * @shost: &scsi host being recovered.
1570 * @work_q: &list_head for pending commands.
1571 * @done_q: &list_head for processed commands.
dc8875e1 1572 */
1da177e4
LT
1573static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1574 struct list_head *work_q,
1575 struct list_head *done_q)
1576{
937abeaa 1577 struct scsi_cmnd *scmd, *chan_scmd, *next;
3eef6257 1578 LIST_HEAD(check_list);
1da177e4
LT
1579 unsigned int channel;
1580 int rtn;
1581
1582 /*
1583 * we really want to loop over the various channels, and do this on
1584 * a channel by channel basis. we should also check to see if any
1585 * of the failed commands are on soft_reset devices, and if so, skip
0bf8c869 1586 * the reset.
1da177e4
LT
1587 */
1588
1589 for (channel = 0; channel <= shost->max_channel; channel++) {
b4562022 1590 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1591 list_splice_init(&check_list, work_q);
1592 SCSI_LOG_ERROR_RECOVERY(3,
1593 shost_printk(KERN_INFO, shost,
1594 "skip %s, past eh deadline\n",
1595 __func__));
1596 return list_empty(work_q);
1597 }
b4562022 1598
1da177e4
LT
1599 chan_scmd = NULL;
1600 list_for_each_entry(scmd, work_q, eh_entry) {
422c0d61 1601 if (channel == scmd_channel(scmd)) {
1da177e4
LT
1602 chan_scmd = scmd;
1603 break;
1604 /*
1605 * FIXME add back in some support for
1606 * soft_reset devices.
1607 */
1608 }
1609 }
1610
1611 if (!chan_scmd)
1612 continue;
91921e01
HR
1613 SCSI_LOG_ERROR_RECOVERY(3,
1614 shost_printk(KERN_INFO, shost,
1615 "%s: Sending BRST chan: %d\n",
1616 current->comm, channel));
1da177e4 1617 rtn = scsi_try_bus_reset(chan_scmd);
2f2eb587 1618 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
937abeaa 1619 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3eef6257
DJ
1620 if (channel == scmd_channel(scmd)) {
1621 if (rtn == FAST_IO_FAIL)
1da177e4
LT
1622 scsi_eh_finish_cmd(scmd,
1623 done_q);
3eef6257
DJ
1624 else
1625 list_move_tail(&scmd->eh_entry,
1626 &check_list);
1627 }
1da177e4
LT
1628 }
1629 } else {
91921e01
HR
1630 SCSI_LOG_ERROR_RECOVERY(3,
1631 shost_printk(KERN_INFO, shost,
1632 "%s: BRST failed chan: %d\n",
1633 current->comm, channel));
1da177e4
LT
1634 }
1635 }
3eef6257 1636 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1637}
1638
1639/**
0bf8c869 1640 * scsi_eh_host_reset - send a host reset
1da177e4
LT
1641 * @work_q: list_head for processed commands.
1642 * @done_q: list_head for processed commands.
dc8875e1 1643 */
91921e01
HR
1644static int scsi_eh_host_reset(struct Scsi_Host *shost,
1645 struct list_head *work_q,
1da177e4
LT
1646 struct list_head *done_q)
1647{
937abeaa 1648 struct scsi_cmnd *scmd, *next;
3eef6257 1649 LIST_HEAD(check_list);
1da177e4 1650 int rtn;
1da177e4
LT
1651
1652 if (!list_empty(work_q)) {
1653 scmd = list_entry(work_q->next,
1654 struct scsi_cmnd, eh_entry);
1655
91921e01
HR
1656 SCSI_LOG_ERROR_RECOVERY(3,
1657 shost_printk(KERN_INFO, shost,
1658 "%s: Sending HRST\n",
1659 current->comm));
1da177e4
LT
1660
1661 rtn = scsi_try_host_reset(scmd);
3eef6257
DJ
1662 if (rtn == SUCCESS) {
1663 list_splice_init(work_q, &check_list);
1664 } else if (rtn == FAST_IO_FAIL) {
937abeaa 1665 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1da177e4
LT
1666 scsi_eh_finish_cmd(scmd, done_q);
1667 }
1668 } else {
91921e01
HR
1669 SCSI_LOG_ERROR_RECOVERY(3,
1670 shost_printk(KERN_INFO, shost,
1671 "%s: HRST failed\n",
1672 current->comm));
1da177e4
LT
1673 }
1674 }
3eef6257 1675 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1da177e4
LT
1676}
1677
1678/**
1679 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1680 * @work_q: list_head for processed commands.
