blob: 2bff9adcacf15de02b7f0f6c35f83116410df606 [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 */
Albert Lee56287762007-04-02 11:30:46 +0800985static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
Tejun Heo022bdb02006-05-15 20:58:22 +0900986{
Albert Lee56287762007-04-02 11:30:46 +0800987 struct ata_device *dev = qc->dev;
988 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
Tejun Heo022bdb02006-05-15 20:58:22 +0900989 struct ata_port *ap = dev->ap;
990 struct ata_taskfile tf;
991 u8 cdb[ATAPI_CDB_LEN];
992
993 DPRINTK("ATAPI request sense\n");
994
Tejun Heo022bdb02006-05-15 20:58:22 +0900995 /* FIXME: is this needed? */
996 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
997
Albert Lee56287762007-04-02 11:30:46 +0800998 /* initialize sense_buf with the error register,
999 * for the case where they are -not- overwritten
1000 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001001 sense_buf[0] = 0x70;
Albert Lee56287762007-04-02 11:30:46 +08001002 sense_buf[2] = qc->result_tf.feature >> 4;
1003
1004 /* some devices time out if garbage left in tf */
1005 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001006
1007 memset(cdb, 0, ATAPI_CDB_LEN);
1008 cdb[0] = REQUEST_SENSE;
1009 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1010
1011 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1012 tf.command = ATA_CMD_PACKET;
1013
1014 /* is it pointless to prefer PIO for "safety reasons"? */
1015 if (ap->flags & ATA_FLAG_PIO_DMA) {
1016 tf.protocol = ATA_PROT_ATAPI_DMA;
1017 tf.feature |= ATAPI_PKT_DMA;
1018 } else {
1019 tf.protocol = ATA_PROT_ATAPI;
1020 tf.lbam = (8 * 1024) & 0xff;
1021 tf.lbah = (8 * 1024) >> 8;
1022 }
1023
1024 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1025 sense_buf, SCSI_SENSE_BUFFERSIZE);
1026}
1027
1028/**
1029 * ata_eh_analyze_serror - analyze SError for a failed port
1030 * @ap: ATA port to analyze SError for
1031 *
1032 * Analyze SError if available and further determine cause of
1033 * failure.
1034 *
1035 * LOCKING:
1036 * None.
1037 */
1038static void ata_eh_analyze_serror(struct ata_port *ap)
1039{
1040 struct ata_eh_context *ehc = &ap->eh_context;
1041 u32 serror = ehc->i.serror;
1042 unsigned int err_mask = 0, action = 0;
1043
1044 if (serror & SERR_PERSISTENT) {
1045 err_mask |= AC_ERR_ATA_BUS;
1046 action |= ATA_EH_HARDRESET;
1047 }
1048 if (serror &
1049 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1050 err_mask |= AC_ERR_ATA_BUS;
1051 action |= ATA_EH_SOFTRESET;
1052 }
1053 if (serror & SERR_PROTOCOL) {
1054 err_mask |= AC_ERR_HSM;
1055 action |= ATA_EH_SOFTRESET;
1056 }
1057 if (serror & SERR_INTERNAL) {
1058 err_mask |= AC_ERR_SYSTEM;
Tejun Heo771b8da2007-03-14 01:20:51 +09001059 action |= ATA_EH_HARDRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001060 }
Tejun Heo084fe632006-05-31 18:28:03 +09001061 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1062 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001063
1064 ehc->i.err_mask |= err_mask;
1065 ehc->i.action |= action;
1066}
1067
1068/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001069 * ata_eh_analyze_ncq_error - analyze NCQ error
1070 * @ap: ATA port to analyze NCQ error for
1071 *
1072 * Read log page 10h, determine the offending qc and acquire
1073 * error status TF. For NCQ device errors, all LLDDs have to do
1074 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1075 * care of the rest.
1076 *
1077 * LOCKING:
1078 * Kernel thread context (may sleep).
1079 */
1080static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1081{
1082 struct ata_eh_context *ehc = &ap->eh_context;
1083 struct ata_device *dev = ap->device;
1084 struct ata_queued_cmd *qc;
1085 struct ata_taskfile tf;
1086 int tag, rc;
1087
1088 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001089 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001090 return;
1091
1092 /* is it NCQ device error? */
1093 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1094 return;
1095
1096 /* has LLDD analyzed already? */
1097 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1098 qc = __ata_qc_from_tag(ap, tag);
1099
1100 if (!(qc->flags & ATA_QCFLAG_FAILED))
1101 continue;
1102
1103 if (qc->err_mask)
1104 return;
1105 }
1106
1107 /* okay, this error is ours */
1108 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1109 if (rc) {
1110 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1111 "(errno=%d)\n", rc);
1112 return;
1113 }
1114
1115 if (!(ap->sactive & (1 << tag))) {
1116 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1117 "inactive tag %d\n", tag);
1118 return;
1119 }
1120
1121 /* we've got the perpetrator, condemn it */
1122 qc = __ata_qc_from_tag(ap, tag);
1123 memcpy(&qc->result_tf, &tf, sizeof(tf));
1124 qc->err_mask |= AC_ERR_DEV;
1125 ehc->i.err_mask &= ~AC_ERR_DEV;
1126}
1127
1128/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001129 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1130 * @qc: qc to analyze
1131 * @tf: Taskfile registers to analyze
1132 *
1133 * Analyze taskfile of @qc and further determine cause of
1134 * failure. This function also requests ATAPI sense data if
1135 * avaliable.
1136 *
1137 * LOCKING:
1138 * Kernel thread context (may sleep).
1139 *
1140 * RETURNS:
1141 * Determined recovery action
1142 */
1143static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1144 const struct ata_taskfile *tf)
1145{
1146 unsigned int tmp, action = 0;
1147 u8 stat = tf->command, err = tf->feature;
1148
1149 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1150 qc->err_mask |= AC_ERR_HSM;
1151 return ATA_EH_SOFTRESET;
1152 }
1153
Tejun Heoa51d6442007-03-20 15:24:11 +09001154 if (stat & (ATA_ERR | ATA_DF))
1155 qc->err_mask |= AC_ERR_DEV;
1156 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001157 return 0;
1158
1159 switch (qc->dev->class) {
1160 case ATA_DEV_ATA:
1161 if (err & ATA_ICRC)
1162 qc->err_mask |= AC_ERR_ATA_BUS;
1163 if (err & ATA_UNC)
1164 qc->err_mask |= AC_ERR_MEDIA;
1165 if (err & ATA_IDNF)
1166 qc->err_mask |= AC_ERR_INVALID;
1167 break;
1168
1169 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001170 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Albert Lee56287762007-04-02 11:30:46 +08001171 tmp = atapi_eh_request_sense(qc);
Tejun Heoa569a302006-11-21 10:40:51 +09001172 if (!tmp) {
1173 /* ATA_QCFLAG_SENSE_VALID is used to
1174 * tell atapi_qc_complete() that sense
1175 * data is already valid.
