| 1 | /* |
| 2 | * libata-eh.c - libata error handling |
| 3 | * |
| 4 | * Maintained by: Tejun Heo <tj@kernel.org> |
| 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/driver-api/libata.rst |
| 29 | * |
| 30 | * Hardware documentation available from http://www.t13.org/ and |
| 31 | * http://www.sata-io.org/ |
| 32 | * |
| 33 | */ |
| 34 | |
| 35 | #include <linux/kernel.h> |
| 36 | #include <linux/blkdev.h> |
| 37 | #include <linux/export.h> |
| 38 | #include <linux/pci.h> |
| 39 | #include <scsi/scsi.h> |
| 40 | #include <scsi/scsi_host.h> |
| 41 | #include <scsi/scsi_eh.h> |
| 42 | #include <scsi/scsi_device.h> |
| 43 | #include <scsi/scsi_cmnd.h> |
| 44 | #include <scsi/scsi_dbg.h> |
| 45 | #include "../scsi/scsi_transport_api.h" |
| 46 | |
| 47 | #include <linux/libata.h> |
| 48 | |
| 49 | #include <trace/events/libata.h> |
| 50 | #include "libata.h" |
| 51 | |
| 52 | enum { |
| 53 | /* speed down verdicts */ |
| 54 | ATA_EH_SPDN_NCQ_OFF = (1 << 0), |
| 55 | ATA_EH_SPDN_SPEED_DOWN = (1 << 1), |
| 56 | ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2), |
| 57 | ATA_EH_SPDN_KEEP_ERRORS = (1 << 3), |
| 58 | |
| 59 | /* error flags */ |
| 60 | ATA_EFLAG_IS_IO = (1 << 0), |
| 61 | ATA_EFLAG_DUBIOUS_XFER = (1 << 1), |
| 62 | ATA_EFLAG_OLD_ER = (1 << 31), |
| 63 | |
| 64 | /* error categories */ |
| 65 | ATA_ECAT_NONE = 0, |
| 66 | ATA_ECAT_ATA_BUS = 1, |
| 67 | ATA_ECAT_TOUT_HSM = 2, |
| 68 | ATA_ECAT_UNK_DEV = 3, |
| 69 | ATA_ECAT_DUBIOUS_NONE = 4, |
| 70 | ATA_ECAT_DUBIOUS_ATA_BUS = 5, |
| 71 | ATA_ECAT_DUBIOUS_TOUT_HSM = 6, |
| 72 | ATA_ECAT_DUBIOUS_UNK_DEV = 7, |
| 73 | ATA_ECAT_NR = 8, |
| 74 | |
| 75 | ATA_EH_CMD_DFL_TIMEOUT = 5000, |
| 76 | |
| 77 | /* always put at least this amount of time between resets */ |
| 78 | ATA_EH_RESET_COOL_DOWN = 5000, |
| 79 | |
| 80 | /* Waiting in ->prereset can never be reliable. It's |
| 81 | * sometimes nice to wait there but it can't be depended upon; |
| 82 | * otherwise, we wouldn't be resetting. Just give it enough |
| 83 | * time for most drives to spin up. |
| 84 | */ |
| 85 | ATA_EH_PRERESET_TIMEOUT = 10000, |
| 86 | ATA_EH_FASTDRAIN_INTERVAL = 3000, |
| 87 | |
| 88 | ATA_EH_UA_TRIES = 5, |
| 89 | |
| 90 | /* probe speed down parameters, see ata_eh_schedule_probe() */ |
| 91 | ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */ |
| 92 | ATA_EH_PROBE_TRIALS = 2, |
| 93 | }; |
| 94 | |
| 95 | /* The following table determines how we sequence resets. Each entry |
| 96 | * represents timeout for that try. The first try can be soft or |
| 97 | * hardreset. All others are hardreset if available. In most cases |
| 98 | * the first reset w/ 10sec timeout should succeed. Following entries |
| 99 | * are mostly for error handling, hotplug and those outlier devices that |
| 100 | * take an exceptionally long time to recover from reset. |
| 101 | */ |
| 102 | static const unsigned long ata_eh_reset_timeouts[] = { |
| 103 | 10000, /* most drives spin up by 10sec */ |
| 104 | 10000, /* > 99% working drives spin up before 20sec */ |
| 105 | 35000, /* give > 30 secs of idleness for outlier devices */ |
| 106 | 5000, /* and sweet one last chance */ |
| 107 | ULONG_MAX, /* > 1 min has elapsed, give up */ |
| 108 | }; |
| 109 | |
| 110 | static const unsigned long ata_eh_identify_timeouts[] = { |
| 111 | 5000, /* covers > 99% of successes and not too boring on failures */ |
| 112 | 10000, /* combined time till here is enough even for media access */ |
| 113 | 30000, /* for true idiots */ |
| 114 | ULONG_MAX, |
| 115 | }; |
| 116 | |
| 117 | static const unsigned long ata_eh_flush_timeouts[] = { |
| 118 | 15000, /* be generous with flush */ |
| 119 | 15000, /* ditto */ |
| 120 | 30000, /* and even more generous */ |
| 121 | ULONG_MAX, |
| 122 | }; |
| 123 | |
| 124 | static const unsigned long ata_eh_other_timeouts[] = { |
| 125 | 5000, /* same rationale as identify timeout */ |
| 126 | 10000, /* ditto */ |
| 127 | /* but no merciful 30sec for other commands, it just isn't worth it */ |
| 128 | ULONG_MAX, |
| 129 | }; |
| 130 | |
| 131 | struct ata_eh_cmd_timeout_ent { |
| 132 | const u8 *commands; |
| 133 | const unsigned long *timeouts; |
| 134 | }; |
| 135 | |
| 136 | /* The following table determines timeouts to use for EH internal |
| 137 | * commands. Each table entry is a command class and matches the |
| 138 | * commands the entry applies to and the timeout table to use. |
| 139 | * |
| 140 | * On the retry after a command timed out, the next timeout value from |
| 141 | * the table is used. If the table doesn't contain further entries, |
| 142 | * the last value is used. |
| 143 | * |
| 144 | * ehc->cmd_timeout_idx keeps track of which timeout to use per |
| 145 | * command class, so if SET_FEATURES times out on the first try, the |
| 146 | * next try will use the second timeout value only for that class. |
| 147 | */ |
| 148 | #define CMDS(cmds...) (const u8 []){ cmds, 0 } |
| 149 | static const struct ata_eh_cmd_timeout_ent |
| 150 | ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = { |
| 151 | { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI), |
| 152 | .timeouts = ata_eh_identify_timeouts, }, |
| 153 | { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT), |
| 154 | .timeouts = ata_eh_other_timeouts, }, |
| 155 | { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT), |
| 156 | .timeouts = ata_eh_other_timeouts, }, |
| 157 | { .commands = CMDS(ATA_CMD_SET_FEATURES), |
| 158 | .timeouts = ata_eh_other_timeouts, }, |
| 159 | { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS), |
| 160 | .timeouts = ata_eh_other_timeouts, }, |
| 161 | { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT), |
| 162 | .timeouts = ata_eh_flush_timeouts }, |
| 163 | }; |
| 164 | #undef CMDS |
| 165 | |
| 166 | static void __ata_port_freeze(struct ata_port *ap); |
| 167 | #ifdef CONFIG_PM |
| 168 | static void ata_eh_handle_port_suspend(struct ata_port *ap); |
| 169 | static void ata_eh_handle_port_resume(struct ata_port *ap); |
| 170 | #else /* CONFIG_PM */ |
| 171 | static void ata_eh_handle_port_suspend(struct ata_port *ap) |
| 172 | { } |
| 173 | |
| 174 | static void ata_eh_handle_port_resume(struct ata_port *ap) |
| 175 | { } |
| 176 | #endif /* CONFIG_PM */ |
| 177 | |
| 178 | static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, |
| 179 | const char *fmt, va_list args) |
| 180 | { |
| 181 | ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len, |
| 182 | ATA_EH_DESC_LEN - ehi->desc_len, |
| 183 | fmt, args); |
| 184 | } |
| 185 | |
| 186 | /** |
| 187 | * __ata_ehi_push_desc - push error description without adding separator |
| 188 | * @ehi: target EHI |
| 189 | * @fmt: printf format string |
| 190 | * |
| 191 | * Format string according to @fmt and append it to @ehi->desc. |
| 192 | * |
| 193 | * LOCKING: |
| 194 | * spin_lock_irqsave(host lock) |
| 195 | */ |
| 196 | void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...) |
| 197 | { |
| 198 | va_list args; |
| 199 | |
| 200 | va_start(args, fmt); |
| 201 | __ata_ehi_pushv_desc(ehi, fmt, args); |
| 202 | va_end(args); |
| 203 | } |
| 204 | |
| 205 | /** |
| 206 | * ata_ehi_push_desc - push error description with separator |
| 207 | * @ehi: target EHI |
| 208 | * @fmt: printf format string |
| 209 | * |
| 210 | * Format string according to @fmt and append it to @ehi->desc. |
| 211 | * If @ehi->desc is not empty, ", " is added in-between. |
| 212 | * |
| 213 | * LOCKING: |
| 214 | * spin_lock_irqsave(host lock) |
| 215 | */ |
| 216 | void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...) |
| 217 | { |
| 218 | va_list args; |
| 219 | |
| 220 | if (ehi->desc_len) |
| 221 | __ata_ehi_push_desc(ehi, ", "); |
| 222 | |
| 223 | va_start(args, fmt); |
| 224 | __ata_ehi_pushv_desc(ehi, fmt, args); |
| 225 | va_end(args); |
| 226 | } |
| 227 | |
| 228 | /** |
| 229 | * ata_ehi_clear_desc - clean error description |
| 230 | * @ehi: target EHI |
| 231 | * |
| 232 | * Clear @ehi->desc. |
| 233 | * |
| 234 | * LOCKING: |
| 235 | * spin_lock_irqsave(host lock) |
| 236 | */ |
| 237 | void ata_ehi_clear_desc(struct ata_eh_info *ehi) |
| 238 | { |
| 239 | ehi->desc[0] = '\0'; |
| 240 | ehi->desc_len = 0; |
| 241 | } |
| 242 | |
| 243 | /** |
| 244 | * ata_port_desc - append port description |
| 245 | * @ap: target ATA port |
| 246 | * @fmt: printf format string |
| 247 | * |
| 248 | * Format string according to @fmt and append it to port |
| 249 | * description. If port description is not empty, " " is added |
| 250 | * in-between. This function is to be used while initializing |
| 251 | * ata_host. The description is printed on host registration. |
| 252 | * |
| 253 | * LOCKING: |
| 254 | * None. |
| 255 | */ |
| 256 | void ata_port_desc(struct ata_port *ap, const char *fmt, ...) |
| 257 | { |
| 258 | va_list args; |
| 259 | |
| 260 | WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING)); |
| 261 | |
| 262 | if (ap->link.eh_info.desc_len) |
| 263 | __ata_ehi_push_desc(&ap->link.eh_info, " "); |
| 264 | |
| 265 | va_start(args, fmt); |
| 266 | __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args); |
| 267 | va_end(args); |
| 268 | } |
| 269 | |
| 270 | #ifdef CONFIG_PCI |
| 271 | |
| 272 | /** |
| 273 | * ata_port_pbar_desc - append PCI BAR description |
| 274 | * @ap: target ATA port |
| 275 | * @bar: target PCI BAR |
| 276 | * @offset: offset into PCI BAR |
| 277 | * @name: name of the area |
| 278 | * |
| 279 | * If @offset is negative, this function formats a string which |
| 280 | * contains the name, address, size and type of the BAR and |
| 281 | * appends it to the port description. If @offset is zero or |
| 282 | * positive, only name and offsetted address is appended. |
| 283 | * |
| 284 | * LOCKING: |
| 285 | * None. |
| 286 | */ |
| 287 | void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset, |
| 288 | const char *name) |
| 289 | { |
| 290 | struct pci_dev *pdev = to_pci_dev(ap->host->dev); |
| 291 | char *type = ""; |
| 292 | unsigned long long start, len; |
| 293 | |
| 294 | if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) |
| 295 | type = "m"; |
| 296 | else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) |
| 297 | type = "i"; |
| 298 | |
| 299 | start = (unsigned long long)pci_resource_start(pdev, bar); |
| 300 | len = (unsigned long long)pci_resource_len(pdev, bar); |
| 301 | |
| 302 | if (offset < 0) |
| 303 | ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start); |
| 304 | else |
| 305 | ata_port_desc(ap, "%s 0x%llx", name, |
| 306 | start + (unsigned long long)offset); |
| 307 | } |
| 308 | |
| 309 | #endif /* CONFIG_PCI */ |
| 310 | |
| 311 | static int ata_lookup_timeout_table(u8 cmd) |
| 312 | { |
| 313 | int i; |
| 314 | |
| 315 | for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) { |
| 316 | const u8 *cur; |
| 317 | |
| 318 | for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++) |
| 319 | if (*cur == cmd) |
| 320 | return i; |
| 321 | } |
| 322 | |
| 323 | return -1; |
| 324 | } |
| 325 | |
| 326 | /** |
| 327 | * ata_internal_cmd_timeout - determine timeout for an internal command |
| 328 | * @dev: target device |
| 329 | * @cmd: internal command to be issued |
| 330 | * |
| 331 | * Determine timeout for internal command @cmd for @dev. |
| 332 | * |
| 333 | * LOCKING: |
| 334 | * EH context. |
| 335 | * |
| 336 | * RETURNS: |
| 337 | * Determined timeout. |
| 338 | */ |
| 339 | unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd) |
| 340 | { |
| 341 | struct ata_eh_context *ehc = &dev->link->eh_context; |
| 342 | int ent = ata_lookup_timeout_table(cmd); |
| 343 | int idx; |
| 344 | |
| 345 | if (ent < 0) |
| 346 | return ATA_EH_CMD_DFL_TIMEOUT; |
| 347 | |
| 348 | idx = ehc->cmd_timeout_idx[dev->devno][ent]; |
| 349 | return ata_eh_cmd_timeout_table[ent].timeouts[idx]; |
| 350 | } |
| 351 | |
| 352 | /** |
| 353 | * ata_internal_cmd_timed_out - notification for internal command timeout |
| 354 | * @dev: target device |
| 355 | * @cmd: internal command which timed out |
| 356 | * |
| 357 | * Notify EH that internal command @cmd for @dev timed out. This |
| 358 | * function should be called only for commands whose timeouts are |
| 359 | * determined using ata_internal_cmd_timeout(). |
| 360 | * |
| 361 | * LOCKING: |
| 362 | * EH context. |
| 363 | */ |
| 364 | void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd) |
| 365 | { |
| 366 | struct ata_eh_context *ehc = &dev->link->eh_context; |
| 367 | int ent = ata_lookup_timeout_table(cmd); |
| 368 | int idx; |
| 369 | |
| 370 | if (ent < 0) |
| 371 | return; |
| 372 | |
| 373 | idx = ehc->cmd_timeout_idx[dev->devno][ent]; |
| 374 | if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX) |
| 375 | ehc->cmd_timeout_idx[dev->devno][ent]++; |
| 376 | } |
| 377 | |
| 378 | static void ata_ering_record(struct ata_ering *ering, unsigned int eflags, |
| 379 | unsigned int err_mask) |
| 380 | { |
| 381 | struct ata_ering_entry *ent; |
| 382 | |
| 383 | WARN_ON(!err_mask); |
| 384 | |
| 385 | ering->cursor++; |
| 386 | ering->cursor %= ATA_ERING_SIZE; |
| 387 | |
| 388 | ent = &ering->ring[ering->cursor]; |
| 389 | ent->eflags = eflags; |
| 390 | ent->err_mask = err_mask; |
| 391 | ent->timestamp = get_jiffies_64(); |
| 392 | } |
| 393 | |
| 394 | static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering) |
| 395 | { |
| 396 | struct ata_ering_entry *ent = &ering->ring[ering->cursor]; |
| 397 | |
| 398 | if (ent->err_mask) |
| 399 | return ent; |
| 400 | return NULL; |
| 401 | } |
| 402 | |
| 403 | int ata_ering_map(struct ata_ering *ering, |
| 404 | int (*map_fn)(struct ata_ering_entry *, void *), |
| 405 | void *arg) |
| 406 | { |
| 407 | int idx, rc = 0; |
| 408 | struct ata_ering_entry *ent; |
| 409 | |
| 410 | idx = ering->cursor; |
| 411 | do { |
| 412 | ent = &ering->ring[idx]; |
| 413 | if (!ent->err_mask) |
| 414 | break; |
| 415 | rc = map_fn(ent, arg); |
| 416 | if (rc) |
| 417 | break; |
| 418 | idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE; |
| 419 | } while (idx != ering->cursor); |
| 420 | |
| 421 | return rc; |
| 422 | } |
| 423 | |
| 424 | static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg) |
| 425 | { |
| 426 | ent->eflags |= ATA_EFLAG_OLD_ER; |
| 427 | return 0; |
| 428 | } |
| 429 | |
| 430 | static void ata_ering_clear(struct ata_ering *ering) |
| 431 | { |
| 432 | ata_ering_map(ering, ata_ering_clear_cb, NULL); |
| 433 | } |
| 434 | |
| 435 | static unsigned int ata_eh_dev_action(struct ata_device *dev) |
| 436 | { |
| 437 | struct ata_eh_context *ehc = &dev->link->eh_context; |
| 438 | |
| 439 | return ehc->i.action | ehc->i.dev_action[dev->devno]; |
| 440 | } |
| 441 | |
| 442 | static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev, |
| 443 | struct ata_eh_info *ehi, unsigned int action) |
| 444 | { |
| 445 | struct ata_device *tdev; |
| 446 | |
| 447 | if (!dev) { |
| 448 | ehi->action &= ~action; |
| 449 | ata_for_each_dev(tdev, link, ALL) |
| 450 | ehi->dev_action[tdev->devno] &= ~action; |
| 451 | } else { |
| 452 | /* doesn't make sense for port-wide EH actions */ |
| 453 | WARN_ON(!(action & ATA_EH_PERDEV_MASK)); |
| 454 | |
| 455 | /* break ehi->action into ehi->dev_action */ |
| 456 | if (ehi->action & action) { |
| 457 | ata_for_each_dev(tdev, link, ALL) |
| 458 | ehi->dev_action[tdev->devno] |= |
| 459 | ehi->action & action; |
| 460 | ehi->action &= ~action; |
| 461 | } |
| 462 | |
| 463 | /* turn off the specified per-dev action */ |
| 464 | ehi->dev_action[dev->devno] &= ~action; |
| 465 | } |
| 466 | } |
| 467 | |
| 468 | /** |
| 469 | * ata_eh_acquire - acquire EH ownership |
| 470 | * @ap: ATA port to acquire EH ownership for |
| 471 | * |
| 472 | * Acquire EH ownership for @ap. This is the basic exclusion |
| 473 | * mechanism for ports sharing a host. Only one port hanging off |
| 474 | * the same host can claim the ownership of EH. |
| 475 | * |
| 476 | * LOCKING: |
| 477 | * EH context. |
| 478 | */ |
| 479 | void ata_eh_acquire(struct ata_port *ap) |
| 480 | { |
| 481 | mutex_lock(&ap->host->eh_mutex); |
| 482 | WARN_ON_ONCE(ap->host->eh_owner); |
| 483 | ap->host->eh_owner = current; |
| 484 | } |
| 485 | |
| 486 | /** |
| 487 | * ata_eh_release - release EH ownership |
| 488 | * @ap: ATA port to release EH ownership for |
| 489 | * |
| 490 | * Release EH ownership for @ap if the caller. The caller must |
| 491 | * have acquired EH ownership using ata_eh_acquire() previously. |
| 492 | * |
| 493 | * LOCKING: |
| 494 | * EH context. |
| 495 | */ |
| 496 | void ata_eh_release(struct ata_port *ap) |
| 497 | { |
| 498 | WARN_ON_ONCE(ap->host->eh_owner != current); |
| 499 | ap->host->eh_owner = NULL; |
| 500 | mutex_unlock(&ap->host->eh_mutex); |
| 501 | } |
| 502 | |
| 503 | /** |
| 504 | * ata_scsi_timed_out - SCSI layer time out callback |
| 505 | * @cmd: timed out SCSI command |
| 506 | * |
| 507 | * Handles SCSI layer timeout. We race with normal completion of |
| 508 | * the qc for @cmd. If the qc is already gone, we lose and let |
| 509 | * the scsi command finish (EH_HANDLED). Otherwise, the qc has |
| 510 | * timed out and EH should be invoked. Prevent ata_qc_complete() |
| 511 | * from finishing it by setting EH_SCHEDULED and return |
| 512 | * EH_NOT_HANDLED. |
| 513 | * |
