blob: 4e31071acc02b15f270a2f76ea752b4a4896f822 [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>
Jeff Garzik28555682007-10-11 17:12:35 -040036#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090037#include <scsi/scsi.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_eh.h>
40#include <scsi/scsi_device.h>
41#include <scsi/scsi_cmnd.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040042#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090043
44#include <linux/libata.h>
45
46#include "libata.h"
47
Tejun Heo7d47e8d2007-02-02 16:22:31 +090048enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090049 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090050 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
51 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
52 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090053 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090054
55 /* error flags */
56 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090057 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Tejun Heo3884f7b2007-11-27 19:28:56 +090058
59 /* error categories */
60 ATA_ECAT_NONE = 0,
61 ATA_ECAT_ATA_BUS = 1,
62 ATA_ECAT_TOUT_HSM = 2,
63 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090064 ATA_ECAT_DUBIOUS_NONE = 4,
65 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
66 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
67 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
68 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090069};
70
Tejun Heo31daabd2007-02-02 16:50:52 +090071/* Waiting in ->prereset can never be reliable. It's sometimes nice
72 * to wait there but it can't be depended upon; otherwise, we wouldn't
73 * be resetting. Just give it enough time for most drives to spin up.
74 */
75enum {
76 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
Tejun Heo5ddf24c2007-07-16 14:29:41 +090077 ATA_EH_FASTDRAIN_INTERVAL = 3 * HZ,
Tejun Heo31daabd2007-02-02 16:50:52 +090078};
79
80/* The following table determines how we sequence resets. Each entry
81 * represents timeout for that try. The first try can be soft or
82 * hardreset. All others are hardreset if available. In most cases
83 * the first reset w/ 10sec timeout should succeed. Following entries
84 * are mostly for error handling, hotplug and retarded devices.
85 */
86static const unsigned long ata_eh_reset_timeouts[] = {
87 10 * HZ, /* most drives spin up by 10sec */
88 10 * HZ, /* > 99% working drives spin up before 20sec */
89 35 * HZ, /* give > 30 secs of idleness for retarded devices */
90 5 * HZ, /* and sweet one last chance */
91 /* > 1 min has elapsed, give up */
92};
93
Tejun Heoad9e2762006-05-15 20:58:12 +090094static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090095#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +090096static void ata_eh_handle_port_suspend(struct ata_port *ap);
97static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090098#else /* CONFIG_PM */
99static void ata_eh_handle_port_suspend(struct ata_port *ap)
100{ }
101
102static void ata_eh_handle_port_resume(struct ata_port *ap)
103{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900104#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900105
Tejun Heob64bbc32007-07-16 14:29:39 +0900106static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
107 va_list args)
108{
109 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
110 ATA_EH_DESC_LEN - ehi->desc_len,
111 fmt, args);
112}
113
114/**
115 * __ata_ehi_push_desc - push error description without adding separator
116 * @ehi: target EHI
117 * @fmt: printf format string
118 *
119 * Format string according to @fmt and append it to @ehi->desc.
120 *
121 * LOCKING:
122 * spin_lock_irqsave(host lock)
123 */
124void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
125{
126 va_list args;
127
128 va_start(args, fmt);
129 __ata_ehi_pushv_desc(ehi, fmt, args);
130 va_end(args);
131}
132
133/**
134 * ata_ehi_push_desc - push error description with separator
135 * @ehi: target EHI
136 * @fmt: printf format string
137 *
138 * Format string according to @fmt and append it to @ehi->desc.
139 * If @ehi->desc is not empty, ", " is added in-between.
140 *
141 * LOCKING:
142 * spin_lock_irqsave(host lock)
143 */
144void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
145{
146 va_list args;
147
148 if (ehi->desc_len)
149 __ata_ehi_push_desc(ehi, ", ");
150
151 va_start(args, fmt);
152 __ata_ehi_pushv_desc(ehi, fmt, args);
153 va_end(args);
154}
155
156/**
157 * ata_ehi_clear_desc - clean error description
158 * @ehi: target EHI
159 *
160 * Clear @ehi->desc.
161 *
162 * LOCKING:
163 * spin_lock_irqsave(host lock)
164 */
165void ata_ehi_clear_desc(struct ata_eh_info *ehi)
166{
167 ehi->desc[0] = '\0';
168 ehi->desc_len = 0;
169}
170
Tejun Heocbcdd872007-08-18 13:14:55 +0900171/**
172 * ata_port_desc - append port description
173 * @ap: target ATA port
174 * @fmt: printf format string
175 *
176 * Format string according to @fmt and append it to port
177 * description. If port description is not empty, " " is added
178 * in-between. This function is to be used while initializing
179 * ata_host. The description is printed on host registration.
180 *
181 * LOCKING:
182 * None.
183 */
184void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
185{
186 va_list args;
187
188 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
189
190 if (ap->link.eh_info.desc_len)
191 __ata_ehi_push_desc(&ap->link.eh_info, " ");
192
193 va_start(args, fmt);
194 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
195 va_end(args);
196}
197
198#ifdef CONFIG_PCI
199
200/**
201 * ata_port_pbar_desc - append PCI BAR description
202 * @ap: target ATA port
203 * @bar: target PCI BAR
204 * @offset: offset into PCI BAR
205 * @name: name of the area
206 *
207 * If @offset is negative, this function formats a string which
208 * contains the name, address, size and type of the BAR and
209 * appends it to the port description. If @offset is zero or
210 * positive, only name and offsetted address is appended.
211 *
212 * LOCKING:
213 * None.
214 */
215void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
216 const char *name)
217{
218 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
219 char *type = "";
220 unsigned long long start, len;
221
222 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
223 type = "m";
224 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
225 type = "i";
226
227 start = (unsigned long long)pci_resource_start(pdev, bar);
228 len = (unsigned long long)pci_resource_len(pdev, bar);
229
230 if (offset < 0)
231 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
232 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800233 ata_port_desc(ap, "%s 0x%llx", name,
234 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900235}
236
237#endif /* CONFIG_PCI */
238
Tejun Heo3884f7b2007-11-27 19:28:56 +0900239static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900240 unsigned int err_mask)
241{
242 struct ata_ering_entry *ent;
243
244 WARN_ON(!err_mask);
245
246 ering->cursor++;
247 ering->cursor %= ATA_ERING_SIZE;
248
249 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900250 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900251 ent->err_mask = err_mask;
252 ent->timestamp = get_jiffies_64();
253}
254
Tejun Heo76326ac2007-11-27 19:28:59 +0900255static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
256{
257 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
258
259 if (ent->err_mask)
260 return ent;
261 return NULL;
262}
263
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900264static void ata_ering_clear(struct ata_ering *ering)
Tejun Heo0c247c52006-05-15 20:58:19 +0900265{
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900266 memset(ering, 0, sizeof(*ering));
Tejun Heo0c247c52006-05-15 20:58:19 +0900267}
268
269static int ata_ering_map(struct ata_ering *ering,
270 int (*map_fn)(struct ata_ering_entry *, void *),
271 void *arg)
272{
273 int idx, rc = 0;
274 struct ata_ering_entry *ent;
275
276 idx = ering->cursor;
277 do {
278 ent = &ering->ring[idx];
279 if (!ent->err_mask)
280 break;
281 rc = map_fn(ent, arg);
282 if (rc)
283 break;
284 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
285 } while (idx != ering->cursor);
286
287 return rc;
288}
289
Tejun Heo64f65ca2006-06-24 20:30:18 +0900290static unsigned int ata_eh_dev_action(struct ata_device *dev)
291{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900292 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900293
294 return ehc->i.action | ehc->i.dev_action[dev->devno];
295}
296
Tejun Heof58229f2007-08-06 18:36:23 +0900297static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900298 struct ata_eh_info *ehi, unsigned int action)
299{
Tejun Heof58229f2007-08-06 18:36:23 +0900300 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900301
302 if (!dev) {
303 ehi->action &= ~action;
Tejun Heof58229f2007-08-06 18:36:23 +0900304 ata_link_for_each_dev(tdev, link)
305 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900306 } else {
307 /* doesn't make sense for port-wide EH actions */
308 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
309
310 /* break ehi->action into ehi->dev_action */
311 if (ehi->action & action) {
Tejun Heof58229f2007-08-06 18:36:23 +0900312 ata_link_for_each_dev(tdev, link)
313 ehi->dev_action[tdev->devno] |=
314 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900315 ehi->action &= ~action;
316 }
317
318 /* turn off the specified per-dev action */
319 ehi->dev_action[dev->devno] &= ~action;
320 }
321}
322
Tejun Heoece1d632006-04-02 18:51:53 +0900323/**
324 * ata_scsi_timed_out - SCSI layer time out callback
325 * @cmd: timed out SCSI command
326 *
327 * Handles SCSI layer timeout. We race with normal completion of
328 * the qc for @cmd. If the qc is already gone, we lose and let
329 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
330 * timed out and EH should be invoked. Prevent ata_qc_complete()
331 * from finishing it by setting EH_SCHEDULED and return
332 * EH_NOT_HANDLED.
333 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900334 * TODO: kill this function once old EH is gone.
335 *
Tejun Heoece1d632006-04-02 18:51:53 +0900336 * LOCKING:
337 * Called from timer context
338 *
339 * RETURNS:
340 * EH_HANDLED or EH_NOT_HANDLED
341 */
342enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
343{
344 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400345 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900346 unsigned long flags;
347 struct ata_queued_cmd *qc;
Tejun Heoad9e2762006-05-15 20:58:12 +0900348 enum scsi_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900349
350 DPRINTK("ENTER\n");
351
Tejun Heoad9e2762006-05-15 20:58:12 +0900352 if (ap->ops->error_handler) {
353 ret = EH_NOT_HANDLED;
354 goto out;
355 }
356
357 ret = EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400358 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900359 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900360 if (qc) {
361 WARN_ON(qc->scsicmd != cmd);
362 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
363 qc->err_mask |= AC_ERR_TIMEOUT;
364 ret = EH_NOT_HANDLED;
365 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400366 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900367
Tejun Heoad9e2762006-05-15 20:58:12 +0900368 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900369 DPRINTK("EXIT, ret=%d\n", ret);
370 return ret;
371}
372
373/**
374 * ata_scsi_error - SCSI layer error handler callback
375 * @host: SCSI host on which error occurred
376 *
377 * Handles SCSI-layer-thrown error events.
378 *
379 * LOCKING:
380 * Inherited from SCSI layer (none, can sleep)
381 *
382 * RETURNS:
383 * Zero.
384 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400385void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900386{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400387 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900388 int i;
Tejun Heoad9e2762006-05-15 20:58:12 +0900389 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900390
391 DPRINTK("ENTER\n");
392
Tejun Heoad9e2762006-05-15 20:58:12 +0900393 /* synchronize with port task */
Tejun Heoece1d632006-04-02 18:51:53 +0900394 ata_port_flush_task(ap);
395
Jeff Garzikcca39742006-08-24 03:19:22 -0400396 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900397
Tejun Heoad9e2762006-05-15 20:58:12 +0900398 /* For new EH, all qcs are finished in one of three ways -
399 * normal completion, error completion, and SCSI timeout.
400 * Both cmpletions can race against SCSI timeout. When normal
401 * completion wins, the qc never reaches EH. When error
402 * completion wins, the qc has ATA_QCFLAG_FAILED set.
403 *
404 * When SCSI timeout wins, things are a bit more complex.
405 * Normal or error completion can occur after the timeout but
406 * before this point. In such cases, both types of
407 * completions are honored. A scmd is determined to have
408 * timed out iff its associated qc is active and not failed.
409 */
410 if (ap->ops->error_handler) {
411 struct scsi_cmnd *scmd, *tmp;
412 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900413
Tejun Heoe30349d2006-07-03 03:02:15 +0900414 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900415
416 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
417 struct ata_queued_cmd *qc;
418
419 for (i = 0; i < ATA_MAX_QUEUE; i++) {
420 qc = __ata_qc_from_tag(ap, i);
421 if (qc->flags & ATA_QCFLAG_ACTIVE &&
422 qc->scsicmd == scmd)
423 break;
424 }
425
426 if (i < ATA_MAX_QUEUE) {
427 /* the scmd has an associated qc */
428 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
429 /* which hasn't failed yet, timeout */
430 qc->err_mask |= AC_ERR_TIMEOUT;
431 qc->flags |= ATA_QCFLAG_FAILED;
432 nr_timedout++;
433 }
434 } else {
435 /* Normal completion occurred after
436 * SCSI timeout but before this point.