1176 *
1177 * TODO: interpret sense data and set
1178 * appropriate err_mask.
1179 */
1180 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1181 } else
1182 qc->err_mask |= tmp;
1183 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001184 }
1185
1186 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1187 action |= ATA_EH_SOFTRESET;
1188
1189 return action;
1190}
1191
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001192static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001193{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001194 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo022bdb02006-05-15 20:58:22 +09001195 return 1;
1196
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001197 if (err_mask & AC_ERR_TIMEOUT)
1198 return 2;
1199
1200 if (is_io) {
1201 if (err_mask & AC_ERR_HSM)
Tejun Heo022bdb02006-05-15 20:58:22 +09001202 return 2;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001203 if ((err_mask &
1204 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1205 return 3;
Tejun Heo022bdb02006-05-15 20:58:22 +09001206 }
1207
1208 return 0;
1209}
1210
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001211struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001212 u64 since;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001213 int nr_errors[4];
Tejun Heo022bdb02006-05-15 20:58:22 +09001214};
1215
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001216static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001217{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001218 struct speed_down_verdict_arg *arg = void_arg;
1219 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
Tejun Heo022bdb02006-05-15 20:58:22 +09001220
1221 if (ent->timestamp < arg->since)
1222 return -1;
1223
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001224 arg->nr_errors[cat]++;
Tejun Heo022bdb02006-05-15 20:58:22 +09001225 return 0;
1226}
1227
1228/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001229 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001230 * @dev: Device of interest
1231 *
1232 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001233 * whether NCQ needs to be turned off, transfer speed should be
1234 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001235 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001236 * Cat-1 is ATA_BUS error for any command.
Tejun Heo022bdb02006-05-15 20:58:22 +09001237 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001238 * Cat-2 is TIMEOUT for any command or HSM violation for known
1239 * supported commands.
1240 *
1241 * Cat-3 is is unclassified DEV error for known supported
Tejun Heo022bdb02006-05-15 20:58:22 +09001242 * command.
1243 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001244 * NCQ needs to be turned off if there have been more than 3
1245 * Cat-2 + Cat-3 errors during last 10 minutes.
1246 *
1247 * Speed down is necessary if there have been more than 3 Cat-1 +
1248 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1249 *
1250 * Falling back to PIO mode is necessary if there have been more
1251 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1252 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001253 * LOCKING:
1254 * Inherited from caller.
1255 *
1256 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001257 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001258 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001259static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001260{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001261 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1262 u64 j64 = get_jiffies_64();
1263 struct speed_down_verdict_arg arg;
1264 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001265
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001266 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001267 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001268 arg.since = j64 - min(j64, j10mins);
1269 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001270
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001271 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1272 verdict |= ATA_EH_SPDN_NCQ_OFF;
1273 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1274 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001275
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001276 /* scan past 3 mins of error history */
1277 memset(&arg, 0, sizeof(arg));
1278 arg.since = j64 - min(j64, j5mins);
1279 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001280
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001281 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1282 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1283
1284 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001285}
1286
1287/**
1288 * ata_eh_speed_down - record error and speed down if necessary
1289 * @dev: Failed device
1290 * @is_io: Did the device fail during normal IO?
1291 * @err_mask: err_mask of the error
1292 *
1293 * Record error and examine error history to determine whether
1294 * adjusting transmission speed is necessary. It also sets
1295 * transmission limits appropriately if such adjustment is
1296 * necessary.
1297 *
1298 * LOCKING:
1299 * Kernel thread context (may sleep).
1300 *
1301 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001302 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001303 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001304static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1305 unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001306{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001307 unsigned int verdict;
1308 unsigned int action = 0;
1309
1310 /* don't bother if Cat-0 error */
1311 if (ata_eh_categorize_error(is_io, err_mask) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001312 return 0;
1313
1314 /* record error and determine whether speed down is necessary */
1315 ata_ering_record(&dev->ering, is_io, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001316 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001317
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001318 /* turn off NCQ? */
1319 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1320 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1321 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1322 dev->flags |= ATA_DFLAG_NCQ_OFF;
1323 ata_dev_printk(dev, KERN_WARNING,
1324 "NCQ disabled due to excessive errors\n");
1325 goto done;
1326 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001327
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001328 /* speed down? */
1329 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1330 /* speed down SATA link speed if possible */
1331 if (sata_down_spd_limit(dev->ap) == 0) {
1332 action |= ATA_EH_HARDRESET;
1333 goto done;
1334 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001335
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001336 /* lower transfer mode */
1337 if (dev->spdn_cnt < 2) {
1338 static const int dma_dnxfer_sel[] =
1339 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1340 static const int pio_dnxfer_sel[] =
1341 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1342 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001343
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001344 if (dev->xfer_shift != ATA_SHIFT_PIO)
1345 sel = dma_dnxfer_sel[dev->spdn_cnt];
1346 else
1347 sel = pio_dnxfer_sel[dev->spdn_cnt];
1348
1349 dev->spdn_cnt++;
1350
1351 if (ata_down_xfermask_limit(dev, sel) == 0) {
1352 action |= ATA_EH_SOFTRESET;
1353 goto done;
1354 }
1355 }
1356 }
1357
1358 /* Fall back to PIO? Slowing down to PIO is meaningless for
1359 * SATA. Consider it only for PATA.
