blob: a5a8f8453061b42a17a16378055542ba6b5f87ea [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 Heo31daabd2007-02-02 16:50:52 +090053/* Waiting in ->prereset can never be reliable. It's sometimes nice
54 * to wait there but it can't be depended upon; otherwise, we wouldn't
55 * be resetting. Just give it enough time for most drives to spin up.
56 */
57enum {
58 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
59};
60
61/* The following table determines how we sequence resets. Each entry
62 * represents timeout for that try. The first try can be soft or
63 * hardreset. All others are hardreset if available. In most cases
64 * the first reset w/ 10sec timeout should succeed. Following entries
65 * are mostly for error handling, hotplug and retarded devices.
66 */
67static const unsigned long ata_eh_reset_timeouts[] = {
68 10 * HZ, /* most drives spin up by 10sec */
69 10 * HZ, /* > 99% working drives spin up before 20sec */
70 35 * HZ, /* give > 30 secs of idleness for retarded devices */
71 5 * HZ, /* and sweet one last chance */
72 /* > 1 min has elapsed, give up */
73};
74
Tejun Heoad9e2762006-05-15 20:58:12 +090075static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo720ba122006-05-31 18:28:13 +090076static void ata_eh_finish(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090077#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +090078static void ata_eh_handle_port_suspend(struct ata_port *ap);
79static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090080#else /* CONFIG_PM */
81static void ata_eh_handle_port_suspend(struct ata_port *ap)
82{ }
83
84static void ata_eh_handle_port_resume(struct ata_port *ap)
85{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +090086#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +090087
Tejun Heob64bbc32007-07-16 14:29:39 +090088static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
89 va_list args)
90{
91 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
92 ATA_EH_DESC_LEN - ehi->desc_len,
93 fmt, args);
94}
95
96/**
97 * __ata_ehi_push_desc - push error description without adding separator
98 * @ehi: target EHI
99 * @fmt: printf format string
100 *
101 * Format string according to @fmt and append it to @ehi->desc.
102 *
103 * LOCKING:
104 * spin_lock_irqsave(host lock)
105 */
106void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
107{
108 va_list args;
109
110 va_start(args, fmt);
111 __ata_ehi_pushv_desc(ehi, fmt, args);
112 va_end(args);
113}
114
115/**
116 * ata_ehi_push_desc - push error description with separator
117 * @ehi: target EHI
118 * @fmt: printf format string
119 *
120 * Format string according to @fmt and append it to @ehi->desc.
121 * If @ehi->desc is not empty, ", " is added in-between.
122 *
123 * LOCKING:
124 * spin_lock_irqsave(host lock)
125 */
126void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
127{
128 va_list args;
129
130 if (ehi->desc_len)
131 __ata_ehi_push_desc(ehi, ", ");
132
133 va_start(args, fmt);
134 __ata_ehi_pushv_desc(ehi, fmt, args);
135 va_end(args);
136}
137
138/**
139 * ata_ehi_clear_desc - clean error description
140 * @ehi: target EHI
141 *
142 * Clear @ehi->desc.
143 *
144 * LOCKING:
145 * spin_lock_irqsave(host lock)
146 */
147void ata_ehi_clear_desc(struct ata_eh_info *ehi)
148{
149 ehi->desc[0] = '\0';
150 ehi->desc_len = 0;
151}
152
Tejun Heo0c247c52006-05-15 20:58:19 +0900153static void ata_ering_record(struct ata_ering *ering, int is_io,
154 unsigned int err_mask)
155{
156 struct ata_ering_entry *ent;
157
158 WARN_ON(!err_mask);
159
160 ering->cursor++;
161 ering->cursor %= ATA_ERING_SIZE;
162
163 ent = &ering->ring[ering->cursor];
164 ent->is_io = is_io;
165 ent->err_mask = err_mask;
166 ent->timestamp = get_jiffies_64();
167}
168
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900169static void ata_ering_clear(struct ata_ering *ering)
Tejun Heo0c247c52006-05-15 20:58:19 +0900170{
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900171 memset(ering, 0, sizeof(*ering));
Tejun Heo0c247c52006-05-15 20:58:19 +0900172}
173
174static int ata_ering_map(struct ata_ering *ering,
175 int (*map_fn)(struct ata_ering_entry *, void *),
176 void *arg)
177{
178 int idx, rc = 0;
179 struct ata_ering_entry *ent;
180
181 idx = ering->cursor;
182 do {
183 ent = &ering->ring[idx];
184 if (!ent->err_mask)
185 break;
186 rc = map_fn(ent, arg);
187 if (rc)
188 break;
189 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
190 } while (idx != ering->cursor);
191
192 return rc;
193}
194
Tejun Heo64f65ca2006-06-24 20:30:18 +0900195static unsigned int ata_eh_dev_action(struct ata_device *dev)
196{
197 struct ata_eh_context *ehc = &dev->ap->eh_context;
198
199 return ehc->i.action | ehc->i.dev_action[dev->devno];
200}
201
Tejun Heoaf181c22006-06-24 20:30:18 +0900202static void ata_eh_clear_action(struct ata_device *dev,
203 struct ata_eh_info *ehi, unsigned int action)
204{
205 int i;
206
207 if (!dev) {
208 ehi->action &= ~action;
209 for (i = 0; i < ATA_MAX_DEVICES; i++)
210 ehi->dev_action[i] &= ~action;
211 } else {
212 /* doesn't make sense for port-wide EH actions */
213 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
214
215 /* break ehi->action into ehi->dev_action */
216 if (ehi->action & action) {
217 for (i = 0; i < ATA_MAX_DEVICES; i++)
218 ehi->dev_action[i] |= ehi->action & action;
219 ehi->action &= ~action;
220 }
221
222 /* turn off the specified per-dev action */
223 ehi->dev_action[dev->devno] &= ~action;
224 }
225}
226
Tejun Heoece1d632006-04-02 18:51:53 +0900227/**
228 * ata_scsi_timed_out - SCSI layer time out callback
229 * @cmd: timed out SCSI command
230 *
231 * Handles SCSI layer timeout. We race with normal completion of
232 * the qc for @cmd. If the qc is already gone, we lose and let
233 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
234 * timed out and EH should be invoked. Prevent ata_qc_complete()
235 * from finishing it by setting EH_SCHEDULED and return
236 * EH_NOT_HANDLED.
237 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900238 * TODO: kill this function once old EH is gone.
239 *
Tejun Heoece1d632006-04-02 18:51:53 +0900240 * LOCKING:
241 * Called from timer context
242 *
243 * RETURNS:
244 * EH_HANDLED or EH_NOT_HANDLED
245 */
246enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
247{
248 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400249 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900250 unsigned long flags;
251 struct ata_queued_cmd *qc;
Tejun Heoad9e2762006-05-15 20:58:12 +0900252 enum scsi_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900253
254 DPRINTK("ENTER\n");
255
Tejun Heoad9e2762006-05-15 20:58:12 +0900256 if (ap->ops->error_handler) {
257 ret = EH_NOT_HANDLED;
258 goto out;
259 }
260
261 ret = EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400262 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900263 qc = ata_qc_from_tag(ap, ap->active_tag);
264 if (qc) {
265 WARN_ON(qc->scsicmd != cmd);
266 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
267 qc->err_mask |= AC_ERR_TIMEOUT;
268 ret = EH_NOT_HANDLED;
269 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400270 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900271
Tejun Heoad9e2762006-05-15 20:58:12 +0900272 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900273 DPRINTK("EXIT, ret=%d\n", ret);
274 return ret;
275}
276
277/**
278 * ata_scsi_error - SCSI layer error handler callback
279 * @host: SCSI host on which error occurred
280 *
281 * Handles SCSI-layer-thrown error events.
282 *
283 * LOCKING:
284 * Inherited from SCSI layer (none, can sleep)
285 *
286 * RETURNS:
287 * Zero.
288 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400289void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900290{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400291 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoad9e2762006-05-15 20:58:12 +0900292 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
293 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900294
295 DPRINTK("ENTER\n");
296
Tejun Heoad9e2762006-05-15 20:58:12 +0900297 /* synchronize with port task */
Tejun Heoece1d632006-04-02 18:51:53 +0900298 ata_port_flush_task(ap);
299
Jeff Garzikcca39742006-08-24 03:19:22 -0400300 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900301
Tejun Heoad9e2762006-05-15 20:58:12 +0900302 /* For new EH, all qcs are finished in one of three ways -
303 * normal completion, error completion, and SCSI timeout.
304 * Both cmpletions can race against SCSI timeout. When normal
305 * completion wins, the qc never reaches EH. When error
306 * completion wins, the qc has ATA_QCFLAG_FAILED set.
307 *
308 * When SCSI timeout wins, things are a bit more complex.
309 * Normal or error completion can occur after the timeout but
310 * before this point. In such cases, both types of
311 * completions are honored. A scmd is determined to have
312 * timed out iff its associated qc is active and not failed.
313 */
314 if (ap->ops->error_handler) {
315 struct scsi_cmnd *scmd, *tmp;
316 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900317
Tejun Heoe30349d2006-07-03 03:02:15 +0900318 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900319
320 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
321 struct ata_queued_cmd *qc;
322
323 for (i = 0; i < ATA_MAX_QUEUE; i++) {
324 qc = __ata_qc_from_tag(ap, i);
325 if (qc->flags & ATA_QCFLAG_ACTIVE &&
326 qc->scsicmd == scmd)
327 break;
328 }
329
330 if (i < ATA_MAX_QUEUE) {
331 /* the scmd has an associated qc */
332 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
333 /* which hasn't failed yet, timeout */
334 qc->err_mask |= AC_ERR_TIMEOUT;
335 qc->flags |= ATA_QCFLAG_FAILED;
336 nr_timedout++;
337 }
338 } else {
339 /* Normal completion occurred after
340 * SCSI timeout but before this point.