437 * Successfully complete it.
438 */
439 scmd->retries = scmd->allowed;
440 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
441 }
442 }
443
444 /* If we have timed out qcs. They belong to EH from
445 * this point but the state of the controller is
446 * unknown. Freeze the port to make sure the IRQ
447 * handler doesn't diddle with those qcs. This must
448 * be done atomically w.r.t. setting QCFLAG_FAILED.
449 */
450 if (nr_timedout)
451 __ata_port_freeze(ap);
452
Tejun Heoe30349d2006-07-03 03:02:15 +0900453 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900454
455 /* initialize eh_tries */
456 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900457 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900458 spin_unlock_wait(ap->lock);
Tejun Heoad9e2762006-05-15 20:58:12 +0900459
460 repeat:
461 /* invoke error handler */
462 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900463 struct ata_link *link;
464
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900465 /* kill fast drain timer */
466 del_timer_sync(&ap->fastdrain_timer);
467
Tejun Heo500530f2006-07-03 16:07:27 +0900468 /* process port resume request */
469 ata_eh_handle_port_resume(ap);
470
Tejun Heof3e81b12006-05-15 20:58:21 +0900471 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900472 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900473
Tejun Heocf1b86c2007-08-06 18:36:23 +0900474 __ata_port_for_each_link(link, ap) {
Tejun Heo00115e02007-11-27 19:28:58 +0900475 struct ata_eh_context *ehc = &link->eh_context;
476 struct ata_device *dev;
477
Tejun Heocf1b86c2007-08-06 18:36:23 +0900478 memset(&link->eh_context, 0, sizeof(link->eh_context));
479 link->eh_context.i = link->eh_info;
480 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900481
482 ata_link_for_each_dev(dev, link) {
483 int devno = dev->devno;
484
485 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
486 if (ata_ncq_enabled(dev))
487 ehc->saved_ncq_enabled |= 1 << devno;
488 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900489 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900490
Tejun Heob51e9e52006-06-29 01:29:30 +0900491 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
492 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900493 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900494
Tejun Heoe30349d2006-07-03 03:02:15 +0900495 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900496
Tejun Heo500530f2006-07-03 16:07:27 +0900497 /* invoke EH, skip if unloading or suspended */
498 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900499 ap->ops->error_handler(ap);
500 else
501 ata_eh_finish(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900502
Tejun Heo500530f2006-07-03 16:07:27 +0900503 /* process port suspend request */
504 ata_eh_handle_port_suspend(ap);
505
Tejun Heoad9e2762006-05-15 20:58:12 +0900506 /* Exception might have happend after ->error_handler
507 * recovered the port but before this point. Repeat
508 * EH in such case.
509 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900510 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900511
Tejun Heob51e9e52006-06-29 01:29:30 +0900512 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900513 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900514 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900515 goto repeat;
516 }
517 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
Tejun Heoa1e10f72007-08-18 13:28:49 +0900518 "tries, giving up\n", ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900519 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900520 }
521
Tejun Heof3e81b12006-05-15 20:58:21 +0900522 /* this run is complete, make sure EH info is clear */
Tejun Heocf1b86c2007-08-06 18:36:23 +0900523 __ata_port_for_each_link(link, ap)
524 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900525
Tejun Heoe30349d2006-07-03 03:02:15 +0900526 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900527 * that if exception occurs after this point but
528 * before EH completion, SCSI midlayer will
529 * re-initiate EH.
530 */
531 host->host_eh_scheduled = 0;
532
Tejun Heoe30349d2006-07-03 03:02:15 +0900533 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900534 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900535 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900536 ap->ops->eng_timeout(ap);
537 }
538
539 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900540 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
541
542 scsi_eh_flush_done_q(&ap->eh_done_q);
543
Tejun Heoad9e2762006-05-15 20:58:12 +0900544 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900545 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900546
Tejun Heo1cdaf532006-07-03 16:07:26 +0900547 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900548 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900549 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
David Howells52bad642006-11-22 14:54:01 +0000550 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900551
552 if (ap->pflags & ATA_PFLAG_RECOVERED)
553 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900554
Tejun Heob51e9e52006-06-29 01:29:30 +0900555 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900556
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900557 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900558 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900559 wake_up_all(&ap->eh_wait_q);
560
Tejun Heoe30349d2006-07-03 03:02:15 +0900561 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900562
Tejun Heoece1d632006-04-02 18:51:53 +0900563 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900564}
565
566/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900567 * ata_port_wait_eh - Wait for the currently pending EH to complete
568 * @ap: Port to wait EH for
569 *
570 * Wait until the currently pending EH is complete.
571 *
572 * LOCKING:
573 * Kernel thread context (may sleep).
574 */
575void ata_port_wait_eh(struct ata_port *ap)
576{
577 unsigned long flags;
578 DEFINE_WAIT(wait);
579
580 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400581 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900582
Tejun Heob51e9e52006-06-29 01:29:30 +0900583 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900584 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400585 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900586 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400587 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900588 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900589 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900590
Jeff Garzikba6a1302006-06-22 23:46:10 -0400591 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900592
593 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400594 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900595 msleep(10);
596 goto retry;
597 }
598}
599
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900600static int ata_eh_nr_in_flight(struct ata_port *ap)
601{
602 unsigned int tag;
603 int nr = 0;
604
605 /* count only non-internal commands */
606 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
607 if (ata_qc_from_tag(ap, tag))
608 nr++;
609
610 return nr;
611}
612
613void ata_eh_fastdrain_timerfn(unsigned long arg)
614{
615 struct ata_port *ap = (void *)arg;
616 unsigned long flags;
617 int cnt;
618
619 spin_lock_irqsave(ap->lock, flags);
620
621 cnt = ata_eh_nr_in_flight(ap);
622
623 /* are we done? */
624 if (!cnt)
625 goto out_unlock;
626
627 if (cnt == ap->fastdrain_cnt) {
628 unsigned int tag;
629
630 /* No progress during the last interval, tag all
631 * in-flight qcs as timed out and freeze the port.
632 */
633 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
634 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
635 if (qc)
636 qc->err_mask |= AC_ERR_TIMEOUT;
637 }
638
639 ata_port_freeze(ap);
640 } else {
641 /* some qcs have finished, give it another chance */
642 ap->fastdrain_cnt = cnt;
643 ap->fastdrain_timer.expires =
644 jiffies + ATA_EH_FASTDRAIN_INTERVAL;
645 add_timer(&ap->fastdrain_timer);
646 }
647
648 out_unlock:
649 spin_unlock_irqrestore(ap->lock, flags);
650}
651
652/**
653 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
654 * @ap: target ATA port
655 * @fastdrain: activate fast drain
656 *
657 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
658 * is non-zero and EH wasn't pending before. Fast drain ensures
659 * that EH kicks in in timely manner.
660 *
661 * LOCKING:
662 * spin_lock_irqsave(host lock)
663 */
664static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
665{
666 int cnt;
667
668 /* already scheduled? */
669 if (ap->pflags & ATA_PFLAG_EH_PENDING)
670 return;
671
672 ap->pflags |= ATA_PFLAG_EH_PENDING;
673
674 if (!fastdrain)
675 return;
676
677 /* do we have in-flight qcs? */
678 cnt = ata_eh_nr_in_flight(ap);
679 if (!cnt)
680 return;
681
682 /* activate fast drain */
683 ap->fastdrain_cnt = cnt;
684 ap->fastdrain_timer.expires = jiffies + ATA_EH_FASTDRAIN_INTERVAL;
685 add_timer(&ap->fastdrain_timer);
686}
687
Tejun Heof686bcb2006-05-15 20:58:05 +0900688/**
689 * ata_qc_schedule_eh - schedule qc for error handling
690 * @qc: command to schedule error handling for
691 *
692 * Schedule error handling for @qc. EH will kick in as soon as
693 * other commands are drained.
694 *
695 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400696 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900697 */
698void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
699{
700 struct ata_port *ap = qc->ap;
701
702 WARN_ON(!ap->ops->error_handler);
703
704 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900705 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900706
707 /* The following will fail if timeout has already expired.
708 * ata_scsi_error() takes care of such scmds on EH entry.
709 * Note that ATA_QCFLAG_FAILED is unconditionally set after
710 * this function completes.
711 */
712 scsi_req_abort_cmd(qc->scsicmd);
713}
714
Tejun Heo7b70fc02006-05-15 20:58:07 +0900715/**
716 * ata_port_schedule_eh - schedule error handling without a qc
717 * @ap: ATA port to schedule EH for
718 *
719 * Schedule error handling for @ap. EH will kick in as soon as
720 * all commands are drained.
721 *
722 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400723 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900724 */
725void ata_port_schedule_eh(struct ata_port *ap)
726{
727 WARN_ON(!ap->ops->error_handler);
728
Tejun Heof4d6d002007-05-01 11:50:15 +0200729 if (ap->pflags & ATA_PFLAG_INITIALIZING)
730 return;
731
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900732 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -0400733 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900734
735 DPRINTK("port EH scheduled\n");
736}
737
Tejun Heodbd82612007-08-06 18:36:23 +0900738static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900739{
740 int tag, nr_aborted = 0;
741
742 WARN_ON(!ap->ops->error_handler);
743
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900744 /* we're gonna abort all commands, no need for fast drain */
745 ata_eh_set_pending(ap, 0);
746
Tejun Heo7b70fc02006-05-15 20:58:07 +0900747 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
748 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
749
Tejun Heodbd82612007-08-06 18:36:23 +0900750 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +0900751 qc->flags |= ATA_QCFLAG_FAILED;
752 ata_qc_complete(qc);
753 nr_aborted++;
754 }
755 }
756
757 if (!nr_aborted)
758 ata_port_schedule_eh(ap);
759
760 return nr_aborted;
761}
762
Tejun Heoe3180492006-05-15 20:58:09 +0900763/**
Tejun Heodbd82612007-08-06 18:36:23 +0900764 * ata_link_abort - abort all qc's on the link
765 * @link: ATA link to abort qc's for
766 *
767 * Abort all active qc's active on @link and schedule EH.
768 *
769 * LOCKING:
770 * spin_lock_irqsave(host lock)
771 *
772 * RETURNS:
773 * Number of aborted qc's.
774 */
775int ata_link_abort(struct ata_link *link)
776{
777 return ata_do_link_abort(link->ap, link);
778}
779
780/**
781 * ata_port_abort - abort all qc's on the port
782 * @ap: ATA port to abort qc's for
783 *
784 * Abort all active qc's of @ap and schedule EH.
785 *
786 * LOCKING:
787 * spin_lock_irqsave(host_set lock)
788 *
789 * RETURNS:
790 * Number of aborted qc's.
791 */
792int ata_port_abort(struct ata_port *ap)
793{
794 return ata_do_link_abort(ap, NULL);
795}
796
797/**
Tejun Heoe3180492006-05-15 20:58:09 +0900798 * __ata_port_freeze - freeze port
799 * @ap: ATA port to freeze
800 *
801 * This function is called when HSM violation or some other
802 * condition disrupts normal operation of the port. Frozen port
803 * is not allowed to perform any operation until the port is
804 * thawed, which usually follows a successful reset.
805 *
806 * ap->ops->freeze() callback can be used for freezing the port
807 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
808 * port cannot be frozen hardware-wise, the interrupt handler
809 * must ack and clear interrupts unconditionally while the port
810 * is frozen.
811 *
812 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400813 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900814 */
815static void __ata_port_freeze(struct ata_port *ap)
816{
817 WARN_ON(!ap->ops->error_handler);
818
819 if (ap->ops->freeze)
820 ap->ops->freeze(ap);
821
Tejun Heob51e9e52006-06-29 01:29:30 +0900822 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900823
Tejun Heo44877b42007-02-21 01:06:51 +0900824 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900825}
826
827/**
828 * ata_port_freeze - abort & freeze port
829 * @ap: ATA port to freeze
830 *
831 * Abort and freeze @ap.
