blob: 361953a502035e4c846196f05ad3ad256ac6220b [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
Tejun Heoece1d632006-04-02 18:51:53 +090035#include <linux/kernel.h>
36#include <scsi/scsi.h>
37#include <scsi/scsi_host.h>
38#include <scsi/scsi_eh.h>
39#include <scsi/scsi_device.h>
40#include <scsi/scsi_cmnd.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040041#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090042
43#include <linux/libata.h>
44
45#include "libata.h"
46
Tejun Heo7d47e8d2007-02-02 16:22:31 +090047enum {
48 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
49 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
50 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
51};
52
Tejun Heoad9e2762006-05-15 20:58:12 +090053static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo720ba122006-05-31 18:28:13 +090054static void ata_eh_finish(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090055#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +090056static void ata_eh_handle_port_suspend(struct ata_port *ap);
57static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090058static int ata_eh_suspend(struct ata_port *ap,
59 struct ata_device **r_failed_dev);
60static void ata_eh_prep_resume(struct ata_port *ap);
61static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev);
62#else /* CONFIG_PM */
63static void ata_eh_handle_port_suspend(struct ata_port *ap)
64{ }
65
66static void ata_eh_handle_port_resume(struct ata_port *ap)
67{ }
68
69static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev)
70{
71 return 0;
72}
73
74static void ata_eh_prep_resume(struct ata_port *ap)
75{ }
76
77static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev)
78{
79 return 0;
80}
81#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +090082
Tejun Heo0c247c52006-05-15 20:58:19 +090083static void ata_ering_record(struct ata_ering *ering, int is_io,
84 unsigned int err_mask)
85{
86 struct ata_ering_entry *ent;
87
88 WARN_ON(!err_mask);
89
90 ering->cursor++;
91 ering->cursor %= ATA_ERING_SIZE;
92
93 ent = &ering->ring[ering->cursor];
94 ent->is_io = is_io;
95 ent->err_mask = err_mask;
96 ent->timestamp = get_jiffies_64();
97}
98
Tejun Heo7d47e8d2007-02-02 16:22:31 +090099static void ata_ering_clear(struct ata_ering *ering)
Tejun Heo0c247c52006-05-15 20:58:19 +0900100{
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900101 memset(ering, 0, sizeof(*ering));
Tejun Heo0c247c52006-05-15 20:58:19 +0900102}
103
104static int ata_ering_map(struct ata_ering *ering,
105 int (*map_fn)(struct ata_ering_entry *, void *),
106 void *arg)
107{
108 int idx, rc = 0;
109 struct ata_ering_entry *ent;
110
111 idx = ering->cursor;
112 do {
113 ent = &ering->ring[idx];
114 if (!ent->err_mask)
115 break;
116 rc = map_fn(ent, arg);
117 if (rc)
118 break;
119 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
120 } while (idx != ering->cursor);
121
122 return rc;
123}
124
Tejun Heo64f65ca2006-06-24 20:30:18 +0900125static unsigned int ata_eh_dev_action(struct ata_device *dev)
126{
127 struct ata_eh_context *ehc = &dev->ap->eh_context;
128
129 return ehc->i.action | ehc->i.dev_action[dev->devno];
130}
131
Tejun Heoaf181c22006-06-24 20:30:18 +0900132static void ata_eh_clear_action(struct ata_device *dev,
133 struct ata_eh_info *ehi, unsigned int action)
134{
135 int i;
136
137 if (!dev) {
138 ehi->action &= ~action;
139 for (i = 0; i < ATA_MAX_DEVICES; i++)
140 ehi->dev_action[i] &= ~action;
141 } else {
142 /* doesn't make sense for port-wide EH actions */
143 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
144
145 /* break ehi->action into ehi->dev_action */
146 if (ehi->action & action) {
147 for (i = 0; i < ATA_MAX_DEVICES; i++)
148 ehi->dev_action[i] |= ehi->action & action;
149 ehi->action &= ~action;
150 }
151
152 /* turn off the specified per-dev action */
153 ehi->dev_action[dev->devno] &= ~action;
154 }
155}
156
Tejun Heoece1d632006-04-02 18:51:53 +0900157/**
158 * ata_scsi_timed_out - SCSI layer time out callback
159 * @cmd: timed out SCSI command
160 *
161 * Handles SCSI layer timeout. We race with normal completion of
162 * the qc for @cmd. If the qc is already gone, we lose and let
163 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
164 * timed out and EH should be invoked. Prevent ata_qc_complete()
165 * from finishing it by setting EH_SCHEDULED and return
166 * EH_NOT_HANDLED.
167 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900168 * TODO: kill this function once old EH is gone.
169 *
Tejun Heoece1d632006-04-02 18:51:53 +0900170 * LOCKING:
171 * Called from timer context
172 *
173 * RETURNS:
174 * EH_HANDLED or EH_NOT_HANDLED
175 */
176enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
177{
178 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400179 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900180 unsigned long flags;
181 struct ata_queued_cmd *qc;
Tejun Heoad9e2762006-05-15 20:58:12 +0900182 enum scsi_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900183
184 DPRINTK("ENTER\n");
185
Tejun Heoad9e2762006-05-15 20:58:12 +0900186 if (ap->ops->error_handler) {
187 ret = EH_NOT_HANDLED;
188 goto out;
189 }
190
191 ret = EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400192 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900193 qc = ata_qc_from_tag(ap, ap->active_tag);
194 if (qc) {
195 WARN_ON(qc->scsicmd != cmd);
196 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
197 qc->err_mask |= AC_ERR_TIMEOUT;
198 ret = EH_NOT_HANDLED;
199 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400200 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900201
Tejun Heoad9e2762006-05-15 20:58:12 +0900202 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900203 DPRINTK("EXIT, ret=%d\n", ret);
204 return ret;
205}
206
207/**
208 * ata_scsi_error - SCSI layer error handler callback
209 * @host: SCSI host on which error occurred
210 *
211 * Handles SCSI-layer-thrown error events.
212 *
213 * LOCKING:
214 * Inherited from SCSI layer (none, can sleep)
215 *
216 * RETURNS:
217 * Zero.
218 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400219void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900220{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400221 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoad9e2762006-05-15 20:58:12 +0900222 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
223 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900224
225 DPRINTK("ENTER\n");
226
Tejun Heoad9e2762006-05-15 20:58:12 +0900227 /* synchronize with port task */
Tejun Heoece1d632006-04-02 18:51:53 +0900228 ata_port_flush_task(ap);
229
Jeff Garzikcca39742006-08-24 03:19:22 -0400230 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900231
Tejun Heoad9e2762006-05-15 20:58:12 +0900232 /* For new EH, all qcs are finished in one of three ways -
233 * normal completion, error completion, and SCSI timeout.
234 * Both cmpletions can race against SCSI timeout. When normal
235 * completion wins, the qc never reaches EH. When error
236 * completion wins, the qc has ATA_QCFLAG_FAILED set.
237 *
238 * When SCSI timeout wins, things are a bit more complex.
239 * Normal or error completion can occur after the timeout but
240 * before this point. In such cases, both types of
241 * completions are honored. A scmd is determined to have
242 * timed out iff its associated qc is active and not failed.
243 */
244 if (ap->ops->error_handler) {
245 struct scsi_cmnd *scmd, *tmp;
246 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900247
Tejun Heoe30349d2006-07-03 03:02:15 +0900248 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900249
250 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
251 struct ata_queued_cmd *qc;
252
253 for (i = 0; i < ATA_MAX_QUEUE; i++) {
254 qc = __ata_qc_from_tag(ap, i);
255 if (qc->flags & ATA_QCFLAG_ACTIVE &&
256 qc->scsicmd == scmd)
257 break;
258 }
259
260 if (i < ATA_MAX_QUEUE) {
261 /* the scmd has an associated qc */
262 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
263 /* which hasn't failed yet, timeout */
264 qc->err_mask |= AC_ERR_TIMEOUT;
265 qc->flags |= ATA_QCFLAG_FAILED;
266 nr_timedout++;
267 }
268 } else {
269 /* Normal completion occurred after
270 * SCSI timeout but before this point.
271 * Successfully complete it.
272 */
273 scmd->retries = scmd->allowed;
274 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
275 }
276 }
277
278 /* If we have timed out qcs. They belong to EH from
279 * this point but the state of the controller is
280 * unknown. Freeze the port to make sure the IRQ
281 * handler doesn't diddle with those qcs. This must
282 * be done atomically w.r.t. setting QCFLAG_FAILED.
283 */
284 if (nr_timedout)
285 __ata_port_freeze(ap);
286
Tejun Heoe30349d2006-07-03 03:02:15 +0900287 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900288 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900289 spin_unlock_wait(ap->lock);
Tejun Heoad9e2762006-05-15 20:58:12 +0900290
291 repeat:
292 /* invoke error handler */
293 if (ap->ops->error_handler) {
Tejun Heo500530f2006-07-03 16:07:27 +0900294 /* process port resume request */
295 ata_eh_handle_port_resume(ap);
296
Tejun Heof3e81b12006-05-15 20:58:21 +0900297 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900298 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900299
300 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
301 ap->eh_context.i = ap->eh_info;
302 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
303
Tejun Heob51e9e52006-06-29 01:29:30 +0900304 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
305 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof3e81b12006-05-15 20:58:21 +0900306
Tejun Heoe30349d2006-07-03 03:02:15 +0900307 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900308
Tejun Heo500530f2006-07-03 16:07:27 +0900309 /* invoke EH, skip if unloading or suspended */
310 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900311 ap->ops->error_handler(ap);
312 else
313 ata_eh_finish(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900314
Tejun Heo500530f2006-07-03 16:07:27 +0900315 /* process port suspend request */
316 ata_eh_handle_port_suspend(ap);
317
Tejun Heoad9e2762006-05-15 20:58:12 +0900318 /* Exception might have happend after ->error_handler
319 * recovered the port but before this point. Repeat
320 * EH in such case.