341 * Successfully complete it.
342 */
343 scmd->retries = scmd->allowed;
344 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
345 }
346 }
347
348 /* If we have timed out qcs. They belong to EH from
349 * this point but the state of the controller is
350 * unknown. Freeze the port to make sure the IRQ
351 * handler doesn't diddle with those qcs. This must
352 * be done atomically w.r.t. setting QCFLAG_FAILED.
353 */
354 if (nr_timedout)
355 __ata_port_freeze(ap);
356
Tejun Heoe30349d2006-07-03 03:02:15 +0900357 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900358 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900359 spin_unlock_wait(ap->lock);
Tejun Heoad9e2762006-05-15 20:58:12 +0900360
361 repeat:
362 /* invoke error handler */
363 if (ap->ops->error_handler) {
Tejun Heo500530f2006-07-03 16:07:27 +0900364 /* process port resume request */
365 ata_eh_handle_port_resume(ap);
366
Tejun Heof3e81b12006-05-15 20:58:21 +0900367 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900368 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900369
370 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
371 ap->eh_context.i = ap->eh_info;
372 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
373
Tejun Heob51e9e52006-06-29 01:29:30 +0900374 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
375 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof3e81b12006-05-15 20:58:21 +0900376
Tejun Heoe30349d2006-07-03 03:02:15 +0900377 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900378
Tejun Heo500530f2006-07-03 16:07:27 +0900379 /* invoke EH, skip if unloading or suspended */
380 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900381 ap->ops->error_handler(ap);
382 else
383 ata_eh_finish(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900384
Tejun Heo500530f2006-07-03 16:07:27 +0900385 /* process port suspend request */
386 ata_eh_handle_port_suspend(ap);
387
Tejun Heoad9e2762006-05-15 20:58:12 +0900388 /* Exception might have happend after ->error_handler
389 * recovered the port but before this point. Repeat
390 * EH in such case.
391 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900392 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900393
Tejun Heob51e9e52006-06-29 01:29:30 +0900394 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoad9e2762006-05-15 20:58:12 +0900395 if (--repeat_cnt) {
396 ata_port_printk(ap, KERN_INFO,
397 "EH pending after completion, "
398 "repeating EH (cnt=%d)\n", repeat_cnt);
Tejun Heoe30349d2006-07-03 03:02:15 +0900399 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900400 goto repeat;
401 }
402 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
403 "tries, giving up\n", ATA_EH_MAX_REPEAT);
Tejun Heo914616a2007-06-25 21:47:11 +0900404 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900405 }
406
Tejun Heof3e81b12006-05-15 20:58:21 +0900407 /* this run is complete, make sure EH info is clear */
408 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
409
Tejun Heoe30349d2006-07-03 03:02:15 +0900410 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900411 * that if exception occurs after this point but
412 * before EH completion, SCSI midlayer will
413 * re-initiate EH.
414 */
415 host->host_eh_scheduled = 0;
416
Tejun Heoe30349d2006-07-03 03:02:15 +0900417 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900418 } else {
419 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
420 ap->ops->eng_timeout(ap);
421 }
422
423 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900424 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
425
426 scsi_eh_flush_done_q(&ap->eh_done_q);
427
Tejun Heoad9e2762006-05-15 20:58:12 +0900428 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900429 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900430
Tejun Heo1cdaf532006-07-03 16:07:26 +0900431 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900432 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900433 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
David Howells52bad642006-11-22 14:54:01 +0000434 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900435
436 if (ap->pflags & ATA_PFLAG_RECOVERED)
437 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b2102006-05-31 18:28:05 +0900438
Tejun Heob51e9e52006-06-29 01:29:30 +0900439 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900440
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900441 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900442 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900443 wake_up_all(&ap->eh_wait_q);
444
Tejun Heoe30349d2006-07-03 03:02:15 +0900445 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900446
Tejun Heoece1d632006-04-02 18:51:53 +0900447 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900448}
449
450/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900451 * ata_port_wait_eh - Wait for the currently pending EH to complete
452 * @ap: Port to wait EH for
453 *
454 * Wait until the currently pending EH is complete.
455 *
456 * LOCKING:
457 * Kernel thread context (may sleep).
458 */
459void ata_port_wait_eh(struct ata_port *ap)
460{
461 unsigned long flags;
462 DEFINE_WAIT(wait);
463
464 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400465 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900466
Tejun Heob51e9e52006-06-29 01:29:30 +0900467 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900468 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400469 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900470 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400471 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900472 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900473 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900474
Jeff Garzikba6a1302006-06-22 23:46:10 -0400475 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900476
477 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400478 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900479 msleep(10);
480 goto retry;
481 }
482}
483
484/**
Tejun Heoece1d632006-04-02 18:51:53 +0900485 * ata_qc_timeout - Handle timeout of queued command
486 * @qc: Command that timed out
487 *
488 * Some part of the kernel (currently, only the SCSI layer)
489 * has noticed that the active command on port @ap has not
490 * completed after a specified length of time. Handle this
491 * condition by disabling DMA (if necessary) and completing
492 * transactions, with error if necessary.
493 *
494 * This also handles the case of the "lost interrupt", where
495 * for some reason (possibly hardware bug, possibly driver bug)
496 * an interrupt was not delivered to the driver, even though the
497 * transaction completed successfully.
498 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900499 * TODO: kill this function once old EH is gone.
500 *
Tejun Heoece1d632006-04-02 18:51:53 +0900501 * LOCKING:
502 * Inherited from SCSI layer (none, can sleep)
503 */
504static void ata_qc_timeout(struct ata_queued_cmd *qc)
505{
506 struct ata_port *ap = qc->ap;
Tejun Heoece1d632006-04-02 18:51:53 +0900507 u8 host_stat = 0, drv_stat;
508 unsigned long flags;
509
510 DPRINTK("ENTER\n");
511
512 ap->hsm_task_state = HSM_ST_IDLE;
513
Jeff Garzikba6a1302006-06-22 23:46:10 -0400514 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900515
516 switch (qc->tf.protocol) {
517
518 case ATA_PROT_DMA:
519 case ATA_PROT_ATAPI_DMA:
520 host_stat = ap->ops->bmdma_status(ap);
521
522 /* before we do anything else, clear DMA-Start bit */
523 ap->ops->bmdma_stop(qc);
524
525 /* fall through */
526
527 default:
528 ata_altstatus(ap);
529 drv_stat = ata_chk_status(ap);
530
531 /* ack bmdma irq events */
532 ap->ops->irq_clear(ap);
533
Tejun Heof15a1da2006-05-15 20:57:56 +0900534 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
535 "stat 0x%x host_stat 0x%x\n",
536 qc->tf.command, drv_stat, host_stat);
Tejun Heoece1d632006-04-02 18:51:53 +0900537
538 /* complete taskfile transaction */
Jeff Garzikc13b56a2006-04-02 10:34:24 -0400539 qc->err_mask |= AC_ERR_TIMEOUT;
Tejun Heoece1d632006-04-02 18:51:53 +0900540 break;
541 }
542
Jeff Garzikba6a1302006-06-22 23:46:10 -0400543 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900544
545 ata_eh_qc_complete(qc);
546
547 DPRINTK("EXIT\n");
548}
549
550/**
551 * ata_eng_timeout - Handle timeout of queued command
552 * @ap: Port on which timed-out command is active
553 *
554 * Some part of the kernel (currently, only the SCSI layer)
555 * has noticed that the active command on port @ap has not
556 * completed after a specified length of time. Handle this
557 * condition by disabling DMA (if necessary) and completing
558 * transactions, with error if necessary.
559 *
560 * This also handles the case of the "lost interrupt", where
561 * for some reason (possibly hardware bug, possibly driver bug)
562 * an interrupt was not delivered to the driver, even though the
563 * transaction completed successfully.
564 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900565 * TODO: kill this function once old EH is gone.
566 *
Tejun Heoece1d632006-04-02 18:51:53 +0900567 * LOCKING:
568 * Inherited from SCSI layer (none, can sleep)
569 */
570void ata_eng_timeout(struct ata_port *ap)
571{
572 DPRINTK("ENTER\n");
573
574 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
575
576 DPRINTK("EXIT\n");
577}
578
Tejun Heof686bcb2006-05-15 20:58:05 +0900579/**
580 * ata_qc_schedule_eh - schedule qc for error handling
581 * @qc: command to schedule error handling for
582 *
583 * Schedule error handling for @qc. EH will kick in as soon as
584 * other commands are drained.
585 *
586 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400587 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900588 */
589void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
590{
591 struct ata_port *ap = qc->ap;
592
593 WARN_ON(!ap->ops->error_handler);
594
595 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heob51e9e52006-06-29 01:29:30 +0900596 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
Tejun Heof686bcb2006-05-15 20:58:05 +0900597
598 /* The following will fail if timeout has already expired.
599 * ata_scsi_error() takes care of such scmds on EH entry.
600 * Note that ATA_QCFLAG_FAILED is unconditionally set after
601 * this function completes.
602 */
603 scsi_req_abort_cmd(qc->scsicmd);
604}
605
Tejun Heo7b70fc02006-05-15 20:58:07 +0900606/**
607 * ata_port_schedule_eh - schedule error handling without a qc
608 * @ap: ATA port to schedule EH for
609 *
610 * Schedule error handling for @ap. EH will kick in as soon as
611 * all commands are drained.
612 *
613 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400614 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900615 */
616void ata_port_schedule_eh(struct ata_port *ap)
617{
618 WARN_ON(!ap->ops->error_handler);
619
Tejun Heof4d6d002007-05-01 11:50:15 +0200620 if (ap->pflags & ATA_PFLAG_INITIALIZING)
621 return;
622
Tejun Heob51e9e52006-06-29 01:29:30 +0900623 ap->pflags |= ATA_PFLAG_EH_PENDING;
Jeff Garzikcca39742006-08-24 03:19:22 -0400624 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900625
626 DPRINTK("port EH scheduled\n");
627}
628
629/**
630 * ata_port_abort - abort all qc's on the port
631 * @ap: ATA port to abort qc's for
632 *
633 * Abort all active qc's of @ap and schedule EH.
