blob: 4b6aa30f4d68529f1a51e0cdaf278c3e2918ec81 [file] [log] [blame]
Tejun Heoece1d632006-04-02 18:51:53 +09001/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
35#include <linux/config.h>
36#include <linux/kernel.h>
37#include <scsi/scsi.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_eh.h>
40#include <scsi/scsi_device.h>
41#include <scsi/scsi_cmnd.h>
Tejun Heof8bbfc22006-05-19 21:07:05 +090042#include "scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090043
44#include <linux/libata.h>
45
46#include "libata.h"
47
Tejun Heoad9e2762006-05-15 20:58:12 +090048static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo720ba122006-05-31 18:28:13 +090049static void ata_eh_finish(struct ata_port *ap);
Tejun Heo500530f2006-07-03 16:07:27 +090050static void ata_eh_handle_port_suspend(struct ata_port *ap);
51static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heoad9e2762006-05-15 20:58:12 +090052
Tejun Heo0c247c52006-05-15 20:58:19 +090053static void ata_ering_record(struct ata_ering *ering, int is_io,
54 unsigned int err_mask)
55{
56 struct ata_ering_entry *ent;
57
58 WARN_ON(!err_mask);
59
60 ering->cursor++;
61 ering->cursor %= ATA_ERING_SIZE;
62
63 ent = &ering->ring[ering->cursor];
64 ent->is_io = is_io;
65 ent->err_mask = err_mask;
66 ent->timestamp = get_jiffies_64();
67}
68
69static struct ata_ering_entry * ata_ering_top(struct ata_ering *ering)
70{
71 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
72 if (!ent->err_mask)
73 return NULL;
74 return ent;
75}
76
77static int ata_ering_map(struct ata_ering *ering,
78 int (*map_fn)(struct ata_ering_entry *, void *),
79 void *arg)
80{
81 int idx, rc = 0;
82 struct ata_ering_entry *ent;
83
84 idx = ering->cursor;
85 do {
86 ent = &ering->ring[idx];
87 if (!ent->err_mask)
88 break;
89 rc = map_fn(ent, arg);
90 if (rc)
91 break;
92 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
93 } while (idx != ering->cursor);
94
95 return rc;
96}
97
Tejun Heo64f65ca2006-06-24 20:30:18 +090098static unsigned int ata_eh_dev_action(struct ata_device *dev)
99{
100 struct ata_eh_context *ehc = &dev->ap->eh_context;
101
102 return ehc->i.action | ehc->i.dev_action[dev->devno];
103}
104
Tejun Heoaf181c22006-06-24 20:30:18 +0900105static void ata_eh_clear_action(struct ata_device *dev,
106 struct ata_eh_info *ehi, unsigned int action)
107{
108 int i;
109
110 if (!dev) {
111 ehi->action &= ~action;
112 for (i = 0; i < ATA_MAX_DEVICES; i++)
113 ehi->dev_action[i] &= ~action;
114 } else {
115 /* doesn't make sense for port-wide EH actions */
116 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
117
118 /* break ehi->action into ehi->dev_action */
119 if (ehi->action & action) {
120 for (i = 0; i < ATA_MAX_DEVICES; i++)
121 ehi->dev_action[i] |= ehi->action & action;
122 ehi->action &= ~action;
123 }
124
125 /* turn off the specified per-dev action */
126 ehi->dev_action[dev->devno] &= ~action;
127 }
128}
129
Tejun Heoece1d632006-04-02 18:51:53 +0900130/**
131 * ata_scsi_timed_out - SCSI layer time out callback
132 * @cmd: timed out SCSI command
133 *
134 * Handles SCSI layer timeout. We race with normal completion of
135 * the qc for @cmd. If the qc is already gone, we lose and let
136 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
137 * timed out and EH should be invoked. Prevent ata_qc_complete()
138 * from finishing it by setting EH_SCHEDULED and return
139 * EH_NOT_HANDLED.
140 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900141 * TODO: kill this function once old EH is gone.
142 *
Tejun Heoece1d632006-04-02 18:51:53 +0900143 * LOCKING:
144 * Called from timer context
145 *
146 * RETURNS:
147 * EH_HANDLED or EH_NOT_HANDLED
148 */
149enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
150{
151 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400152 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900153 unsigned long flags;
154 struct ata_queued_cmd *qc;
Tejun Heoad9e2762006-05-15 20:58:12 +0900155 enum scsi_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900156
157 DPRINTK("ENTER\n");
158
Tejun Heoad9e2762006-05-15 20:58:12 +0900159 if (ap->ops->error_handler) {
160 ret = EH_NOT_HANDLED;
161 goto out;
162 }
163
164 ret = EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400165 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900166 qc = ata_qc_from_tag(ap, ap->active_tag);
167 if (qc) {
168 WARN_ON(qc->scsicmd != cmd);
169 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
170 qc->err_mask |= AC_ERR_TIMEOUT;
171 ret = EH_NOT_HANDLED;
172 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400173 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900174
Tejun Heoad9e2762006-05-15 20:58:12 +0900175 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900176 DPRINTK("EXIT, ret=%d\n", ret);
177 return ret;
178}
179
180/**
181 * ata_scsi_error - SCSI layer error handler callback
182 * @host: SCSI host on which error occurred
183 *
184 * Handles SCSI-layer-thrown error events.
185 *
186 * LOCKING:
187 * Inherited from SCSI layer (none, can sleep)
188 *
189 * RETURNS:
190 * Zero.
191 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400192void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900193{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400194 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoad9e2762006-05-15 20:58:12 +0900195 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
196 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900197
198 DPRINTK("ENTER\n");
199
Tejun Heoad9e2762006-05-15 20:58:12 +0900200 /* synchronize with port task */
Tejun Heoece1d632006-04-02 18:51:53 +0900201 ata_port_flush_task(ap);
202
Tejun Heoad9e2762006-05-15 20:58:12 +0900203 /* synchronize with host_set lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900204
Tejun Heoad9e2762006-05-15 20:58:12 +0900205 /* For new EH, all qcs are finished in one of three ways -
206 * normal completion, error completion, and SCSI timeout.
207 * Both cmpletions can race against SCSI timeout. When normal
208 * completion wins, the qc never reaches EH. When error
209 * completion wins, the qc has ATA_QCFLAG_FAILED set.
210 *
211 * When SCSI timeout wins, things are a bit more complex.
212 * Normal or error completion can occur after the timeout but
213 * before this point. In such cases, both types of
214 * completions are honored. A scmd is determined to have
215 * timed out iff its associated qc is active and not failed.
216 */
217 if (ap->ops->error_handler) {
218 struct scsi_cmnd *scmd, *tmp;
219 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900220
Tejun Heoe30349d2006-07-03 03:02:15 +0900221 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900222
223 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
224 struct ata_queued_cmd *qc;
225
226 for (i = 0; i < ATA_MAX_QUEUE; i++) {
227 qc = __ata_qc_from_tag(ap, i);
228 if (qc->flags & ATA_QCFLAG_ACTIVE &&
229 qc->scsicmd == scmd)
230 break;
231 }
232
233 if (i < ATA_MAX_QUEUE) {
234 /* the scmd has an associated qc */
235 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
236 /* which hasn't failed yet, timeout */
237 qc->err_mask |= AC_ERR_TIMEOUT;
238 qc->flags |= ATA_QCFLAG_FAILED;
239 nr_timedout++;
240 }
241 } else {
242 /* Normal completion occurred after
243 * SCSI timeout but before this point.
244 * Successfully complete it.
245 */
246 scmd->retries = scmd->allowed;
247 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
248 }
249 }
250
251 /* If we have timed out qcs. They belong to EH from
252 * this point but the state of the controller is
253 * unknown. Freeze the port to make sure the IRQ
254 * handler doesn't diddle with those qcs. This must
255 * be done atomically w.r.t. setting QCFLAG_FAILED.
256 */
257 if (nr_timedout)
258 __ata_port_freeze(ap);
259
Tejun Heoe30349d2006-07-03 03:02:15 +0900260 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900261 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900262 spin_unlock_wait(ap->lock);
Tejun Heoad9e2762006-05-15 20:58:12 +0900263
264 repeat:
265 /* invoke error handler */
266 if (ap->ops->error_handler) {
Tejun Heo500530f2006-07-03 16:07:27 +0900267 /* process port resume request */
268 ata_eh_handle_port_resume(ap);
269
Tejun Heof3e81b12006-05-15 20:58:21 +0900270 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900271 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900272
273 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
274 ap->eh_context.i = ap->eh_info;
275 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
276
Tejun Heob51e9e52006-06-29 01:29:30 +0900277 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
278 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof3e81b12006-05-15 20:58:21 +0900279
Tejun Heoe30349d2006-07-03 03:02:15 +0900280 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900281
Tejun Heo500530f2006-07-03 16:07:27 +0900282 /* invoke EH, skip if unloading or suspended */
283 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900284 ap->ops->error_handler(ap);
285 else
286 ata_eh_finish(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900287
Tejun Heo500530f2006-07-03 16:07:27 +0900288 /* process port suspend request */
289 ata_eh_handle_port_suspend(ap);
290
Tejun Heoad9e2762006-05-15 20:58:12 +0900291 /* Exception might have happend after ->error_handler
292 * recovered the port but before this point. Repeat
293 * EH in such case.