832 *
833 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400834 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900835 *
836 * RETURNS:
837 * Number of aborted commands.
838 */
839int ata_port_freeze(struct ata_port *ap)
840{
841 int nr_aborted;
842
843 WARN_ON(!ap->ops->error_handler);
844
845 nr_aborted = ata_port_abort(ap);
846 __ata_port_freeze(ap);
847
848 return nr_aborted;
849}
850
851/**
Tejun Heo7d77b242007-09-23 13:14:13 +0900852 * sata_async_notification - SATA async notification handler
853 * @ap: ATA port where async notification is received
854 *
855 * Handler to be called when async notification via SDB FIS is
856 * received. This function schedules EH if necessary.
857 *
858 * LOCKING:
859 * spin_lock_irqsave(host lock)
860 *
861 * RETURNS:
862 * 1 if EH is scheduled, 0 otherwise.
863 */
864int sata_async_notification(struct ata_port *ap)
865{
866 u32 sntf;
867 int rc;
868
869 if (!(ap->flags & ATA_FLAG_AN))
870 return 0;
871
872 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
873 if (rc == 0)
874 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
875
876 if (!ap->nr_pmp_links || rc) {
877 /* PMP is not attached or SNTF is not available */
878 if (!ap->nr_pmp_links) {
879 /* PMP is not attached. Check whether ATAPI
880 * AN is configured. If so, notify media
881 * change.
882 */
883 struct ata_device *dev = ap->link.device;
884
885 if ((dev->class == ATA_DEV_ATAPI) &&
886 (dev->flags & ATA_DFLAG_AN))
887 ata_scsi_media_change_notify(dev);
888 return 0;
889 } else {
890 /* PMP is attached but SNTF is not available.
891 * ATAPI async media change notification is
892 * not used. The PMP must be reporting PHY
893 * status change, schedule EH.
894 */
895 ata_port_schedule_eh(ap);
896 return 1;
897 }
898 } else {
899 /* PMP is attached and SNTF is available */
900 struct ata_link *link;
901
902 /* check and notify ATAPI AN */
903 ata_port_for_each_link(link, ap) {
904 if (!(sntf & (1 << link->pmp)))
905 continue;
906
907 if ((link->device->class == ATA_DEV_ATAPI) &&
908 (link->device->flags & ATA_DFLAG_AN))
909 ata_scsi_media_change_notify(link->device);
910 }
911
912 /* If PMP is reporting that PHY status of some
913 * downstream ports has changed, schedule EH.
914 */
915 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
916 ata_port_schedule_eh(ap);
917 return 1;
918 }
919
920 return 0;
921 }
922}
923
924/**
Tejun Heoe3180492006-05-15 20:58:09 +0900925 * ata_eh_freeze_port - EH helper to freeze port
926 * @ap: ATA port to freeze
927 *
928 * Freeze @ap.
929 *
930 * LOCKING:
931 * None.
932 */
933void ata_eh_freeze_port(struct ata_port *ap)
934{
935 unsigned long flags;
936
937 if (!ap->ops->error_handler)
938 return;
939
Jeff Garzikba6a1302006-06-22 23:46:10 -0400940 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900941 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400942 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900943}
944
945/**
946 * ata_port_thaw_port - EH helper to thaw port
947 * @ap: ATA port to thaw
948 *
949 * Thaw frozen port @ap.
950 *
951 * LOCKING:
952 * None.
953 */
954void ata_eh_thaw_port(struct ata_port *ap)
955{
956 unsigned long flags;
957
958 if (!ap->ops->error_handler)
959 return;
960
Jeff Garzikba6a1302006-06-22 23:46:10 -0400961 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900962
Tejun Heob51e9e52006-06-29 01:29:30 +0900963 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900964
965 if (ap->ops->thaw)
966 ap->ops->thaw(ap);
967
Jeff Garzikba6a1302006-06-22 23:46:10 -0400968 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900969
Tejun Heo44877b42007-02-21 01:06:51 +0900970 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900971}
972
Tejun Heoece1d632006-04-02 18:51:53 +0900973static void ata_eh_scsidone(struct scsi_cmnd *scmd)
974{
975 /* nada */
976}
977
978static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
979{
980 struct ata_port *ap = qc->ap;
981 struct scsi_cmnd *scmd = qc->scsicmd;
982 unsigned long flags;
983
Jeff Garzikba6a1302006-06-22 23:46:10 -0400984 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900985 qc->scsidone = ata_eh_scsidone;
986 __ata_qc_complete(qc);
987 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -0400988 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900989
990 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
991}
992
993/**
994 * ata_eh_qc_complete - Complete an active ATA command from EH
995 * @qc: Command to complete
996 *
997 * Indicate to the mid and upper layers that an ATA command has
998 * completed. To be used from EH.
999 */
1000void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1001{
1002 struct scsi_cmnd *scmd = qc->scsicmd;
1003 scmd->retries = scmd->allowed;
1004 __ata_eh_qc_complete(qc);
1005}
1006
1007/**
1008 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1009 * @qc: Command to retry
1010 *
1011 * Indicate to the mid and upper layers that an ATA command
1012 * should be retried. To be used from EH.
1013 *
1014 * SCSI midlayer limits the number of retries to scmd->allowed.
1015 * scmd->retries is decremented for commands which get retried
1016 * due to unrelated failures (qc->err_mask is zero).
1017 */
1018void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1019{
1020 struct scsi_cmnd *scmd = qc->scsicmd;
1021 if (!qc->err_mask && scmd->retries)
1022 scmd->retries--;
1023 __ata_eh_qc_complete(qc);
1024}
Tejun Heo022bdb02006-05-15 20:58:22 +09001025
1026/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001027 * ata_eh_detach_dev - detach ATA device
1028 * @dev: ATA device to detach
1029 *
1030 * Detach @dev.
1031 *
1032 * LOCKING:
1033 * None.
1034 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001035void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001036{
Tejun Heof58229f2007-08-06 18:36:23 +09001037 struct ata_link *link = dev->link;
1038 struct ata_port *ap = link->ap;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001039 unsigned long flags;
1040
1041 ata_dev_disable(dev);
1042
Jeff Garzikba6a1302006-06-22 23:46:10 -04001043 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001044
1045 dev->flags &= ~ATA_DFLAG_DETACH;
1046
1047 if (ata_scsi_offline_dev(dev)) {
1048 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001049 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001050 }
1051
Tejun Heobeb07c12006-06-24 20:30:19 +09001052 /* clear per-dev EH actions */
Tejun Heof58229f2007-08-06 18:36:23 +09001053 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1054 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heobeb07c12006-06-24 20:30:19 +09001055
Jeff Garzikba6a1302006-06-22 23:46:10 -04001056 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001057}
1058
1059/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001060 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001061 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001062 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001063 * @action: action about to be performed
1064 *
1065 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001066 * in @link->eh_info such that eh actions are not unnecessarily
1067 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001068 *
1069 * LOCKING:
1070 * None.
1071 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001072void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1073 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001074{
Tejun Heo955e57d2007-08-06 18:36:23 +09001075 struct ata_port *ap = link->ap;
1076 struct ata_eh_info *ehi = &link->eh_info;
1077 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001078 unsigned long flags;
1079
Jeff Garzikba6a1302006-06-22 23:46:10 -04001080 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001081
Tejun Heo13abf502006-07-10 23:18:46 +09001082 /* Reset is represented by combination of actions and EHI
1083 * flags. Suck in all related bits before clearing eh_info to
1084 * avoid losing requested action.
1085 */
1086 if (action & ATA_EH_RESET_MASK) {
1087 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
1088 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
Tejun Heo1cdaf532006-07-03 16:07:26 +09001089
Tejun Heo13abf502006-07-10 23:18:46 +09001090 /* make sure all reset actions are cleared & clear EHI flags */
1091 action |= ATA_EH_RESET_MASK;
1092 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
1093 }
1094
Tejun Heo955e57d2007-08-06 18:36:23 +09001095 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001096
1097 if (!(ehc->i.flags & ATA_EHI_QUIET))
Tejun Heo1cdaf532006-07-03 16:07:26 +09001098 ap->pflags |= ATA_PFLAG_RECOVERED;
1099
Jeff Garzikba6a1302006-06-22 23:46:10 -04001100 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001101}
1102
1103/**
Tejun Heo47005f22006-06-19 18:27:23 +09001104 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001105* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001106 * @dev: target ATA dev for per-dev action (can be NULL)
1107 * @action: action just completed
1108 *
1109 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001110 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001111 *
1112 * LOCKING:
1113 * None.
1114 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001115void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1116 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001117{
Tejun Heo955e57d2007-08-06 18:36:23 +09001118 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001119
Tejun Heo13abf502006-07-10 23:18:46 +09001120 /* if reset is complete, clear all reset actions & reset modifier */
1121 if (action & ATA_EH_RESET_MASK) {
1122 action |= ATA_EH_RESET_MASK;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001123 ehc->i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
Tejun Heo13abf502006-07-10 23:18:46 +09001124 }
1125
Tejun Heo955e57d2007-08-06 18:36:23 +09001126 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001127}
1128
1129/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001130 * ata_err_string - convert err_mask to descriptive string
1131 * @err_mask: error mask to convert to string
1132 *
1133 * Convert @err_mask to descriptive string. Errors are
1134 * prioritized according to severity and only the most severe
1135 * error is reported.
1136 *
1137 * LOCKING:
1138 * None.
1139 *
1140 * RETURNS:
1141 * Descriptive string for @err_mask
1142 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001143static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001144{
1145 if (err_mask & AC_ERR_HOST_BUS)
1146 return "host bus error";
1147 if (err_mask & AC_ERR_ATA_BUS)
1148 return "ATA bus error";
1149 if (err_mask & AC_ERR_TIMEOUT)
1150 return "timeout";
1151 if (err_mask & AC_ERR_HSM)
1152 return "HSM violation";
1153 if (err_mask & AC_ERR_SYSTEM)
1154 return "internal error";
1155 if (err_mask & AC_ERR_MEDIA)
1156 return "media error";
1157 if (err_mask & AC_ERR_INVALID)
1158 return "invalid argument";
1159 if (err_mask & AC_ERR_DEV)
1160 return "device error";
1161 return "unknown error";
1162}
1163
1164/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001165 * ata_read_log_page - read a specific log page
1166 * @dev: target device
1167 * @page: page to read
1168 * @buf: buffer to store read page
1169 * @sectors: number of sectors to read
1170 *
1171 * Read log page using READ_LOG_EXT command.
1172 *
1173 * LOCKING:
1174 * Kernel thread context (may sleep).
1175 *
1176 * RETURNS:
1177 * 0 on success, AC_ERR_* mask otherwise.
1178 */
1179static unsigned int ata_read_log_page(struct ata_device *dev,
1180 u8 page, void *buf, unsigned int sectors)
1181{
1182 struct ata_taskfile tf;
1183 unsigned int err_mask;
1184
1185 DPRINTK("read log page - page %d\n", page);
1186
1187 ata_tf_init(dev, &tf);
1188 tf.command = ATA_CMD_READ_LOG_EXT;
1189 tf.lbal = page;
1190 tf.nsect = sectors;
1191 tf.hob_nsect = sectors >> 8;
1192 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1193 tf.protocol = ATA_PROT_PIO;
1194
1195 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001196 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001197
1198 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1199 return err_mask;
1200}
1201
1202/**
1203 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1204 * @dev: Device to read log page 10h from
1205 * @tag: Resulting tag of the failed command
1206 * @tf: Resulting taskfile registers of the failed command
1207 *
1208 * Read log page 10h to obtain NCQ error details and clear error
1209 * condition.
1210 *
1211 * LOCKING:
1212 * Kernel thread context (may sleep).
1213 *
1214 * RETURNS:
1215 * 0 on success, -errno otherwise.