634 *
635 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400636 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900637 *
638 * RETURNS:
639 * Number of aborted qc's.
640 */
641int ata_port_abort(struct ata_port *ap)
642{
643 int tag, nr_aborted = 0;
644
645 WARN_ON(!ap->ops->error_handler);
646
647 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
648 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
649
650 if (qc) {
651 qc->flags |= ATA_QCFLAG_FAILED;
652 ata_qc_complete(qc);
653 nr_aborted++;
654 }
655 }
656
657 if (!nr_aborted)
658 ata_port_schedule_eh(ap);
659
660 return nr_aborted;
661}
662
Tejun Heoe3180492006-05-15 20:58:09 +0900663/**
664 * __ata_port_freeze - freeze port
665 * @ap: ATA port to freeze
666 *
667 * This function is called when HSM violation or some other
668 * condition disrupts normal operation of the port. Frozen port
669 * is not allowed to perform any operation until the port is
670 * thawed, which usually follows a successful reset.
671 *
672 * ap->ops->freeze() callback can be used for freezing the port
673 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
674 * port cannot be frozen hardware-wise, the interrupt handler
675 * must ack and clear interrupts unconditionally while the port
676 * is frozen.
677 *
678 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400679 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900680 */
681static void __ata_port_freeze(struct ata_port *ap)
682{
683 WARN_ON(!ap->ops->error_handler);
684
685 if (ap->ops->freeze)
686 ap->ops->freeze(ap);
687
Tejun Heob51e9e52006-06-29 01:29:30 +0900688 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900689
Tejun Heo44877b42007-02-21 01:06:51 +0900690 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900691}
692
693/**
694 * ata_port_freeze - abort & freeze port
695 * @ap: ATA port to freeze
696 *
697 * Abort and freeze @ap.
698 *
699 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400700 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900701 *
702 * RETURNS:
703 * Number of aborted commands.
704 */
705int ata_port_freeze(struct ata_port *ap)
706{
707 int nr_aborted;
708
709 WARN_ON(!ap->ops->error_handler);
710
711 nr_aborted = ata_port_abort(ap);
712 __ata_port_freeze(ap);
713
714 return nr_aborted;
715}
716
717/**
718 * ata_eh_freeze_port - EH helper to freeze port
719 * @ap: ATA port to freeze
720 *
721 * Freeze @ap.
722 *
723 * LOCKING:
724 * None.
725 */
726void ata_eh_freeze_port(struct ata_port *ap)
727{
728 unsigned long flags;
729
730 if (!ap->ops->error_handler)
731 return;
732
Jeff Garzikba6a1302006-06-22 23:46:10 -0400733 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900734 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400735 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900736}
737
738/**
739 * ata_port_thaw_port - EH helper to thaw port
740 * @ap: ATA port to thaw
741 *
742 * Thaw frozen port @ap.
743 *
744 * LOCKING:
745 * None.
746 */
747void ata_eh_thaw_port(struct ata_port *ap)
748{
749 unsigned long flags;
750
751 if (!ap->ops->error_handler)
752 return;
753
Jeff Garzikba6a1302006-06-22 23:46:10 -0400754 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900755
Tejun Heob51e9e52006-06-29 01:29:30 +0900756 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900757
758 if (ap->ops->thaw)
759 ap->ops->thaw(ap);
760
Jeff Garzikba6a1302006-06-22 23:46:10 -0400761 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900762
Tejun Heo44877b42007-02-21 01:06:51 +0900763 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900764}
765
Tejun Heoece1d632006-04-02 18:51:53 +0900766static void ata_eh_scsidone(struct scsi_cmnd *scmd)
767{
768 /* nada */
769}
770
771static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
772{
773 struct ata_port *ap = qc->ap;
774 struct scsi_cmnd *scmd = qc->scsicmd;
775 unsigned long flags;
776
Jeff Garzikba6a1302006-06-22 23:46:10 -0400777 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900778 qc->scsidone = ata_eh_scsidone;
779 __ata_qc_complete(qc);
780 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -0400781 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900782
783 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
784}
785
786/**
787 * ata_eh_qc_complete - Complete an active ATA command from EH
788 * @qc: Command to complete
789 *
790 * Indicate to the mid and upper layers that an ATA command has
791 * completed. To be used from EH.
792 */
793void ata_eh_qc_complete(struct ata_queued_cmd *qc)
794{
795 struct scsi_cmnd *scmd = qc->scsicmd;
796 scmd->retries = scmd->allowed;
797 __ata_eh_qc_complete(qc);
798}
799
800/**
801 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
802 * @qc: Command to retry
803 *
804 * Indicate to the mid and upper layers that an ATA command
805 * should be retried. To be used from EH.
806 *
807 * SCSI midlayer limits the number of retries to scmd->allowed.
808 * scmd->retries is decremented for commands which get retried
809 * due to unrelated failures (qc->err_mask is zero).
810 */
811void ata_eh_qc_retry(struct ata_queued_cmd *qc)
812{
813 struct scsi_cmnd *scmd = qc->scsicmd;
814 if (!qc->err_mask && scmd->retries)
815 scmd->retries--;
816 __ata_eh_qc_complete(qc);
817}
Tejun Heo022bdb02006-05-15 20:58:22 +0900818
819/**
Tejun Heo0ea035a2006-05-31 18:28:01 +0900820 * ata_eh_detach_dev - detach ATA device
821 * @dev: ATA device to detach
822 *
823 * Detach @dev.
824 *
825 * LOCKING:
826 * None.
827 */
828static void ata_eh_detach_dev(struct ata_device *dev)
829{
830 struct ata_port *ap = dev->ap;
831 unsigned long flags;
832
833 ata_dev_disable(dev);
834
Jeff Garzikba6a1302006-06-22 23:46:10 -0400835 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +0900836
837 dev->flags &= ~ATA_DFLAG_DETACH;
838
839 if (ata_scsi_offline_dev(dev)) {
840 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +0900841 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +0900842 }
843
Tejun Heobeb07c12006-06-24 20:30:19 +0900844 /* clear per-dev EH actions */
845 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
846 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
847
Jeff Garzikba6a1302006-06-22 23:46:10 -0400848 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +0900849}
850
851/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900852 * ata_eh_about_to_do - about to perform eh_action
853 * @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +0900854 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +0900855 * @action: action about to be performed
856 *
857 * Called just before performing EH actions to clear related bits
858 * in @ap->eh_info such that eh actions are not unnecessarily
859 * repeated.
860 *
861 * LOCKING:
862 * None.
863 */
Tejun Heo47005f22006-06-19 18:27:23 +0900864static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
865 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +0900866{
867 unsigned long flags;
Tejun Heo13abf502006-07-10 23:18:46 +0900868 struct ata_eh_info *ehi = &ap->eh_info;
869 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +0900870
Jeff Garzikba6a1302006-06-22 23:46:10 -0400871 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900872
Tejun Heo13abf502006-07-10 23:18:46 +0900873 /* Reset is represented by combination of actions and EHI
874 * flags. Suck in all related bits before clearing eh_info to
875 * avoid losing requested action.
876 */
877 if (action & ATA_EH_RESET_MASK) {
878 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
879 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900880
Tejun Heo13abf502006-07-10 23:18:46 +0900881 /* make sure all reset actions are cleared & clear EHI flags */
882 action |= ATA_EH_RESET_MASK;
883 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
884 }
885
886 ata_eh_clear_action(dev, ehi, action);
887
888 if (!(ehc->i.flags & ATA_EHI_QUIET))
Tejun Heo1cdaf532006-07-03 16:07:26 +0900889 ap->pflags |= ATA_PFLAG_RECOVERED;
890
Jeff Garzikba6a1302006-06-22 23:46:10 -0400891 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +0900892}
893
894/**
Tejun Heo47005f22006-06-19 18:27:23 +0900895 * ata_eh_done - EH action complete
896 * @ap: target ATA port
897 * @dev: target ATA dev for per-dev action (can be NULL)
898 * @action: action just completed
899 *
900 * Called right after performing EH actions to clear related bits
901 * in @ap->eh_context.
902 *
903 * LOCKING:
904 * None.
905 */
906static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
907 unsigned int action)
908{
Tejun Heo13abf502006-07-10 23:18:46 +0900909 /* if reset is complete, clear all reset actions & reset modifier */
910 if (action & ATA_EH_RESET_MASK) {
911 action |= ATA_EH_RESET_MASK;
912 ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
913 }
914
Tejun Heo47005f22006-06-19 18:27:23 +0900915 ata_eh_clear_action(dev, &ap->eh_context.i, action);
916}
917
918/**
Tejun Heo022bdb02006-05-15 20:58:22 +0900919 * ata_err_string - convert err_mask to descriptive string
920 * @err_mask: error mask to convert to string
921 *
922 * Convert @err_mask to descriptive string. Errors are
923 * prioritized according to severity and only the most severe
924 * error is reported.
925 *
926 * LOCKING:
927 * None.
928 *
929 * RETURNS:
930 * Descriptive string for @err_mask
931 */
932static const char * ata_err_string(unsigned int err_mask)
933{
934 if (err_mask & AC_ERR_HOST_BUS)
935 return "host bus error";
936 if (err_mask & AC_ERR_ATA_BUS)
937 return "ATA bus error";
938 if (err_mask & AC_ERR_TIMEOUT)
939 return "timeout";
940 if (err_mask & AC_ERR_HSM)
941 return "HSM violation";
942 if (err_mask & AC_ERR_SYSTEM)
943 return "internal error";
944 if (err_mask & AC_ERR_MEDIA)
945 return "media error";
946 if (err_mask & AC_ERR_INVALID)
947 return "invalid argument";
948 if (err_mask & AC_ERR_DEV)
949 return "device error";
950 return "unknown error";
951}
952
953/**
Tejun Heoe8ee8452006-05-15 21:03:46 +0900954 * ata_read_log_page - read a specific log page
955 * @dev: target device
956 * @page: page to read
957 * @buf: buffer to store read page
958 * @sectors: number of sectors to read
959 *
960 * Read log page using READ_LOG_EXT command.