294 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900295 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900296
Tejun Heob51e9e52006-06-29 01:29:30 +0900297 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoad9e2762006-05-15 20:58:12 +0900298 if (--repeat_cnt) {
299 ata_port_printk(ap, KERN_INFO,
300 "EH pending after completion, "
301 "repeating EH (cnt=%d)\n", repeat_cnt);
Tejun Heoe30349d2006-07-03 03:02:15 +0900302 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900303 goto repeat;
304 }
305 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
306 "tries, giving up\n", ATA_EH_MAX_REPEAT);
307 }
308
Tejun Heof3e81b12006-05-15 20:58:21 +0900309 /* this run is complete, make sure EH info is clear */
310 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
311
Tejun Heoe30349d2006-07-03 03:02:15 +0900312 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900313 * that if exception occurs after this point but
314 * before EH completion, SCSI midlayer will
315 * re-initiate EH.
316 */
317 host->host_eh_scheduled = 0;
318
Tejun Heoe30349d2006-07-03 03:02:15 +0900319 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900320 } else {
321 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
322 ap->ops->eng_timeout(ap);
323 }
324
325 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900326 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
327
328 scsi_eh_flush_done_q(&ap->eh_done_q);
329
Tejun Heoad9e2762006-05-15 20:58:12 +0900330 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900331 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900332
Tejun Heo1cdaf532006-07-03 16:07:26 +0900333 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900334 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900335 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
336 queue_work(ata_aux_wq, &ap->hotplug_task);
337
338 if (ap->pflags & ATA_PFLAG_RECOVERED)
339 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900340
Tejun Heob51e9e52006-06-29 01:29:30 +0900341 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900342
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900343 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900344 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900345 wake_up_all(&ap->eh_wait_q);
346
Tejun Heoe30349d2006-07-03 03:02:15 +0900347 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900348
Tejun Heoece1d632006-04-02 18:51:53 +0900349 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900350}
351
352/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900353 * ata_port_wait_eh - Wait for the currently pending EH to complete
354 * @ap: Port to wait EH for
355 *
356 * Wait until the currently pending EH is complete.
357 *
358 * LOCKING:
359 * Kernel thread context (may sleep).
360 */
361void ata_port_wait_eh(struct ata_port *ap)
362{
363 unsigned long flags;
364 DEFINE_WAIT(wait);
365
366 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400367 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900368
Tejun Heob51e9e52006-06-29 01:29:30 +0900369 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900370 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400371 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900372 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400373 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900374 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900375 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900376
Jeff Garzikba6a1302006-06-22 23:46:10 -0400377 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900378
379 /* make sure SCSI EH is complete */
380 if (scsi_host_in_recovery(ap->host)) {
381 msleep(10);
382 goto retry;
383 }
384}
385
386/**
Tejun Heoece1d632006-04-02 18:51:53 +0900387 * ata_qc_timeout - Handle timeout of queued command
388 * @qc: Command that timed out
389 *
390 * Some part of the kernel (currently, only the SCSI layer)
391 * has noticed that the active command on port @ap has not
392 * completed after a specified length of time. Handle this
393 * condition by disabling DMA (if necessary) and completing
394 * transactions, with error if necessary.
395 *
396 * This also handles the case of the "lost interrupt", where
397 * for some reason (possibly hardware bug, possibly driver bug)
398 * an interrupt was not delivered to the driver, even though the
399 * transaction completed successfully.
400 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900401 * TODO: kill this function once old EH is gone.
402 *
Tejun Heoece1d632006-04-02 18:51:53 +0900403 * LOCKING:
404 * Inherited from SCSI layer (none, can sleep)
405 */
406static void ata_qc_timeout(struct ata_queued_cmd *qc)
407{
408 struct ata_port *ap = qc->ap;
Tejun Heoece1d632006-04-02 18:51:53 +0900409 u8 host_stat = 0, drv_stat;
410 unsigned long flags;
411
412 DPRINTK("ENTER\n");
413
414 ap->hsm_task_state = HSM_ST_IDLE;
415
Jeff Garzikba6a1302006-06-22 23:46:10 -0400416 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900417
418 switch (qc->tf.protocol) {
419
420 case ATA_PROT_DMA:
421 case ATA_PROT_ATAPI_DMA:
422 host_stat = ap->ops->bmdma_status(ap);
423
424 /* before we do anything else, clear DMA-Start bit */
425 ap->ops->bmdma_stop(qc);
426
427 /* fall through */
428
429 default:
430 ata_altstatus(ap);
431 drv_stat = ata_chk_status(ap);
432
433 /* ack bmdma irq events */
434 ap->ops->irq_clear(ap);
435
Tejun Heof15a1da2006-05-15 20:57:56 +0900436 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
437 "stat 0x%x host_stat 0x%x\n",
438 qc->tf.command, drv_stat, host_stat);
Tejun Heoece1d632006-04-02 18:51:53 +0900439
440 /* complete taskfile transaction */
Jeff Garzikc13b56a2006-04-02 10:34:24 -0400441 qc->err_mask |= AC_ERR_TIMEOUT;
Tejun Heoece1d632006-04-02 18:51:53 +0900442 break;
443 }
444
Jeff Garzikba6a1302006-06-22 23:46:10 -0400445 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900446
447 ata_eh_qc_complete(qc);
448
449 DPRINTK("EXIT\n");
450}
451
452/**
453 * ata_eng_timeout - Handle timeout of queued command
454 * @ap: Port on which timed-out command is active
455 *
456 * Some part of the kernel (currently, only the SCSI layer)
457 * has noticed that the active command on port @ap has not
458 * completed after a specified length of time. Handle this
459 * condition by disabling DMA (if necessary) and completing
460 * transactions, with error if necessary.
461 *
462 * This also handles the case of the "lost interrupt", where
463 * for some reason (possibly hardware bug, possibly driver bug)
464 * an interrupt was not delivered to the driver, even though the
465 * transaction completed successfully.
466 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900467 * TODO: kill this function once old EH is gone.
468 *
Tejun Heoece1d632006-04-02 18:51:53 +0900469 * LOCKING:
470 * Inherited from SCSI layer (none, can sleep)
471 */
472void ata_eng_timeout(struct ata_port *ap)
473{
474 DPRINTK("ENTER\n");
475
476 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
477
478 DPRINTK("EXIT\n");
479}
480
Tejun Heof686bcb2006-05-15 20:58:05 +0900481/**
482 * ata_qc_schedule_eh - schedule qc for error handling
483 * @qc: command to schedule error handling for
484 *
485 * Schedule error handling for @qc. EH will kick in as soon as
486 * other commands are drained.
487 *
488 * LOCKING:
489 * spin_lock_irqsave(host_set lock)
490 */
491void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
492{
493 struct ata_port *ap = qc->ap;
494
495 WARN_ON(!ap->ops->error_handler);
496
497 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heob51e9e52006-06-29 01:29:30 +0900498 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
Tejun Heof686bcb2006-05-15 20:58:05 +0900499
500 /* The following will fail if timeout has already expired.
501 * ata_scsi_error() takes care of such scmds on EH entry.
502 * Note that ATA_QCFLAG_FAILED is unconditionally set after
503 * this function completes.
504 */
505 scsi_req_abort_cmd(qc->scsicmd);
506}
507
Tejun Heo7b70fc02006-05-15 20:58:07 +0900508/**
509 * ata_port_schedule_eh - schedule error handling without a qc
510 * @ap: ATA port to schedule EH for
511 *
512 * Schedule error handling for @ap. EH will kick in as soon as
513 * all commands are drained.
514 *
515 * LOCKING:
516 * spin_lock_irqsave(host_set lock)
517 */
518void ata_port_schedule_eh(struct ata_port *ap)
519{
520 WARN_ON(!ap->ops->error_handler);
521
Tejun Heob51e9e52006-06-29 01:29:30 +0900522 ap->pflags |= ATA_PFLAG_EH_PENDING;
Tejun Heof8bbfc22006-05-19 21:07:05 +0900523 scsi_schedule_eh(ap->host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900524
525 DPRINTK("port EH scheduled\n");
526}
527
528/**
529 * ata_port_abort - abort all qc's on the port
530 * @ap: ATA port to abort qc's for
531 *
532 * Abort all active qc's of @ap and schedule EH.
533 *
534 * LOCKING:
535 * spin_lock_irqsave(host_set lock)
536 *
537 * RETURNS:
538 * Number of aborted qc's.
539 */
540int ata_port_abort(struct ata_port *ap)
541{
542 int tag, nr_aborted = 0;
543
544 WARN_ON(!ap->ops->error_handler);
545
546 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
547 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
548
549 if (qc) {
550 qc->flags |= ATA_QCFLAG_FAILED;
551 ata_qc_complete(qc);
552 nr_aborted++;
553 }
554 }
555
556 if (!nr_aborted)
557 ata_port_schedule_eh(ap);
558
559 return nr_aborted;
560}
561
Tejun Heoe3180492006-05-15 20:58:09 +0900562/**
563 * __ata_port_freeze - freeze port
564 * @ap: ATA port to freeze
565 *
566 * This function is called when HSM violation or some other
567 * condition disrupts normal operation of the port. Frozen port
568 * is not allowed to perform any operation until the port is
569 * thawed, which usually follows a successful reset.