| 514 | * TODO: kill this function once old EH is gone. |
| 515 | * |
| 516 | * LOCKING: |
| 517 | * Called from timer context |
| 518 | * |
| 519 | * RETURNS: |
| 520 | * EH_HANDLED or EH_NOT_HANDLED |
| 521 | */ |
| 522 | enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd) |
| 523 | { |
| 524 | struct Scsi_Host *host = cmd->device->host; |
| 525 | struct ata_port *ap = ata_shost_to_port(host); |
| 526 | unsigned long flags; |
| 527 | struct ata_queued_cmd *qc; |
| 528 | enum blk_eh_timer_return ret; |
| 529 | |
| 530 | DPRINTK("ENTER\n"); |
| 531 | |
| 532 | if (ap->ops->error_handler) { |
| 533 | ret = BLK_EH_NOT_HANDLED; |
| 534 | goto out; |
| 535 | } |
| 536 | |
| 537 | ret = BLK_EH_HANDLED; |
| 538 | spin_lock_irqsave(ap->lock, flags); |
| 539 | qc = ata_qc_from_tag(ap, ap->link.active_tag); |
| 540 | if (qc) { |
| 541 | WARN_ON(qc->scsicmd != cmd); |
| 542 | qc->flags |= ATA_QCFLAG_EH_SCHEDULED; |
| 543 | qc->err_mask |= AC_ERR_TIMEOUT; |
| 544 | ret = BLK_EH_NOT_HANDLED; |
| 545 | } |
| 546 | spin_unlock_irqrestore(ap->lock, flags); |
| 547 | |
| 548 | out: |
| 549 | DPRINTK("EXIT, ret=%d\n", ret); |
| 550 | return ret; |
| 551 | } |
| 552 | EXPORT_SYMBOL(ata_scsi_timed_out); |
| 553 | |
| 554 | static void ata_eh_unload(struct ata_port *ap) |
| 555 | { |
| 556 | struct ata_link *link; |
| 557 | struct ata_device *dev; |
| 558 | unsigned long flags; |
| 559 | |
| 560 | /* Restore SControl IPM and SPD for the next driver and |
| 561 | * disable attached devices. |
| 562 | */ |
| 563 | ata_for_each_link(link, ap, PMP_FIRST) { |
| 564 | sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0); |
| 565 | ata_for_each_dev(dev, link, ALL) |
| 566 | ata_dev_disable(dev); |
| 567 | } |
| 568 | |
| 569 | /* freeze and set UNLOADED */ |
| 570 | spin_lock_irqsave(ap->lock, flags); |
| 571 | |
| 572 | ata_port_freeze(ap); /* won't be thawed */ |
| 573 | ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */ |
| 574 | ap->pflags |= ATA_PFLAG_UNLOADED; |
| 575 | |
| 576 | spin_unlock_irqrestore(ap->lock, flags); |
| 577 | } |
| 578 | |
| 579 | /** |
| 580 | * ata_scsi_error - SCSI layer error handler callback |
| 581 | * @host: SCSI host on which error occurred |
| 582 | * |
| 583 | * Handles SCSI-layer-thrown error events. |
| 584 | * |
| 585 | * LOCKING: |
| 586 | * Inherited from SCSI layer (none, can sleep) |
| 587 | * |
| 588 | * RETURNS: |
| 589 | * Zero. |
| 590 | */ |
| 591 | void ata_scsi_error(struct Scsi_Host *host) |
| 592 | { |
| 593 | struct ata_port *ap = ata_shost_to_port(host); |
| 594 | unsigned long flags; |
| 595 | LIST_HEAD(eh_work_q); |
| 596 | |
| 597 | DPRINTK("ENTER\n"); |
| 598 | |
| 599 | spin_lock_irqsave(host->host_lock, flags); |
| 600 | list_splice_init(&host->eh_cmd_q, &eh_work_q); |
| 601 | spin_unlock_irqrestore(host->host_lock, flags); |
| 602 | |
| 603 | ata_scsi_cmd_error_handler(host, ap, &eh_work_q); |
| 604 | |
| 605 | /* If we timed raced normal completion and there is nothing to |
| 606 | recover nr_timedout == 0 why exactly are we doing error recovery ? */ |
| 607 | ata_scsi_port_error_handler(host, ap); |
| 608 | |
| 609 | /* finish or retry handled scmd's and clean up */ |
| 610 | WARN_ON(!list_empty(&eh_work_q)); |
| 611 | |
| 612 | DPRINTK("EXIT\n"); |
| 613 | } |
| 614 | |
| 615 | /** |
| 616 | * ata_scsi_cmd_error_handler - error callback for a list of commands |
| 617 | * @host: scsi host containing the port |
| 618 | * @ap: ATA port within the host |
| 619 | * @eh_work_q: list of commands to process |
| 620 | * |
| 621 | * process the given list of commands and return those finished to the |
| 622 | * ap->eh_done_q. This function is the first part of the libata error |
| 623 | * handler which processes a given list of failed commands. |
| 624 | */ |
| 625 | void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap, |
| 626 | struct list_head *eh_work_q) |
| 627 | { |
| 628 | int i; |
| 629 | unsigned long flags; |
| 630 | |
| 631 | /* make sure sff pio task is not running */ |
| 632 | ata_sff_flush_pio_task(ap); |
| 633 | |
| 634 | /* synchronize with host lock and sort out timeouts */ |
| 635 | |
| 636 | /* For new EH, all qcs are finished in one of three ways - |
| 637 | * normal completion, error completion, and SCSI timeout. |
| 638 | * Both completions can race against SCSI timeout. When normal |
| 639 | * completion wins, the qc never reaches EH. When error |
| 640 | * completion wins, the qc has ATA_QCFLAG_FAILED set. |
| 641 | * |
| 642 | * When SCSI timeout wins, things are a bit more complex. |
| 643 | * Normal or error completion can occur after the timeout but |
| 644 | * before this point. In such cases, both types of |
| 645 | * completions are honored. A scmd is determined to have |
| 646 | * timed out iff its associated qc is active and not failed. |
| 647 | */ |
| 648 | spin_lock_irqsave(ap->lock, flags); |
| 649 | if (ap->ops->error_handler) { |
| 650 | struct scsi_cmnd *scmd, *tmp; |
| 651 | int nr_timedout = 0; |
| 652 | |
| 653 | /* This must occur under the ap->lock as we don't want |
| 654 | a polled recovery to race the real interrupt handler |
| 655 | |
| 656 | The lost_interrupt handler checks for any completed but |
| 657 | non-notified command and completes much like an IRQ handler. |
| 658 | |
| 659 | We then fall into the error recovery code which will treat |
| 660 | this as if normal completion won the race */ |
| 661 | |
| 662 | if (ap->ops->lost_interrupt) |
| 663 | ap->ops->lost_interrupt(ap); |
| 664 | |
| 665 | list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) { |
| 666 | struct ata_queued_cmd *qc; |
| 667 | |
| 668 | for (i = 0; i < ATA_MAX_QUEUE; i++) { |
| 669 | qc = __ata_qc_from_tag(ap, i); |
| 670 | if (qc->flags & ATA_QCFLAG_ACTIVE && |
| 671 | qc->scsicmd == scmd) |
| 672 | break; |
| 673 | } |
| 674 | |
| 675 | if (i < ATA_MAX_QUEUE) { |
| 676 | /* the scmd has an associated qc */ |
| 677 | if (!(qc->flags & ATA_QCFLAG_FAILED)) { |
| 678 | /* which hasn't failed yet, timeout */ |
| 679 | qc->err_mask |= AC_ERR_TIMEOUT; |
| 680 | qc->flags |= ATA_QCFLAG_FAILED; |
| 681 | nr_timedout++; |
| 682 | } |
| 683 | } else { |
| 684 | /* Normal completion occurred after |
| 685 | * SCSI timeout but before this point. |
| 686 | * Successfully complete it. |
| 687 | */ |
| 688 | scmd->retries = scmd->allowed; |
| 689 | scsi_eh_finish_cmd(scmd, &ap->eh_done_q); |
| 690 | } |
| 691 | } |
| 692 | |
| 693 | /* If we have timed out qcs. They belong to EH from |
| 694 | * this point but the state of the controller is |
| 695 | * unknown. Freeze the port to make sure the IRQ |
| 696 | * handler doesn't diddle with those qcs. This must |
| 697 | * be done atomically w.r.t. setting QCFLAG_FAILED. |
| 698 | */ |
| 699 | if (nr_timedout) |
| 700 | __ata_port_freeze(ap); |
| 701 | |
| 702 | |
| 703 | /* initialize eh_tries */ |
| 704 | ap->eh_tries = ATA_EH_MAX_TRIES; |
| 705 | } |
| 706 | spin_unlock_irqrestore(ap->lock, flags); |
| 707 | |
| 708 | } |
| 709 | EXPORT_SYMBOL(ata_scsi_cmd_error_handler); |
| 710 | |
| 711 | /** |
| 712 | * ata_scsi_port_error_handler - recover the port after the commands |
| 713 | * @host: SCSI host containing the port |
| 714 | * @ap: the ATA port |
| 715 | * |
| 716 | * Handle the recovery of the port @ap after all the commands |
| 717 | * have been recovered. |
| 718 | */ |
| 719 | void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap) |
| 720 | { |
| 721 | unsigned long flags; |
| 722 | |
| 723 | /* invoke error handler */ |
| 724 | if (ap->ops->error_handler) { |
| 725 | struct ata_link *link; |
| 726 | |
| 727 | /* acquire EH ownership */ |
| 728 | ata_eh_acquire(ap); |
| 729 | repeat: |
| 730 | /* kill fast drain timer */ |
| 731 | del_timer_sync(&ap->fastdrain_timer); |
| 732 | |
| 733 | /* process port resume request */ |
| 734 | ata_eh_handle_port_resume(ap); |
| 735 | |
| 736 | /* fetch & clear EH info */ |
| 737 | spin_lock_irqsave(ap->lock, flags); |
| 738 | |
| 739 | ata_for_each_link(link, ap, HOST_FIRST) { |
| 740 | struct ata_eh_context *ehc = &link->eh_context; |
| 741 | struct ata_device *dev; |
| 742 | |
| 743 | memset(&link->eh_context, 0, sizeof(link->eh_context)); |
| 744 | link->eh_context.i = link->eh_info; |
| 745 | memset(&link->eh_info, 0, sizeof(link->eh_info)); |
| 746 | |
| 747 | ata_for_each_dev(dev, link, ENABLED) { |
| 748 | int devno = dev->devno; |
| 749 | |
| 750 | ehc->saved_xfer_mode[devno] = dev->xfer_mode; |
| 751 | if (ata_ncq_enabled(dev)) |
| 752 | ehc->saved_ncq_enabled |= 1 << devno; |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS; |
| 757 | ap->pflags &= ~ATA_PFLAG_EH_PENDING; |
| 758 | ap->excl_link = NULL; /* don't maintain exclusion over EH */ |
| 759 | |
| 760 | spin_unlock_irqrestore(ap->lock, flags); |
| 761 | |
| 762 | /* invoke EH, skip if unloading or suspended */ |
| 763 | if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED))) |
| 764 | ap->ops->error_handler(ap); |
| 765 | else { |
| 766 | /* if unloading, commence suicide */ |
| 767 | if ((ap->pflags & ATA_PFLAG_UNLOADING) && |
| 768 | !(ap->pflags & ATA_PFLAG_UNLOADED)) |
| 769 | ata_eh_unload(ap); |
| 770 | ata_eh_finish(ap); |
| 771 | } |
| 772 | |
| 773 | /* process port suspend request */ |
| 774 | ata_eh_handle_port_suspend(ap); |
| 775 | |
| 776 | /* Exception might have happened after ->error_handler |
| 777 | * recovered the port but before this point. Repeat |
| 778 | * EH in such case. |
| 779 | */ |
| 780 | spin_lock_irqsave(ap->lock, flags); |
| 781 | |
| 782 | if (ap->pflags & ATA_PFLAG_EH_PENDING) { |
| 783 | if (--ap->eh_tries) { |
| 784 | spin_unlock_irqrestore(ap->lock, flags); |
| 785 | goto repeat; |
| 786 | } |
| 787 | ata_port_err(ap, |
| 788 | "EH pending after %d tries, giving up\n", |
| 789 | ATA_EH_MAX_TRIES); |
| 790 | ap->pflags &= ~ATA_PFLAG_EH_PENDING; |
| 791 | } |
| 792 | |
| 793 | /* this run is complete, make sure EH info is clear */ |
| 794 | ata_for_each_link(link, ap, HOST_FIRST) |
| 795 | memset(&link->eh_info, 0, sizeof(link->eh_info)); |
| 796 | |
| 797 | /* end eh (clear host_eh_scheduled) while holding |
| 798 | * ap->lock such that if exception occurs after this |
| 799 | * point but before EH completion, SCSI midlayer will |
| 800 | * re-initiate EH. |
| 801 | */ |
| 802 | ap->ops->end_eh(ap); |
| 803 | |
| 804 | spin_unlock_irqrestore(ap->lock, flags); |
| 805 | ata_eh_release(ap); |
| 806 | } else { |
| 807 | WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL); |
| 808 | ap->ops->eng_timeout(ap); |
| 809 | } |
| 810 | |
| 811 | scsi_eh_flush_done_q(&ap->eh_done_q); |
| 812 | |
| 813 | /* clean up */ |
| 814 | spin_lock_irqsave(ap->lock, flags); |
| 815 | |
| 816 | if (ap->pflags & ATA_PFLAG_LOADING) |
| 817 | ap->pflags &= ~ATA_PFLAG_LOADING; |
| 818 | else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) |
| 819 | schedule_delayed_work(&ap->hotplug_task, 0); |
| 820 | |
| 821 | if (ap->pflags & ATA_PFLAG_RECOVERED) |
| 822 | ata_port_info(ap, "EH complete\n"); |
| 823 | |
| 824 | ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED); |
| 825 | |
| 826 | /* tell wait_eh that we're done */ |
| 827 | ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS; |
| 828 | wake_up_all(&ap->eh_wait_q); |
| 829 | |
| 830 | spin_unlock_irqrestore(ap->lock, flags); |
| 831 | } |
| 832 | EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler); |
| 833 | |
| 834 | /** |
| 835 | * ata_port_wait_eh - Wait for the currently pending EH to complete |
| 836 | * @ap: Port to wait EH for |
| 837 | * |
| 838 | * Wait until the currently pending EH is complete. |
| 839 | * |
| 840 | * LOCKING: |
| 841 | * Kernel thread context (may sleep). |
| 842 | */ |
| 843 | void ata_port_wait_eh(struct ata_port *ap) |
| 844 | { |
| 845 | unsigned long flags; |
| 846 | DEFINE_WAIT(wait); |
| 847 | |
| 848 | retry: |
| 849 | spin_lock_irqsave(ap->lock, flags); |
| 850 | |
| 851 | while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) { |
| 852 | prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE); |
| 853 | spin_unlock_irqrestore(ap->lock, flags); |
| 854 | schedule(); |
| 855 | spin_lock_irqsave(ap->lock, flags); |
| 856 | } |
| 857 | finish_wait(&ap->eh_wait_q, &wait); |
| 858 | |
| 859 | spin_unlock_irqrestore(ap->lock, flags); |
| 860 | |
| 861 | /* make sure SCSI EH is complete */ |
| 862 | if (scsi_host_in_recovery(ap->scsi_host)) { |
| 863 | ata_msleep(ap, 10); |
| 864 | goto retry; |
| 865 | } |
| 866 | } |
| 867 | EXPORT_SYMBOL_GPL(ata_port_wait_eh); |
| 868 | |
| 869 | static int ata_eh_nr_in_flight(struct ata_port *ap) |
| 870 | { |
| 871 | unsigned int tag; |
| 872 | int nr = 0; |
| 873 | |
| 874 | /* count only non-internal commands */ |
| 875 | for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) |
| 876 | if (ata_qc_from_tag(ap, tag)) |
| 877 | nr++; |
| 878 | |
| 879 | return nr; |
| 880 | } |
| 881 | |
| 882 | void ata_eh_fastdrain_timerfn(unsigned long arg) |
| 883 | { |
| 884 | struct ata_port *ap = (void *)arg; |
| 885 | unsigned long flags; |
| 886 | int cnt; |
| 887 | |
| 888 | spin_lock_irqsave(ap->lock, flags); |
| 889 | |
| 890 | cnt = ata_eh_nr_in_flight(ap); |
| 891 | |
| 892 | /* are we done? */ |
| 893 | if (!cnt) |
| 894 | goto out_unlock; |
| 895 | |
| 896 | if (cnt == ap->fastdrain_cnt) { |
| 897 | unsigned int tag; |
| 898 | |
| 899 | /* No progress during the last interval, tag all |
| 900 | * in-flight qcs as timed out and freeze the port. |
| 901 | */ |
| 902 | for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) { |
| 903 | struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag); |
| 904 | if (qc) |
| 905 | qc->err_mask |= AC_ERR_TIMEOUT; |
| 906 | } |
| 907 | |
| 908 | ata_port_freeze(ap); |
| 909 | } else { |
| 910 | /* some qcs have finished, give it another chance */ |
| 911 | ap->fastdrain_cnt = cnt; |
| 912 | ap->fastdrain_timer.expires = |
| 913 | ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL); |
| 914 | add_timer(&ap->fastdrain_timer); |
| 915 | } |
| 916 | |
| 917 | out_unlock: |
| 918 | spin_unlock_irqrestore(ap->lock, flags); |
| 919 | } |
| 920 | |
| 921 | /** |
| 922 | * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain |
| 923 | * @ap: target ATA port |
| 924 | * @fastdrain: activate fast drain |
| 925 | * |
| 926 | * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain |
| 927 | * is non-zero and EH wasn't pending before. Fast drain ensures |
| 928 | * that EH kicks in in timely manner. |
| 929 | * |
| 930 | * LOCKING: |
| 931 | * spin_lock_irqsave(host lock) |
| 932 | */ |
| 933 | static void ata_eh_set_pending(struct ata_port *ap, int fastdrain) |
| 934 | { |
| 935 | int cnt; |
| 936 | |
| 937 | /* already scheduled? */ |
| 938 | if (ap->pflags & ATA_PFLAG_EH_PENDING) |
| 939 | return; |
| 940 | |
| 941 | ap->pflags |= ATA_PFLAG_EH_PENDING; |
| 942 | |
| 943 | if (!fastdrain) |
| 944 | return; |
| 945 | |
| 946 | /* do we have in-flight qcs? */ |
| 947 | cnt = ata_eh_nr_in_flight(ap); |
| 948 | if (!cnt) |
| 949 | return; |
| 950 | |
| 951 | /* activate fast drain */ |
| 952 | ap->fastdrain_cnt = cnt; |
| 953 | ap->fastdrain_timer.expires = |
| 954 | ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL); |
| 955 | add_timer(&ap->fastdrain_timer); |
| 956 | } |
| 957 | |
| 958 | /** |
| 959 | * ata_qc_schedule_eh - schedule qc for error handling |
| 960 | * @qc: command to schedule error handling for |
| 961 | * |
| 962 | * Schedule error handling for @qc. EH will kick in as soon as |
| 963 | * other commands are drained. |
| 964 | * |
| 965 | * LOCKING: |
| 966 | * spin_lock_irqsave(host lock) |
| 967 | */ |
| 968 | void ata_qc_schedule_eh(struct ata_queued_cmd *qc) |
| 969 | { |
| 970 | struct ata_port *ap = qc->ap; |
| 971 | struct request_queue *q = qc->scsicmd->device->request_queue; |
| 972 | unsigned long flags; |
| 973 | |
| 974 | WARN_ON(!ap->ops->error_handler); |
| 975 | |
| 976 | qc->flags |= ATA_QCFLAG_FAILED; |
| 977 | ata_eh_set_pending(ap, 1); |
| 978 | |
| 979 | /* The following will fail if timeout has already expired. |
| 980 | * ata_scsi_error() takes care of such scmds on EH entry. |
| 981 | * Note that ATA_QCFLAG_FAILED is unconditionally set after |
| 982 | * this function completes. |
| 983 | */ |
| 984 | spin_lock_irqsave(q->queue_lock, flags); |
| 985 | blk_abort_request(qc->scsicmd->request); |
| 986 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 987 | } |
| 988 | |
| 989 | /** |
| 990 | * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine |
| 991 | * @ap: ATA port to schedule EH for |
| 992 | * |
| 993 | * LOCKING: inherited from ata_port_schedule_eh |
| 994 | * spin_lock_irqsave(host lock) |
| 995 | */ |
| 996 | void ata_std_sched_eh(struct ata_port *ap) |
| 997 | { |
| 998 | WARN_ON(!ap->ops->error_handler); |
| 999 | |
| 1000 | if (ap->pflags & ATA_PFLAG_INITIALIZING) |
| 1001 | return; |
| 1002 | |
| 1003 | ata_eh_set_pending(ap, 1); |
| 1004 | scsi_schedule_eh(ap->scsi_host); |
| 1005 | |
| 1006 | DPRINTK("port EH scheduled\n"); |
| 1007 | } |
| 1008 | EXPORT_SYMBOL_GPL(ata_std_sched_eh); |
| 1009 | |
| 1010 | /** |
| 1011 | * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine |
| 1012 | * @ap: ATA port to end EH for |
| 1013 | * |
| 1014 | * In the libata object model there is a 1:1 mapping of ata_port to |