321 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900322 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900323
Tejun Heob51e9e52006-06-29 01:29:30 +0900324 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoad9e2762006-05-15 20:58:12 +0900325 if (--repeat_cnt) {
326 ata_port_printk(ap, KERN_INFO,
327 "EH pending after completion, "
328 "repeating EH (cnt=%d)\n", repeat_cnt);
Tejun Heoe30349d2006-07-03 03:02:15 +0900329 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900330 goto repeat;
331 }
332 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
333 "tries, giving up\n", ATA_EH_MAX_REPEAT);
334 }
335
Tejun Heof3e81b12006-05-15 20:58:21 +0900336 /* this run is complete, make sure EH info is clear */
337 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
338
Tejun Heoe30349d2006-07-03 03:02:15 +0900339 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900340 * that if exception occurs after this point but
341 * before EH completion, SCSI midlayer will
342 * re-initiate EH.
343 */
344 host->host_eh_scheduled = 0;
345
Tejun Heoe30349d2006-07-03 03:02:15 +0900346 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900347 } else {
348 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
349 ap->ops->eng_timeout(ap);
350 }
351
352 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900353 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
354
355 scsi_eh_flush_done_q(&ap->eh_done_q);
356
Tejun Heoad9e2762006-05-15 20:58:12 +0900357 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900358 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900359
Tejun Heo1cdaf532006-07-03 16:07:26 +0900360 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900361 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900362 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
David Howells52bad642006-11-22 14:54:01 +0000363 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900364
365 if (ap->pflags & ATA_PFLAG_RECOVERED)
366 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900367
Tejun Heob51e9e52006-06-29 01:29:30 +0900368 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900369
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900370 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900371 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900372 wake_up_all(&ap->eh_wait_q);
373
Tejun Heoe30349d2006-07-03 03:02:15 +0900374 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900375
Tejun Heoece1d632006-04-02 18:51:53 +0900376 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900377}
378
379/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900380 * ata_port_wait_eh - Wait for the currently pending EH to complete
381 * @ap: Port to wait EH for
382 *
383 * Wait until the currently pending EH is complete.
384 *
385 * LOCKING:
386 * Kernel thread context (may sleep).
387 */
388void ata_port_wait_eh(struct ata_port *ap)
389{
390 unsigned long flags;
391 DEFINE_WAIT(wait);
392
393 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400394 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900395
Tejun Heob51e9e52006-06-29 01:29:30 +0900396 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900397 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400398 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900399 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400400 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900401 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900402 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900403
Jeff Garzikba6a1302006-06-22 23:46:10 -0400404 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900405
406 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400407 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900408 msleep(10);
409 goto retry;
410 }
411}
412
413/**
Tejun Heoece1d632006-04-02 18:51:53 +0900414 * ata_qc_timeout - Handle timeout of queued command
415 * @qc: Command that timed out
416 *
417 * Some part of the kernel (currently, only the SCSI layer)
418 * has noticed that the active command on port @ap has not
419 * completed after a specified length of time. Handle this
420 * condition by disabling DMA (if necessary) and completing
421 * transactions, with error if necessary.
422 *
423 * This also handles the case of the "lost interrupt", where
424 * for some reason (possibly hardware bug, possibly driver bug)
425 * an interrupt was not delivered to the driver, even though the
426 * transaction completed successfully.
427 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900428 * TODO: kill this function once old EH is gone.
429 *
Tejun Heoece1d632006-04-02 18:51:53 +0900430 * LOCKING:
431 * Inherited from SCSI layer (none, can sleep)
432 */
433static void ata_qc_timeout(struct ata_queued_cmd *qc)
434{
435 struct ata_port *ap = qc->ap;
Tejun Heoece1d632006-04-02 18:51:53 +0900436 u8 host_stat = 0, drv_stat;
437 unsigned long flags;
438
439 DPRINTK("ENTER\n");
440
441 ap->hsm_task_state = HSM_ST_IDLE;
442
Jeff Garzikba6a1302006-06-22 23:46:10 -0400443 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900444
445 switch (qc->tf.protocol) {
446
447 case ATA_PROT_DMA:
448 case ATA_PROT_ATAPI_DMA:
449 host_stat = ap->ops->bmdma_status(ap);
450
451 /* before we do anything else, clear DMA-Start bit */
452 ap->ops->bmdma_stop(qc);
453
454 /* fall through */
455
456 default:
457 ata_altstatus(ap);
458 drv_stat = ata_chk_status(ap);
459
460 /* ack bmdma irq events */
461 ap->ops->irq_clear(ap);
462
Tejun Heof15a1da2006-05-15 20:57:56 +0900463 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
464 "stat 0x%x host_stat 0x%x\n",
465 qc->tf.command, drv_stat, host_stat);
Tejun Heoece1d632006-04-02 18:51:53 +0900466
467 /* complete taskfile transaction */
Jeff Garzikc13b56a2006-04-02 10:34:24 -0400468 qc->err_mask |= AC_ERR_TIMEOUT;
Tejun Heoece1d632006-04-02 18:51:53 +0900469 break;
470 }
471
Jeff Garzikba6a1302006-06-22 23:46:10 -0400472 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900473
474 ata_eh_qc_complete(qc);
475
476 DPRINTK("EXIT\n");
477}
478
479/**
480 * ata_eng_timeout - Handle timeout of queued command
481 * @ap: Port on which timed-out command is active
482 *
483 * Some part of the kernel (currently, only the SCSI layer)
484 * has noticed that the active command on port @ap has not
485 * completed after a specified length of time. Handle this
486 * condition by disabling DMA (if necessary) and completing
487 * transactions, with error if necessary.
488 *
489 * This also handles the case of the "lost interrupt", where
490 * for some reason (possibly hardware bug, possibly driver bug)
491 * an interrupt was not delivered to the driver, even though the
492 * transaction completed successfully.
493 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900494 * TODO: kill this function once old EH is gone.
495 *
Tejun Heoece1d632006-04-02 18:51:53 +0900496 * LOCKING:
497 * Inherited from SCSI layer (none, can sleep)
498 */
499void ata_eng_timeout(struct ata_port *ap)
500{
501 DPRINTK("ENTER\n");
502
503 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
504
505 DPRINTK("EXIT\n");
506}
507
Tejun Heof686bcb2006-05-15 20:58:05 +0900508/**
509 * ata_qc_schedule_eh - schedule qc for error handling
510 * @qc: command to schedule error handling for
511 *
512 * Schedule error handling for @qc. EH will kick in as soon as
513 * other commands are drained.
514 *
515 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400516 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900517 */
518void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
519{
520 struct ata_port *ap = qc->ap;
521
522 WARN_ON(!ap->ops->error_handler);
523
524 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heob51e9e52006-06-29 01:29:30 +0900525 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
Tejun Heof686bcb2006-05-15 20:58:05 +0900526
527 /* The following will fail if timeout has already expired.
528 * ata_scsi_error() takes care of such scmds on EH entry.
529 * Note that ATA_QCFLAG_FAILED is unconditionally set after
530 * this function completes.
531 */
532 scsi_req_abort_cmd(qc->scsicmd);
533}
534
Tejun Heo7b70fc02006-05-15 20:58:07 +0900535/**
536 * ata_port_schedule_eh - schedule error handling without a qc
537 * @ap: ATA port to schedule EH for
538 *
539 * Schedule error handling for @ap. EH will kick in as soon as
540 * all commands are drained.
541 *
542 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400543 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900544 */
545void ata_port_schedule_eh(struct ata_port *ap)
546{
547 WARN_ON(!ap->ops->error_handler);
548
Tejun Heob51e9e52006-06-29 01:29:30 +0900549 ap->pflags |= ATA_PFLAG_EH_PENDING;
Jeff Garzikcca39742006-08-24 03:19:22 -0400550 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900551
552 DPRINTK("port EH scheduled\n");
553}
554
555/**
556 * ata_port_abort - abort all qc's on the port
557 * @ap: ATA port to abort qc's for
558 *
559 * Abort all active qc's of @ap and schedule EH.
560 *
561 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400562 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900563 *
564 * RETURNS:
565 * Number of aborted qc's.
566 */
567int ata_port_abort(struct ata_port *ap)
568{
569 int tag, nr_aborted = 0;
570
571 WARN_ON(!ap->ops->error_handler);
572
573 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
574 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
575
576 if (qc) {
577 qc->flags |= ATA_QCFLAG_FAILED;
578 ata_qc_complete(qc);
579 nr_aborted++;
580 }
581 }
582
583 if (!nr_aborted)
584 ata_port_schedule_eh(ap);
585
586 return nr_aborted;
587}
588
Tejun Heoe3180492006-05-15 20:58:09 +0900589/**
590 * __ata_port_freeze - freeze port
591 * @ap: ATA port to freeze
592 *
593 * This function is called when HSM violation or some other
594 * condition disrupts normal operation of the port. Frozen port
595 * is not allowed to perform any operation until the port is
596 * thawed, which usually follows a successful reset.
597 *
598 * ap->ops->freeze() callback can be used for freezing the port
599 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
600 * port cannot be frozen hardware-wise, the interrupt handler
601 * must ack and clear interrupts unconditionally while the port
602 * is frozen.
603 *
604 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400605 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900606 */
607static void __ata_port_freeze(struct ata_port *ap)
608{
609 WARN_ON(!ap->ops->error_handler);
610
611 if (ap->ops->freeze)
612 ap->ops->freeze(ap);
613
Tejun Heob51e9e52006-06-29 01:29:30 +0900614 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900615
Tejun Heo44877b42007-02-21 01:06:51 +0900616 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900617}
618
619/**
620 * ata_port_freeze - abort & freeze port
621 * @ap: ATA port to freeze
622 *
623 * Abort and freeze @ap.
624 *
625 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400626 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900627 *
628 * RETURNS:
629 * Number of aborted commands.
