blob: 95838b3822313bebb6282642a716c486a9d332a2 [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>
Jens Axboe242f9dc2008-09-14 05:55:09 -070036#include <linux/blkdev.h>
Jeff Garzik28555682007-10-11 17:12:35 -040037#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090038#include <scsi/scsi.h>
39#include <scsi/scsi_host.h>
40#include <scsi/scsi_eh.h>
41#include <scsi/scsi_device.h>
42#include <scsi/scsi_cmnd.h>
Robert Hancock65211482009-07-14 20:43:39 -060043#include <scsi/scsi_dbg.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040044#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090045
46#include <linux/libata.h>
47
48#include "libata.h"
49
Tejun Heo7d47e8d2007-02-02 16:22:31 +090050enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090051 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090052 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090055 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090056
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090059 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Gwendal Grignoud9027472010-05-25 12:31:38 -070060 ATA_EFLAG_OLD_ER = (1 << 31),
Tejun Heo3884f7b2007-11-27 19:28:56 +090061
62 /* error categories */
63 ATA_ECAT_NONE = 0,
64 ATA_ECAT_ATA_BUS = 1,
65 ATA_ECAT_TOUT_HSM = 2,
66 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090067 ATA_ECAT_DUBIOUS_NONE = 4,
68 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
69 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
70 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090072
Tejun Heo87fbc5a2008-05-20 02:17:54 +090073 ATA_EH_CMD_DFL_TIMEOUT = 5000,
74
Tejun Heo0a2c0f52008-05-20 02:17:52 +090075 /* always put at least this amount of time between resets */
76 ATA_EH_RESET_COOL_DOWN = 5000,
77
Tejun Heo341c2c92008-05-20 02:17:51 +090078 /* Waiting in ->prereset can never be reliable. It's
79 * sometimes nice to wait there but it can't be depended upon;
80 * otherwise, we wouldn't be resetting. Just give it enough
81 * time for most drives to spin up.
82 */
83 ATA_EH_PRERESET_TIMEOUT = 10000,
84 ATA_EH_FASTDRAIN_INTERVAL = 3000,
Tejun Heo11fc33d2008-08-30 14:20:01 +020085
86 ATA_EH_UA_TRIES = 5,
Tejun Heoc2c7a892009-01-29 20:31:34 +090087
88 /* probe speed down parameters, see ata_eh_schedule_probe() */
89 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
90 ATA_EH_PROBE_TRIALS = 2,
Tejun Heo31daabd2007-02-02 16:50:52 +090091};
92
93/* The following table determines how we sequence resets. Each entry
94 * represents timeout for that try. The first try can be soft or
95 * hardreset. All others are hardreset if available. In most cases
96 * the first reset w/ 10sec timeout should succeed. Following entries
97 * are mostly for error handling, hotplug and retarded devices.
98 */
99static const unsigned long ata_eh_reset_timeouts[] = {
Tejun Heo341c2c92008-05-20 02:17:51 +0900100 10000, /* most drives spin up by 10sec */
101 10000, /* > 99% working drives spin up before 20sec */
102 35000, /* give > 30 secs of idleness for retarded devices */
103 5000, /* and sweet one last chance */
Tejun Heod8af0eb2008-05-20 02:17:53 +0900104 ULONG_MAX, /* > 1 min has elapsed, give up */
Tejun Heo31daabd2007-02-02 16:50:52 +0900105};
106
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900107static const unsigned long ata_eh_identify_timeouts[] = {
108 5000, /* covers > 99% of successes and not too boring on failures */
109 10000, /* combined time till here is enough even for media access */
110 30000, /* for true idiots */
111 ULONG_MAX,
112};
113
Tejun Heo6013efd2009-11-19 15:36:45 +0900114static const unsigned long ata_eh_flush_timeouts[] = {
115 15000, /* be generous with flush */
116 15000, /* ditto */
117 30000, /* and even more generous */
118 ULONG_MAX,
119};
120
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900121static const unsigned long ata_eh_other_timeouts[] = {
122 5000, /* same rationale as identify timeout */
123 10000, /* ditto */
124 /* but no merciful 30sec for other commands, it just isn't worth it */
125 ULONG_MAX,
126};
127
128struct ata_eh_cmd_timeout_ent {
129 const u8 *commands;
130 const unsigned long *timeouts;
131};
132
133/* The following table determines timeouts to use for EH internal
134 * commands. Each table entry is a command class and matches the
135 * commands the entry applies to and the timeout table to use.
136 *
137 * On the retry after a command timed out, the next timeout value from
138 * the table is used. If the table doesn't contain further entries,
139 * the last value is used.
140 *
141 * ehc->cmd_timeout_idx keeps track of which timeout to use per
142 * command class, so if SET_FEATURES times out on the first try, the
143 * next try will use the second timeout value only for that class.
144 */
145#define CMDS(cmds...) (const u8 []){ cmds, 0 }
146static const struct ata_eh_cmd_timeout_ent
147ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
148 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
149 .timeouts = ata_eh_identify_timeouts, },
150 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
151 .timeouts = ata_eh_other_timeouts, },
152 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_SET_FEATURES),
155 .timeouts = ata_eh_other_timeouts, },
156 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
157 .timeouts = ata_eh_other_timeouts, },
Tejun Heo6013efd2009-11-19 15:36:45 +0900158 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
159 .timeouts = ata_eh_flush_timeouts },
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900160};
161#undef CMDS
162
Tejun Heoad9e2762006-05-15 20:58:12 +0900163static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900164#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +0900165static void ata_eh_handle_port_suspend(struct ata_port *ap);
166static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900167#else /* CONFIG_PM */
168static void ata_eh_handle_port_suspend(struct ata_port *ap)
169{ }
170
171static void ata_eh_handle_port_resume(struct ata_port *ap)
172{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900173#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900174
Tejun Heob64bbc32007-07-16 14:29:39 +0900175static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
176 va_list args)
177{
178 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
179 ATA_EH_DESC_LEN - ehi->desc_len,
180 fmt, args);
181}
182
183/**
184 * __ata_ehi_push_desc - push error description without adding separator
185 * @ehi: target EHI
186 * @fmt: printf format string
187 *
188 * Format string according to @fmt and append it to @ehi->desc.
189 *
190 * LOCKING:
191 * spin_lock_irqsave(host lock)
192 */
193void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
194{
195 va_list args;
196
197 va_start(args, fmt);
198 __ata_ehi_pushv_desc(ehi, fmt, args);
199 va_end(args);
200}
201
202/**
203 * ata_ehi_push_desc - push error description with separator
204 * @ehi: target EHI
205 * @fmt: printf format string
206 *
207 * Format string according to @fmt and append it to @ehi->desc.
208 * If @ehi->desc is not empty, ", " is added in-between.
209 *
210 * LOCKING:
211 * spin_lock_irqsave(host lock)
212 */
213void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
214{
215 va_list args;
216
217 if (ehi->desc_len)
218 __ata_ehi_push_desc(ehi, ", ");
219
220 va_start(args, fmt);
221 __ata_ehi_pushv_desc(ehi, fmt, args);
222 va_end(args);
223}
224
225/**
226 * ata_ehi_clear_desc - clean error description
227 * @ehi: target EHI
228 *
229 * Clear @ehi->desc.
230 *
231 * LOCKING:
232 * spin_lock_irqsave(host lock)
233 */
234void ata_ehi_clear_desc(struct ata_eh_info *ehi)
235{
236 ehi->desc[0] = '\0';
237 ehi->desc_len = 0;
238}
239
Tejun Heocbcdd872007-08-18 13:14:55 +0900240/**
241 * ata_port_desc - append port description
242 * @ap: target ATA port
243 * @fmt: printf format string
244 *
245 * Format string according to @fmt and append it to port
246 * description. If port description is not empty, " " is added
247 * in-between. This function is to be used while initializing
248 * ata_host. The description is printed on host registration.
249 *
250 * LOCKING:
251 * None.
252 */
253void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
254{
255 va_list args;
256
257 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
258
259 if (ap->link.eh_info.desc_len)
260 __ata_ehi_push_desc(&ap->link.eh_info, " ");
261
262 va_start(args, fmt);
263 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
264 va_end(args);
265}
266
267#ifdef CONFIG_PCI
268
269/**
270 * ata_port_pbar_desc - append PCI BAR description
271 * @ap: target ATA port
272 * @bar: target PCI BAR
273 * @offset: offset into PCI BAR
274 * @name: name of the area
275 *
276 * If @offset is negative, this function formats a string which
277 * contains the name, address, size and type of the BAR and
278 * appends it to the port description. If @offset is zero or
279 * positive, only name and offsetted address is appended.
280 *
281 * LOCKING:
282 * None.
283 */
284void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
285 const char *name)
286{
287 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
288 char *type = "";
289 unsigned long long start, len;
290
291 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
292 type = "m";
293 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
294 type = "i";
295
296 start = (unsigned long long)pci_resource_start(pdev, bar);
297 len = (unsigned long long)pci_resource_len(pdev, bar);
298
299 if (offset < 0)
300 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
301 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800302 ata_port_desc(ap, "%s 0x%llx", name,
303 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900304}
305
306#endif /* CONFIG_PCI */
307
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900308static int ata_lookup_timeout_table(u8 cmd)
309{
310 int i;
311
312 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
313 const u8 *cur;
314
315 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
316 if (*cur == cmd)
317 return i;
318 }
319
320 return -1;
321}
322
323/**
324 * ata_internal_cmd_timeout - determine timeout for an internal command
325 * @dev: target device
326 * @cmd: internal command to be issued
327 *
328 * Determine timeout for internal command @cmd for @dev.
329 *
330 * LOCKING:
331 * EH context.
332 *
333 * RETURNS:
334 * Determined timeout.
335 */
336unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
337{
338 struct ata_eh_context *ehc = &dev->link->eh_context;
339 int ent = ata_lookup_timeout_table(cmd);
340 int idx;
341
342 if (ent < 0)
343 return ATA_EH_CMD_DFL_TIMEOUT;
344
345 idx = ehc->cmd_timeout_idx[dev->devno][ent];
346 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
347}
348
349/**
350 * ata_internal_cmd_timed_out - notification for internal command timeout
351 * @dev: target device
352 * @cmd: internal command which timed out
353 *
354 * Notify EH that internal command @cmd for @dev timed out. This
355 * function should be called only for commands whose timeouts are
356 * determined using ata_internal_cmd_timeout().
357 *
358 * LOCKING:
359 * EH context.
360 */
361void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
362{
363 struct ata_eh_context *ehc = &dev->link->eh_context;
364 int ent = ata_lookup_timeout_table(cmd);
365 int idx;
366
367 if (ent < 0)
368 return;
369
370 idx = ehc->cmd_timeout_idx[dev->devno][ent];
371 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
372 ehc->cmd_timeout_idx[dev->devno][ent]++;
373}
374
Tejun Heo3884f7b2007-11-27 19:28:56 +0900375static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900376 unsigned int err_mask)
377{
378 struct ata_ering_entry *ent;
379
380 WARN_ON(!err_mask);
381
382 ering->cursor++;
383 ering->cursor %= ATA_ERING_SIZE;
384
385 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900386 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900387 ent->err_mask = err_mask;
388 ent->timestamp = get_jiffies_64();
389}
390
Tejun Heo76326ac2007-11-27 19:28:59 +0900391static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
392{
393 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
394
395 if (ent->err_mask)
396 return ent;
397 return NULL;
398}
399
Gwendal Grignoud9027472010-05-25 12:31:38 -0700400int ata_ering_map(struct ata_ering *ering,
401 int (*map_fn)(struct ata_ering_entry *, void *),
402 void *arg)
Tejun Heo0c247c52006-05-15 20:58:19 +0900403{
404 int idx, rc = 0;
405 struct ata_ering_entry *ent;
406
407 idx = ering->cursor;
408 do {
409 ent = &ering->ring[idx];
410 if (!ent->err_mask)
411 break;
412 rc = map_fn(ent, arg);
413 if (rc)
414 break;
415 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
416 } while (idx != ering->cursor);
417
418 return rc;
419}
420
Gwendal Grignoud9027472010-05-25 12:31:38 -0700421int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
422{
423 ent->eflags |= ATA_EFLAG_OLD_ER;
424 return 0;
425}
426
427static void ata_ering_clear(struct ata_ering *ering)
428{
429 ata_ering_map(ering, ata_ering_clear_cb, NULL);
430}
431
Tejun Heo64f65ca2006-06-24 20:30:18 +0900432static unsigned int ata_eh_dev_action(struct ata_device *dev)
433{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900434 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900435
436 return ehc->i.action | ehc->i.dev_action[dev->devno];
437}
438
Tejun Heof58229f2007-08-06 18:36:23 +0900439static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900440 struct ata_eh_info *ehi, unsigned int action)
441{
Tejun Heof58229f2007-08-06 18:36:23 +0900442 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900443
444 if (!dev) {
445 ehi->action &= ~action;
Tejun Heo1eca4362008-11-03 20:03:17 +0900446 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900447 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900448 } else {
449 /* doesn't make sense for port-wide EH actions */
450 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
451
452 /* break ehi->action into ehi->dev_action */
453 if (ehi->action & action) {
Tejun Heo1eca4362008-11-03 20:03:17 +0900454 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900455 ehi->dev_action[tdev->devno] |=
456 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900457 ehi->action &= ~action;
458 }
459
460 /* turn off the specified per-dev action */
461 ehi->dev_action[dev->devno] &= ~action;
462 }
463}
464
Tejun Heoece1d632006-04-02 18:51:53 +0900465/**
466 * ata_scsi_timed_out - SCSI layer time out callback
467 * @cmd: timed out SCSI command
468 *
469 * Handles SCSI layer timeout. We race with normal completion of
470 * the qc for @cmd. If the qc is already gone, we lose and let
471 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
472 * timed out and EH should be invoked. Prevent ata_qc_complete()
473 * from finishing it by setting EH_SCHEDULED and return
474 * EH_NOT_HANDLED.
475 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900476 * TODO: kill this function once old EH is gone.
477 *
Tejun Heoece1d632006-04-02 18:51:53 +0900478 * LOCKING:
479 * Called from timer context
480 *
481 * RETURNS:
482 * EH_HANDLED or EH_NOT_HANDLED
483 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700484enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
Tejun Heoece1d632006-04-02 18:51:53 +0900485{
486 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400487 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900488 unsigned long flags;
489 struct ata_queued_cmd *qc;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700490 enum blk_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900491
492 DPRINTK("ENTER\n");
493
Tejun Heoad9e2762006-05-15 20:58:12 +0900494 if (ap->ops->error_handler) {
Jens Axboe242f9dc2008-09-14 05:55:09 -0700495 ret = BLK_EH_NOT_HANDLED;
Tejun Heoad9e2762006-05-15 20:58:12 +0900496 goto out;
497 }
498
Jens Axboe242f9dc2008-09-14 05:55:09 -0700499 ret = BLK_EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400500 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900501 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900502 if (qc) {
503 WARN_ON(qc->scsicmd != cmd);
504 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
505 qc->err_mask |= AC_ERR_TIMEOUT;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700506 ret = BLK_EH_NOT_HANDLED;
Tejun Heoece1d632006-04-02 18:51:53 +0900507 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400508 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900509
Tejun Heoad9e2762006-05-15 20:58:12 +0900510 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900511 DPRINTK("EXIT, ret=%d\n", ret);
512 return ret;
513}
514
Tejun Heoece180d2008-11-03 20:04:37 +0900515static void ata_eh_unload(struct ata_port *ap)
516{
517 struct ata_link *link;
518 struct ata_device *dev;
519 unsigned long flags;
520
521 /* Restore SControl IPM and SPD for the next driver and
522 * disable attached devices.