| 1015 | * shost, so host fields can be directly manipulated under ap->lock, in |
| 1016 | * the libsas case we need to hold a lock at the ha->level to coordinate |
| 1017 | * these events. |
| 1018 | * |
| 1019 | * LOCKING: |
| 1020 | * spin_lock_irqsave(host lock) |
| 1021 | */ |
| 1022 | void ata_std_end_eh(struct ata_port *ap) |
| 1023 | { |
| 1024 | struct Scsi_Host *host = ap->scsi_host; |
| 1025 | |
| 1026 | host->host_eh_scheduled = 0; |
| 1027 | } |
| 1028 | EXPORT_SYMBOL(ata_std_end_eh); |
| 1029 | |
| 1030 | |
| 1031 | /** |
| 1032 | * ata_port_schedule_eh - schedule error handling without a qc |
| 1033 | * @ap: ATA port to schedule EH for |
| 1034 | * |
| 1035 | * Schedule error handling for @ap. EH will kick in as soon as |
| 1036 | * all commands are drained. |
| 1037 | * |
| 1038 | * LOCKING: |
| 1039 | * spin_lock_irqsave(host lock) |
| 1040 | */ |
| 1041 | void ata_port_schedule_eh(struct ata_port *ap) |
| 1042 | { |
| 1043 | /* see: ata_std_sched_eh, unless you know better */ |
| 1044 | ap->ops->sched_eh(ap); |
| 1045 | } |
| 1046 | |
| 1047 | static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link) |
| 1048 | { |
| 1049 | int tag, nr_aborted = 0; |
| 1050 | |
| 1051 | WARN_ON(!ap->ops->error_handler); |
| 1052 | |
| 1053 | /* we're gonna abort all commands, no need for fast drain */ |
| 1054 | ata_eh_set_pending(ap, 0); |
| 1055 | |
| 1056 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 1057 | struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag); |
| 1058 | |
| 1059 | if (qc && (!link || qc->dev->link == link)) { |
| 1060 | qc->flags |= ATA_QCFLAG_FAILED; |
| 1061 | ata_qc_complete(qc); |
| 1062 | nr_aborted++; |
| 1063 | } |
| 1064 | } |
| 1065 | |
| 1066 | if (!nr_aborted) |
| 1067 | ata_port_schedule_eh(ap); |
| 1068 | |
| 1069 | return nr_aborted; |
| 1070 | } |
| 1071 | |
| 1072 | /** |
| 1073 | * ata_link_abort - abort all qc's on the link |
| 1074 | * @link: ATA link to abort qc's for |
| 1075 | * |
| 1076 | * Abort all active qc's active on @link and schedule EH. |
| 1077 | * |
| 1078 | * LOCKING: |
| 1079 | * spin_lock_irqsave(host lock) |
| 1080 | * |
| 1081 | * RETURNS: |
| 1082 | * Number of aborted qc's. |
| 1083 | */ |
| 1084 | int ata_link_abort(struct ata_link *link) |
| 1085 | { |
| 1086 | return ata_do_link_abort(link->ap, link); |
| 1087 | } |
| 1088 | |
| 1089 | /** |
| 1090 | * ata_port_abort - abort all qc's on the port |
| 1091 | * @ap: ATA port to abort qc's for |
| 1092 | * |
| 1093 | * Abort all active qc's of @ap and schedule EH. |
| 1094 | * |
| 1095 | * LOCKING: |
| 1096 | * spin_lock_irqsave(host_set lock) |
| 1097 | * |
| 1098 | * RETURNS: |
| 1099 | * Number of aborted qc's. |
| 1100 | */ |
| 1101 | int ata_port_abort(struct ata_port *ap) |
| 1102 | { |
| 1103 | return ata_do_link_abort(ap, NULL); |
| 1104 | } |
| 1105 | |
| 1106 | /** |
| 1107 | * __ata_port_freeze - freeze port |
| 1108 | * @ap: ATA port to freeze |
| 1109 | * |
| 1110 | * This function is called when HSM violation or some other |
| 1111 | * condition disrupts normal operation of the port. Frozen port |
| 1112 | * is not allowed to perform any operation until the port is |
| 1113 | * thawed, which usually follows a successful reset. |
| 1114 | * |
| 1115 | * ap->ops->freeze() callback can be used for freezing the port |
| 1116 | * hardware-wise (e.g. mask interrupt and stop DMA engine). If a |
| 1117 | * port cannot be frozen hardware-wise, the interrupt handler |
| 1118 | * must ack and clear interrupts unconditionally while the port |
| 1119 | * is frozen. |
| 1120 | * |
| 1121 | * LOCKING: |
| 1122 | * spin_lock_irqsave(host lock) |
| 1123 | */ |
| 1124 | static void __ata_port_freeze(struct ata_port *ap) |
| 1125 | { |
| 1126 | WARN_ON(!ap->ops->error_handler); |
| 1127 | |
| 1128 | if (ap->ops->freeze) |
| 1129 | ap->ops->freeze(ap); |
| 1130 | |
| 1131 | ap->pflags |= ATA_PFLAG_FROZEN; |
| 1132 | |
| 1133 | DPRINTK("ata%u port frozen\n", ap->print_id); |
| 1134 | } |
| 1135 | |
| 1136 | /** |
| 1137 | * ata_port_freeze - abort & freeze port |
| 1138 | * @ap: ATA port to freeze |
| 1139 | * |
| 1140 | * Abort and freeze @ap. The freeze operation must be called |
| 1141 | * first, because some hardware requires special operations |
| 1142 | * before the taskfile registers are accessible. |
| 1143 | * |
| 1144 | * LOCKING: |
| 1145 | * spin_lock_irqsave(host lock) |
| 1146 | * |
| 1147 | * RETURNS: |
| 1148 | * Number of aborted commands. |
| 1149 | */ |
| 1150 | int ata_port_freeze(struct ata_port *ap) |
| 1151 | { |
| 1152 | int nr_aborted; |
| 1153 | |
| 1154 | WARN_ON(!ap->ops->error_handler); |
| 1155 | |
| 1156 | __ata_port_freeze(ap); |
| 1157 | nr_aborted = ata_port_abort(ap); |
| 1158 | |
| 1159 | return nr_aborted; |
| 1160 | } |
| 1161 | |
| 1162 | /** |
| 1163 | * sata_async_notification - SATA async notification handler |
| 1164 | * @ap: ATA port where async notification is received |
| 1165 | * |
| 1166 | * Handler to be called when async notification via SDB FIS is |
| 1167 | * received. This function schedules EH if necessary. |
| 1168 | * |
| 1169 | * LOCKING: |
| 1170 | * spin_lock_irqsave(host lock) |
| 1171 | * |
| 1172 | * RETURNS: |
| 1173 | * 1 if EH is scheduled, 0 otherwise. |
| 1174 | */ |
| 1175 | int sata_async_notification(struct ata_port *ap) |
| 1176 | { |
| 1177 | u32 sntf; |
| 1178 | int rc; |
| 1179 | |
| 1180 | if (!(ap->flags & ATA_FLAG_AN)) |
| 1181 | return 0; |
| 1182 | |
| 1183 | rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf); |
| 1184 | if (rc == 0) |
| 1185 | sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf); |
| 1186 | |
| 1187 | if (!sata_pmp_attached(ap) || rc) { |
| 1188 | /* PMP is not attached or SNTF is not available */ |
| 1189 | if (!sata_pmp_attached(ap)) { |
| 1190 | /* PMP is not attached. Check whether ATAPI |
| 1191 | * AN is configured. If so, notify media |
| 1192 | * change. |
| 1193 | */ |
| 1194 | struct ata_device *dev = ap->link.device; |
| 1195 | |
| 1196 | if ((dev->class == ATA_DEV_ATAPI) && |
| 1197 | (dev->flags & ATA_DFLAG_AN)) |
| 1198 | ata_scsi_media_change_notify(dev); |
| 1199 | return 0; |
| 1200 | } else { |
| 1201 | /* PMP is attached but SNTF is not available. |
| 1202 | * ATAPI async media change notification is |
| 1203 | * not used. The PMP must be reporting PHY |
| 1204 | * status change, schedule EH. |
| 1205 | */ |
| 1206 | ata_port_schedule_eh(ap); |
| 1207 | return 1; |
| 1208 | } |
| 1209 | } else { |
| 1210 | /* PMP is attached and SNTF is available */ |
| 1211 | struct ata_link *link; |
| 1212 | |
| 1213 | /* check and notify ATAPI AN */ |
| 1214 | ata_for_each_link(link, ap, EDGE) { |
| 1215 | if (!(sntf & (1 << link->pmp))) |
| 1216 | continue; |
| 1217 | |
| 1218 | if ((link->device->class == ATA_DEV_ATAPI) && |
| 1219 | (link->device->flags & ATA_DFLAG_AN)) |
| 1220 | ata_scsi_media_change_notify(link->device); |
| 1221 | } |
| 1222 | |
| 1223 | /* If PMP is reporting that PHY status of some |
| 1224 | * downstream ports has changed, schedule EH. |
| 1225 | */ |
| 1226 | if (sntf & (1 << SATA_PMP_CTRL_PORT)) { |
| 1227 | ata_port_schedule_eh(ap); |
| 1228 | return 1; |
| 1229 | } |
| 1230 | |
| 1231 | return 0; |
| 1232 | } |
| 1233 | } |
| 1234 | |
| 1235 | /** |
| 1236 | * ata_eh_freeze_port - EH helper to freeze port |
| 1237 | * @ap: ATA port to freeze |
| 1238 | * |
| 1239 | * Freeze @ap. |
| 1240 | * |
| 1241 | * LOCKING: |
| 1242 | * None. |
| 1243 | */ |
| 1244 | void ata_eh_freeze_port(struct ata_port *ap) |
| 1245 | { |
| 1246 | unsigned long flags; |
| 1247 | |
| 1248 | if (!ap->ops->error_handler) |
| 1249 | return; |
| 1250 | |
| 1251 | spin_lock_irqsave(ap->lock, flags); |
| 1252 | __ata_port_freeze(ap); |
| 1253 | spin_unlock_irqrestore(ap->lock, flags); |
| 1254 | } |
| 1255 | |
| 1256 | /** |
| 1257 | * ata_port_thaw_port - EH helper to thaw port |
| 1258 | * @ap: ATA port to thaw |
| 1259 | * |
| 1260 | * Thaw frozen port @ap. |
| 1261 | * |
| 1262 | * LOCKING: |
| 1263 | * None. |
| 1264 | */ |
| 1265 | void ata_eh_thaw_port(struct ata_port *ap) |
| 1266 | { |
| 1267 | unsigned long flags; |
| 1268 | |
| 1269 | if (!ap->ops->error_handler) |
| 1270 | return; |
| 1271 | |
| 1272 | spin_lock_irqsave(ap->lock, flags); |
| 1273 | |
| 1274 | ap->pflags &= ~ATA_PFLAG_FROZEN; |
| 1275 | |
| 1276 | if (ap->ops->thaw) |
| 1277 | ap->ops->thaw(ap); |
| 1278 | |
| 1279 | spin_unlock_irqrestore(ap->lock, flags); |
| 1280 | |
| 1281 | DPRINTK("ata%u port thawed\n", ap->print_id); |
| 1282 | } |
| 1283 | |
| 1284 | static void ata_eh_scsidone(struct scsi_cmnd *scmd) |
| 1285 | { |
| 1286 | /* nada */ |
| 1287 | } |
| 1288 | |
| 1289 | static void __ata_eh_qc_complete(struct ata_queued_cmd *qc) |
| 1290 | { |
| 1291 | struct ata_port *ap = qc->ap; |
| 1292 | struct scsi_cmnd *scmd = qc->scsicmd; |
| 1293 | unsigned long flags; |
| 1294 | |
| 1295 | spin_lock_irqsave(ap->lock, flags); |
| 1296 | qc->scsidone = ata_eh_scsidone; |
| 1297 | __ata_qc_complete(qc); |
| 1298 | WARN_ON(ata_tag_valid(qc->tag)); |
| 1299 | spin_unlock_irqrestore(ap->lock, flags); |
| 1300 | |
| 1301 | scsi_eh_finish_cmd(scmd, &ap->eh_done_q); |
| 1302 | } |
| 1303 | |
| 1304 | /** |
| 1305 | * ata_eh_qc_complete - Complete an active ATA command from EH |
| 1306 | * @qc: Command to complete |
| 1307 | * |
| 1308 | * Indicate to the mid and upper layers that an ATA command has |
| 1309 | * completed. To be used from EH. |
| 1310 | */ |
| 1311 | void ata_eh_qc_complete(struct ata_queued_cmd *qc) |
| 1312 | { |
| 1313 | struct scsi_cmnd *scmd = qc->scsicmd; |
| 1314 | scmd->retries = scmd->allowed; |
| 1315 | __ata_eh_qc_complete(qc); |
| 1316 | } |
| 1317 | |
| 1318 | /** |
| 1319 | * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH |
| 1320 | * @qc: Command to retry |
| 1321 | * |
| 1322 | * Indicate to the mid and upper layers that an ATA command |
| 1323 | * should be retried. To be used from EH. |
| 1324 | * |
| 1325 | * SCSI midlayer limits the number of retries to scmd->allowed. |
| 1326 | * scmd->allowed is incremented for commands which get retried |
| 1327 | * due to unrelated failures (qc->err_mask is zero). |
| 1328 | */ |
| 1329 | void ata_eh_qc_retry(struct ata_queued_cmd *qc) |
| 1330 | { |
| 1331 | struct scsi_cmnd *scmd = qc->scsicmd; |
| 1332 | if (!qc->err_mask) |
| 1333 | scmd->allowed++; |
| 1334 | __ata_eh_qc_complete(qc); |
| 1335 | } |
| 1336 | |
| 1337 | /** |
| 1338 | * ata_dev_disable - disable ATA device |
| 1339 | * @dev: ATA device to disable |
| 1340 | * |
| 1341 | * Disable @dev. |
| 1342 | * |
| 1343 | * Locking: |
| 1344 | * EH context. |
| 1345 | */ |
| 1346 | void ata_dev_disable(struct ata_device *dev) |
| 1347 | { |
| 1348 | if (!ata_dev_enabled(dev)) |
| 1349 | return; |
| 1350 | |
| 1351 | if (ata_msg_drv(dev->link->ap)) |
| 1352 | ata_dev_warn(dev, "disabled\n"); |
| 1353 | ata_acpi_on_disable(dev); |
| 1354 | ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET); |
| 1355 | dev->class++; |
| 1356 | |
| 1357 | /* From now till the next successful probe, ering is used to |
| 1358 | * track probe failures. Clear accumulated device error info. |
| 1359 | */ |
| 1360 | ata_ering_clear(&dev->ering); |
| 1361 | } |
| 1362 | |
| 1363 | /** |
| 1364 | * ata_eh_detach_dev - detach ATA device |
| 1365 | * @dev: ATA device to detach |
| 1366 | * |
| 1367 | * Detach @dev. |
| 1368 | * |
| 1369 | * LOCKING: |
| 1370 | * None. |
| 1371 | */ |
| 1372 | void ata_eh_detach_dev(struct ata_device *dev) |
| 1373 | { |
| 1374 | struct ata_link *link = dev->link; |
| 1375 | struct ata_port *ap = link->ap; |
| 1376 | struct ata_eh_context *ehc = &link->eh_context; |
| 1377 | unsigned long flags; |
| 1378 | |
| 1379 | ata_dev_disable(dev); |
| 1380 | |
| 1381 | spin_lock_irqsave(ap->lock, flags); |
| 1382 | |
| 1383 | dev->flags &= ~ATA_DFLAG_DETACH; |
| 1384 | |
| 1385 | if (ata_scsi_offline_dev(dev)) { |
| 1386 | dev->flags |= ATA_DFLAG_DETACHED; |
| 1387 | ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; |
| 1388 | } |
| 1389 | |
| 1390 | /* clear per-dev EH info */ |
| 1391 | ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK); |
| 1392 | ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK); |
| 1393 | ehc->saved_xfer_mode[dev->devno] = 0; |
| 1394 | ehc->saved_ncq_enabled &= ~(1 << dev->devno); |
| 1395 | |
| 1396 | spin_unlock_irqrestore(ap->lock, flags); |
| 1397 | } |
| 1398 | |
| 1399 | /** |
| 1400 | * ata_eh_about_to_do - about to perform eh_action |
| 1401 | * @link: target ATA link |
| 1402 | * @dev: target ATA dev for per-dev action (can be NULL) |
| 1403 | * @action: action about to be performed |
| 1404 | * |
| 1405 | * Called just before performing EH actions to clear related bits |
| 1406 | * in @link->eh_info such that eh actions are not unnecessarily |
| 1407 | * repeated. |
| 1408 | * |
| 1409 | * LOCKING: |
| 1410 | * None. |
| 1411 | */ |
| 1412 | void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev, |
| 1413 | unsigned int action) |
| 1414 | { |
| 1415 | struct ata_port *ap = link->ap; |
| 1416 | struct ata_eh_info *ehi = &link->eh_info; |
| 1417 | struct ata_eh_context *ehc = &link->eh_context; |
| 1418 | unsigned long flags; |
| 1419 | |
| 1420 | spin_lock_irqsave(ap->lock, flags); |
| 1421 | |
| 1422 | ata_eh_clear_action(link, dev, ehi, action); |
| 1423 | |
| 1424 | /* About to take EH action, set RECOVERED. Ignore actions on |
| 1425 | * slave links as master will do them again. |
| 1426 | */ |
| 1427 | if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link) |
| 1428 | ap->pflags |= ATA_PFLAG_RECOVERED; |
| 1429 | |
| 1430 | spin_unlock_irqrestore(ap->lock, flags); |
| 1431 | } |
| 1432 | |
| 1433 | /** |
| 1434 | * ata_eh_done - EH action complete |
| 1435 | * @link: ATA link for which EH actions are complete |
| 1436 | * @dev: target ATA dev for per-dev action (can be NULL) |
| 1437 | * @action: action just completed |
| 1438 | * |
| 1439 | * Called right after performing EH actions to clear related bits |
| 1440 | * in @link->eh_context. |
| 1441 | * |
| 1442 | * LOCKING: |
| 1443 | * None. |
| 1444 | */ |
| 1445 | void ata_eh_done(struct ata_link *link, struct ata_device *dev, |
| 1446 | unsigned int action) |
| 1447 | { |
| 1448 | struct ata_eh_context *ehc = &link->eh_context; |
| 1449 | |
| 1450 | ata_eh_clear_action(link, dev, &ehc->i, action); |
| 1451 | } |
| 1452 | |
| 1453 | /** |
| 1454 | * ata_err_string - convert err_mask to descriptive string |
| 1455 | * @err_mask: error mask to convert to string |
| 1456 | * |
| 1457 | * Convert @err_mask to descriptive string. Errors are |
| 1458 | * prioritized according to severity and only the most severe |
| 1459 | * error is reported. |
| 1460 | * |
| 1461 | * LOCKING: |
| 1462 | * None. |
| 1463 | * |
| 1464 | * RETURNS: |
| 1465 | * Descriptive string for @err_mask |
| 1466 | */ |
| 1467 | static const char *ata_err_string(unsigned int err_mask) |
| 1468 | { |
| 1469 | if (err_mask & AC_ERR_HOST_BUS) |
| 1470 | return "host bus error"; |
| 1471 | if (err_mask & AC_ERR_ATA_BUS) |
| 1472 | return "ATA bus error"; |
| 1473 | if (err_mask & AC_ERR_TIMEOUT) |
| 1474 | return "timeout"; |
| 1475 | if (err_mask & AC_ERR_HSM) |
| 1476 | return "HSM violation"; |
| 1477 | if (err_mask & AC_ERR_SYSTEM) |
| 1478 | return "internal error"; |
| 1479 | if (err_mask & AC_ERR_MEDIA) |
| 1480 | return "media error"; |
| 1481 | if (err_mask & AC_ERR_INVALID) |
| 1482 | return "invalid argument"; |
| 1483 | if (err_mask & AC_ERR_DEV) |
| 1484 | return "device error"; |
| 1485 | return "unknown error"; |
| 1486 | } |
| 1487 | |
| 1488 | /** |
| 1489 | * ata_eh_read_log_10h - Read log page 10h for NCQ error details |
| 1490 | * @dev: Device to read log page 10h from |
| 1491 | * @tag: Resulting tag of the failed command |
| 1492 | * @tf: Resulting taskfile registers of the failed command |
| 1493 | * |
| 1494 | * Read log page 10h to obtain NCQ error details and clear error |
| 1495 | * condition. |
| 1496 | * |
| 1497 | * LOCKING: |
| 1498 | * Kernel thread context (may sleep). |
| 1499 | * |
| 1500 | * RETURNS: |
| 1501 | * 0 on success, -errno otherwise. |
| 1502 | */ |
| 1503 | static int ata_eh_read_log_10h(struct ata_device *dev, |
| 1504 | int *tag, struct ata_taskfile *tf) |
| 1505 | { |
| 1506 | u8 *buf = dev->link->ap->sector_buf; |
| 1507 | unsigned int err_mask; |
| 1508 | u8 csum; |
| 1509 | int i; |
| 1510 | |
| 1511 | err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1); |
| 1512 | if (err_mask) |
| 1513 | return -EIO; |
| 1514 | |
| 1515 | csum = 0; |
| 1516 | for (i = 0; i < ATA_SECT_SIZE; i++) |
| 1517 | csum += buf[i]; |
| 1518 | if (csum) |
| 1519 | ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n", |
| 1520 | csum); |
| 1521 | |
| 1522 | if (buf[0] & 0x80) |
| 1523 | return -ENOENT; |
| 1524 | |
| 1525 | *tag = buf[0] & 0x1f; |
| 1526 | |
| 1527 | tf->command = buf[2]; |
| 1528 | tf->feature = buf[3]; |
| 1529 | tf->lbal = buf[4]; |
| 1530 | tf->lbam = buf[5]; |
| 1531 | tf->lbah = buf[6]; |
| 1532 | tf->device = buf[7]; |
| 1533 | tf->hob_lbal = buf[8]; |
| 1534 | tf->hob_lbam = buf[9]; |
| 1535 | tf->hob_lbah = buf[10]; |
| 1536 | tf->nsect = buf[12]; |
| 1537 | tf->hob_nsect = buf[13]; |
| 1538 | if (ata_id_has_ncq_autosense(dev->id)) |