1216 */
1217static int ata_eh_read_log_10h(struct ata_device *dev,
1218 int *tag, struct ata_taskfile *tf)
1219{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001220 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001221 unsigned int err_mask;
1222 u8 csum;
1223 int i;
1224
1225 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1226 if (err_mask)
1227 return -EIO;
1228
1229 csum = 0;
1230 for (i = 0; i < ATA_SECT_SIZE; i++)
1231 csum += buf[i];
1232 if (csum)
1233 ata_dev_printk(dev, KERN_WARNING,
1234 "invalid checksum 0x%x on log page 10h\n", csum);
1235
1236 if (buf[0] & 0x80)
1237 return -ENOENT;
1238
1239 *tag = buf[0] & 0x1f;
1240
1241 tf->command = buf[2];
1242 tf->feature = buf[3];
1243 tf->lbal = buf[4];
1244 tf->lbam = buf[5];
1245 tf->lbah = buf[6];
1246 tf->device = buf[7];
1247 tf->hob_lbal = buf[8];
1248 tf->hob_lbam = buf[9];
1249 tf->hob_lbah = buf[10];
1250 tf->nsect = buf[12];
1251 tf->hob_nsect = buf[13];
1252
1253 return 0;
1254}
1255
1256/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001257 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1258 * @dev: device to perform REQUEST_SENSE to
1259 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1260 *
1261 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1262 * SENSE. This function is EH helper.
1263 *
1264 * LOCKING:
1265 * Kernel thread context (may sleep).
1266 *
1267 * RETURNS:
1268 * 0 on success, AC_ERR_* mask on failure
1269 */
Albert Lee56287762007-04-02 11:30:46 +08001270static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
Tejun Heo022bdb02006-05-15 20:58:22 +09001271{
Albert Lee56287762007-04-02 11:30:46 +08001272 struct ata_device *dev = qc->dev;
1273 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001274 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001275 struct ata_taskfile tf;
1276 u8 cdb[ATAPI_CDB_LEN];
1277
1278 DPRINTK("ATAPI request sense\n");
1279
Tejun Heo022bdb02006-05-15 20:58:22 +09001280 /* FIXME: is this needed? */
1281 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1282
Albert Lee56287762007-04-02 11:30:46 +08001283 /* initialize sense_buf with the error register,
1284 * for the case where they are -not- overwritten
1285 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001286 sense_buf[0] = 0x70;
Albert Lee56287762007-04-02 11:30:46 +08001287 sense_buf[2] = qc->result_tf.feature >> 4;
1288
Jeff Garzika617c092007-05-21 20:14:23 -04001289 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001290 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001291
1292 memset(cdb, 0, ATAPI_CDB_LEN);
1293 cdb[0] = REQUEST_SENSE;
1294 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1295
1296 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1297 tf.command = ATA_CMD_PACKET;
1298
1299 /* is it pointless to prefer PIO for "safety reasons"? */
1300 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001301 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001302 tf.feature |= ATAPI_PKT_DMA;
1303 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001304 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001305 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1306 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001307 }
1308
1309 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001310 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001311}
1312
1313/**
1314 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001315 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001316 *
1317 * Analyze SError if available and further determine cause of
1318 * failure.
1319 *
1320 * LOCKING:
1321 * None.
1322 */
Tejun Heo02607312007-08-06 18:36:23 +09001323static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001324{
Tejun Heo02607312007-08-06 18:36:23 +09001325 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001326 u32 serror = ehc->i.serror;
1327 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001328 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001329
1330 if (serror & SERR_PERSISTENT) {
1331 err_mask |= AC_ERR_ATA_BUS;
1332 action |= ATA_EH_HARDRESET;
1333 }
1334 if (serror &
1335 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1336 err_mask |= AC_ERR_ATA_BUS;
1337 action |= ATA_EH_SOFTRESET;
1338 }
1339 if (serror & SERR_PROTOCOL) {
1340 err_mask |= AC_ERR_HSM;
1341 action |= ATA_EH_SOFTRESET;
1342 }
1343 if (serror & SERR_INTERNAL) {
1344 err_mask |= AC_ERR_SYSTEM;
Tejun Heo771b8da2007-03-14 01:20:51 +09001345 action |= ATA_EH_HARDRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001346 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001347
1348 /* Determine whether a hotplug event has occurred. Both
1349 * SError.N/X are considered hotplug events for enabled or
1350 * host links. For disabled PMP links, only N bit is
1351 * considered as X bit is left at 1 for link plugging.
1352 */
1353 hotplug_mask = 0;
1354
1355 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1356 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1357 else
1358 hotplug_mask = SERR_PHYRDY_CHG;
1359
1360 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001361 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001362
1363 ehc->i.err_mask |= err_mask;
1364 ehc->i.action |= action;
1365}
1366
1367/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001368 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001369 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001370 *
1371 * Read log page 10h, determine the offending qc and acquire
1372 * error status TF. For NCQ device errors, all LLDDs have to do
1373 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1374 * care of the rest.
1375 *
1376 * LOCKING:
1377 * Kernel thread context (may sleep).
1378 */
Tejun Heo02607312007-08-06 18:36:23 +09001379static void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001380{
Tejun Heo02607312007-08-06 18:36:23 +09001381 struct ata_port *ap = link->ap;
1382 struct ata_eh_context *ehc = &link->eh_context;
1383 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001384 struct ata_queued_cmd *qc;
1385 struct ata_taskfile tf;
1386 int tag, rc;
1387
1388 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001389 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001390 return;
1391
1392 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001393 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001394 return;
1395
1396 /* has LLDD analyzed already? */
1397 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1398 qc = __ata_qc_from_tag(ap, tag);
1399
1400 if (!(qc->flags & ATA_QCFLAG_FAILED))
1401 continue;
1402
1403 if (qc->err_mask)
1404 return;
1405 }
1406
1407 /* okay, this error is ours */
1408 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1409 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001410 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001411 "(errno=%d)\n", rc);
1412 return;
1413 }
1414
Tejun Heo02607312007-08-06 18:36:23 +09001415 if (!(link->sactive & (1 << tag))) {
1416 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001417 "inactive tag %d\n", tag);
1418 return;
1419 }
1420
1421 /* we've got the perpetrator, condemn it */
1422 qc = __ata_qc_from_tag(ap, tag);
1423 memcpy(&qc->result_tf, &tf, sizeof(tf));
Tejun Heo5335b722007-07-16 14:29:40 +09001424 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001425 ehc->i.err_mask &= ~AC_ERR_DEV;
1426}
1427
1428/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001429 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1430 * @qc: qc to analyze
1431 * @tf: Taskfile registers to analyze
1432 *
1433 * Analyze taskfile of @qc and further determine cause of
1434 * failure. This function also requests ATAPI sense data if
1435 * avaliable.
1436 *
1437 * LOCKING:
1438 * Kernel thread context (may sleep).
1439 *
1440 * RETURNS:
1441 * Determined recovery action
1442 */
1443static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1444 const struct ata_taskfile *tf)
1445{
1446 unsigned int tmp, action = 0;
1447 u8 stat = tf->command, err = tf->feature;
1448
1449 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1450 qc->err_mask |= AC_ERR_HSM;
1451 return ATA_EH_SOFTRESET;
1452 }
1453
Tejun Heoa51d6442007-03-20 15:24:11 +09001454 if (stat & (ATA_ERR | ATA_DF))
1455 qc->err_mask |= AC_ERR_DEV;
1456 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001457 return 0;
1458
1459 switch (qc->dev->class) {
1460 case ATA_DEV_ATA:
1461 if (err & ATA_ICRC)
1462 qc->err_mask |= AC_ERR_ATA_BUS;
1463 if (err & ATA_UNC)
1464 qc->err_mask |= AC_ERR_MEDIA;
1465 if (err & ATA_IDNF)
1466 qc->err_mask |= AC_ERR_INVALID;
1467 break;
1468
1469 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001470 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Albert Lee56287762007-04-02 11:30:46 +08001471 tmp = atapi_eh_request_sense(qc);
Tejun Heoa569a302006-11-21 10:40:51 +09001472 if (!tmp) {
1473 /* ATA_QCFLAG_SENSE_VALID is used to
1474 * tell atapi_qc_complete() that sense
1475 * data is already valid.
1476 *
1477 * TODO: interpret sense data and set
1478 * appropriate err_mask.
1479 */
1480 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1481 } else
1482 qc->err_mask |= tmp;
1483 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001484 }
1485
1486 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1487 action |= ATA_EH_SOFTRESET;
1488
1489 return action;
1490}
1491
Tejun Heo76326ac2007-11-27 19:28:59 +09001492static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1493 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001494{
Tejun Heo76326ac2007-11-27 19:28:59 +09001495 int base = 0;
1496
1497 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1498 *xfer_ok = 1;
1499
1500 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001501 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001502
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001503 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001504 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001505
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001506 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001507 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001508
Tejun Heo3884f7b2007-11-27 19:28:56 +09001509 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001510 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001511 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001512 if ((err_mask &
1513 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001514 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001515 }
1516
1517 return 0;
1518}
1519
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001520struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001521 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001522 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001523 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001524};
1525
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001526static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001527{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001528 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001529 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001530
1531 if (ent->timestamp < arg->since)
1532 return -1;
1533
Tejun Heo76326ac2007-11-27 19:28:59 +09001534 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1535 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001536 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001537
Tejun Heo022bdb02006-05-15 20:58:22 +09001538 return 0;
1539}
1540
1541/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001542 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001543 * @dev: Device of interest
1544 *
1545 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001546 * whether NCQ needs to be turned off, transfer speed should be
1547 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001548 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001549 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001550 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001551 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1552 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001553 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001554 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001555 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001556 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1557 * data transfer hasn't been verified.
1558 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001559 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001560 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001561 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001562 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001563 * SPEED_DOWN : Speed down transfer speed but don't fall back
1564 * to PIO.
1565 *
1566 * FALLBACK_TO_PIO : Fall back to PIO.
1567 *
1568 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001569 * taken per error. An action triggered by non-DUBIOUS errors
1570 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1571 * This is to expedite speed down decisions right after device is
1572 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001573 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001574 * The followings are speed down rules. #1 and #2 deal with
1575 * DUBIOUS errors.
1576 *
1577 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1578 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1579 *
1580 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1581 * occurred during last 5 mins, NCQ_OFF.
1582 *
1583 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Tejun Heo3884f7b2007-11-27 19:28:56 +09001584 * ocurred during last 5 mins, FALLBACK_TO_PIO
1585 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001586 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001587 * during last 10 mins, NCQ_OFF.
1588 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001589 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001590 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001591 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001592 * LOCKING:
1593 * Inherited from caller.
1594 *
1595 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001596 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001597 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001598static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001599{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001600 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1601 u64 j64 = get_jiffies_64();
1602 struct speed_down_verdict_arg arg;
1603 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001604
Tejun Heo3884f7b2007-11-27 19:28:56 +09001605 /* scan past 5 mins of error history */
1606 memset(&arg, 0, sizeof(arg));
1607 arg.since = j64 - min(j64, j5mins);
1608 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1609
Tejun Heo76326ac2007-11-27 19:28:59 +09001610 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1611 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1612 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1613 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1614
1615 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1616 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1617 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1618
Tejun Heo3884f7b2007-11-27 19:28:56 +09001619 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1620 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001621 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001622 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1623
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001624 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001625 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001626 arg.since = j64 - min(j64, j10mins);
1627 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001628
Tejun Heo3884f7b2007-11-27 19:28:56 +09001629 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1630 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001631 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001632
1633 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1634 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001635 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001636 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001637
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001638 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001639}
1640
1641/**
1642 * ata_eh_speed_down - record error and speed down if necessary
1643 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001644 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001645 * @err_mask: err_mask of the error
1646 *
1647 * Record error and examine error history to determine whether
1648 * adjusting transmission speed is necessary. It also sets
1649 * transmission limits appropriately if such adjustment is
1650 * necessary.
1651 *
1652 * LOCKING:
1653 * Kernel thread context (may sleep).