523 */
524 ata_for_each_link(link, ap, PMP_FIRST) {
525 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
526 ata_for_each_dev(dev, link, ALL)
527 ata_dev_disable(dev);
528 }
529
530 /* freeze and set UNLOADED */
531 spin_lock_irqsave(ap->lock, flags);
532
533 ata_port_freeze(ap); /* won't be thawed */
534 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
535 ap->pflags |= ATA_PFLAG_UNLOADED;
536
537 spin_unlock_irqrestore(ap->lock, flags);
538}
539
Tejun Heoece1d632006-04-02 18:51:53 +0900540/**
541 * ata_scsi_error - SCSI layer error handler callback
542 * @host: SCSI host on which error occurred
543 *
544 * Handles SCSI-layer-thrown error events.
545 *
546 * LOCKING:
547 * Inherited from SCSI layer (none, can sleep)
548 *
549 * RETURNS:
550 * Zero.
551 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400552void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900553{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400554 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900555 int i;
Tejun Heoad9e2762006-05-15 20:58:12 +0900556 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900557
558 DPRINTK("ENTER\n");
559
Tejun Heoc4291372010-05-10 21:41:38 +0200560 /* make sure sff pio task is not running */
561 ata_sff_flush_pio_task(ap);
Tejun Heoece1d632006-04-02 18:51:53 +0900562
Jeff Garzikcca39742006-08-24 03:19:22 -0400563 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900564
Tejun Heoad9e2762006-05-15 20:58:12 +0900565 /* For new EH, all qcs are finished in one of three ways -
566 * normal completion, error completion, and SCSI timeout.
Alan Coxc96f1732009-03-24 10:23:46 +0000567 * Both completions can race against SCSI timeout. When normal
Tejun Heoad9e2762006-05-15 20:58:12 +0900568 * completion wins, the qc never reaches EH. When error
569 * completion wins, the qc has ATA_QCFLAG_FAILED set.
570 *
571 * When SCSI timeout wins, things are a bit more complex.
572 * Normal or error completion can occur after the timeout but
573 * before this point. In such cases, both types of
574 * completions are honored. A scmd is determined to have
575 * timed out iff its associated qc is active and not failed.
576 */
577 if (ap->ops->error_handler) {
578 struct scsi_cmnd *scmd, *tmp;
579 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900580
Tejun Heoe30349d2006-07-03 03:02:15 +0900581 spin_lock_irqsave(ap->lock, flags);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700582
Alan Coxc96f1732009-03-24 10:23:46 +0000583 /* This must occur under the ap->lock as we don't want
584 a polled recovery to race the real interrupt handler
Gwendal Grignoud9027472010-05-25 12:31:38 -0700585
Alan Coxc96f1732009-03-24 10:23:46 +0000586 The lost_interrupt handler checks for any completed but
587 non-notified command and completes much like an IRQ handler.
Gwendal Grignoud9027472010-05-25 12:31:38 -0700588
Alan Coxc96f1732009-03-24 10:23:46 +0000589 We then fall into the error recovery code which will treat
590 this as if normal completion won the race */
Tejun Heoad9e2762006-05-15 20:58:12 +0900591
Alan Coxc96f1732009-03-24 10:23:46 +0000592 if (ap->ops->lost_interrupt)
593 ap->ops->lost_interrupt(ap);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700594
Tejun Heoad9e2762006-05-15 20:58:12 +0900595 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
596 struct ata_queued_cmd *qc;
597
598 for (i = 0; i < ATA_MAX_QUEUE; i++) {
599 qc = __ata_qc_from_tag(ap, i);
600 if (qc->flags & ATA_QCFLAG_ACTIVE &&
601 qc->scsicmd == scmd)
602 break;
603 }
604
605 if (i < ATA_MAX_QUEUE) {
606 /* the scmd has an associated qc */
607 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
608 /* which hasn't failed yet, timeout */
609 qc->err_mask |= AC_ERR_TIMEOUT;
610 qc->flags |= ATA_QCFLAG_FAILED;
611 nr_timedout++;
612 }
613 } else {
614 /* Normal completion occurred after
615 * SCSI timeout but before this point.
616 * Successfully complete it.
617 */
618 scmd->retries = scmd->allowed;
619 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
620 }
621 }
622
623 /* If we have timed out qcs. They belong to EH from
624 * this point but the state of the controller is
625 * unknown. Freeze the port to make sure the IRQ
626 * handler doesn't diddle with those qcs. This must
627 * be done atomically w.r.t. setting QCFLAG_FAILED.
628 */
629 if (nr_timedout)
630 __ata_port_freeze(ap);
631
Tejun Heoe30349d2006-07-03 03:02:15 +0900632 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900633
634 /* initialize eh_tries */
635 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900636 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900637 spin_unlock_wait(ap->lock);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700638
Alan Coxc96f1732009-03-24 10:23:46 +0000639 /* If we timed raced normal completion and there is nothing to
640 recover nr_timedout == 0 why exactly are we doing error recovery ? */
Tejun Heoad9e2762006-05-15 20:58:12 +0900641
642 repeat:
643 /* invoke error handler */
644 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900645 struct ata_link *link;
646
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900647 /* kill fast drain timer */
648 del_timer_sync(&ap->fastdrain_timer);
649
Tejun Heo500530f2006-07-03 16:07:27 +0900650 /* process port resume request */
651 ata_eh_handle_port_resume(ap);
652
Tejun Heof3e81b12006-05-15 20:58:21 +0900653 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900654 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900655
Tejun Heo1eca4362008-11-03 20:03:17 +0900656 ata_for_each_link(link, ap, HOST_FIRST) {
Tejun Heo00115e02007-11-27 19:28:58 +0900657 struct ata_eh_context *ehc = &link->eh_context;
658 struct ata_device *dev;
659
Tejun Heocf1b86c2007-08-06 18:36:23 +0900660 memset(&link->eh_context, 0, sizeof(link->eh_context));
661 link->eh_context.i = link->eh_info;
662 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900663
Tejun Heo1eca4362008-11-03 20:03:17 +0900664 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +0900665 int devno = dev->devno;
666
667 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
668 if (ata_ncq_enabled(dev))
669 ehc->saved_ncq_enabled |= 1 << devno;
670 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900671 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900672
Tejun Heob51e9e52006-06-29 01:29:30 +0900673 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
674 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900675 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900676
Tejun Heoe30349d2006-07-03 03:02:15 +0900677 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900678
Tejun Heo500530f2006-07-03 16:07:27 +0900679 /* invoke EH, skip if unloading or suspended */
680 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900681 ap->ops->error_handler(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900682 else {
683 /* if unloading, commence suicide */
684 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
685 !(ap->pflags & ATA_PFLAG_UNLOADED))
686 ata_eh_unload(ap);
Tejun Heo720ba122006-05-31 18:28:13 +0900687 ata_eh_finish(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900688 }
Tejun Heoad9e2762006-05-15 20:58:12 +0900689
Tejun Heo500530f2006-07-03 16:07:27 +0900690 /* process port suspend request */
691 ata_eh_handle_port_suspend(ap);
692
Tejun Heoad9e2762006-05-15 20:58:12 +0900693 /* Exception might have happend after ->error_handler
694 * recovered the port but before this point. Repeat
695 * EH in such case.
696 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900697 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900698
Tejun Heob51e9e52006-06-29 01:29:30 +0900699 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900700 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900701 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900702 goto repeat;
703 }
704 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
Tejun Heoa1e10f72007-08-18 13:28:49 +0900705 "tries, giving up\n", ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900706 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900707 }
708
Tejun Heof3e81b12006-05-15 20:58:21 +0900709 /* this run is complete, make sure EH info is clear */
Tejun Heo1eca4362008-11-03 20:03:17 +0900710 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heocf1b86c2007-08-06 18:36:23 +0900711 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900712
Tejun Heoe30349d2006-07-03 03:02:15 +0900713 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900714 * that if exception occurs after this point but
715 * before EH completion, SCSI midlayer will
716 * re-initiate EH.
717 */
718 host->host_eh_scheduled = 0;
719
Tejun Heoe30349d2006-07-03 03:02:15 +0900720 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900721 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900722 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900723 ap->ops->eng_timeout(ap);
724 }
725
726 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900727 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
728
729 scsi_eh_flush_done_q(&ap->eh_done_q);
730
Tejun Heoad9e2762006-05-15 20:58:12 +0900731 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900732 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900733
Tejun Heo1cdaf532006-07-03 16:07:26 +0900734 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900735 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900736 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
Tejun Heoad72cf92010-07-02 10:03:52 +0200737 schedule_delayed_work(&ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900738
739 if (ap->pflags & ATA_PFLAG_RECOVERED)
740 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900741
Tejun Heob51e9e52006-06-29 01:29:30 +0900742 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900743
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900744 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900745 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900746 wake_up_all(&ap->eh_wait_q);
747
Tejun Heoe30349d2006-07-03 03:02:15 +0900748 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900749
Tejun Heoece1d632006-04-02 18:51:53 +0900750 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900751}
752
753/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900754 * ata_port_wait_eh - Wait for the currently pending EH to complete
755 * @ap: Port to wait EH for
756 *
757 * Wait until the currently pending EH is complete.
758 *
759 * LOCKING:
760 * Kernel thread context (may sleep).
761 */
762void ata_port_wait_eh(struct ata_port *ap)
763{
764 unsigned long flags;
765 DEFINE_WAIT(wait);
766
767 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400768 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900769
Tejun Heob51e9e52006-06-29 01:29:30 +0900770 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900771 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400772 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900773 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400774 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900775 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900776 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900777
Jeff Garzikba6a1302006-06-22 23:46:10 -0400778 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900779
780 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400781 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900782 msleep(10);
783 goto retry;
784 }
785}
786
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900787static int ata_eh_nr_in_flight(struct ata_port *ap)
788{
789 unsigned int tag;
790 int nr = 0;
791
792 /* count only non-internal commands */
793 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
794 if (ata_qc_from_tag(ap, tag))
795 nr++;
796
797 return nr;
798}
799
800void ata_eh_fastdrain_timerfn(unsigned long arg)
801{
802 struct ata_port *ap = (void *)arg;
803 unsigned long flags;
804 int cnt;
805
806 spin_lock_irqsave(ap->lock, flags);
807
808 cnt = ata_eh_nr_in_flight(ap);
809
810 /* are we done? */
811 if (!cnt)
812 goto out_unlock;
813
814 if (cnt == ap->fastdrain_cnt) {
815 unsigned int tag;
816
817 /* No progress during the last interval, tag all
818 * in-flight qcs as timed out and freeze the port.
819 */
820 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
821 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
822 if (qc)
823 qc->err_mask |= AC_ERR_TIMEOUT;
824 }
825
826 ata_port_freeze(ap);
827 } else {
828 /* some qcs have finished, give it another chance */
829 ap->fastdrain_cnt = cnt;
830 ap->fastdrain_timer.expires =
Tejun Heo341c2c92008-05-20 02:17:51 +0900831 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900832 add_timer(&ap->fastdrain_timer);
833 }
834
835 out_unlock:
836 spin_unlock_irqrestore(ap->lock, flags);
837}
838
839/**
840 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
841 * @ap: target ATA port
842 * @fastdrain: activate fast drain
843 *
844 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
845 * is non-zero and EH wasn't pending before. Fast drain ensures
846 * that EH kicks in in timely manner.
847 *
848 * LOCKING:
849 * spin_lock_irqsave(host lock)
850 */
851static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
852{
853 int cnt;
854
855 /* already scheduled? */
856 if (ap->pflags & ATA_PFLAG_EH_PENDING)
857 return;
858
859 ap->pflags |= ATA_PFLAG_EH_PENDING;
860
861 if (!fastdrain)
862 return;
863
864 /* do we have in-flight qcs? */
865 cnt = ata_eh_nr_in_flight(ap);
866 if (!cnt)
867 return;
868
869 /* activate fast drain */
870 ap->fastdrain_cnt = cnt;
Tejun Heo341c2c92008-05-20 02:17:51 +0900871 ap->fastdrain_timer.expires =
872 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900873 add_timer(&ap->fastdrain_timer);
874}
875
Tejun Heof686bcb2006-05-15 20:58:05 +0900876/**
877 * ata_qc_schedule_eh - schedule qc for error handling
878 * @qc: command to schedule error handling for
879 *
880 * Schedule error handling for @qc. EH will kick in as soon as
881 * other commands are drained.
882 *
883 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400884 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900885 */
886void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
887{
888 struct ata_port *ap = qc->ap;
Tejun Heofa41efd2010-04-15 08:57:37 +0900889 struct request_queue *q = qc->scsicmd->device->request_queue;
890 unsigned long flags;
Tejun Heof686bcb2006-05-15 20:58:05 +0900891
892 WARN_ON(!ap->ops->error_handler);
893
894 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900895 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900896
897 /* The following will fail if timeout has already expired.
898 * ata_scsi_error() takes care of such scmds on EH entry.
899 * Note that ATA_QCFLAG_FAILED is unconditionally set after
900 * this function completes.
901 */
Tejun Heofa41efd2010-04-15 08:57:37 +0900902 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700903 blk_abort_request(qc->scsicmd->request);
Tejun Heofa41efd2010-04-15 08:57:37 +0900904 spin_unlock_irqrestore(q->queue_lock, flags);
Tejun Heof686bcb2006-05-15 20:58:05 +0900905}
906
Tejun Heo7b70fc02006-05-15 20:58:07 +0900907/**
908 * ata_port_schedule_eh - schedule error handling without a qc
909 * @ap: ATA port to schedule EH for
910 *
911 * Schedule error handling for @ap. EH will kick in as soon as
912 * all commands are drained.
913 *
914 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400915 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900916 */
917void ata_port_schedule_eh(struct ata_port *ap)
918{
919 WARN_ON(!ap->ops->error_handler);
920
Tejun Heof4d6d002007-05-01 11:50:15 +0200921 if (ap->pflags & ATA_PFLAG_INITIALIZING)
922 return;
923
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900924 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -0400925 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900926
927 DPRINTK("port EH scheduled\n");
928}
929
Tejun Heodbd82612007-08-06 18:36:23 +0900930static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900931{
932 int tag, nr_aborted = 0;
933
934 WARN_ON(!ap->ops->error_handler);
935
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900936 /* we're gonna abort all commands, no need for fast drain */
937 ata_eh_set_pending(ap, 0);
938
Tejun Heo7b70fc02006-05-15 20:58:07 +0900939 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
940 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
941
Tejun Heodbd82612007-08-06 18:36:23 +0900942 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +0900943 qc->flags |= ATA_QCFLAG_FAILED;
944 ata_qc_complete(qc);
945 nr_aborted++;
946 }
947 }
948
949 if (!nr_aborted)
950 ata_port_schedule_eh(ap);
951
952 return nr_aborted;
953}
954
Tejun Heoe3180492006-05-15 20:58:09 +0900955/**
Tejun Heodbd82612007-08-06 18:36:23 +0900956 * ata_link_abort - abort all qc's on the link
957 * @link: ATA link to abort qc's for
958 *
959 * Abort all active qc's active on @link and schedule EH.