| 1539 | tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16]; |
| 1540 | |
| 1541 | return 0; |
| 1542 | } |
| 1543 | |
| 1544 | /** |
| 1545 | * atapi_eh_tur - perform ATAPI TEST_UNIT_READY |
| 1546 | * @dev: target ATAPI device |
| 1547 | * @r_sense_key: out parameter for sense_key |
| 1548 | * |
| 1549 | * Perform ATAPI TEST_UNIT_READY. |
| 1550 | * |
| 1551 | * LOCKING: |
| 1552 | * EH context (may sleep). |
| 1553 | * |
| 1554 | * RETURNS: |
| 1555 | * 0 on success, AC_ERR_* mask on failure. |
| 1556 | */ |
| 1557 | unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key) |
| 1558 | { |
| 1559 | u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 }; |
| 1560 | struct ata_taskfile tf; |
| 1561 | unsigned int err_mask; |
| 1562 | |
| 1563 | ata_tf_init(dev, &tf); |
| 1564 | |
| 1565 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; |
| 1566 | tf.command = ATA_CMD_PACKET; |
| 1567 | tf.protocol = ATAPI_PROT_NODATA; |
| 1568 | |
| 1569 | err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0); |
| 1570 | if (err_mask == AC_ERR_DEV) |
| 1571 | *r_sense_key = tf.feature >> 4; |
| 1572 | return err_mask; |
| 1573 | } |
| 1574 | |
| 1575 | /** |
| 1576 | * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT |
| 1577 | * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to |
| 1578 | * @cmd: scsi command for which the sense code should be set |
| 1579 | * |
| 1580 | * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK |
| 1581 | * SENSE. This function is an EH helper. |
| 1582 | * |
| 1583 | * LOCKING: |
| 1584 | * Kernel thread context (may sleep). |
| 1585 | */ |
| 1586 | static void ata_eh_request_sense(struct ata_queued_cmd *qc, |
| 1587 | struct scsi_cmnd *cmd) |
| 1588 | { |
| 1589 | struct ata_device *dev = qc->dev; |
| 1590 | struct ata_taskfile tf; |
| 1591 | unsigned int err_mask; |
| 1592 | |
| 1593 | if (qc->ap->pflags & ATA_PFLAG_FROZEN) { |
| 1594 | ata_dev_warn(dev, "sense data available but port frozen\n"); |
| 1595 | return; |
| 1596 | } |
| 1597 | |
| 1598 | if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID) |
| 1599 | return; |
| 1600 | |
| 1601 | if (!ata_id_sense_reporting_enabled(dev->id)) { |
| 1602 | ata_dev_warn(qc->dev, "sense data reporting disabled\n"); |
| 1603 | return; |
| 1604 | } |
| 1605 | |
| 1606 | DPRINTK("ATA request sense\n"); |
| 1607 | |
| 1608 | ata_tf_init(dev, &tf); |
| 1609 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; |
| 1610 | tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48; |
| 1611 | tf.command = ATA_CMD_REQ_SENSE_DATA; |
| 1612 | tf.protocol = ATA_PROT_NODATA; |
| 1613 | |
| 1614 | err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0); |
| 1615 | /* Ignore err_mask; ATA_ERR might be set */ |
| 1616 | if (tf.command & ATA_SENSE) { |
| 1617 | ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal); |
| 1618 | qc->flags |= ATA_QCFLAG_SENSE_VALID; |
| 1619 | } else { |
| 1620 | ata_dev_warn(dev, "request sense failed stat %02x emask %x\n", |
| 1621 | tf.command, err_mask); |
| 1622 | } |
| 1623 | } |
| 1624 | |
| 1625 | /** |
| 1626 | * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE |
| 1627 | * @dev: device to perform REQUEST_SENSE to |
| 1628 | * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long) |
| 1629 | * @dfl_sense_key: default sense key to use |
| 1630 | * |
| 1631 | * Perform ATAPI REQUEST_SENSE after the device reported CHECK |
| 1632 | * SENSE. This function is EH helper. |
| 1633 | * |
| 1634 | * LOCKING: |
| 1635 | * Kernel thread context (may sleep). |
| 1636 | * |
| 1637 | * RETURNS: |
| 1638 | * 0 on success, AC_ERR_* mask on failure |
| 1639 | */ |
| 1640 | unsigned int atapi_eh_request_sense(struct ata_device *dev, |
| 1641 | u8 *sense_buf, u8 dfl_sense_key) |
| 1642 | { |
| 1643 | u8 cdb[ATAPI_CDB_LEN] = |
| 1644 | { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 }; |
| 1645 | struct ata_port *ap = dev->link->ap; |
| 1646 | struct ata_taskfile tf; |
| 1647 | |
| 1648 | DPRINTK("ATAPI request sense\n"); |
| 1649 | |
| 1650 | memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE); |
| 1651 | |
| 1652 | /* initialize sense_buf with the error register, |
| 1653 | * for the case where they are -not- overwritten |
| 1654 | */ |
| 1655 | sense_buf[0] = 0x70; |
| 1656 | sense_buf[2] = dfl_sense_key; |
| 1657 | |
| 1658 | /* some devices time out if garbage left in tf */ |
| 1659 | ata_tf_init(dev, &tf); |
| 1660 | |
| 1661 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; |
| 1662 | tf.command = ATA_CMD_PACKET; |
| 1663 | |
| 1664 | /* is it pointless to prefer PIO for "safety reasons"? */ |
| 1665 | if (ap->flags & ATA_FLAG_PIO_DMA) { |
| 1666 | tf.protocol = ATAPI_PROT_DMA; |
| 1667 | tf.feature |= ATAPI_PKT_DMA; |
| 1668 | } else { |
| 1669 | tf.protocol = ATAPI_PROT_PIO; |
| 1670 | tf.lbam = SCSI_SENSE_BUFFERSIZE; |
| 1671 | tf.lbah = 0; |
| 1672 | } |
| 1673 | |
| 1674 | return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE, |
| 1675 | sense_buf, SCSI_SENSE_BUFFERSIZE, 0); |
| 1676 | } |
| 1677 | |
| 1678 | /** |
| 1679 | * ata_eh_analyze_serror - analyze SError for a failed port |
| 1680 | * @link: ATA link to analyze SError for |
| 1681 | * |
| 1682 | * Analyze SError if available and further determine cause of |
| 1683 | * failure. |
| 1684 | * |
| 1685 | * LOCKING: |
| 1686 | * None. |
| 1687 | */ |
| 1688 | static void ata_eh_analyze_serror(struct ata_link *link) |
| 1689 | { |
| 1690 | struct ata_eh_context *ehc = &link->eh_context; |
| 1691 | u32 serror = ehc->i.serror; |
| 1692 | unsigned int err_mask = 0, action = 0; |
| 1693 | u32 hotplug_mask; |
| 1694 | |
| 1695 | if (serror & (SERR_PERSISTENT | SERR_DATA)) { |
| 1696 | err_mask |= AC_ERR_ATA_BUS; |
| 1697 | action |= ATA_EH_RESET; |
| 1698 | } |
| 1699 | if (serror & SERR_PROTOCOL) { |
| 1700 | err_mask |= AC_ERR_HSM; |
| 1701 | action |= ATA_EH_RESET; |
| 1702 | } |
| 1703 | if (serror & SERR_INTERNAL) { |
| 1704 | err_mask |= AC_ERR_SYSTEM; |
| 1705 | action |= ATA_EH_RESET; |
| 1706 | } |
| 1707 | |
| 1708 | /* Determine whether a hotplug event has occurred. Both |
| 1709 | * SError.N/X are considered hotplug events for enabled or |
| 1710 | * host links. For disabled PMP links, only N bit is |
| 1711 | * considered as X bit is left at 1 for link plugging. |
| 1712 | */ |
| 1713 | if (link->lpm_policy > ATA_LPM_MAX_POWER) |
| 1714 | hotplug_mask = 0; /* hotplug doesn't work w/ LPM */ |
| 1715 | else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link)) |
| 1716 | hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG; |
| 1717 | else |
| 1718 | hotplug_mask = SERR_PHYRDY_CHG; |
| 1719 | |
| 1720 | if (serror & hotplug_mask) |
| 1721 | ata_ehi_hotplugged(&ehc->i); |
| 1722 | |
| 1723 | ehc->i.err_mask |= err_mask; |
| 1724 | ehc->i.action |= action; |
| 1725 | } |
| 1726 | |
| 1727 | /** |
| 1728 | * ata_eh_analyze_ncq_error - analyze NCQ error |
| 1729 | * @link: ATA link to analyze NCQ error for |
| 1730 | * |
| 1731 | * Read log page 10h, determine the offending qc and acquire |
| 1732 | * error status TF. For NCQ device errors, all LLDDs have to do |
| 1733 | * is setting AC_ERR_DEV in ehi->err_mask. This function takes |
| 1734 | * care of the rest. |
| 1735 | * |
| 1736 | * LOCKING: |
| 1737 | * Kernel thread context (may sleep). |
| 1738 | */ |
| 1739 | void ata_eh_analyze_ncq_error(struct ata_link *link) |
| 1740 | { |
| 1741 | struct ata_port *ap = link->ap; |
| 1742 | struct ata_eh_context *ehc = &link->eh_context; |
| 1743 | struct ata_device *dev = link->device; |
| 1744 | struct ata_queued_cmd *qc; |
| 1745 | struct ata_taskfile tf; |
| 1746 | int tag, rc; |
| 1747 | |
| 1748 | /* if frozen, we can't do much */ |
| 1749 | if (ap->pflags & ATA_PFLAG_FROZEN) |
| 1750 | return; |
| 1751 | |
| 1752 | /* is it NCQ device error? */ |
| 1753 | if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV)) |
| 1754 | return; |
| 1755 | |
| 1756 | /* has LLDD analyzed already? */ |
| 1757 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 1758 | qc = __ata_qc_from_tag(ap, tag); |
| 1759 | |
| 1760 | if (!(qc->flags & ATA_QCFLAG_FAILED)) |
| 1761 | continue; |
| 1762 | |
| 1763 | if (qc->err_mask) |
| 1764 | return; |
| 1765 | } |
| 1766 | |
| 1767 | /* okay, this error is ours */ |
| 1768 | memset(&tf, 0, sizeof(tf)); |
| 1769 | rc = ata_eh_read_log_10h(dev, &tag, &tf); |
| 1770 | if (rc) { |
| 1771 | ata_link_err(link, "failed to read log page 10h (errno=%d)\n", |
| 1772 | rc); |
| 1773 | return; |
| 1774 | } |
| 1775 | |
| 1776 | if (!(link->sactive & (1 << tag))) { |
| 1777 | ata_link_err(link, "log page 10h reported inactive tag %d\n", |
| 1778 | tag); |
| 1779 | return; |
| 1780 | } |
| 1781 | |
| 1782 | /* we've got the perpetrator, condemn it */ |
| 1783 | qc = __ata_qc_from_tag(ap, tag); |
| 1784 | memcpy(&qc->result_tf, &tf, sizeof(tf)); |
| 1785 | qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48; |
| 1786 | qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ; |
| 1787 | if ((qc->result_tf.command & ATA_SENSE) || qc->result_tf.auxiliary) { |
| 1788 | char sense_key, asc, ascq; |
| 1789 | |
| 1790 | sense_key = (qc->result_tf.auxiliary >> 16) & 0xff; |
| 1791 | asc = (qc->result_tf.auxiliary >> 8) & 0xff; |
| 1792 | ascq = qc->result_tf.auxiliary & 0xff; |
| 1793 | ata_scsi_set_sense(dev, qc->scsicmd, sense_key, asc, ascq); |
| 1794 | ata_scsi_set_sense_information(dev, qc->scsicmd, |
| 1795 | &qc->result_tf); |
| 1796 | qc->flags |= ATA_QCFLAG_SENSE_VALID; |
| 1797 | } |
| 1798 | |
| 1799 | ehc->i.err_mask &= ~AC_ERR_DEV; |
| 1800 | } |
| 1801 | |
| 1802 | /** |
| 1803 | * ata_eh_analyze_tf - analyze taskfile of a failed qc |
| 1804 | * @qc: qc to analyze |
| 1805 | * @tf: Taskfile registers to analyze |
| 1806 | * |
| 1807 | * Analyze taskfile of @qc and further determine cause of |
| 1808 | * failure. This function also requests ATAPI sense data if |
| 1809 | * available. |
| 1810 | * |
| 1811 | * LOCKING: |
| 1812 | * Kernel thread context (may sleep). |
| 1813 | * |
| 1814 | * RETURNS: |
| 1815 | * Determined recovery action |
| 1816 | */ |
| 1817 | static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc, |
| 1818 | const struct ata_taskfile *tf) |
| 1819 | { |
| 1820 | unsigned int tmp, action = 0; |
| 1821 | u8 stat = tf->command, err = tf->feature; |
| 1822 | |
| 1823 | if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) { |
| 1824 | qc->err_mask |= AC_ERR_HSM; |
| 1825 | return ATA_EH_RESET; |
| 1826 | } |
| 1827 | |
| 1828 | if (stat & (ATA_ERR | ATA_DF)) { |
| 1829 | qc->err_mask |= AC_ERR_DEV; |
| 1830 | /* |
| 1831 | * Sense data reporting does not work if the |
| 1832 | * device fault bit is set. |
| 1833 | */ |
| 1834 | if (stat & ATA_DF) |
| 1835 | stat &= ~ATA_SENSE; |
| 1836 | } else { |
| 1837 | return 0; |
| 1838 | } |
| 1839 | |
| 1840 | switch (qc->dev->class) { |
| 1841 | case ATA_DEV_ATA: |
| 1842 | case ATA_DEV_ZAC: |
| 1843 | if (stat & ATA_SENSE) |
| 1844 | ata_eh_request_sense(qc, qc->scsicmd); |
| 1845 | if (err & ATA_ICRC) |
| 1846 | qc->err_mask |= AC_ERR_ATA_BUS; |
| 1847 | if (err & (ATA_UNC | ATA_AMNF)) |
| 1848 | qc->err_mask |= AC_ERR_MEDIA; |
| 1849 | if (err & ATA_IDNF) |
| 1850 | qc->err_mask |= AC_ERR_INVALID; |
| 1851 | break; |
| 1852 | |
| 1853 | case ATA_DEV_ATAPI: |
| 1854 | if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) { |
| 1855 | tmp = atapi_eh_request_sense(qc->dev, |
| 1856 | qc->scsicmd->sense_buffer, |
| 1857 | qc->result_tf.feature >> 4); |
| 1858 | if (!tmp) |
| 1859 | qc->flags |= ATA_QCFLAG_SENSE_VALID; |
| 1860 | else |
| 1861 | qc->err_mask |= tmp; |
| 1862 | } |
| 1863 | } |
| 1864 | |
| 1865 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) { |
| 1866 | int ret = scsi_check_sense(qc->scsicmd); |
| 1867 | /* |
| 1868 | * SUCCESS here means that the sense code could |
| 1869 | * evaluated and should be passed to the upper layers |
| 1870 | * for correct evaluation. |
| 1871 | * FAILED means the sense code could not interpreted |
| 1872 | * and the device would need to be reset. |
| 1873 | * NEEDS_RETRY and ADD_TO_MLQUEUE means that the |
| 1874 | * command would need to be retried. |
| 1875 | */ |
| 1876 | if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) { |
| 1877 | qc->flags |= ATA_QCFLAG_RETRY; |
| 1878 | qc->err_mask |= AC_ERR_OTHER; |
| 1879 | } else if (ret != SUCCESS) { |
| 1880 | qc->err_mask |= AC_ERR_HSM; |
| 1881 | } |
| 1882 | } |
| 1883 | if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS)) |
| 1884 | action |= ATA_EH_RESET; |
| 1885 | |
| 1886 | return action; |
| 1887 | } |
| 1888 | |
| 1889 | static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask, |
| 1890 | int *xfer_ok) |
| 1891 | { |
| 1892 | int base = 0; |
| 1893 | |
| 1894 | if (!(eflags & ATA_EFLAG_DUBIOUS_XFER)) |
| 1895 | *xfer_ok = 1; |
| 1896 | |
| 1897 | if (!*xfer_ok) |
| 1898 | base = ATA_ECAT_DUBIOUS_NONE; |
| 1899 | |
| 1900 | if (err_mask & AC_ERR_ATA_BUS) |
| 1901 | return base + ATA_ECAT_ATA_BUS; |
| 1902 | |
| 1903 | if (err_mask & AC_ERR_TIMEOUT) |
| 1904 | return base + ATA_ECAT_TOUT_HSM; |
| 1905 | |
| 1906 | if (eflags & ATA_EFLAG_IS_IO) { |
| 1907 | if (err_mask & AC_ERR_HSM) |
| 1908 | return base + ATA_ECAT_TOUT_HSM; |
| 1909 | if ((err_mask & |
| 1910 | (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV) |
| 1911 | return base + ATA_ECAT_UNK_DEV; |
| 1912 | } |
| 1913 | |
| 1914 | return 0; |
| 1915 | } |
| 1916 | |
| 1917 | struct speed_down_verdict_arg { |
| 1918 | u64 since; |
| 1919 | int xfer_ok; |
| 1920 | int nr_errors[ATA_ECAT_NR]; |
| 1921 | }; |
| 1922 | |
| 1923 | static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg) |
| 1924 | { |
| 1925 | struct speed_down_verdict_arg *arg = void_arg; |
| 1926 | int cat; |
| 1927 | |
| 1928 | if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since)) |
| 1929 | return -1; |
| 1930 | |
| 1931 | cat = ata_eh_categorize_error(ent->eflags, ent->err_mask, |
| 1932 | &arg->xfer_ok); |
| 1933 | arg->nr_errors[cat]++; |
| 1934 | |
| 1935 | return 0; |
| 1936 | } |
| 1937 | |
| 1938 | /** |
| 1939 | * ata_eh_speed_down_verdict - Determine speed down verdict |
| 1940 | * @dev: Device of interest |
| 1941 | * |
| 1942 | * This function examines error ring of @dev and determines |
| 1943 | * whether NCQ needs to be turned off, transfer speed should be |
| 1944 | * stepped down, or falling back to PIO is necessary. |
| 1945 | * |
| 1946 | * ECAT_ATA_BUS : ATA_BUS error for any command |
| 1947 | * |
| 1948 | * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for |
| 1949 | * IO commands |
| 1950 | * |
| 1951 | * ECAT_UNK_DEV : Unknown DEV error for IO commands |
| 1952 | * |
| 1953 | * ECAT_DUBIOUS_* : Identical to above three but occurred while |
| 1954 | * data transfer hasn't been verified. |
| 1955 | * |
| 1956 | * Verdicts are |
| 1957 | * |
| 1958 | * NCQ_OFF : Turn off NCQ. |
| 1959 | * |
| 1960 | * SPEED_DOWN : Speed down transfer speed but don't fall back |
| 1961 | * to PIO. |
| 1962 | * |
| 1963 | * FALLBACK_TO_PIO : Fall back to PIO. |
| 1964 | * |
| 1965 | * Even if multiple verdicts are returned, only one action is |
| 1966 | * taken per error. An action triggered by non-DUBIOUS errors |
| 1967 | * clears ering, while one triggered by DUBIOUS_* errors doesn't. |
| 1968 | * This is to expedite speed down decisions right after device is |
| 1969 | * initially configured. |
| 1970 | * |
| 1971 | * The following are speed down rules. #1 and #2 deal with |
| 1972 | * DUBIOUS errors. |
| 1973 | * |
| 1974 | * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors |
| 1975 | * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO. |
| 1976 | * |
| 1977 | * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors |
| 1978 | * occurred during last 5 mins, NCQ_OFF. |
| 1979 | * |
| 1980 | * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors |
| 1981 | * occurred during last 5 mins, FALLBACK_TO_PIO |
| 1982 | * |
| 1983 | * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred |
| 1984 | * during last 10 mins, NCQ_OFF. |
| 1985 | * |
| 1986 | * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6 |
| 1987 | * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN. |
| 1988 | * |
| 1989 | * LOCKING: |
| 1990 | * Inherited from caller. |
| 1991 | * |
| 1992 | * RETURNS: |
| 1993 | * OR of ATA_EH_SPDN_* flags. |
| 1994 | */ |
| 1995 | static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev) |
| 1996 | { |
| 1997 | const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ; |
| 1998 | u64 j64 = get_jiffies_64(); |
| 1999 | struct speed_down_verdict_arg arg; |
| 2000 | unsigned int verdict = 0; |
| 2001 | |
| 2002 | /* scan past 5 mins of error history */ |
| 2003 | memset(&arg, 0, sizeof(arg)); |
| 2004 | arg.since = j64 - min(j64, j5mins); |
| 2005 | ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg); |
| 2006 | |
| 2007 | if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] + |
| 2008 | arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1) |
| 2009 | verdict |= ATA_EH_SPDN_SPEED_DOWN | |
| 2010 | ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS; |
| 2011 | |
| 2012 | if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] + |
| 2013 | arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1) |