1681 * @done_q: list_head for processed commands.
dc8875e1 1682 */
1da177e4
LT
1683static void scsi_eh_offline_sdevs(struct list_head *work_q,
1684 struct list_head *done_q)
1685{
937abeaa 1686 struct scsi_cmnd *scmd, *next;
1da177e4 1687
937abeaa 1688 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
31765d7d
MW
1689 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1690 "not ready after error recovery\n");
1da177e4 1691 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
3111b0d1 1692 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1da177e4
LT
1693 /*
1694 * FIXME: Handle lost cmds.
1695 */
1696 }
1697 scsi_eh_finish_cmd(scmd, done_q);
1698 }
1699 return;
1700}
1701
4a27446f 1702/**
e494f6a7 1703 * scsi_noretry_cmd - determine if command should be failed fast
4a27446f
MC
1704 * @scmd: SCSI cmd to examine.
1705 */
1706int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1707{
1708 switch (host_byte(scmd->result)) {
1709 case DID_OK:
1710 break;
e494f6a7
HR
1711 case DID_TIME_OUT:
1712 goto check_type;
4a27446f 1713 case DID_BUS_BUSY:
33659ebb 1714 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
4a27446f 1715 case DID_PARITY:
33659ebb 1716 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
4a27446f
MC
1717 case DID_ERROR:
1718 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1719 status_byte(scmd->result) == RESERVATION_CONFLICT)
1720 return 0;
1721 /* fall through */
1722 case DID_SOFT_ERROR:
33659ebb 1723 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
4a27446f
MC
1724 }
1725
e494f6a7
HR
1726 if (status_byte(scmd->result) != CHECK_CONDITION)
1727 return 0;
4a27446f 1728
e494f6a7
HR
1729check_type:
1730 /*
1731 * assume caller has checked sense and determined
1732 * the check condition was retryable.
1733 */
1734 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1735 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1736 return 1;
1737 else
1738 return 0;
4a27446f
MC
1739}
1740
1da177e4
LT
1741/**
1742 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1743 * @scmd: SCSI cmd to examine.
1744 *
1745 * Notes:
1746 * This is *only* called when we are examining the status after sending
1747 * out the actual data command. any commands that are queued for error
1748 * recovery (e.g. test_unit_ready) do *not* come through here.
1749 *
1750 * When this routine returns failed, it means the error handler thread
1751 * is woken. In cases where the error code indicates an error that
1752 * doesn't require the error handler read (i.e. we don't need to
1753 * abort/reset), this function should return SUCCESS.
dc8875e1 1754 */
1da177e4
LT
1755int scsi_decide_disposition(struct scsi_cmnd *scmd)
1756{
1757 int rtn;
1758
1759 /*
1760 * if the device is offline, then we clearly just pass the result back
1761 * up to the top level.
1762 */
1763 if (!scsi_device_online(scmd->device)) {
91921e01
HR
1764 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1765 "%s: device offline - report as SUCCESS\n", __func__));
1da177e4
LT
1766 return SUCCESS;
1767 }
1768
1769 /*
1770 * first check the host byte, to see if there is anything in there
1771 * that would indicate what we need to do.
1772 */
1773 switch (host_byte(scmd->result)) {
1774 case DID_PASSTHROUGH:
1775 /*
1776 * no matter what, pass this through to the upper layer.
1777 * nuke this special code so that it looks like we are saying
1778 * did_ok.
1779 */
1780 scmd->result &= 0xff00ffff;
1781 return SUCCESS;
1782 case DID_OK:
1783 /*
1784 * looks good. drop through, and check the next byte.
1785 */
1786 break;
e494f6a7
HR
1787 case DID_ABORT:
1788 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
8922a908 1789 set_host_byte(scmd, DID_TIME_OUT);
e494f6a7
HR
1790 return SUCCESS;
1791 }
1da177e4
LT
1792 case DID_NO_CONNECT:
1793 case DID_BAD_TARGET:
1da177e4
LT
1794 /*
1795 * note - this means that we just report the status back
1796 * to the top level driver, not that we actually think
1797 * that it indicates SUCCESS.
1798 */
1799 return SUCCESS;
1800 /*
1801 * when the low level driver returns did_soft_error,
0bf8c869 1802 * it is responsible for keeping an internal retry counter
1da177e4
LT
1803 * in order to avoid endless loops (db)
1804 *
1805 * actually this is a bug in this function here. we should
1806 * be mindful of the maximum number of retries specified
1807 * and not get stuck in a loop.
1808 */
1809 case DID_SOFT_ERROR:
1810 goto maybe_retry;
1811 case DID_IMM_RETRY:
1812 return NEEDS_RETRY;
1813
bf341919
JB
1814 case DID_REQUEUE:
1815 return ADD_TO_MLQUEUE;
a4dfaa6f
MC
1816 case DID_TRANSPORT_DISRUPTED:
1817 /*
1818 * LLD/transport was disrupted during processing of the IO.