570 *
571 * ap->ops->freeze() callback can be used for freezing the port
572 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
573 * port cannot be frozen hardware-wise, the interrupt handler
574 * must ack and clear interrupts unconditionally while the port
575 * is frozen.
576 *
577 * LOCKING:
578 * spin_lock_irqsave(host_set lock)
579 */
580static void __ata_port_freeze(struct ata_port *ap)
581{
582 WARN_ON(!ap->ops->error_handler);
583
584 if (ap->ops->freeze)
585 ap->ops->freeze(ap);
586
Tejun Heob51e9e52006-06-29 01:29:30 +0900587 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900588
589 DPRINTK("ata%u port frozen\n", ap->id);
590}
591
592/**
593 * ata_port_freeze - abort & freeze port
594 * @ap: ATA port to freeze
595 *
596 * Abort and freeze @ap.
597 *
598 * LOCKING:
599 * spin_lock_irqsave(host_set lock)
600 *
601 * RETURNS:
602 * Number of aborted commands.
603 */
604int ata_port_freeze(struct ata_port *ap)
605{
606 int nr_aborted;
607
608 WARN_ON(!ap->ops->error_handler);
609
610 nr_aborted = ata_port_abort(ap);
611 __ata_port_freeze(ap);
612
613 return nr_aborted;
614}
615
616/**
617 * ata_eh_freeze_port - EH helper to freeze port
618 * @ap: ATA port to freeze
619 *
620 * Freeze @ap.
621 *
622 * LOCKING:
623 * None.
624 */
625void ata_eh_freeze_port(struct ata_port *ap)
626{
627 unsigned long flags;
628
629 if (!ap->ops->error_handler)
630 return;
631
Jeff Garzikba6a1302006-06-22 23:46:10 -0400632 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900633 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400634 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900635}
636
637/**
638 * ata_port_thaw_port - EH helper to thaw port
639 * @ap: ATA port to thaw
640 *
641 * Thaw frozen port @ap.
642 *
643 * LOCKING:
644 * None.
645 */
646void ata_eh_thaw_port(struct ata_port *ap)
647{
648 unsigned long flags;
649
650 if (!ap->ops->error_handler)
651 return;
652
Jeff Garzikba6a1302006-06-22 23:46:10 -0400653 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900654
Tejun Heob51e9e52006-06-29 01:29:30 +0900655 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900656
657 if (ap->ops->thaw)
658 ap->ops->thaw(ap);
659
Jeff Garzikba6a1302006-06-22 23:46:10 -0400660 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900661
662 DPRINTK("ata%u port thawed\n", ap->id);
663}
664
Tejun Heoece1d632006-04-02 18:51:53 +0900665static void ata_eh_scsidone(struct scsi_cmnd *scmd)
666{
667 /* nada */
668}
669
670static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
671{
672 struct ata_port *ap = qc->ap;
673 struct scsi_cmnd *scmd = qc->scsicmd;
674 unsigned long flags;
675
Jeff Garzikba6a1302006-06-22 23:46:10 -0400676 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900677 qc->scsidone = ata_eh_scsidone;
678 __ata_qc_complete(qc);
679 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -0400680 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900681
682 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
683}
684
685/**
686 * ata_eh_qc_complete - Complete an active ATA command from EH
687 * @qc: Command to complete
688 *
689 * Indicate to the mid and upper layers that an ATA command has
690 * completed. To be used from EH.
691 */
692void ata_eh_qc_complete(struct ata_queued_cmd *qc)
693{
694 struct scsi_cmnd *scmd = qc->scsicmd;
695 scmd->retries = scmd->allowed;
696 __ata_eh_qc_complete(qc);
697}
698
699/**
700 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
701 * @qc: Command to retry
702 *
703 * Indicate to the mid and upper layers that an ATA command
704 * should be retried. To be used from EH.
705 *
706 * SCSI midlayer limits the number of retries to scmd->allowed.
707 * scmd->retries is decremented for commands which get retried
708 * due to unrelated failures (qc->err_mask is zero).
709 */
710void ata_eh_qc_retry(struct ata_queued_cmd *qc)
711{
712 struct scsi_cmnd *scmd = qc->scsicmd;
713 if (!qc->err_mask && scmd->retries)
714 scmd->retries--;
715 __ata_eh_qc_complete(qc);
716}
Tejun Heo022bdb02006-05-15 20:58:22 +0900717
718/**
Tejun Heo0ea035a2006-05-31 18:28:01 +0900719 * ata_eh_detach_dev - detach ATA device
720 * @dev: ATA device to detach
721 *
722 * Detach @dev.
723 *
724 * LOCKING:
725 * None.
726 */
727static void ata_eh_detach_dev(struct ata_device *dev)
728{
729 struct ata_port *ap = dev->ap;
730 unsigned long flags;
731
732 ata_dev_disable(dev);
733
Jeff Garzikba6a1302006-06-22 23:46:10 -0400734 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +0900735
736 dev->flags &= ~ATA_DFLAG_DETACH;
737
738 if (ata_scsi_offline_dev(dev)) {
739 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +0900740 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +0900741 }
742
Tejun Heobeb07c12006-06-24 20:30:19 +0900743 /* clear per-dev EH actions */
744 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
745 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
746
Jeff Garzikba6a1302006-06-22 23:46:10 -0400747 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +0900748}
749
750/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900751 * ata_eh_about_to_do - about to perform eh_action
752 * @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +0900753 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +0900754 * @action: action about to be performed
755 *
756 * Called just before performing EH actions to clear related bits
757 * in @ap->eh_info such that eh actions are not unnecessarily
758 * repeated.
759 *
760 * LOCKING:
761 * None.
762 */
Tejun Heo47005f22006-06-19 18:27:23 +0900763static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
764 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +0900765{
766 unsigned long flags;
767
Jeff Garzikba6a1302006-06-22 23:46:10 -0400768 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900769
Tejun Heo47005f22006-06-19 18:27:23 +0900770 ata_eh_clear_action(dev, &ap->eh_info, action);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900771
772 if (!(ap->eh_context.i.flags & ATA_EHI_QUIET))
773 ap->pflags |= ATA_PFLAG_RECOVERED;
774
Jeff Garzikba6a1302006-06-22 23:46:10 -0400775 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +0900776}
777
778/**
Tejun Heo47005f22006-06-19 18:27:23 +0900779 * ata_eh_done - EH action complete
780 * @ap: target ATA port
781 * @dev: target ATA dev for per-dev action (can be NULL)
782 * @action: action just completed
783 *
784 * Called right after performing EH actions to clear related bits
785 * in @ap->eh_context.
786 *
787 * LOCKING:
788 * None.
789 */
790static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
791 unsigned int action)
792{
793 ata_eh_clear_action(dev, &ap->eh_context.i, action);
794}
795
796/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900797 * ata_err_string - convert err_mask to descriptive string
798 * @err_mask: error mask to convert to string
799 *
800 * Convert @err_mask to descriptive string. Errors are
801 * prioritized according to severity and only the most severe
802 * error is reported.
803 *
804 * LOCKING:
805 * None.
806 *
807 * RETURNS:
808 * Descriptive string for @err_mask
809 */
810static const char * ata_err_string(unsigned int err_mask)
811{
812 if (err_mask & AC_ERR_HOST_BUS)
813 return "host bus error";
814 if (err_mask & AC_ERR_ATA_BUS)
815 return "ATA bus error";
816 if (err_mask & AC_ERR_TIMEOUT)
817 return "timeout";
818 if (err_mask & AC_ERR_HSM)
819 return "HSM violation";
820 if (err_mask & AC_ERR_SYSTEM)
821 return "internal error";
822 if (err_mask & AC_ERR_MEDIA)
823 return "media error";
824 if (err_mask & AC_ERR_INVALID)
825 return "invalid argument";
826 if (err_mask & AC_ERR_DEV)
827 return "device error";
828 return "unknown error";
829}
830
831/**
Tejun Heoe8ee8452006-05-15 21:03:46 +0900832 * ata_read_log_page - read a specific log page
833 * @dev: target device
834 * @page: page to read
835 * @buf: buffer to store read page
836 * @sectors: number of sectors to read
837 *
838 * Read log page using READ_LOG_EXT command.
839 *
840 * LOCKING:
841 * Kernel thread context (may sleep).
842 *
843 * RETURNS:
844 * 0 on success, AC_ERR_* mask otherwise.
845 */
846static unsigned int ata_read_log_page(struct ata_device *dev,
847 u8 page, void *buf, unsigned int sectors)
848{
849 struct ata_taskfile tf;
850 unsigned int err_mask;
851
852 DPRINTK("read log page - page %d\n", page);
853
854 ata_tf_init(dev, &tf);
855 tf.command = ATA_CMD_READ_LOG_EXT;
856 tf.lbal = page;
857 tf.nsect = sectors;
858 tf.hob_nsect = sectors >> 8;
859 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
860 tf.protocol = ATA_PROT_PIO;
861
862 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
863 buf, sectors * ATA_SECT_SIZE);
864
865 DPRINTK("EXIT, err_mask=%x\n", err_mask);
866 return err_mask;
867}
868
869/**
870 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
871 * @dev: Device to read log page 10h from
872 * @tag: Resulting tag of the failed command
873 * @tf: Resulting taskfile registers of the failed command
874 *
875 * Read log page 10h to obtain NCQ error details and clear error
876 * condition.