1654 *
1655 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001656 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001657 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001658static unsigned int ata_eh_speed_down(struct ata_device *dev,
1659 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001660{
Tejun Heo3884f7b2007-11-27 19:28:56 +09001661 struct ata_link *link = dev->link;
Tejun Heo76326ac2007-11-27 19:28:59 +09001662 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001663 unsigned int verdict;
1664 unsigned int action = 0;
1665
1666 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001667 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001668 return 0;
1669
1670 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001671 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001672 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001673
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001674 /* turn off NCQ? */
1675 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1676 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1677 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1678 dev->flags |= ATA_DFLAG_NCQ_OFF;
1679 ata_dev_printk(dev, KERN_WARNING,
1680 "NCQ disabled due to excessive errors\n");
1681 goto done;
1682 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001683
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001684 /* speed down? */
1685 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1686 /* speed down SATA link speed if possible */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001687 if (sata_down_spd_limit(link) == 0) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001688 action |= ATA_EH_HARDRESET;
1689 goto done;
1690 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001691
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001692 /* lower transfer mode */
1693 if (dev->spdn_cnt < 2) {
1694 static const int dma_dnxfer_sel[] =
1695 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1696 static const int pio_dnxfer_sel[] =
1697 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1698 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001699
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001700 if (dev->xfer_shift != ATA_SHIFT_PIO)
1701 sel = dma_dnxfer_sel[dev->spdn_cnt];
1702 else
1703 sel = pio_dnxfer_sel[dev->spdn_cnt];
1704
1705 dev->spdn_cnt++;
1706
1707 if (ata_down_xfermask_limit(dev, sel) == 0) {
1708 action |= ATA_EH_SOFTRESET;
1709 goto done;
1710 }
1711 }
1712 }
1713
1714 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09001715 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001716 */
1717 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09001718 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001719 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1720 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1721 dev->spdn_cnt = 0;
1722 action |= ATA_EH_SOFTRESET;
1723 goto done;
1724 }
1725 }
1726
Tejun Heo022bdb02006-05-15 20:58:22 +09001727 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001728 done:
1729 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09001730 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1731 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001732 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001733}
1734
1735/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001736 * ata_eh_link_autopsy - analyze error and determine recovery action
1737 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09001738 *
Tejun Heo02607312007-08-06 18:36:23 +09001739 * Analyze why @link failed and determine which recovery actions
1740 * are needed. This function also sets more detailed AC_ERR_*
1741 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09001742 *
1743 * LOCKING:
1744 * Kernel thread context (may sleep).
1745 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001746static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001747{
Tejun Heo02607312007-08-06 18:36:23 +09001748 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09001749 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09001750 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001751 unsigned int all_err_mask = 0, eflags = 0;
1752 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09001753 u32 serror;
1754 int rc;
1755
1756 DPRINTK("ENTER\n");
1757
Tejun Heo1cdaf532006-07-03 16:07:26 +09001758 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1759 return;
1760
Tejun Heo022bdb02006-05-15 20:58:22 +09001761 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09001762 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09001763 if (rc == 0) {
1764 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09001765 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09001766 } else if (rc != -EOPNOTSUPP) {
1767 /* SError read failed, force hardreset and probing */
1768 ata_ehi_schedule_probe(&ehc->i);
Tejun Heo4528e4d2006-07-08 20:17:26 +09001769 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09001770 ehc->i.err_mask |= AC_ERR_OTHER;
1771 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001772
Tejun Heoe8ee8452006-05-15 21:03:46 +09001773 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09001774 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001775
Tejun Heo022bdb02006-05-15 20:58:22 +09001776 /* any real error trumps AC_ERR_OTHER */
1777 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1778 ehc->i.err_mask &= ~AC_ERR_OTHER;
1779
1780 all_err_mask |= ehc->i.err_mask;
1781
1782 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1783 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1784
Tejun Heo02607312007-08-06 18:36:23 +09001785 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001786 continue;
1787
1788 /* inherit upper level err_mask */
1789 qc->err_mask |= ehc->i.err_mask;
1790
Tejun Heo022bdb02006-05-15 20:58:22 +09001791 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001792 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001793
1794 /* DEV errors are probably spurious in case of ATA_BUS error */
1795 if (qc->err_mask & AC_ERR_ATA_BUS)
1796 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1797 AC_ERR_INVALID);
1798
1799 /* any real error trumps unknown error */
1800 if (qc->err_mask & ~AC_ERR_OTHER)
1801 qc->err_mask &= ~AC_ERR_OTHER;
1802
1803 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09001804 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09001805 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09001806
1807 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001808 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001809 all_err_mask |= qc->err_mask;
1810 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001811 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09001812 }
1813
Tejun Heoa20f33f2006-05-16 12:58:24 +09001814 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001815 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001816 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001817 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001818 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1819 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001820 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001821
Tejun Heodfcc1732007-10-31 10:17:05 +09001822 /* If we have offending qcs and the associated failed device,
1823 * perform per-dev EH action only on the offending device.
1824 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001825 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09001826 ehc->i.dev_action[ehc->i.dev->devno] |=
1827 ehc->i.action & ATA_EH_PERDEV_MASK;
1828 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001829 }
1830
Tejun Heo2695e362008-01-10 13:41:23 +09001831 /* propagate timeout to host link */
1832 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1833 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1834
1835 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09001836 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09001837 if (!dev && ((ata_link_max_devices(link) == 1 &&
1838 ata_dev_enabled(link->device))))
1839 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09001840
Tejun Heo76326ac2007-11-27 19:28:59 +09001841 if (dev) {
1842 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1843 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001844 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09001845 }
Tejun Heodfcc1732007-10-31 10:17:05 +09001846
Tejun Heo022bdb02006-05-15 20:58:22 +09001847 DPRINTK("EXIT\n");
1848}
1849
1850/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001851 * ata_eh_autopsy - analyze error and determine recovery action
1852 * @ap: host port to perform autopsy on
1853 *
1854 * Analyze all links of @ap and determine why they failed and
1855 * which recovery actions are needed.
1856 *
1857 * LOCKING:
1858 * Kernel thread context (may sleep).
1859 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001860void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09001861{
1862 struct ata_link *link;
1863
Tejun Heo2695e362008-01-10 13:41:23 +09001864 ata_port_for_each_link(link, ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09001865 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09001866
1867 /* Autopsy of fanout ports can affect host link autopsy.
1868 * Perform host link autopsy last.
1869 */
1870 if (ap->nr_pmp_links)
1871 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09001872}
1873
1874/**
1875 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09001876 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09001877 *
1878 * Report EH to user.
1879 *
1880 * LOCKING:
1881 * None.
1882 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001883static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001884{
Tejun Heo02607312007-08-06 18:36:23 +09001885 struct ata_port *ap = link->ap;
1886 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001887 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09001888 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09001889 int tag, nr_failed = 0;
1890
Tejun Heo94ff3d52007-10-09 14:57:56 +09001891 if (ehc->i.flags & ATA_EHI_QUIET)
1892 return;
1893
Tejun Heo022bdb02006-05-15 20:58:22 +09001894 desc = NULL;
1895 if (ehc->i.desc[0] != '\0')
1896 desc = ehc->i.desc;
1897
1898 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1899 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1900
Tejun Heoe027bd32007-10-26 16:19:26 +09001901 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link ||
1902 ((qc->flags & ATA_QCFLAG_QUIET) &&
1903 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09001904 continue;
1905 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1906 continue;
1907
1908 nr_failed++;
1909 }
1910
1911 if (!nr_failed && !ehc->i.err_mask)
1912 return;
1913
1914 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001915 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001916 frozen = " frozen";
1917
Tejun Heoa1e10f72007-08-18 13:28:49 +09001918 memset(tries_buf, 0, sizeof(tries_buf));
1919 if (ap->eh_tries < ATA_EH_MAX_TRIES)
1920 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
1921 ap->eh_tries);
1922
Tejun Heo022bdb02006-05-15 20:58:22 +09001923 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001924 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09001925 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1926 ehc->i.err_mask, link->sactive, ehc->i.serror,
1927 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001928 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09001929 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001930 } else {
Tejun Heo02607312007-08-06 18:36:23 +09001931 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09001932 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1933 ehc->i.err_mask, link->sactive, ehc->i.serror,
1934 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001935 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09001936 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001937 }
1938
Robert Hancock1333e192007-10-02 11:22:02 -04001939 if (ehc->i.serror)
1940 ata_port_printk(ap, KERN_ERR,
1941 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1942 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
1943 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
1944 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
1945 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
1946 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
1947 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
1948 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
1949 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
1950 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
1951 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
1952 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
1953 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
1954 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
1955 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
1956 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
1957 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001958 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04001959
Tejun Heo022bdb02006-05-15 20:58:22 +09001960 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1961 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09001962 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09001963 const u8 *cdb = qc->cdb;
1964 char data_buf[20] = "";
1965 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09001966
Tejun Heo02607312007-08-06 18:36:23 +09001967 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1968 qc->dev->link != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001969 continue;
1970
Tejun Heoabb6a882007-11-28 23:16:09 +09001971 if (qc->dma_dir != DMA_NONE) {
1972 static const char *dma_str[] = {
1973 [DMA_BIDIRECTIONAL] = "bidi",
1974 [DMA_TO_DEVICE] = "out",
1975 [DMA_FROM_DEVICE] = "in",
1976 };
1977 static const char *prot_str[] = {
1978 [ATA_PROT_PIO] = "pio",
1979 [ATA_PROT_DMA] = "dma",
1980 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05001981 [ATAPI_PROT_PIO] = "pio",
1982 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09001983 };
1984
1985 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
1986 prot_str[qc->tf.protocol], qc->nbytes,
1987 dma_str[qc->dma_dir]);
1988 }
1989
Jeff Garzike39eec12007-12-01 18:05:39 -05001990 if (ata_is_atapi(qc->tf.protocol))
Tejun Heoabb6a882007-11-28 23:16:09 +09001991 snprintf(cdb_buf, sizeof(cdb_buf),
1992 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
1993 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
1994 cdb[0], cdb[1], cdb[2], cdb[3],
1995 cdb[4], cdb[5], cdb[6], cdb[7],
1996 cdb[8], cdb[9], cdb[10], cdb[11],
1997 cdb[12], cdb[13], cdb[14], cdb[15]);
1998
Tejun Heo8a937582006-11-14 22:36:12 +09001999 ata_dev_printk(qc->dev, KERN_ERR,
2000 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002001 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002002 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002003 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002004 cmd->command, cmd->feature, cmd->nsect,
2005 cmd->lbal, cmd->lbam, cmd->lbah,
2006 cmd->hob_feature, cmd->hob_nsect,
2007 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002008 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002009 res->command, res->feature, res->nsect,
2010 res->lbal, res->lbam, res->lbah,
2011 res->hob_feature, res->hob_nsect,
2012 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002013 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2014 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002015
2016 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002017 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002018 if (res->command & ATA_BUSY)
2019 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002020 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002021 else
2022 ata_dev_printk(qc->dev, KERN_ERR,
2023 "status: { %s%s%s%s}\n",
2024 res->command & ATA_DRDY ? "DRDY " : "",
2025 res->command & ATA_DF ? "DF " : "",
2026 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002027 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002028 }
2029
2030 if (cmd->command != ATA_CMD_PACKET &&
2031 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2032 ATA_ABORTED)))
2033 ata_dev_printk(qc->dev, KERN_ERR,
2034 "error: { %s%s%s%s}\n",
2035 res->feature & ATA_ICRC ? "ICRC " : "",
2036 res->feature & ATA_UNC ? "UNC " : "",
2037 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002038 res->feature & ATA_ABORTED ? "ABRT " : "");
Tejun Heo022bdb02006-05-15 20:58:22 +09002039 }
2040}
2041
Tejun Heo9b1e2652007-08-06 18:36:24 +09002042/**
2043 * ata_eh_report - report error handling to user
2044 * @ap: ATA port to report EH about
2045 *
2046 * Report EH to user.
2047 *
2048 * LOCKING:
2049 * None.