960 *
961 * LOCKING:
962 * spin_lock_irqsave(host lock)
963 *
964 * RETURNS:
965 * Number of aborted qc's.
966 */
967int ata_link_abort(struct ata_link *link)
968{
969 return ata_do_link_abort(link->ap, link);
970}
971
972/**
973 * ata_port_abort - abort all qc's on the port
974 * @ap: ATA port to abort qc's for
975 *
976 * Abort all active qc's of @ap and schedule EH.
977 *
978 * LOCKING:
979 * spin_lock_irqsave(host_set lock)
980 *
981 * RETURNS:
982 * Number of aborted qc's.
983 */
984int ata_port_abort(struct ata_port *ap)
985{
986 return ata_do_link_abort(ap, NULL);
987}
988
989/**
Tejun Heoe3180492006-05-15 20:58:09 +0900990 * __ata_port_freeze - freeze port
991 * @ap: ATA port to freeze
992 *
993 * This function is called when HSM violation or some other
994 * condition disrupts normal operation of the port. Frozen port
995 * is not allowed to perform any operation until the port is
996 * thawed, which usually follows a successful reset.
997 *
998 * ap->ops->freeze() callback can be used for freezing the port
999 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1000 * port cannot be frozen hardware-wise, the interrupt handler
1001 * must ack and clear interrupts unconditionally while the port
1002 * is frozen.
1003 *
1004 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001005 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001006 */
1007static void __ata_port_freeze(struct ata_port *ap)
1008{
1009 WARN_ON(!ap->ops->error_handler);
1010
1011 if (ap->ops->freeze)
1012 ap->ops->freeze(ap);
1013
Tejun Heob51e9e52006-06-29 01:29:30 +09001014 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001015
Tejun Heo44877b42007-02-21 01:06:51 +09001016 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001017}
1018
1019/**
1020 * ata_port_freeze - abort & freeze port
1021 * @ap: ATA port to freeze
1022 *
Jeff Garzik54c38442009-04-07 19:13:15 -04001023 * Abort and freeze @ap. The freeze operation must be called
1024 * first, because some hardware requires special operations
1025 * before the taskfile registers are accessible.
Tejun Heoe3180492006-05-15 20:58:09 +09001026 *
1027 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001028 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001029 *
1030 * RETURNS:
1031 * Number of aborted commands.
1032 */
1033int ata_port_freeze(struct ata_port *ap)
1034{
1035 int nr_aborted;
1036
1037 WARN_ON(!ap->ops->error_handler);
1038
Tejun Heoe3180492006-05-15 20:58:09 +09001039 __ata_port_freeze(ap);
Jeff Garzik54c38442009-04-07 19:13:15 -04001040 nr_aborted = ata_port_abort(ap);
Tejun Heoe3180492006-05-15 20:58:09 +09001041
1042 return nr_aborted;
1043}
1044
1045/**
Tejun Heo7d77b242007-09-23 13:14:13 +09001046 * sata_async_notification - SATA async notification handler
1047 * @ap: ATA port where async notification is received
1048 *
1049 * Handler to be called when async notification via SDB FIS is
1050 * received. This function schedules EH if necessary.
1051 *
1052 * LOCKING:
1053 * spin_lock_irqsave(host lock)
1054 *
1055 * RETURNS:
1056 * 1 if EH is scheduled, 0 otherwise.
1057 */
1058int sata_async_notification(struct ata_port *ap)
1059{
1060 u32 sntf;
1061 int rc;
1062
1063 if (!(ap->flags & ATA_FLAG_AN))
1064 return 0;
1065
1066 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1067 if (rc == 0)
1068 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1069
Tejun Heo071f44b2008-04-07 22:47:22 +09001070 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001071 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001072 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001073 /* PMP is not attached. Check whether ATAPI
1074 * AN is configured. If so, notify media
1075 * change.
1076 */
1077 struct ata_device *dev = ap->link.device;
1078
1079 if ((dev->class == ATA_DEV_ATAPI) &&
1080 (dev->flags & ATA_DFLAG_AN))
1081 ata_scsi_media_change_notify(dev);
1082 return 0;
1083 } else {
1084 /* PMP is attached but SNTF is not available.
1085 * ATAPI async media change notification is
1086 * not used. The PMP must be reporting PHY
1087 * status change, schedule EH.
1088 */
1089 ata_port_schedule_eh(ap);
1090 return 1;
1091 }
1092 } else {
1093 /* PMP is attached and SNTF is available */
1094 struct ata_link *link;
1095
1096 /* check and notify ATAPI AN */
Tejun Heo1eca4362008-11-03 20:03:17 +09001097 ata_for_each_link(link, ap, EDGE) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001098 if (!(sntf & (1 << link->pmp)))
1099 continue;
1100
1101 if ((link->device->class == ATA_DEV_ATAPI) &&
1102 (link->device->flags & ATA_DFLAG_AN))
1103 ata_scsi_media_change_notify(link->device);
1104 }
1105
1106 /* If PMP is reporting that PHY status of some
1107 * downstream ports has changed, schedule EH.
1108 */
1109 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1110 ata_port_schedule_eh(ap);
1111 return 1;
1112 }
1113
1114 return 0;
1115 }
1116}
1117
1118/**
Tejun Heoe3180492006-05-15 20:58:09 +09001119 * ata_eh_freeze_port - EH helper to freeze port
1120 * @ap: ATA port to freeze
1121 *
1122 * Freeze @ap.
1123 *
1124 * LOCKING:
1125 * None.
1126 */
1127void ata_eh_freeze_port(struct ata_port *ap)
1128{
1129 unsigned long flags;
1130
1131 if (!ap->ops->error_handler)
1132 return;
1133
Jeff Garzikba6a1302006-06-22 23:46:10 -04001134 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001135 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001136 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001137}
1138
1139/**
1140 * ata_port_thaw_port - EH helper to thaw port
1141 * @ap: ATA port to thaw
1142 *
1143 * Thaw frozen port @ap.
1144 *
1145 * LOCKING:
1146 * None.
1147 */
1148void ata_eh_thaw_port(struct ata_port *ap)
1149{
1150 unsigned long flags;
1151
1152 if (!ap->ops->error_handler)
1153 return;
1154
Jeff Garzikba6a1302006-06-22 23:46:10 -04001155 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001156
Tejun Heob51e9e52006-06-29 01:29:30 +09001157 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001158
1159 if (ap->ops->thaw)
1160 ap->ops->thaw(ap);
1161
Jeff Garzikba6a1302006-06-22 23:46:10 -04001162 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001163
Tejun Heo44877b42007-02-21 01:06:51 +09001164 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001165}
1166
Tejun Heoece1d632006-04-02 18:51:53 +09001167static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1168{
1169 /* nada */
1170}
1171
1172static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1173{
1174 struct ata_port *ap = qc->ap;
1175 struct scsi_cmnd *scmd = qc->scsicmd;
1176 unsigned long flags;
1177
Jeff Garzikba6a1302006-06-22 23:46:10 -04001178 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001179 qc->scsidone = ata_eh_scsidone;
1180 __ata_qc_complete(qc);
1181 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001182 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001183
1184 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1185}
1186
1187/**
1188 * ata_eh_qc_complete - Complete an active ATA command from EH
1189 * @qc: Command to complete
1190 *
1191 * Indicate to the mid and upper layers that an ATA command has
1192 * completed. To be used from EH.
1193 */
1194void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1195{
1196 struct scsi_cmnd *scmd = qc->scsicmd;
1197 scmd->retries = scmd->allowed;
1198 __ata_eh_qc_complete(qc);
1199}
1200
1201/**
1202 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1203 * @qc: Command to retry
1204 *
1205 * Indicate to the mid and upper layers that an ATA command
1206 * should be retried. To be used from EH.
1207 *
1208 * SCSI midlayer limits the number of retries to scmd->allowed.
1209 * scmd->retries is decremented for commands which get retried
1210 * due to unrelated failures (qc->err_mask is zero).
1211 */
1212void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1213{
1214 struct scsi_cmnd *scmd = qc->scsicmd;
1215 if (!qc->err_mask && scmd->retries)
1216 scmd->retries--;
1217 __ata_eh_qc_complete(qc);
1218}
Tejun Heo022bdb02006-05-15 20:58:22 +09001219
1220/**
Tejun Heo678afac2009-01-29 20:31:30 +09001221 * ata_dev_disable - disable ATA device
1222 * @dev: ATA device to disable
1223 *
1224 * Disable @dev.
1225 *
1226 * Locking:
1227 * EH context.
1228 */
1229void ata_dev_disable(struct ata_device *dev)
1230{
1231 if (!ata_dev_enabled(dev))
1232 return;
1233
1234 if (ata_msg_drv(dev->link->ap))
1235 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1236 ata_acpi_on_disable(dev);
1237 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1238 dev->class++;
Tejun Heo99cf6102009-01-29 20:31:32 +09001239
1240 /* From now till the next successful probe, ering is used to
1241 * track probe failures. Clear accumulated device error info.
1242 */
1243 ata_ering_clear(&dev->ering);
Tejun Heo678afac2009-01-29 20:31:30 +09001244}
1245
1246/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001247 * ata_eh_detach_dev - detach ATA device
1248 * @dev: ATA device to detach
1249 *
1250 * Detach @dev.
1251 *
1252 * LOCKING:
1253 * None.
1254 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001255void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001256{
Tejun Heof58229f2007-08-06 18:36:23 +09001257 struct ata_link *link = dev->link;
1258 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001259 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001260 unsigned long flags;
1261
1262 ata_dev_disable(dev);
1263
Jeff Garzikba6a1302006-06-22 23:46:10 -04001264 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001265
1266 dev->flags &= ~ATA_DFLAG_DETACH;
1267
1268 if (ata_scsi_offline_dev(dev)) {
1269 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001270 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001271 }
1272
Tejun Heo90484eb2008-10-26 15:43:03 +09001273 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001274 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1275 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001276 ehc->saved_xfer_mode[dev->devno] = 0;
1277 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001278
Jeff Garzikba6a1302006-06-22 23:46:10 -04001279 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001280}
1281
1282/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001283 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001284 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001285 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001286 * @action: action about to be performed
1287 *
1288 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001289 * in @link->eh_info such that eh actions are not unnecessarily
1290 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001291 *
1292 * LOCKING:
1293 * None.
1294 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001295void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1296 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001297{
Tejun Heo955e57d2007-08-06 18:36:23 +09001298 struct ata_port *ap = link->ap;
1299 struct ata_eh_info *ehi = &link->eh_info;
1300 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001301 unsigned long flags;
1302
Jeff Garzikba6a1302006-06-22 23:46:10 -04001303 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001304
Tejun Heo955e57d2007-08-06 18:36:23 +09001305 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001306
Tejun Heoa568d1d2008-10-21 20:37:21 +09001307 /* About to take EH action, set RECOVERED. Ignore actions on
1308 * slave links as master will do them again.
1309 */
1310 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001311 ap->pflags |= ATA_PFLAG_RECOVERED;
1312
Jeff Garzikba6a1302006-06-22 23:46:10 -04001313 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001314}
1315
1316/**
Tejun Heo47005f22006-06-19 18:27:23 +09001317 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001318* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001319 * @dev: target ATA dev for per-dev action (can be NULL)
1320 * @action: action just completed
1321 *
1322 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001323 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001324 *
1325 * LOCKING:
1326 * None.
1327 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001328void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1329 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001330{
Tejun Heo955e57d2007-08-06 18:36:23 +09001331 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001332
Tejun Heo955e57d2007-08-06 18:36:23 +09001333 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001334}
1335
1336/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001337 * ata_err_string - convert err_mask to descriptive string
1338 * @err_mask: error mask to convert to string
1339 *
1340 * Convert @err_mask to descriptive string. Errors are
1341 * prioritized according to severity and only the most severe
1342 * error is reported.
1343 *
1344 * LOCKING:
1345 * None.
1346 *
1347 * RETURNS:
1348 * Descriptive string for @err_mask
1349 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001350static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001351{
1352 if (err_mask & AC_ERR_HOST_BUS)
1353 return "host bus error";
1354 if (err_mask & AC_ERR_ATA_BUS)
1355 return "ATA bus error";
1356 if (err_mask & AC_ERR_TIMEOUT)
1357 return "timeout";
1358 if (err_mask & AC_ERR_HSM)
1359 return "HSM violation";
1360 if (err_mask & AC_ERR_SYSTEM)
1361 return "internal error";
1362 if (err_mask & AC_ERR_MEDIA)
1363 return "media error";
1364 if (err_mask & AC_ERR_INVALID)
1365 return "invalid argument";
1366 if (err_mask & AC_ERR_DEV)
1367 return "device error";
1368 return "unknown error";
1369}
1370
1371/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001372 * ata_read_log_page - read a specific log page
1373 * @dev: target device
1374 * @page: page to read
1375 * @buf: buffer to store read page
1376 * @sectors: number of sectors to read
1377 *
1378 * Read log page using READ_LOG_EXT command.
1379 *
1380 * LOCKING:
1381 * Kernel thread context (may sleep).
1382 *
1383 * RETURNS:
1384 * 0 on success, AC_ERR_* mask otherwise.
1385 */
1386static unsigned int ata_read_log_page(struct ata_device *dev,
1387 u8 page, void *buf, unsigned int sectors)
1388{
1389 struct ata_taskfile tf;
1390 unsigned int err_mask;
1391
1392 DPRINTK("read log page - page %d\n", page);
1393
1394 ata_tf_init(dev, &tf);
1395 tf.command = ATA_CMD_READ_LOG_EXT;
1396 tf.lbal = page;
1397 tf.nsect = sectors;
1398 tf.hob_nsect = sectors >> 8;
1399 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1400 tf.protocol = ATA_PROT_PIO;
1401
1402 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001403 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001404
1405 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1406 return err_mask;
1407}
1408
1409/**
1410 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1411 * @dev: Device to read log page 10h from
1412 * @tag: Resulting tag of the failed command
1413 * @tf: Resulting taskfile registers of the failed command
1414 *
1415 * Read log page 10h to obtain NCQ error details and clear error
1416 * condition.
1417 *
1418 * LOCKING:
1419 * Kernel thread context (may sleep).
1420 *
1421 * RETURNS:
1422 * 0 on success, -errno otherwise.
1423 */
1424static int ata_eh_read_log_10h(struct ata_device *dev,
1425 int *tag, struct ata_taskfile *tf)
1426{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001427 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001428 unsigned int err_mask;
1429 u8 csum;
1430 int i;
1431
1432 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1433 if (err_mask)
1434 return -EIO;
1435
1436 csum = 0;
1437 for (i = 0; i < ATA_SECT_SIZE; i++)
1438 csum += buf[i];
1439 if (csum)
1440 ata_dev_printk(dev, KERN_WARNING,
1441 "invalid checksum 0x%x on log page 10h\n", csum);
1442
1443 if (buf[0] & 0x80)
1444 return -ENOENT;
1445
1446 *tag = buf[0] & 0x1f;
1447
1448 tf->command = buf[2];
1449 tf->feature = buf[3];
1450 tf->lbal = buf[4];
1451 tf->lbam = buf[5];
1452 tf->lbah = buf[6];
1453 tf->device = buf[7];
1454 tf->hob_lbal = buf[8];
1455 tf->hob_lbam = buf[9];
1456 tf->hob_lbah = buf[10];
1457 tf->nsect = buf[12];
1458 tf->hob_nsect = buf[13];
1459
1460 return 0;
1461}
1462
1463/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001464 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1465 * @dev: target ATAPI device
1466 * @r_sense_key: out parameter for sense_key
1467 *
1468 * Perform ATAPI TEST_UNIT_READY.