630 */
631int ata_port_freeze(struct ata_port *ap)
632{
633 int nr_aborted;
634
635 WARN_ON(!ap->ops->error_handler);
636
637 nr_aborted = ata_port_abort(ap);
638 __ata_port_freeze(ap);
639
640 return nr_aborted;
641}
642
643/**
644 * ata_eh_freeze_port - EH helper to freeze port
645 * @ap: ATA port to freeze
646 *
647 * Freeze @ap.
648 *
649 * LOCKING:
650 * None.
651 */
652void ata_eh_freeze_port(struct ata_port *ap)
653{
654 unsigned long flags;
655
656 if (!ap->ops->error_handler)
657 return;
658
Jeff Garzikba6a1302006-06-22 23:46:10 -0400659 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900660 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400661 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900662}
663
664/**
665 * ata_port_thaw_port - EH helper to thaw port
666 * @ap: ATA port to thaw
667 *
668 * Thaw frozen port @ap.
669 *
670 * LOCKING:
671 * None.
672 */
673void ata_eh_thaw_port(struct ata_port *ap)
674{
675 unsigned long flags;
676
677 if (!ap->ops->error_handler)
678 return;
679
Jeff Garzikba6a1302006-06-22 23:46:10 -0400680 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900681
Tejun Heob51e9e52006-06-29 01:29:30 +0900682 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900683
684 if (ap->ops->thaw)
685 ap->ops->thaw(ap);
686
Jeff Garzikba6a1302006-06-22 23:46:10 -0400687 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900688
Tejun Heo44877b42007-02-21 01:06:51 +0900689 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900690}
691
Tejun Heoece1d632006-04-02 18:51:53 +0900692static void ata_eh_scsidone(struct scsi_cmnd *scmd)
693{
694 /* nada */
695}
696
697static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
698{
699 struct ata_port *ap = qc->ap;
700 struct scsi_cmnd *scmd = qc->scsicmd;
701 unsigned long flags;
702
Jeff Garzikba6a1302006-06-22 23:46:10 -0400703 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900704 qc->scsidone = ata_eh_scsidone;
705 __ata_qc_complete(qc);
706 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -0400707 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900708
709 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
710}
711
712/**
713 * ata_eh_qc_complete - Complete an active ATA command from EH
714 * @qc: Command to complete
715 *
716 * Indicate to the mid and upper layers that an ATA command has
717 * completed. To be used from EH.
718 */
719void ata_eh_qc_complete(struct ata_queued_cmd *qc)
720{
721 struct scsi_cmnd *scmd = qc->scsicmd;
722 scmd->retries = scmd->allowed;
723 __ata_eh_qc_complete(qc);
724}
725
726/**
727 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
728 * @qc: Command to retry
729 *
730 * Indicate to the mid and upper layers that an ATA command
731 * should be retried. To be used from EH.
732 *
733 * SCSI midlayer limits the number of retries to scmd->allowed.
734 * scmd->retries is decremented for commands which get retried
735 * due to unrelated failures (qc->err_mask is zero).
736 */
737void ata_eh_qc_retry(struct ata_queued_cmd *qc)
738{
739 struct scsi_cmnd *scmd = qc->scsicmd;
740 if (!qc->err_mask && scmd->retries)
741 scmd->retries--;
742 __ata_eh_qc_complete(qc);
743}
Tejun Heo022bdb02006-05-15 20:58:22 +0900744
745/**
Tejun Heo0ea035a2006-05-31 18:28:01 +0900746 * ata_eh_detach_dev - detach ATA device
747 * @dev: ATA device to detach
748 *
749 * Detach @dev.
750 *
751 * LOCKING:
752 * None.
753 */
754static void ata_eh_detach_dev(struct ata_device *dev)
755{
756 struct ata_port *ap = dev->ap;
757 unsigned long flags;
758
759 ata_dev_disable(dev);
760
Jeff Garzikba6a1302006-06-22 23:46:10 -0400761 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +0900762
763 dev->flags &= ~ATA_DFLAG_DETACH;
764
765 if (ata_scsi_offline_dev(dev)) {
766 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +0900767 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +0900768 }
769
Tejun Heobeb07c12006-06-24 20:30:19 +0900770 /* clear per-dev EH actions */
771 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
772 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
773
Jeff Garzikba6a1302006-06-22 23:46:10 -0400774 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +0900775}
776
777/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900778 * ata_eh_about_to_do - about to perform eh_action
779 * @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +0900780 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +0900781 * @action: action about to be performed
782 *
783 * Called just before performing EH actions to clear related bits
784 * in @ap->eh_info such that eh actions are not unnecessarily
785 * repeated.
786 *
787 * LOCKING:
788 * None.
789 */
Tejun Heo47005f22006-06-19 18:27:23 +0900790static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
791 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +0900792{
793 unsigned long flags;
Tejun Heo13abf502006-07-10 23:18:46 +0900794 struct ata_eh_info *ehi = &ap->eh_info;
795 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +0900796
Jeff Garzikba6a1302006-06-22 23:46:10 -0400797 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900798
Tejun Heo13abf502006-07-10 23:18:46 +0900799 /* Reset is represented by combination of actions and EHI
800 * flags. Suck in all related bits before clearing eh_info to
801 * avoid losing requested action.
802 */
803 if (action & ATA_EH_RESET_MASK) {
804 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
805 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900806
Tejun Heo13abf502006-07-10 23:18:46 +0900807 /* make sure all reset actions are cleared & clear EHI flags */
808 action |= ATA_EH_RESET_MASK;
809 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
810 }
811
812 ata_eh_clear_action(dev, ehi, action);
813
814 if (!(ehc->i.flags & ATA_EHI_QUIET))
Tejun Heo1cdaf532006-07-03 16:07:26 +0900815 ap->pflags |= ATA_PFLAG_RECOVERED;
816
Jeff Garzikba6a1302006-06-22 23:46:10 -0400817 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +0900818}
819
820/**
Tejun Heo47005f22006-06-19 18:27:23 +0900821 * ata_eh_done - EH action complete
822 * @ap: target ATA port
823 * @dev: target ATA dev for per-dev action (can be NULL)
824 * @action: action just completed
825 *
826 * Called right after performing EH actions to clear related bits
827 * in @ap->eh_context.
828 *
829 * LOCKING:
830 * None.
831 */
832static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
833 unsigned int action)
834{
Tejun Heo13abf502006-07-10 23:18:46 +0900835 /* if reset is complete, clear all reset actions & reset modifier */
836 if (action & ATA_EH_RESET_MASK) {
837 action |= ATA_EH_RESET_MASK;
838 ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
839 }
840
Tejun Heo47005f22006-06-19 18:27:23 +0900841 ata_eh_clear_action(dev, &ap->eh_context.i, action);
842}
843
844/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900845 * ata_err_string - convert err_mask to descriptive string
846 * @err_mask: error mask to convert to string
847 *
848 * Convert @err_mask to descriptive string. Errors are
849 * prioritized according to severity and only the most severe
850 * error is reported.
851 *
852 * LOCKING:
853 * None.
854 *
855 * RETURNS:
856 * Descriptive string for @err_mask
857 */
858static const char * ata_err_string(unsigned int err_mask)
859{
860 if (err_mask & AC_ERR_HOST_BUS)
861 return "host bus error";
862 if (err_mask & AC_ERR_ATA_BUS)
863 return "ATA bus error";
864 if (err_mask & AC_ERR_TIMEOUT)
865 return "timeout";
866 if (err_mask & AC_ERR_HSM)
867 return "HSM violation";
868 if (err_mask & AC_ERR_SYSTEM)
869 return "internal error";
870 if (err_mask & AC_ERR_MEDIA)
871 return "media error";
872 if (err_mask & AC_ERR_INVALID)
873 return "invalid argument";
874 if (err_mask & AC_ERR_DEV)
875 return "device error";
876 return "unknown error";
877}
878
879/**
Tejun Heoe8ee8452006-05-15 21:03:46 +0900880 * ata_read_log_page - read a specific log page
881 * @dev: target device
882 * @page: page to read
883 * @buf: buffer to store read page
884 * @sectors: number of sectors to read
885 *
886 * Read log page using READ_LOG_EXT command.
887 *
888 * LOCKING:
889 * Kernel thread context (may sleep).
890 *
891 * RETURNS:
892 * 0 on success, AC_ERR_* mask otherwise.
893 */
894static unsigned int ata_read_log_page(struct ata_device *dev,
895 u8 page, void *buf, unsigned int sectors)
896{
897 struct ata_taskfile tf;
898 unsigned int err_mask;
899
900 DPRINTK("read log page - page %d\n", page);
901
902 ata_tf_init(dev, &tf);
903 tf.command = ATA_CMD_READ_LOG_EXT;
904 tf.lbal = page;
905 tf.nsect = sectors;
906 tf.hob_nsect = sectors >> 8;
907 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
908 tf.protocol = ATA_PROT_PIO;
909
910 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
911 buf, sectors * ATA_SECT_SIZE);
912
913 DPRINTK("EXIT, err_mask=%x\n", err_mask);
914 return err_mask;
915}
916
917/**
918 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
919 * @dev: Device to read log page 10h from
920 * @tag: Resulting tag of the failed command
921 * @tf: Resulting taskfile registers of the failed command
922 *
923 * Read log page 10h to obtain NCQ error details and clear error
924 * condition.
925 *
926 * LOCKING:
927 * Kernel thread context (may sleep).
928 *
929 * RETURNS:
930 * 0 on success, -errno otherwise.
931 */
932static int ata_eh_read_log_10h(struct ata_device *dev,
933 int *tag, struct ata_taskfile *tf)
934{
935 u8 *buf = dev->ap->sector_buf;
936 unsigned int err_mask;
937 u8 csum;
938 int i;
939
940 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
941 if (err_mask)
942 return -EIO;
943
944 csum = 0;
945 for (i = 0; i < ATA_SECT_SIZE; i++)
946 csum += buf[i];
947 if (csum)
948 ata_dev_printk(dev, KERN_WARNING,
949 "invalid checksum 0x%x on log page 10h\n", csum);
950
951 if (buf[0] & 0x80)
952 return -ENOENT;
953
954 *tag = buf[0] & 0x1f;
955
956 tf->command = buf[2];
957 tf->feature = buf[3];
958 tf->lbal = buf[4];
959 tf->lbam = buf[5];
960 tf->lbah = buf[6];
961 tf->device = buf[7];
962 tf->hob_lbal = buf[8];
963 tf->hob_lbam = buf[9];
964 tf->hob_lbah = buf[10];
965 tf->nsect = buf[12];
966 tf->hob_nsect = buf[13];
967
968 return 0;
969}
970
971/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900972 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
973 * @dev: device to perform REQUEST_SENSE to
974 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
975 *
976 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
977 * SENSE. This function is EH helper.