| 2014 | verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS; |
| 2015 | |
| 2016 | if (arg.nr_errors[ATA_ECAT_ATA_BUS] + |
| 2017 | arg.nr_errors[ATA_ECAT_TOUT_HSM] + |
| 2018 | arg.nr_errors[ATA_ECAT_UNK_DEV] > 6) |
| 2019 | verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO; |
| 2020 | |
| 2021 | /* scan past 10 mins of error history */ |
| 2022 | memset(&arg, 0, sizeof(arg)); |
| 2023 | arg.since = j64 - min(j64, j10mins); |
| 2024 | ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg); |
| 2025 | |
| 2026 | if (arg.nr_errors[ATA_ECAT_TOUT_HSM] + |
| 2027 | arg.nr_errors[ATA_ECAT_UNK_DEV] > 3) |
| 2028 | verdict |= ATA_EH_SPDN_NCQ_OFF; |
| 2029 | |
| 2030 | if (arg.nr_errors[ATA_ECAT_ATA_BUS] + |
| 2031 | arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 || |
| 2032 | arg.nr_errors[ATA_ECAT_UNK_DEV] > 6) |
| 2033 | verdict |= ATA_EH_SPDN_SPEED_DOWN; |
| 2034 | |
| 2035 | return verdict; |
| 2036 | } |
| 2037 | |
| 2038 | /** |
| 2039 | * ata_eh_speed_down - record error and speed down if necessary |
| 2040 | * @dev: Failed device |
| 2041 | * @eflags: mask of ATA_EFLAG_* flags |
| 2042 | * @err_mask: err_mask of the error |
| 2043 | * |
| 2044 | * Record error and examine error history to determine whether |
| 2045 | * adjusting transmission speed is necessary. It also sets |
| 2046 | * transmission limits appropriately if such adjustment is |
| 2047 | * necessary. |
| 2048 | * |
| 2049 | * LOCKING: |
| 2050 | * Kernel thread context (may sleep). |
| 2051 | * |
| 2052 | * RETURNS: |
| 2053 | * Determined recovery action. |
| 2054 | */ |
| 2055 | static unsigned int ata_eh_speed_down(struct ata_device *dev, |
| 2056 | unsigned int eflags, unsigned int err_mask) |
| 2057 | { |
| 2058 | struct ata_link *link = ata_dev_phys_link(dev); |
| 2059 | int xfer_ok = 0; |
| 2060 | unsigned int verdict; |
| 2061 | unsigned int action = 0; |
| 2062 | |
| 2063 | /* don't bother if Cat-0 error */ |
| 2064 | if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0) |
| 2065 | return 0; |
| 2066 | |
| 2067 | /* record error and determine whether speed down is necessary */ |
| 2068 | ata_ering_record(&dev->ering, eflags, err_mask); |
| 2069 | verdict = ata_eh_speed_down_verdict(dev); |
| 2070 | |
| 2071 | /* turn off NCQ? */ |
| 2072 | if ((verdict & ATA_EH_SPDN_NCQ_OFF) && |
| 2073 | (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ | |
| 2074 | ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) { |
| 2075 | dev->flags |= ATA_DFLAG_NCQ_OFF; |
| 2076 | ata_dev_warn(dev, "NCQ disabled due to excessive errors\n"); |
| 2077 | goto done; |
| 2078 | } |
| 2079 | |
| 2080 | /* speed down? */ |
| 2081 | if (verdict & ATA_EH_SPDN_SPEED_DOWN) { |
| 2082 | /* speed down SATA link speed if possible */ |
| 2083 | if (sata_down_spd_limit(link, 0) == 0) { |
| 2084 | action |= ATA_EH_RESET; |
| 2085 | goto done; |
| 2086 | } |
| 2087 | |
| 2088 | /* lower transfer mode */ |
| 2089 | if (dev->spdn_cnt < 2) { |
| 2090 | static const int dma_dnxfer_sel[] = |
| 2091 | { ATA_DNXFER_DMA, ATA_DNXFER_40C }; |
| 2092 | static const int pio_dnxfer_sel[] = |
| 2093 | { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 }; |
| 2094 | int sel; |
| 2095 | |
| 2096 | if (dev->xfer_shift != ATA_SHIFT_PIO) |
| 2097 | sel = dma_dnxfer_sel[dev->spdn_cnt]; |
| 2098 | else |
| 2099 | sel = pio_dnxfer_sel[dev->spdn_cnt]; |
| 2100 | |
| 2101 | dev->spdn_cnt++; |
| 2102 | |
| 2103 | if (ata_down_xfermask_limit(dev, sel) == 0) { |
| 2104 | action |= ATA_EH_RESET; |
| 2105 | goto done; |
| 2106 | } |
| 2107 | } |
| 2108 | } |
| 2109 | |
| 2110 | /* Fall back to PIO? Slowing down to PIO is meaningless for |
| 2111 | * SATA ATA devices. Consider it only for PATA and SATAPI. |
| 2112 | */ |
| 2113 | if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) && |
| 2114 | (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) && |
| 2115 | (dev->xfer_shift != ATA_SHIFT_PIO)) { |
| 2116 | if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) { |
| 2117 | dev->spdn_cnt = 0; |
| 2118 | action |= ATA_EH_RESET; |
| 2119 | goto done; |
| 2120 | } |
| 2121 | } |
| 2122 | |
| 2123 | return 0; |
| 2124 | done: |
| 2125 | /* device has been slowed down, blow error history */ |
| 2126 | if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS)) |
| 2127 | ata_ering_clear(&dev->ering); |
| 2128 | return action; |
| 2129 | } |
| 2130 | |
| 2131 | /** |
| 2132 | * ata_eh_worth_retry - analyze error and decide whether to retry |
| 2133 | * @qc: qc to possibly retry |
| 2134 | * |
| 2135 | * Look at the cause of the error and decide if a retry |
| 2136 | * might be useful or not. We don't want to retry media errors |
| 2137 | * because the drive itself has probably already taken 10-30 seconds |
| 2138 | * doing its own internal retries before reporting the failure. |
| 2139 | */ |
| 2140 | static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc) |
| 2141 | { |
| 2142 | if (qc->err_mask & AC_ERR_MEDIA) |
| 2143 | return 0; /* don't retry media errors */ |
| 2144 | if (qc->flags & ATA_QCFLAG_IO) |
| 2145 | return 1; /* otherwise retry anything from fs stack */ |
| 2146 | if (qc->err_mask & AC_ERR_INVALID) |
| 2147 | return 0; /* don't retry these */ |
| 2148 | return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */ |
| 2149 | } |
| 2150 | |
| 2151 | /** |
| 2152 | * ata_eh_link_autopsy - analyze error and determine recovery action |
| 2153 | * @link: host link to perform autopsy on |
| 2154 | * |
| 2155 | * Analyze why @link failed and determine which recovery actions |
| 2156 | * are needed. This function also sets more detailed AC_ERR_* |
| 2157 | * values and fills sense data for ATAPI CHECK SENSE. |
| 2158 | * |
| 2159 | * LOCKING: |
| 2160 | * Kernel thread context (may sleep). |
| 2161 | */ |
| 2162 | static void ata_eh_link_autopsy(struct ata_link *link) |
| 2163 | { |
| 2164 | struct ata_port *ap = link->ap; |
| 2165 | struct ata_eh_context *ehc = &link->eh_context; |
| 2166 | struct ata_device *dev; |
| 2167 | unsigned int all_err_mask = 0, eflags = 0; |
| 2168 | int tag; |
| 2169 | u32 serror; |
| 2170 | int rc; |
| 2171 | |
| 2172 | DPRINTK("ENTER\n"); |
| 2173 | |
| 2174 | if (ehc->i.flags & ATA_EHI_NO_AUTOPSY) |
| 2175 | return; |
| 2176 | |
| 2177 | /* obtain and analyze SError */ |
| 2178 | rc = sata_scr_read(link, SCR_ERROR, &serror); |
| 2179 | if (rc == 0) { |
| 2180 | ehc->i.serror |= serror; |
| 2181 | ata_eh_analyze_serror(link); |
| 2182 | } else if (rc != -EOPNOTSUPP) { |
| 2183 | /* SError read failed, force reset and probing */ |
| 2184 | ehc->i.probe_mask |= ATA_ALL_DEVICES; |
| 2185 | ehc->i.action |= ATA_EH_RESET; |
| 2186 | ehc->i.err_mask |= AC_ERR_OTHER; |
| 2187 | } |
| 2188 | |
| 2189 | /* analyze NCQ failure */ |
| 2190 | ata_eh_analyze_ncq_error(link); |
| 2191 | |
| 2192 | /* any real error trumps AC_ERR_OTHER */ |
| 2193 | if (ehc->i.err_mask & ~AC_ERR_OTHER) |
| 2194 | ehc->i.err_mask &= ~AC_ERR_OTHER; |
| 2195 | |
| 2196 | all_err_mask |= ehc->i.err_mask; |
| 2197 | |
| 2198 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 2199 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 2200 | |
| 2201 | if (!(qc->flags & ATA_QCFLAG_FAILED) || |
| 2202 | ata_dev_phys_link(qc->dev) != link) |
| 2203 | continue; |
| 2204 | |
| 2205 | /* inherit upper level err_mask */ |
| 2206 | qc->err_mask |= ehc->i.err_mask; |
| 2207 | |
| 2208 | /* analyze TF */ |
| 2209 | ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf); |
| 2210 | |
| 2211 | /* DEV errors are probably spurious in case of ATA_BUS error */ |
| 2212 | if (qc->err_mask & AC_ERR_ATA_BUS) |
| 2213 | qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA | |
| 2214 | AC_ERR_INVALID); |
| 2215 | |
| 2216 | /* any real error trumps unknown error */ |
| 2217 | if (qc->err_mask & ~AC_ERR_OTHER) |
| 2218 | qc->err_mask &= ~AC_ERR_OTHER; |
| 2219 | |
| 2220 | /* |
| 2221 | * SENSE_VALID trumps dev/unknown error and revalidation. Upper |
| 2222 | * layers will determine whether the command is worth retrying |
| 2223 | * based on the sense data and device class/type. Otherwise, |
| 2224 | * determine directly if the command is worth retrying using its |
| 2225 | * error mask and flags. |
| 2226 | */ |
| 2227 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) |
| 2228 | qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER); |
| 2229 | else if (ata_eh_worth_retry(qc)) |
| 2230 | qc->flags |= ATA_QCFLAG_RETRY; |
| 2231 | |
| 2232 | /* accumulate error info */ |
| 2233 | ehc->i.dev = qc->dev; |
| 2234 | all_err_mask |= qc->err_mask; |
| 2235 | if (qc->flags & ATA_QCFLAG_IO) |
| 2236 | eflags |= ATA_EFLAG_IS_IO; |
| 2237 | trace_ata_eh_link_autopsy_qc(qc); |
| 2238 | } |
| 2239 | |
| 2240 | /* enforce default EH actions */ |
| 2241 | if (ap->pflags & ATA_PFLAG_FROZEN || |
| 2242 | all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT)) |
| 2243 | ehc->i.action |= ATA_EH_RESET; |
| 2244 | else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) || |
| 2245 | (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV))) |
| 2246 | ehc->i.action |= ATA_EH_REVALIDATE; |
| 2247 | |
| 2248 | /* If we have offending qcs and the associated failed device, |
| 2249 | * perform per-dev EH action only on the offending device. |
| 2250 | */ |
| 2251 | if (ehc->i.dev) { |
| 2252 | ehc->i.dev_action[ehc->i.dev->devno] |= |
| 2253 | ehc->i.action & ATA_EH_PERDEV_MASK; |
| 2254 | ehc->i.action &= ~ATA_EH_PERDEV_MASK; |
| 2255 | } |
| 2256 | |
| 2257 | /* propagate timeout to host link */ |
| 2258 | if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link)) |
| 2259 | ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT; |
| 2260 | |
| 2261 | /* record error and consider speeding down */ |
| 2262 | dev = ehc->i.dev; |
| 2263 | if (!dev && ((ata_link_max_devices(link) == 1 && |
| 2264 | ata_dev_enabled(link->device)))) |
| 2265 | dev = link->device; |
| 2266 | |
| 2267 | if (dev) { |
| 2268 | if (dev->flags & ATA_DFLAG_DUBIOUS_XFER) |
| 2269 | eflags |= ATA_EFLAG_DUBIOUS_XFER; |
| 2270 | ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask); |
| 2271 | trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask); |
| 2272 | } |
| 2273 | DPRINTK("EXIT\n"); |
| 2274 | } |
| 2275 | |
| 2276 | /** |
| 2277 | * ata_eh_autopsy - analyze error and determine recovery action |
| 2278 | * @ap: host port to perform autopsy on |
| 2279 | * |
| 2280 | * Analyze all links of @ap and determine why they failed and |
| 2281 | * which recovery actions are needed. |
| 2282 | * |
| 2283 | * LOCKING: |
| 2284 | * Kernel thread context (may sleep). |
| 2285 | */ |
| 2286 | void ata_eh_autopsy(struct ata_port *ap) |
| 2287 | { |
| 2288 | struct ata_link *link; |
| 2289 | |
| 2290 | ata_for_each_link(link, ap, EDGE) |
| 2291 | ata_eh_link_autopsy(link); |
| 2292 | |
| 2293 | /* Handle the frigging slave link. Autopsy is done similarly |
| 2294 | * but actions and flags are transferred over to the master |
| 2295 | * link and handled from there. |
| 2296 | */ |
| 2297 | if (ap->slave_link) { |
| 2298 | struct ata_eh_context *mehc = &ap->link.eh_context; |
| 2299 | struct ata_eh_context *sehc = &ap->slave_link->eh_context; |
| 2300 | |
| 2301 | /* transfer control flags from master to slave */ |
| 2302 | sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK; |
| 2303 | |
| 2304 | /* perform autopsy on the slave link */ |
| 2305 | ata_eh_link_autopsy(ap->slave_link); |
| 2306 | |
| 2307 | /* transfer actions from slave to master and clear slave */ |
| 2308 | ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS); |
| 2309 | mehc->i.action |= sehc->i.action; |
| 2310 | mehc->i.dev_action[1] |= sehc->i.dev_action[1]; |
| 2311 | mehc->i.flags |= sehc->i.flags; |
| 2312 | ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS); |
| 2313 | } |
| 2314 | |
| 2315 | /* Autopsy of fanout ports can affect host link autopsy. |
| 2316 | * Perform host link autopsy last. |
| 2317 | */ |
| 2318 | if (sata_pmp_attached(ap)) |
| 2319 | ata_eh_link_autopsy(&ap->link); |
| 2320 | } |
| 2321 | |
| 2322 | /** |
| 2323 | * ata_get_cmd_descript - get description for ATA command |
| 2324 | * @command: ATA command code to get description for |
| 2325 | * |
| 2326 | * Return a textual description of the given command, or NULL if the |
| 2327 | * command is not known. |
| 2328 | * |
| 2329 | * LOCKING: |
| 2330 | * None |
| 2331 | */ |
| 2332 | const char *ata_get_cmd_descript(u8 command) |
| 2333 | { |
| 2334 | #ifdef CONFIG_ATA_VERBOSE_ERROR |
| 2335 | static const struct |
| 2336 | { |
| 2337 | u8 command; |
| 2338 | const char *text; |
| 2339 | } cmd_descr[] = { |
| 2340 | { ATA_CMD_DEV_RESET, "DEVICE RESET" }, |
| 2341 | { ATA_CMD_CHK_POWER, "CHECK POWER MODE" }, |
| 2342 | { ATA_CMD_STANDBY, "STANDBY" }, |
| 2343 | { ATA_CMD_IDLE, "IDLE" }, |
| 2344 | { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" }, |
| 2345 | { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" }, |
| 2346 | { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" }, |
| 2347 | { ATA_CMD_NOP, "NOP" }, |
| 2348 | { ATA_CMD_FLUSH, "FLUSH CACHE" }, |
| 2349 | { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" }, |
| 2350 | { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" }, |
| 2351 | { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" }, |
| 2352 | { ATA_CMD_SERVICE, "SERVICE" }, |
| 2353 | { ATA_CMD_READ, "READ DMA" }, |
| 2354 | { ATA_CMD_READ_EXT, "READ DMA EXT" }, |
| 2355 | { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" }, |
| 2356 | { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" }, |
| 2357 | { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" }, |
| 2358 | { ATA_CMD_WRITE, "WRITE DMA" }, |
| 2359 | { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" }, |
| 2360 | { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" }, |
| 2361 | { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" }, |
| 2362 | { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" }, |
| 2363 | { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" }, |
| 2364 | { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" }, |
| 2365 | { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" }, |
| 2366 | { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" }, |
| 2367 | { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" }, |
| 2368 | { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" }, |
| 2369 | { ATA_CMD_PIO_READ, "READ SECTOR(S)" }, |
| 2370 | { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" }, |
| 2371 | { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" }, |
| 2372 | { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" }, |
| 2373 | { ATA_CMD_READ_MULTI, "READ MULTIPLE" }, |
| 2374 | { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" }, |
| 2375 | { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" }, |
| 2376 | { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" }, |
| 2377 | { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" }, |
| 2378 | { ATA_CMD_SET_FEATURES, "SET FEATURES" }, |
| 2379 | { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" }, |
| 2380 | { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" }, |
| 2381 | { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" }, |
| 2382 | { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" }, |
| 2383 | { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" }, |
| 2384 | { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" }, |
| 2385 | { ATA_CMD_SLEEP, "SLEEP" }, |
| 2386 | { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" }, |
| 2387 | { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" }, |
| 2388 | { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" }, |
| 2389 | { ATA_CMD_SET_MAX, "SET MAX ADDRESS" }, |
| 2390 | { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" }, |
| 2391 | { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" }, |
| 2392 | { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" }, |
| 2393 | { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" }, |
| 2394 | { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" }, |
| 2395 | { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" }, |
| 2396 | { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" }, |
| 2397 | { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" }, |
| 2398 | { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" }, |
| 2399 | { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" }, |
| 2400 | { ATA_CMD_PMP_READ, "READ BUFFER" }, |
| 2401 | { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" }, |
| 2402 | { ATA_CMD_PMP_WRITE, "WRITE BUFFER" }, |
| 2403 | { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" }, |
| 2404 | { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" }, |
| 2405 | { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" }, |
| 2406 | { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" }, |
| 2407 | { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" }, |
| 2408 | { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" }, |
| 2409 | { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" }, |
| 2410 | { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" }, |
| 2411 | { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" }, |
| 2412 | { ATA_CMD_SMART, "SMART" }, |