1360 */
1361 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1362 (dev->ap->cbl != ATA_CBL_SATA) &&
1363 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1364 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1365 dev->spdn_cnt = 0;
1366 action |= ATA_EH_SOFTRESET;
1367 goto done;
1368 }
1369 }
1370
Tejun Heo022bdb02006-05-15 20:58:22 +09001371 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001372 done:
1373 /* device has been slowed down, blow error history */
1374 ata_ering_clear(&dev->ering);
1375 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001376}
1377
1378/**
1379 * ata_eh_autopsy - analyze error and determine recovery action
1380 * @ap: ATA port to perform autopsy on
1381 *
1382 * Analyze why @ap failed and determine which recovery action is
1383 * needed. This function also sets more detailed AC_ERR_* values
1384 * and fills sense data for ATAPI CHECK SENSE.
1385 *
1386 * LOCKING:
1387 * Kernel thread context (may sleep).
1388 */
1389static void ata_eh_autopsy(struct ata_port *ap)
1390{
1391 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001392 unsigned int all_err_mask = 0;
1393 int tag, is_io = 0;
1394 u32 serror;
1395 int rc;
1396
1397 DPRINTK("ENTER\n");
1398
Tejun Heo1cdaf532006-07-03 16:07:26 +09001399 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1400 return;
1401
Tejun Heo022bdb02006-05-15 20:58:22 +09001402 /* obtain and analyze SError */
1403 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1404 if (rc == 0) {
1405 ehc->i.serror |= serror;
1406 ata_eh_analyze_serror(ap);
1407 } else if (rc != -EOPNOTSUPP)
Tejun Heo4528e4d2006-07-08 20:17:26 +09001408 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001409
Tejun Heoe8ee8452006-05-15 21:03:46 +09001410 /* analyze NCQ failure */
1411 ata_eh_analyze_ncq_error(ap);
1412
Tejun Heo022bdb02006-05-15 20:58:22 +09001413 /* any real error trumps AC_ERR_OTHER */
1414 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1415 ehc->i.err_mask &= ~AC_ERR_OTHER;
1416
1417 all_err_mask |= ehc->i.err_mask;
1418
1419 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1420 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1421
1422 if (!(qc->flags & ATA_QCFLAG_FAILED))
1423 continue;
1424
1425 /* inherit upper level err_mask */
1426 qc->err_mask |= ehc->i.err_mask;
1427
Tejun Heo022bdb02006-05-15 20:58:22 +09001428 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001429 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001430
1431 /* DEV errors are probably spurious in case of ATA_BUS error */
1432 if (qc->err_mask & AC_ERR_ATA_BUS)
1433 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1434 AC_ERR_INVALID);
1435
1436 /* any real error trumps unknown error */
1437 if (qc->err_mask & ~AC_ERR_OTHER)
1438 qc->err_mask &= ~AC_ERR_OTHER;
1439
1440 /* SENSE_VALID trumps dev/unknown error and revalidation */
1441 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1442 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo4528e4d2006-07-08 20:17:26 +09001443 ehc->i.action &= ~ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001444 }
1445
1446 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001447 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001448 all_err_mask |= qc->err_mask;
1449 if (qc->flags & ATA_QCFLAG_IO)
1450 is_io = 1;
1451 }
1452
Tejun Heoa20f33f2006-05-16 12:58:24 +09001453 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001454 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001455 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001456 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heoa20f33f2006-05-16 12:58:24 +09001457 else if (all_err_mask)
Tejun Heo4528e4d2006-07-08 20:17:26 +09001458 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001459
Tejun Heo47005f22006-06-19 18:27:23 +09001460 /* if we have offending qcs and the associated failed device */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001461 if (ehc->i.dev) {
Tejun Heo47005f22006-06-19 18:27:23 +09001462 /* speed down */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001463 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1464 all_err_mask);
Tejun Heo47005f22006-06-19 18:27:23 +09001465
1466 /* perform per-dev EH action only on the offending device */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001467 ehc->i.dev_action[ehc->i.dev->devno] |=
1468 ehc->i.action & ATA_EH_PERDEV_MASK;
1469 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001470 }
1471
Tejun Heo022bdb02006-05-15 20:58:22 +09001472 DPRINTK("EXIT\n");
1473}
1474
1475/**
1476 * ata_eh_report - report error handling to user
1477 * @ap: ATA port EH is going on
1478 *
1479 * Report EH to user.
1480 *
1481 * LOCKING:
1482 * None.
1483 */
1484static void ata_eh_report(struct ata_port *ap)
1485{
1486 struct ata_eh_context *ehc = &ap->eh_context;
1487 const char *frozen, *desc;
1488 int tag, nr_failed = 0;
1489
1490 desc = NULL;
1491 if (ehc->i.desc[0] != '\0')
1492 desc = ehc->i.desc;
1493
1494 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1495 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1496
1497 if (!(qc->flags & ATA_QCFLAG_FAILED))
1498 continue;
1499 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1500 continue;
1501
1502 nr_failed++;
1503 }
1504
1505 if (!nr_failed && !ehc->i.err_mask)
1506 return;
1507
1508 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001509 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001510 frozen = " frozen";
1511
1512 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001513 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1514 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1515 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1516 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001517 if (desc)
1518 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1519 } else {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001520 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1521 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1522 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1523 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001524 if (desc)
1525 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1526 }
1527
1528 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
Tejun Heo8a937582006-11-14 22:36:12 +09001529 static const char *dma_str[] = {
1530 [DMA_BIDIRECTIONAL] = "bidi",
1531 [DMA_TO_DEVICE] = "out",
1532 [DMA_FROM_DEVICE] = "in",
1533 [DMA_NONE] = "",
1534 };
Tejun Heo022bdb02006-05-15 20:58:22 +09001535 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09001536 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001537
1538 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1539 continue;
1540
Tejun Heo8a937582006-11-14 22:36:12 +09001541 ata_dev_printk(qc->dev, KERN_ERR,
1542 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo664e8502006-11-20 16:05:34 +09001543 "tag %d cdb 0x%x data %u %s\n "
Tejun Heo8a937582006-11-14 22:36:12 +09001544 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1545 "Emask 0x%x (%s)\n",
1546 cmd->command, cmd->feature, cmd->nsect,
1547 cmd->lbal, cmd->lbam, cmd->lbah,
1548 cmd->hob_feature, cmd->hob_nsect,
1549 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heo726f0782007-01-03 17:30:39 +09001550 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
Tejun Heo664e8502006-11-20 16:05:34 +09001551 dma_str[qc->dma_dir],
Tejun Heo8a937582006-11-14 22:36:12 +09001552 res->command, res->feature, res->nsect,
1553 res->lbal, res->lbam, res->lbah,
1554 res->hob_feature, res->hob_nsect,
1555 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1556 res->device, qc->err_mask, ata_err_string(qc->err_mask));
Tejun Heo022bdb02006-05-15 20:58:22 +09001557 }
1558}
1559
Tejun Heod87fa382006-05-31 18:28:24 +09001560static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1561 unsigned int *classes)
1562{
1563 int i, rc;
1564
1565 for (i = 0; i < ATA_MAX_DEVICES; i++)
1566 classes[i] = ATA_DEV_UNKNOWN;
1567
1568 rc = reset(ap, classes);
1569 if (rc)
1570 return rc;
1571
1572 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1573 * is complete and convert all ATA_DEV_UNKNOWN to
1574 * ATA_DEV_NONE.