1469 *
1470 * LOCKING:
1471 * EH context (may sleep).
1472 *
1473 * RETURNS:
1474 * 0 on success, AC_ERR_* mask on failure.
1475 */
1476static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1477{
1478 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1479 struct ata_taskfile tf;
1480 unsigned int err_mask;
1481
1482 ata_tf_init(dev, &tf);
1483
1484 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1485 tf.command = ATA_CMD_PACKET;
1486 tf.protocol = ATAPI_PROT_NODATA;
1487
1488 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1489 if (err_mask == AC_ERR_DEV)
1490 *r_sense_key = tf.feature >> 4;
1491 return err_mask;
1492}
1493
1494/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001495 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1496 * @dev: device to perform REQUEST_SENSE to
1497 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001498 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001499 *
1500 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1501 * SENSE. This function is EH helper.
1502 *
1503 * LOCKING:
1504 * Kernel thread context (may sleep).
1505 *
1506 * RETURNS:
1507 * 0 on success, AC_ERR_* mask on failure
1508 */
Tejun Heo3eabddb2008-06-10 18:28:05 +09001509static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1510 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001511{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001512 u8 cdb[ATAPI_CDB_LEN] =
1513 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001514 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001515 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001516
1517 DPRINTK("ATAPI request sense\n");
1518
Tejun Heo022bdb02006-05-15 20:58:22 +09001519 /* FIXME: is this needed? */
1520 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1521
Albert Lee56287762007-04-02 11:30:46 +08001522 /* initialize sense_buf with the error register,
1523 * for the case where they are -not- overwritten
1524 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001525 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001526 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001527
Jeff Garzika617c092007-05-21 20:14:23 -04001528 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001529 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001530
Tejun Heo022bdb02006-05-15 20:58:22 +09001531 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1532 tf.command = ATA_CMD_PACKET;
1533
1534 /* is it pointless to prefer PIO for "safety reasons"? */
1535 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001536 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001537 tf.feature |= ATAPI_PKT_DMA;
1538 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001539 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001540 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1541 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001542 }
1543
1544 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001545 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001546}
1547
1548/**
1549 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001550 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001551 *
1552 * Analyze SError if available and further determine cause of
1553 * failure.
1554 *
1555 * LOCKING:
1556 * None.
1557 */
Tejun Heo02607312007-08-06 18:36:23 +09001558static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001559{
Tejun Heo02607312007-08-06 18:36:23 +09001560 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001561 u32 serror = ehc->i.serror;
1562 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001563 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001564
Tejun Heoe0614db2008-05-19 01:15:09 +09001565 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001566 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001567 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001568 }
1569 if (serror & SERR_PROTOCOL) {
1570 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001571 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001572 }
1573 if (serror & SERR_INTERNAL) {
1574 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001575 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001576 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001577
1578 /* Determine whether a hotplug event has occurred. Both
1579 * SError.N/X are considered hotplug events for enabled or
1580 * host links. For disabled PMP links, only N bit is
1581 * considered as X bit is left at 1 for link plugging.
1582 */
1583 hotplug_mask = 0;
1584
1585 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1586 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1587 else
1588 hotplug_mask = SERR_PHYRDY_CHG;
1589
1590 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001591 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001592
1593 ehc->i.err_mask |= err_mask;
1594 ehc->i.action |= action;
1595}
1596
1597/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001598 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001599 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001600 *
1601 * Read log page 10h, determine the offending qc and acquire
1602 * error status TF. For NCQ device errors, all LLDDs have to do
1603 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1604 * care of the rest.
1605 *
1606 * LOCKING:
1607 * Kernel thread context (may sleep).
1608 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001609void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001610{
Tejun Heo02607312007-08-06 18:36:23 +09001611 struct ata_port *ap = link->ap;
1612 struct ata_eh_context *ehc = &link->eh_context;
1613 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001614 struct ata_queued_cmd *qc;
1615 struct ata_taskfile tf;
1616 int tag, rc;
1617
1618 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001619 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001620 return;
1621
1622 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001623 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001624 return;
1625
1626 /* has LLDD analyzed already? */
1627 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1628 qc = __ata_qc_from_tag(ap, tag);
1629
1630 if (!(qc->flags & ATA_QCFLAG_FAILED))
1631 continue;
1632
1633 if (qc->err_mask)
1634 return;
1635 }
1636
1637 /* okay, this error is ours */
Jeff Garzika09bf4c2010-04-22 21:59:13 -04001638 memset(&tf, 0, sizeof(tf));
Tejun Heoe8ee8452006-05-15 21:03:46 +09001639 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1640 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001641 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001642 "(errno=%d)\n", rc);
1643 return;
1644 }
1645
Tejun Heo02607312007-08-06 18:36:23 +09001646 if (!(link->sactive & (1 << tag))) {
1647 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001648 "inactive tag %d\n", tag);
1649 return;
1650 }
1651
1652 /* we've got the perpetrator, condemn it */
1653 qc = __ata_qc_from_tag(ap, tag);
1654 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001655 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001656 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001657 ehc->i.err_mask &= ~AC_ERR_DEV;
1658}
1659
1660/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001661 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1662 * @qc: qc to analyze
1663 * @tf: Taskfile registers to analyze
1664 *
1665 * Analyze taskfile of @qc and further determine cause of
1666 * failure. This function also requests ATAPI sense data if
1667 * avaliable.
1668 *
1669 * LOCKING:
1670 * Kernel thread context (may sleep).
1671 *
1672 * RETURNS:
1673 * Determined recovery action
1674 */
1675static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1676 const struct ata_taskfile *tf)
1677{
1678 unsigned int tmp, action = 0;
1679 u8 stat = tf->command, err = tf->feature;
1680
1681 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1682 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001683 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001684 }
1685
Tejun Heoa51d6442007-03-20 15:24:11 +09001686 if (stat & (ATA_ERR | ATA_DF))
1687 qc->err_mask |= AC_ERR_DEV;
1688 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001689 return 0;
1690
1691 switch (qc->dev->class) {
1692 case ATA_DEV_ATA:
1693 if (err & ATA_ICRC)
1694 qc->err_mask |= AC_ERR_ATA_BUS;
1695 if (err & ATA_UNC)
1696 qc->err_mask |= AC_ERR_MEDIA;
1697 if (err & ATA_IDNF)
1698 qc->err_mask |= AC_ERR_INVALID;
1699 break;
1700
1701 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001702 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001703 tmp = atapi_eh_request_sense(qc->dev,
1704 qc->scsicmd->sense_buffer,
1705 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001706 if (!tmp) {
1707 /* ATA_QCFLAG_SENSE_VALID is used to
1708 * tell atapi_qc_complete() that sense
1709 * data is already valid.
1710 *
1711 * TODO: interpret sense data and set
1712 * appropriate err_mask.
1713 */
1714 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1715 } else
1716 qc->err_mask |= tmp;
1717 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001718 }
1719
1720 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001721 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001722
1723 return action;
1724}
1725
Tejun Heo76326ac2007-11-27 19:28:59 +09001726static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1727 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001728{
Tejun Heo76326ac2007-11-27 19:28:59 +09001729 int base = 0;
1730
1731 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1732 *xfer_ok = 1;
1733
1734 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001735 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001736
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001737 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001738 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001739
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001740 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001741 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001742
Tejun Heo3884f7b2007-11-27 19:28:56 +09001743 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001744 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001745 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001746 if ((err_mask &
1747 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001748 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001749 }
1750
1751 return 0;
1752}
1753
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001754struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001755 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001756 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001757 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001758};
1759
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001760static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001761{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001762 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001763 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001764
Gwendal Grignoud9027472010-05-25 12:31:38 -07001765 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
Tejun Heo022bdb02006-05-15 20:58:22 +09001766 return -1;
1767
Tejun Heo76326ac2007-11-27 19:28:59 +09001768 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1769 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001770 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001771
Tejun Heo022bdb02006-05-15 20:58:22 +09001772 return 0;
1773}
1774
1775/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001776 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001777 * @dev: Device of interest
1778 *
1779 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001780 * whether NCQ needs to be turned off, transfer speed should be
1781 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001782 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001783 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001784 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001785 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1786 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001787 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001788 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001789 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001790 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1791 * data transfer hasn't been verified.
1792 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001793 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001794 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001795 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001796 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001797 * SPEED_DOWN : Speed down transfer speed but don't fall back
1798 * to PIO.
1799 *
1800 * FALLBACK_TO_PIO : Fall back to PIO.
1801 *
1802 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001803 * taken per error. An action triggered by non-DUBIOUS errors
1804 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1805 * This is to expedite speed down decisions right after device is
1806 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001807 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001808 * The followings are speed down rules. #1 and #2 deal with
1809 * DUBIOUS errors.
1810 *
1811 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1812 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1813 *
1814 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1815 * occurred during last 5 mins, NCQ_OFF.
1816 *
1817 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Tejun Heo3884f7b2007-11-27 19:28:56 +09001818 * ocurred during last 5 mins, FALLBACK_TO_PIO
1819 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001820 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001821 * during last 10 mins, NCQ_OFF.
1822 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001823 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001824 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001825 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001826 * LOCKING:
1827 * Inherited from caller.
1828 *
1829 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001830 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001831 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001832static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001833{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001834 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1835 u64 j64 = get_jiffies_64();
1836 struct speed_down_verdict_arg arg;
1837 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001838
Tejun Heo3884f7b2007-11-27 19:28:56 +09001839 /* scan past 5 mins of error history */
1840 memset(&arg, 0, sizeof(arg));
1841 arg.since = j64 - min(j64, j5mins);
1842 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1843
Tejun Heo76326ac2007-11-27 19:28:59 +09001844 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1845 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1846 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1847 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1848
1849 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1850 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1851 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1852
Tejun Heo3884f7b2007-11-27 19:28:56 +09001853 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1854 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001855 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001856 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1857
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001858 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001859 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001860 arg.since = j64 - min(j64, j10mins);
1861 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001862
Tejun Heo3884f7b2007-11-27 19:28:56 +09001863 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1864 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001865 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001866
1867 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1868 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001869 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001870 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001871
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001872 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001873}
1874
1875/**
1876 * ata_eh_speed_down - record error and speed down if necessary
1877 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001878 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001879 * @err_mask: err_mask of the error
1880 *
1881 * Record error and examine error history to determine whether
1882 * adjusting transmission speed is necessary. It also sets
1883 * transmission limits appropriately if such adjustment is
1884 * necessary.
1885 *
1886 * LOCKING:
1887 * Kernel thread context (may sleep).
1888 *
1889 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001890 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001891 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001892static unsigned int ata_eh_speed_down(struct ata_device *dev,
1893 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001894{
Tejun Heob1c72912008-07-31 17:02:43 +09001895 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09001896 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001897 unsigned int verdict;
1898 unsigned int action = 0;
1899
1900 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001901 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001902 return 0;
1903
1904 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001905 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001906 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001907
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001908 /* turn off NCQ? */
1909 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1910 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1911 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1912 dev->flags |= ATA_DFLAG_NCQ_OFF;
1913 ata_dev_printk(dev, KERN_WARNING,
1914 "NCQ disabled due to excessive errors\n");
1915 goto done;
1916 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001917
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001918 /* speed down? */
1919 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1920 /* speed down SATA link speed if possible */
Tejun Heoa07d4992009-01-29 20:31:33 +09001921 if (sata_down_spd_limit(link, 0) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001922 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001923 goto done;
1924 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001925
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001926 /* lower transfer mode */
1927 if (dev->spdn_cnt < 2) {
1928 static const int dma_dnxfer_sel[] =
1929 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1930 static const int pio_dnxfer_sel[] =
1931 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1932 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001933
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001934 if (dev->xfer_shift != ATA_SHIFT_PIO)
1935 sel = dma_dnxfer_sel[dev->spdn_cnt];
1936 else
1937 sel = pio_dnxfer_sel[dev->spdn_cnt];
1938
1939 dev->spdn_cnt++;
1940
1941 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001942 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001943 goto done;
1944 }
1945 }
1946 }
1947
1948 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09001949 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001950 */
1951 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09001952 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001953 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1954 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1955 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09001956 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001957 goto done;
1958 }
1959 }
1960
Tejun Heo022bdb02006-05-15 20:58:22 +09001961 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001962 done:
1963 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09001964 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1965 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001966 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001967}
1968
1969/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001970 * ata_eh_link_autopsy - analyze error and determine recovery action
1971 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09001972 *
Tejun Heo02607312007-08-06 18:36:23 +09001973 * Analyze why @link failed and determine which recovery actions
1974 * are needed. This function also sets more detailed AC_ERR_*
1975 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09001976 *
1977 * LOCKING:
1978 * Kernel thread context (may sleep).
1979 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001980static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001981{
Tejun Heo02607312007-08-06 18:36:23 +09001982 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09001983 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09001984 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001985 unsigned int all_err_mask = 0, eflags = 0;
1986 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09001987 u32 serror;
1988 int rc;
1989
1990 DPRINTK("ENTER\n");
1991
Tejun Heo1cdaf532006-07-03 16:07:26 +09001992 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1993 return;
1994
Tejun Heo022bdb02006-05-15 20:58:22 +09001995 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09001996 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09001997 if (rc == 0) {
1998 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09001999 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09002000 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09002001 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09002002 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09002003 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09002004 ehc->i.err_mask |= AC_ERR_OTHER;
2005 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002006
Tejun Heoe8ee8452006-05-15 21:03:46 +09002007 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09002008 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09002009
Tejun Heo022bdb02006-05-15 20:58:22 +09002010 /* any real error trumps AC_ERR_OTHER */
2011 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2012 ehc->i.err_mask &= ~AC_ERR_OTHER;
2013
2014 all_err_mask |= ehc->i.err_mask;
2015
2016 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2017 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2018
Tejun Heob1c72912008-07-31 17:02:43 +09002019 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2020 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002021 continue;
2022
2023 /* inherit upper level err_mask */
2024 qc->err_mask |= ehc->i.err_mask;
2025
Tejun Heo022bdb02006-05-15 20:58:22 +09002026 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002027 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002028
2029 /* DEV errors are probably spurious in case of ATA_BUS error */
2030 if (qc->err_mask & AC_ERR_ATA_BUS)
2031 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2032 AC_ERR_INVALID);
2033
2034 /* any real error trumps unknown error */
2035 if (qc->err_mask & ~AC_ERR_OTHER)
2036 qc->err_mask &= ~AC_ERR_OTHER;
2037
2038 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09002039 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09002040 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09002041
Tejun Heo03faab72008-03-27 19:14:24 +09002042 /* determine whether the command is worth retrying */
Tejun Heo534ead72010-01-14 16:18:09 +09002043 if (qc->flags & ATA_QCFLAG_IO ||
2044 (!(qc->err_mask & AC_ERR_INVALID) &&
2045 qc->err_mask != AC_ERR_DEV))
Tejun Heo03faab72008-03-27 19:14:24 +09002046 qc->flags |= ATA_QCFLAG_RETRY;
2047
Tejun Heo022bdb02006-05-15 20:58:22 +09002048 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002049 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09002050 all_err_mask |= qc->err_mask;
2051 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09002052 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09002053 }
2054
Tejun Heoa20f33f2006-05-16 12:58:24 +09002055 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09002056 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09002057 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09002058 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002059 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2060 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09002061 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002062
Tejun Heodfcc1732007-10-31 10:17:05 +09002063 /* If we have offending qcs and the associated failed device,
2064 * perform per-dev EH action only on the offending device.