978 *
979 * LOCKING:
980 * Kernel thread context (may sleep).
981 *
982 * RETURNS:
983 * 0 on success, AC_ERR_* mask on failure
984 */
985static unsigned int atapi_eh_request_sense(struct ata_device *dev,
986 unsigned char *sense_buf)
987{
988 struct ata_port *ap = dev->ap;
989 struct ata_taskfile tf;
990 u8 cdb[ATAPI_CDB_LEN];
991
992 DPRINTK("ATAPI request sense\n");
993
994 ata_tf_init(dev, &tf);
995
996 /* FIXME: is this needed? */
997 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
998
999 /* XXX: why tf_read here? */
1000 ap->ops->tf_read(ap, &tf);
1001
1002 /* fill these in, for the case where they are -not- overwritten */
1003 sense_buf[0] = 0x70;
1004 sense_buf[2] = tf.feature >> 4;
1005
1006 memset(cdb, 0, ATAPI_CDB_LEN);
1007 cdb[0] = REQUEST_SENSE;
1008 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1009
1010 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1011 tf.command = ATA_CMD_PACKET;
1012
1013 /* is it pointless to prefer PIO for "safety reasons"? */
1014 if (ap->flags & ATA_FLAG_PIO_DMA) {
1015 tf.protocol = ATA_PROT_ATAPI_DMA;
1016 tf.feature |= ATAPI_PKT_DMA;
1017 } else {
1018 tf.protocol = ATA_PROT_ATAPI;
1019 tf.lbam = (8 * 1024) & 0xff;
1020 tf.lbah = (8 * 1024) >> 8;
1021 }
1022
1023 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1024 sense_buf, SCSI_SENSE_BUFFERSIZE);
1025}
1026
1027/**
1028 * ata_eh_analyze_serror - analyze SError for a failed port
1029 * @ap: ATA port to analyze SError for
1030 *
1031 * Analyze SError if available and further determine cause of
1032 * failure.
1033 *
1034 * LOCKING:
1035 * None.
1036 */
1037static void ata_eh_analyze_serror(struct ata_port *ap)
1038{
1039 struct ata_eh_context *ehc = &ap->eh_context;
1040 u32 serror = ehc->i.serror;
1041 unsigned int err_mask = 0, action = 0;
1042
1043 if (serror & SERR_PERSISTENT) {
1044 err_mask |= AC_ERR_ATA_BUS;
1045 action |= ATA_EH_HARDRESET;
1046 }
1047 if (serror &
1048 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1049 err_mask |= AC_ERR_ATA_BUS;
1050 action |= ATA_EH_SOFTRESET;
1051 }
1052 if (serror & SERR_PROTOCOL) {
1053 err_mask |= AC_ERR_HSM;
1054 action |= ATA_EH_SOFTRESET;
1055 }
1056 if (serror & SERR_INTERNAL) {
1057 err_mask |= AC_ERR_SYSTEM;
1058 action |= ATA_EH_SOFTRESET;
1059 }
Tejun Heo084fe632006-05-31 18:28:03 +09001060 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1061 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001062
1063 ehc->i.err_mask |= err_mask;
1064 ehc->i.action |= action;
1065}
1066
1067/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001068 * ata_eh_analyze_ncq_error - analyze NCQ error
1069 * @ap: ATA port to analyze NCQ error for
1070 *
1071 * Read log page 10h, determine the offending qc and acquire
1072 * error status TF. For NCQ device errors, all LLDDs have to do
1073 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1074 * care of the rest.
1075 *
1076 * LOCKING:
1077 * Kernel thread context (may sleep).
1078 */
1079static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1080{
1081 struct ata_eh_context *ehc = &ap->eh_context;
1082 struct ata_device *dev = ap->device;
1083 struct ata_queued_cmd *qc;
1084 struct ata_taskfile tf;
1085 int tag, rc;
1086
1087 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001088 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001089 return;
1090
1091 /* is it NCQ device error? */
1092 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1093 return;
1094
1095 /* has LLDD analyzed already? */
1096 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1097 qc = __ata_qc_from_tag(ap, tag);
1098
1099 if (!(qc->flags & ATA_QCFLAG_FAILED))
1100 continue;
1101
1102 if (qc->err_mask)
1103 return;
1104 }
1105
1106 /* okay, this error is ours */
1107 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1108 if (rc) {
1109 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1110 "(errno=%d)\n", rc);
1111 return;
1112 }
1113
1114 if (!(ap->sactive & (1 << tag))) {
1115 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1116 "inactive tag %d\n", tag);
1117 return;
1118 }
1119
1120 /* we've got the perpetrator, condemn it */
1121 qc = __ata_qc_from_tag(ap, tag);
1122 memcpy(&qc->result_tf, &tf, sizeof(tf));
1123 qc->err_mask |= AC_ERR_DEV;
1124 ehc->i.err_mask &= ~AC_ERR_DEV;
1125}
1126
1127/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001128 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1129 * @qc: qc to analyze
1130 * @tf: Taskfile registers to analyze
1131 *
1132 * Analyze taskfile of @qc and further determine cause of
1133 * failure. This function also requests ATAPI sense data if
1134 * avaliable.
1135 *
1136 * LOCKING:
1137 * Kernel thread context (may sleep).
1138 *
1139 * RETURNS:
1140 * Determined recovery action
1141 */
1142static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1143 const struct ata_taskfile *tf)
1144{
1145 unsigned int tmp, action = 0;
1146 u8 stat = tf->command, err = tf->feature;
1147
1148 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1149 qc->err_mask |= AC_ERR_HSM;
1150 return ATA_EH_SOFTRESET;
1151 }
1152
1153 if (!(qc->err_mask & AC_ERR_DEV))
1154 return 0;
1155
1156 switch (qc->dev->class) {
1157 case ATA_DEV_ATA:
1158 if (err & ATA_ICRC)
1159 qc->err_mask |= AC_ERR_ATA_BUS;
1160 if (err & ATA_UNC)
1161 qc->err_mask |= AC_ERR_MEDIA;
1162 if (err & ATA_IDNF)
1163 qc->err_mask |= AC_ERR_INVALID;
1164 break;
1165
1166 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001167 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1168 tmp = atapi_eh_request_sense(qc->dev,
1169 qc->scsicmd->sense_buffer);
1170 if (!tmp) {
1171 /* ATA_QCFLAG_SENSE_VALID is used to
1172 * tell atapi_qc_complete() that sense
1173 * data is already valid.
1174 *
1175 * TODO: interpret sense data and set
1176 * appropriate err_mask.
1177 */
1178 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1179 } else
1180 qc->err_mask |= tmp;
1181 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001182 }
1183
1184 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1185 action |= ATA_EH_SOFTRESET;
1186
1187 return action;
1188}
1189
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001190static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001191{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001192 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo022bdb02006-05-15 20:58:22 +09001193 return 1;
1194
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001195 if (err_mask & AC_ERR_TIMEOUT)
1196 return 2;
1197
1198 if (is_io) {
1199 if (err_mask & AC_ERR_HSM)
Tejun Heo022bdb02006-05-15 20:58:22 +09001200 return 2;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001201 if ((err_mask &
1202 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1203 return 3;
Tejun Heo022bdb02006-05-15 20:58:22 +09001204 }
1205
1206 return 0;
1207}
1208
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001209struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001210 u64 since;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001211 int nr_errors[4];
Tejun Heo022bdb02006-05-15 20:58:22 +09001212};
1213
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001214static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001215{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001216 struct speed_down_verdict_arg *arg = void_arg;
1217 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
Tejun Heo022bdb02006-05-15 20:58:22 +09001218
1219 if (ent->timestamp < arg->since)
1220 return -1;
1221
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001222 arg->nr_errors[cat]++;
Tejun Heo022bdb02006-05-15 20:58:22 +09001223 return 0;
1224}
1225
1226/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001227 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001228 * @dev: Device of interest
1229 *
1230 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001231 * whether NCQ needs to be turned off, transfer speed should be
1232 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001233 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001234 * Cat-1 is ATA_BUS error for any command.
Tejun Heo022bdb02006-05-15 20:58:22 +09001235 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001236 * Cat-2 is TIMEOUT for any command or HSM violation for known
1237 * supported commands.
1238 *
1239 * Cat-3 is is unclassified DEV error for known supported
Tejun Heo022bdb02006-05-15 20:58:22 +09001240 * command.
1241 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001242 * NCQ needs to be turned off if there have been more than 3
1243 * Cat-2 + Cat-3 errors during last 10 minutes.
1244 *
1245 * Speed down is necessary if there have been more than 3 Cat-1 +
1246 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1247 *
1248 * Falling back to PIO mode is necessary if there have been more
1249 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1250 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001251 * LOCKING:
1252 * Inherited from caller.
1253 *
1254 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001255 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001256 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001257static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001258{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001259 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1260 u64 j64 = get_jiffies_64();
1261 struct speed_down_verdict_arg arg;
1262 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001263
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001264 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001265 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001266 arg.since = j64 - min(j64, j10mins);
1267 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001268
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001269 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1270 verdict |= ATA_EH_SPDN_NCQ_OFF;
1271 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1272 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001273
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001274 /* scan past 3 mins of error history */
1275 memset(&arg, 0, sizeof(arg));
1276 arg.since = j64 - min(j64, j5mins);
1277 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001278
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001279 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1280 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1281
1282 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001283}
1284
1285/**
1286 * ata_eh_speed_down - record error and speed down if necessary
1287 * @dev: Failed device
1288 * @is_io: Did the device fail during normal IO?