1575 */
1576 for (i = 0; i < ATA_MAX_DEVICES; i++)
1577 if (classes[i] != ATA_DEV_UNKNOWN)
1578 break;
1579
1580 if (i < ATA_MAX_DEVICES)
1581 for (i = 0; i < ATA_MAX_DEVICES; i++)
1582 if (classes[i] == ATA_DEV_UNKNOWN)
1583 classes[i] = ATA_DEV_NONE;
1584
1585 return 0;
1586}
1587
Tejun Heo664faf02006-05-31 18:27:50 +09001588static int ata_eh_followup_srst_needed(int rc, int classify,
1589 const unsigned int *classes)
1590{
1591 if (rc == -EAGAIN)
1592 return 1;
1593 if (rc != 0)
1594 return 0;
1595 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1596 return 1;
1597 return 0;
1598}
1599
1600static int ata_eh_reset(struct ata_port *ap, int classify,
Tejun Heof5914a42006-05-31 18:27:48 +09001601 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09001602 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1603{
1604 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09001605 unsigned int *classes = ehc->classes;
Tejun Heo022bdb02006-05-15 20:58:22 +09001606 int tries = ATA_EH_RESET_TRIES;
Tejun Heo1cdaf532006-07-03 16:07:26 +09001607 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heof5914a42006-05-31 18:27:48 +09001608 unsigned int action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001609 ata_reset_fn_t reset;
Tejun Heo664faf02006-05-31 18:27:50 +09001610 int i, did_followup_srst, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09001611
Tejun Heo13abf502006-07-10 23:18:46 +09001612 /* about to reset */
1613 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1614
Tejun Heof5914a42006-05-31 18:27:48 +09001615 /* Determine which reset to use and record in ehc->i.action.
1616 * prereset() may examine and modify it.
1617 */
1618 action = ehc->i.action;
1619 ehc->i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo022bdb02006-05-15 20:58:22 +09001620 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
Tejun Heof5914a42006-05-31 18:27:48 +09001621 !(action & ATA_EH_HARDRESET))))
1622 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001623 else
Tejun Heof5914a42006-05-31 18:27:48 +09001624 ehc->i.action |= ATA_EH_HARDRESET;
1625
1626 if (prereset) {
1627 rc = prereset(ap);
1628 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01001629 if (rc == -ENOENT) {
Tejun Heo4aa9ab62007-03-12 17:24:08 +09001630 ata_port_printk(ap, KERN_DEBUG,
1631 "port disabled. ignoring.\n");
Alan Coxc9619222006-09-26 17:53:38 +01001632 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09001633
1634 for (i = 0; i < ATA_MAX_DEVICES; i++)
1635 classes[i] = ATA_DEV_NONE;
1636
1637 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01001638 } else
1639 ata_port_printk(ap, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09001640 "prereset failed (errno=%d)\n", rc);
1641 return rc;
1642 }
1643 }
1644
1645 /* prereset() might have modified ehc->i.action */
1646 if (ehc->i.action & ATA_EH_HARDRESET)
Tejun Heo022bdb02006-05-15 20:58:22 +09001647 reset = hardreset;
Tejun Heof5914a42006-05-31 18:27:48 +09001648 else if (ehc->i.action & ATA_EH_SOFTRESET)
1649 reset = softreset;
1650 else {
1651 /* prereset told us not to reset, bang classes and return */
1652 for (i = 0; i < ATA_MAX_DEVICES; i++)
1653 classes[i] = ATA_DEV_NONE;
1654 return 0;
1655 }
1656
1657 /* did prereset() screw up? if so, fix up to avoid oopsing */
1658 if (!reset) {
1659 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1660 "invalid reset type\n");
1661 if (softreset)
1662 reset = softreset;
1663 else
1664 reset = hardreset;
1665 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001666
1667 retry:
Tejun Heo3e706392006-05-31 18:28:11 +09001668 /* shut up during boot probing */
1669 if (verbose)
1670 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1671 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09001672
Tejun Heo13abf502006-07-10 23:18:46 +09001673 /* mark that this EH session started with reset */
Tejun Heo0d64a232007-04-23 02:41:05 +09001674 if (reset == hardreset)
1675 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1676 else
1677 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001678
1679 rc = ata_do_reset(ap, reset, classes);
1680
Tejun Heo664faf02006-05-31 18:27:50 +09001681 did_followup_srst = 0;
1682 if (reset == hardreset &&
1683 ata_eh_followup_srst_needed(rc, classify, classes)) {
1684 /* okay, let's do follow-up softreset */
1685 did_followup_srst = 1;
1686 reset = softreset;
1687
1688 if (!reset) {
1689 ata_port_printk(ap, KERN_ERR,
1690 "follow-up softreset required "
1691 "but no softreset avaliable\n");
1692 return -EINVAL;
1693 }
1694
Tejun Heo47005f22006-06-19 18:27:23 +09001695 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
Tejun Heo664faf02006-05-31 18:27:50 +09001696 rc = ata_do_reset(ap, reset, classes);
1697
1698 if (rc == 0 && classify &&
1699 classes[0] == ATA_DEV_UNKNOWN) {
1700 ata_port_printk(ap, KERN_ERR,
1701 "classification failed\n");
1702 return -EINVAL;
1703 }
1704 }
1705
Tejun Heo022bdb02006-05-15 20:58:22 +09001706 if (rc && --tries) {
Tejun Heo664faf02006-05-31 18:27:50 +09001707 const char *type;
1708
1709 if (reset == softreset) {
1710 if (did_followup_srst)
1711 type = "follow-up soft";
1712 else
1713 type = "soft";
1714 } else
1715 type = "hard";
1716
Tejun Heo022bdb02006-05-15 20:58:22 +09001717 ata_port_printk(ap, KERN_WARNING,
Tejun Heo664faf02006-05-31 18:27:50 +09001718 "%sreset failed, retrying in 5 secs\n", type);
Tejun Heo022bdb02006-05-15 20:58:22 +09001719 ssleep(5);
1720
1721 if (reset == hardreset)
1722 sata_down_spd_limit(ap);
1723 if (hardreset)
1724 reset = hardreset;
1725 goto retry;
1726 }
1727
1728 if (rc == 0) {
Tejun Heo20952b62006-05-31 18:27:23 +09001729 /* After the reset, the device state is PIO 0 and the
1730 * controller state is undefined. Record the mode.