2065 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002066 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09002067 ehc->i.dev_action[ehc->i.dev->devno] |=
2068 ehc->i.action & ATA_EH_PERDEV_MASK;
2069 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09002070 }
2071
Tejun Heo2695e362008-01-10 13:41:23 +09002072 /* propagate timeout to host link */
2073 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2074 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2075
2076 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09002077 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09002078 if (!dev && ((ata_link_max_devices(link) == 1 &&
2079 ata_dev_enabled(link->device))))
2080 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09002081
Tejun Heo76326ac2007-11-27 19:28:59 +09002082 if (dev) {
2083 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2084 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002085 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09002086 }
Tejun Heodfcc1732007-10-31 10:17:05 +09002087
Tejun Heo022bdb02006-05-15 20:58:22 +09002088 DPRINTK("EXIT\n");
2089}
2090
2091/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002092 * ata_eh_autopsy - analyze error and determine recovery action
2093 * @ap: host port to perform autopsy on
2094 *
2095 * Analyze all links of @ap and determine why they failed and
2096 * which recovery actions are needed.
2097 *
2098 * LOCKING:
2099 * Kernel thread context (may sleep).
2100 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002101void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002102{
2103 struct ata_link *link;
2104
Tejun Heo1eca4362008-11-03 20:03:17 +09002105 ata_for_each_link(link, ap, EDGE)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002106 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002107
Tejun Heob1c72912008-07-31 17:02:43 +09002108 /* Handle the frigging slave link. Autopsy is done similarly
2109 * but actions and flags are transferred over to the master
2110 * link and handled from there.
2111 */
2112 if (ap->slave_link) {
2113 struct ata_eh_context *mehc = &ap->link.eh_context;
2114 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2115
Tejun Heo848e4c62008-10-21 14:26:39 +09002116 /* transfer control flags from master to slave */
2117 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2118
2119 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002120 ata_eh_link_autopsy(ap->slave_link);
2121
Tejun Heo848e4c62008-10-21 14:26:39 +09002122 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002123 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2124 mehc->i.action |= sehc->i.action;
2125 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2126 mehc->i.flags |= sehc->i.flags;
2127 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2128 }
2129
Tejun Heo2695e362008-01-10 13:41:23 +09002130 /* Autopsy of fanout ports can affect host link autopsy.
2131 * Perform host link autopsy last.
2132 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002133 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002134 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002135}
2136
2137/**
Robert Hancock65211482009-07-14 20:43:39 -06002138 * ata_get_cmd_descript - get description for ATA command
2139 * @command: ATA command code to get description for
2140 *
2141 * Return a textual description of the given command, or NULL if the
2142 * command is not known.
2143 *
2144 * LOCKING:
2145 * None
2146 */
2147const char *ata_get_cmd_descript(u8 command)
2148{
2149#ifdef CONFIG_ATA_VERBOSE_ERROR
2150 static const struct
2151 {
2152 u8 command;
2153 const char *text;
2154 } cmd_descr[] = {
2155 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2156 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2157 { ATA_CMD_STANDBY, "STANDBY" },
2158 { ATA_CMD_IDLE, "IDLE" },
2159 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2160 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2161 { ATA_CMD_NOP, "NOP" },
2162 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2163 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2164 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2165 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2166 { ATA_CMD_SERVICE, "SERVICE" },
2167 { ATA_CMD_READ, "READ DMA" },
2168 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2169 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2170 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2171 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2172 { ATA_CMD_WRITE, "WRITE DMA" },
2173 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2174 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2175 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2176 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2177 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2178 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2179 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2180 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2181 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2182 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2183 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2184 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2185 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2186 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2187 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2188 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2189 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2190 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2191 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2192 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2193 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2194 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2195 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2196 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2197 { ATA_CMD_SLEEP, "SLEEP" },
2198 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2199 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2200 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2201 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2202 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2203 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2204 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2205 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2206 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2207 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2208 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2209 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2210 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2211 { ATA_CMD_PMP_READ, "READ BUFFER" },
2212 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2213 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2214 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2215 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2216 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2217 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2218 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2219 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2220 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2221 { ATA_CMD_SMART, "SMART" },
2222 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2223 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
FUJITA Tomonoriacad762722010-07-05 15:45:20 +09002224 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
Robert Hancock65211482009-07-14 20:43:39 -06002225 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2226 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2227 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2228 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2229 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2230 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2231 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2232 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2233 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2234 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2235 { ATA_CMD_RESTORE, "RECALIBRATE" },
2236 { 0, NULL } /* terminate list */
2237 };
2238
2239 unsigned int i;
2240 for (i = 0; cmd_descr[i].text; i++)
2241 if (cmd_descr[i].command == command)
2242 return cmd_descr[i].text;
2243#endif
2244
2245 return NULL;
2246}
2247
2248/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002249 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002250 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002251 *
2252 * Report EH to user.
2253 *
2254 * LOCKING:
2255 * None.
2256 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002257static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002258{
Tejun Heo02607312007-08-06 18:36:23 +09002259 struct ata_port *ap = link->ap;
2260 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002261 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09002262 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09002263 int tag, nr_failed = 0;
2264
Tejun Heo94ff3d52007-10-09 14:57:56 +09002265 if (ehc->i.flags & ATA_EHI_QUIET)
2266 return;
2267
Tejun Heo022bdb02006-05-15 20:58:22 +09002268 desc = NULL;
2269 if (ehc->i.desc[0] != '\0')
2270 desc = ehc->i.desc;
2271
2272 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2273 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2274
Tejun Heob1c72912008-07-31 17:02:43 +09002275 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2276 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002277 ((qc->flags & ATA_QCFLAG_QUIET) &&
2278 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002279 continue;
2280 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2281 continue;
2282
2283 nr_failed++;
2284 }
2285
2286 if (!nr_failed && !ehc->i.err_mask)
2287 return;
2288
2289 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002290 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002291 frozen = " frozen";
2292
Tejun Heoa1e10f72007-08-18 13:28:49 +09002293 memset(tries_buf, 0, sizeof(tries_buf));
2294 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2295 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2296 ap->eh_tries);
2297
Tejun Heo022bdb02006-05-15 20:58:22 +09002298 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09002299 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002300 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2301 ehc->i.err_mask, link->sactive, ehc->i.serror,
2302 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002303 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09002304 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002305 } else {
Tejun Heo02607312007-08-06 18:36:23 +09002306 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002307 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2308 ehc->i.err_mask, link->sactive, ehc->i.serror,
2309 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002310 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09002311 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002312 }
2313
Robert Hancock65211482009-07-14 20:43:39 -06002314#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002315 if (ehc->i.serror)
Tejun Heoda0e21d2008-07-31 16:08:38 +09002316 ata_link_printk(link, KERN_ERR,
Robert Hancock1333e192007-10-02 11:22:02 -04002317 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2318 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2319 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2320 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2321 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2322 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2323 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2324 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2325 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2326 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2327 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2328 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2329 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2330 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2331 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2332 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2333 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002334 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002335#endif
Robert Hancock1333e192007-10-02 11:22:02 -04002336
Tejun Heo022bdb02006-05-15 20:58:22 +09002337 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2338 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002339 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002340 const u8 *cdb = qc->cdb;
2341 char data_buf[20] = "";
2342 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002343
Tejun Heo02607312007-08-06 18:36:23 +09002344 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002345 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002346 continue;
2347
Tejun Heoabb6a882007-11-28 23:16:09 +09002348 if (qc->dma_dir != DMA_NONE) {
2349 static const char *dma_str[] = {
2350 [DMA_BIDIRECTIONAL] = "bidi",
2351 [DMA_TO_DEVICE] = "out",
2352 [DMA_FROM_DEVICE] = "in",
2353 };
2354 static const char *prot_str[] = {
2355 [ATA_PROT_PIO] = "pio",
2356 [ATA_PROT_DMA] = "dma",
2357 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002358 [ATAPI_PROT_PIO] = "pio",
2359 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002360 };
2361
2362 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2363 prot_str[qc->tf.protocol], qc->nbytes,
2364 dma_str[qc->dma_dir]);
2365 }
2366
Robert Hancock65211482009-07-14 20:43:39 -06002367 if (ata_is_atapi(qc->tf.protocol)) {
2368 if (qc->scsicmd)
2369 scsi_print_command(qc->scsicmd);
2370 else
2371 snprintf(cdb_buf, sizeof(cdb_buf),
Tejun Heoabb6a882007-11-28 23:16:09 +09002372 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2373 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2374 cdb[0], cdb[1], cdb[2], cdb[3],
2375 cdb[4], cdb[5], cdb[6], cdb[7],
2376 cdb[8], cdb[9], cdb[10], cdb[11],
2377 cdb[12], cdb[13], cdb[14], cdb[15]);
Robert Hancock65211482009-07-14 20:43:39 -06002378 } else {
2379 const char *descr = ata_get_cmd_descript(cmd->command);
2380 if (descr)
2381 ata_dev_printk(qc->dev, KERN_ERR,
2382 "failed command: %s\n", descr);
2383 }
Tejun Heoabb6a882007-11-28 23:16:09 +09002384
Tejun Heo8a937582006-11-14 22:36:12 +09002385 ata_dev_printk(qc->dev, KERN_ERR,
2386 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002387 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002388 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002389 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002390 cmd->command, cmd->feature, cmd->nsect,
2391 cmd->lbal, cmd->lbam, cmd->lbah,
2392 cmd->hob_feature, cmd->hob_nsect,
2393 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002394 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002395 res->command, res->feature, res->nsect,
2396 res->lbal, res->lbam, res->lbah,
2397 res->hob_feature, res->hob_nsect,
2398 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002399 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2400 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002401
Robert Hancock65211482009-07-14 20:43:39 -06002402#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002403 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002404 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002405 if (res->command & ATA_BUSY)
2406 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002407 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002408 else
2409 ata_dev_printk(qc->dev, KERN_ERR,
2410 "status: { %s%s%s%s}\n",
2411 res->command & ATA_DRDY ? "DRDY " : "",
2412 res->command & ATA_DF ? "DF " : "",
2413 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002414 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002415 }
2416
2417 if (cmd->command != ATA_CMD_PACKET &&
2418 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2419 ATA_ABORTED)))
2420 ata_dev_printk(qc->dev, KERN_ERR,
2421 "error: { %s%s%s%s}\n",
2422 res->feature & ATA_ICRC ? "ICRC " : "",
2423 res->feature & ATA_UNC ? "UNC " : "",
2424 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002425 res->feature & ATA_ABORTED ? "ABRT " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002426#endif
Tejun Heo022bdb02006-05-15 20:58:22 +09002427 }
2428}
2429
Tejun Heo9b1e2652007-08-06 18:36:24 +09002430/**
2431 * ata_eh_report - report error handling to user
2432 * @ap: ATA port to report EH about
2433 *
2434 * Report EH to user.
2435 *
2436 * LOCKING:
2437 * None.
2438 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002439void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002440{
2441 struct ata_link *link;
2442
Tejun Heo1eca4362008-11-03 20:03:17 +09002443 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002444 ata_eh_link_report(link);
2445}
2446
Tejun Heocc0680a2007-08-06 18:36:23 +09002447static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002448 unsigned int *classes, unsigned long deadline,
2449 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002450{
Tejun Heof58229f2007-08-06 18:36:23 +09002451 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002452
Tejun Heob1c72912008-07-31 17:02:43 +09002453 if (clear_classes)
Tejun Heo1eca4362008-11-03 20:03:17 +09002454 ata_for_each_dev(dev, link, ALL)
Tejun Heob1c72912008-07-31 17:02:43 +09002455 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002456
Tejun Heof0465192008-05-19 01:15:08 +09002457 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002458}
2459
Tejun Heoae791c02007-09-23 13:14:12 +09002460static int ata_eh_followup_srst_needed(struct ata_link *link,
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002461 int rc, const unsigned int *classes)
Tejun Heo664faf02006-05-31 18:27:50 +09002462{
Tejun Heo45db2f62008-04-08 01:46:56 +09002463 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002464 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002465 if (rc == -EAGAIN)
2466 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002467 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002468 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002469 return 0;
2470}
2471
Tejun Heofb7fd612007-09-23 13:14:12 +09002472int ata_eh_reset(struct ata_link *link, int classify,
2473 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2474 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002475{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002476 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002477 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002478 struct ata_eh_context *ehc = &link->eh_context;
Bartlomiej Zolnierkiewicz705d2012009-07-26 16:21:01 +02002479 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
Tejun Heo664faf02006-05-31 18:27:50 +09002480 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002481 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002482 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002483 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002484 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002485 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002486 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002487 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002488 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002489 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002490 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002491
Tejun Heo932648b2008-05-19 01:15:06 +09002492 /*
2493 * Prepare to reset
2494 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002495 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2496 max_tries++;
Tejun Heo05944bd2008-08-13 20:19:09 +09002497 if (link->flags & ATA_LFLAG_NO_HRST)
2498 hardreset = NULL;
2499 if (link->flags & ATA_LFLAG_NO_SRST)
2500 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002501
Tejun Heo19b72322008-11-04 17:08:40 +09002502 /* make sure each reset attemp is at least COOL_DOWN apart */
2503 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2504 now = jiffies;
2505 WARN_ON(time_after(ehc->last_reset, now));
2506 deadline = ata_deadline(ehc->last_reset,
2507 ATA_EH_RESET_COOL_DOWN);
2508 if (time_before(now, deadline))
2509 schedule_timeout_uninterruptible(deadline - now);
2510 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002511
Tejun Heoafaa5c32007-10-09 15:06:10 +09002512 spin_lock_irqsave(ap->lock, flags);
2513 ap->pflags |= ATA_PFLAG_RESETTING;
2514 spin_unlock_irqrestore(ap->lock, flags);
2515
Tejun Heocf480622008-01-24 00:05:14 +09002516 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002517
Tejun Heo1eca4362008-11-03 20:03:17 +09002518 ata_for_each_dev(dev, link, ALL) {
Tejun Heocdeab112007-10-29 16:41:09 +09002519 /* If we issue an SRST then an ATA drive (not ATAPI)
2520 * may change configuration and be in PIO0 timing. If
2521 * we do a hard reset (or are coming from power on)
2522 * this is true for ATA or ATAPI. Until we've set a
2523 * suitable controller mode we should not touch the
2524 * bus as we may be talking too fast.
2525 */
2526 dev->pio_mode = XFER_PIO_0;
2527
2528 /* If the controller has a pio mode setup function
2529 * then use it to set the chipset to rights. Don't
2530 * touch the DMA setup as that will be dealt with when
2531 * configuring devices.