877 *
878 * LOCKING:
879 * Kernel thread context (may sleep).
880 *
881 * RETURNS:
882 * 0 on success, -errno otherwise.
883 */
884static int ata_eh_read_log_10h(struct ata_device *dev,
885 int *tag, struct ata_taskfile *tf)
886{
887 u8 *buf = dev->ap->sector_buf;
888 unsigned int err_mask;
889 u8 csum;
890 int i;
891
892 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
893 if (err_mask)
894 return -EIO;
895
896 csum = 0;
897 for (i = 0; i < ATA_SECT_SIZE; i++)
898 csum += buf[i];
899 if (csum)
900 ata_dev_printk(dev, KERN_WARNING,
901 "invalid checksum 0x%x on log page 10h\n", csum);
902
903 if (buf[0] & 0x80)
904 return -ENOENT;
905
906 *tag = buf[0] & 0x1f;
907
908 tf->command = buf[2];
909 tf->feature = buf[3];
910 tf->lbal = buf[4];
911 tf->lbam = buf[5];
912 tf->lbah = buf[6];
913 tf->device = buf[7];
914 tf->hob_lbal = buf[8];
915 tf->hob_lbam = buf[9];
916 tf->hob_lbah = buf[10];
917 tf->nsect = buf[12];
918 tf->hob_nsect = buf[13];
919
920 return 0;
921}
922
923/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900924 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
925 * @dev: device to perform REQUEST_SENSE to
926 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
927 *
928 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
929 * SENSE. This function is EH helper.
930 *
931 * LOCKING:
932 * Kernel thread context (may sleep).
933 *
934 * RETURNS:
935 * 0 on success, AC_ERR_* mask on failure
936 */
937static unsigned int atapi_eh_request_sense(struct ata_device *dev,
938 unsigned char *sense_buf)
939{
940 struct ata_port *ap = dev->ap;
941 struct ata_taskfile tf;
942 u8 cdb[ATAPI_CDB_LEN];
943
944 DPRINTK("ATAPI request sense\n");
945
946 ata_tf_init(dev, &tf);
947
948 /* FIXME: is this needed? */
949 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
950
951 /* XXX: why tf_read here? */
952 ap->ops->tf_read(ap, &tf);
953
954 /* fill these in, for the case where they are -not- overwritten */
955 sense_buf[0] = 0x70;
956 sense_buf[2] = tf.feature >> 4;
957
958 memset(cdb, 0, ATAPI_CDB_LEN);
959 cdb[0] = REQUEST_SENSE;
960 cdb[4] = SCSI_SENSE_BUFFERSIZE;
961
962 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
963 tf.command = ATA_CMD_PACKET;
964
965 /* is it pointless to prefer PIO for "safety reasons"? */
966 if (ap->flags & ATA_FLAG_PIO_DMA) {
967 tf.protocol = ATA_PROT_ATAPI_DMA;
968 tf.feature |= ATAPI_PKT_DMA;
969 } else {
970 tf.protocol = ATA_PROT_ATAPI;
971 tf.lbam = (8 * 1024) & 0xff;
972 tf.lbah = (8 * 1024) >> 8;
973 }
974
975 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
976 sense_buf, SCSI_SENSE_BUFFERSIZE);
977}
978
979/**
980 * ata_eh_analyze_serror - analyze SError for a failed port
981 * @ap: ATA port to analyze SError for
982 *
983 * Analyze SError if available and further determine cause of
984 * failure.
985 *
986 * LOCKING:
987 * None.
988 */
989static void ata_eh_analyze_serror(struct ata_port *ap)
990{
991 struct ata_eh_context *ehc = &ap->eh_context;
992 u32 serror = ehc->i.serror;
993 unsigned int err_mask = 0, action = 0;
994
995 if (serror & SERR_PERSISTENT) {
996 err_mask |= AC_ERR_ATA_BUS;
997 action |= ATA_EH_HARDRESET;
998 }
999 if (serror &
1000 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1001 err_mask |= AC_ERR_ATA_BUS;
1002 action |= ATA_EH_SOFTRESET;
1003 }
1004 if (serror & SERR_PROTOCOL) {
1005 err_mask |= AC_ERR_HSM;
1006 action |= ATA_EH_SOFTRESET;
1007 }
1008 if (serror & SERR_INTERNAL) {
1009 err_mask |= AC_ERR_SYSTEM;
1010 action |= ATA_EH_SOFTRESET;
1011 }
Tejun Heo084fe632006-05-31 18:28:03 +09001012 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1013 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001014
1015 ehc->i.err_mask |= err_mask;
1016 ehc->i.action |= action;
1017}
1018
1019/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001020 * ata_eh_analyze_ncq_error - analyze NCQ error
1021 * @ap: ATA port to analyze NCQ error for
1022 *
1023 * Read log page 10h, determine the offending qc and acquire
1024 * error status TF. For NCQ device errors, all LLDDs have to do
1025 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1026 * care of the rest.
1027 *
1028 * LOCKING:
1029 * Kernel thread context (may sleep).
1030 */
1031static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1032{
1033 struct ata_eh_context *ehc = &ap->eh_context;
1034 struct ata_device *dev = ap->device;
1035 struct ata_queued_cmd *qc;
1036 struct ata_taskfile tf;
1037 int tag, rc;
1038
1039 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001040 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001041 return;
1042
1043 /* is it NCQ device error? */
1044 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1045 return;
1046
1047 /* has LLDD analyzed already? */
1048 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1049 qc = __ata_qc_from_tag(ap, tag);
1050
1051 if (!(qc->flags & ATA_QCFLAG_FAILED))
1052 continue;
1053
1054 if (qc->err_mask)
1055 return;
1056 }
1057
1058 /* okay, this error is ours */
1059 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1060 if (rc) {
1061 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1062 "(errno=%d)\n", rc);
1063 return;
1064 }
1065
1066 if (!(ap->sactive & (1 << tag))) {
1067 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1068 "inactive tag %d\n", tag);
1069 return;
1070 }
1071
1072 /* we've got the perpetrator, condemn it */
1073 qc = __ata_qc_from_tag(ap, tag);
1074 memcpy(&qc->result_tf, &tf, sizeof(tf));
1075 qc->err_mask |= AC_ERR_DEV;
1076 ehc->i.err_mask &= ~AC_ERR_DEV;
1077}
1078
1079/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001080 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1081 * @qc: qc to analyze
1082 * @tf: Taskfile registers to analyze
1083 *
1084 * Analyze taskfile of @qc and further determine cause of
1085 * failure. This function also requests ATAPI sense data if
1086 * avaliable.
1087 *
1088 * LOCKING:
1089 * Kernel thread context (may sleep).
1090 *
1091 * RETURNS:
1092 * Determined recovery action
1093 */
1094static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1095 const struct ata_taskfile *tf)
1096{
1097 unsigned int tmp, action = 0;
1098 u8 stat = tf->command, err = tf->feature;
1099
1100 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1101 qc->err_mask |= AC_ERR_HSM;
1102 return ATA_EH_SOFTRESET;
1103 }
1104
1105 if (!(qc->err_mask & AC_ERR_DEV))
1106 return 0;
1107
1108 switch (qc->dev->class) {
1109 case ATA_DEV_ATA:
1110 if (err & ATA_ICRC)
1111 qc->err_mask |= AC_ERR_ATA_BUS;
1112 if (err & ATA_UNC)
1113 qc->err_mask |= AC_ERR_MEDIA;
1114 if (err & ATA_IDNF)
1115 qc->err_mask |= AC_ERR_INVALID;
1116 break;
1117
1118 case ATA_DEV_ATAPI:
1119 tmp = atapi_eh_request_sense(qc->dev,
1120 qc->scsicmd->sense_buffer);
1121 if (!tmp) {
1122 /* ATA_QCFLAG_SENSE_VALID is used to tell
1123 * atapi_qc_complete() that sense data is
1124 * already valid.
1125 *
1126 * TODO: interpret sense data and set
1127 * appropriate err_mask.
1128 */
1129 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1130 } else
1131 qc->err_mask |= tmp;
1132 }
1133
1134 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1135 action |= ATA_EH_SOFTRESET;
1136
1137 return action;
1138}
1139
1140static int ata_eh_categorize_ering_entry(struct ata_ering_entry *ent)
1141{
1142 if (ent->err_mask & (AC_ERR_ATA_BUS | AC_ERR_TIMEOUT))
1143 return 1;
1144
1145 if (ent->is_io) {
1146 if (ent->err_mask & AC_ERR_HSM)
1147 return 1;
1148 if ((ent->err_mask &
1149 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1150 return 2;
1151 }
1152
1153 return 0;
1154}
1155
1156struct speed_down_needed_arg {
1157 u64 since;
1158 int nr_errors[3];
1159};
1160
1161static int speed_down_needed_cb(struct ata_ering_entry *ent, void *void_arg)
1162{
1163 struct speed_down_needed_arg *arg = void_arg;
1164
1165 if (ent->timestamp < arg->since)
1166 return -1;
1167
1168 arg->nr_errors[ata_eh_categorize_ering_entry(ent)]++;
1169 return 0;
1170}
1171
1172/**
1173 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1174 * @dev: Device of interest
1175 *
1176 * This function examines error ring of @dev and determines
1177 * whether speed down is necessary. Speed down is necessary if
1178 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1179 * errors during last 15 minutes.