1819 * The transport class is now blocked/blocking,
1820 * and the transport will decide what to do with the IO
939c2288
MC
1821 * based on its timers and recovery capablilities if
1822 * there are enough retries.
a4dfaa6f 1823 */
939c2288 1824 goto maybe_retry;
a4dfaa6f
MC
1825 case DID_TRANSPORT_FAILFAST:
1826 /*
1827 * The transport decided to failfast the IO (most likely
1828 * the fast io fail tmo fired), so send IO directly upwards.
1829 */
1830 return SUCCESS;
1da177e4
LT
1831 case DID_ERROR:
1832 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1833 status_byte(scmd->result) == RESERVATION_CONFLICT)
1834 /*
1835 * execute reservation conflict processing code
1836 * lower down
1837 */
1838 break;
1839 /* fallthrough */
1da177e4
LT
1840 case DID_BUS_BUSY:
1841 case DID_PARITY:
1842 goto maybe_retry;
1843 case DID_TIME_OUT:
1844 /*
1845 * when we scan the bus, we get timeout messages for
1846 * these commands if there is no device available.
1847 * other hosts report did_no_connect for the same thing.
1848 */
1849 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1850 scmd->cmnd[0] == INQUIRY)) {
1851 return SUCCESS;
1852 } else {
1853 return FAILED;
1854 }
1855 case DID_RESET:
1856 return SUCCESS;
1857 default:
1858 return FAILED;
1859 }
1860
1861 /*
1862 * next, check the message byte.
1863 */
1864 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1865 return FAILED;
1866
1867 /*
1868 * check the status byte to see if this indicates anything special.
1869 */
1870 switch (status_byte(scmd->result)) {
1871 case QUEUE_FULL:
42a6a918 1872 scsi_handle_queue_full(scmd->device);
1da177e4
LT
1873 /*
1874 * the case of trying to send too many commands to a
1875 * tagged queueing device.
1876 */
1877 case BUSY:
1878 /*
1879 * device can't talk to us at the moment. Should only
1880 * occur (SAM-3) when the task queue is empty, so will cause
1881 * the empty queue handling to trigger a stall in the
1882 * device.
1883 */
1884 return ADD_TO_MLQUEUE;
1885 case GOOD:
279afdfe
EM
1886 if (scmd->cmnd[0] == REPORT_LUNS)
1887 scmd->device->sdev_target->expecting_lun_change = 0;
4a84067d 1888 scsi_handle_queue_ramp_up(scmd->device);
1da177e4 1889 case COMMAND_TERMINATED:
1da177e4 1890 return SUCCESS;
a9b589d9
VB
1891 case TASK_ABORTED:
1892 goto maybe_retry;
1da177e4
LT
1893 case CHECK_CONDITION:
1894 rtn = scsi_check_sense(scmd);
1895 if (rtn == NEEDS_RETRY)
1896 goto maybe_retry;
1897 /* if rtn == FAILED, we have no sense information;
1898 * returning FAILED will wake the error handler thread
1899 * to collect the sense and redo the decide
1900 * disposition */
1901 return rtn;
1902 case CONDITION_GOOD:
1903 case INTERMEDIATE_GOOD:
1904 case INTERMEDIATE_C_GOOD:
1905 case ACA_ACTIVE:
1906 /*
1907 * who knows? FIXME(eric)
1908 */
1909 return SUCCESS;
1910
1911 case RESERVATION_CONFLICT:
9ccfc756
JB
1912 sdev_printk(KERN_INFO, scmd->device,
1913 "reservation conflict\n");
2082ebc4 1914 set_host_byte(scmd, DID_NEXUS_FAILURE);
1da177e4
LT
1915 return SUCCESS; /* causes immediate i/o error */
1916 default:
1917 return FAILED;
1918 }
1919 return FAILED;
1920
1921 maybe_retry:
1922
1923 /* we requeue for retry because the error was retryable, and
1924 * the request was not marked fast fail. Note that above,
1925 * even if the request is marked fast fail, we still requeue
1926 * for queue congestion conditions (QUEUE_FULL or BUSY) */
8884efab 1927 if ((++scmd->retries) <= scmd->allowed
4a27446f 1928 && !scsi_noretry_cmd(scmd)) {
1da177e4
LT
1929 return NEEDS_RETRY;
1930 } else {
1931 /*
1932 * no more retries - report this one back to upper level.
1933 */
1934 return SUCCESS;
1935 }
1936}
1937
f078727b
FT
1938static void eh_lock_door_done(struct request *req, int uptodate)
1939{
1940 __blk_put_request(req->q, req);
1941}
1942
1da177e4
LT
1943/**
1944 * scsi_eh_lock_door - Prevent medium removal for the specified device
1945 * @sdev: SCSI device to prevent medium removal
1946 *
1947 * Locking:
91bc31fb 1948 * We must be called from process context.