2050 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002051void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002052{
2053 struct ata_link *link;
2054
2055 __ata_port_for_each_link(link, ap)
2056 ata_eh_link_report(link);
2057}
2058
Tejun Heocc0680a2007-08-06 18:36:23 +09002059static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heod4b2bab2007-02-02 16:50:52 +09002060 unsigned int *classes, unsigned long deadline)
Tejun Heod87fa382006-05-31 18:28:24 +09002061{
Tejun Heof58229f2007-08-06 18:36:23 +09002062 struct ata_device *dev;
2063 int rc;
Tejun Heod87fa382006-05-31 18:28:24 +09002064
Tejun Heocc0680a2007-08-06 18:36:23 +09002065 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002066 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002067
Tejun Heocc0680a2007-08-06 18:36:23 +09002068 rc = reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002069 if (rc)
2070 return rc;
2071
2072 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
2073 * is complete and convert all ATA_DEV_UNKNOWN to
2074 * ATA_DEV_NONE.
2075 */
Tejun Heocc0680a2007-08-06 18:36:23 +09002076 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002077 if (classes[dev->devno] != ATA_DEV_UNKNOWN)
Tejun Heod87fa382006-05-31 18:28:24 +09002078 break;
2079
Tejun Heof58229f2007-08-06 18:36:23 +09002080 if (dev) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002081 ata_link_for_each_dev(dev, link) {
Tejun Heof58229f2007-08-06 18:36:23 +09002082 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2083 classes[dev->devno] = ATA_DEV_NONE;
2084 }
2085 }
Tejun Heod87fa382006-05-31 18:28:24 +09002086
2087 return 0;
2088}
2089
Tejun Heoae791c02007-09-23 13:14:12 +09002090static int ata_eh_followup_srst_needed(struct ata_link *link,
2091 int rc, int classify,
Tejun Heo664faf02006-05-31 18:27:50 +09002092 const unsigned int *classes)
2093{
Tejun Heoae791c02007-09-23 13:14:12 +09002094 if (link->flags & ATA_LFLAG_NO_SRST)
2095 return 0;
Tejun Heo664faf02006-05-31 18:27:50 +09002096 if (rc == -EAGAIN)
2097 return 1;
2098 if (rc != 0)
2099 return 0;
Tejun Heo3495de72007-09-23 13:19:53 +09002100 if ((link->ap->flags & ATA_FLAG_PMP) && ata_is_host_link(link))
2101 return 1;
Tejun Heoae791c02007-09-23 13:14:12 +09002102 if (classify && !(link->flags & ATA_LFLAG_ASSUME_CLASS) &&
2103 classes[0] == ATA_DEV_UNKNOWN)
Tejun Heo664faf02006-05-31 18:27:50 +09002104 return 1;
2105 return 0;
2106}
2107
Tejun Heofb7fd612007-09-23 13:14:12 +09002108int ata_eh_reset(struct ata_link *link, int classify,
2109 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2110 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002111{
Tejun Heo416dc9e2007-10-31 10:17:03 +09002112 const int max_tries = ARRAY_SIZE(ata_eh_reset_timeouts);
Tejun Heoafaa5c32007-10-09 15:06:10 +09002113 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002114 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09002115 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002116 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002117 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heo31daabd2007-02-02 16:50:52 +09002118 int try = 0;
Tejun Heof58229f2007-08-06 18:36:23 +09002119 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002120 unsigned long deadline, now;
Tejun Heo0e06d9c2007-10-24 15:21:26 +09002121 unsigned int tmp_action;
Tejun Heo022bdb02006-05-15 20:58:22 +09002122 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002123 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002124 u32 sstatus;
Tejun Heof58229f2007-08-06 18:36:23 +09002125 int rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002126
Tejun Heo13abf502006-07-10 23:18:46 +09002127 /* about to reset */
Tejun Heoafaa5c32007-10-09 15:06:10 +09002128 spin_lock_irqsave(ap->lock, flags);
2129 ap->pflags |= ATA_PFLAG_RESETTING;
2130 spin_unlock_irqrestore(ap->lock, flags);
2131
Tejun Heo955e57d2007-08-06 18:36:23 +09002132 ata_eh_about_to_do(link, NULL, ehc->i.action & ATA_EH_RESET_MASK);
Tejun Heo13abf502006-07-10 23:18:46 +09002133
Tejun Heocdeab112007-10-29 16:41:09 +09002134 ata_link_for_each_dev(dev, link) {
2135 /* If we issue an SRST then an ATA drive (not ATAPI)
2136 * may change configuration and be in PIO0 timing. If
2137 * we do a hard reset (or are coming from power on)
2138 * this is true for ATA or ATAPI. Until we've set a
2139 * suitable controller mode we should not touch the
2140 * bus as we may be talking too fast.
2141 */
2142 dev->pio_mode = XFER_PIO_0;
2143
2144 /* If the controller has a pio mode setup function
2145 * then use it to set the chipset to rights. Don't
2146 * touch the DMA setup as that will be dealt with when
2147 * configuring devices.
2148 */
2149 if (ap->ops->set_piomode)
2150 ap->ops->set_piomode(ap, dev);
2151 }
2152
Tejun Heof5914a42006-05-31 18:27:48 +09002153 /* Determine which reset to use and record in ehc->i.action.
2154 * prereset() may examine and modify it.
2155 */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002156 if (softreset && (!hardreset || (!(lflags & ATA_LFLAG_NO_SRST) &&
Tejun Heoae791c02007-09-23 13:14:12 +09002157 !sata_set_spd_needed(link) &&
Tejun Heo0e06d9c2007-10-24 15:21:26 +09002158 !(ehc->i.action & ATA_EH_HARDRESET))))
2159 tmp_action = ATA_EH_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002160 else
Tejun Heo0e06d9c2007-10-24 15:21:26 +09002161 tmp_action = ATA_EH_HARDRESET;
2162
2163 ehc->i.action = (ehc->i.action & ~ATA_EH_RESET_MASK) | tmp_action;
Tejun Heof5914a42006-05-31 18:27:48 +09002164
2165 if (prereset) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002166 rc = prereset(link, jiffies + ATA_EH_PRERESET_TIMEOUT);
Tejun Heof5914a42006-05-31 18:27:48 +09002167 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002168 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002169 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002170 "port disabled. ignoring.\n");
Tejun Heo9af5c9c2007-08-06 18:36:22 +09002171 ehc->i.action &= ~ATA_EH_RESET_MASK;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002172
Tejun Heo936fd732007-08-06 18:36:23 +09002173 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002174 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002175
2176 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002177 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002178 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002179 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002180 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002181 }
2182 }
2183
2184 /* prereset() might have modified ehc->i.action */
2185 if (ehc->i.action & ATA_EH_HARDRESET)
Tejun Heo022bdb02006-05-15 20:58:22 +09002186 reset = hardreset;
Tejun Heof5914a42006-05-31 18:27:48 +09002187 else if (ehc->i.action & ATA_EH_SOFTRESET)
2188 reset = softreset;
2189 else {
2190 /* prereset told us not to reset, bang classes and return */
Tejun Heo936fd732007-08-06 18:36:23 +09002191 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002192 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002193 rc = 0;
2194 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002195 }
2196
2197 /* did prereset() screw up? if so, fix up to avoid oopsing */
2198 if (!reset) {
Tejun Heof5914a42006-05-31 18:27:48 +09002199 if (softreset)
2200 reset = softreset;
2201 else
2202 reset = hardreset;
2203 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002204
2205 retry:
Tejun Heo31daabd2007-02-02 16:50:52 +09002206 deadline = jiffies + ata_eh_reset_timeouts[try++];
2207
Tejun Heo3e706392006-05-31 18:28:11 +09002208 /* shut up during boot probing */
2209 if (verbose)
Tejun Heocc0680a2007-08-06 18:36:23 +09002210 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
Tejun Heo3e706392006-05-31 18:28:11 +09002211 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002212
Tejun Heo13abf502006-07-10 23:18:46 +09002213 /* mark that this EH session started with reset */
Tejun Heo0d64a232007-04-23 02:41:05 +09002214 if (reset == hardreset)
2215 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2216 else
2217 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002218
Tejun Heocc0680a2007-08-06 18:36:23 +09002219 rc = ata_do_reset(link, reset, classes, deadline);
Tejun Heo022bdb02006-05-15 20:58:22 +09002220
Tejun Heo664faf02006-05-31 18:27:50 +09002221 if (reset == hardreset &&
Tejun Heoae791c02007-09-23 13:14:12 +09002222 ata_eh_followup_srst_needed(link, rc, classify, classes)) {
Tejun Heo664faf02006-05-31 18:27:50 +09002223 /* okay, let's do follow-up softreset */
Tejun Heo664faf02006-05-31 18:27:50 +09002224 reset = softreset;
2225
2226 if (!reset) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002227 ata_link_printk(link, KERN_ERR,
Tejun Heo664faf02006-05-31 18:27:50 +09002228 "follow-up softreset required "
2229 "but no softreset avaliable\n");
Tejun Heofccb6ea2007-07-16 14:29:41 +09002230 rc = -EINVAL;
Tejun Heo08cf69d2007-10-31 10:17:04 +09002231 goto fail;
Tejun Heo664faf02006-05-31 18:27:50 +09002232 }
2233
Tejun Heo955e57d2007-08-06 18:36:23 +09002234 ata_eh_about_to_do(link, NULL, ATA_EH_RESET_MASK);
Tejun Heocc0680a2007-08-06 18:36:23 +09002235 rc = ata_do_reset(link, reset, classes, deadline);
Tejun Heo664faf02006-05-31 18:27:50 +09002236 }
2237
Tejun Heo416dc9e2007-10-31 10:17:03 +09002238 /* -EAGAIN can happen if we skipped followup SRST */
2239 if (rc && rc != -EAGAIN)
2240 goto fail;
Tejun Heoae791c02007-09-23 13:14:12 +09002241
Tejun Heo08cf69d2007-10-31 10:17:04 +09002242 /* was classification successful? */
2243 if (classify && classes[0] == ATA_DEV_UNKNOWN &&
2244 !(lflags & ATA_LFLAG_ASSUME_CLASS)) {
2245 if (try < max_tries) {
2246 ata_link_printk(link, KERN_WARNING,
2247 "classification failed\n");
2248 rc = -EINVAL;
2249 goto fail;
2250 }
2251
2252 ata_link_printk(link, KERN_WARNING,
2253 "classfication failed, assuming ATA\n");
2254 lflags |= ATA_LFLAG_ASSUME_ATA;
2255 }
2256
Tejun Heo416dc9e2007-10-31 10:17:03 +09002257 ata_link_for_each_dev(dev, link) {
2258 /* After the reset, the device state is PIO 0 and the
2259 * controller state is undefined. Reset also wakes up
2260 * drives from sleeping mode.
2261 */
2262 dev->pio_mode = XFER_PIO_0;
2263 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002264
Tejun Heo416dc9e2007-10-31 10:17:03 +09002265 if (ata_link_offline(link))
2266 continue;
Tejun Heo664faf02006-05-31 18:27:50 +09002267
Tejun Heo4ccd3322008-01-08 20:26:12 +09002268 /* apply class override */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002269 if (lflags & ATA_LFLAG_ASSUME_ATA)
2270 classes[dev->devno] = ATA_DEV_ATA;
2271 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2272 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
Tejun Heo022bdb02006-05-15 20:58:22 +09002273 }
2274
Tejun Heo416dc9e2007-10-31 10:17:03 +09002275 /* record current link speed */
2276 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2277 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002278
Tejun Heo416dc9e2007-10-31 10:17:03 +09002279 if (postreset)
2280 postreset(link, classes);
Tejun Heo20952b62006-05-31 18:27:23 +09002281
Tejun Heo416dc9e2007-10-31 10:17:03 +09002282 /* reset successful, schedule revalidation */
2283 ata_eh_done(link, NULL, ehc->i.action & ATA_EH_RESET_MASK);
2284 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heoae791c02007-09-23 13:14:12 +09002285
Tejun Heo416dc9e2007-10-31 10:17:03 +09002286 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002287 out:
2288 /* clear hotplug flag */
2289 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002290
2291 spin_lock_irqsave(ap->lock, flags);
2292 ap->pflags &= ~ATA_PFLAG_RESETTING;
2293 spin_unlock_irqrestore(ap->lock, flags);
2294
Tejun Heo022bdb02006-05-15 20:58:22 +09002295 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002296
2297 fail:
2298 if (rc == -ERESTART || try >= max_tries)
2299 goto out;
2300
2301 now = jiffies;
2302 if (time_before(now, deadline)) {
2303 unsigned long delta = deadline - now;
2304
2305 ata_link_printk(link, KERN_WARNING, "reset failed "
2306 "(errno=%d), retrying in %u secs\n",
2307 rc, (jiffies_to_msecs(delta) + 999) / 1000);
2308
2309 while (delta)
2310 delta = schedule_timeout_uninterruptible(delta);
2311 }
2312
2313 if (rc == -EPIPE || try == max_tries - 1)
2314 sata_down_spd_limit(link);
2315 if (hardreset)
2316 reset = hardreset;
2317 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002318}
2319
Tejun Heo02607312007-08-06 18:36:23 +09002320static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002321 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002322{
Tejun Heo02607312007-08-06 18:36:23 +09002323 struct ata_port *ap = link->ap;
2324 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002325 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002326 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002327 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09002328 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002329
2330 DPRINTK("ENTER\n");
2331
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002332 /* For PATA drive side cable detection to work, IDENTIFY must
2333 * be done backwards such that PDIAG- is released by the slave
2334 * device before the master device is identified.