1180 *
1181 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1182 * violation for known supported commands.
1183 *
1184 * Cat-2 errors are unclassified DEV error for known supported
1185 * command.
1186 *
1187 * LOCKING:
1188 * Inherited from caller.
1189 *
1190 * RETURNS:
1191 * 1 if speed down is necessary, 0 otherwise
1192 */
1193static int ata_eh_speed_down_needed(struct ata_device *dev)
1194{
1195 const u64 interval = 15LLU * 60 * HZ;
1196 static const int err_limits[3] = { -1, 3, 10 };
1197 struct speed_down_needed_arg arg;
1198 struct ata_ering_entry *ent;
1199 int err_cat;
1200 u64 j64;
1201
1202 ent = ata_ering_top(&dev->ering);
1203 if (!ent)
1204 return 0;
1205
1206 err_cat = ata_eh_categorize_ering_entry(ent);
1207 if (err_cat == 0)
1208 return 0;
1209
1210 memset(&arg, 0, sizeof(arg));
1211
1212 j64 = get_jiffies_64();
1213 if (j64 >= interval)
1214 arg.since = j64 - interval;
1215 else
1216 arg.since = 0;
1217
1218 ata_ering_map(&dev->ering, speed_down_needed_cb, &arg);
1219
1220 return arg.nr_errors[err_cat] > err_limits[err_cat];
1221}
1222
1223/**
1224 * ata_eh_speed_down - record error and speed down if necessary
1225 * @dev: Failed device
1226 * @is_io: Did the device fail during normal IO?
1227 * @err_mask: err_mask of the error
1228 *
1229 * Record error and examine error history to determine whether
1230 * adjusting transmission speed is necessary. It also sets
1231 * transmission limits appropriately if such adjustment is
1232 * necessary.
1233 *
1234 * LOCKING:
1235 * Kernel thread context (may sleep).
1236 *
1237 * RETURNS:
1238 * 0 on success, -errno otherwise
1239 */
1240static int ata_eh_speed_down(struct ata_device *dev, int is_io,
1241 unsigned int err_mask)
1242{
1243 if (!err_mask)
1244 return 0;
1245
1246 /* record error and determine whether speed down is necessary */
1247 ata_ering_record(&dev->ering, is_io, err_mask);
1248
1249 if (!ata_eh_speed_down_needed(dev))
1250 return 0;
1251
1252 /* speed down SATA link speed if possible */
1253 if (sata_down_spd_limit(dev->ap) == 0)
1254 return ATA_EH_HARDRESET;
1255
1256 /* lower transfer mode */
1257 if (ata_down_xfermask_limit(dev, 0) == 0)
1258 return ATA_EH_SOFTRESET;
1259
1260 ata_dev_printk(dev, KERN_ERR,
1261 "speed down requested but no transfer mode left\n");
1262 return 0;
1263}
1264
1265/**
1266 * ata_eh_autopsy - analyze error and determine recovery action
1267 * @ap: ATA port to perform autopsy on
1268 *
1269 * Analyze why @ap failed and determine which recovery action is
1270 * needed. This function also sets more detailed AC_ERR_* values
1271 * and fills sense data for ATAPI CHECK SENSE.
1272 *
1273 * LOCKING:
1274 * Kernel thread context (may sleep).
1275 */
1276static void ata_eh_autopsy(struct ata_port *ap)
1277{
1278 struct ata_eh_context *ehc = &ap->eh_context;
1279 unsigned int action = ehc->i.action;
1280 struct ata_device *failed_dev = NULL;
1281 unsigned int all_err_mask = 0;
1282 int tag, is_io = 0;
1283 u32 serror;
1284 int rc;
1285
1286 DPRINTK("ENTER\n");
1287
Tejun Heo1cdaf532006-07-03 16:07:26 +09001288 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1289 return;
1290
Tejun Heo022bdb02006-05-15 20:58:22 +09001291 /* obtain and analyze SError */
1292 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1293 if (rc == 0) {
1294 ehc->i.serror |= serror;
1295 ata_eh_analyze_serror(ap);
1296 } else if (rc != -EOPNOTSUPP)
1297 action |= ATA_EH_HARDRESET;
1298
Tejun Heoe8ee8452006-05-15 21:03:46 +09001299 /* analyze NCQ failure */
1300 ata_eh_analyze_ncq_error(ap);
1301
Tejun Heo022bdb02006-05-15 20:58:22 +09001302 /* any real error trumps AC_ERR_OTHER */
1303 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1304 ehc->i.err_mask &= ~AC_ERR_OTHER;
1305
1306 all_err_mask |= ehc->i.err_mask;
1307
1308 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1309 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1310
1311 if (!(qc->flags & ATA_QCFLAG_FAILED))
1312 continue;
1313
1314 /* inherit upper level err_mask */
1315 qc->err_mask |= ehc->i.err_mask;
1316
Tejun Heo022bdb02006-05-15 20:58:22 +09001317 /* analyze TF */
1318 action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1319
1320 /* DEV errors are probably spurious in case of ATA_BUS error */
1321 if (qc->err_mask & AC_ERR_ATA_BUS)
1322 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1323 AC_ERR_INVALID);
1324
1325 /* any real error trumps unknown error */
1326 if (qc->err_mask & ~AC_ERR_OTHER)
1327 qc->err_mask &= ~AC_ERR_OTHER;
1328
1329 /* SENSE_VALID trumps dev/unknown error and revalidation */
1330 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1331 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1332 action &= ~ATA_EH_REVALIDATE;
1333 }
1334
1335 /* accumulate error info */
1336 failed_dev = qc->dev;
1337 all_err_mask |= qc->err_mask;
1338 if (qc->flags & ATA_QCFLAG_IO)
1339 is_io = 1;
1340 }
1341
Tejun Heoa20f33f2006-05-16 12:58:24 +09001342 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001343 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001344 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1345 action |= ATA_EH_SOFTRESET;
1346 else if (all_err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001347 action |= ATA_EH_REVALIDATE;
1348
Tejun Heo47005f22006-06-19 18:27:23 +09001349 /* if we have offending qcs and the associated failed device */
1350 if (failed_dev) {
1351 /* speed down */
1352 action |= ata_eh_speed_down(failed_dev, is_io, all_err_mask);
1353
1354 /* perform per-dev EH action only on the offending device */
1355 ehc->i.dev_action[failed_dev->devno] |=
1356 action & ATA_EH_PERDEV_MASK;
1357 action &= ~ATA_EH_PERDEV_MASK;
1358 }
1359
Tejun Heoa20f33f2006-05-16 12:58:24 +09001360 /* record autopsy result */
Tejun Heo022bdb02006-05-15 20:58:22 +09001361 ehc->i.dev = failed_dev;
Tejun Heo0662c582006-07-03 02:54:58 +09001362 ehc->i.action |= action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001363
1364 DPRINTK("EXIT\n");
1365}
1366
1367/**
1368 * ata_eh_report - report error handling to user
1369 * @ap: ATA port EH is going on
1370 *
1371 * Report EH to user.
1372 *
1373 * LOCKING:
1374 * None.
1375 */
1376static void ata_eh_report(struct ata_port *ap)
1377{
1378 struct ata_eh_context *ehc = &ap->eh_context;
1379 const char *frozen, *desc;
1380 int tag, nr_failed = 0;
1381
1382 desc = NULL;
1383 if (ehc->i.desc[0] != '\0')
1384 desc = ehc->i.desc;
1385
1386 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1387 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1388
1389 if (!(qc->flags & ATA_QCFLAG_FAILED))
1390 continue;
1391 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1392 continue;
1393
1394 nr_failed++;
1395 }
1396
1397 if (!nr_failed && !ehc->i.err_mask)
1398 return;
1399
1400 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001401 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001402 frozen = " frozen";
1403
1404 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001405 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1406 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1407 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1408 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001409 if (desc)
1410 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1411 } else {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001412 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1413 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1414 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1415 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001416 if (desc)
1417 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1418 }
1419
1420 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1421 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1422
1423 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1424 continue;
1425
1426 ata_dev_printk(qc->dev, KERN_ERR, "tag %d cmd 0x%x "
1427 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1428 qc->tag, qc->tf.command, qc->err_mask,
1429 qc->result_tf.command, qc->result_tf.feature,
1430 ata_err_string(qc->err_mask));
1431 }
1432}
1433
Tejun Heod87fa382006-05-31 18:28:24 +09001434static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1435 unsigned int *classes)
1436{
1437 int i, rc;
1438
1439 for (i = 0; i < ATA_MAX_DEVICES; i++)
1440 classes[i] = ATA_DEV_UNKNOWN;
1441
1442 rc = reset(ap, classes);
1443 if (rc)
1444 return rc;
1445
1446 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1447 * is complete and convert all ATA_DEV_UNKNOWN to
1448 * ATA_DEV_NONE.