2532 */
2533 if (ap->ops->set_piomode)
2534 ap->ops->set_piomode(ap, dev);
2535 }
2536
Tejun Heocf480622008-01-24 00:05:14 +09002537 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002538 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002539 ehc->i.action &= ~ATA_EH_RESET;
2540 if (hardreset) {
2541 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002542 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002543 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002544 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002545 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002546 }
Tejun Heof5914a42006-05-31 18:27:48 +09002547
2548 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002549 unsigned long deadline = ata_deadline(jiffies,
2550 ATA_EH_PRERESET_TIMEOUT);
2551
2552 if (slave) {
2553 sehc->i.action &= ~ATA_EH_RESET;
2554 sehc->i.action |= ehc->i.action;
2555 }
2556
2557 rc = prereset(link, deadline);
2558
2559 /* If present, do prereset on slave link too. Reset
2560 * is skipped iff both master and slave links report
2561 * -ENOENT or clear ATA_EH_RESET.
2562 */
2563 if (slave && (rc == 0 || rc == -ENOENT)) {
2564 int tmp;
2565
2566 tmp = prereset(slave, deadline);
2567 if (tmp != -ENOENT)
2568 rc = tmp;
2569
2570 ehc->i.action |= sehc->i.action;
2571 }
2572
Tejun Heof5914a42006-05-31 18:27:48 +09002573 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002574 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002575 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002576 "port disabled. ignoring.\n");
Tejun Heocf480622008-01-24 00:05:14 +09002577 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002578
Tejun Heo1eca4362008-11-03 20:03:17 +09002579 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09002580 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002581
2582 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002583 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002584 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002585 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002586 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002587 }
Tejun Heof5914a42006-05-31 18:27:48 +09002588
Tejun Heo932648b2008-05-19 01:15:06 +09002589 /* prereset() might have cleared ATA_EH_RESET. If so,
Tejun Heod6515e62009-03-04 15:59:30 +09002590 * bang classes, thaw and return.
Tejun Heo932648b2008-05-19 01:15:06 +09002591 */
2592 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
Tejun Heo1eca4362008-11-03 20:03:17 +09002593 ata_for_each_dev(dev, link, ALL)
Tejun Heo932648b2008-05-19 01:15:06 +09002594 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heod6515e62009-03-04 15:59:30 +09002595 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2596 ata_is_host_link(link))
2597 ata_eh_thaw_port(ap);
Tejun Heo932648b2008-05-19 01:15:06 +09002598 rc = 0;
2599 goto out;
2600 }
Tejun Heof5914a42006-05-31 18:27:48 +09002601 }
2602
Tejun Heo022bdb02006-05-15 20:58:22 +09002603 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002604 /*
2605 * Perform reset
2606 */
Tejun Heodc98c322008-05-19 01:15:07 +09002607 if (ata_is_host_link(link))
2608 ata_eh_freeze_port(ap);
2609
Tejun Heo341c2c92008-05-20 02:17:51 +09002610 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002611
Tejun Heo932648b2008-05-19 01:15:06 +09002612 if (reset) {
2613 if (verbose)
2614 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2615 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002616
Tejun Heo932648b2008-05-19 01:15:06 +09002617 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002618 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002619 if (reset == hardreset)
2620 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2621 else
2622 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002623
Tejun Heob1c72912008-07-31 17:02:43 +09002624 rc = ata_do_reset(link, reset, classes, deadline, true);
2625 if (rc && rc != -EAGAIN) {
2626 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002627 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002628 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002629
Tejun Heob1c72912008-07-31 17:02:43 +09002630 /* hardreset slave link if existent */
2631 if (slave && reset == hardreset) {
2632 int tmp;
2633
2634 if (verbose)
2635 ata_link_printk(slave, KERN_INFO,
2636 "hard resetting link\n");
2637
2638 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2639 tmp = ata_do_reset(slave, reset, classes, deadline,
2640 false);
2641 switch (tmp) {
2642 case -EAGAIN:
2643 rc = -EAGAIN;
2644 case 0:
2645 break;
2646 default:
2647 failed_link = slave;
2648 rc = tmp;
2649 goto fail;
2650 }
2651 }
2652
2653 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002654 if (reset == hardreset &&
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002655 ata_eh_followup_srst_needed(link, rc, classes)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002656 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002657
Tejun Heo932648b2008-05-19 01:15:06 +09002658 if (!reset) {
2659 ata_link_printk(link, KERN_ERR,
2660 "follow-up softreset required "
2661 "but no softreset avaliable\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002662 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002663 rc = -EINVAL;
2664 goto fail;
2665 }
2666
2667 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002668 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heofe2c4d02009-07-08 12:16:37 +09002669 if (rc) {
2670 failed_link = link;
2671 goto fail;
2672 }
Tejun Heo664faf02006-05-31 18:27:50 +09002673 }
Tejun Heo932648b2008-05-19 01:15:06 +09002674 } else {
2675 if (verbose)
2676 ata_link_printk(link, KERN_INFO, "no reset method "
2677 "available, skipping reset\n");
2678 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2679 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002680 }
2681
Tejun Heo932648b2008-05-19 01:15:06 +09002682 /*
2683 * Post-reset processing
2684 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002685 ata_for_each_dev(dev, link, ALL) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002686 /* After the reset, the device state is PIO 0 and the
2687 * controller state is undefined. Reset also wakes up
2688 * drives from sleeping mode.
2689 */
2690 dev->pio_mode = XFER_PIO_0;
2691 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002692
Tejun Heo3b761d32009-10-06 17:08:40 +09002693 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2694 continue;
2695
2696 /* apply class override */
2697 if (lflags & ATA_LFLAG_ASSUME_ATA)
2698 classes[dev->devno] = ATA_DEV_ATA;
2699 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2700 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
Tejun Heo022bdb02006-05-15 20:58:22 +09002701 }
2702
Tejun Heo416dc9e2007-10-31 10:17:03 +09002703 /* record current link speed */
2704 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2705 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002706 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2707 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002708
Tejun Heodc98c322008-05-19 01:15:07 +09002709 /* thaw the port */
2710 if (ata_is_host_link(link))
2711 ata_eh_thaw_port(ap);
2712
Tejun Heof0465192008-05-19 01:15:08 +09002713 /* postreset() should clear hardware SError. Although SError
2714 * is cleared during link resume, clearing SError here is
2715 * necessary as some PHYs raise hotplug events after SRST.
2716 * This introduces race condition where hotplug occurs between
2717 * reset and here. This race is mediated by cross checking
2718 * link onlineness and classification result later.
2719 */
Tejun Heob1c72912008-07-31 17:02:43 +09002720 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002721 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002722 if (slave)
2723 postreset(slave, classes);
2724 }
Tejun Heo20952b62006-05-31 18:27:23 +09002725
Tejun Heo1e641062009-07-17 11:58:33 +09002726 /*
2727 * Some controllers can't be frozen very well and may set
2728 * spuruious error conditions during reset. Clear accumulated
2729 * error information. As reset is the final recovery action,
2730 * nothing is lost by doing this.
2731 */
Tejun Heof0465192008-05-19 01:15:08 +09002732 spin_lock_irqsave(link->ap->lock, flags);
Tejun Heo1e641062009-07-17 11:58:33 +09002733 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heob1c72912008-07-31 17:02:43 +09002734 if (slave)
Tejun Heo1e641062009-07-17 11:58:33 +09002735 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2736 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof0465192008-05-19 01:15:08 +09002737 spin_unlock_irqrestore(link->ap->lock, flags);
2738
Tejun Heo3b761d32009-10-06 17:08:40 +09002739 /*
2740 * Make sure onlineness and classification result correspond.
Tejun Heof0465192008-05-19 01:15:08 +09002741 * Hotplug could have happened during reset and some
2742 * controllers fail to wait while a drive is spinning up after
2743 * being hotplugged causing misdetection. By cross checking
Tejun Heo3b761d32009-10-06 17:08:40 +09002744 * link on/offlineness and classification result, those
2745 * conditions can be reliably detected and retried.
Tejun Heof0465192008-05-19 01:15:08 +09002746 */
Tejun Heob1c72912008-07-31 17:02:43 +09002747 nr_unknown = 0;
Tejun Heo1eca4362008-11-03 20:03:17 +09002748 ata_for_each_dev(dev, link, ALL) {
Tejun Heo3b761d32009-10-06 17:08:40 +09002749 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2750 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2751 ata_dev_printk(dev, KERN_DEBUG, "link online "
2752 "but device misclassifed\n");
2753 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002754 nr_unknown++;
Tejun Heo3b761d32009-10-06 17:08:40 +09002755 }
2756 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2757 if (ata_class_enabled(classes[dev->devno]))
2758 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2759 "clearing class %d to NONE\n",
2760 classes[dev->devno]);
2761 classes[dev->devno] = ATA_DEV_NONE;
2762 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2763 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2764 "clearing UNKNOWN to NONE\n");
2765 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002766 }
Tejun Heof0465192008-05-19 01:15:08 +09002767 }
2768
Tejun Heob1c72912008-07-31 17:02:43 +09002769 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002770 if (try < max_tries) {
2771 ata_link_printk(link, KERN_WARNING, "link online but "
Tejun Heo3b761d32009-10-06 17:08:40 +09002772 "%d devices misclassified, retrying\n",
2773 nr_unknown);
Tejun Heob1c72912008-07-31 17:02:43 +09002774 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002775 rc = -EAGAIN;
2776 goto fail;
2777 }
2778 ata_link_printk(link, KERN_WARNING,
Tejun Heo3b761d32009-10-06 17:08:40 +09002779 "link online but %d devices misclassified, "
2780 "device detection might fail\n", nr_unknown);
Tejun Heof0465192008-05-19 01:15:08 +09002781 }
2782
Tejun Heo416dc9e2007-10-31 10:17:03 +09002783 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002784 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002785 if (slave)
2786 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo19b72322008-11-04 17:08:40 +09002787 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002788 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heoae791c02007-09-23 13:14:12 +09002789
Tejun Heo416dc9e2007-10-31 10:17:03 +09002790 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002791 out:
2792 /* clear hotplug flag */
2793 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002794 if (slave)
2795 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002796
2797 spin_lock_irqsave(ap->lock, flags);
2798 ap->pflags &= ~ATA_PFLAG_RESETTING;
2799 spin_unlock_irqrestore(ap->lock, flags);
2800
Tejun Heo022bdb02006-05-15 20:58:22 +09002801 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002802
2803 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002804 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2805 if (!ata_is_host_link(link) &&
2806 sata_scr_read(link, SCR_STATUS, &sstatus))
2807 rc = -ERESTART;
2808
Tejun Heo416dc9e2007-10-31 10:17:03 +09002809 if (rc == -ERESTART || try >= max_tries)
2810 goto out;
2811
2812 now = jiffies;
2813 if (time_before(now, deadline)) {
2814 unsigned long delta = deadline - now;
2815
Tejun Heob1c72912008-07-31 17:02:43 +09002816 ata_link_printk(failed_link, KERN_WARNING,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002817 "reset failed (errno=%d), retrying in %u secs\n",
2818 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002819
2820 while (delta)
2821 delta = schedule_timeout_uninterruptible(delta);
2822 }
2823
Tejun Heob1c72912008-07-31 17:02:43 +09002824 if (try == max_tries - 1) {
Tejun Heoa07d4992009-01-29 20:31:33 +09002825 sata_down_spd_limit(link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002826 if (slave)
Tejun Heoa07d4992009-01-29 20:31:33 +09002827 sata_down_spd_limit(slave, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002828 } else if (rc == -EPIPE)
Tejun Heoa07d4992009-01-29 20:31:33 +09002829 sata_down_spd_limit(failed_link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002830
Tejun Heo416dc9e2007-10-31 10:17:03 +09002831 if (hardreset)
2832 reset = hardreset;
2833 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002834}
2835
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002836static inline void ata_eh_pull_park_action(struct ata_port *ap)
2837{
2838 struct ata_link *link;
2839 struct ata_device *dev;
2840 unsigned long flags;
2841
2842 /*
2843 * This function can be thought of as an extended version of
2844 * ata_eh_about_to_do() specially crafted to accommodate the
2845 * requirements of ATA_EH_PARK handling. Since the EH thread
2846 * does not leave the do {} while () loop in ata_eh_recover as
2847 * long as the timeout for a park request to *one* device on
2848 * the port has not expired, and since we still want to pick
2849 * up park requests to other devices on the same port or
2850 * timeout updates for the same device, we have to pull
2851 * ATA_EH_PARK actions from eh_info into eh_context.i
2852 * ourselves at the beginning of each pass over the loop.
2853 *
2854 * Additionally, all write accesses to &ap->park_req_pending
2855 * through INIT_COMPLETION() (see below) or complete_all()
2856 * (see ata_scsi_park_store()) are protected by the host lock.
2857 * As a result we have that park_req_pending.done is zero on
2858 * exit from this function, i.e. when ATA_EH_PARK actions for
2859 * *all* devices on port ap have been pulled into the
2860 * respective eh_context structs. If, and only if,
2861 * park_req_pending.done is non-zero by the time we reach
2862 * wait_for_completion_timeout(), another ATA_EH_PARK action
2863 * has been scheduled for at least one of the devices on port
2864 * ap and we have to cycle over the do {} while () loop in
2865 * ata_eh_recover() again.
2866 */
2867
2868 spin_lock_irqsave(ap->lock, flags);
2869 INIT_COMPLETION(ap->park_req_pending);
Tejun Heo1eca4362008-11-03 20:03:17 +09002870 ata_for_each_link(link, ap, EDGE) {
2871 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002872 struct ata_eh_info *ehi = &link->eh_info;
2873
2874 link->eh_context.i.dev_action[dev->devno] |=
2875 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2876 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2877 }
2878 }
2879 spin_unlock_irqrestore(ap->lock, flags);
2880}
2881
2882static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2883{
2884 struct ata_eh_context *ehc = &dev->link->eh_context;
2885 struct ata_taskfile tf;
2886 unsigned int err_mask;
2887
2888 ata_tf_init(dev, &tf);
2889 if (park) {
2890 ehc->unloaded_mask |= 1 << dev->devno;
2891 tf.command = ATA_CMD_IDLEIMMEDIATE;
2892 tf.feature = 0x44;
2893 tf.lbal = 0x4c;
2894 tf.lbam = 0x4e;
2895 tf.lbah = 0x55;
2896 } else {
2897 ehc->unloaded_mask &= ~(1 << dev->devno);
2898 tf.command = ATA_CMD_CHK_POWER;
2899 }
2900
2901 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2902 tf.protocol |= ATA_PROT_NODATA;
2903 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2904 if (park && (err_mask || tf.lbal != 0xc4)) {
2905 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2906 ehc->unloaded_mask &= ~(1 << dev->devno);
2907 }
2908}
2909
Tejun Heo02607312007-08-06 18:36:23 +09002910static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002911 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002912{
Tejun Heo02607312007-08-06 18:36:23 +09002913 struct ata_port *ap = link->ap;
2914 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002915 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002916 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002917 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09002918 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002919
2920 DPRINTK("ENTER\n");
2921
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002922 /* For PATA drive side cable detection to work, IDENTIFY must
2923 * be done backwards such that PDIAG- is released by the slave
2924 * device before the master device is identified.