1289 * @err_mask: err_mask of the error
1290 *
1291 * Record error and examine error history to determine whether
1292 * adjusting transmission speed is necessary. It also sets
1293 * transmission limits appropriately if such adjustment is
1294 * necessary.
1295 *
1296 * LOCKING:
1297 * Kernel thread context (may sleep).
1298 *
1299 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001300 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001301 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001302static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1303 unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001304{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001305 unsigned int verdict;
1306 unsigned int action = 0;
1307
1308 /* don't bother if Cat-0 error */
1309 if (ata_eh_categorize_error(is_io, err_mask) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001310 return 0;
1311
1312 /* record error and determine whether speed down is necessary */
1313 ata_ering_record(&dev->ering, is_io, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001314 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001315
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001316 /* turn off NCQ? */
1317 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1318 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1319 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1320 dev->flags |= ATA_DFLAG_NCQ_OFF;
1321 ata_dev_printk(dev, KERN_WARNING,
1322 "NCQ disabled due to excessive errors\n");
1323 goto done;
1324 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001325
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001326 /* speed down? */
1327 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1328 /* speed down SATA link speed if possible */
1329 if (sata_down_spd_limit(dev->ap) == 0) {
1330 action |= ATA_EH_HARDRESET;
1331 goto done;
1332 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001333
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001334 /* lower transfer mode */
1335 if (dev->spdn_cnt < 2) {
1336 static const int dma_dnxfer_sel[] =
1337 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1338 static const int pio_dnxfer_sel[] =
1339 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1340 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001341
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001342 if (dev->xfer_shift != ATA_SHIFT_PIO)
1343 sel = dma_dnxfer_sel[dev->spdn_cnt];
1344 else
1345 sel = pio_dnxfer_sel[dev->spdn_cnt];
1346
1347 dev->spdn_cnt++;
1348
1349 if (ata_down_xfermask_limit(dev, sel) == 0) {
1350 action |= ATA_EH_SOFTRESET;
1351 goto done;
1352 }
1353 }
1354 }
1355
1356 /* Fall back to PIO? Slowing down to PIO is meaningless for
1357 * SATA. Consider it only for PATA.
1358 */
1359 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1360 (dev->ap->cbl != ATA_CBL_SATA) &&
1361 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1362 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1363 dev->spdn_cnt = 0;
1364 action |= ATA_EH_SOFTRESET;
1365 goto done;
1366 }
1367 }
1368
Tejun Heo022bdb02006-05-15 20:58:22 +09001369 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001370 done:
1371 /* device has been slowed down, blow error history */
1372 ata_ering_clear(&dev->ering);
1373 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001374}
1375
1376/**
1377 * ata_eh_autopsy - analyze error and determine recovery action
1378 * @ap: ATA port to perform autopsy on
1379 *
1380 * Analyze why @ap failed and determine which recovery action is
1381 * needed. This function also sets more detailed AC_ERR_* values
1382 * and fills sense data for ATAPI CHECK SENSE.
1383 *
1384 * LOCKING:
1385 * Kernel thread context (may sleep).
1386 */
1387static void ata_eh_autopsy(struct ata_port *ap)
1388{
1389 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001390 unsigned int all_err_mask = 0;
1391 int tag, is_io = 0;
1392 u32 serror;
1393 int rc;
1394
1395 DPRINTK("ENTER\n");
1396
Tejun Heo1cdaf532006-07-03 16:07:26 +09001397 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1398 return;
1399
Tejun Heo022bdb02006-05-15 20:58:22 +09001400 /* obtain and analyze SError */
1401 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1402 if (rc == 0) {
1403 ehc->i.serror |= serror;
1404 ata_eh_analyze_serror(ap);
1405 } else if (rc != -EOPNOTSUPP)
Tejun Heo4528e4d2006-07-08 20:17:26 +09001406 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001407
Tejun Heoe8ee8452006-05-15 21:03:46 +09001408 /* analyze NCQ failure */
1409 ata_eh_analyze_ncq_error(ap);
1410
Tejun Heo022bdb02006-05-15 20:58:22 +09001411 /* any real error trumps AC_ERR_OTHER */
1412 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1413 ehc->i.err_mask &= ~AC_ERR_OTHER;
1414
1415 all_err_mask |= ehc->i.err_mask;
1416
1417 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1418 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1419
1420 if (!(qc->flags & ATA_QCFLAG_FAILED))
1421 continue;
1422
1423 /* inherit upper level err_mask */
1424 qc->err_mask |= ehc->i.err_mask;
1425
Tejun Heo022bdb02006-05-15 20:58:22 +09001426 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001427 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001428
1429 /* DEV errors are probably spurious in case of ATA_BUS error */
1430 if (qc->err_mask & AC_ERR_ATA_BUS)
1431 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1432 AC_ERR_INVALID);
1433
1434 /* any real error trumps unknown error */
1435 if (qc->err_mask & ~AC_ERR_OTHER)
1436 qc->err_mask &= ~AC_ERR_OTHER;
1437
1438 /* SENSE_VALID trumps dev/unknown error and revalidation */
1439 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1440 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo4528e4d2006-07-08 20:17:26 +09001441 ehc->i.action &= ~ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001442 }
1443
1444 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001445 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001446 all_err_mask |= qc->err_mask;
1447 if (qc->flags & ATA_QCFLAG_IO)
1448 is_io = 1;
1449 }
1450
Tejun Heoa20f33f2006-05-16 12:58:24 +09001451 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001452 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001453 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001454 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heoa20f33f2006-05-16 12:58:24 +09001455 else if (all_err_mask)
Tejun Heo4528e4d2006-07-08 20:17:26 +09001456 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001457
Tejun Heo47005f22006-06-19 18:27:23 +09001458 /* if we have offending qcs and the associated failed device */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001459 if (ehc->i.dev) {
Tejun Heo47005f22006-06-19 18:27:23 +09001460 /* speed down */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001461 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1462 all_err_mask);
Tejun Heo47005f22006-06-19 18:27:23 +09001463
1464 /* perform per-dev EH action only on the offending device */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001465 ehc->i.dev_action[ehc->i.dev->devno] |=
1466 ehc->i.action & ATA_EH_PERDEV_MASK;
1467 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001468 }
1469
Tejun Heo022bdb02006-05-15 20:58:22 +09001470 DPRINTK("EXIT\n");
1471}
1472
1473/**
1474 * ata_eh_report - report error handling to user
1475 * @ap: ATA port EH is going on
1476 *
1477 * Report EH to user.
1478 *
1479 * LOCKING:
1480 * None.
1481 */
1482static void ata_eh_report(struct ata_port *ap)
1483{
1484 struct ata_eh_context *ehc = &ap->eh_context;
1485 const char *frozen, *desc;
1486 int tag, nr_failed = 0;
1487
1488 desc = NULL;
1489 if (ehc->i.desc[0] != '\0')
1490 desc = ehc->i.desc;
1491
1492 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1493 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1494
1495 if (!(qc->flags & ATA_QCFLAG_FAILED))
1496 continue;
1497 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1498 continue;
1499
1500 nr_failed++;
1501 }
1502
1503 if (!nr_failed && !ehc->i.err_mask)
1504 return;
1505
1506 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001507 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001508 frozen = " frozen";
1509
1510 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001511 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1512 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1513 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1514 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001515 if (desc)
1516 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1517 } else {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001518 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1519 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1520 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1521 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001522 if (desc)
1523 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1524 }
1525
1526 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
Tejun Heo8a937582006-11-14 22:36:12 +09001527 static const char *dma_str[] = {
1528 [DMA_BIDIRECTIONAL] = "bidi",
1529 [DMA_TO_DEVICE] = "out",
1530 [DMA_FROM_DEVICE] = "in",
1531 [DMA_NONE] = "",
1532 };
Tejun Heo022bdb02006-05-15 20:58:22 +09001533 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09001534 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001535
1536 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1537 continue;
1538
Tejun Heo8a937582006-11-14 22:36:12 +09001539 ata_dev_printk(qc->dev, KERN_ERR,
1540 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo664e8502006-11-20 16:05:34 +09001541 "tag %d cdb 0x%x data %u %s\n "
Tejun Heo8a937582006-11-14 22:36:12 +09001542 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1543 "Emask 0x%x (%s)\n",
1544 cmd->command, cmd->feature, cmd->nsect,
1545 cmd->lbal, cmd->lbam, cmd->lbah,
1546 cmd->hob_feature, cmd->hob_nsect,
1547 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heo726f0782007-01-03 17:30:39 +09001548 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
Tejun Heo664e8502006-11-20 16:05:34 +09001549 dma_str[qc->dma_dir],
Tejun Heo8a937582006-11-14 22:36:12 +09001550 res->command, res->feature, res->nsect,
1551 res->lbal, res->lbam, res->lbah,
1552 res->hob_feature, res->hob_nsect,
1553 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1554 res->device, qc->err_mask, ata_err_string(qc->err_mask));
Tejun Heo022bdb02006-05-15 20:58:22 +09001555 }
1556}
1557
Tejun Heod87fa382006-05-31 18:28:24 +09001558static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1559 unsigned int *classes)
1560{
1561 int i, rc;
1562
1563 for (i = 0; i < ATA_MAX_DEVICES; i++)
1564 classes[i] = ATA_DEV_UNKNOWN;
1565
1566 rc = reset(ap, classes);
1567 if (rc)
1568 return rc;
1569
1570 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1571 * is complete and convert all ATA_DEV_UNKNOWN to
1572 * ATA_DEV_NONE.