961 *
962 * LOCKING:
963 * Kernel thread context (may sleep).
964 *
965 * RETURNS:
966 * 0 on success, AC_ERR_* mask otherwise.
967 */
968static unsigned int ata_read_log_page(struct ata_device *dev,
969 u8 page, void *buf, unsigned int sectors)
970{
971 struct ata_taskfile tf;
972 unsigned int err_mask;
973
974 DPRINTK("read log page - page %d\n", page);
975
976 ata_tf_init(dev, &tf);
977 tf.command = ATA_CMD_READ_LOG_EXT;
978 tf.lbal = page;
979 tf.nsect = sectors;
980 tf.hob_nsect = sectors >> 8;
981 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
982 tf.protocol = ATA_PROT_PIO;
983
984 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
985 buf, sectors * ATA_SECT_SIZE);
986
987 DPRINTK("EXIT, err_mask=%x\n", err_mask);
988 return err_mask;
989}
990
991/**
992 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
993 * @dev: Device to read log page 10h from
994 * @tag: Resulting tag of the failed command
995 * @tf: Resulting taskfile registers of the failed command
996 *
997 * Read log page 10h to obtain NCQ error details and clear error
998 * condition.
999 *
1000 * LOCKING:
1001 * Kernel thread context (may sleep).
1002 *
1003 * RETURNS:
1004 * 0 on success, -errno otherwise.
1005 */
1006static int ata_eh_read_log_10h(struct ata_device *dev,
1007 int *tag, struct ata_taskfile *tf)
1008{
1009 u8 *buf = dev->ap->sector_buf;
1010 unsigned int err_mask;
1011 u8 csum;
1012 int i;
1013
1014 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1015 if (err_mask)
1016 return -EIO;
1017
1018 csum = 0;
1019 for (i = 0; i < ATA_SECT_SIZE; i++)
1020 csum += buf[i];
1021 if (csum)
1022 ata_dev_printk(dev, KERN_WARNING,
1023 "invalid checksum 0x%x on log page 10h\n", csum);
1024
1025 if (buf[0] & 0x80)
1026 return -ENOENT;
1027
1028 *tag = buf[0] & 0x1f;
1029
1030 tf->command = buf[2];
1031 tf->feature = buf[3];
1032 tf->lbal = buf[4];
1033 tf->lbam = buf[5];
1034 tf->lbah = buf[6];
1035 tf->device = buf[7];
1036 tf->hob_lbal = buf[8];
1037 tf->hob_lbam = buf[9];
1038 tf->hob_lbah = buf[10];
1039 tf->nsect = buf[12];
1040 tf->hob_nsect = buf[13];
1041
1042 return 0;
1043}
1044
1045/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001046 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1047 * @dev: device to perform REQUEST_SENSE to
1048 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1049 *
1050 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1051 * SENSE. This function is EH helper.
1052 *
1053 * LOCKING:
1054 * Kernel thread context (may sleep).
1055 *
1056 * RETURNS:
1057 * 0 on success, AC_ERR_* mask on failure
1058 */
Albert Lee56287762007-04-02 11:30:46 +08001059static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
Tejun Heo022bdb02006-05-15 20:58:22 +09001060{
Albert Lee56287762007-04-02 11:30:46 +08001061 struct ata_device *dev = qc->dev;
1062 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
Tejun Heo022bdb02006-05-15 20:58:22 +09001063 struct ata_port *ap = dev->ap;
1064 struct ata_taskfile tf;
1065 u8 cdb[ATAPI_CDB_LEN];
1066
1067 DPRINTK("ATAPI request sense\n");
1068
Tejun Heo022bdb02006-05-15 20:58:22 +09001069 /* FIXME: is this needed? */
1070 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1071
Albert Lee56287762007-04-02 11:30:46 +08001072 /* initialize sense_buf with the error register,
1073 * for the case where they are -not- overwritten
1074 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001075 sense_buf[0] = 0x70;
Albert Lee56287762007-04-02 11:30:46 +08001076 sense_buf[2] = qc->result_tf.feature >> 4;
1077
Jeff Garzika617c092007-05-21 20:14:23 -04001078 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001079 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001080
1081 memset(cdb, 0, ATAPI_CDB_LEN);
1082 cdb[0] = REQUEST_SENSE;
1083 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1084
1085 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1086 tf.command = ATA_CMD_PACKET;
1087
1088 /* is it pointless to prefer PIO for "safety reasons"? */
1089 if (ap->flags & ATA_FLAG_PIO_DMA) {
1090 tf.protocol = ATA_PROT_ATAPI_DMA;
1091 tf.feature |= ATAPI_PKT_DMA;
1092 } else {
1093 tf.protocol = ATA_PROT_ATAPI;
1094 tf.lbam = (8 * 1024) & 0xff;
1095 tf.lbah = (8 * 1024) >> 8;
1096 }
1097
1098 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1099 sense_buf, SCSI_SENSE_BUFFERSIZE);
1100}
1101
1102/**
1103 * ata_eh_analyze_serror - analyze SError for a failed port
1104 * @ap: ATA port to analyze SError for
1105 *
1106 * Analyze SError if available and further determine cause of
1107 * failure.
1108 *
1109 * LOCKING:
1110 * None.
1111 */
1112static void ata_eh_analyze_serror(struct ata_port *ap)
1113{
1114 struct ata_eh_context *ehc = &ap->eh_context;
1115 u32 serror = ehc->i.serror;
1116 unsigned int err_mask = 0, action = 0;
1117
1118 if (serror & SERR_PERSISTENT) {
1119 err_mask |= AC_ERR_ATA_BUS;
1120 action |= ATA_EH_HARDRESET;
1121 }
1122 if (serror &
1123 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1124 err_mask |= AC_ERR_ATA_BUS;
1125 action |= ATA_EH_SOFTRESET;
1126 }
1127 if (serror & SERR_PROTOCOL) {
1128 err_mask |= AC_ERR_HSM;
1129 action |= ATA_EH_SOFTRESET;
1130 }
1131 if (serror & SERR_INTERNAL) {
1132 err_mask |= AC_ERR_SYSTEM;
Tejun Heo771b8da2007-03-14 01:20:51 +09001133 action |= ATA_EH_HARDRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001134 }
Tejun Heo084fe632006-05-31 18:28:03 +09001135 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1136 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001137
1138 ehc->i.err_mask |= err_mask;
1139 ehc->i.action |= action;
1140}
1141
1142/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001143 * ata_eh_analyze_ncq_error - analyze NCQ error
1144 * @ap: ATA port to analyze NCQ error for
1145 *
1146 * Read log page 10h, determine the offending qc and acquire
1147 * error status TF. For NCQ device errors, all LLDDs have to do
1148 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1149 * care of the rest.
1150 *
1151 * LOCKING:
1152 * Kernel thread context (may sleep).
1153 */
1154static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1155{
1156 struct ata_eh_context *ehc = &ap->eh_context;
1157 struct ata_device *dev = ap->device;
1158 struct ata_queued_cmd *qc;
1159 struct ata_taskfile tf;
1160 int tag, rc;
1161
1162 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001163 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001164 return;
1165
1166 /* is it NCQ device error? */
1167 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1168 return;
1169
1170 /* has LLDD analyzed already? */
1171 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1172 qc = __ata_qc_from_tag(ap, tag);
1173
1174 if (!(qc->flags & ATA_QCFLAG_FAILED))
1175 continue;
1176
1177 if (qc->err_mask)
1178 return;
1179 }
1180
1181 /* okay, this error is ours */
1182 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1183 if (rc) {
1184 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1185 "(errno=%d)\n", rc);
1186 return;
1187 }
1188
1189 if (!(ap->sactive & (1 << tag))) {
1190 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1191 "inactive tag %d\n", tag);
1192 return;
1193 }
1194
1195 /* we've got the perpetrator, condemn it */
1196 qc = __ata_qc_from_tag(ap, tag);
1197 memcpy(&qc->result_tf, &tf, sizeof(tf));
Tejun Heo5335b722007-07-16 14:29:40 +09001198 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001199 ehc->i.err_mask &= ~AC_ERR_DEV;
1200}
1201
1202/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001203 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1204 * @qc: qc to analyze
1205 * @tf: Taskfile registers to analyze
1206 *
1207 * Analyze taskfile of @qc and further determine cause of
1208 * failure. This function also requests ATAPI sense data if
1209 * avaliable.
1210 *
1211 * LOCKING:
1212 * Kernel thread context (may sleep).
1213 *
1214 * RETURNS:
1215 * Determined recovery action
1216 */
1217static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1218 const struct ata_taskfile *tf)
1219{
1220 unsigned int tmp, action = 0;
1221 u8 stat = tf->command, err = tf->feature;
1222
1223 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1224 qc->err_mask |= AC_ERR_HSM;
1225 return ATA_EH_SOFTRESET;
1226 }
1227
Tejun Heoa51d6442007-03-20 15:24:11 +09001228 if (stat & (ATA_ERR | ATA_DF))
1229 qc->err_mask |= AC_ERR_DEV;
1230 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001231 return 0;
1232
1233 switch (qc->dev->class) {
1234 case ATA_DEV_ATA:
1235 if (err & ATA_ICRC)
1236 qc->err_mask |= AC_ERR_ATA_BUS;
1237 if (err & ATA_UNC)
1238 qc->err_mask |= AC_ERR_MEDIA;
1239 if (err & ATA_IDNF)
1240 qc->err_mask |= AC_ERR_INVALID;
1241 break;
1242
1243 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001244 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Albert Lee56287762007-04-02 11:30:46 +08001245 tmp = atapi_eh_request_sense(qc);
Tejun Heoa569a302006-11-21 10:40:51 +09001246 if (!tmp) {
1247 /* ATA_QCFLAG_SENSE_VALID is used to
1248 * tell atapi_qc_complete() that sense
1249 * data is already valid.