1731 */
1732 for (i = 0; i < ATA_MAX_DEVICES; i++)
1733 ap->device[i].pio_mode = XFER_PIO_0;
1734
Tejun Heo022bdb02006-05-15 20:58:22 +09001735 if (postreset)
1736 postreset(ap, classes);
1737
1738 /* reset successful, schedule revalidation */
Tejun Heo13abf502006-07-10 23:18:46 +09001739 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
Tejun Heo022bdb02006-05-15 20:58:22 +09001740 ehc->i.action |= ATA_EH_REVALIDATE;
1741 }
1742
1743 return rc;
1744}
1745
Tejun Heo084fe632006-05-31 18:28:03 +09001746static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1747 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001748{
1749 struct ata_eh_context *ehc = &ap->eh_context;
1750 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001751 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09001752 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09001753 int i, rc = 0;
1754
1755 DPRINTK("ENTER\n");
1756
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001757 /* For PATA drive side cable detection to work, IDENTIFY must
1758 * be done backwards such that PDIAG- is released by the slave
1759 * device before the master device is identified.
1760 */
1761 for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) {
Tejun Heobff04642006-11-10 18:08:10 +09001762 unsigned int action, readid_flags = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001763
Tejun Heo47005f22006-06-19 18:27:23 +09001764 dev = &ap->device[i];
Tejun Heo64f65ca2006-06-24 20:30:18 +09001765 action = ata_eh_dev_action(dev);
Tejun Heo47005f22006-06-19 18:27:23 +09001766
Tejun Heobff04642006-11-10 18:08:10 +09001767 if (ehc->i.flags & ATA_EHI_DID_RESET)
1768 readid_flags |= ATA_READID_POSTRESET;
1769
Tejun Heo02670bf2006-07-03 16:07:26 +09001770 if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001771 if (ata_port_offline(ap)) {
1772 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001773 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09001774 }
1775
Tejun Heo47005f22006-06-19 18:27:23 +09001776 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
Tejun Heobff04642006-11-10 18:08:10 +09001777 rc = ata_dev_revalidate(dev, readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001778 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001779 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09001780
Tejun Heo47005f22006-06-19 18:27:23 +09001781 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1782
Tejun Heobaa1e782006-11-01 18:39:27 +09001783 /* Configuration may have changed, reconfigure
1784 * transfer mode.
1785 */
1786 ehc->i.flags |= ATA_EHI_SETMODE;
1787
zhao, forrest3057ac32006-06-12 12:01:34 +08001788 /* schedule the scsi_rescan_device() here */
1789 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09001790 } else if (dev->class == ATA_DEV_UNKNOWN &&
1791 ehc->tries[dev->devno] &&
1792 ata_class_enabled(ehc->classes[dev->devno])) {
1793 dev->class = ehc->classes[dev->devno];
1794
Tejun Heobff04642006-11-10 18:08:10 +09001795 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1796 dev->id);
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001797 switch (rc) {
1798 case 0:
1799 new_mask |= 1 << i;
1800 break;
1801 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001802 /* IDENTIFY was issued to non-existent
1803 * device. No need to reset. Just
1804 * thaw and kill the device.
1805 */
1806 ata_eh_thaw_port(ap);
1807 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo084fe632006-05-31 18:28:03 +09001808 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001809 default:
1810 dev->class = ATA_DEV_UNKNOWN;
1811 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001812 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001813 }
1814 }
1815
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09001816 /* PDIAG- should have been released, ask cable type if post-reset */
1817 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
1818 ap->cbl = ap->ops->cable_detect(ap);
1819
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001820 /* Configure new devices forward such that user doesn't see
1821 * device detection messages backwards.
1822 */
1823 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1824 dev = &ap->device[i];
Tejun Heo022bdb02006-05-15 20:58:22 +09001825
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001826 if (!(new_mask & (1 << i)))
1827 continue;
1828
1829 ehc->i.flags |= ATA_EHI_PRINTINFO;
1830 rc = ata_dev_configure(dev);
1831 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1832 if (rc)
1833 goto err;
1834
1835 spin_lock_irqsave(ap->lock, flags);
1836 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1837 spin_unlock_irqrestore(ap->lock, flags);
1838
1839 /* new device discovered, configure xfermode */
1840 ehc->i.flags |= ATA_EHI_SETMODE;
1841 }
1842
1843 return 0;
1844
1845 err:
1846 *r_failed_dev = dev;
1847 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001848 return rc;
1849}
1850
Tejun Heo6ffa01d2007-03-02 17:32:47 +09001851#ifdef CONFIG_PM
Tejun Heo02670bf2006-07-03 16:07:26 +09001852/**
1853 * ata_eh_suspend - handle suspend EH action
1854 * @ap: target host port
1855 * @r_failed_dev: result parameter to indicate failing device
1856 *
1857 * Handle suspend EH action. Disk devices are spinned down and
1858 * other types of devices are just marked suspended. Once
1859 * suspended, no EH action to the device is allowed until it is
1860 * resumed.
1861 *
1862 * LOCKING:
1863 * Kernel thread context (may sleep).