2335 */
Tejun Heo02607312007-08-06 18:36:23 +09002336 ata_link_for_each_dev_reverse(dev, link) {
Tejun Heof58229f2007-08-06 18:36:23 +09002337 unsigned int action = ata_eh_dev_action(dev);
2338 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09002339
Tejun Heobff04642006-11-10 18:08:10 +09002340 if (ehc->i.flags & ATA_EHI_DID_RESET)
2341 readid_flags |= ATA_READID_POSTRESET;
2342
Tejun Heo9666f402007-05-04 21:27:47 +02002343 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a2007-09-23 13:19:54 +09002344 WARN_ON(dev->class == ATA_DEV_PMP);
2345
Tejun Heo02607312007-08-06 18:36:23 +09002346 if (ata_link_offline(link)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002347 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002348 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002349 }
2350
Tejun Heo02607312007-08-06 18:36:23 +09002351 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09002352 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2353 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09002354 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002355 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002356
Tejun Heo02607312007-08-06 18:36:23 +09002357 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09002358
Tejun Heobaa1e782006-11-01 18:39:27 +09002359 /* Configuration may have changed, reconfigure
2360 * transfer mode.
2361 */
2362 ehc->i.flags |= ATA_EHI_SETMODE;
2363
zhao, forrest3057ac32006-06-12 12:01:34 +08002364 /* schedule the scsi_rescan_device() here */
2365 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09002366 } else if (dev->class == ATA_DEV_UNKNOWN &&
2367 ehc->tries[dev->devno] &&
2368 ata_class_enabled(ehc->classes[dev->devno])) {
2369 dev->class = ehc->classes[dev->devno];
2370
Tejun Heo633273a2007-09-23 13:19:54 +09002371 if (dev->class == ATA_DEV_PMP)
2372 rc = sata_pmp_attach(dev);
2373 else
2374 rc = ata_dev_read_id(dev, &dev->class,
2375 readid_flags, dev->id);
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002376 switch (rc) {
2377 case 0:
Tejun Heof58229f2007-08-06 18:36:23 +09002378 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002379 break;
2380 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002381 /* IDENTIFY was issued to non-existent
2382 * device. No need to reset. Just
2383 * thaw and kill the device.
2384 */
2385 ata_eh_thaw_port(ap);
2386 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo084fe632006-05-31 18:28:03 +09002387 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002388 default:
2389 dev->class = ATA_DEV_UNKNOWN;
2390 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002391 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002392 }
2393 }
2394
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002395 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002396 if (ata_is_host_link(link) && ap->ops->cable_detect &&
2397 (ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002398 ap->cbl = ap->ops->cable_detect(ap);
2399
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002400 /* Configure new devices forward such that user doesn't see
2401 * device detection messages backwards.
2402 */
Tejun Heo02607312007-08-06 18:36:23 +09002403 ata_link_for_each_dev(dev, link) {
Tejun Heo633273a2007-09-23 13:19:54 +09002404 if (!(new_mask & (1 << dev->devno)) ||
2405 dev->class == ATA_DEV_PMP)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002406 continue;
2407
2408 ehc->i.flags |= ATA_EHI_PRINTINFO;
2409 rc = ata_dev_configure(dev);
2410 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2411 if (rc)
2412 goto err;
2413
2414 spin_lock_irqsave(ap->lock, flags);
2415 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2416 spin_unlock_irqrestore(ap->lock, flags);
2417
2418 /* new device discovered, configure xfermode */
2419 ehc->i.flags |= ATA_EHI_SETMODE;
2420 }
2421
2422 return 0;
2423
2424 err:
2425 *r_failed_dev = dev;
2426 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002427 return rc;
2428}
2429
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002430/**
2431 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2432 * @link: link on which timings will be programmed
2433 * @r_failed_dev: out paramter for failed device
2434 *
2435 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2436 * ata_set_mode() fails, pointer to the failing device is
2437 * returned in @r_failed_dev.
2438 *
2439 * LOCKING:
2440 * PCI/etc. bus probe sem.
2441 *
2442 * RETURNS:
2443 * 0 on success, negative errno otherwise
2444 */
2445int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2446{
2447 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09002448 struct ata_device *dev;
2449 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002450
Tejun Heo76326ac2007-11-27 19:28:59 +09002451 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2452 ata_link_for_each_dev(dev, link) {
2453 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2454 struct ata_ering_entry *ent;
2455
2456 ent = ata_ering_top(&dev->ering);
2457 if (ent)
2458 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2459 }
2460 }
2461
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002462 /* has private set_mode? */
2463 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09002464 rc = ap->ops->set_mode(link, r_failed_dev);
2465 else
2466 rc = ata_do_set_mode(link, r_failed_dev);
2467
2468 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2469 ata_link_for_each_dev(dev, link) {
2470 struct ata_eh_context *ehc = &link->eh_context;
2471 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2472 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2473
2474 if (dev->xfer_mode != saved_xfer_mode ||
2475 ata_ncq_enabled(dev) != saved_ncq)
2476 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2477 }
2478
2479 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002480}
2481
Tejun Heo02607312007-08-06 18:36:23 +09002482static int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002483{
Tejun Heof58229f2007-08-06 18:36:23 +09002484 struct ata_device *dev;
2485 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002486
Tejun Heo02607312007-08-06 18:36:23 +09002487 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002488 if (ata_dev_enabled(dev))
Tejun Heo022bdb02006-05-15 20:58:22 +09002489 cnt++;
2490 return cnt;
2491}
2492
Tejun Heo02607312007-08-06 18:36:23 +09002493static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002494{
Tejun Heof58229f2007-08-06 18:36:23 +09002495 struct ata_device *dev;
2496 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002497
Tejun Heo02607312007-08-06 18:36:23 +09002498 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002499 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09002500 cnt++;
2501 return cnt;
2502}
2503
Tejun Heo02607312007-08-06 18:36:23 +09002504static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002505{
Tejun Heo02607312007-08-06 18:36:23 +09002506 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09002507 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09002508
Tejun Heof9df58c2007-09-23 13:14:13 +09002509 /* skip disabled links */
2510 if (link->flags & ATA_LFLAG_DISABLED)
2511 return 1;
2512
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09002513 /* thaw frozen port, resume link and recover failed devices */
Tejun Heo02607312007-08-06 18:36:23 +09002514 if ((link->ap->pflags & ATA_PFLAG_FROZEN) ||
2515 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_link_nr_enabled(link))
Tejun Heo084fe632006-05-31 18:28:03 +09002516 return 0;
2517
2518 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo02607312007-08-06 18:36:23 +09002519 ata_link_for_each_dev(dev, link) {
Tejun Heo084fe632006-05-31 18:28:03 +09002520 if (dev->class == ATA_DEV_UNKNOWN &&
2521 ehc->classes[dev->devno] != ATA_DEV_NONE)
2522 return 0;
2523 }
2524
2525 return 1;
2526}
2527
Tejun Heo02c05a22007-11-27 19:28:54 +09002528static int ata_eh_schedule_probe(struct ata_device *dev)
2529{
2530 struct ata_eh_context *ehc = &dev->link->eh_context;
2531
2532 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2533 (ehc->did_probe_mask & (1 << dev->devno)))
2534 return 0;
2535
2536 ata_eh_detach_dev(dev);
2537 ata_dev_init(dev);
2538 ehc->did_probe_mask |= (1 << dev->devno);
2539 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo00115e02007-11-27 19:28:58 +09002540 ehc->saved_xfer_mode[dev->devno] = 0;
2541 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09002542
2543 return 1;
2544}
2545
Tejun Heo9b1e2652007-08-06 18:36:24 +09002546static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09002547{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09002548 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09002549
2550 ehc->tries[dev->devno]--;
2551
2552 switch (err) {
2553 case -ENODEV:
2554 /* device missing or wrong IDENTIFY data, schedule probing */
2555 ehc->i.probe_mask |= (1 << dev->devno);
2556 case -EINVAL:
2557 /* give it just one more chance */
2558 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2559 case -EIO:
Tejun Heo4fb46152007-10-29 16:45:05 +09002560 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
Tejun Heofee7ca72007-07-01 19:05:58 +09002561 /* This is the last chance, better to slow
2562 * down than lose it.
2563 */
Tejun Heo936fd732007-08-06 18:36:23 +09002564 sata_down_spd_limit(dev->link);
Tejun Heofee7ca72007-07-01 19:05:58 +09002565 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2566 }
2567 }
2568
2569 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2570 /* disable device if it has used up all its chances */
2571 ata_dev_disable(dev);
2572
2573 /* detach if offline */
Tejun Heo936fd732007-08-06 18:36:23 +09002574 if (ata_link_offline(dev->link))
Tejun Heofee7ca72007-07-01 19:05:58 +09002575 ata_eh_detach_dev(dev);
2576
Tejun Heo02c05a22007-11-27 19:28:54 +09002577 /* schedule probe if necessary */
2578 if (ata_eh_schedule_probe(dev))
Tejun Heofee7ca72007-07-01 19:05:58 +09002579 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002580
2581 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09002582 } else {
2583 /* soft didn't work? be haaaaard */
2584 if (ehc->i.flags & ATA_EHI_DID_RESET)
2585 ehc->i.action |= ATA_EH_HARDRESET;
2586 else
2587 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002588
2589 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09002590 }
2591}
2592
Tejun Heo022bdb02006-05-15 20:58:22 +09002593/**
2594 * ata_eh_recover - recover host port after error
2595 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002596 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002597 * @softreset: softreset method (can be NULL)
2598 * @hardreset: hardreset method (can be NULL)
2599 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002600 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09002601 *
2602 * This is the alpha and omega, eum and yang, heart and soul of
2603 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09002604 * recover each link and hotplug requests are recorded in the
2605 * link's eh_context. This function executes all the operations
2606 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09002607 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09002608 *
2609 * LOCKING:
2610 * Kernel thread context (may sleep).
2611 *
2612 * RETURNS:
2613 * 0 on success, -errno on failure.