1da177e4
LT
1949 *
1950 * Notes:
1951 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1952 * head of the devices request queue, and continue.
dc8875e1 1953 */
1da177e4
LT
1954static void scsi_eh_lock_door(struct scsi_device *sdev)
1955{
f078727b 1956 struct request *req;
1da177e4 1957
91bc31fb
JB
1958 /*
1959 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1960 * request becomes available
1961 */
f078727b 1962 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
a492f075 1963 if (IS_ERR(req))
eb571eea 1964 return;
1da177e4 1965
f27b087b
JA
1966 blk_rq_set_block_pc(req);
1967
f078727b
FT
1968 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1969 req->cmd[1] = 0;
1970 req->cmd[2] = 0;
1971 req->cmd[3] = 0;
1972 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1973 req->cmd[5] = 0;
1da177e4 1974
f078727b
FT
1975 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1976
f078727b
FT
1977 req->cmd_flags |= REQ_QUIET;
1978 req->timeout = 10 * HZ;
1979 req->retries = 5;
1980
1981 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1982}
1da177e4
LT
1983
1984/**
1985 * scsi_restart_operations - restart io operations to the specified host.
1986 * @shost: Host we are restarting.
1987 *
1988 * Notes:
1989 * When we entered the error handler, we blocked all further i/o to
1990 * this device. we need to 'reverse' this process.
dc8875e1 1991 */
1da177e4
LT
1992static void scsi_restart_operations(struct Scsi_Host *shost)
1993{
1994 struct scsi_device *sdev;
939647ee 1995 unsigned long flags;
1da177e4
LT
1996
1997 /*
1998 * If the door was locked, we need to insert a door lock request
1999 * onto the head of the SCSI request queue for the device. There
2000 * is no point trying to lock the door of an off-line device.
2001 */
2002 shost_for_each_device(sdev, shost) {
2003 if (scsi_device_online(sdev) && sdev->locked)
2004 scsi_eh_lock_door(sdev);
2005 }
2006
2007 /*
2008 * next free up anything directly waiting upon the host. this
2009 * will be requests for character device operations, and also for
2010 * ioctls to queued block devices.
2011 */
b4562022 2012 SCSI_LOG_ERROR_RECOVERY(3,
91921e01 2013 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1da177e4 2014
939647ee
JB
2015 spin_lock_irqsave(shost->host_lock, flags);
2016 if (scsi_host_set_state(shost, SHOST_RUNNING))
2017 if (scsi_host_set_state(shost, SHOST_CANCEL))
2018 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2019 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2020
2021 wake_up(&shost->host_wait);
2022
2023 /*
2024 * finally we need to re-initiate requests that may be pending. we will
2025 * have had everything blocked while error handling is taking place, and
2026 * now that error recovery is done, we will need to ensure that these
2027 * requests are started.
2028 */
2029 scsi_run_host_queues(shost);
57fc2e33
DW
2030
2031 /*
2032 * if eh is active and host_eh_scheduled is pending we need to re-run
2033 * recovery. we do this check after scsi_run_host_queues() to allow
2034 * everything pent up since the last eh run a chance to make forward
2035 * progress before we sync again. Either we'll immediately re-run
2036 * recovery or scsi_device_unbusy() will wake us again when these
2037 * pending commands complete.
2038 */
2039 spin_lock_irqsave(shost->host_lock, flags);
2040 if (shost->host_eh_scheduled)
2041 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2042 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2043 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2044}
2045
2046/**
2047 * scsi_eh_ready_devs - check device ready state and recover if not.
2048 * @shost: host to be recovered.
eb44820c
RL
2049 * @work_q: &list_head for pending commands.
2050 * @done_q: &list_head for processed commands.
dc8875e1 2051 */
dca84e46
DW
2052void scsi_eh_ready_devs(struct Scsi_Host *shost,
2053 struct list_head *work_q,
2054 struct list_head *done_q)
1da177e4
LT
2055{
2056 if (!scsi_eh_stu(shost, work_q, done_q))
2057 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
30bd7df8
MC
2058 if (!scsi_eh_target_reset(shost, work_q, done_q))
2059 if (!scsi_eh_bus_reset(shost, work_q, done_q))
91921e01 2060 if (!scsi_eh_host_reset(shost, work_q, done_q))
30bd7df8
MC
2061 scsi_eh_offline_sdevs(work_q,
2062 done_q);
1da177e4 2063}
dca84e46 2064EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1da177e4
LT
2065
2066/**
2067 * scsi_eh_flush_done_q - finish processed commands or retry them.