2925 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002926 ata_for_each_dev(dev, link, ALL_REVERSE) {
Tejun Heof58229f2007-08-06 18:36:23 +09002927 unsigned int action = ata_eh_dev_action(dev);
2928 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09002929
Tejun Heobff04642006-11-10 18:08:10 +09002930 if (ehc->i.flags & ATA_EHI_DID_RESET)
2931 readid_flags |= ATA_READID_POSTRESET;
2932
Tejun Heo9666f402007-05-04 21:27:47 +02002933 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a32007-09-23 13:19:54 +09002934 WARN_ON(dev->class == ATA_DEV_PMP);
2935
Tejun Heob1c72912008-07-31 17:02:43 +09002936 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002937 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002938 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002939 }
2940
Tejun Heo02607312007-08-06 18:36:23 +09002941 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09002942 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2943 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09002944 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002945 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002946
Tejun Heo02607312007-08-06 18:36:23 +09002947 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09002948
Tejun Heobaa1e782006-11-01 18:39:27 +09002949 /* Configuration may have changed, reconfigure
2950 * transfer mode.
2951 */
2952 ehc->i.flags |= ATA_EHI_SETMODE;
2953
zhao, forrest3057ac32006-06-12 12:01:34 +08002954 /* schedule the scsi_rescan_device() here */
Tejun Heoad72cf92010-07-02 10:03:52 +02002955 schedule_work(&(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09002956 } else if (dev->class == ATA_DEV_UNKNOWN &&
2957 ehc->tries[dev->devno] &&
2958 ata_class_enabled(ehc->classes[dev->devno])) {
Tejun Heo842faa62009-05-10 01:06:54 +09002959 /* Temporarily set dev->class, it will be
2960 * permanently set once all configurations are
2961 * complete. This is necessary because new
2962 * device configuration is done in two
2963 * separate loops.
2964 */
Tejun Heo084fe632006-05-31 18:28:03 +09002965 dev->class = ehc->classes[dev->devno];
2966
Tejun Heo633273a32007-09-23 13:19:54 +09002967 if (dev->class == ATA_DEV_PMP)
2968 rc = sata_pmp_attach(dev);
2969 else
2970 rc = ata_dev_read_id(dev, &dev->class,
2971 readid_flags, dev->id);
Tejun Heo842faa62009-05-10 01:06:54 +09002972
2973 /* read_id might have changed class, store and reset */
2974 ehc->classes[dev->devno] = dev->class;
2975 dev->class = ATA_DEV_UNKNOWN;
2976
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002977 switch (rc) {
2978 case 0:
Tejun Heo99cf6102009-01-29 20:31:32 +09002979 /* clear error info accumulated during probe */
2980 ata_ering_clear(&dev->ering);
Tejun Heof58229f2007-08-06 18:36:23 +09002981 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002982 break;
2983 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002984 /* IDENTIFY was issued to non-existent
2985 * device. No need to reset. Just
Tejun Heo842faa62009-05-10 01:06:54 +09002986 * thaw and ignore the device.
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002987 */
2988 ata_eh_thaw_port(ap);
Tejun Heo084fe632006-05-31 18:28:03 +09002989 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002990 default:
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002991 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002992 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002993 }
2994 }
2995
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002996 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09002997 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2998 if (ap->ops->cable_detect)
2999 ap->cbl = ap->ops->cable_detect(ap);
3000 ata_force_cbl(ap);
3001 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003002
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003003 /* Configure new devices forward such that user doesn't see
3004 * device detection messages backwards.
3005 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003006 ata_for_each_dev(dev, link, ALL) {
Tejun Heo4f7c2872009-10-15 23:37:32 +09003007 if (!(new_mask & (1 << dev->devno)))
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003008 continue;
3009
Tejun Heo842faa62009-05-10 01:06:54 +09003010 dev->class = ehc->classes[dev->devno];
3011
Tejun Heo4f7c2872009-10-15 23:37:32 +09003012 if (dev->class == ATA_DEV_PMP)
3013 continue;
3014
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003015 ehc->i.flags |= ATA_EHI_PRINTINFO;
3016 rc = ata_dev_configure(dev);
3017 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo842faa62009-05-10 01:06:54 +09003018 if (rc) {
3019 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003020 goto err;
Tejun Heo842faa62009-05-10 01:06:54 +09003021 }
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003022
3023 spin_lock_irqsave(ap->lock, flags);
3024 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3025 spin_unlock_irqrestore(ap->lock, flags);
3026
3027 /* new device discovered, configure xfermode */
3028 ehc->i.flags |= ATA_EHI_SETMODE;
3029 }
3030
3031 return 0;
3032
3033 err:
3034 *r_failed_dev = dev;
3035 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003036 return rc;
3037}
3038
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003039/**
3040 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3041 * @link: link on which timings will be programmed
Martin Olsson98a17082009-04-22 18:21:29 +02003042 * @r_failed_dev: out parameter for failed device
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003043 *
3044 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3045 * ata_set_mode() fails, pointer to the failing device is
3046 * returned in @r_failed_dev.
3047 *
3048 * LOCKING:
3049 * PCI/etc. bus probe sem.
3050 *
3051 * RETURNS:
3052 * 0 on success, negative errno otherwise
3053 */
3054int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3055{
3056 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09003057 struct ata_device *dev;
3058 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003059
Tejun Heo76326ac2007-11-27 19:28:59 +09003060 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
Tejun Heo1eca4362008-11-03 20:03:17 +09003061 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo76326ac2007-11-27 19:28:59 +09003062 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3063 struct ata_ering_entry *ent;
3064
3065 ent = ata_ering_top(&dev->ering);
3066 if (ent)
3067 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3068 }
3069 }
3070
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003071 /* has private set_mode? */
3072 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09003073 rc = ap->ops->set_mode(link, r_failed_dev);
3074 else
3075 rc = ata_do_set_mode(link, r_failed_dev);
3076
3077 /* if transfer mode has changed, set DUBIOUS_XFER on device */
Tejun Heo1eca4362008-11-03 20:03:17 +09003078 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +09003079 struct ata_eh_context *ehc = &link->eh_context;
3080 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3081 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3082
3083 if (dev->xfer_mode != saved_xfer_mode ||
3084 ata_ncq_enabled(dev) != saved_ncq)
3085 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3086 }
3087
3088 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003089}
3090
Tejun Heo11fc33d2008-08-30 14:20:01 +02003091/**
3092 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3093 * @dev: ATAPI device to clear UA for
3094 *
3095 * Resets and other operations can make an ATAPI device raise
3096 * UNIT ATTENTION which causes the next operation to fail. This
3097 * function clears UA.
3098 *
3099 * LOCKING:
3100 * EH context (may sleep).
3101 *
3102 * RETURNS:
3103 * 0 on success, -errno on failure.
3104 */
3105static int atapi_eh_clear_ua(struct ata_device *dev)
3106{
3107 int i;
3108
3109 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
Tejun Heob5357082009-03-02 18:55:16 +09003110 u8 *sense_buffer = dev->link->ap->sector_buf;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003111 u8 sense_key = 0;
3112 unsigned int err_mask;
3113
3114 err_mask = atapi_eh_tur(dev, &sense_key);
3115 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3116 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3117 "failed (err_mask=0x%x)\n", err_mask);
3118 return -EIO;
3119 }
3120
3121 if (!err_mask || sense_key != UNIT_ATTENTION)
3122 return 0;
3123
3124 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3125 if (err_mask) {
3126 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3127 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3128 return -EIO;
3129 }
3130 }
3131
3132 ata_dev_printk(dev, KERN_WARNING,
3133 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3134
3135 return 0;
3136}
3137
Tejun Heo6013efd2009-11-19 15:36:45 +09003138/**
3139 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3140 * @dev: ATA device which may need FLUSH retry
3141 *
3142 * If @dev failed FLUSH, it needs to be reported upper layer
3143 * immediately as it means that @dev failed to remap and already
3144 * lost at least a sector and further FLUSH retrials won't make
3145 * any difference to the lost sector. However, if FLUSH failed
3146 * for other reasons, for example transmission error, FLUSH needs
3147 * to be retried.
3148 *
3149 * This function determines whether FLUSH failure retry is
3150 * necessary and performs it if so.
3151 *
3152 * RETURNS:
3153 * 0 if EH can continue, -errno if EH needs to be repeated.
3154 */
3155static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3156{
3157 struct ata_link *link = dev->link;
3158 struct ata_port *ap = link->ap;
3159 struct ata_queued_cmd *qc;
3160 struct ata_taskfile tf;
3161 unsigned int err_mask;
3162 int rc = 0;
3163
3164 /* did flush fail for this device? */
3165 if (!ata_tag_valid(link->active_tag))
3166 return 0;
3167
3168 qc = __ata_qc_from_tag(ap, link->active_tag);
3169 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3170 qc->tf.command != ATA_CMD_FLUSH))
3171 return 0;
3172
3173 /* if the device failed it, it should be reported to upper layers */
3174 if (qc->err_mask & AC_ERR_DEV)
3175 return 0;
3176
3177 /* flush failed for some other reason, give it another shot */
3178 ata_tf_init(dev, &tf);
3179
3180 tf.command = qc->tf.command;
3181 tf.flags |= ATA_TFLAG_DEVICE;
3182 tf.protocol = ATA_PROT_NODATA;
3183
3184 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3185 tf.command, qc->err_mask);
3186
3187 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3188 if (!err_mask) {
3189 /*
3190 * FLUSH is complete but there's no way to
3191 * successfully complete a failed command from EH.
3192 * Making sure retry is allowed at least once and
3193 * retrying it should do the trick - whatever was in
3194 * the cache is already on the platter and this won't
3195 * cause infinite loop.
3196 */
3197 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3198 } else {
3199 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3200 err_mask);
3201 rc = -EIO;
3202
3203 /* if device failed it, report it to upper layers */
3204 if (err_mask & AC_ERR_DEV) {
3205 qc->err_mask |= AC_ERR_DEV;
3206 qc->result_tf = tf;
3207 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3208 rc = 0;
3209 }
3210 }
3211 return rc;
3212}
3213
Tejun Heo02607312007-08-06 18:36:23 +09003214static int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003215{
Tejun Heof58229f2007-08-06 18:36:23 +09003216 struct ata_device *dev;
3217 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003218
Tejun Heo1eca4362008-11-03 20:03:17 +09003219 ata_for_each_dev(dev, link, ENABLED)
3220 cnt++;
Tejun Heo022bdb02006-05-15 20:58:22 +09003221 return cnt;
3222}
3223
Tejun Heo02607312007-08-06 18:36:23 +09003224static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003225{
Tejun Heof58229f2007-08-06 18:36:23 +09003226 struct ata_device *dev;
3227 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003228
Tejun Heo1eca4362008-11-03 20:03:17 +09003229 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09003230 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09003231 cnt++;
3232 return cnt;
3233}
3234
Tejun Heo02607312007-08-06 18:36:23 +09003235static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003236{
Tejun Heo672b2d62008-01-24 00:05:14 +09003237 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09003238 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09003239 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09003240
Tejun Heof9df58c2007-09-23 13:14:13 +09003241 /* skip disabled links */
3242 if (link->flags & ATA_LFLAG_DISABLED)
3243 return 1;
3244
Tejun Heoe2f3d752010-09-07 14:05:31 +02003245 /* skip if explicitly requested */
3246 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3247 return 1;
3248
Tejun Heo672b2d62008-01-24 00:05:14 +09003249 /* thaw frozen port and recover failed devices */
3250 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3251 return 0;
3252
3253 /* reset at least once if reset is requested */
3254 if ((ehc->i.action & ATA_EH_RESET) &&
3255 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09003256 return 0;
3257
3258 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo1eca4362008-11-03 20:03:17 +09003259 ata_for_each_dev(dev, link, ALL) {
Tejun Heo084fe632006-05-31 18:28:03 +09003260 if (dev->class == ATA_DEV_UNKNOWN &&
3261 ehc->classes[dev->devno] != ATA_DEV_NONE)
3262 return 0;
3263 }
3264
3265 return 1;
3266}
3267
Tejun Heoc2c7a892009-01-29 20:31:34 +09003268static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3269{
3270 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3271 u64 now = get_jiffies_64();
3272 int *trials = void_arg;
3273
3274 if (ent->timestamp < now - min(now, interval))
3275 return -1;
3276
3277 (*trials)++;
3278 return 0;
3279}
3280
Tejun Heo02c05a22007-11-27 19:28:54 +09003281static int ata_eh_schedule_probe(struct ata_device *dev)
3282{
3283 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heoc2c7a892009-01-29 20:31:34 +09003284 struct ata_link *link = ata_dev_phys_link(dev);
3285 int trials = 0;
Tejun Heo02c05a22007-11-27 19:28:54 +09003286
3287 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3288 (ehc->did_probe_mask & (1 << dev->devno)))
3289 return 0;
3290
3291 ata_eh_detach_dev(dev);
3292 ata_dev_init(dev);
3293 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09003294 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09003295 ehc->saved_xfer_mode[dev->devno] = 0;
3296 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09003297
Tejun Heoc2c7a892009-01-29 20:31:34 +09003298 /* Record and count probe trials on the ering. The specific
3299 * error mask used is irrelevant. Because a successful device
3300 * detection clears the ering, this count accumulates only if
3301 * there are consecutive failed probes.
3302 *
3303 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3304 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3305 * forced to 1.5Gbps.
3306 *
3307 * This is to work around cases where failed link speed
3308 * negotiation results in device misdetection leading to
3309 * infinite DEVXCHG or PHRDY CHG events.
3310 */
3311 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3312 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3313
3314 if (trials > ATA_EH_PROBE_TRIALS)
3315 sata_down_spd_limit(link, 1);
3316
Tejun Heo02c05a22007-11-27 19:28:54 +09003317 return 1;
3318}
3319
Tejun Heo9b1e2652007-08-06 18:36:24 +09003320static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09003321{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09003322 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09003323
Tejun Heocf9a5902009-01-29 20:31:35 +09003324 /* -EAGAIN from EH routine indicates retry without prejudice.
3325 * The requester is responsible for ensuring forward progress.
3326 */
3327 if (err != -EAGAIN)
3328 ehc->tries[dev->devno]--;
Tejun Heofee7ca72007-07-01 19:05:58 +09003329
3330 switch (err) {
3331 case -ENODEV:
3332 /* device missing or wrong IDENTIFY data, schedule probing */
3333 ehc->i.probe_mask |= (1 << dev->devno);
3334 case -EINVAL:
3335 /* give it just one more chance */
3336 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3337 case -EIO:
Tejun Heod89293a2009-01-29 20:31:29 +09003338 if (ehc->tries[dev->devno] == 1) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003339 /* This is the last chance, better to slow
3340 * down than lose it.
3341 */
Tejun Heoa07d4992009-01-29 20:31:33 +09003342 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
Tejun Heod89293a2009-01-29 20:31:29 +09003343 if (dev->pio_mode > XFER_PIO_0)
3344 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
Tejun Heofee7ca72007-07-01 19:05:58 +09003345 }
3346 }
3347
3348 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3349 /* disable device if it has used up all its chances */
3350 ata_dev_disable(dev);
3351
3352 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09003353 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09003354 ata_eh_detach_dev(dev);
3355
Tejun Heo02c05a22007-11-27 19:28:54 +09003356 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003357 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003358 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003359 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3360 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3361 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09003362
3363 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09003364 } else {
Tejun Heocf480622008-01-24 00:05:14 +09003365 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003366 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09003367 }
3368}
3369
Tejun Heo022bdb02006-05-15 20:58:22 +09003370/**
3371 * ata_eh_recover - recover host port after error
3372 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09003373 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003374 * @softreset: softreset method (can be NULL)
3375 * @hardreset: hardreset method (can be NULL)
3376 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003377 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003378 *
3379 * This is the alpha and omega, eum and yang, heart and soul of
3380 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003381 * recover each link and hotplug requests are recorded in the
3382 * link's eh_context. This function executes all the operations
3383 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003384 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003385 *
3386 * LOCKING:
3387 * Kernel thread context (may sleep).