1573 */
1574 for (i = 0; i < ATA_MAX_DEVICES; i++)
1575 if (classes[i] != ATA_DEV_UNKNOWN)
1576 break;
1577
1578 if (i < ATA_MAX_DEVICES)
1579 for (i = 0; i < ATA_MAX_DEVICES; i++)
1580 if (classes[i] == ATA_DEV_UNKNOWN)
1581 classes[i] = ATA_DEV_NONE;
1582
1583 return 0;
1584}
1585
Tejun Heo664faf02006-05-31 18:27:50 +09001586static int ata_eh_followup_srst_needed(int rc, int classify,
1587 const unsigned int *classes)
1588{
1589 if (rc == -EAGAIN)
1590 return 1;
1591 if (rc != 0)
1592 return 0;
1593 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1594 return 1;
1595 return 0;
1596}
1597
1598static int ata_eh_reset(struct ata_port *ap, int classify,
Tejun Heof5914a42006-05-31 18:27:48 +09001599 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09001600 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1601{
1602 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09001603 unsigned int *classes = ehc->classes;
Tejun Heo022bdb02006-05-15 20:58:22 +09001604 int tries = ATA_EH_RESET_TRIES;
Tejun Heo1cdaf532006-07-03 16:07:26 +09001605 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heof5914a42006-05-31 18:27:48 +09001606 unsigned int action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001607 ata_reset_fn_t reset;
Tejun Heo664faf02006-05-31 18:27:50 +09001608 int i, did_followup_srst, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09001609
Tejun Heo13abf502006-07-10 23:18:46 +09001610 /* about to reset */
1611 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1612
Tejun Heof5914a42006-05-31 18:27:48 +09001613 /* Determine which reset to use and record in ehc->i.action.
1614 * prereset() may examine and modify it.
1615 */
1616 action = ehc->i.action;
1617 ehc->i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo022bdb02006-05-15 20:58:22 +09001618 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
Tejun Heof5914a42006-05-31 18:27:48 +09001619 !(action & ATA_EH_HARDRESET))))
1620 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001621 else
Tejun Heof5914a42006-05-31 18:27:48 +09001622 ehc->i.action |= ATA_EH_HARDRESET;
1623
1624 if (prereset) {
1625 rc = prereset(ap);
1626 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01001627 if (rc == -ENOENT) {
Tejun Heo4aa9ab62007-03-12 17:24:08 +09001628 ata_port_printk(ap, KERN_DEBUG,
1629 "port disabled. ignoring.\n");
Alan Coxc9619222006-09-26 17:53:38 +01001630 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09001631
1632 for (i = 0; i < ATA_MAX_DEVICES; i++)
1633 classes[i] = ATA_DEV_NONE;
1634
1635 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01001636 } else
1637 ata_port_printk(ap, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09001638 "prereset failed (errno=%d)\n", rc);
1639 return rc;
1640 }
1641 }
1642
1643 /* prereset() might have modified ehc->i.action */
1644 if (ehc->i.action & ATA_EH_HARDRESET)
Tejun Heo022bdb02006-05-15 20:58:22 +09001645 reset = hardreset;
Tejun Heof5914a42006-05-31 18:27:48 +09001646 else if (ehc->i.action & ATA_EH_SOFTRESET)
1647 reset = softreset;
1648 else {
1649 /* prereset told us not to reset, bang classes and return */
1650 for (i = 0; i < ATA_MAX_DEVICES; i++)
1651 classes[i] = ATA_DEV_NONE;
1652 return 0;
1653 }
1654
1655 /* did prereset() screw up? if so, fix up to avoid oopsing */
1656 if (!reset) {
1657 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1658 "invalid reset type\n");
1659 if (softreset)
1660 reset = softreset;
1661 else
1662 reset = hardreset;
1663 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001664
1665 retry:
Tejun Heo3e706392006-05-31 18:28:11 +09001666 /* shut up during boot probing */
1667 if (verbose)
1668 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1669 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09001670
Tejun Heo13abf502006-07-10 23:18:46 +09001671 /* mark that this EH session started with reset */
Tejun Heo022bdb02006-05-15 20:58:22 +09001672 ehc->i.flags |= ATA_EHI_DID_RESET;
1673
1674 rc = ata_do_reset(ap, reset, classes);
1675
Tejun Heo664faf02006-05-31 18:27:50 +09001676 did_followup_srst = 0;
1677 if (reset == hardreset &&
1678 ata_eh_followup_srst_needed(rc, classify, classes)) {
1679 /* okay, let's do follow-up softreset */
1680 did_followup_srst = 1;
1681 reset = softreset;
1682
1683 if (!reset) {
1684 ata_port_printk(ap, KERN_ERR,
1685 "follow-up softreset required "
1686 "but no softreset avaliable\n");
1687 return -EINVAL;
1688 }
1689
Tejun Heo47005f22006-06-19 18:27:23 +09001690 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
Tejun Heo664faf02006-05-31 18:27:50 +09001691 rc = ata_do_reset(ap, reset, classes);
1692
1693 if (rc == 0 && classify &&
1694 classes[0] == ATA_DEV_UNKNOWN) {
1695 ata_port_printk(ap, KERN_ERR,
1696 "classification failed\n");
1697 return -EINVAL;
1698 }
1699 }
1700
Tejun Heo022bdb02006-05-15 20:58:22 +09001701 if (rc && --tries) {
Tejun Heo664faf02006-05-31 18:27:50 +09001702 const char *type;
1703
1704 if (reset == softreset) {
1705 if (did_followup_srst)
1706 type = "follow-up soft";
1707 else
1708 type = "soft";
1709 } else
1710 type = "hard";
1711
Tejun Heo022bdb02006-05-15 20:58:22 +09001712 ata_port_printk(ap, KERN_WARNING,
Tejun Heo664faf02006-05-31 18:27:50 +09001713 "%sreset failed, retrying in 5 secs\n", type);
Tejun Heo022bdb02006-05-15 20:58:22 +09001714 ssleep(5);
1715
1716 if (reset == hardreset)
1717 sata_down_spd_limit(ap);
1718 if (hardreset)
1719 reset = hardreset;
1720 goto retry;
1721 }
1722
1723 if (rc == 0) {
Tejun Heo20952b62006-05-31 18:27:23 +09001724 /* After the reset, the device state is PIO 0 and the
1725 * controller state is undefined. Record the mode.
1726 */
1727 for (i = 0; i < ATA_MAX_DEVICES; i++)
1728 ap->device[i].pio_mode = XFER_PIO_0;
1729
Tejun Heo022bdb02006-05-15 20:58:22 +09001730 if (postreset)
1731 postreset(ap, classes);
1732
1733 /* reset successful, schedule revalidation */
Tejun Heo13abf502006-07-10 23:18:46 +09001734 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
Tejun Heo022bdb02006-05-15 20:58:22 +09001735 ehc->i.action |= ATA_EH_REVALIDATE;
1736 }
1737
1738 return rc;
1739}
1740
Tejun Heo084fe632006-05-31 18:28:03 +09001741static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1742 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001743{
1744 struct ata_eh_context *ehc = &ap->eh_context;
1745 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09001746 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09001747 int i, rc = 0;
1748
1749 DPRINTK("ENTER\n");
1750
1751 for (i = 0; i < ATA_MAX_DEVICES; i++) {
Tejun Heobff04642006-11-10 18:08:10 +09001752 unsigned int action, readid_flags = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001753
Tejun Heo47005f22006-06-19 18:27:23 +09001754 dev = &ap->device[i];
Tejun Heo64f65ca2006-06-24 20:30:18 +09001755 action = ata_eh_dev_action(dev);
Tejun Heo47005f22006-06-19 18:27:23 +09001756
Tejun Heobff04642006-11-10 18:08:10 +09001757 if (ehc->i.flags & ATA_EHI_DID_RESET)
1758 readid_flags |= ATA_READID_POSTRESET;
1759
Tejun Heo02670bf2006-07-03 16:07:26 +09001760 if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001761 if (ata_port_offline(ap)) {
1762 rc = -EIO;
1763 break;
1764 }
1765
Tejun Heo47005f22006-06-19 18:27:23 +09001766 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
Tejun Heobff04642006-11-10 18:08:10 +09001767 rc = ata_dev_revalidate(dev, readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001768 if (rc)
1769 break;
1770
Tejun Heo47005f22006-06-19 18:27:23 +09001771 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1772
Tejun Heobaa1e782006-11-01 18:39:27 +09001773 /* Configuration may have changed, reconfigure
1774 * transfer mode.
1775 */
1776 ehc->i.flags |= ATA_EHI_SETMODE;
1777
zhao, forrest3057ac32006-06-12 12:01:34 +08001778 /* schedule the scsi_rescan_device() here */
1779 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09001780 } else if (dev->class == ATA_DEV_UNKNOWN &&
1781 ehc->tries[dev->devno] &&
1782 ata_class_enabled(ehc->classes[dev->devno])) {
1783 dev->class = ehc->classes[dev->devno];
1784
Tejun Heobff04642006-11-10 18:08:10 +09001785 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1786 dev->id);
Tejun Heoefdaedc2006-11-01 18:38:52 +09001787 if (rc == 0) {
1788 ehc->i.flags |= ATA_EHI_PRINTINFO;
1789 rc = ata_dev_configure(dev);
1790 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001791 } else if (rc == -ENOENT) {
1792 /* IDENTIFY was issued to non-existent
1793 * device. No need to reset. Just
1794 * thaw and kill the device.
1795 */
1796 ata_eh_thaw_port(ap);
1797 dev->class = ATA_DEV_UNKNOWN;
1798 rc = 0;
Tejun Heoefdaedc2006-11-01 18:38:52 +09001799 }
Tejun Heo084fe632006-05-31 18:28:03 +09001800
1801 if (rc) {
1802 dev->class = ATA_DEV_UNKNOWN;
1803 break;
1804 }
1805
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001806 if (ata_dev_enabled(dev)) {
1807 spin_lock_irqsave(ap->lock, flags);
1808 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1809 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heobaa1e782006-11-01 18:39:27 +09001810
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001811 /* new device discovered, configure xfermode */
1812 ehc->i.flags |= ATA_EHI_SETMODE;
1813 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001814 }
1815 }
1816
Tejun Heo47005f22006-06-19 18:27:23 +09001817 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09001818 *r_failed_dev = dev;
1819
1820 DPRINTK("EXIT\n");
1821 return rc;
1822}
1823
Tejun Heo6ffa01d2007-03-02 17:32:47 +09001824#ifdef CONFIG_PM
Tejun Heo02670bf2006-07-03 16:07:26 +09001825/**
1826 * ata_eh_suspend - handle suspend EH action
1827 * @ap: target host port
1828 * @r_failed_dev: result parameter to indicate failing device
1829 *
1830 * Handle suspend EH action. Disk devices are spinned down and
1831 * other types of devices are just marked suspended. Once
1832 * suspended, no EH action to the device is allowed until it is
1833 * resumed.