1864 *
1865 * RETURNS:
1866 * 0 on success, -errno otherwise
1867 */
1868static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev)
1869{
1870 struct ata_device *dev;
1871 int i, rc = 0;
1872
1873 DPRINTK("ENTER\n");
1874
1875 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1876 unsigned long flags;
1877 unsigned int action, err_mask;
1878
1879 dev = &ap->device[i];
1880 action = ata_eh_dev_action(dev);
1881
1882 if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND))
1883 continue;
1884
1885 WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED);
1886
1887 ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND);
1888
1889 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1890 /* flush cache */
1891 rc = ata_flush_cache(dev);
1892 if (rc)
1893 break;
1894
1895 /* spin down */
1896 err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1);
1897 if (err_mask) {
1898 ata_dev_printk(dev, KERN_ERR, "failed to "
1899 "spin down (err_mask=0x%x)\n",
1900 err_mask);
1901 rc = -EIO;
1902 break;
1903 }
1904 }
1905
1906 spin_lock_irqsave(ap->lock, flags);
1907 dev->flags |= ATA_DFLAG_SUSPENDED;
1908 spin_unlock_irqrestore(ap->lock, flags);
1909
1910 ata_eh_done(ap, dev, ATA_EH_SUSPEND);
1911 }
1912
1913 if (rc)
1914 *r_failed_dev = dev;
1915
1916 DPRINTK("EXIT\n");
Tejun Heo03ee5b12007-01-26 20:10:25 +09001917 return rc;
Tejun Heo02670bf2006-07-03 16:07:26 +09001918}
1919
1920/**
1921 * ata_eh_prep_resume - prep for resume EH action
1922 * @ap: target host port
1923 *
1924 * Clear SUSPENDED in preparation for scheduled resume actions.
1925 * This allows other parts of EH to access the devices being
1926 * resumed.
1927 *
1928 * LOCKING:
1929 * Kernel thread context (may sleep).
1930 */
1931static void ata_eh_prep_resume(struct ata_port *ap)
1932{
1933 struct ata_device *dev;
1934 unsigned long flags;
1935 int i;
1936
1937 DPRINTK("ENTER\n");
1938
1939 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1940 unsigned int action;
1941
1942 dev = &ap->device[i];
1943 action = ata_eh_dev_action(dev);
1944
1945 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1946 continue;
1947
1948 spin_lock_irqsave(ap->lock, flags);
1949 dev->flags &= ~ATA_DFLAG_SUSPENDED;
1950 spin_unlock_irqrestore(ap->lock, flags);
1951 }
1952
1953 DPRINTK("EXIT\n");
1954}
1955
1956/**
1957 * ata_eh_resume - handle resume EH action
1958 * @ap: target host port
1959 * @r_failed_dev: result parameter to indicate failing device
1960 *
1961 * Handle resume EH action. Target devices are already reset and
1962 * revalidated. Spinning up is the only operation left.
1963 *
1964 * LOCKING:
1965 * Kernel thread context (may sleep).
1966 *
1967 * RETURNS:
1968 * 0 on success, -errno otherwise
1969 */
1970static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev)
1971{
1972 struct ata_device *dev;
1973 int i, rc = 0;
1974
1975 DPRINTK("ENTER\n");
1976
1977 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1978 unsigned int action, err_mask;
1979
1980 dev = &ap->device[i];
1981 action = ata_eh_dev_action(dev);
1982
1983 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1984 continue;
1985
1986 ata_eh_about_to_do(ap, dev, ATA_EH_RESUME);
1987
1988 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1989 err_mask = ata_do_simple_cmd(dev,
1990 ATA_CMD_IDLEIMMEDIATE);
1991 if (err_mask) {
1992 ata_dev_printk(dev, KERN_ERR, "failed to "
1993 "spin up (err_mask=0x%x)\n",
1994 err_mask);
1995 rc = -EIO;
1996 break;
1997 }
1998 }
1999
2000 ata_eh_done(ap, dev, ATA_EH_RESUME);
2001 }
2002
2003 if (rc)
2004 *r_failed_dev = dev;
2005
2006 DPRINTK("EXIT\n");
2007 return 0;
2008}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002009#endif /* CONFIG_PM */
Tejun Heo02670bf2006-07-03 16:07:26 +09002010
Tejun Heo022bdb02006-05-15 20:58:22 +09002011static int ata_port_nr_enabled(struct ata_port *ap)
2012{
2013 int i, cnt = 0;
2014
2015 for (i = 0; i < ATA_MAX_DEVICES; i++)
2016 if (ata_dev_enabled(&ap->device[i]))
2017 cnt++;
2018 return cnt;
2019}
2020
Tejun Heo084fe632006-05-31 18:28:03 +09002021static int ata_port_nr_vacant(struct ata_port *ap)
2022{
2023 int i, cnt = 0;
2024
2025 for (i = 0; i < ATA_MAX_DEVICES; i++)
2026 if (ap->device[i].class == ATA_DEV_UNKNOWN)
2027 cnt++;
2028 return cnt;
2029}
2030
2031static int ata_eh_skip_recovery(struct ata_port *ap)
2032{
2033 struct ata_eh_context *ehc = &ap->eh_context;
2034 int i;
2035
Tejun Heo02670bf2006-07-03 16:07:26 +09002036 /* skip if all possible devices are suspended */
2037 for (i = 0; i < ata_port_max_devices(ap); i++) {
2038 struct ata_device *dev = &ap->device[i];
2039
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09002040 if (!(dev->flags & ATA_DFLAG_SUSPENDED))
Tejun Heo02670bf2006-07-03 16:07:26 +09002041 break;
2042 }
2043
2044 if (i == ata_port_max_devices(ap))
2045 return 1;
2046
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09002047 /* thaw frozen port, resume link and recover failed devices */
2048 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
2049 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
Tejun Heo084fe632006-05-31 18:28:03 +09002050 return 0;
2051
2052 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2053 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2054 struct ata_device *dev = &ap->device[i];
2055
2056 if (dev->class == ATA_DEV_UNKNOWN &&
2057 ehc->classes[dev->devno] != ATA_DEV_NONE)
2058 return 0;
2059 }
2060
2061 return 1;
2062}
2063
Tejun Heo022bdb02006-05-15 20:58:22 +09002064/**
2065 * ata_eh_recover - recover host port after error
2066 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002067 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002068 * @softreset: softreset method (can be NULL)
2069 * @hardreset: hardreset method (can be NULL)
2070 * @postreset: postreset method (can be NULL)
2071 *
2072 * This is the alpha and omega, eum and yang, heart and soul of
2073 * libata exception handling. On entry, actions required to
Tejun Heo084fe632006-05-31 18:28:03 +09002074 * recover the port and hotplug requests are recorded in
2075 * eh_context. This function executes all the operations with
2076 * appropriate retrials and fallbacks to resurrect failed
2077 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09002078 *
2079 * LOCKING:
2080 * Kernel thread context (may sleep).