3388 *
3389 * RETURNS:
3390 * 0 on success, -errno on failure.
3391 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003392int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3393 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3394 ata_postreset_fn_t postreset,
3395 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003396{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003397 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003398 struct ata_device *dev;
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003399 int nr_failed_devs;
Tejun Heodc98c322008-05-19 01:15:07 +09003400 int rc;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003401 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003402
3403 DPRINTK("ENTER\n");
3404
3405 /* prep for recovery */
Tejun Heo1eca4362008-11-03 20:03:17 +09003406 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003407 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003408
Tejun Heof9df58c2007-09-23 13:14:13 +09003409 /* re-enable link? */
3410 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3411 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3412 spin_lock_irqsave(ap->lock, flags);
3413 link->flags &= ~ATA_LFLAG_DISABLED;
3414 spin_unlock_irqrestore(ap->lock, flags);
3415 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3416 }
3417
Tejun Heo1eca4362008-11-03 20:03:17 +09003418 ata_for_each_dev(dev, link, ALL) {
Tejun Heofd995f72007-09-23 13:14:12 +09003419 if (link->flags & ATA_LFLAG_NO_RETRY)
3420 ehc->tries[dev->devno] = 1;
3421 else
3422 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003423
Tejun Heo9b1e2652007-08-06 18:36:24 +09003424 /* collect port action mask recorded in dev actions */
3425 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3426 ~ATA_EH_PERDEV_MASK;
3427 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003428
Tejun Heo9b1e2652007-08-06 18:36:24 +09003429 /* process hotplug request */
3430 if (dev->flags & ATA_DFLAG_DETACH)
3431 ata_eh_detach_dev(dev);
3432
Tejun Heo02c05a22007-11-27 19:28:54 +09003433 /* schedule probe if necessary */
3434 if (!ata_dev_enabled(dev))
3435 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003436 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003437 }
3438
3439 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003440 rc = 0;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003441 nr_failed_devs = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003442
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003443 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003444 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003445 goto out;
3446
Tejun Heo9b1e2652007-08-06 18:36:24 +09003447 /* prep for EH */
Tejun Heo1eca4362008-11-03 20:03:17 +09003448 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003449 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003450
Tejun Heo9b1e2652007-08-06 18:36:24 +09003451 /* skip EH if possible. */
3452 if (ata_eh_skip_recovery(link))
3453 ehc->i.action = 0;
3454
Tejun Heo1eca4362008-11-03 20:03:17 +09003455 ata_for_each_dev(dev, link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003456 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3457 }
Tejun Heo084fe632006-05-31 18:28:03 +09003458
Tejun Heo022bdb02006-05-15 20:58:22 +09003459 /* reset */
Tejun Heo1eca4362008-11-03 20:03:17 +09003460 ata_for_each_link(link, ap, EDGE) {
Tejun Heodc98c322008-05-19 01:15:07 +09003461 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003462
Tejun Heodc98c322008-05-19 01:15:07 +09003463 if (!(ehc->i.action & ATA_EH_RESET))
3464 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003465
Tejun Heodc98c322008-05-19 01:15:07 +09003466 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3467 prereset, softreset, hardreset, postreset);
3468 if (rc) {
3469 ata_link_printk(link, KERN_ERR,
3470 "reset failed, giving up\n");
3471 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003472 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003473 }
3474
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003475 do {
3476 unsigned long now;
3477
3478 /*
3479 * clears ATA_EH_PARK in eh_info and resets
3480 * ap->park_req_pending
3481 */
3482 ata_eh_pull_park_action(ap);
3483
3484 deadline = jiffies;
Tejun Heo1eca4362008-11-03 20:03:17 +09003485 ata_for_each_link(link, ap, EDGE) {
3486 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003487 struct ata_eh_context *ehc = &link->eh_context;
3488 unsigned long tmp;
3489
3490 if (dev->class != ATA_DEV_ATA)
3491 continue;
3492 if (!(ehc->i.dev_action[dev->devno] &
3493 ATA_EH_PARK))
3494 continue;
3495 tmp = dev->unpark_deadline;
3496 if (time_before(deadline, tmp))
3497 deadline = tmp;
3498 else if (time_before_eq(tmp, jiffies))
3499 continue;
3500 if (ehc->unloaded_mask & (1 << dev->devno))
3501 continue;
3502
3503 ata_eh_park_issue_cmd(dev, 1);
3504 }
3505 }
3506
3507 now = jiffies;
3508 if (time_before_eq(deadline, now))
3509 break;
3510
3511 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3512 deadline - now);
3513 } while (deadline);
Tejun Heo1eca4362008-11-03 20:03:17 +09003514 ata_for_each_link(link, ap, EDGE) {
3515 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003516 if (!(link->eh_context.unloaded_mask &
3517 (1 << dev->devno)))
3518 continue;
3519
3520 ata_eh_park_issue_cmd(dev, 0);
3521 ata_eh_done(link, dev, ATA_EH_PARK);
3522 }
3523 }
3524
Tejun Heo9b1e2652007-08-06 18:36:24 +09003525 /* the rest */
Tejun Heo1eca4362008-11-03 20:03:17 +09003526 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003527 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003528
Tejun Heo9b1e2652007-08-06 18:36:24 +09003529 /* revalidate existing devices and attach new ones */
3530 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003531 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09003532 goto dev_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003533
Tejun Heo633273a32007-09-23 13:19:54 +09003534 /* if PMP got attached, return, pmp EH will take care of it */
3535 if (link->device->class == ATA_DEV_PMP) {
3536 ehc->i.action = 0;
3537 return 0;
3538 }
3539
Tejun Heo9b1e2652007-08-06 18:36:24 +09003540 /* configure transfer mode if necessary */
3541 if (ehc->i.flags & ATA_EHI_SETMODE) {
3542 rc = ata_set_mode(link, &dev);
3543 if (rc)
3544 goto dev_fail;
3545 ehc->i.flags &= ~ATA_EHI_SETMODE;
3546 }
3547
Tejun Heo11fc33d2008-08-30 14:20:01 +02003548 /* If reset has been issued, clear UA to avoid
3549 * disrupting the current users of the device.
3550 */
3551 if (ehc->i.flags & ATA_EHI_DID_RESET) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003552 ata_for_each_dev(dev, link, ALL) {
Tejun Heo11fc33d2008-08-30 14:20:01 +02003553 if (dev->class != ATA_DEV_ATAPI)
3554 continue;
3555 rc = atapi_eh_clear_ua(dev);
3556 if (rc)
3557 goto dev_fail;
3558 }
3559 }
3560
Tejun Heo6013efd2009-11-19 15:36:45 +09003561 /* retry flush if necessary */
3562 ata_for_each_dev(dev, link, ALL) {
3563 if (dev->class != ATA_DEV_ATA)
3564 continue;
3565 rc = ata_eh_maybe_retry_flush(dev);
3566 if (rc)
3567 goto dev_fail;
3568 }
3569
Tejun Heo11fc33d2008-08-30 14:20:01 +02003570 /* configure link power saving */
Tejun Heo3ec25eb2008-03-27 18:37:14 +09003571 if (ehc->i.action & ATA_EH_LPM)
Tejun Heo1eca4362008-11-03 20:03:17 +09003572 ata_for_each_dev(dev, link, ALL)
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003573 ata_dev_enable_pm(dev, ap->pm_policy);
3574
Tejun Heo9b1e2652007-08-06 18:36:24 +09003575 /* this link is okay now */
3576 ehc->i.flags = 0;
3577 continue;
3578
Jeff Garzik2dcb4072007-10-19 06:42:56 -04003579dev_fail:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003580 nr_failed_devs++;
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003581 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003582
Tejun Heob06ce3e2007-10-09 15:06:48 +09003583 if (ap->pflags & ATA_PFLAG_FROZEN) {
3584 /* PMP reset requires working host port.
3585 * Can't retry if it's frozen.
3586 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003587 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003588 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003589 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003590 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003591 }
3592
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003593 if (nr_failed_devs)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003594 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003595
Tejun Heo022bdb02006-05-15 20:58:22 +09003596 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003597 if (rc && r_failed_link)
3598 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003599
3600 DPRINTK("EXIT, rc=%d\n", rc);
3601 return rc;
3602}
3603
3604/**
3605 * ata_eh_finish - finish up EH
3606 * @ap: host port to finish EH for
3607 *
3608 * Recovery is complete. Clean up EH states and retry or finish
3609 * failed qcs.
3610 *
3611 * LOCKING:
3612 * None.
3613 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003614void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003615{
3616 int tag;
3617
3618 /* retry or finish qcs */
3619 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3620 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3621
3622 if (!(qc->flags & ATA_QCFLAG_FAILED))
3623 continue;
3624
3625 if (qc->err_mask) {
3626 /* FIXME: Once EH migration is complete,
3627 * generate sense data in this function,
3628 * considering both err_mask and tf.
3629 */
Tejun Heo03faab72008-03-27 19:14:24 +09003630 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003631 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003632 else
3633 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003634 } else {
3635 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3636 ata_eh_qc_complete(qc);
3637 } else {
3638 /* feed zero TF to sense generation */
3639 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3640 ata_eh_qc_retry(qc);
3641 }
3642 }
3643 }
Tejun Heoda917d62007-09-23 13:14:12 +09003644
3645 /* make sure nr_active_links is zero after EH */
3646 WARN_ON(ap->nr_active_links);
3647 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003648}
3649
3650/**
3651 * ata_do_eh - do standard error handling
3652 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003653 *
Tejun Heof5914a42006-05-31 18:27:48 +09003654 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003655 * @softreset: softreset method (can be NULL)
3656 * @hardreset: hardreset method (can be NULL)
3657 * @postreset: postreset method (can be NULL)
3658 *
3659 * Perform standard error handling sequence.
3660 *
3661 * LOCKING:
3662 * Kernel thread context (may sleep).
3663 */
Tejun Heof5914a42006-05-31 18:27:48 +09003664void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3665 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3666 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003667{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003668 struct ata_device *dev;
3669 int rc;
3670
3671 ata_eh_autopsy(ap);
3672 ata_eh_report(ap);
3673
3674 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3675 NULL);
3676 if (rc) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003677 ata_for_each_dev(dev, &ap->link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003678 ata_dev_disable(dev);
3679 }
3680
Tejun Heo022bdb02006-05-15 20:58:22 +09003681 ata_eh_finish(ap);
3682}
Tejun Heo500530f2006-07-03 16:07:27 +09003683
Tejun Heoa1efdab2008-03-25 12:22:50 +09003684/**
3685 * ata_std_error_handler - standard error handler
3686 * @ap: host port to handle error for
3687 *
3688 * Standard error handler
3689 *
3690 * LOCKING:
3691 * Kernel thread context (may sleep).
3692 */
3693void ata_std_error_handler(struct ata_port *ap)
3694{
3695 struct ata_port_operations *ops = ap->ops;
3696 ata_reset_fn_t hardreset = ops->hardreset;
3697
Tejun Heo57c9efd2008-04-07 22:47:19 +09003698 /* ignore built-in hardreset if SCR access is not available */
Tejun Heofe06e5f2010-05-10 21:41:39 +02003699 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09003700 hardreset = NULL;
3701
3702 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3703}
3704
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003705#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09003706/**
3707 * ata_eh_handle_port_suspend - perform port suspend operation
3708 * @ap: port to suspend
3709 *
3710 * Suspend @ap.
3711 *
3712 * LOCKING:
3713 * Kernel thread context (may sleep).
3714 */
3715static void ata_eh_handle_port_suspend(struct ata_port *ap)
3716{
3717 unsigned long flags;
3718 int rc = 0;
3719
3720 /* are we suspending? */
3721 spin_lock_irqsave(ap->lock, flags);
3722 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3723 ap->pm_mesg.event == PM_EVENT_ON) {
3724 spin_unlock_irqrestore(ap->lock, flags);
3725 return;
3726 }
3727 spin_unlock_irqrestore(ap->lock, flags);
3728
3729 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3730
Tejun Heo64578a32007-05-15 03:28:16 +09003731 /* tell ACPI we're suspending */
3732 rc = ata_acpi_on_suspend(ap);
3733 if (rc)
3734 goto out;
3735
Tejun Heo500530f2006-07-03 16:07:27 +09003736 /* suspend */
3737 ata_eh_freeze_port(ap);
3738
3739 if (ap->ops->port_suspend)
3740 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3741
Shaohua Libd3adca2007-11-02 09:32:38 +08003742 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a32007-05-15 03:28:16 +09003743 out:
Tejun Heo500530f2006-07-03 16:07:27 +09003744 /* report result */
3745 spin_lock_irqsave(ap->lock, flags);
3746
3747 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3748 if (rc == 0)
3749 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09003750 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09003751 ata_port_schedule_eh(ap);
3752
3753 if (ap->pm_result) {
3754 *ap->pm_result = rc;
3755 ap->pm_result = NULL;
3756 }
3757
3758 spin_unlock_irqrestore(ap->lock, flags);
3759
3760 return;
3761}
3762
3763/**
3764 * ata_eh_handle_port_resume - perform port resume operation
3765 * @ap: port to resume
3766 *
3767 * Resume @ap.
3768 *
Tejun Heo500530f2006-07-03 16:07:27 +09003769 * LOCKING:
3770 * Kernel thread context (may sleep).
3771 */
3772static void ata_eh_handle_port_resume(struct ata_port *ap)
3773{
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09003774 struct ata_link *link;
3775 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09003776 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02003777 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09003778
3779 /* are we resuming? */
3780 spin_lock_irqsave(ap->lock, flags);
3781 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3782 ap->pm_mesg.event != PM_EVENT_ON) {
3783 spin_unlock_irqrestore(ap->lock, flags);
3784 return;
3785 }
3786 spin_unlock_irqrestore(ap->lock, flags);
3787
Tejun Heo9666f402007-05-04 21:27:47 +02003788 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09003789
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09003790 /*
3791 * Error timestamps are in jiffies which doesn't run while
3792 * suspended and PHY events during resume isn't too uncommon.
3793 * When the two are combined, it can lead to unnecessary speed
3794 * downs if the machine is suspended and resumed repeatedly.
3795 * Clear error history.
3796 */
3797 ata_for_each_link(link, ap, HOST_FIRST)
3798 ata_for_each_dev(dev, link, ALL)
3799 ata_ering_clear(&dev->ering);
3800
Shaohua Libd3adca2007-11-02 09:32:38 +08003801 ata_acpi_set_state(ap, PMSG_ON);
3802
Tejun Heo500530f2006-07-03 16:07:27 +09003803 if (ap->ops->port_resume)
3804 rc = ap->ops->port_resume(ap);
3805
Tejun Heo67465442007-05-15 03:28:16 +09003806 /* tell ACPI that we're resuming */
3807 ata_acpi_on_resume(ap);
3808
Tejun Heo9666f402007-05-04 21:27:47 +02003809 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09003810 spin_lock_irqsave(ap->lock, flags);
3811 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3812 if (ap->pm_result) {
3813 *ap->pm_result = rc;
3814 ap->pm_result = NULL;
3815 }
3816 spin_unlock_irqrestore(ap->lock, flags);
3817}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003818#endif /* CONFIG_PM */