1449 */
1450 for (i = 0; i < ATA_MAX_DEVICES; i++)
1451 if (classes[i] != ATA_DEV_UNKNOWN)
1452 break;
1453
1454 if (i < ATA_MAX_DEVICES)
1455 for (i = 0; i < ATA_MAX_DEVICES; i++)
1456 if (classes[i] == ATA_DEV_UNKNOWN)
1457 classes[i] = ATA_DEV_NONE;
1458
1459 return 0;
1460}
1461
Tejun Heo664faf02006-05-31 18:27:50 +09001462static int ata_eh_followup_srst_needed(int rc, int classify,
1463 const unsigned int *classes)
1464{
1465 if (rc == -EAGAIN)
1466 return 1;
1467 if (rc != 0)
1468 return 0;
1469 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1470 return 1;
1471 return 0;
1472}
1473
1474static int ata_eh_reset(struct ata_port *ap, int classify,
Tejun Heof5914a42006-05-31 18:27:48 +09001475 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09001476 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1477{
1478 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09001479 unsigned int *classes = ehc->classes;
Tejun Heo022bdb02006-05-15 20:58:22 +09001480 int tries = ATA_EH_RESET_TRIES;
Tejun Heo1cdaf532006-07-03 16:07:26 +09001481 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heof5914a42006-05-31 18:27:48 +09001482 unsigned int action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001483 ata_reset_fn_t reset;
Tejun Heo664faf02006-05-31 18:27:50 +09001484 int i, did_followup_srst, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09001485
Tejun Heof5914a42006-05-31 18:27:48 +09001486 /* Determine which reset to use and record in ehc->i.action.
1487 * prereset() may examine and modify it.
1488 */
1489 action = ehc->i.action;
1490 ehc->i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo022bdb02006-05-15 20:58:22 +09001491 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
Tejun Heof5914a42006-05-31 18:27:48 +09001492 !(action & ATA_EH_HARDRESET))))
1493 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001494 else
Tejun Heof5914a42006-05-31 18:27:48 +09001495 ehc->i.action |= ATA_EH_HARDRESET;
1496
1497 if (prereset) {
1498 rc = prereset(ap);
1499 if (rc) {
1500 ata_port_printk(ap, KERN_ERR,
1501 "prereset failed (errno=%d)\n", rc);
1502 return rc;
1503 }
1504 }
1505
1506 /* prereset() might have modified ehc->i.action */
1507 if (ehc->i.action & ATA_EH_HARDRESET)
Tejun Heo022bdb02006-05-15 20:58:22 +09001508 reset = hardreset;
Tejun Heof5914a42006-05-31 18:27:48 +09001509 else if (ehc->i.action & ATA_EH_SOFTRESET)
1510 reset = softreset;
1511 else {
1512 /* prereset told us not to reset, bang classes and return */
1513 for (i = 0; i < ATA_MAX_DEVICES; i++)
1514 classes[i] = ATA_DEV_NONE;
1515 return 0;
1516 }
1517
1518 /* did prereset() screw up? if so, fix up to avoid oopsing */
1519 if (!reset) {
1520 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1521 "invalid reset type\n");
1522 if (softreset)
1523 reset = softreset;
1524 else
1525 reset = hardreset;
1526 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001527
1528 retry:
Tejun Heo3e706392006-05-31 18:28:11 +09001529 /* shut up during boot probing */
1530 if (verbose)
1531 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1532 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09001533
1534 /* reset */
Tejun Heo47005f22006-06-19 18:27:23 +09001535 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
Tejun Heo022bdb02006-05-15 20:58:22 +09001536 ehc->i.flags |= ATA_EHI_DID_RESET;
1537
1538 rc = ata_do_reset(ap, reset, classes);
1539
Tejun Heo664faf02006-05-31 18:27:50 +09001540 did_followup_srst = 0;
1541 if (reset == hardreset &&
1542 ata_eh_followup_srst_needed(rc, classify, classes)) {
1543 /* okay, let's do follow-up softreset */
1544 did_followup_srst = 1;
1545 reset = softreset;
1546
1547 if (!reset) {
1548 ata_port_printk(ap, KERN_ERR,
1549 "follow-up softreset required "
1550 "but no softreset avaliable\n");
1551 return -EINVAL;
1552 }
1553
Tejun Heo47005f22006-06-19 18:27:23 +09001554 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
Tejun Heo664faf02006-05-31 18:27:50 +09001555 rc = ata_do_reset(ap, reset, classes);
1556
1557 if (rc == 0 && classify &&
1558 classes[0] == ATA_DEV_UNKNOWN) {
1559 ata_port_printk(ap, KERN_ERR,
1560 "classification failed\n");
1561 return -EINVAL;
1562 }
1563 }
1564
Tejun Heo022bdb02006-05-15 20:58:22 +09001565 if (rc && --tries) {
Tejun Heo664faf02006-05-31 18:27:50 +09001566 const char *type;
1567
1568 if (reset == softreset) {
1569 if (did_followup_srst)
1570 type = "follow-up soft";
1571 else
1572 type = "soft";
1573 } else
1574 type = "hard";
1575
Tejun Heo022bdb02006-05-15 20:58:22 +09001576 ata_port_printk(ap, KERN_WARNING,
Tejun Heo664faf02006-05-31 18:27:50 +09001577 "%sreset failed, retrying in 5 secs\n", type);
Tejun Heo022bdb02006-05-15 20:58:22 +09001578 ssleep(5);
1579
1580 if (reset == hardreset)
1581 sata_down_spd_limit(ap);
1582 if (hardreset)
1583 reset = hardreset;
1584 goto retry;
1585 }
1586
1587 if (rc == 0) {
Tejun Heo20952b62006-05-31 18:27:23 +09001588 /* After the reset, the device state is PIO 0 and the
1589 * controller state is undefined. Record the mode.
1590 */
1591 for (i = 0; i < ATA_MAX_DEVICES; i++)
1592 ap->device[i].pio_mode = XFER_PIO_0;
1593
Tejun Heo022bdb02006-05-15 20:58:22 +09001594 if (postreset)
1595 postreset(ap, classes);
1596
1597 /* reset successful, schedule revalidation */
Tejun Heo47005f22006-06-19 18:27:23 +09001598 ata_eh_done(ap, NULL, ATA_EH_RESET_MASK);
Tejun Heo022bdb02006-05-15 20:58:22 +09001599 ehc->i.action |= ATA_EH_REVALIDATE;
1600 }
1601
1602 return rc;
1603}
1604
Tejun Heo084fe632006-05-31 18:28:03 +09001605static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1606 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001607{
1608 struct ata_eh_context *ehc = &ap->eh_context;
1609 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09001610 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09001611 int i, rc = 0;
1612
1613 DPRINTK("ENTER\n");
1614
1615 for (i = 0; i < ATA_MAX_DEVICES; i++) {
Tejun Heo47005f22006-06-19 18:27:23 +09001616 unsigned int action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001617
Tejun Heo47005f22006-06-19 18:27:23 +09001618 dev = &ap->device[i];
Tejun Heo64f65ca2006-06-24 20:30:18 +09001619 action = ata_eh_dev_action(dev);
Tejun Heo47005f22006-06-19 18:27:23 +09001620
Tejun Heo02670bf2006-07-03 16:07:26 +09001621 if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001622 if (ata_port_offline(ap)) {
1623 rc = -EIO;
1624 break;
1625 }
1626
Tejun Heo47005f22006-06-19 18:27:23 +09001627 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
Tejun Heo022bdb02006-05-15 20:58:22 +09001628 rc = ata_dev_revalidate(dev,
1629 ehc->i.flags & ATA_EHI_DID_RESET);
1630 if (rc)
1631 break;
1632
Tejun Heo47005f22006-06-19 18:27:23 +09001633 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1634
zhao, forrest3057ac32006-06-12 12:01:34 +08001635 /* schedule the scsi_rescan_device() here */
1636 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09001637 } else if (dev->class == ATA_DEV_UNKNOWN &&
1638 ehc->tries[dev->devno] &&
1639 ata_class_enabled(ehc->classes[dev->devno])) {
1640 dev->class = ehc->classes[dev->devno];
1641
1642 rc = ata_dev_read_id(dev, &dev->class, 1, dev->id);
1643 if (rc == 0)
1644 rc = ata_dev_configure(dev, 1);
1645
1646 if (rc) {
1647 dev->class = ATA_DEV_UNKNOWN;
1648 break;
1649 }
1650
Jeff Garzikba6a1302006-06-22 23:46:10 -04001651 spin_lock_irqsave(ap->lock, flags);
Tejun Heob51e9e52006-06-29 01:29:30 +09001652 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Jeff Garzikba6a1302006-06-22 23:46:10 -04001653 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001654 }
1655 }
1656
Tejun Heo47005f22006-06-19 18:27:23 +09001657 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09001658 *r_failed_dev = dev;
1659
1660 DPRINTK("EXIT\n");
1661 return rc;
1662}
1663
Tejun Heo02670bf2006-07-03 16:07:26 +09001664/**
1665 * ata_eh_suspend - handle suspend EH action
1666 * @ap: target host port
1667 * @r_failed_dev: result parameter to indicate failing device
1668 *
1669 * Handle suspend EH action. Disk devices are spinned down and
1670 * other types of devices are just marked suspended. Once
1671 * suspended, no EH action to the device is allowed until it is
1672 * resumed.