2068 * @done_q: list_head of processed commands.
dc8875e1 2069 */
041c5fc3 2070void scsi_eh_flush_done_q(struct list_head *done_q)
1da177e4 2071{
937abeaa 2072 struct scsi_cmnd *scmd, *next;
1da177e4 2073
937abeaa
CH
2074 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2075 list_del_init(&scmd->eh_entry);
1da177e4 2076 if (scsi_device_online(scmd->device) &&
4a27446f 2077 !scsi_noretry_cmd(scmd) &&
8884efab 2078 (++scmd->retries <= scmd->allowed)) {
91921e01
HR
2079 SCSI_LOG_ERROR_RECOVERY(3,
2080 scmd_printk(KERN_INFO, scmd,
2081 "%s: flush retry cmd: %p\n",
2082 current->comm, scmd));
1da177e4
LT
2083 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2084 } else {
793698ce
PM
2085 /*
2086 * If just we got sense for the device (called
2087 * scsi_eh_get_sense), scmd->result is already
2088 * set, do not set DRIVER_TIMEOUT.
2089 */
1da177e4
LT
2090 if (!scmd->result)
2091 scmd->result |= (DRIVER_TIMEOUT << 24);
91921e01
HR
2092 SCSI_LOG_ERROR_RECOVERY(3,
2093 scmd_printk(KERN_INFO, scmd,
2094 "%s: flush finish cmd: %p\n",
2095 current->comm, scmd));
1da177e4
LT
2096 scsi_finish_command(scmd);
2097 }
2098 }
2099}
041c5fc3 2100EXPORT_SYMBOL(scsi_eh_flush_done_q);
1da177e4
LT
2101
2102/**
2103 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2104 * @shost: Host to unjam.
2105 *
2106 * Notes:
2107 * When we come in here, we *know* that all commands on the bus have
2108 * either completed, failed or timed out. we also know that no further
2109 * commands are being sent to the host, so things are relatively quiet
2110 * and we have freedom to fiddle with things as we wish.
2111 *
2112 * This is only the *default* implementation. it is possible for
2113 * individual drivers to supply their own version of this function, and
2114 * if the maintainer wishes to do this, it is strongly suggested that
2115 * this function be taken as a template and modified. this function
2116 * was designed to correctly handle problems for about 95% of the
2117 * different cases out there, and it should always provide at least a
2118 * reasonable amount of error recovery.
2119 *
2120 * Any command marked 'failed' or 'timeout' must eventually have
2121 * scsi_finish_cmd() called for it. we do all of the retry stuff
2122 * here, so when we restart the host after we return it should have an
2123 * empty queue.
dc8875e1 2124 */
1da177e4
LT
2125static void scsi_unjam_host(struct Scsi_Host *shost)
2126{
2127 unsigned long flags;
2128 LIST_HEAD(eh_work_q);
2129 LIST_HEAD(eh_done_q);
2130
2131 spin_lock_irqsave(shost->host_lock, flags);
2132 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2133 spin_unlock_irqrestore(shost->host_lock, flags);
2134
2135 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2136
2137 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2138 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2139 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2140
b4562022 2141 spin_lock_irqsave(shost->host_lock, flags);
bb3b621a 2142 if (shost->eh_deadline != -1)
b4562022
HR
2143 shost->last_reset = 0;
2144 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2145 scsi_eh_flush_done_q(&eh_done_q);
2146}
2147
2148/**
ad42eb1b 2149 * scsi_error_handler - SCSI error handler thread
1da177e4
LT
2150 * @data: Host for which we are running.
2151 *
2152 * Notes:
ad42eb1b
CH
2153 * This is the main error handling loop. This is run as a kernel thread
2154 * for every SCSI host and handles all error handling activity.
dc8875e1 2155 */
1da177e4
LT
2156int scsi_error_handler(void *data)
2157{
ad42eb1b 2158 struct Scsi_Host *shost = data;
1da177e4 2159
1da177e4 2160 /*
ad42eb1b
CH
2161 * We use TASK_INTERRUPTIBLE so that the thread is not
2162 * counted against the load average as a running process.
2163 * We never actually get interrupted because kthread_run
c03264a7 2164 * disables signal delivery for the created thread.
1da177e4 2165 */
3ed7a470 2166 while (!kthread_should_stop()) {
b9d5c6b7 2167 set_current_state(TASK_INTERRUPTIBLE);
ee7863bc 2168 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
74665016 2169 shost->host_failed != atomic_read(&shost->host_busy)) {
ad42eb1b 2170 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2171 shost_printk(KERN_INFO, shost,
2172 "scsi_eh_%d: sleeping\n",
2173 shost->host_no));
3ed7a470 2174 schedule();
3ed7a470
JB
2175 continue;
2176 }
1da177e4 2177
3ed7a470 2178 __set_current_state(TASK_RUNNING);
ad42eb1b 2179 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2180 shost_printk(KERN_INFO, shost,
2181 "scsi_eh_%d: waking up %d/%d/%d\n",
2182 shost->host_no, shost->host_eh_scheduled,
74665016
CH
2183 shost->host_failed,
2184 atomic_read(&shost->host_busy)));
1da177e4 2185
1da177e4
LT
2186 /*
2187 * We have a host that is failing for some reason. Figure out
2188 * what we need to do to get it up and online again (if we can).