| 2413 | { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" }, |
| 2414 | { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" }, |
| 2415 | { ATA_CMD_DSM, "DATA SET MANAGEMENT" }, |
| 2416 | { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" }, |
| 2417 | { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" }, |
| 2418 | { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" }, |
| 2419 | { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" }, |
| 2420 | { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" }, |
| 2421 | { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" }, |
| 2422 | { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" }, |
| 2423 | { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" }, |
| 2424 | { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" }, |
| 2425 | { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" }, |
| 2426 | { ATA_CMD_READ_LONG, "READ LONG (with retries)" }, |
| 2427 | { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" }, |
| 2428 | { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" }, |
| 2429 | { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" }, |
| 2430 | { ATA_CMD_RESTORE, "RECALIBRATE" }, |
| 2431 | { 0, NULL } /* terminate list */ |
| 2432 | }; |
| 2433 | |
| 2434 | unsigned int i; |
| 2435 | for (i = 0; cmd_descr[i].text; i++) |
| 2436 | if (cmd_descr[i].command == command) |
| 2437 | return cmd_descr[i].text; |
| 2438 | #endif |
| 2439 | |
| 2440 | return NULL; |
| 2441 | } |
| 2442 | EXPORT_SYMBOL_GPL(ata_get_cmd_descript); |
| 2443 | |
| 2444 | /** |
| 2445 | * ata_eh_link_report - report error handling to user |
| 2446 | * @link: ATA link EH is going on |
| 2447 | * |
| 2448 | * Report EH to user. |
| 2449 | * |
| 2450 | * LOCKING: |
| 2451 | * None. |
| 2452 | */ |
| 2453 | static void ata_eh_link_report(struct ata_link *link) |
| 2454 | { |
| 2455 | struct ata_port *ap = link->ap; |
| 2456 | struct ata_eh_context *ehc = &link->eh_context; |
| 2457 | const char *frozen, *desc; |
| 2458 | char tries_buf[6] = ""; |
| 2459 | int tag, nr_failed = 0; |
| 2460 | |
| 2461 | if (ehc->i.flags & ATA_EHI_QUIET) |
| 2462 | return; |
| 2463 | |
| 2464 | desc = NULL; |
| 2465 | if (ehc->i.desc[0] != '\0') |
| 2466 | desc = ehc->i.desc; |
| 2467 | |
| 2468 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 2469 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 2470 | |
| 2471 | if (!(qc->flags & ATA_QCFLAG_FAILED) || |
| 2472 | ata_dev_phys_link(qc->dev) != link || |
| 2473 | ((qc->flags & ATA_QCFLAG_QUIET) && |
| 2474 | qc->err_mask == AC_ERR_DEV)) |
| 2475 | continue; |
| 2476 | if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask) |
| 2477 | continue; |
| 2478 | |
| 2479 | nr_failed++; |
| 2480 | } |
| 2481 | |
| 2482 | if (!nr_failed && !ehc->i.err_mask) |
| 2483 | return; |
| 2484 | |
| 2485 | frozen = ""; |
| 2486 | if (ap->pflags & ATA_PFLAG_FROZEN) |
| 2487 | frozen = " frozen"; |
| 2488 | |
| 2489 | if (ap->eh_tries < ATA_EH_MAX_TRIES) |
| 2490 | snprintf(tries_buf, sizeof(tries_buf), " t%d", |
| 2491 | ap->eh_tries); |
| 2492 | |
| 2493 | if (ehc->i.dev) { |
| 2494 | ata_dev_err(ehc->i.dev, "exception Emask 0x%x " |
| 2495 | "SAct 0x%x SErr 0x%x action 0x%x%s%s\n", |
| 2496 | ehc->i.err_mask, link->sactive, ehc->i.serror, |
| 2497 | ehc->i.action, frozen, tries_buf); |
| 2498 | if (desc) |
| 2499 | ata_dev_err(ehc->i.dev, "%s\n", desc); |
| 2500 | } else { |
| 2501 | ata_link_err(link, "exception Emask 0x%x " |
| 2502 | "SAct 0x%x SErr 0x%x action 0x%x%s%s\n", |
| 2503 | ehc->i.err_mask, link->sactive, ehc->i.serror, |
| 2504 | ehc->i.action, frozen, tries_buf); |
| 2505 | if (desc) |
| 2506 | ata_link_err(link, "%s\n", desc); |
| 2507 | } |
| 2508 | |
| 2509 | #ifdef CONFIG_ATA_VERBOSE_ERROR |
| 2510 | if (ehc->i.serror) |
| 2511 | ata_link_err(link, |
| 2512 | "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n", |
| 2513 | ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "", |
| 2514 | ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "", |
| 2515 | ehc->i.serror & SERR_DATA ? "UnrecovData " : "", |
| 2516 | ehc->i.serror & SERR_PERSISTENT ? "Persist " : "", |
| 2517 | ehc->i.serror & SERR_PROTOCOL ? "Proto " : "", |
| 2518 | ehc->i.serror & SERR_INTERNAL ? "HostInt " : "", |
| 2519 | ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "", |
| 2520 | ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "", |
| 2521 | ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "", |
| 2522 | ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "", |
| 2523 | ehc->i.serror & SERR_DISPARITY ? "Dispar " : "", |
| 2524 | ehc->i.serror & SERR_CRC ? "BadCRC " : "", |
| 2525 | ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "", |
| 2526 | ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "", |
| 2527 | ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "", |
| 2528 | ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "", |
| 2529 | ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : ""); |
| 2530 | #endif |
| 2531 | |
| 2532 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 2533 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 2534 | struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf; |
| 2535 | char data_buf[20] = ""; |
| 2536 | char cdb_buf[70] = ""; |
| 2537 | |
| 2538 | if (!(qc->flags & ATA_QCFLAG_FAILED) || |
| 2539 | ata_dev_phys_link(qc->dev) != link || !qc->err_mask) |
| 2540 | continue; |
| 2541 | |
| 2542 | if (qc->dma_dir != DMA_NONE) { |
| 2543 | static const char *dma_str[] = { |
| 2544 | [DMA_BIDIRECTIONAL] = "bidi", |
| 2545 | [DMA_TO_DEVICE] = "out", |
| 2546 | [DMA_FROM_DEVICE] = "in", |
| 2547 | }; |
| 2548 | const char *prot_str = NULL; |
| 2549 | |
| 2550 | switch (qc->tf.protocol) { |
| 2551 | case ATA_PROT_UNKNOWN: |
| 2552 | prot_str = "unknown"; |
| 2553 | break; |
| 2554 | case ATA_PROT_NODATA: |
| 2555 | prot_str = "nodata"; |
| 2556 | break; |
| 2557 | case ATA_PROT_PIO: |
| 2558 | prot_str = "pio"; |
| 2559 | break; |
| 2560 | case ATA_PROT_DMA: |
| 2561 | prot_str = "dma"; |
| 2562 | break; |
| 2563 | case ATA_PROT_NCQ: |
| 2564 | prot_str = "ncq dma"; |
| 2565 | break; |
| 2566 | case ATA_PROT_NCQ_NODATA: |
| 2567 | prot_str = "ncq nodata"; |
| 2568 | break; |
| 2569 | case ATAPI_PROT_NODATA: |
| 2570 | prot_str = "nodata"; |
| 2571 | break; |
| 2572 | case ATAPI_PROT_PIO: |
| 2573 | prot_str = "pio"; |
| 2574 | break; |
| 2575 | case ATAPI_PROT_DMA: |
| 2576 | prot_str = "dma"; |
| 2577 | break; |
| 2578 | } |
| 2579 | snprintf(data_buf, sizeof(data_buf), " %s %u %s", |
| 2580 | prot_str, qc->nbytes, dma_str[qc->dma_dir]); |
| 2581 | } |
| 2582 | |
| 2583 | if (ata_is_atapi(qc->tf.protocol)) { |
| 2584 | const u8 *cdb = qc->cdb; |
| 2585 | size_t cdb_len = qc->dev->cdb_len; |
| 2586 | |
| 2587 | if (qc->scsicmd) { |
| 2588 | cdb = qc->scsicmd->cmnd; |
| 2589 | cdb_len = qc->scsicmd->cmd_len; |
| 2590 | } |
| 2591 | __scsi_format_command(cdb_buf, sizeof(cdb_buf), |
| 2592 | cdb, cdb_len); |
| 2593 | } else { |
| 2594 | const char *descr = ata_get_cmd_descript(cmd->command); |
| 2595 | if (descr) |
| 2596 | ata_dev_err(qc->dev, "failed command: %s\n", |
| 2597 | descr); |
| 2598 | } |
| 2599 | |
| 2600 | ata_dev_err(qc->dev, |
| 2601 | "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " |
| 2602 | "tag %d%s\n %s" |
| 2603 | "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " |
| 2604 | "Emask 0x%x (%s)%s\n", |
| 2605 | cmd->command, cmd->feature, cmd->nsect, |
| 2606 | cmd->lbal, cmd->lbam, cmd->lbah, |
| 2607 | cmd->hob_feature, cmd->hob_nsect, |
| 2608 | cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah, |
| 2609 | cmd->device, qc->tag, data_buf, cdb_buf, |
| 2610 | res->command, res->feature, res->nsect, |
| 2611 | res->lbal, res->lbam, res->lbah, |
| 2612 | res->hob_feature, res->hob_nsect, |
| 2613 | res->hob_lbal, res->hob_lbam, res->hob_lbah, |
| 2614 | res->device, qc->err_mask, ata_err_string(qc->err_mask), |
| 2615 | qc->err_mask & AC_ERR_NCQ ? " <F>" : ""); |
| 2616 | |
| 2617 | #ifdef CONFIG_ATA_VERBOSE_ERROR |
| 2618 | if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ | |
| 2619 | ATA_SENSE | ATA_ERR)) { |
| 2620 | if (res->command & ATA_BUSY) |
| 2621 | ata_dev_err(qc->dev, "status: { Busy }\n"); |
| 2622 | else |
| 2623 | ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n", |
| 2624 | res->command & ATA_DRDY ? "DRDY " : "", |
| 2625 | res->command & ATA_DF ? "DF " : "", |
| 2626 | res->command & ATA_DRQ ? "DRQ " : "", |
| 2627 | res->command & ATA_SENSE ? "SENSE " : "", |
| 2628 | res->command & ATA_ERR ? "ERR " : ""); |
| 2629 | } |
| 2630 | |
| 2631 | if (cmd->command != ATA_CMD_PACKET && |
| 2632 | (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF | |
| 2633 | ATA_IDNF | ATA_ABORTED))) |
| 2634 | ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n", |
| 2635 | res->feature & ATA_ICRC ? "ICRC " : "", |
| 2636 | res->feature & ATA_UNC ? "UNC " : "", |
| 2637 | res->feature & ATA_AMNF ? "AMNF " : "", |
| 2638 | res->feature & ATA_IDNF ? "IDNF " : "", |
| 2639 | res->feature & ATA_ABORTED ? "ABRT " : ""); |
| 2640 | #endif |
| 2641 | } |
| 2642 | } |
| 2643 | |
| 2644 | /** |
| 2645 | * ata_eh_report - report error handling to user |
| 2646 | * @ap: ATA port to report EH about |
| 2647 | * |
| 2648 | * Report EH to user. |
| 2649 | * |
| 2650 | * LOCKING: |
| 2651 | * None. |
| 2652 | */ |
| 2653 | void ata_eh_report(struct ata_port *ap) |
| 2654 | { |
| 2655 | struct ata_link *link; |
| 2656 | |
| 2657 | ata_for_each_link(link, ap, HOST_FIRST) |
| 2658 | ata_eh_link_report(link); |
| 2659 | } |
| 2660 | |
| 2661 | static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset, |
| 2662 | unsigned int *classes, unsigned long deadline, |
| 2663 | bool clear_classes) |
| 2664 | { |
| 2665 | struct ata_device *dev; |
| 2666 | |
| 2667 | if (clear_classes) |
| 2668 | ata_for_each_dev(dev, link, ALL) |
| 2669 | classes[dev->devno] = ATA_DEV_UNKNOWN; |
| 2670 | |
| 2671 | return reset(link, classes, deadline); |
| 2672 | } |
| 2673 | |
| 2674 | static int ata_eh_followup_srst_needed(struct ata_link *link, int rc) |
| 2675 | { |
| 2676 | if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link)) |
| 2677 | return 0; |
| 2678 | if (rc == -EAGAIN) |
| 2679 | return 1; |
| 2680 | if (sata_pmp_supported(link->ap) && ata_is_host_link(link)) |
| 2681 | return 1; |
| 2682 | return 0; |
| 2683 | } |
| 2684 | |
| 2685 | int ata_eh_reset(struct ata_link *link, int classify, |
| 2686 | ata_prereset_fn_t prereset, ata_reset_fn_t softreset, |
| 2687 | ata_reset_fn_t hardreset, ata_postreset_fn_t postreset) |
| 2688 | { |
| 2689 | struct ata_port *ap = link->ap; |
| 2690 | struct ata_link *slave = ap->slave_link; |
| 2691 | struct ata_eh_context *ehc = &link->eh_context; |
| 2692 | struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL; |
| 2693 | unsigned int *classes = ehc->classes; |
| 2694 | unsigned int lflags = link->flags; |
| 2695 | int verbose = !(ehc->i.flags & ATA_EHI_QUIET); |
| 2696 | int max_tries = 0, try = 0; |
| 2697 | struct ata_link *failed_link; |
| 2698 | struct ata_device *dev; |
| 2699 | unsigned long deadline, now; |
| 2700 | ata_reset_fn_t reset; |
| 2701 | unsigned long flags; |
| 2702 | u32 sstatus; |
| 2703 | int nr_unknown, rc; |
| 2704 | |
| 2705 | /* |
| 2706 | * Prepare to reset |
| 2707 | */ |
| 2708 | while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX) |
| 2709 | max_tries++; |
| 2710 | if (link->flags & ATA_LFLAG_RST_ONCE) |
| 2711 | max_tries = 1; |
| 2712 | if (link->flags & ATA_LFLAG_NO_HRST) |
| 2713 | hardreset = NULL; |
| 2714 | if (link->flags & ATA_LFLAG_NO_SRST) |
| 2715 | softreset = NULL; |
| 2716 | |
| 2717 | /* make sure each reset attempt is at least COOL_DOWN apart */ |
| 2718 | if (ehc->i.flags & ATA_EHI_DID_RESET) { |
| 2719 | now = jiffies; |
| 2720 | WARN_ON(time_after(ehc->last_reset, now)); |
| 2721 | deadline = ata_deadline(ehc->last_reset, |
| 2722 | ATA_EH_RESET_COOL_DOWN); |
| 2723 | if (time_before(now, deadline)) |
| 2724 | schedule_timeout_uninterruptible(deadline - now); |
| 2725 | } |
| 2726 | |
| 2727 | spin_lock_irqsave(ap->lock, flags); |
| 2728 | ap->pflags |= ATA_PFLAG_RESETTING; |
| 2729 | spin_unlock_irqrestore(ap->lock, flags); |
| 2730 | |
| 2731 | ata_eh_about_to_do(link, NULL, ATA_EH_RESET); |
| 2732 | |
| 2733 | ata_for_each_dev(dev, link, ALL) { |
| 2734 | /* If we issue an SRST then an ATA drive (not ATAPI) |
| 2735 | * may change configuration and be in PIO0 timing. If |
| 2736 | * we do a hard reset (or are coming from power on) |
| 2737 | * this is true for ATA or ATAPI. Until we've set a |
| 2738 | * suitable controller mode we should not touch the |
| 2739 | * bus as we may be talking too fast. |
| 2740 | */ |
| 2741 | dev->pio_mode = XFER_PIO_0; |
| 2742 | dev->dma_mode = 0xff; |
| 2743 | |
| 2744 | /* If the controller has a pio mode setup function |
| 2745 | * then use it to set the chipset to rights. Don't |
| 2746 | * touch the DMA setup as that will be dealt with when |
| 2747 | * configuring devices. |
| 2748 | */ |
| 2749 | if (ap->ops->set_piomode) |
| 2750 | ap->ops->set_piomode(ap, dev); |
| 2751 | } |
| 2752 | |
| 2753 | /* prefer hardreset */ |
| 2754 | reset = NULL; |
| 2755 | ehc->i.action &= ~ATA_EH_RESET; |
| 2756 | if (hardreset) { |
| 2757 | reset = hardreset; |
| 2758 | ehc->i.action |= ATA_EH_HARDRESET; |
| 2759 | } else if (softreset) { |
| 2760 | reset = softreset; |
| 2761 | ehc->i.action |= ATA_EH_SOFTRESET; |
| 2762 | } |
| 2763 | |
| 2764 | if (prereset) { |
| 2765 | unsigned long deadline = ata_deadline(jiffies, |
| 2766 | ATA_EH_PRERESET_TIMEOUT); |
| 2767 | |
| 2768 | if (slave) { |
| 2769 | sehc->i.action &= ~ATA_EH_RESET; |
| 2770 | sehc->i.action |= ehc->i.action; |
| 2771 | } |
| 2772 | |
| 2773 | rc = prereset(link, deadline); |
| 2774 | |
| 2775 | /* If present, do prereset on slave link too. Reset |
| 2776 | * is skipped iff both master and slave links report |
| 2777 | * -ENOENT or clear ATA_EH_RESET. |
| 2778 | */ |
| 2779 | if (slave && (rc == 0 || rc == -ENOENT)) { |
| 2780 | int tmp; |
| 2781 | |
| 2782 | tmp = prereset(slave, deadline); |
| 2783 | if (tmp != -ENOENT) |
| 2784 | rc = tmp; |
| 2785 | |
| 2786 | ehc->i.action |= sehc->i.action; |
| 2787 | } |
| 2788 | |
| 2789 | if (rc) { |
| 2790 | if (rc == -ENOENT) { |
| 2791 | ata_link_dbg(link, "port disabled--ignoring\n"); |
| 2792 | ehc->i.action &= ~ATA_EH_RESET; |
| 2793 | |
| 2794 | ata_for_each_dev(dev, link, ALL) |
| 2795 | classes[dev->devno] = ATA_DEV_NONE; |
| 2796 | |
| 2797 | rc = 0; |
| 2798 | } else |
| 2799 | ata_link_err(link, |
| 2800 | "prereset failed (errno=%d)\n", |
| 2801 | rc); |
| 2802 | goto out; |
| 2803 | } |
| 2804 | |
| 2805 | /* prereset() might have cleared ATA_EH_RESET. If so, |
| 2806 | * bang classes, thaw and return. |
| 2807 | */ |
| 2808 | if (reset && !(ehc->i.action & ATA_EH_RESET)) { |
| 2809 | ata_for_each_dev(dev, link, ALL) |
| 2810 | classes[dev->devno] = ATA_DEV_NONE; |
| 2811 | if ((ap->pflags & ATA_PFLAG_FROZEN) && |
| 2812 | ata_is_host_link(link)) |
| 2813 | ata_eh_thaw_port(ap); |
| 2814 | rc = 0; |
| 2815 | goto out; |
| 2816 | } |
| 2817 | } |
| 2818 | |
| 2819 | retry: |
| 2820 | /* |
| 2821 | * Perform reset |
| 2822 | */ |
| 2823 | if (ata_is_host_link(link)) |
| 2824 | ata_eh_freeze_port(ap); |
| 2825 | |
| 2826 | deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]); |
| 2827 | |
| 2828 | if (reset) { |
| 2829 | if (verbose) |
| 2830 | ata_link_info(link, "%s resetting link\n", |
| 2831 | reset == softreset ? "soft" : "hard"); |
| 2832 | |
| 2833 | /* mark that this EH session started with reset */ |
| 2834 | ehc->last_reset = jiffies; |
| 2835 | if (reset == hardreset) |
| 2836 | ehc->i.flags |= ATA_EHI_DID_HARDRESET; |
| 2837 | else |
| 2838 | ehc->i.flags |= ATA_EHI_DID_SOFTRESET; |
| 2839 | |
| 2840 | rc = ata_do_reset(link, reset, classes, deadline, true); |
| 2841 | if (rc && rc != -EAGAIN) { |
| 2842 | failed_link = link; |
| 2843 | goto fail; |
| 2844 | } |
| 2845 | |
| 2846 | /* hardreset slave link if existent */ |
| 2847 | if (slave && reset == hardreset) { |
| 2848 | int tmp; |
| 2849 | |
| 2850 | if (verbose) |
| 2851 | ata_link_info(slave, "hard resetting link\n"); |
| 2852 | |
| 2853 | ata_eh_about_to_do(slave, NULL, ATA_EH_RESET); |
| 2854 | tmp = ata_do_reset(slave, reset, classes, deadline, |
| 2855 | false); |
| 2856 | switch (tmp) { |
| 2857 | case -EAGAIN: |
| 2858 | rc = -EAGAIN; |
| 2859 | case 0: |
| 2860 | break; |
| 2861 | default: |
| 2862 | failed_link = slave; |
| 2863 | rc = tmp; |
| 2864 | goto fail; |
| 2865 | } |
| 2866 | } |
| 2867 | |
| 2868 | /* perform follow-up SRST if necessary */ |
| 2869 | if (reset == hardreset && |
| 2870 | ata_eh_followup_srst_needed(link, rc)) { |
| 2871 | reset = softreset; |
| 2872 | |
| 2873 | if (!reset) { |
| 2874 | ata_link_err(link, |
| 2875 | "follow-up softreset required but no softreset available\n"); |
| 2876 | failed_link = link; |
| 2877 | rc = -EINVAL; |
| 2878 | goto fail; |
| 2879 | } |
| 2880 | |
| 2881 | ata_eh_about_to_do(link, NULL, ATA_EH_RESET); |
| 2882 | rc = ata_do_reset(link, reset, classes, deadline, true); |
| 2883 | if (rc) { |
| 2884 | failed_link = link; |
| 2885 | goto fail; |