2081 *
2082 * RETURNS:
2083 * 0 on success, -errno on failure.
2084 */
Tejun Heof5914a42006-05-31 18:27:48 +09002085static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2086 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09002087 ata_postreset_fn_t postreset)
2088{
2089 struct ata_eh_context *ehc = &ap->eh_context;
2090 struct ata_device *dev;
Tejun Heo4ae72a12007-02-02 16:22:30 +09002091 int i, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002092
2093 DPRINTK("ENTER\n");
2094
2095 /* prep for recovery */
2096 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2097 dev = &ap->device[i];
2098
2099 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo084fe632006-05-31 18:28:03 +09002100
Tejun Heo79a55b72007-01-18 17:22:18 +09002101 /* collect port action mask recorded in dev actions */
2102 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
2103 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
2104
Tejun Heo084fe632006-05-31 18:28:03 +09002105 /* process hotplug request */
2106 if (dev->flags & ATA_DFLAG_DETACH)
2107 ata_eh_detach_dev(dev);
2108
2109 if (!ata_dev_enabled(dev) &&
2110 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2111 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2112 ata_eh_detach_dev(dev);
2113 ata_dev_init(dev);
2114 ehc->did_probe_mask |= (1 << dev->devno);
2115 ehc->i.action |= ATA_EH_SOFTRESET;
2116 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002117 }
2118
2119 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09002120 rc = 0;
2121
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002122 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09002123 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002124 goto out;
2125
Tejun Heo02670bf2006-07-03 16:07:26 +09002126 /* prep for resume */
2127 ata_eh_prep_resume(ap);
2128
Tejun Heo022bdb02006-05-15 20:58:22 +09002129 /* skip EH if possible. */
Tejun Heo084fe632006-05-31 18:28:03 +09002130 if (ata_eh_skip_recovery(ap))
Tejun Heo022bdb02006-05-15 20:58:22 +09002131 ehc->i.action = 0;
2132
Tejun Heo084fe632006-05-31 18:28:03 +09002133 for (i = 0; i < ATA_MAX_DEVICES; i++)
2134 ehc->classes[i] = ATA_DEV_UNKNOWN;
2135
Tejun Heo022bdb02006-05-15 20:58:22 +09002136 /* reset */
2137 if (ehc->i.action & ATA_EH_RESET_MASK) {
2138 ata_eh_freeze_port(ap);
2139
Tejun Heo084fe632006-05-31 18:28:03 +09002140 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2141 softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002142 if (rc) {
2143 ata_port_printk(ap, KERN_ERR,
2144 "reset failed, giving up\n");
2145 goto out;
2146 }
2147
2148 ata_eh_thaw_port(ap);
2149 }
2150
Tejun Heo084fe632006-05-31 18:28:03 +09002151 /* revalidate existing devices and attach new ones */
2152 rc = ata_eh_revalidate_and_attach(ap, &dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09002153 if (rc)
2154 goto dev_fail;
2155
Tejun Heo02670bf2006-07-03 16:07:26 +09002156 /* resume devices */
2157 rc = ata_eh_resume(ap, &dev);
2158 if (rc)
2159 goto dev_fail;
2160
Tejun Heobaa1e782006-11-01 18:39:27 +09002161 /* configure transfer mode if necessary */
2162 if (ehc->i.flags & ATA_EHI_SETMODE) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002163 rc = ata_set_mode(ap, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09002164 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09002165 goto dev_fail;
Tejun Heobaa1e782006-11-01 18:39:27 +09002166 ehc->i.flags &= ~ATA_EHI_SETMODE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002167 }
2168
Tejun Heo02670bf2006-07-03 16:07:26 +09002169 /* suspend devices */
2170 rc = ata_eh_suspend(ap, &dev);
2171 if (rc)
2172 goto dev_fail;
2173
Tejun Heo022bdb02006-05-15 20:58:22 +09002174 goto out;
2175
2176 dev_fail:
Tejun Heo4ae72a12007-02-02 16:22:30 +09002177 ehc->tries[dev->devno]--;
2178
Tejun Heo022bdb02006-05-15 20:58:22 +09002179 switch (rc) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002180 case -EINVAL:
Tejun Heo4ae72a12007-02-02 16:22:30 +09002181 /* eeek, something went very wrong, give up */
Tejun Heo022bdb02006-05-15 20:58:22 +09002182 ehc->tries[dev->devno] = 0;
2183 break;
Tejun Heo4ae72a12007-02-02 16:22:30 +09002184
2185 case -ENODEV:
2186 /* device missing or wrong IDENTIFY data, schedule probing */
2187 ehc->i.probe_mask |= (1 << dev->devno);
2188 /* give it just one more chance */
2189 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
Tejun Heo022bdb02006-05-15 20:58:22 +09002190 case -EIO:
Tejun Heo4ae72a12007-02-02 16:22:30 +09002191 if (ehc->tries[dev->devno] == 1) {
2192 /* This is the last chance, better to slow
2193 * down than lose it.
2194 */
2195 sata_down_spd_limit(ap);
2196 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2197 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002198 }
2199
Tejun Heo084fe632006-05-31 18:28:03 +09002200 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2201 /* disable device if it has used up all its chances */
Tejun Heo022bdb02006-05-15 20:58:22 +09002202 ata_dev_disable(dev);
2203
Tejun Heo084fe632006-05-31 18:28:03 +09002204 /* detach if offline */
2205 if (ata_port_offline(ap))
2206 ata_eh_detach_dev(dev);
2207
2208 /* probe if requested */
2209 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2210 !(ehc->did_probe_mask & (1 << dev->devno))) {
2211 ata_eh_detach_dev(dev);
2212 ata_dev_init(dev);
2213
2214 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2215 ehc->did_probe_mask |= (1 << dev->devno);
2216 ehc->i.action |= ATA_EH_SOFTRESET;
2217 }
2218 } else {
2219 /* soft didn't work? be haaaaard */
2220 if (ehc->i.flags & ATA_EHI_DID_RESET)
2221 ehc->i.action |= ATA_EH_HARDRESET;
2222 else
2223 ehc->i.action |= ATA_EH_SOFTRESET;
2224 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002225
2226 if (ata_port_nr_enabled(ap)) {
2227 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2228 "devices, retrying in 5 secs\n");
2229 ssleep(5);
2230 } else {
2231 /* no device left, repeat fast */
2232 msleep(500);
2233 }
2234
2235 goto retry;
2236
2237 out:
2238 if (rc) {
2239 for (i = 0; i < ATA_MAX_DEVICES; i++)
2240 ata_dev_disable(&ap->device[i]);
2241 }
2242
2243 DPRINTK("EXIT, rc=%d\n", rc);
2244 return rc;
2245}
2246
2247/**
2248 * ata_eh_finish - finish up EH
2249 * @ap: host port to finish EH for
2250 *
2251 * Recovery is complete. Clean up EH states and retry or finish
2252 * failed qcs.