2189 * If we fail, we end up taking the thing offline.
2190 */
ae0751ff 2191 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
bc4f2401
AS
2192 SCSI_LOG_ERROR_RECOVERY(1,
2193 printk(KERN_ERR "Error handler scsi_eh_%d "
2194 "unable to autoresume\n",
2195 shost->host_no));
2196 continue;
2197 }
2198
9227c33d
CH
2199 if (shost->transportt->eh_strategy_handler)
2200 shost->transportt->eh_strategy_handler(shost);
1da177e4
LT
2201 else
2202 scsi_unjam_host(shost);
2203
1da177e4
LT
2204 /*
2205 * Note - if the above fails completely, the action is to take
2206 * individual devices offline and flush the queue of any
2207 * outstanding requests that may have been pending. When we
2208 * restart, we restart any I/O to any other devices on the bus
2209 * which are still online.
2210 */
2211 scsi_restart_operations(shost);
ae0751ff
LM
2212 if (!shost->eh_noresume)
2213 scsi_autopm_put_host(shost);
1da177e4 2214 }
461a0ffb
SR
2215 __set_current_state(TASK_RUNNING);
2216
ad42eb1b 2217 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2218 shost_printk(KERN_INFO, shost,
2219 "Error handler scsi_eh_%d exiting\n",
2220 shost->host_no));
3ed7a470 2221 shost->ehandler = NULL;
1da177e4
LT
2222 return 0;
2223}
2224
2225/*
2226 * Function: scsi_report_bus_reset()
2227 *
2228 * Purpose: Utility function used by low-level drivers to report that
2229 * they have observed a bus reset on the bus being handled.
2230 *
2231 * Arguments: shost - Host in question
2232 * channel - channel on which reset was observed.
2233 *
2234 * Returns: Nothing
2235 *
2236 * Lock status: Host lock must be held.
2237 *
2238 * Notes: This only needs to be called if the reset is one which
2239 * originates from an unknown location. Resets originated
2240 * by the mid-level itself don't need to call this, but there
2241 * should be no harm.
2242 *
2243 * The main purpose of this is to make sure that a CHECK_CONDITION
2244 * is properly treated.
2245 */
2246void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2247{
2248 struct scsi_device *sdev;
2249
2250 __shost_for_each_device(sdev, shost) {
30bd7df8
MC
2251 if (channel == sdev_channel(sdev))
2252 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2253 }
2254}
2255EXPORT_SYMBOL(scsi_report_bus_reset);
2256
2257/*
2258 * Function: scsi_report_device_reset()
2259 *
2260 * Purpose: Utility function used by low-level drivers to report that
2261 * they have observed a device reset on the device being handled.
2262 *
2263 * Arguments: shost - Host in question
2264 * channel - channel on which reset was observed
2265 * target - target on which reset was observed
2266 *
2267 * Returns: Nothing
2268 *
2269 * Lock status: Host lock must be held
2270 *
2271 * Notes: This only needs to be called if the reset is one which
2272 * originates from an unknown location. Resets originated
2273 * by the mid-level itself don't need to call this, but there
2274 * should be no harm.
2275 *
2276 * The main purpose of this is to make sure that a CHECK_CONDITION
2277 * is properly treated.
2278 */
2279void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2280{
2281 struct scsi_device *sdev;
2282
2283 __shost_for_each_device(sdev, shost) {
422c0d61 2284 if (channel == sdev_channel(sdev) &&
30bd7df8
MC
2285 target == sdev_id(sdev))
2286 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2287 }
2288}
2289EXPORT_SYMBOL(scsi_report_device_reset);
2290
2291static void
2292scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2293{
2294}
2295
2296/*
2297 * Function: scsi_reset_provider
2298 *
2299 * Purpose: Send requested reset to a bus or device at any phase.
2300 *
2301 * Arguments: device - device to send reset to
2302 * flag - reset type (see scsi.h)
2303 *
2304 * Returns: SUCCESS/FAILURE.
2305 *
2306 * Notes: This is used by the SCSI Generic driver to provide
2307 * Bus/Device reset capability.