1834 *
1835 * LOCKING:
1836 * Kernel thread context (may sleep).
1837 *
1838 * RETURNS:
1839 * 0 on success, -errno otherwise
1840 */
1841static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev)
1842{
1843 struct ata_device *dev;
1844 int i, rc = 0;
1845
1846 DPRINTK("ENTER\n");
1847
1848 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1849 unsigned long flags;
1850 unsigned int action, err_mask;
1851
1852 dev = &ap->device[i];
1853 action = ata_eh_dev_action(dev);
1854
1855 if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND))
1856 continue;
1857
1858 WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED);
1859
1860 ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND);
1861
1862 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1863 /* flush cache */
1864 rc = ata_flush_cache(dev);
1865 if (rc)
1866 break;
1867
1868 /* spin down */
1869 err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1);
1870 if (err_mask) {
1871 ata_dev_printk(dev, KERN_ERR, "failed to "
1872 "spin down (err_mask=0x%x)\n",
1873 err_mask);
1874 rc = -EIO;
1875 break;
1876 }
1877 }
1878
1879 spin_lock_irqsave(ap->lock, flags);
1880 dev->flags |= ATA_DFLAG_SUSPENDED;
1881 spin_unlock_irqrestore(ap->lock, flags);
1882
1883 ata_eh_done(ap, dev, ATA_EH_SUSPEND);
1884 }
1885
1886 if (rc)
1887 *r_failed_dev = dev;
1888
1889 DPRINTK("EXIT\n");
Tejun Heo03ee5b12007-01-26 20:10:25 +09001890 return rc;
Tejun Heo02670bf2006-07-03 16:07:26 +09001891}
1892
1893/**
1894 * ata_eh_prep_resume - prep for resume EH action
1895 * @ap: target host port
1896 *
1897 * Clear SUSPENDED in preparation for scheduled resume actions.
1898 * This allows other parts of EH to access the devices being
1899 * resumed.
1900 *
1901 * LOCKING:
1902 * Kernel thread context (may sleep).
1903 */
1904static void ata_eh_prep_resume(struct ata_port *ap)
1905{
1906 struct ata_device *dev;
1907 unsigned long flags;
1908 int i;
1909
1910 DPRINTK("ENTER\n");
1911
1912 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1913 unsigned int action;
1914
1915 dev = &ap->device[i];
1916 action = ata_eh_dev_action(dev);
1917
1918 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1919 continue;
1920
1921 spin_lock_irqsave(ap->lock, flags);
1922 dev->flags &= ~ATA_DFLAG_SUSPENDED;
1923 spin_unlock_irqrestore(ap->lock, flags);
1924 }
1925
1926 DPRINTK("EXIT\n");
1927}
1928
1929/**
1930 * ata_eh_resume - handle resume EH action
1931 * @ap: target host port
1932 * @r_failed_dev: result parameter to indicate failing device
1933 *
1934 * Handle resume EH action. Target devices are already reset and
1935 * revalidated. Spinning up is the only operation left.
1936 *
1937 * LOCKING:
1938 * Kernel thread context (may sleep).
1939 *
1940 * RETURNS:
1941 * 0 on success, -errno otherwise
1942 */
1943static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev)
1944{
1945 struct ata_device *dev;
1946 int i, rc = 0;
1947
1948 DPRINTK("ENTER\n");
1949
1950 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1951 unsigned int action, err_mask;
1952
1953 dev = &ap->device[i];
1954 action = ata_eh_dev_action(dev);
1955
1956 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1957 continue;
1958
1959 ata_eh_about_to_do(ap, dev, ATA_EH_RESUME);
1960
1961 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1962 err_mask = ata_do_simple_cmd(dev,
1963 ATA_CMD_IDLEIMMEDIATE);
1964 if (err_mask) {
1965 ata_dev_printk(dev, KERN_ERR, "failed to "
1966 "spin up (err_mask=0x%x)\n",
1967 err_mask);
1968 rc = -EIO;
1969 break;
1970 }
1971 }
1972
1973 ata_eh_done(ap, dev, ATA_EH_RESUME);
1974 }
1975
1976 if (rc)
1977 *r_failed_dev = dev;
1978
1979 DPRINTK("EXIT\n");
1980 return 0;
1981}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09001982#endif /* CONFIG_PM */
Tejun Heo02670bf2006-07-03 16:07:26 +09001983
Tejun Heo022bdb02006-05-15 20:58:22 +09001984static int ata_port_nr_enabled(struct ata_port *ap)
1985{
1986 int i, cnt = 0;
1987
1988 for (i = 0; i < ATA_MAX_DEVICES; i++)
1989 if (ata_dev_enabled(&ap->device[i]))
1990 cnt++;
1991 return cnt;
1992}
1993
Tejun Heo084fe632006-05-31 18:28:03 +09001994static int ata_port_nr_vacant(struct ata_port *ap)
1995{
1996 int i, cnt = 0;
1997
1998 for (i = 0; i < ATA_MAX_DEVICES; i++)
1999 if (ap->device[i].class == ATA_DEV_UNKNOWN)
2000 cnt++;
2001 return cnt;
2002}
2003
2004static int ata_eh_skip_recovery(struct ata_port *ap)
2005{
2006 struct ata_eh_context *ehc = &ap->eh_context;
2007 int i;
2008
Tejun Heo02670bf2006-07-03 16:07:26 +09002009 /* skip if all possible devices are suspended */
2010 for (i = 0; i < ata_port_max_devices(ap); i++) {
2011 struct ata_device *dev = &ap->device[i];
2012
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09002013 if (!(dev->flags & ATA_DFLAG_SUSPENDED))
Tejun Heo02670bf2006-07-03 16:07:26 +09002014 break;
2015 }
2016
2017 if (i == ata_port_max_devices(ap))
2018 return 1;
2019
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09002020 /* thaw frozen port, resume link and recover failed devices */
2021 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
2022 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
Tejun Heo084fe632006-05-31 18:28:03 +09002023 return 0;
2024
2025 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2026 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2027 struct ata_device *dev = &ap->device[i];
2028
2029 if (dev->class == ATA_DEV_UNKNOWN &&
2030 ehc->classes[dev->devno] != ATA_DEV_NONE)
2031 return 0;
2032 }
2033
2034 return 1;
2035}
2036
Tejun Heo022bdb02006-05-15 20:58:22 +09002037/**
2038 * ata_eh_recover - recover host port after error
2039 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002040 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002041 * @softreset: softreset method (can be NULL)
2042 * @hardreset: hardreset method (can be NULL)
2043 * @postreset: postreset method (can be NULL)
2044 *
2045 * This is the alpha and omega, eum and yang, heart and soul of
2046 * libata exception handling. On entry, actions required to
Tejun Heo084fe632006-05-31 18:28:03 +09002047 * recover the port and hotplug requests are recorded in
2048 * eh_context. This function executes all the operations with
2049 * appropriate retrials and fallbacks to resurrect failed
2050 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09002051 *
2052 * LOCKING:
2053 * Kernel thread context (may sleep).
2054 *
2055 * RETURNS:
2056 * 0 on success, -errno on failure.