1673 *
1674 * LOCKING:
1675 * Kernel thread context (may sleep).
1676 *
1677 * RETURNS:
1678 * 0 on success, -errno otherwise
1679 */
1680static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev)
1681{
1682 struct ata_device *dev;
1683 int i, rc = 0;
1684
1685 DPRINTK("ENTER\n");
1686
1687 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1688 unsigned long flags;
1689 unsigned int action, err_mask;
1690
1691 dev = &ap->device[i];
1692 action = ata_eh_dev_action(dev);
1693
1694 if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND))
1695 continue;
1696
1697 WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED);
1698
1699 ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND);
1700
1701 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1702 /* flush cache */
1703 rc = ata_flush_cache(dev);
1704 if (rc)
1705 break;
1706
1707 /* spin down */
1708 err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1);
1709 if (err_mask) {
1710 ata_dev_printk(dev, KERN_ERR, "failed to "
1711 "spin down (err_mask=0x%x)\n",
1712 err_mask);
1713 rc = -EIO;
1714 break;
1715 }
1716 }
1717
1718 spin_lock_irqsave(ap->lock, flags);
1719 dev->flags |= ATA_DFLAG_SUSPENDED;
1720 spin_unlock_irqrestore(ap->lock, flags);
1721
1722 ata_eh_done(ap, dev, ATA_EH_SUSPEND);
1723 }
1724
1725 if (rc)
1726 *r_failed_dev = dev;
1727
1728 DPRINTK("EXIT\n");
1729 return 0;
1730}
1731
1732/**
1733 * ata_eh_prep_resume - prep for resume EH action
1734 * @ap: target host port
1735 *
1736 * Clear SUSPENDED in preparation for scheduled resume actions.
1737 * This allows other parts of EH to access the devices being
1738 * resumed.
1739 *
1740 * LOCKING:
1741 * Kernel thread context (may sleep).
1742 */
1743static void ata_eh_prep_resume(struct ata_port *ap)
1744{
1745 struct ata_device *dev;
1746 unsigned long flags;
1747 int i;
1748
1749 DPRINTK("ENTER\n");
1750
1751 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1752 unsigned int action;
1753
1754 dev = &ap->device[i];
1755 action = ata_eh_dev_action(dev);
1756
1757 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1758 continue;
1759
1760 spin_lock_irqsave(ap->lock, flags);
1761 dev->flags &= ~ATA_DFLAG_SUSPENDED;
1762 spin_unlock_irqrestore(ap->lock, flags);
1763 }
1764
1765 DPRINTK("EXIT\n");
1766}
1767
1768/**
1769 * ata_eh_resume - handle resume EH action
1770 * @ap: target host port
1771 * @r_failed_dev: result parameter to indicate failing device
1772 *
1773 * Handle resume EH action. Target devices are already reset and
1774 * revalidated. Spinning up is the only operation left.
1775 *
1776 * LOCKING:
1777 * Kernel thread context (may sleep).
1778 *
1779 * RETURNS:
1780 * 0 on success, -errno otherwise
1781 */
1782static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev)
1783{
1784 struct ata_device *dev;
1785 int i, rc = 0;
1786
1787 DPRINTK("ENTER\n");
1788
1789 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1790 unsigned int action, err_mask;
1791
1792 dev = &ap->device[i];
1793 action = ata_eh_dev_action(dev);
1794
1795 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1796 continue;
1797
1798 ata_eh_about_to_do(ap, dev, ATA_EH_RESUME);
1799
1800 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1801 err_mask = ata_do_simple_cmd(dev,
1802 ATA_CMD_IDLEIMMEDIATE);
1803 if (err_mask) {
1804 ata_dev_printk(dev, KERN_ERR, "failed to "
1805 "spin up (err_mask=0x%x)\n",
1806 err_mask);
1807 rc = -EIO;
1808 break;
1809 }
1810 }
1811
1812 ata_eh_done(ap, dev, ATA_EH_RESUME);
1813 }
1814
1815 if (rc)
1816 *r_failed_dev = dev;
1817
1818 DPRINTK("EXIT\n");
1819 return 0;
1820}
1821
Tejun Heo022bdb02006-05-15 20:58:22 +09001822static int ata_port_nr_enabled(struct ata_port *ap)
1823{
1824 int i, cnt = 0;
1825
1826 for (i = 0; i < ATA_MAX_DEVICES; i++)
1827 if (ata_dev_enabled(&ap->device[i]))
1828 cnt++;
1829 return cnt;
1830}
1831
Tejun Heo084fe632006-05-31 18:28:03 +09001832static int ata_port_nr_vacant(struct ata_port *ap)
1833{
1834 int i, cnt = 0;
1835
1836 for (i = 0; i < ATA_MAX_DEVICES; i++)
1837 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1838 cnt++;
1839 return cnt;
1840}
1841
1842static int ata_eh_skip_recovery(struct ata_port *ap)
1843{
1844 struct ata_eh_context *ehc = &ap->eh_context;
1845 int i;
1846
Tejun Heo02670bf2006-07-03 16:07:26 +09001847 /* skip if all possible devices are suspended */
1848 for (i = 0; i < ata_port_max_devices(ap); i++) {
1849 struct ata_device *dev = &ap->device[i];
1850
1851 if (ata_dev_absent(dev) || ata_dev_ready(dev))
1852 break;
1853 }
1854
1855 if (i == ata_port_max_devices(ap))
1856 return 1;
1857
1858 /* always thaw frozen port and recover failed devices */
Tejun Heob51e9e52006-06-29 01:29:30 +09001859 if (ap->pflags & ATA_PFLAG_FROZEN || ata_port_nr_enabled(ap))
Tejun Heo084fe632006-05-31 18:28:03 +09001860 return 0;
1861
1862 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1863 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1864 struct ata_device *dev = &ap->device[i];
1865
1866 if (dev->class == ATA_DEV_UNKNOWN &&
1867 ehc->classes[dev->devno] != ATA_DEV_NONE)
1868 return 0;
1869 }
1870
1871 return 1;
1872}
1873
Tejun Heo022bdb02006-05-15 20:58:22 +09001874/**
1875 * ata_eh_recover - recover host port after error
1876 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09001877 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001878 * @softreset: softreset method (can be NULL)
1879 * @hardreset: hardreset method (can be NULL)
1880 * @postreset: postreset method (can be NULL)
1881 *
1882 * This is the alpha and omega, eum and yang, heart and soul of
1883 * libata exception handling. On entry, actions required to
Tejun Heo084fe632006-05-31 18:28:03 +09001884 * recover the port and hotplug requests are recorded in
1885 * eh_context. This function executes all the operations with
1886 * appropriate retrials and fallbacks to resurrect failed
1887 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09001888 *
1889 * LOCKING:
1890 * Kernel thread context (may sleep).
1891 *
1892 * RETURNS:
1893 * 0 on success, -errno on failure.