1250 *
1251 * TODO: interpret sense data and set
1252 * appropriate err_mask.
1253 */
1254 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1255 } else
1256 qc->err_mask |= tmp;
1257 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001258 }
1259
1260 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1261 action |= ATA_EH_SOFTRESET;
1262
1263 return action;
1264}
1265
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001266static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001267{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001268 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo022bdb02006-05-15 20:58:22 +09001269 return 1;
1270
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001271 if (err_mask & AC_ERR_TIMEOUT)
1272 return 2;
1273
1274 if (is_io) {
1275 if (err_mask & AC_ERR_HSM)
Tejun Heo022bdb02006-05-15 20:58:22 +09001276 return 2;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001277 if ((err_mask &
1278 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1279 return 3;
Tejun Heo022bdb02006-05-15 20:58:22 +09001280 }
1281
1282 return 0;
1283}
1284
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001285struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001286 u64 since;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001287 int nr_errors[4];
Tejun Heo022bdb02006-05-15 20:58:22 +09001288};
1289
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001290static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001291{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001292 struct speed_down_verdict_arg *arg = void_arg;
1293 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
Tejun Heo022bdb02006-05-15 20:58:22 +09001294
1295 if (ent->timestamp < arg->since)
1296 return -1;
1297
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001298 arg->nr_errors[cat]++;
Tejun Heo022bdb02006-05-15 20:58:22 +09001299 return 0;
1300}
1301
1302/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001303 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001304 * @dev: Device of interest
1305 *
1306 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001307 * whether NCQ needs to be turned off, transfer speed should be
1308 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001309 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001310 * Cat-1 is ATA_BUS error for any command.
Tejun Heo022bdb02006-05-15 20:58:22 +09001311 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001312 * Cat-2 is TIMEOUT for any command or HSM violation for known
1313 * supported commands.
1314 *
1315 * Cat-3 is is unclassified DEV error for known supported
Tejun Heo022bdb02006-05-15 20:58:22 +09001316 * command.
1317 *
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001318 * NCQ needs to be turned off if there have been more than 3
1319 * Cat-2 + Cat-3 errors during last 10 minutes.
1320 *
1321 * Speed down is necessary if there have been more than 3 Cat-1 +
1322 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1323 *
1324 * Falling back to PIO mode is necessary if there have been more
1325 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1326 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001327 * LOCKING:
1328 * Inherited from caller.
1329 *
1330 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001331 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001332 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001333static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001334{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001335 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1336 u64 j64 = get_jiffies_64();
1337 struct speed_down_verdict_arg arg;
1338 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001339
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001340 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001341 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001342 arg.since = j64 - min(j64, j10mins);
1343 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001344
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001345 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1346 verdict |= ATA_EH_SPDN_NCQ_OFF;
1347 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1348 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001349
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001350 /* scan past 3 mins of error history */
1351 memset(&arg, 0, sizeof(arg));
1352 arg.since = j64 - min(j64, j5mins);
1353 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001354
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001355 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1356 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1357
1358 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001359}
1360
1361/**
1362 * ata_eh_speed_down - record error and speed down if necessary
1363 * @dev: Failed device
1364 * @is_io: Did the device fail during normal IO?
1365 * @err_mask: err_mask of the error
1366 *
1367 * Record error and examine error history to determine whether
1368 * adjusting transmission speed is necessary. It also sets
1369 * transmission limits appropriately if such adjustment is
1370 * necessary.
1371 *
1372 * LOCKING:
1373 * Kernel thread context (may sleep).
1374 *
1375 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001376 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001377 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001378static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1379 unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001380{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001381 unsigned int verdict;
1382 unsigned int action = 0;
1383
1384 /* don't bother if Cat-0 error */
1385 if (ata_eh_categorize_error(is_io, err_mask) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001386 return 0;
1387
1388 /* record error and determine whether speed down is necessary */
1389 ata_ering_record(&dev->ering, is_io, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001390 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001391
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001392 /* turn off NCQ? */
1393 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1394 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1395 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1396 dev->flags |= ATA_DFLAG_NCQ_OFF;
1397 ata_dev_printk(dev, KERN_WARNING,
1398 "NCQ disabled due to excessive errors\n");
1399 goto done;
1400 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001401
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001402 /* speed down? */
1403 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1404 /* speed down SATA link speed if possible */
1405 if (sata_down_spd_limit(dev->ap) == 0) {
1406 action |= ATA_EH_HARDRESET;
1407 goto done;
1408 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001409
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001410 /* lower transfer mode */
1411 if (dev->spdn_cnt < 2) {
1412 static const int dma_dnxfer_sel[] =
1413 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1414 static const int pio_dnxfer_sel[] =
1415 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1416 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001417
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001418 if (dev->xfer_shift != ATA_SHIFT_PIO)
1419 sel = dma_dnxfer_sel[dev->spdn_cnt];
1420 else
1421 sel = pio_dnxfer_sel[dev->spdn_cnt];
1422
1423 dev->spdn_cnt++;
1424
1425 if (ata_down_xfermask_limit(dev, sel) == 0) {
1426 action |= ATA_EH_SOFTRESET;
1427 goto done;
1428 }
1429 }
1430 }
1431
1432 /* Fall back to PIO? Slowing down to PIO is meaningless for
1433 * SATA. Consider it only for PATA.
1434 */
1435 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1436 (dev->ap->cbl != ATA_CBL_SATA) &&
1437 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1438 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1439 dev->spdn_cnt = 0;
1440 action |= ATA_EH_SOFTRESET;
1441 goto done;
1442 }
1443 }
1444
Tejun Heo022bdb02006-05-15 20:58:22 +09001445 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001446 done:
1447 /* device has been slowed down, blow error history */
1448 ata_ering_clear(&dev->ering);
1449 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001450}
1451
1452/**
1453 * ata_eh_autopsy - analyze error and determine recovery action
1454 * @ap: ATA port to perform autopsy on
1455 *
1456 * Analyze why @ap failed and determine which recovery action is
1457 * needed. This function also sets more detailed AC_ERR_* values
1458 * and fills sense data for ATAPI CHECK SENSE.
1459 *
1460 * LOCKING:
1461 * Kernel thread context (may sleep).
1462 */
1463static void ata_eh_autopsy(struct ata_port *ap)
1464{
1465 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001466 unsigned int all_err_mask = 0;
1467 int tag, is_io = 0;
1468 u32 serror;
1469 int rc;
1470
1471 DPRINTK("ENTER\n");
1472
Tejun Heo1cdaf532006-07-03 16:07:26 +09001473 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1474 return;
1475
Tejun Heo022bdb02006-05-15 20:58:22 +09001476 /* obtain and analyze SError */
1477 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1478 if (rc == 0) {
1479 ehc->i.serror |= serror;
1480 ata_eh_analyze_serror(ap);
1481 } else if (rc != -EOPNOTSUPP)
Tejun Heo4528e4d2006-07-08 20:17:26 +09001482 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001483
Tejun Heoe8ee8452006-05-15 21:03:46 +09001484 /* analyze NCQ failure */
1485 ata_eh_analyze_ncq_error(ap);
1486
Tejun Heo022bdb02006-05-15 20:58:22 +09001487 /* any real error trumps AC_ERR_OTHER */
1488 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1489 ehc->i.err_mask &= ~AC_ERR_OTHER;
1490
1491 all_err_mask |= ehc->i.err_mask;
1492
1493 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1494 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1495
1496 if (!(qc->flags & ATA_QCFLAG_FAILED))
1497 continue;
1498
1499 /* inherit upper level err_mask */
1500 qc->err_mask |= ehc->i.err_mask;
1501
Tejun Heo022bdb02006-05-15 20:58:22 +09001502 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001503 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001504
1505 /* DEV errors are probably spurious in case of ATA_BUS error */
1506 if (qc->err_mask & AC_ERR_ATA_BUS)
1507 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1508 AC_ERR_INVALID);
1509
1510 /* any real error trumps unknown error */
1511 if (qc->err_mask & ~AC_ERR_OTHER)
1512 qc->err_mask &= ~AC_ERR_OTHER;
1513
1514 /* SENSE_VALID trumps dev/unknown error and revalidation */
1515 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1516 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo4528e4d2006-07-08 20:17:26 +09001517 ehc->i.action &= ~ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001518 }
1519
1520 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001521 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001522 all_err_mask |= qc->err_mask;
1523 if (qc->flags & ATA_QCFLAG_IO)
1524 is_io = 1;
1525 }
1526
Tejun Heoa20f33f2006-05-16 12:58:24 +09001527 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001528 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001529 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001530 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heoa20f33f2006-05-16 12:58:24 +09001531 else if (all_err_mask)
Tejun Heo4528e4d2006-07-08 20:17:26 +09001532 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001533
Tejun Heo47005f22006-06-19 18:27:23 +09001534 /* if we have offending qcs and the associated failed device */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001535 if (ehc->i.dev) {
Tejun Heo47005f22006-06-19 18:27:23 +09001536 /* speed down */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001537 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1538 all_err_mask);
Tejun Heo47005f22006-06-19 18:27:23 +09001539
1540 /* perform per-dev EH action only on the offending device */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001541 ehc->i.dev_action[ehc->i.dev->devno] |=
1542 ehc->i.action & ATA_EH_PERDEV_MASK;
1543 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001544 }
1545
Tejun Heo022bdb02006-05-15 20:58:22 +09001546 DPRINTK("EXIT\n");
1547}
1548
1549/**
1550 * ata_eh_report - report error handling to user
1551 * @ap: ATA port EH is going on
1552 *
1553 * Report EH to user.