| 2886 | } |
| 2887 | } |
| 2888 | } else { |
| 2889 | if (verbose) |
| 2890 | ata_link_info(link, |
| 2891 | "no reset method available, skipping reset\n"); |
| 2892 | if (!(lflags & ATA_LFLAG_ASSUME_CLASS)) |
| 2893 | lflags |= ATA_LFLAG_ASSUME_ATA; |
| 2894 | } |
| 2895 | |
| 2896 | /* |
| 2897 | * Post-reset processing |
| 2898 | */ |
| 2899 | ata_for_each_dev(dev, link, ALL) { |
| 2900 | /* After the reset, the device state is PIO 0 and the |
| 2901 | * controller state is undefined. Reset also wakes up |
| 2902 | * drives from sleeping mode. |
| 2903 | */ |
| 2904 | dev->pio_mode = XFER_PIO_0; |
| 2905 | dev->flags &= ~ATA_DFLAG_SLEEPING; |
| 2906 | |
| 2907 | if (ata_phys_link_offline(ata_dev_phys_link(dev))) |
| 2908 | continue; |
| 2909 | |
| 2910 | /* apply class override */ |
| 2911 | if (lflags & ATA_LFLAG_ASSUME_ATA) |
| 2912 | classes[dev->devno] = ATA_DEV_ATA; |
| 2913 | else if (lflags & ATA_LFLAG_ASSUME_SEMB) |
| 2914 | classes[dev->devno] = ATA_DEV_SEMB_UNSUP; |
| 2915 | } |
| 2916 | |
| 2917 | /* record current link speed */ |
| 2918 | if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0) |
| 2919 | link->sata_spd = (sstatus >> 4) & 0xf; |
| 2920 | if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0) |
| 2921 | slave->sata_spd = (sstatus >> 4) & 0xf; |
| 2922 | |
| 2923 | /* thaw the port */ |
| 2924 | if (ata_is_host_link(link)) |
| 2925 | ata_eh_thaw_port(ap); |
| 2926 | |
| 2927 | /* postreset() should clear hardware SError. Although SError |
| 2928 | * is cleared during link resume, clearing SError here is |
| 2929 | * necessary as some PHYs raise hotplug events after SRST. |
| 2930 | * This introduces race condition where hotplug occurs between |
| 2931 | * reset and here. This race is mediated by cross checking |
| 2932 | * link onlineness and classification result later. |
| 2933 | */ |
| 2934 | if (postreset) { |
| 2935 | postreset(link, classes); |
| 2936 | if (slave) |
| 2937 | postreset(slave, classes); |
| 2938 | } |
| 2939 | |
| 2940 | /* |
| 2941 | * Some controllers can't be frozen very well and may set spurious |
| 2942 | * error conditions during reset. Clear accumulated error |
| 2943 | * information and re-thaw the port if frozen. As reset is the |
| 2944 | * final recovery action and we cross check link onlineness against |
| 2945 | * device classification later, no hotplug event is lost by this. |
| 2946 | */ |
| 2947 | spin_lock_irqsave(link->ap->lock, flags); |
| 2948 | memset(&link->eh_info, 0, sizeof(link->eh_info)); |
| 2949 | if (slave) |
| 2950 | memset(&slave->eh_info, 0, sizeof(link->eh_info)); |
| 2951 | ap->pflags &= ~ATA_PFLAG_EH_PENDING; |
| 2952 | spin_unlock_irqrestore(link->ap->lock, flags); |
| 2953 | |
| 2954 | if (ap->pflags & ATA_PFLAG_FROZEN) |
| 2955 | ata_eh_thaw_port(ap); |
| 2956 | |
| 2957 | /* |
| 2958 | * Make sure onlineness and classification result correspond. |
| 2959 | * Hotplug could have happened during reset and some |
| 2960 | * controllers fail to wait while a drive is spinning up after |
| 2961 | * being hotplugged causing misdetection. By cross checking |
| 2962 | * link on/offlineness and classification result, those |
| 2963 | * conditions can be reliably detected and retried. |
| 2964 | */ |
| 2965 | nr_unknown = 0; |
| 2966 | ata_for_each_dev(dev, link, ALL) { |
| 2967 | if (ata_phys_link_online(ata_dev_phys_link(dev))) { |
| 2968 | if (classes[dev->devno] == ATA_DEV_UNKNOWN) { |
| 2969 | ata_dev_dbg(dev, "link online but device misclassified\n"); |
| 2970 | classes[dev->devno] = ATA_DEV_NONE; |
| 2971 | nr_unknown++; |
| 2972 | } |
| 2973 | } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) { |
| 2974 | if (ata_class_enabled(classes[dev->devno])) |
| 2975 | ata_dev_dbg(dev, |
| 2976 | "link offline, clearing class %d to NONE\n", |
| 2977 | classes[dev->devno]); |
| 2978 | classes[dev->devno] = ATA_DEV_NONE; |
| 2979 | } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) { |
| 2980 | ata_dev_dbg(dev, |
| 2981 | "link status unknown, clearing UNKNOWN to NONE\n"); |
| 2982 | classes[dev->devno] = ATA_DEV_NONE; |
| 2983 | } |
| 2984 | } |
| 2985 | |
| 2986 | if (classify && nr_unknown) { |
| 2987 | if (try < max_tries) { |
| 2988 | ata_link_warn(link, |
| 2989 | "link online but %d devices misclassified, retrying\n", |
| 2990 | nr_unknown); |
| 2991 | failed_link = link; |
| 2992 | rc = -EAGAIN; |
| 2993 | goto fail; |
| 2994 | } |
| 2995 | ata_link_warn(link, |
| 2996 | "link online but %d devices misclassified, " |
| 2997 | "device detection might fail\n", nr_unknown); |
| 2998 | } |
| 2999 | |
| 3000 | /* reset successful, schedule revalidation */ |
| 3001 | ata_eh_done(link, NULL, ATA_EH_RESET); |
| 3002 | if (slave) |
| 3003 | ata_eh_done(slave, NULL, ATA_EH_RESET); |
| 3004 | ehc->last_reset = jiffies; /* update to completion time */ |
| 3005 | ehc->i.action |= ATA_EH_REVALIDATE; |
| 3006 | link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */ |
| 3007 | |
| 3008 | rc = 0; |
| 3009 | out: |
| 3010 | /* clear hotplug flag */ |
| 3011 | ehc->i.flags &= ~ATA_EHI_HOTPLUGGED; |
| 3012 | if (slave) |
| 3013 | sehc->i.flags &= ~ATA_EHI_HOTPLUGGED; |
| 3014 | |
| 3015 | spin_lock_irqsave(ap->lock, flags); |
| 3016 | ap->pflags &= ~ATA_PFLAG_RESETTING; |
| 3017 | spin_unlock_irqrestore(ap->lock, flags); |
| 3018 | |
| 3019 | return rc; |
| 3020 | |
| 3021 | fail: |
| 3022 | /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */ |
| 3023 | if (!ata_is_host_link(link) && |
| 3024 | sata_scr_read(link, SCR_STATUS, &sstatus)) |
| 3025 | rc = -ERESTART; |
| 3026 | |
| 3027 | if (try >= max_tries) { |
| 3028 | /* |
| 3029 | * Thaw host port even if reset failed, so that the port |
| 3030 | * can be retried on the next phy event. This risks |
| 3031 | * repeated EH runs but seems to be a better tradeoff than |
| 3032 | * shutting down a port after a botched hotplug attempt. |
| 3033 | */ |
| 3034 | if (ata_is_host_link(link)) |
| 3035 | ata_eh_thaw_port(ap); |
| 3036 | goto out; |
| 3037 | } |
| 3038 | |
| 3039 | now = jiffies; |
| 3040 | if (time_before(now, deadline)) { |
| 3041 | unsigned long delta = deadline - now; |
| 3042 | |
| 3043 | ata_link_warn(failed_link, |
| 3044 | "reset failed (errno=%d), retrying in %u secs\n", |
| 3045 | rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000)); |
| 3046 | |
| 3047 | ata_eh_release(ap); |
| 3048 | while (delta) |
| 3049 | delta = schedule_timeout_uninterruptible(delta); |
| 3050 | ata_eh_acquire(ap); |
| 3051 | } |
| 3052 | |
| 3053 | /* |
| 3054 | * While disks spinup behind PMP, some controllers fail sending SRST. |
| 3055 | * They need to be reset - as well as the PMP - before retrying. |
| 3056 | */ |
| 3057 | if (rc == -ERESTART) { |
| 3058 | if (ata_is_host_link(link)) |
| 3059 | ata_eh_thaw_port(ap); |
| 3060 | goto out; |
| 3061 | } |
| 3062 | |
| 3063 | if (try == max_tries - 1) { |
| 3064 | sata_down_spd_limit(link, 0); |
| 3065 | if (slave) |
| 3066 | sata_down_spd_limit(slave, 0); |
| 3067 | } else if (rc == -EPIPE) |
| 3068 | sata_down_spd_limit(failed_link, 0); |
| 3069 | |
| 3070 | if (hardreset) |
| 3071 | reset = hardreset; |
| 3072 | goto retry; |
| 3073 | } |
| 3074 | |
| 3075 | static inline void ata_eh_pull_park_action(struct ata_port *ap) |
| 3076 | { |
| 3077 | struct ata_link *link; |
| 3078 | struct ata_device *dev; |
| 3079 | unsigned long flags; |
| 3080 | |
| 3081 | /* |
| 3082 | * This function can be thought of as an extended version of |
| 3083 | * ata_eh_about_to_do() specially crafted to accommodate the |
| 3084 | * requirements of ATA_EH_PARK handling. Since the EH thread |
| 3085 | * does not leave the do {} while () loop in ata_eh_recover as |
| 3086 | * long as the timeout for a park request to *one* device on |
| 3087 | * the port has not expired, and since we still want to pick |
| 3088 | * up park requests to other devices on the same port or |
| 3089 | * timeout updates for the same device, we have to pull |
| 3090 | * ATA_EH_PARK actions from eh_info into eh_context.i |
| 3091 | * ourselves at the beginning of each pass over the loop. |
| 3092 | * |
| 3093 | * Additionally, all write accesses to &ap->park_req_pending |
| 3094 | * through reinit_completion() (see below) or complete_all() |
| 3095 | * (see ata_scsi_park_store()) are protected by the host lock. |
| 3096 | * As a result we have that park_req_pending.done is zero on |
| 3097 | * exit from this function, i.e. when ATA_EH_PARK actions for |
| 3098 | * *all* devices on port ap have been pulled into the |
| 3099 | * respective eh_context structs. If, and only if, |
| 3100 | * park_req_pending.done is non-zero by the time we reach |
| 3101 | * wait_for_completion_timeout(), another ATA_EH_PARK action |
| 3102 | * has been scheduled for at least one of the devices on port |
| 3103 | * ap and we have to cycle over the do {} while () loop in |
| 3104 | * ata_eh_recover() again. |
| 3105 | */ |
| 3106 | |
| 3107 | spin_lock_irqsave(ap->lock, flags); |
| 3108 | reinit_completion(&ap->park_req_pending); |
| 3109 | ata_for_each_link(link, ap, EDGE) { |
| 3110 | ata_for_each_dev(dev, link, ALL) { |
| 3111 | struct ata_eh_info *ehi = &link->eh_info; |
| 3112 | |
| 3113 | link->eh_context.i.dev_action[dev->devno] |= |
| 3114 | ehi->dev_action[dev->devno] & ATA_EH_PARK; |
| 3115 | ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK); |
| 3116 | } |
| 3117 | } |
| 3118 | spin_unlock_irqrestore(ap->lock, flags); |
| 3119 | } |
| 3120 | |
| 3121 | static void ata_eh_park_issue_cmd(struct ata_device *dev, int park) |
| 3122 | { |
| 3123 | struct ata_eh_context *ehc = &dev->link->eh_context; |
| 3124 | struct ata_taskfile tf; |
| 3125 | unsigned int err_mask; |
| 3126 | |
| 3127 | ata_tf_init(dev, &tf); |
| 3128 | if (park) { |
| 3129 | ehc->unloaded_mask |= 1 << dev->devno; |
| 3130 | tf.command = ATA_CMD_IDLEIMMEDIATE; |
| 3131 | tf.feature = 0x44; |
| 3132 | tf.lbal = 0x4c; |
| 3133 | tf.lbam = 0x4e; |
| 3134 | tf.lbah = 0x55; |
| 3135 | } else { |
| 3136 | ehc->unloaded_mask &= ~(1 << dev->devno); |
| 3137 | tf.command = ATA_CMD_CHK_POWER; |
| 3138 | } |
| 3139 | |
| 3140 | tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR; |
| 3141 | tf.protocol = ATA_PROT_NODATA; |
| 3142 | err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0); |
| 3143 | if (park && (err_mask || tf.lbal != 0xc4)) { |
| 3144 | ata_dev_err(dev, "head unload failed!\n"); |
| 3145 | ehc->unloaded_mask &= ~(1 << dev->devno); |
| 3146 | } |
| 3147 | } |
| 3148 | |
| 3149 | static int ata_eh_revalidate_and_attach(struct ata_link *link, |
| 3150 | struct ata_device **r_failed_dev) |
| 3151 | { |
| 3152 | struct ata_port *ap = link->ap; |
| 3153 | struct ata_eh_context *ehc = &link->eh_context; |
| 3154 | struct ata_device *dev; |
| 3155 | unsigned int new_mask = 0; |
| 3156 | unsigned long flags; |
| 3157 | int rc = 0; |
| 3158 | |
| 3159 | DPRINTK("ENTER\n"); |
| 3160 | |
| 3161 | /* For PATA drive side cable detection to work, IDENTIFY must |
| 3162 | * be done backwards such that PDIAG- is released by the slave |
| 3163 | * device before the master device is identified. |
| 3164 | */ |
| 3165 | ata_for_each_dev(dev, link, ALL_REVERSE) { |
| 3166 | unsigned int action = ata_eh_dev_action(dev); |
| 3167 | unsigned int readid_flags = 0; |
| 3168 | |
| 3169 | if (ehc->i.flags & ATA_EHI_DID_RESET) |
| 3170 | readid_flags |= ATA_READID_POSTRESET; |
| 3171 | |
| 3172 | if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) { |
| 3173 | WARN_ON(dev->class == ATA_DEV_PMP); |
| 3174 | |
| 3175 | if (ata_phys_link_offline(ata_dev_phys_link(dev))) { |
| 3176 | rc = -EIO; |
| 3177 | goto err; |
| 3178 | } |
| 3179 | |
| 3180 | ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE); |
| 3181 | rc = ata_dev_revalidate(dev, ehc->classes[dev->devno], |
| 3182 | readid_flags); |
| 3183 | if (rc) |
| 3184 | goto err; |
| 3185 | |
| 3186 | ata_eh_done(link, dev, ATA_EH_REVALIDATE); |
| 3187 | |
| 3188 | /* Configuration may have changed, reconfigure |
| 3189 | * transfer mode. |
| 3190 | */ |
| 3191 | ehc->i.flags |= ATA_EHI_SETMODE; |
| 3192 | |
| 3193 | /* schedule the scsi_rescan_device() here */ |
| 3194 | schedule_work(&(ap->scsi_rescan_task)); |
| 3195 | } else if (dev->class == ATA_DEV_UNKNOWN && |
| 3196 | ehc->tries[dev->devno] && |
| 3197 | ata_class_enabled(ehc->classes[dev->devno])) { |
| 3198 | /* Temporarily set dev->class, it will be |
| 3199 | * permanently set once all configurations are |
| 3200 | * complete. This is necessary because new |
| 3201 | * device configuration is done in two |
| 3202 | * separate loops. |
| 3203 | */ |
| 3204 | dev->class = ehc->classes[dev->devno]; |
| 3205 | |
| 3206 | if (dev->class == ATA_DEV_PMP) |
| 3207 | rc = sata_pmp_attach(dev); |
| 3208 | else |
| 3209 | rc = ata_dev_read_id(dev, &dev->class, |
| 3210 | readid_flags, dev->id); |
| 3211 | |
| 3212 | /* read_id might have changed class, store and reset */ |
| 3213 | ehc->classes[dev->devno] = dev->class; |
| 3214 | dev->class = ATA_DEV_UNKNOWN; |
| 3215 | |
| 3216 | switch (rc) { |
| 3217 | case 0: |
| 3218 | /* clear error info accumulated during probe */ |
| 3219 | ata_ering_clear(&dev->ering); |
| 3220 | new_mask |= 1 << dev->devno; |
| 3221 | break; |
| 3222 | case -ENOENT: |
| 3223 | /* IDENTIFY was issued to non-existent |
| 3224 | * device. No need to reset. Just |
| 3225 | * thaw and ignore the device. |
| 3226 | */ |
| 3227 | ata_eh_thaw_port(ap); |
| 3228 | break; |
| 3229 | default: |
| 3230 | goto err; |
| 3231 | } |
| 3232 | } |
| 3233 | } |
| 3234 | |
| 3235 | /* PDIAG- should have been released, ask cable type if post-reset */ |
| 3236 | if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) { |
| 3237 | if (ap->ops->cable_detect) |
| 3238 | ap->cbl = ap->ops->cable_detect(ap); |
| 3239 | ata_force_cbl(ap); |
| 3240 | } |
| 3241 | |
| 3242 | /* Configure new devices forward such that user doesn't see |
| 3243 | * device detection messages backwards. |
| 3244 | */ |
| 3245 | ata_for_each_dev(dev, link, ALL) { |
| 3246 | if (!(new_mask & (1 << dev->devno))) |
| 3247 | continue; |
| 3248 | |
| 3249 | dev->class = ehc->classes[dev->devno]; |
| 3250 | |
| 3251 | if (dev->class == ATA_DEV_PMP) |
| 3252 | continue; |
| 3253 | |
| 3254 | ehc->i.flags |= ATA_EHI_PRINTINFO; |
| 3255 | rc = ata_dev_configure(dev); |
| 3256 | ehc->i.flags &= ~ATA_EHI_PRINTINFO; |
| 3257 | if (rc) { |
| 3258 | dev->class = ATA_DEV_UNKNOWN; |
| 3259 | goto err; |
| 3260 | } |
| 3261 | |
| 3262 | spin_lock_irqsave(ap->lock, flags); |
| 3263 | ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; |
| 3264 | spin_unlock_irqrestore(ap->lock, flags); |
| 3265 | |
| 3266 | /* new device discovered, configure xfermode */ |
| 3267 | ehc->i.flags |= ATA_EHI_SETMODE; |
| 3268 | } |
| 3269 | |
| 3270 | return 0; |
| 3271 | |
| 3272 | err: |
| 3273 | *r_failed_dev = dev; |
| 3274 | DPRINTK("EXIT rc=%d\n", rc); |
| 3275 | return rc; |
| 3276 | } |
| 3277 | |
| 3278 | /** |
| 3279 | * ata_set_mode - Program timings and issue SET FEATURES - XFER |
| 3280 | * @link: link on which timings will be programmed |
| 3281 | * @r_failed_dev: out parameter for failed device |
| 3282 | * |
| 3283 | * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If |
| 3284 | * ata_set_mode() fails, pointer to the failing device is |
| 3285 | * returned in @r_failed_dev. |
| 3286 | * |
| 3287 | * LOCKING: |
| 3288 | * PCI/etc. bus probe sem. |
| 3289 | * |
| 3290 | * RETURNS: |
| 3291 | * 0 on success, negative errno otherwise |
| 3292 | */ |
| 3293 | int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev) |
| 3294 | { |
| 3295 | struct ata_port *ap = link->ap; |
| 3296 | struct ata_device *dev; |
| 3297 | int rc; |
| 3298 | |
| 3299 | /* if data transfer is verified, clear DUBIOUS_XFER on ering top */ |
| 3300 | ata_for_each_dev(dev, link, ENABLED) { |
| 3301 | if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) { |
| 3302 | struct ata_ering_entry *ent; |
| 3303 | |
| 3304 | ent = ata_ering_top(&dev->ering); |
| 3305 | if (ent) |
| 3306 | ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER; |
| 3307 | } |
| 3308 | } |
| 3309 | |
| 3310 | /* has private set_mode? */ |
| 3311 | if (ap->ops->set_mode) |
| 3312 | rc = ap->ops->set_mode(link, r_failed_dev); |
| 3313 | else |
| 3314 | rc = ata_do_set_mode(link, r_failed_dev); |
| 3315 | |
| 3316 | /* if transfer mode has changed, set DUBIOUS_XFER on device */ |
| 3317 | ata_for_each_dev(dev, link, ENABLED) { |
| 3318 | struct ata_eh_context *ehc = &link->eh_context; |
| 3319 | u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno]; |
| 3320 | u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno)); |
| 3321 | |
| 3322 | if (dev->xfer_mode != saved_xfer_mode || |
| 3323 | ata_ncq_enabled(dev) != saved_ncq) |
| 3324 | dev->flags |= ATA_DFLAG_DUBIOUS_XFER; |
| 3325 | } |
| 3326 | |
| 3327 | return rc; |
| 3328 | } |
| 3329 | |
| 3330 | /** |
| 3331 | * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset |
| 3332 | * @dev: ATAPI device to clear UA for |
| 3333 | * |
| 3334 | * Resets and other operations can make an ATAPI device raise |
| 3335 | * UNIT ATTENTION which causes the next operation to fail. This |
| 3336 | * function clears UA. |
| 3337 | * |
| 3338 | * LOCKING: |
| 3339 | * EH context (may sleep). |