2614 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002615int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2616 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2617 ata_postreset_fn_t postreset,
2618 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002619{
Tejun Heo9b1e2652007-08-06 18:36:24 +09002620 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09002621 struct ata_device *dev;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002622 int nr_failed_devs, nr_disabled_devs;
2623 int reset, rc;
Tejun Heof9df58c2007-09-23 13:14:13 +09002624 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09002625
2626 DPRINTK("ENTER\n");
2627
2628 /* prep for recovery */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002629 ata_port_for_each_link(link, ap) {
2630 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09002631
Tejun Heof9df58c2007-09-23 13:14:13 +09002632 /* re-enable link? */
2633 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
2634 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
2635 spin_lock_irqsave(ap->lock, flags);
2636 link->flags &= ~ATA_LFLAG_DISABLED;
2637 spin_unlock_irqrestore(ap->lock, flags);
2638 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
2639 }
2640
Tejun Heo9b1e2652007-08-06 18:36:24 +09002641 ata_link_for_each_dev(dev, link) {
Tejun Heofd995f72007-09-23 13:14:12 +09002642 if (link->flags & ATA_LFLAG_NO_RETRY)
2643 ehc->tries[dev->devno] = 1;
2644 else
2645 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09002646
Tejun Heo9b1e2652007-08-06 18:36:24 +09002647 /* collect port action mask recorded in dev actions */
2648 ehc->i.action |= ehc->i.dev_action[dev->devno] &
2649 ~ATA_EH_PERDEV_MASK;
2650 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09002651
Tejun Heo9b1e2652007-08-06 18:36:24 +09002652 /* process hotplug request */
2653 if (dev->flags & ATA_DFLAG_DETACH)
2654 ata_eh_detach_dev(dev);
2655
Tejun Heo02c05a22007-11-27 19:28:54 +09002656 /* schedule probe if necessary */
2657 if (!ata_dev_enabled(dev))
2658 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09002659 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002660 }
2661
2662 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09002663 rc = 0;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002664 nr_failed_devs = 0;
2665 nr_disabled_devs = 0;
2666 reset = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002667
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002668 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09002669 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002670 goto out;
2671
Tejun Heo9b1e2652007-08-06 18:36:24 +09002672 /* prep for EH */
2673 ata_port_for_each_link(link, ap) {
2674 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002675
Tejun Heo9b1e2652007-08-06 18:36:24 +09002676 /* skip EH if possible. */
2677 if (ata_eh_skip_recovery(link))
2678 ehc->i.action = 0;
2679
2680 /* do we need to reset? */
2681 if (ehc->i.action & ATA_EH_RESET_MASK)
2682 reset = 1;
2683
2684 ata_link_for_each_dev(dev, link)
2685 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2686 }
Tejun Heo084fe632006-05-31 18:28:03 +09002687
Tejun Heo022bdb02006-05-15 20:58:22 +09002688 /* reset */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002689 if (reset) {
Tejun Heob06ce3e2007-10-09 15:06:48 +09002690 /* if PMP is attached, this function only deals with
2691 * downstream links, port should stay thawed.
2692 */
2693 if (!ap->nr_pmp_links)
2694 ata_eh_freeze_port(ap);
Tejun Heo022bdb02006-05-15 20:58:22 +09002695
Tejun Heo9b1e2652007-08-06 18:36:24 +09002696 ata_port_for_each_link(link, ap) {
2697 struct ata_eh_context *ehc = &link->eh_context;
2698
2699 if (!(ehc->i.action & ATA_EH_RESET_MASK))
2700 continue;
2701
2702 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
2703 prereset, softreset, hardreset,
2704 postreset);
2705 if (rc) {
2706 ata_link_printk(link, KERN_ERR,
2707 "reset failed, giving up\n");
2708 goto out;
2709 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002710 }
2711
Tejun Heob06ce3e2007-10-09 15:06:48 +09002712 if (!ap->nr_pmp_links)
2713 ata_eh_thaw_port(ap);
Tejun Heo022bdb02006-05-15 20:58:22 +09002714 }
2715
Tejun Heo9b1e2652007-08-06 18:36:24 +09002716 /* the rest */
2717 ata_port_for_each_link(link, ap) {
2718 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002719
Tejun Heo9b1e2652007-08-06 18:36:24 +09002720 /* revalidate existing devices and attach new ones */
2721 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09002722 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09002723 goto dev_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002724
Tejun Heo633273a2007-09-23 13:19:54 +09002725 /* if PMP got attached, return, pmp EH will take care of it */
2726 if (link->device->class == ATA_DEV_PMP) {
2727 ehc->i.action = 0;
2728 return 0;
2729 }
2730
Tejun Heo9b1e2652007-08-06 18:36:24 +09002731 /* configure transfer mode if necessary */
2732 if (ehc->i.flags & ATA_EHI_SETMODE) {
2733 rc = ata_set_mode(link, &dev);
2734 if (rc)
2735 goto dev_fail;
2736 ehc->i.flags &= ~ATA_EHI_SETMODE;
2737 }
2738
Kristen Carlson Accardica773292007-10-25 00:58:59 -04002739 if (ehc->i.action & ATA_EHI_LPM)
2740 ata_link_for_each_dev(dev, link)
2741 ata_dev_enable_pm(dev, ap->pm_policy);
2742
Tejun Heo9b1e2652007-08-06 18:36:24 +09002743 /* this link is okay now */
2744 ehc->i.flags = 0;
2745 continue;
2746
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002747dev_fail:
Tejun Heo9b1e2652007-08-06 18:36:24 +09002748 nr_failed_devs++;
2749 if (ata_eh_handle_dev_fail(dev, rc))
2750 nr_disabled_devs++;
2751
Tejun Heob06ce3e2007-10-09 15:06:48 +09002752 if (ap->pflags & ATA_PFLAG_FROZEN) {
2753 /* PMP reset requires working host port.
2754 * Can't retry if it's frozen.
2755 */
2756 if (ap->nr_pmp_links)
2757 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002758 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09002759 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002760 }
2761
Tejun Heo9b1e2652007-08-06 18:36:24 +09002762 if (nr_failed_devs) {
2763 if (nr_failed_devs != nr_disabled_devs) {
2764 ata_port_printk(ap, KERN_WARNING, "failed to recover "
2765 "some devices, retrying in 5 secs\n");
2766 ssleep(5);
2767 } else {
2768 /* no device left to recover, repeat fast */
2769 msleep(500);
2770 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002771
Tejun Heo9b1e2652007-08-06 18:36:24 +09002772 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002773 }
2774
Tejun Heo022bdb02006-05-15 20:58:22 +09002775 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09002776 if (rc && r_failed_link)
2777 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09002778
2779 DPRINTK("EXIT, rc=%d\n", rc);
2780 return rc;
2781}
2782
2783/**
2784 * ata_eh_finish - finish up EH
2785 * @ap: host port to finish EH for
2786 *
2787 * Recovery is complete. Clean up EH states and retry or finish
2788 * failed qcs.
2789 *
2790 * LOCKING:
2791 * None.
2792 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002793void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09002794{
2795 int tag;
2796
2797 /* retry or finish qcs */
2798 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2799 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2800
2801 if (!(qc->flags & ATA_QCFLAG_FAILED))
2802 continue;
2803
2804 if (qc->err_mask) {
2805 /* FIXME: Once EH migration is complete,
2806 * generate sense data in this function,
2807 * considering both err_mask and tf.
Tejun Heof90f0822007-10-26 16:12:41 +09002808 *
2809 * There's no point in retrying invalid
2810 * (detected by libata) and non-IO device
2811 * errors (rejected by device). Finish them
2812 * immediately.
Tejun Heo022bdb02006-05-15 20:58:22 +09002813 */
Tejun Heof90f0822007-10-26 16:12:41 +09002814 if ((qc->err_mask & AC_ERR_INVALID) ||
2815 (!(qc->flags & ATA_QCFLAG_IO) &&
2816 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002817 ata_eh_qc_complete(qc);
2818 else
2819 ata_eh_qc_retry(qc);
2820 } else {
2821 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2822 ata_eh_qc_complete(qc);
2823 } else {
2824 /* feed zero TF to sense generation */
2825 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2826 ata_eh_qc_retry(qc);
2827 }
2828 }
2829 }
Tejun Heoda917d62007-09-23 13:14:12 +09002830
2831 /* make sure nr_active_links is zero after EH */
2832 WARN_ON(ap->nr_active_links);
2833 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002834}
2835
2836/**
2837 * ata_do_eh - do standard error handling
2838 * @ap: host port to handle error for
Tejun Heof5914a42006-05-31 18:27:48 +09002839 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002840 * @softreset: softreset method (can be NULL)
2841 * @hardreset: hardreset method (can be NULL)
2842 * @postreset: postreset method (can be NULL)
2843 *
2844 * Perform standard error handling sequence.
2845 *
2846 * LOCKING:
2847 * Kernel thread context (may sleep).
2848 */
Tejun Heof5914a42006-05-31 18:27:48 +09002849void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2850 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2851 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002852{
Tejun Heo9b1e2652007-08-06 18:36:24 +09002853 struct ata_device *dev;
2854 int rc;
2855
2856 ata_eh_autopsy(ap);
2857 ata_eh_report(ap);
2858
2859 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
2860 NULL);
2861 if (rc) {
2862 ata_link_for_each_dev(dev, &ap->link)
2863 ata_dev_disable(dev);
2864 }
2865
Tejun Heo022bdb02006-05-15 20:58:22 +09002866 ata_eh_finish(ap);
2867}
Tejun Heo500530f2006-07-03 16:07:27 +09002868
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002869#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09002870/**
2871 * ata_eh_handle_port_suspend - perform port suspend operation
2872 * @ap: port to suspend
2873 *
2874 * Suspend @ap.
2875 *
2876 * LOCKING:
2877 * Kernel thread context (may sleep).
2878 */
2879static void ata_eh_handle_port_suspend(struct ata_port *ap)
2880{
2881 unsigned long flags;
2882 int rc = 0;
2883
2884 /* are we suspending? */
2885 spin_lock_irqsave(ap->lock, flags);
2886 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2887 ap->pm_mesg.event == PM_EVENT_ON) {
2888 spin_unlock_irqrestore(ap->lock, flags);
2889 return;
2890 }
2891 spin_unlock_irqrestore(ap->lock, flags);
2892
2893 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2894
Tejun Heo64578a32007-05-15 03:28:16 +09002895 /* tell ACPI we're suspending */
2896 rc = ata_acpi_on_suspend(ap);
2897 if (rc)
2898 goto out;
2899
Tejun Heo500530f2006-07-03 16:07:27 +09002900 /* suspend */
2901 ata_eh_freeze_port(ap);
2902
2903 if (ap->ops->port_suspend)
2904 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2905
Shaohua Libd3adca2007-11-02 09:32:38 +08002906 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a32007-05-15 03:28:16 +09002907 out:
Tejun Heo500530f2006-07-03 16:07:27 +09002908 /* report result */
2909 spin_lock_irqsave(ap->lock, flags);
2910
2911 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2912 if (rc == 0)
2913 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09002914 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09002915 ata_port_schedule_eh(ap);
2916
2917 if (ap->pm_result) {
2918 *ap->pm_result = rc;
2919 ap->pm_result = NULL;
2920 }
2921
2922 spin_unlock_irqrestore(ap->lock, flags);
2923
2924 return;
2925}
2926
2927/**
2928 * ata_eh_handle_port_resume - perform port resume operation
2929 * @ap: port to resume
2930 *
2931 * Resume @ap.
2932 *
Tejun Heo500530f2006-07-03 16:07:27 +09002933 * LOCKING:
2934 * Kernel thread context (may sleep).
2935 */
2936static void ata_eh_handle_port_resume(struct ata_port *ap)
2937{
Tejun Heo500530f2006-07-03 16:07:27 +09002938 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02002939 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09002940
2941 /* are we resuming? */
2942 spin_lock_irqsave(ap->lock, flags);
2943 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2944 ap->pm_mesg.event != PM_EVENT_ON) {
2945 spin_unlock_irqrestore(ap->lock, flags);
2946 return;
2947 }
2948 spin_unlock_irqrestore(ap->lock, flags);
2949
Tejun Heo9666f402007-05-04 21:27:47 +02002950 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09002951
Shaohua Libd3adca2007-11-02 09:32:38 +08002952 ata_acpi_set_state(ap, PMSG_ON);
2953
Tejun Heo500530f2006-07-03 16:07:27 +09002954 if (ap->ops->port_resume)
2955 rc = ap->ops->port_resume(ap);
2956
Tejun Heo67465442007-05-15 03:28:16 +09002957 /* tell ACPI that we're resuming */
2958 ata_acpi_on_resume(ap);
2959
Tejun Heo9666f402007-05-04 21:27:47 +02002960 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09002961 spin_lock_irqsave(ap->lock, flags);
2962 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2963 if (ap->pm_result) {
2964 *ap->pm_result = rc;
2965 ap->pm_result = NULL;
2966 }
2967 spin_unlock_irqrestore(ap->lock, flags);
2968}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002969#endif /* CONFIG_PM */