1554 *
1555 * LOCKING:
1556 * None.
1557 */
1558static void ata_eh_report(struct ata_port *ap)
1559{
1560 struct ata_eh_context *ehc = &ap->eh_context;
1561 const char *frozen, *desc;
1562 int tag, nr_failed = 0;
1563
1564 desc = NULL;
1565 if (ehc->i.desc[0] != '\0')
1566 desc = ehc->i.desc;
1567
1568 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1569 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1570
1571 if (!(qc->flags & ATA_QCFLAG_FAILED))
1572 continue;
1573 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1574 continue;
1575
1576 nr_failed++;
1577 }
1578
1579 if (!nr_failed && !ehc->i.err_mask)
1580 return;
1581
1582 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001583 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001584 frozen = " frozen";
1585
1586 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001587 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1588 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1589 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1590 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001591 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09001592 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001593 } else {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001594 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1595 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1596 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1597 ehc->i.action, frozen);
Tejun Heo022bdb02006-05-15 20:58:22 +09001598 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09001599 ata_port_printk(ap, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001600 }
1601
1602 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
Tejun Heo8a937582006-11-14 22:36:12 +09001603 static const char *dma_str[] = {
1604 [DMA_BIDIRECTIONAL] = "bidi",
1605 [DMA_TO_DEVICE] = "out",
1606 [DMA_FROM_DEVICE] = "in",
1607 [DMA_NONE] = "",
1608 };
Tejun Heo022bdb02006-05-15 20:58:22 +09001609 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09001610 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001611
1612 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1613 continue;
1614
Tejun Heo8a937582006-11-14 22:36:12 +09001615 ata_dev_printk(qc->dev, KERN_ERR,
1616 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo664e8502006-11-20 16:05:34 +09001617 "tag %d cdb 0x%x data %u %s\n "
Tejun Heo8a937582006-11-14 22:36:12 +09001618 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09001619 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09001620 cmd->command, cmd->feature, cmd->nsect,
1621 cmd->lbal, cmd->lbam, cmd->lbah,
1622 cmd->hob_feature, cmd->hob_nsect,
1623 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heo726f0782007-01-03 17:30:39 +09001624 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
Tejun Heo664e8502006-11-20 16:05:34 +09001625 dma_str[qc->dma_dir],
Tejun Heo8a937582006-11-14 22:36:12 +09001626 res->command, res->feature, res->nsect,
1627 res->lbal, res->lbam, res->lbah,
1628 res->hob_feature, res->hob_nsect,
1629 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09001630 res->device, qc->err_mask, ata_err_string(qc->err_mask),
1631 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Tejun Heo022bdb02006-05-15 20:58:22 +09001632 }
1633}
1634
Tejun Heod87fa382006-05-31 18:28:24 +09001635static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
Tejun Heod4b2bab2007-02-02 16:50:52 +09001636 unsigned int *classes, unsigned long deadline)
Tejun Heod87fa382006-05-31 18:28:24 +09001637{
1638 int i, rc;
1639
1640 for (i = 0; i < ATA_MAX_DEVICES; i++)
1641 classes[i] = ATA_DEV_UNKNOWN;
1642
Tejun Heod4b2bab2007-02-02 16:50:52 +09001643 rc = reset(ap, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09001644 if (rc)
1645 return rc;
1646
1647 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1648 * is complete and convert all ATA_DEV_UNKNOWN to
1649 * ATA_DEV_NONE.
1650 */
1651 for (i = 0; i < ATA_MAX_DEVICES; i++)
1652 if (classes[i] != ATA_DEV_UNKNOWN)
1653 break;
1654
1655 if (i < ATA_MAX_DEVICES)
1656 for (i = 0; i < ATA_MAX_DEVICES; i++)
1657 if (classes[i] == ATA_DEV_UNKNOWN)
1658 classes[i] = ATA_DEV_NONE;
1659
1660 return 0;
1661}
1662
Tejun Heo664faf02006-05-31 18:27:50 +09001663static int ata_eh_followup_srst_needed(int rc, int classify,
1664 const unsigned int *classes)
1665{
1666 if (rc == -EAGAIN)
1667 return 1;
1668 if (rc != 0)
1669 return 0;
1670 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1671 return 1;
1672 return 0;
1673}
1674
1675static int ata_eh_reset(struct ata_port *ap, int classify,
Tejun Heof5914a42006-05-31 18:27:48 +09001676 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09001677 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1678{
1679 struct ata_eh_context *ehc = &ap->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09001680 unsigned int *classes = ehc->classes;
Tejun Heo1cdaf532006-07-03 16:07:26 +09001681 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heo31daabd2007-02-02 16:50:52 +09001682 int try = 0;
1683 unsigned long deadline;
Tejun Heof5914a42006-05-31 18:27:48 +09001684 unsigned int action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001685 ata_reset_fn_t reset;
Tejun Heo8b5bb2f2007-06-25 21:43:04 +09001686 int i, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09001687
Tejun Heo13abf502006-07-10 23:18:46 +09001688 /* about to reset */
1689 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1690
Tejun Heof5914a42006-05-31 18:27:48 +09001691 /* Determine which reset to use and record in ehc->i.action.
1692 * prereset() may examine and modify it.
1693 */
1694 action = ehc->i.action;
1695 ehc->i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo022bdb02006-05-15 20:58:22 +09001696 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
Tejun Heof5914a42006-05-31 18:27:48 +09001697 !(action & ATA_EH_HARDRESET))))
1698 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001699 else
Tejun Heof5914a42006-05-31 18:27:48 +09001700 ehc->i.action |= ATA_EH_HARDRESET;
1701
1702 if (prereset) {
Tejun Heo31daabd2007-02-02 16:50:52 +09001703 rc = prereset(ap, jiffies + ATA_EH_PRERESET_TIMEOUT);
Tejun Heof5914a42006-05-31 18:27:48 +09001704 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01001705 if (rc == -ENOENT) {
Tejun Heo4aa9ab62007-03-12 17:24:08 +09001706 ata_port_printk(ap, KERN_DEBUG,
1707 "port disabled. ignoring.\n");
Alan Coxc9619222006-09-26 17:53:38 +01001708 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09001709
1710 for (i = 0; i < ATA_MAX_DEVICES; i++)
1711 classes[i] = ATA_DEV_NONE;
1712
1713 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01001714 } else
1715 ata_port_printk(ap, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09001716 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09001717 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09001718 }
1719 }
1720
1721 /* prereset() might have modified ehc->i.action */
1722 if (ehc->i.action & ATA_EH_HARDRESET)
Tejun Heo022bdb02006-05-15 20:58:22 +09001723 reset = hardreset;
Tejun Heof5914a42006-05-31 18:27:48 +09001724 else if (ehc->i.action & ATA_EH_SOFTRESET)
1725 reset = softreset;
1726 else {
1727 /* prereset told us not to reset, bang classes and return */
1728 for (i = 0; i < ATA_MAX_DEVICES; i++)
1729 classes[i] = ATA_DEV_NONE;
Tejun Heofccb6ea2007-07-16 14:29:41 +09001730 rc = 0;
1731 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09001732 }
1733
1734 /* did prereset() screw up? if so, fix up to avoid oopsing */
1735 if (!reset) {
Tejun Heof5914a42006-05-31 18:27:48 +09001736 if (softreset)
1737 reset = softreset;
1738 else
1739 reset = hardreset;
1740 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001741
1742 retry:
Tejun Heo31daabd2007-02-02 16:50:52 +09001743 deadline = jiffies + ata_eh_reset_timeouts[try++];
1744
Tejun Heo3e706392006-05-31 18:28:11 +09001745 /* shut up during boot probing */
1746 if (verbose)
1747 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1748 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09001749
Tejun Heo13abf502006-07-10 23:18:46 +09001750 /* mark that this EH session started with reset */
Tejun Heo0d64a232007-04-23 02:41:05 +09001751 if (reset == hardreset)
1752 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1753 else
1754 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001755
Tejun Heo31daabd2007-02-02 16:50:52 +09001756 rc = ata_do_reset(ap, reset, classes, deadline);
Tejun Heo022bdb02006-05-15 20:58:22 +09001757
Tejun Heo664faf02006-05-31 18:27:50 +09001758 if (reset == hardreset &&
1759 ata_eh_followup_srst_needed(rc, classify, classes)) {
1760 /* okay, let's do follow-up softreset */
Tejun Heo664faf02006-05-31 18:27:50 +09001761 reset = softreset;
1762
1763 if (!reset) {
1764 ata_port_printk(ap, KERN_ERR,
1765 "follow-up softreset required "
1766 "but no softreset avaliable\n");
Tejun Heofccb6ea2007-07-16 14:29:41 +09001767 rc = -EINVAL;
1768 goto out;
Tejun Heo664faf02006-05-31 18:27:50 +09001769 }
1770
Tejun Heo47005f22006-06-19 18:27:23 +09001771 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
Tejun Heo31daabd2007-02-02 16:50:52 +09001772 rc = ata_do_reset(ap, reset, classes, deadline);
Tejun Heo664faf02006-05-31 18:27:50 +09001773
1774 if (rc == 0 && classify &&
1775 classes[0] == ATA_DEV_UNKNOWN) {
1776 ata_port_printk(ap, KERN_ERR,
1777 "classification failed\n");
Tejun Heofccb6ea2007-07-16 14:29:41 +09001778 rc = -EINVAL;
1779 goto out;
Tejun Heo664faf02006-05-31 18:27:50 +09001780 }
1781 }
1782
Tejun Heo31daabd2007-02-02 16:50:52 +09001783 if (rc && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
1784 unsigned long now = jiffies;
Tejun Heo664faf02006-05-31 18:27:50 +09001785
Tejun Heo31daabd2007-02-02 16:50:52 +09001786 if (time_before(now, deadline)) {
1787 unsigned long delta = deadline - jiffies;
Tejun Heo664faf02006-05-31 18:27:50 +09001788
Tejun Heo31daabd2007-02-02 16:50:52 +09001789 ata_port_printk(ap, KERN_WARNING, "reset failed "
1790 "(errno=%d), retrying in %u secs\n",
1791 rc, (jiffies_to_msecs(delta) + 999) / 1000);
Tejun Heo022bdb02006-05-15 20:58:22 +09001792
Tejun Heo31daabd2007-02-02 16:50:52 +09001793 schedule_timeout_uninterruptible(delta);
1794 }
1795
Tejun Heof1545152007-07-16 14:29:40 +09001796 if (rc == -EPIPE ||
Tejun Heo31daabd2007-02-02 16:50:52 +09001797 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
Tejun Heo022bdb02006-05-15 20:58:22 +09001798 sata_down_spd_limit(ap);
1799 if (hardreset)
1800 reset = hardreset;
1801 goto retry;
1802 }
1803
1804 if (rc == 0) {
Tejun Heo008a7892007-07-16 14:29:40 +09001805 u32 sstatus;
1806
Tejun Heo20952b62006-05-31 18:27:23 +09001807 /* After the reset, the device state is PIO 0 and the
1808 * controller state is undefined. Record the mode.