2308 */
2309int
2310scsi_reset_provider(struct scsi_device *dev, int flag)
2311{
bc4f2401 2312 struct scsi_cmnd *scmd;
d7a1bb0a 2313 struct Scsi_Host *shost = dev->host;
1da177e4 2314 struct request req;
d7a1bb0a 2315 unsigned long flags;
1da177e4
LT
2316 int rtn;
2317
bc4f2401
AS
2318 if (scsi_autopm_get_host(shost) < 0)
2319 return FAILED;
2320
04796336
CH
2321 if (!get_device(&dev->sdev_gendev)) {
2322 rtn = FAILED;
2323 goto out_put_autopm_host;
2324 }
2325
bc4f2401 2326 scmd = scsi_get_command(dev, GFP_KERNEL);
95eeb5f5
CH
2327 if (!scmd) {
2328 rtn = FAILED;
2329 put_device(&dev->sdev_gendev);
2330 goto out_put_autopm_host;
2331 }
2332
4f54eec8 2333 blk_rq_init(NULL, &req);
1da177e4 2334 scmd->request = &req;
b4edcbca 2335
64a87b24
BH
2336 scmd->cmnd = req.cmd;
2337
1da177e4 2338 scmd->scsi_done = scsi_reset_provider_done_command;
30b0c37b 2339 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1da177e4
LT
2340
2341 scmd->cmd_len = 0;
2342
2343 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1da177e4 2344
d7a1bb0a
JS
2345 spin_lock_irqsave(shost->host_lock, flags);
2346 shost->tmf_in_progress = 1;
2347 spin_unlock_irqrestore(shost->host_lock, flags);
2348
1da177e4
LT
2349 switch (flag) {
2350 case SCSI_TRY_RESET_DEVICE:
2351 rtn = scsi_try_bus_device_reset(scmd);
2352 if (rtn == SUCCESS)
2353 break;
2354 /* FALLTHROUGH */
30bd7df8
MC
2355 case SCSI_TRY_RESET_TARGET:
2356 rtn = scsi_try_target_reset(scmd);
2357 if (rtn == SUCCESS)
2358 break;
2359 /* FALLTHROUGH */
1da177e4
LT
2360 case SCSI_TRY_RESET_BUS:
2361 rtn = scsi_try_bus_reset(scmd);
2362 if (rtn == SUCCESS)
2363 break;
2364 /* FALLTHROUGH */
2365 case SCSI_TRY_RESET_HOST:
2366 rtn = scsi_try_host_reset(scmd);
2367 break;
2368 default:
2369 rtn = FAILED;
2370 }
2371
d7a1bb0a
JS
2372 spin_lock_irqsave(shost->host_lock, flags);
2373 shost->tmf_in_progress = 0;
2374 spin_unlock_irqrestore(shost->host_lock, flags);
2375
2376 /*
2377 * be sure to wake up anyone who was sleeping or had their queue
2378 * suspended while we performed the TMF.
2379 */
2380 SCSI_LOG_ERROR_RECOVERY(3,
91921e01
HR
2381 shost_printk(KERN_INFO, shost,
2382 "waking up host to restart after TMF\n"));
d7a1bb0a
JS
2383
2384 wake_up(&shost->host_wait);
2385
2386 scsi_run_host_queues(shost);
2387
1da177e4 2388 scsi_next_command(scmd);
04796336 2389out_put_autopm_host:
bc4f2401 2390 scsi_autopm_put_host(shost);
1da177e4
LT
2391 return rtn;
2392}
2393EXPORT_SYMBOL(scsi_reset_provider);
2394
2395/**
2396 * scsi_normalize_sense - normalize main elements from either fixed or
2397 * descriptor sense data format into a common format.
2398 *
2399 * @sense_buffer: byte array containing sense data returned by device
2400 * @sb_len: number of valid bytes in sense_buffer
2401 * @sshdr: pointer to instance of structure that common
2402 * elements are written to.
2403 *
2404 * Notes:
2405 * The "main elements" from sense data are: response_code, sense_key,
2406 * asc, ascq and additional_length (only for descriptor format).
2407 *
2408 * Typically this function can be called after a device has
2409 * responded to a SCSI command with the CHECK_CONDITION status.