1894 */
Tejun Heof5914a42006-05-31 18:27:48 +09001895static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
1896 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09001897 ata_postreset_fn_t postreset)
1898{
1899 struct ata_eh_context *ehc = &ap->eh_context;
1900 struct ata_device *dev;
1901 int down_xfermask, i, rc;
1902
1903 DPRINTK("ENTER\n");
1904
1905 /* prep for recovery */
1906 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1907 dev = &ap->device[i];
1908
1909 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo084fe632006-05-31 18:28:03 +09001910
1911 /* process hotplug request */
1912 if (dev->flags & ATA_DFLAG_DETACH)
1913 ata_eh_detach_dev(dev);
1914
1915 if (!ata_dev_enabled(dev) &&
1916 ((ehc->i.probe_mask & (1 << dev->devno)) &&
1917 !(ehc->did_probe_mask & (1 << dev->devno)))) {
1918 ata_eh_detach_dev(dev);
1919 ata_dev_init(dev);
1920 ehc->did_probe_mask |= (1 << dev->devno);
1921 ehc->i.action |= ATA_EH_SOFTRESET;
1922 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001923 }
1924
1925 retry:
1926 down_xfermask = 0;
1927 rc = 0;
1928
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09001929 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09001930 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09001931 goto out;
1932
Tejun Heo02670bf2006-07-03 16:07:26 +09001933 /* prep for resume */
1934 ata_eh_prep_resume(ap);
1935
Tejun Heo022bdb02006-05-15 20:58:22 +09001936 /* skip EH if possible. */
Tejun Heo084fe632006-05-31 18:28:03 +09001937 if (ata_eh_skip_recovery(ap))
Tejun Heo022bdb02006-05-15 20:58:22 +09001938 ehc->i.action = 0;
1939
Tejun Heo084fe632006-05-31 18:28:03 +09001940 for (i = 0; i < ATA_MAX_DEVICES; i++)
1941 ehc->classes[i] = ATA_DEV_UNKNOWN;
1942
Tejun Heo022bdb02006-05-15 20:58:22 +09001943 /* reset */
1944 if (ehc->i.action & ATA_EH_RESET_MASK) {
1945 ata_eh_freeze_port(ap);
1946
Tejun Heo084fe632006-05-31 18:28:03 +09001947 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
1948 softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09001949 if (rc) {
1950 ata_port_printk(ap, KERN_ERR,
1951 "reset failed, giving up\n");
1952 goto out;
1953 }
1954
1955 ata_eh_thaw_port(ap);
1956 }
1957
Tejun Heo084fe632006-05-31 18:28:03 +09001958 /* revalidate existing devices and attach new ones */
1959 rc = ata_eh_revalidate_and_attach(ap, &dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001960 if (rc)
1961 goto dev_fail;
1962
Tejun Heo02670bf2006-07-03 16:07:26 +09001963 /* resume devices */
1964 rc = ata_eh_resume(ap, &dev);
1965 if (rc)
1966 goto dev_fail;
1967
Tejun Heo022bdb02006-05-15 20:58:22 +09001968 /* configure transfer mode if the port has been reset */
1969 if (ehc->i.flags & ATA_EHI_DID_RESET) {
1970 rc = ata_set_mode(ap, &dev);
1971 if (rc) {
1972 down_xfermask = 1;
1973 goto dev_fail;
1974 }
1975 }
1976
Tejun Heo02670bf2006-07-03 16:07:26 +09001977 /* suspend devices */
1978 rc = ata_eh_suspend(ap, &dev);
1979 if (rc)
1980 goto dev_fail;
1981
Tejun Heo022bdb02006-05-15 20:58:22 +09001982 goto out;
1983
1984 dev_fail:
1985 switch (rc) {
1986 case -ENODEV:
Tejun Heo084fe632006-05-31 18:28:03 +09001987 /* device missing, schedule probing */
1988 ehc->i.probe_mask |= (1 << dev->devno);
Tejun Heo022bdb02006-05-15 20:58:22 +09001989 case -EINVAL:
1990 ehc->tries[dev->devno] = 0;
1991 break;
1992 case -EIO:
1993 sata_down_spd_limit(ap);
1994 default:
1995 ehc->tries[dev->devno]--;
1996 if (down_xfermask &&
1997 ata_down_xfermask_limit(dev, ehc->tries[dev->devno] == 1))
1998 ehc->tries[dev->devno] = 0;
1999 }
2000
Tejun Heo084fe632006-05-31 18:28:03 +09002001 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2002 /* disable device if it has used up all its chances */
Tejun Heo022bdb02006-05-15 20:58:22 +09002003 ata_dev_disable(dev);
2004
Tejun Heo084fe632006-05-31 18:28:03 +09002005 /* detach if offline */
2006 if (ata_port_offline(ap))
2007 ata_eh_detach_dev(dev);
2008
2009 /* probe if requested */
2010 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2011 !(ehc->did_probe_mask & (1 << dev->devno))) {
2012 ata_eh_detach_dev(dev);
2013 ata_dev_init(dev);
2014
2015 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2016 ehc->did_probe_mask |= (1 << dev->devno);
2017 ehc->i.action |= ATA_EH_SOFTRESET;
2018 }
2019 } else {
2020 /* soft didn't work? be haaaaard */
2021 if (ehc->i.flags & ATA_EHI_DID_RESET)
2022 ehc->i.action |= ATA_EH_HARDRESET;
2023 else
2024 ehc->i.action |= ATA_EH_SOFTRESET;
2025 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002026
2027 if (ata_port_nr_enabled(ap)) {
2028 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2029 "devices, retrying in 5 secs\n");
2030 ssleep(5);
2031 } else {
2032 /* no device left, repeat fast */
2033 msleep(500);
2034 }
2035
2036 goto retry;
2037
2038 out:
2039 if (rc) {
2040 for (i = 0; i < ATA_MAX_DEVICES; i++)
2041 ata_dev_disable(&ap->device[i]);
2042 }
2043
2044 DPRINTK("EXIT, rc=%d\n", rc);
2045 return rc;
2046}
2047
2048/**
2049 * ata_eh_finish - finish up EH
2050 * @ap: host port to finish EH for
2051 *
2052 * Recovery is complete. Clean up EH states and retry or finish
2053 * failed qcs.
2054 *
2055 * LOCKING:
2056 * None.
2057 */
2058static void ata_eh_finish(struct ata_port *ap)
2059{
2060 int tag;
2061
2062 /* retry or finish qcs */
2063 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2064 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2065
2066 if (!(qc->flags & ATA_QCFLAG_FAILED))
2067 continue;
2068
2069 if (qc->err_mask) {
2070 /* FIXME: Once EH migration is complete,
2071 * generate sense data in this function,
2072 * considering both err_mask and tf.
2073 */
2074 if (qc->err_mask & AC_ERR_INVALID)
2075 ata_eh_qc_complete(qc);
2076 else
2077 ata_eh_qc_retry(qc);
2078 } else {
2079 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2080 ata_eh_qc_complete(qc);
2081 } else {
2082 /* feed zero TF to sense generation */
2083 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2084 ata_eh_qc_retry(qc);
2085 }
2086 }
2087 }
2088}
2089
2090/**
2091 * ata_do_eh - do standard error handling
2092 * @ap: host port to handle error for
Tejun Heof5914a42006-05-31 18:27:48 +09002093 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002094 * @softreset: softreset method (can be NULL)
2095 * @hardreset: hardreset method (can be NULL)
2096 * @postreset: postreset method (can be NULL)
2097 *
2098 * Perform standard error handling sequence.
2099 *
2100 * LOCKING:
2101 * Kernel thread context (may sleep).
2102 */
Tejun Heof5914a42006-05-31 18:27:48 +09002103void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2104 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2105 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002106{
Tejun Heo1cdaf532006-07-03 16:07:26 +09002107 ata_eh_autopsy(ap);
2108 ata_eh_report(ap);
Tejun Heof5914a42006-05-31 18:27:48 +09002109 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002110 ata_eh_finish(ap);
2111}
Tejun Heo500530f2006-07-03 16:07:27 +09002112
2113/**
2114 * ata_eh_handle_port_suspend - perform port suspend operation
2115 * @ap: port to suspend
2116 *
2117 * Suspend @ap.
2118 *
2119 * LOCKING:
2120 * Kernel thread context (may sleep).
2121 */
2122static void ata_eh_handle_port_suspend(struct ata_port *ap)
2123{
2124 unsigned long flags;
2125 int rc = 0;
2126
2127 /* are we suspending? */
2128 spin_lock_irqsave(ap->lock, flags);
2129 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2130 ap->pm_mesg.event == PM_EVENT_ON) {
2131 spin_unlock_irqrestore(ap->lock, flags);
2132 return;
2133 }
2134 spin_unlock_irqrestore(ap->lock, flags);
2135
2136 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2137
2138 /* suspend */
2139 ata_eh_freeze_port(ap);
2140
2141 if (ap->ops->port_suspend)
2142 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2143
2144 /* report result */
2145 spin_lock_irqsave(ap->lock, flags);
2146
2147 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2148 if (rc == 0)
2149 ap->pflags |= ATA_PFLAG_SUSPENDED;
2150 else
2151 ata_port_schedule_eh(ap);
2152
2153 if (ap->pm_result) {
2154 *ap->pm_result = rc;
2155 ap->pm_result = NULL;
2156 }
2157
2158 spin_unlock_irqrestore(ap->lock, flags);
2159
2160 return;
2161}
2162
2163/**
2164 * ata_eh_handle_port_resume - perform port resume operation
2165 * @ap: port to resume
2166 *
2167 * Resume @ap.
2168 *
2169 * This function also waits upto one second until all devices
2170 * hanging off this port requests resume EH action. This is to
2171 * prevent invoking EH and thus reset multiple times on resume.
2172 *
2173 * On DPM resume, where some of devices might not be resumed
2174 * together, this may delay port resume upto one second, but such
2175 * DPM resumes are rare and 1 sec delay isn't too bad.
2176 *
2177 * LOCKING:
2178 * Kernel thread context (may sleep).
2179 */
2180static void ata_eh_handle_port_resume(struct ata_port *ap)
2181{
2182 unsigned long timeout;
2183 unsigned long flags;
2184 int i, rc = 0;
2185
2186 /* are we resuming? */
2187 spin_lock_irqsave(ap->lock, flags);
2188 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2189 ap->pm_mesg.event != PM_EVENT_ON) {
2190 spin_unlock_irqrestore(ap->lock, flags);
2191 return;
2192 }
2193 spin_unlock_irqrestore(ap->lock, flags);
2194
2195 /* spurious? */
2196 if (!(ap->pflags & ATA_PFLAG_SUSPENDED))
2197 goto done;
2198
2199 if (ap->ops->port_resume)
2200 rc = ap->ops->port_resume(ap);
2201
2202 /* give devices time to request EH */
2203 timeout = jiffies + HZ; /* 1s max */
2204 while (1) {
2205 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2206 struct ata_device *dev = &ap->device[i];
2207 unsigned int action = ata_eh_dev_action(dev);
2208
2209 if ((dev->flags & ATA_DFLAG_SUSPENDED) &&
2210 !(action & ATA_EH_RESUME))
2211 break;
2212 }
2213
2214 if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout))
2215 break;
2216 msleep(10);
2217 }
2218
2219 done:
2220 spin_lock_irqsave(ap->lock, flags);
2221 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2222 if (ap->pm_result) {
2223 *ap->pm_result = rc;
2224 ap->pm_result = NULL;
2225 }
2226 spin_unlock_irqrestore(ap->lock, flags);
2227}