1809 */
1810 for (i = 0; i < ATA_MAX_DEVICES; i++)
1811 ap->device[i].pio_mode = XFER_PIO_0;
1812
Tejun Heo008a7892007-07-16 14:29:40 +09001813 /* record current link speed */
1814 if (sata_scr_read(ap, SCR_STATUS, &sstatus) == 0)
1815 ap->sata_spd = (sstatus >> 4) & 0xf;
1816
Tejun Heo022bdb02006-05-15 20:58:22 +09001817 if (postreset)
1818 postreset(ap, classes);
1819
1820 /* reset successful, schedule revalidation */
Tejun Heo13abf502006-07-10 23:18:46 +09001821 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
Tejun Heo022bdb02006-05-15 20:58:22 +09001822 ehc->i.action |= ATA_EH_REVALIDATE;
1823 }
Tejun Heofccb6ea2007-07-16 14:29:41 +09001824 out:
1825 /* clear hotplug flag */
1826 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heo022bdb02006-05-15 20:58:22 +09001827 return rc;
1828}
1829
Tejun Heo084fe632006-05-31 18:28:03 +09001830static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1831 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001832{
1833 struct ata_eh_context *ehc = &ap->eh_context;
1834 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001835 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09001836 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09001837 int i, rc = 0;
1838
1839 DPRINTK("ENTER\n");
1840
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001841 /* For PATA drive side cable detection to work, IDENTIFY must
1842 * be done backwards such that PDIAG- is released by the slave
1843 * device before the master device is identified.
1844 */
1845 for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) {
Tejun Heobff04642006-11-10 18:08:10 +09001846 unsigned int action, readid_flags = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001847
Tejun Heo47005f22006-06-19 18:27:23 +09001848 dev = &ap->device[i];
Tejun Heo64f65ca2006-06-24 20:30:18 +09001849 action = ata_eh_dev_action(dev);
Tejun Heo47005f22006-06-19 18:27:23 +09001850
Tejun Heobff04642006-11-10 18:08:10 +09001851 if (ehc->i.flags & ATA_EHI_DID_RESET)
1852 readid_flags |= ATA_READID_POSTRESET;
1853
Tejun Heo9666f402007-05-04 21:27:47 +02001854 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001855 if (ata_port_offline(ap)) {
1856 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001857 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09001858 }
1859
Tejun Heo47005f22006-06-19 18:27:23 +09001860 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
Tejun Heobff04642006-11-10 18:08:10 +09001861 rc = ata_dev_revalidate(dev, readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001862 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001863 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09001864
Tejun Heo47005f22006-06-19 18:27:23 +09001865 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1866
Tejun Heobaa1e782006-11-01 18:39:27 +09001867 /* Configuration may have changed, reconfigure
1868 * transfer mode.
1869 */
1870 ehc->i.flags |= ATA_EHI_SETMODE;
1871
zhao, forrest3057ac32006-06-12 12:01:34 +08001872 /* schedule the scsi_rescan_device() here */
1873 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09001874 } else if (dev->class == ATA_DEV_UNKNOWN &&
1875 ehc->tries[dev->devno] &&
1876 ata_class_enabled(ehc->classes[dev->devno])) {
1877 dev->class = ehc->classes[dev->devno];
1878
Tejun Heobff04642006-11-10 18:08:10 +09001879 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1880 dev->id);
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001881 switch (rc) {
1882 case 0:
1883 new_mask |= 1 << i;
1884 break;
1885 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001886 /* IDENTIFY was issued to non-existent
1887 * device. No need to reset. Just
1888 * thaw and kill the device.
1889 */
1890 ata_eh_thaw_port(ap);
1891 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo084fe632006-05-31 18:28:03 +09001892 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001893 default:
1894 dev->class = ATA_DEV_UNKNOWN;
1895 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09001896 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001897 }
1898 }
1899
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09001900 /* PDIAG- should have been released, ask cable type if post-reset */
1901 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
1902 ap->cbl = ap->ops->cable_detect(ap);
1903
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001904 /* Configure new devices forward such that user doesn't see
1905 * device detection messages backwards.
1906 */
1907 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1908 dev = &ap->device[i];
Tejun Heo022bdb02006-05-15 20:58:22 +09001909
Tejun Heo8c3c52a2007-03-22 22:24:19 +09001910 if (!(new_mask & (1 << i)))
1911 continue;
1912
1913 ehc->i.flags |= ATA_EHI_PRINTINFO;
1914 rc = ata_dev_configure(dev);
1915 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1916 if (rc)
1917 goto err;
1918
1919 spin_lock_irqsave(ap->lock, flags);
1920 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1921 spin_unlock_irqrestore(ap->lock, flags);
1922
1923 /* new device discovered, configure xfermode */
1924 ehc->i.flags |= ATA_EHI_SETMODE;
1925 }
1926
1927 return 0;
1928
1929 err:
1930 *r_failed_dev = dev;
1931 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001932 return rc;
1933}
1934
1935static int ata_port_nr_enabled(struct ata_port *ap)
1936{
1937 int i, cnt = 0;
1938
1939 for (i = 0; i < ATA_MAX_DEVICES; i++)
1940 if (ata_dev_enabled(&ap->device[i]))
1941 cnt++;
1942 return cnt;
1943}
1944
Tejun Heo084fe632006-05-31 18:28:03 +09001945static int ata_port_nr_vacant(struct ata_port *ap)
1946{
1947 int i, cnt = 0;
1948
1949 for (i = 0; i < ATA_MAX_DEVICES; i++)
1950 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1951 cnt++;
1952 return cnt;
1953}
1954
1955static int ata_eh_skip_recovery(struct ata_port *ap)
1956{
1957 struct ata_eh_context *ehc = &ap->eh_context;
1958 int i;
1959
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09001960 /* thaw frozen port, resume link and recover failed devices */
1961 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
1962 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
Tejun Heo084fe632006-05-31 18:28:03 +09001963 return 0;
1964
1965 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1966 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1967 struct ata_device *dev = &ap->device[i];
1968
1969 if (dev->class == ATA_DEV_UNKNOWN &&
1970 ehc->classes[dev->devno] != ATA_DEV_NONE)
1971 return 0;
1972 }
1973
1974 return 1;
1975}
1976
Tejun Heofee7ca72007-07-01 19:05:58 +09001977static void ata_eh_handle_dev_fail(struct ata_device *dev, int err)
1978{
1979 struct ata_port *ap = dev->ap;
1980 struct ata_eh_context *ehc = &ap->eh_context;
1981
1982 ehc->tries[dev->devno]--;
1983
1984 switch (err) {
1985 case -ENODEV:
1986 /* device missing or wrong IDENTIFY data, schedule probing */
1987 ehc->i.probe_mask |= (1 << dev->devno);
1988 case -EINVAL:
1989 /* give it just one more chance */
1990 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
1991 case -EIO:
1992 if (ehc->tries[dev->devno] == 1) {
1993 /* This is the last chance, better to slow
1994 * down than lose it.
1995 */
1996 sata_down_spd_limit(ap);
1997 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
1998 }
1999 }
2000
2001 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2002 /* disable device if it has used up all its chances */
2003 ata_dev_disable(dev);
2004
2005 /* detach if offline */
2006 if (ata_port_offline(ap))
2007 ata_eh_detach_dev(dev);
2008
2009 /* probe if requested */
2010 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2011 !(ehc->did_probe_mask & (1 << dev->devno))) {
2012 ata_eh_detach_dev(dev);
2013 ata_dev_init(dev);
2014
2015 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2016 ehc->did_probe_mask |= (1 << dev->devno);
2017 ehc->i.action |= ATA_EH_SOFTRESET;
2018 }
2019 } else {
2020 /* soft didn't work? be haaaaard */
2021 if (ehc->i.flags & ATA_EHI_DID_RESET)
2022 ehc->i.action |= ATA_EH_HARDRESET;
2023 else
2024 ehc->i.action |= ATA_EH_SOFTRESET;
2025 }
2026}
2027
Tejun Heo022bdb02006-05-15 20:58:22 +09002028/**
2029 * ata_eh_recover - recover host port after error
2030 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002031 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002032 * @softreset: softreset method (can be NULL)
2033 * @hardreset: hardreset method (can be NULL)
2034 * @postreset: postreset method (can be NULL)
2035 *
2036 * This is the alpha and omega, eum and yang, heart and soul of
2037 * libata exception handling. On entry, actions required to
Tejun Heo084fe632006-05-31 18:28:03 +09002038 * recover the port and hotplug requests are recorded in
2039 * eh_context. This function executes all the operations with
2040 * appropriate retrials and fallbacks to resurrect failed
2041 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09002042 *
2043 * LOCKING:
2044 * Kernel thread context (may sleep).