2410 *
2411 * Return value:
2412 * 1 if valid sense data information found, else 0;
dc8875e1 2413 */
1da177e4
LT
2414int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2415 struct scsi_sense_hdr *sshdr)
2416{
33aa687d 2417 if (!sense_buffer || !sb_len)
1da177e4
LT
2418 return 0;
2419
2420 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2421
2422 sshdr->response_code = (sense_buffer[0] & 0x7f);
33aa687d
JB
2423
2424 if (!scsi_sense_valid(sshdr))
2425 return 0;
2426
1da177e4
LT
2427 if (sshdr->response_code >= 0x72) {
2428 /*
2429 * descriptor format
2430 */
2431 if (sb_len > 1)
2432 sshdr->sense_key = (sense_buffer[1] & 0xf);
2433 if (sb_len > 2)
2434 sshdr->asc = sense_buffer[2];
2435 if (sb_len > 3)
2436 sshdr->ascq = sense_buffer[3];
2437 if (sb_len > 7)
2438 sshdr->additional_length = sense_buffer[7];
2439 } else {
0bf8c869 2440 /*
1da177e4
LT
2441 * fixed format
2442 */
2443 if (sb_len > 2)
2444 sshdr->sense_key = (sense_buffer[2] & 0xf);
2445 if (sb_len > 7) {
2446 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2447 sb_len : (sense_buffer[7] + 8);
2448 if (sb_len > 12)
2449 sshdr->asc = sense_buffer[12];
2450 if (sb_len > 13)
2451 sshdr->ascq = sense_buffer[13];
2452 }
2453 }
2454
2455 return 1;
2456}
2457EXPORT_SYMBOL(scsi_normalize_sense);
2458
1da177e4
LT
2459int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2460 struct scsi_sense_hdr *sshdr)
2461{
2462 return scsi_normalize_sense(cmd->sense_buffer,
b80ca4f7 2463 SCSI_SENSE_BUFFERSIZE, sshdr);
1da177e4
LT
2464}
2465EXPORT_SYMBOL(scsi_command_normalize_sense);
2466
2467/**
eb44820c 2468 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1da177e4
LT
2469 * @sense_buffer: byte array of descriptor format sense data
2470 * @sb_len: number of valid bytes in sense_buffer
2471 * @desc_type: value of descriptor type to find
2472 * (e.g. 0 -> information)
2473 *
2474 * Notes:
2475 * only valid when sense data is in descriptor format
2476 *
2477 * Return value:
2478 * pointer to start of (first) descriptor if found else NULL
dc8875e1 2479 */
1da177e4
LT
2480const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2481 int desc_type)
2482{
2483 int add_sen_len, add_len, desc_len, k;
2484 const u8 * descp;
2485
2486 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2487 return NULL;
2488 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2489 return NULL;
2490 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2491 add_sen_len : (sb_len - 8);
2492 descp = &sense_buffer[8];
2493 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2494 descp += desc_len;
2495 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2496 desc_len = add_len + 2;
2497 if (descp[0] == desc_type)
2498 return descp;
2499 if (add_len < 0) // short descriptor ??
2500 break;
2501 }
2502 return NULL;
2503}
2504EXPORT_SYMBOL(scsi_sense_desc_find);
2505
2506/**
eb44820c 2507 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1da177e4
LT
2508 * @sense_buffer: byte array of sense data
2509 * @sb_len: number of valid bytes in sense_buffer
2510 * @info_out: pointer to 64 integer where 8 or 4 byte information
2511 * field will be placed if found.
2512 *
2513 * Return value:
2514 * 1 if information field found, 0 if not found.
dc8875e1 2515 */
1da177e4
LT
2516int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2517 u64 * info_out)
2518{
2519 int j;
2520 const u8 * ucp;
2521 u64 ull;
2522
2523 if (sb_len < 7)
2524 return 0;
2525 switch (sense_buffer[0] & 0x7f) {
2526 case 0x70:
2527 case 0x71:
2528 if (sense_buffer[0] & 0x80) {
2529 *info_out = (sense_buffer[3] << 24) +
2530 (sense_buffer[4] << 16) +
2531 (sense_buffer[5] << 8) + sense_buffer[6];
2532 return 1;
2533 } else
2534 return 0;
2535 case 0x72:
2536 case 0x73:
2537 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2538 0 /* info desc */);
2539 if (ucp && (0xa == ucp[1])) {
2540 ull = 0;
2541 for (j = 0; j < 8; ++j) {
2542 if (j > 0)
2543 ull <<= 8;
2544 ull |= ucp[4 + j];
2545 }
2546 *info_out = ull;
2547 return 1;
2548 } else
2549 return 0;
2550 default:
2551 return 0;
2552 }
2553}
2554EXPORT_SYMBOL(scsi_get_sense_info_fld);
3bc6a261
FT
2555
2556/**
2557 * scsi_build_sense_buffer - build sense data in a buffer
2558 * @desc: Sense format (non zero == descriptor format,
2559 * 0 == fixed format)
2560 * @buf: Where to build sense data
2561 * @key: Sense key
2562 * @asc: Additional sense code
2563 * @ascq: Additional sense code qualifier
2564 *
2565 **/
2566void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2567{
2568 if (desc) {
2569 buf[0] = 0x72; /* descriptor, current */
2570 buf[1] = key;
2571 buf[2] = asc;
2572 buf[3] = ascq;
2573 buf[7] = 0;
2574 } else {
2575 buf[0] = 0x70; /* fixed, current */
2576 buf[2] = key;
2577 buf[7] = 0xa;
2578 buf[12] = asc;
2579 buf[13] = ascq;
2580 }
2581}
2582EXPORT_SYMBOL(scsi_build_sense_buffer);