2057 */
Tejun Heof5914a42006-05-31 18:27:48 +09002058static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2059 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09002060 ata_postreset_fn_t postreset)
2061{
2062 struct ata_eh_context *ehc = &ap->eh_context;
2063 struct ata_device *dev;
Tejun Heo4ae72a12007-02-02 16:22:30 +09002064 int i, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002065
2066 DPRINTK("ENTER\n");
2067
2068 /* prep for recovery */
2069 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2070 dev = &ap->device[i];
2071
2072 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo084fe632006-05-31 18:28:03 +09002073
Tejun Heo79a55b72007-01-18 17:22:18 +09002074 /* collect port action mask recorded in dev actions */
2075 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
2076 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
2077
Tejun Heo084fe632006-05-31 18:28:03 +09002078 /* process hotplug request */
2079 if (dev->flags & ATA_DFLAG_DETACH)
2080 ata_eh_detach_dev(dev);
2081
2082 if (!ata_dev_enabled(dev) &&
2083 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2084 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2085 ata_eh_detach_dev(dev);
2086 ata_dev_init(dev);
2087 ehc->did_probe_mask |= (1 << dev->devno);
2088 ehc->i.action |= ATA_EH_SOFTRESET;
2089 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002090 }
2091
2092 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09002093 rc = 0;
2094
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002095 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09002096 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002097 goto out;
2098
Tejun Heo02670bf2006-07-03 16:07:26 +09002099 /* prep for resume */
2100 ata_eh_prep_resume(ap);
2101
Tejun Heo022bdb02006-05-15 20:58:22 +09002102 /* skip EH if possible. */
Tejun Heo084fe632006-05-31 18:28:03 +09002103 if (ata_eh_skip_recovery(ap))
Tejun Heo022bdb02006-05-15 20:58:22 +09002104 ehc->i.action = 0;
2105
Tejun Heo084fe632006-05-31 18:28:03 +09002106 for (i = 0; i < ATA_MAX_DEVICES; i++)
2107 ehc->classes[i] = ATA_DEV_UNKNOWN;
2108
Tejun Heo022bdb02006-05-15 20:58:22 +09002109 /* reset */
2110 if (ehc->i.action & ATA_EH_RESET_MASK) {
2111 ata_eh_freeze_port(ap);
2112
Tejun Heo084fe632006-05-31 18:28:03 +09002113 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2114 softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002115 if (rc) {
2116 ata_port_printk(ap, KERN_ERR,
2117 "reset failed, giving up\n");
2118 goto out;
2119 }
2120
2121 ata_eh_thaw_port(ap);
2122 }
2123
Tejun Heo084fe632006-05-31 18:28:03 +09002124 /* revalidate existing devices and attach new ones */
2125 rc = ata_eh_revalidate_and_attach(ap, &dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09002126 if (rc)
2127 goto dev_fail;
2128
Tejun Heo02670bf2006-07-03 16:07:26 +09002129 /* resume devices */
2130 rc = ata_eh_resume(ap, &dev);
2131 if (rc)
2132 goto dev_fail;
2133
Tejun Heobaa1e782006-11-01 18:39:27 +09002134 /* configure transfer mode if necessary */
2135 if (ehc->i.flags & ATA_EHI_SETMODE) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002136 rc = ata_set_mode(ap, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09002137 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09002138 goto dev_fail;
Tejun Heobaa1e782006-11-01 18:39:27 +09002139 ehc->i.flags &= ~ATA_EHI_SETMODE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002140 }
2141
Tejun Heo02670bf2006-07-03 16:07:26 +09002142 /* suspend devices */
2143 rc = ata_eh_suspend(ap, &dev);
2144 if (rc)
2145 goto dev_fail;
2146
Tejun Heo022bdb02006-05-15 20:58:22 +09002147 goto out;
2148
2149 dev_fail:
Tejun Heo4ae72a12007-02-02 16:22:30 +09002150 ehc->tries[dev->devno]--;
2151
Tejun Heo022bdb02006-05-15 20:58:22 +09002152 switch (rc) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002153 case -EINVAL:
Tejun Heo4ae72a12007-02-02 16:22:30 +09002154 /* eeek, something went very wrong, give up */
Tejun Heo022bdb02006-05-15 20:58:22 +09002155 ehc->tries[dev->devno] = 0;
2156 break;
Tejun Heo4ae72a12007-02-02 16:22:30 +09002157
2158 case -ENODEV:
2159 /* device missing or wrong IDENTIFY data, schedule probing */
2160 ehc->i.probe_mask |= (1 << dev->devno);
2161 /* give it just one more chance */
2162 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
Tejun Heo022bdb02006-05-15 20:58:22 +09002163 case -EIO:
Tejun Heo4ae72a12007-02-02 16:22:30 +09002164 if (ehc->tries[dev->devno] == 1) {
2165 /* This is the last chance, better to slow
2166 * down than lose it.
2167 */
2168 sata_down_spd_limit(ap);
2169 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2170 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002171 }
2172
Tejun Heo084fe632006-05-31 18:28:03 +09002173 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2174 /* disable device if it has used up all its chances */
Tejun Heo022bdb02006-05-15 20:58:22 +09002175 ata_dev_disable(dev);
2176
Tejun Heo084fe632006-05-31 18:28:03 +09002177 /* detach if offline */
2178 if (ata_port_offline(ap))
2179 ata_eh_detach_dev(dev);
2180
2181 /* probe if requested */
2182 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2183 !(ehc->did_probe_mask & (1 << dev->devno))) {
2184 ata_eh_detach_dev(dev);
2185 ata_dev_init(dev);
2186
2187 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2188 ehc->did_probe_mask |= (1 << dev->devno);
2189 ehc->i.action |= ATA_EH_SOFTRESET;
2190 }
2191 } else {
2192 /* soft didn't work? be haaaaard */
2193 if (ehc->i.flags & ATA_EHI_DID_RESET)
2194 ehc->i.action |= ATA_EH_HARDRESET;
2195 else
2196 ehc->i.action |= ATA_EH_SOFTRESET;
2197 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002198
2199 if (ata_port_nr_enabled(ap)) {
2200 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2201 "devices, retrying in 5 secs\n");
2202 ssleep(5);
2203 } else {
2204 /* no device left, repeat fast */
2205 msleep(500);
2206 }
2207
2208 goto retry;
2209
2210 out:
2211 if (rc) {
2212 for (i = 0; i < ATA_MAX_DEVICES; i++)
2213 ata_dev_disable(&ap->device[i]);
2214 }
2215
2216 DPRINTK("EXIT, rc=%d\n", rc);
2217 return rc;
2218}
2219
2220/**
2221 * ata_eh_finish - finish up EH
2222 * @ap: host port to finish EH for
2223 *
2224 * Recovery is complete. Clean up EH states and retry or finish
2225 * failed qcs.
2226 *
2227 * LOCKING:
2228 * None.
2229 */
2230static void ata_eh_finish(struct ata_port *ap)
2231{
2232 int tag;
2233
2234 /* retry or finish qcs */
2235 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2236 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2237
2238 if (!(qc->flags & ATA_QCFLAG_FAILED))
2239 continue;
2240
2241 if (qc->err_mask) {
2242 /* FIXME: Once EH migration is complete,
2243 * generate sense data in this function,
2244 * considering both err_mask and tf.
2245 */
2246 if (qc->err_mask & AC_ERR_INVALID)
2247 ata_eh_qc_complete(qc);
2248 else
2249 ata_eh_qc_retry(qc);
2250 } else {
2251 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2252 ata_eh_qc_complete(qc);
2253 } else {
2254 /* feed zero TF to sense generation */
2255 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2256 ata_eh_qc_retry(qc);
2257 }
2258 }
2259 }
2260}
2261
2262/**
2263 * ata_do_eh - do standard error handling
2264 * @ap: host port to handle error for
Tejun Heof5914a42006-05-31 18:27:48 +09002265 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002266 * @softreset: softreset method (can be NULL)
2267 * @hardreset: hardreset method (can be NULL)
2268 * @postreset: postreset method (can be NULL)
2269 *
2270 * Perform standard error handling sequence.
2271 *
2272 * LOCKING:
2273 * Kernel thread context (may sleep).
2274 */
Tejun Heof5914a42006-05-31 18:27:48 +09002275void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2276 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2277 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002278{
Tejun Heo1cdaf532006-07-03 16:07:26 +09002279 ata_eh_autopsy(ap);
2280 ata_eh_report(ap);
Tejun Heof5914a42006-05-31 18:27:48 +09002281 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002282 ata_eh_finish(ap);
2283}
Tejun Heo500530f2006-07-03 16:07:27 +09002284
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002285#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09002286/**
2287 * ata_eh_handle_port_suspend - perform port suspend operation
2288 * @ap: port to suspend
2289 *
2290 * Suspend @ap.
2291 *
2292 * LOCKING:
2293 * Kernel thread context (may sleep).
2294 */
2295static void ata_eh_handle_port_suspend(struct ata_port *ap)
2296{
2297 unsigned long flags;
2298 int rc = 0;
2299
2300 /* are we suspending? */
2301 spin_lock_irqsave(ap->lock, flags);
2302 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2303 ap->pm_mesg.event == PM_EVENT_ON) {
2304 spin_unlock_irqrestore(ap->lock, flags);
2305 return;
2306 }
2307 spin_unlock_irqrestore(ap->lock, flags);
2308
2309 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2310
2311 /* suspend */
2312 ata_eh_freeze_port(ap);
2313
2314 if (ap->ops->port_suspend)
2315 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2316
2317 /* report result */
2318 spin_lock_irqsave(ap->lock, flags);
2319
2320 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2321 if (rc == 0)
2322 ap->pflags |= ATA_PFLAG_SUSPENDED;
2323 else
2324 ata_port_schedule_eh(ap);
2325
2326 if (ap->pm_result) {
2327 *ap->pm_result = rc;
2328 ap->pm_result = NULL;
2329 }
2330
2331 spin_unlock_irqrestore(ap->lock, flags);
2332
2333 return;
2334}
2335
2336/**
2337 * ata_eh_handle_port_resume - perform port resume operation
2338 * @ap: port to resume
2339 *
2340 * Resume @ap.
2341 *
2342 * This function also waits upto one second until all devices
2343 * hanging off this port requests resume EH action. This is to
2344 * prevent invoking EH and thus reset multiple times on resume.
2345 *
2346 * On DPM resume, where some of devices might not be resumed
2347 * together, this may delay port resume upto one second, but such
2348 * DPM resumes are rare and 1 sec delay isn't too bad.
2349 *
2350 * LOCKING:
2351 * Kernel thread context (may sleep).
2352 */
2353static void ata_eh_handle_port_resume(struct ata_port *ap)
2354{
2355 unsigned long timeout;
2356 unsigned long flags;
2357 int i, rc = 0;
2358
2359 /* are we resuming? */
2360 spin_lock_irqsave(ap->lock, flags);
2361 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2362 ap->pm_mesg.event != PM_EVENT_ON) {
2363 spin_unlock_irqrestore(ap->lock, flags);
2364 return;
2365 }
2366 spin_unlock_irqrestore(ap->lock, flags);
2367
2368 /* spurious? */
2369 if (!(ap->pflags & ATA_PFLAG_SUSPENDED))
2370 goto done;
2371
2372 if (ap->ops->port_resume)
2373 rc = ap->ops->port_resume(ap);
2374
2375 /* give devices time to request EH */
2376 timeout = jiffies + HZ; /* 1s max */
2377 while (1) {
2378 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2379 struct ata_device *dev = &ap->device[i];
2380 unsigned int action = ata_eh_dev_action(dev);
2381
2382 if ((dev->flags & ATA_DFLAG_SUSPENDED) &&
2383 !(action & ATA_EH_RESUME))
2384 break;
2385 }
2386
2387 if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout))
2388 break;
2389 msleep(10);
2390 }
2391
2392 done:
2393 spin_lock_irqsave(ap->lock, flags);
2394 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2395 if (ap->pm_result) {
2396 *ap->pm_result = rc;
2397 ap->pm_result = NULL;
2398 }
2399 spin_unlock_irqrestore(ap->lock, flags);
2400}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002401#endif /* CONFIG_PM */