2045 *
2046 * RETURNS:
2047 * 0 on success, -errno on failure.
2048 */
Tejun Heof5914a42006-05-31 18:27:48 +09002049static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2050 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
Tejun Heo022bdb02006-05-15 20:58:22 +09002051 ata_postreset_fn_t postreset)
2052{
2053 struct ata_eh_context *ehc = &ap->eh_context;
2054 struct ata_device *dev;
Tejun Heo4ae72a12007-02-02 16:22:30 +09002055 int i, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002056
2057 DPRINTK("ENTER\n");
2058
2059 /* prep for recovery */
2060 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2061 dev = &ap->device[i];
2062
2063 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo084fe632006-05-31 18:28:03 +09002064
Tejun Heo79a55b72007-01-18 17:22:18 +09002065 /* collect port action mask recorded in dev actions */
2066 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
2067 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
2068
Tejun Heo084fe632006-05-31 18:28:03 +09002069 /* process hotplug request */
2070 if (dev->flags & ATA_DFLAG_DETACH)
2071 ata_eh_detach_dev(dev);
2072
2073 if (!ata_dev_enabled(dev) &&
2074 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2075 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2076 ata_eh_detach_dev(dev);
2077 ata_dev_init(dev);
2078 ehc->did_probe_mask |= (1 << dev->devno);
2079 ehc->i.action |= ATA_EH_SOFTRESET;
2080 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002081 }
2082
2083 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09002084 rc = 0;
2085
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002086 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09002087 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002088 goto out;
2089
Tejun Heo022bdb02006-05-15 20:58:22 +09002090 /* skip EH if possible. */
Tejun Heo084fe632006-05-31 18:28:03 +09002091 if (ata_eh_skip_recovery(ap))
Tejun Heo022bdb02006-05-15 20:58:22 +09002092 ehc->i.action = 0;
2093
Tejun Heo084fe632006-05-31 18:28:03 +09002094 for (i = 0; i < ATA_MAX_DEVICES; i++)
2095 ehc->classes[i] = ATA_DEV_UNKNOWN;
2096
Tejun Heo022bdb02006-05-15 20:58:22 +09002097 /* reset */
2098 if (ehc->i.action & ATA_EH_RESET_MASK) {
2099 ata_eh_freeze_port(ap);
2100
Tejun Heo084fe632006-05-31 18:28:03 +09002101 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2102 softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002103 if (rc) {
2104 ata_port_printk(ap, KERN_ERR,
2105 "reset failed, giving up\n");
2106 goto out;
2107 }
2108
2109 ata_eh_thaw_port(ap);
2110 }
2111
Tejun Heo084fe632006-05-31 18:28:03 +09002112 /* revalidate existing devices and attach new ones */
2113 rc = ata_eh_revalidate_and_attach(ap, &dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09002114 if (rc)
2115 goto dev_fail;
2116
Tejun Heobaa1e782006-11-01 18:39:27 +09002117 /* configure transfer mode if necessary */
2118 if (ehc->i.flags & ATA_EHI_SETMODE) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002119 rc = ata_set_mode(ap, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09002120 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09002121 goto dev_fail;
Tejun Heobaa1e782006-11-01 18:39:27 +09002122 ehc->i.flags &= ~ATA_EHI_SETMODE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002123 }
2124
2125 goto out;
2126
2127 dev_fail:
Tejun Heofee7ca72007-07-01 19:05:58 +09002128 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002129
2130 if (ata_port_nr_enabled(ap)) {
2131 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2132 "devices, retrying in 5 secs\n");
2133 ssleep(5);
2134 } else {
2135 /* no device left, repeat fast */
2136 msleep(500);
2137 }
2138
2139 goto retry;
2140
2141 out:
2142 if (rc) {
2143 for (i = 0; i < ATA_MAX_DEVICES; i++)
2144 ata_dev_disable(&ap->device[i]);
2145 }
2146
2147 DPRINTK("EXIT, rc=%d\n", rc);
2148 return rc;
2149}
2150
2151/**
2152 * ata_eh_finish - finish up EH
2153 * @ap: host port to finish EH for
2154 *
2155 * Recovery is complete. Clean up EH states and retry or finish
2156 * failed qcs.
2157 *
2158 * LOCKING:
2159 * None.
2160 */
2161static void ata_eh_finish(struct ata_port *ap)
2162{
2163 int tag;
2164
2165 /* retry or finish qcs */
2166 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2167 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2168
2169 if (!(qc->flags & ATA_QCFLAG_FAILED))
2170 continue;
2171
2172 if (qc->err_mask) {
2173 /* FIXME: Once EH migration is complete,
2174 * generate sense data in this function,
2175 * considering both err_mask and tf.
2176 */
2177 if (qc->err_mask & AC_ERR_INVALID)
2178 ata_eh_qc_complete(qc);
2179 else
2180 ata_eh_qc_retry(qc);
2181 } else {
2182 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2183 ata_eh_qc_complete(qc);
2184 } else {
2185 /* feed zero TF to sense generation */
2186 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2187 ata_eh_qc_retry(qc);
2188 }
2189 }
2190 }
2191}
2192
2193/**
2194 * ata_do_eh - do standard error handling
2195 * @ap: host port to handle error for
Tejun Heof5914a42006-05-31 18:27:48 +09002196 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002197 * @softreset: softreset method (can be NULL)
2198 * @hardreset: hardreset method (can be NULL)
2199 * @postreset: postreset method (can be NULL)
2200 *
2201 * Perform standard error handling sequence.
2202 *
2203 * LOCKING:
2204 * Kernel thread context (may sleep).
2205 */
Tejun Heof5914a42006-05-31 18:27:48 +09002206void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2207 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2208 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002209{
Tejun Heo1cdaf532006-07-03 16:07:26 +09002210 ata_eh_autopsy(ap);
2211 ata_eh_report(ap);
Tejun Heof5914a42006-05-31 18:27:48 +09002212 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
Tejun Heo022bdb02006-05-15 20:58:22 +09002213 ata_eh_finish(ap);
2214}
Tejun Heo500530f2006-07-03 16:07:27 +09002215
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002216#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09002217/**
2218 * ata_eh_handle_port_suspend - perform port suspend operation
2219 * @ap: port to suspend
2220 *
2221 * Suspend @ap.
2222 *
2223 * LOCKING:
2224 * Kernel thread context (may sleep).
2225 */
2226static void ata_eh_handle_port_suspend(struct ata_port *ap)
2227{
2228 unsigned long flags;
2229 int rc = 0;
2230
2231 /* are we suspending? */
2232 spin_lock_irqsave(ap->lock, flags);
2233 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2234 ap->pm_mesg.event == PM_EVENT_ON) {
2235 spin_unlock_irqrestore(ap->lock, flags);
2236 return;
2237 }
2238 spin_unlock_irqrestore(ap->lock, flags);
2239
2240 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2241
Tejun Heo64578a32007-05-15 03:28:16 +09002242 /* tell ACPI we're suspending */
2243 rc = ata_acpi_on_suspend(ap);
2244 if (rc)
2245 goto out;
2246
Tejun Heo500530f2006-07-03 16:07:27 +09002247 /* suspend */
2248 ata_eh_freeze_port(ap);
2249
2250 if (ap->ops->port_suspend)
2251 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2252
Tejun Heo64578a32007-05-15 03:28:16 +09002253 out:
Tejun Heo500530f2006-07-03 16:07:27 +09002254 /* report result */
2255 spin_lock_irqsave(ap->lock, flags);
2256
2257 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2258 if (rc == 0)
2259 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09002260 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09002261 ata_port_schedule_eh(ap);
2262
2263 if (ap->pm_result) {
2264 *ap->pm_result = rc;
2265 ap->pm_result = NULL;
2266 }
2267
2268 spin_unlock_irqrestore(ap->lock, flags);
2269
2270 return;
2271}
2272
2273/**
2274 * ata_eh_handle_port_resume - perform port resume operation
2275 * @ap: port to resume
2276 *
2277 * Resume @ap.
2278 *
Tejun Heo500530f2006-07-03 16:07:27 +09002279 * LOCKING:
2280 * Kernel thread context (may sleep).
2281 */
2282static void ata_eh_handle_port_resume(struct ata_port *ap)
2283{
Tejun Heo500530f2006-07-03 16:07:27 +09002284 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02002285 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09002286
2287 /* are we resuming? */
2288 spin_lock_irqsave(ap->lock, flags);
2289 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2290 ap->pm_mesg.event != PM_EVENT_ON) {
2291 spin_unlock_irqrestore(ap->lock, flags);
2292 return;
2293 }
2294 spin_unlock_irqrestore(ap->lock, flags);
2295
Tejun Heo9666f402007-05-04 21:27:47 +02002296 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09002297
2298 if (ap->ops->port_resume)
2299 rc = ap->ops->port_resume(ap);
2300
Tejun Heo67465442007-05-15 03:28:16 +09002301 /* tell ACPI that we're resuming */
2302 ata_acpi_on_resume(ap);
2303
Tejun Heo9666f402007-05-04 21:27:47 +02002304 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09002305 spin_lock_irqsave(ap->lock, flags);
2306 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2307 if (ap->pm_result) {
2308 *ap->pm_result = rc;
2309 ap->pm_result = NULL;
2310 }
2311 spin_unlock_irqrestore(ap->lock, flags);
2312}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002313#endif /* CONFIG_PM */