2253 *
2254 * LOCKING:
2255 * None.
2256 */
2257static void ata_eh_finish(struct ata_port *ap)
2258{
2259 int tag;
2260
2261 /* retry or finish qcs */
2262 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2263 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2264
2265 if (!(qc->flags & ATA_QCFLAG_FAILED))
2266 continue;
2267
2268 if (qc->err_mask) {
2269 /* FIXME: Once EH migration is complete,
2270 * generate sense data in this function,
2271 * considering both err_mask and tf.
2272 */
2273 if (qc->err_mask & AC_ERR_INVALID)
2274 ata_eh_qc_complete(qc);
2275 else
2276 ata_eh_qc_retry(qc);
2277 } else {
2278 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2279 ata_eh_qc_complete(qc);
2280 } else {
2281 /* feed zero TF to sense generation */
2282 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2283 ata_eh_qc_retry(qc);
2284 }
2285 }
2286 }
2287}
2288
2289/**
2290 * ata_do_eh - do standard error handling
2291 * @ap: host port to handle error for
Tejun Heof5914a42006-05-31 18:27:48 +09002292 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002293 * @softreset: softreset method (can be NULL)
2294 * @hardreset: hardreset method (can be NULL)
2295 * @postreset: postreset method (can be NULL)
2296 *
2297 * Perform standard error handling sequence.
2298 *
2299 * LOCKING:
2300 * Kernel thread context (may sleep).
2301 */
Tejun Heof5914a42006-05-31 18:27:48 +09002302void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2303 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2304 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002305{
Tejun Heo1cdaf532006-07-03 16:07:26 +09002306 ata_eh_autopsy(ap);
2307 ata_eh_report(ap);
Tejun Heof5914a42006-05-31 18:27:48 +09002308 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002309 ata_eh_finish(ap);
2310}
Tejun Heo500530f2006-07-03 16:07:27 +09002311
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002312#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09002313/**
2314 * ata_eh_handle_port_suspend - perform port suspend operation
2315 * @ap: port to suspend
2316 *
2317 * Suspend @ap.
2318 *
2319 * LOCKING:
2320 * Kernel thread context (may sleep).
2321 */
2322static void ata_eh_handle_port_suspend(struct ata_port *ap)
2323{
2324 unsigned long flags;
2325 int rc = 0;
2326
2327 /* are we suspending? */
2328 spin_lock_irqsave(ap->lock, flags);
2329 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2330 ap->pm_mesg.event == PM_EVENT_ON) {
2331 spin_unlock_irqrestore(ap->lock, flags);
2332 return;
2333 }
2334 spin_unlock_irqrestore(ap->lock, flags);
2335
2336 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2337
2338 /* suspend */
2339 ata_eh_freeze_port(ap);
2340
2341 if (ap->ops->port_suspend)
2342 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2343
2344 /* report result */
2345 spin_lock_irqsave(ap->lock, flags);
2346
2347 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2348 if (rc == 0)
2349 ap->pflags |= ATA_PFLAG_SUSPENDED;
2350 else
2351 ata_port_schedule_eh(ap);
2352
2353 if (ap->pm_result) {
2354 *ap->pm_result = rc;
2355 ap->pm_result = NULL;
2356 }
2357
2358 spin_unlock_irqrestore(ap->lock, flags);
2359
2360 return;
2361}
2362
2363/**
2364 * ata_eh_handle_port_resume - perform port resume operation
2365 * @ap: port to resume
2366 *
2367 * Resume @ap.
2368 *
2369 * This function also waits upto one second until all devices
2370 * hanging off this port requests resume EH action. This is to
2371 * prevent invoking EH and thus reset multiple times on resume.
2372 *
2373 * On DPM resume, where some of devices might not be resumed
2374 * together, this may delay port resume upto one second, but such
2375 * DPM resumes are rare and 1 sec delay isn't too bad.
2376 *
2377 * LOCKING:
2378 * Kernel thread context (may sleep).
2379 */
2380static void ata_eh_handle_port_resume(struct ata_port *ap)
2381{
2382 unsigned long timeout;
2383 unsigned long flags;
2384 int i, rc = 0;
2385
2386 /* are we resuming? */
2387 spin_lock_irqsave(ap->lock, flags);
2388 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2389 ap->pm_mesg.event != PM_EVENT_ON) {
2390 spin_unlock_irqrestore(ap->lock, flags);
2391 return;
2392 }
2393 spin_unlock_irqrestore(ap->lock, flags);
2394
2395 /* spurious? */
2396 if (!(ap->pflags & ATA_PFLAG_SUSPENDED))
2397 goto done;
2398
2399 if (ap->ops->port_resume)
2400 rc = ap->ops->port_resume(ap);
2401
2402 /* give devices time to request EH */
2403 timeout = jiffies + HZ; /* 1s max */
2404 while (1) {
2405 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2406 struct ata_device *dev = &ap->device[i];
2407 unsigned int action = ata_eh_dev_action(dev);
2408
2409 if ((dev->flags & ATA_DFLAG_SUSPENDED) &&
2410 !(action & ATA_EH_RESUME))
2411 break;
2412 }
2413
2414 if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout))
2415 break;
2416 msleep(10);
2417 }
2418
2419 done:
2420 spin_lock_irqsave(ap->lock, flags);
2421 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2422 if (ap->pm_result) {
2423 *ap->pm_result = rc;
2424 ap->pm_result = NULL;
2425 }
2426 spin_unlock_irqrestore(ap->lock, flags);
2427}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002428#endif /* CONFIG_PM */