| 3340 | * |
| 3341 | * RETURNS: |
| 3342 | * 0 on success, -errno on failure. |
| 3343 | */ |
| 3344 | static int atapi_eh_clear_ua(struct ata_device *dev) |
| 3345 | { |
| 3346 | int i; |
| 3347 | |
| 3348 | for (i = 0; i < ATA_EH_UA_TRIES; i++) { |
| 3349 | u8 *sense_buffer = dev->link->ap->sector_buf; |
| 3350 | u8 sense_key = 0; |
| 3351 | unsigned int err_mask; |
| 3352 | |
| 3353 | err_mask = atapi_eh_tur(dev, &sense_key); |
| 3354 | if (err_mask != 0 && err_mask != AC_ERR_DEV) { |
| 3355 | ata_dev_warn(dev, |
| 3356 | "TEST_UNIT_READY failed (err_mask=0x%x)\n", |
| 3357 | err_mask); |
| 3358 | return -EIO; |
| 3359 | } |
| 3360 | |
| 3361 | if (!err_mask || sense_key != UNIT_ATTENTION) |
| 3362 | return 0; |
| 3363 | |
| 3364 | err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key); |
| 3365 | if (err_mask) { |
| 3366 | ata_dev_warn(dev, "failed to clear " |
| 3367 | "UNIT ATTENTION (err_mask=0x%x)\n", err_mask); |
| 3368 | return -EIO; |
| 3369 | } |
| 3370 | } |
| 3371 | |
| 3372 | ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n", |
| 3373 | ATA_EH_UA_TRIES); |
| 3374 | |
| 3375 | return 0; |
| 3376 | } |
| 3377 | |
| 3378 | /** |
| 3379 | * ata_eh_maybe_retry_flush - Retry FLUSH if necessary |
| 3380 | * @dev: ATA device which may need FLUSH retry |
| 3381 | * |
| 3382 | * If @dev failed FLUSH, it needs to be reported upper layer |
| 3383 | * immediately as it means that @dev failed to remap and already |
| 3384 | * lost at least a sector and further FLUSH retrials won't make |
| 3385 | * any difference to the lost sector. However, if FLUSH failed |
| 3386 | * for other reasons, for example transmission error, FLUSH needs |
| 3387 | * to be retried. |
| 3388 | * |
| 3389 | * This function determines whether FLUSH failure retry is |
| 3390 | * necessary and performs it if so. |
| 3391 | * |
| 3392 | * RETURNS: |
| 3393 | * 0 if EH can continue, -errno if EH needs to be repeated. |
| 3394 | */ |
| 3395 | static int ata_eh_maybe_retry_flush(struct ata_device *dev) |
| 3396 | { |
| 3397 | struct ata_link *link = dev->link; |
| 3398 | struct ata_port *ap = link->ap; |
| 3399 | struct ata_queued_cmd *qc; |
| 3400 | struct ata_taskfile tf; |
| 3401 | unsigned int err_mask; |
| 3402 | int rc = 0; |
| 3403 | |
| 3404 | /* did flush fail for this device? */ |
| 3405 | if (!ata_tag_valid(link->active_tag)) |
| 3406 | return 0; |
| 3407 | |
| 3408 | qc = __ata_qc_from_tag(ap, link->active_tag); |
| 3409 | if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT && |
| 3410 | qc->tf.command != ATA_CMD_FLUSH)) |
| 3411 | return 0; |
| 3412 | |
| 3413 | /* if the device failed it, it should be reported to upper layers */ |
| 3414 | if (qc->err_mask & AC_ERR_DEV) |
| 3415 | return 0; |
| 3416 | |
| 3417 | /* flush failed for some other reason, give it another shot */ |
| 3418 | ata_tf_init(dev, &tf); |
| 3419 | |
| 3420 | tf.command = qc->tf.command; |
| 3421 | tf.flags |= ATA_TFLAG_DEVICE; |
| 3422 | tf.protocol = ATA_PROT_NODATA; |
| 3423 | |
| 3424 | ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n", |
| 3425 | tf.command, qc->err_mask); |
| 3426 | |
| 3427 | err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0); |
| 3428 | if (!err_mask) { |
| 3429 | /* |
| 3430 | * FLUSH is complete but there's no way to |
| 3431 | * successfully complete a failed command from EH. |
| 3432 | * Making sure retry is allowed at least once and |
| 3433 | * retrying it should do the trick - whatever was in |
| 3434 | * the cache is already on the platter and this won't |
| 3435 | * cause infinite loop. |
| 3436 | */ |
| 3437 | qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1); |
| 3438 | } else { |
| 3439 | ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n", |
| 3440 | err_mask); |
| 3441 | rc = -EIO; |
| 3442 | |
| 3443 | /* if device failed it, report it to upper layers */ |
| 3444 | if (err_mask & AC_ERR_DEV) { |
| 3445 | qc->err_mask |= AC_ERR_DEV; |
| 3446 | qc->result_tf = tf; |
| 3447 | if (!(ap->pflags & ATA_PFLAG_FROZEN)) |
| 3448 | rc = 0; |
| 3449 | } |
| 3450 | } |
| 3451 | return rc; |
| 3452 | } |
| 3453 | |
| 3454 | /** |
| 3455 | * ata_eh_set_lpm - configure SATA interface power management |
| 3456 | * @link: link to configure power management |
| 3457 | * @policy: the link power management policy |
| 3458 | * @r_failed_dev: out parameter for failed device |
| 3459 | * |
| 3460 | * Enable SATA Interface power management. This will enable |
| 3461 | * Device Interface Power Management (DIPM) for min_power |
| 3462 | * policy, and then call driver specific callbacks for |
| 3463 | * enabling Host Initiated Power management. |
| 3464 | * |
| 3465 | * LOCKING: |
| 3466 | * EH context. |
| 3467 | * |
| 3468 | * RETURNS: |
| 3469 | * 0 on success, -errno on failure. |
| 3470 | */ |
| 3471 | static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy, |
| 3472 | struct ata_device **r_failed_dev) |
| 3473 | { |
| 3474 | struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL; |
| 3475 | struct ata_eh_context *ehc = &link->eh_context; |
| 3476 | struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL; |
| 3477 | enum ata_lpm_policy old_policy = link->lpm_policy; |
| 3478 | bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM; |
| 3479 | unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM; |
| 3480 | unsigned int err_mask; |
| 3481 | int rc; |
| 3482 | |
| 3483 | /* if the link or host doesn't do LPM, noop */ |
| 3484 | if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm)) |
| 3485 | return 0; |
| 3486 | |
| 3487 | /* |
| 3488 | * DIPM is enabled only for MIN_POWER as some devices |
| 3489 | * misbehave when the host NACKs transition to SLUMBER. Order |
| 3490 | * device and link configurations such that the host always |
| 3491 | * allows DIPM requests. |
| 3492 | */ |
| 3493 | ata_for_each_dev(dev, link, ENABLED) { |
| 3494 | bool hipm = ata_id_has_hipm(dev->id); |
| 3495 | bool dipm = ata_id_has_dipm(dev->id) && !no_dipm; |
| 3496 | |
| 3497 | /* find the first enabled and LPM enabled devices */ |
| 3498 | if (!link_dev) |
| 3499 | link_dev = dev; |
| 3500 | |
| 3501 | if (!lpm_dev && (hipm || dipm)) |
| 3502 | lpm_dev = dev; |
| 3503 | |
| 3504 | hints &= ~ATA_LPM_EMPTY; |
| 3505 | if (!hipm) |
| 3506 | hints &= ~ATA_LPM_HIPM; |
| 3507 | |
| 3508 | /* disable DIPM before changing link config */ |
| 3509 | if (policy != ATA_LPM_MIN_POWER && dipm) { |
| 3510 | err_mask = ata_dev_set_feature(dev, |
| 3511 | SETFEATURES_SATA_DISABLE, SATA_DIPM); |
| 3512 | if (err_mask && err_mask != AC_ERR_DEV) { |
| 3513 | ata_dev_warn(dev, |
| 3514 | "failed to disable DIPM, Emask 0x%x\n", |
| 3515 | err_mask); |
| 3516 | rc = -EIO; |
| 3517 | goto fail; |
| 3518 | } |
| 3519 | } |
| 3520 | } |
| 3521 | |
| 3522 | if (ap) { |
| 3523 | rc = ap->ops->set_lpm(link, policy, hints); |
| 3524 | if (!rc && ap->slave_link) |
| 3525 | rc = ap->ops->set_lpm(ap->slave_link, policy, hints); |
| 3526 | } else |
| 3527 | rc = sata_pmp_set_lpm(link, policy, hints); |
| 3528 | |
| 3529 | /* |
| 3530 | * Attribute link config failure to the first (LPM) enabled |
| 3531 | * device on the link. |
| 3532 | */ |
| 3533 | if (rc) { |
| 3534 | if (rc == -EOPNOTSUPP) { |
| 3535 | link->flags |= ATA_LFLAG_NO_LPM; |
| 3536 | return 0; |
| 3537 | } |
| 3538 | dev = lpm_dev ? lpm_dev : link_dev; |
| 3539 | goto fail; |
| 3540 | } |
| 3541 | |
| 3542 | /* |
| 3543 | * Low level driver acked the transition. Issue DIPM command |
| 3544 | * with the new policy set. |
| 3545 | */ |
| 3546 | link->lpm_policy = policy; |
| 3547 | if (ap && ap->slave_link) |
| 3548 | ap->slave_link->lpm_policy = policy; |
| 3549 | |
| 3550 | /* host config updated, enable DIPM if transitioning to MIN_POWER */ |
| 3551 | ata_for_each_dev(dev, link, ENABLED) { |
| 3552 | if (policy == ATA_LPM_MIN_POWER && !no_dipm && |
| 3553 | ata_id_has_dipm(dev->id)) { |
| 3554 | err_mask = ata_dev_set_feature(dev, |
| 3555 | SETFEATURES_SATA_ENABLE, SATA_DIPM); |
| 3556 | if (err_mask && err_mask != AC_ERR_DEV) { |
| 3557 | ata_dev_warn(dev, |
| 3558 | "failed to enable DIPM, Emask 0x%x\n", |
| 3559 | err_mask); |
| 3560 | rc = -EIO; |
| 3561 | goto fail; |
| 3562 | } |
| 3563 | } |
| 3564 | } |
| 3565 | |
| 3566 | link->last_lpm_change = jiffies; |
| 3567 | link->flags |= ATA_LFLAG_CHANGED; |
| 3568 | |
| 3569 | return 0; |
| 3570 | |
| 3571 | fail: |
| 3572 | /* restore the old policy */ |
| 3573 | link->lpm_policy = old_policy; |
| 3574 | if (ap && ap->slave_link) |
| 3575 | ap->slave_link->lpm_policy = old_policy; |
| 3576 | |
| 3577 | /* if no device or only one more chance is left, disable LPM */ |
| 3578 | if (!dev || ehc->tries[dev->devno] <= 2) { |
| 3579 | ata_link_warn(link, "disabling LPM on the link\n"); |
| 3580 | link->flags |= ATA_LFLAG_NO_LPM; |
| 3581 | } |
| 3582 | if (r_failed_dev) |
| 3583 | *r_failed_dev = dev; |
| 3584 | return rc; |
| 3585 | } |
| 3586 | |
| 3587 | int ata_link_nr_enabled(struct ata_link *link) |
| 3588 | { |
| 3589 | struct ata_device *dev; |
| 3590 | int cnt = 0; |
| 3591 | |
| 3592 | ata_for_each_dev(dev, link, ENABLED) |
| 3593 | cnt++; |
| 3594 | return cnt; |
| 3595 | } |
| 3596 | |
| 3597 | static int ata_link_nr_vacant(struct ata_link *link) |
| 3598 | { |
| 3599 | struct ata_device *dev; |
| 3600 | int cnt = 0; |
| 3601 | |
| 3602 | ata_for_each_dev(dev, link, ALL) |
| 3603 | if (dev->class == ATA_DEV_UNKNOWN) |
| 3604 | cnt++; |
| 3605 | return cnt; |
| 3606 | } |
| 3607 | |
| 3608 | static int ata_eh_skip_recovery(struct ata_link *link) |
| 3609 | { |
| 3610 | struct ata_port *ap = link->ap; |
| 3611 | struct ata_eh_context *ehc = &link->eh_context; |
| 3612 | struct ata_device *dev; |
| 3613 | |
| 3614 | /* skip disabled links */ |
| 3615 | if (link->flags & ATA_LFLAG_DISABLED) |
| 3616 | return 1; |
| 3617 | |
| 3618 | /* skip if explicitly requested */ |
| 3619 | if (ehc->i.flags & ATA_EHI_NO_RECOVERY) |
| 3620 | return 1; |
| 3621 | |
| 3622 | /* thaw frozen port and recover failed devices */ |
| 3623 | if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link)) |
| 3624 | return 0; |
| 3625 | |
| 3626 | /* reset at least once if reset is requested */ |
| 3627 | if ((ehc->i.action & ATA_EH_RESET) && |
| 3628 | !(ehc->i.flags & ATA_EHI_DID_RESET)) |
| 3629 | return 0; |
| 3630 | |
| 3631 | /* skip if class codes for all vacant slots are ATA_DEV_NONE */ |
| 3632 | ata_for_each_dev(dev, link, ALL) { |
| 3633 | if (dev->class == ATA_DEV_UNKNOWN && |
| 3634 | ehc->classes[dev->devno] != ATA_DEV_NONE) |
| 3635 | return 0; |
| 3636 | } |
| 3637 | |
| 3638 | return 1; |
| 3639 | } |
| 3640 | |
| 3641 | static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg) |
| 3642 | { |
| 3643 | u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL); |
| 3644 | u64 now = get_jiffies_64(); |
| 3645 | int *trials = void_arg; |
| 3646 | |
| 3647 | if ((ent->eflags & ATA_EFLAG_OLD_ER) || |
| 3648 | (ent->timestamp < now - min(now, interval))) |
| 3649 | return -1; |
| 3650 | |
| 3651 | (*trials)++; |
| 3652 | return 0; |
| 3653 | } |
| 3654 | |
| 3655 | static int ata_eh_schedule_probe(struct ata_device *dev) |
| 3656 | { |
| 3657 | struct ata_eh_context *ehc = &dev->link->eh_context; |
| 3658 | struct ata_link *link = ata_dev_phys_link(dev); |
| 3659 | int trials = 0; |
| 3660 | |
| 3661 | if (!(ehc->i.probe_mask & (1 << dev->devno)) || |
| 3662 | (ehc->did_probe_mask & (1 << dev->devno))) |
| 3663 | return 0; |
| 3664 | |
| 3665 | ata_eh_detach_dev(dev); |
| 3666 | ata_dev_init(dev); |
| 3667 | ehc->did_probe_mask |= (1 << dev->devno); |
| 3668 | ehc->i.action |= ATA_EH_RESET; |
| 3669 | ehc->saved_xfer_mode[dev->devno] = 0; |
| 3670 | ehc->saved_ncq_enabled &= ~(1 << dev->devno); |
| 3671 | |
| 3672 | /* the link maybe in a deep sleep, wake it up */ |
| 3673 | if (link->lpm_policy > ATA_LPM_MAX_POWER) { |
| 3674 | if (ata_is_host_link(link)) |
| 3675 | link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER, |
| 3676 | ATA_LPM_EMPTY); |
| 3677 | else |
| 3678 | sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER, |
| 3679 | ATA_LPM_EMPTY); |
| 3680 | } |
| 3681 | |
| 3682 | /* Record and count probe trials on the ering. The specific |
| 3683 | * error mask used is irrelevant. Because a successful device |
| 3684 | * detection clears the ering, this count accumulates only if |
| 3685 | * there are consecutive failed probes. |
| 3686 | * |
| 3687 | * If the count is equal to or higher than ATA_EH_PROBE_TRIALS |
| 3688 | * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is |
| 3689 | * forced to 1.5Gbps. |
| 3690 | * |
| 3691 | * This is to work around cases where failed link speed |
| 3692 | * negotiation results in device misdetection leading to |
| 3693 | * infinite DEVXCHG or PHRDY CHG events. |
| 3694 | */ |
| 3695 | ata_ering_record(&dev->ering, 0, AC_ERR_OTHER); |
| 3696 | ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials); |
| 3697 | |
| 3698 | if (trials > ATA_EH_PROBE_TRIALS) |
| 3699 | sata_down_spd_limit(link, 1); |
| 3700 | |
| 3701 | return 1; |
| 3702 | } |
| 3703 | |
| 3704 | static int ata_eh_handle_dev_fail(struct ata_device *dev, int err) |
| 3705 | { |
| 3706 | struct ata_eh_context *ehc = &dev->link->eh_context; |
| 3707 | |
| 3708 | /* -EAGAIN from EH routine indicates retry without prejudice. |
| 3709 | * The requester is responsible for ensuring forward progress. |
| 3710 | */ |
| 3711 | if (err != -EAGAIN) |
| 3712 | ehc->tries[dev->devno]--; |
| 3713 | |
| 3714 | switch (err) { |
| 3715 | case -ENODEV: |
| 3716 | /* device missing or wrong IDENTIFY data, schedule probing */ |
| 3717 | ehc->i.probe_mask |= (1 << dev->devno); |
| 3718 | case -EINVAL: |
| 3719 | /* give it just one more chance */ |
| 3720 | ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1); |
| 3721 | case -EIO: |
| 3722 | if (ehc->tries[dev->devno] == 1) { |
| 3723 | /* This is the last chance, better to slow |
| 3724 | * down than lose it. |
| 3725 | */ |
| 3726 | sata_down_spd_limit(ata_dev_phys_link(dev), 0); |
| 3727 | if (dev->pio_mode > XFER_PIO_0) |
| 3728 | ata_down_xfermask_limit(dev, ATA_DNXFER_PIO); |
| 3729 | } |
| 3730 | } |
| 3731 | |
| 3732 | if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) { |
| 3733 | /* disable device if it has used up all its chances */ |
| 3734 | ata_dev_disable(dev); |
| 3735 | |
| 3736 | /* detach if offline */ |
| 3737 | if (ata_phys_link_offline(ata_dev_phys_link(dev))) |
| 3738 | ata_eh_detach_dev(dev); |
| 3739 | |
| 3740 | /* schedule probe if necessary */ |
| 3741 | if (ata_eh_schedule_probe(dev)) { |
| 3742 | ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; |
| 3743 | memset(ehc->cmd_timeout_idx[dev->devno], 0, |
| 3744 | sizeof(ehc->cmd_timeout_idx[dev->devno])); |
| 3745 | } |
| 3746 | |
| 3747 | return 1; |
| 3748 | } else { |
| 3749 | ehc->i.action |= ATA_EH_RESET; |
| 3750 | return 0; |
| 3751 | } |
| 3752 | } |
| 3753 | |
| 3754 | /** |
| 3755 | * ata_eh_recover - recover host port after error |
| 3756 | * @ap: host port to recover |
| 3757 | * @prereset: prereset method (can be NULL) |
| 3758 | * @softreset: softreset method (can be NULL) |
| 3759 | * @hardreset: hardreset method (can be NULL) |
| 3760 | * @postreset: postreset method (can be NULL) |
| 3761 | * @r_failed_link: out parameter for failed link |
| 3762 | * |
| 3763 | * This is the alpha and omega, eum and yang, heart and soul of |
| 3764 | * libata exception handling. On entry, actions required to |
| 3765 | * recover each link and hotplug requests are recorded in the |
| 3766 | * link's eh_context. This function executes all the operations |
| 3767 | * with appropriate retrials and fallbacks to resurrect failed |
| 3768 | * devices, detach goners and greet newcomers. |
| 3769 | * |
| 3770 | * LOCKING: |
| 3771 | * Kernel thread context (may sleep). |
| 3772 | * |
| 3773 | * RETURNS: |
| 3774 | * 0 on success, -errno on failure. |
| 3775 | */ |
| 3776 | int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset, |
| 3777 | ata_reset_fn_t softreset, ata_reset_fn_t hardreset, |
| 3778 | ata_postreset_fn_t postreset, |
| 3779 | struct ata_link **r_failed_link) |
| 3780 | { |
| 3781 | struct ata_link *link; |
| 3782 | struct ata_device *dev; |
| 3783 | int rc, nr_fails; |
| 3784 | unsigned long flags, deadline; |
| 3785 | |
| 3786 | DPRINTK("ENTER\n"); |
| 3787 | |
| 3788 | /* prep for recovery */ |
| 3789 | ata_for_each_link(link, ap, EDGE) { |
| 3790 | struct ata_eh_context *ehc = &link->eh_context; |
| 3791 | |
| 3792 | /* re-enable link? */ |
| 3793 | if (ehc->i.action & ATA_EH_ENABLE_LINK) { |
| 3794 | ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK); |
| 3795 | spin_lock_irqsave(ap->lock, flags); |
| 3796 | link->flags &= ~ATA_LFLAG_DISABLED; |
| 3797 | spin_unlock_irqrestore(ap->lock, flags); |
| 3798 | ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK); |
| 3799 | } |
| 3800 | |
| 3801 | ata_for_each_dev(dev, link, ALL) { |
| 3802 | if (link->flags & ATA_LFLAG_NO_RETRY) |
| 3803 | ehc->tries[dev->devno] = 1; |
| 3804 | else |
| 3805 | ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; |
| 3806 | |
| 3807 | /* collect port action mask recorded in dev actions */ |
| 3808 | ehc->i.action |= ehc->i.dev_action[dev->devno] & |
| 3809 | ~ATA_EH_PERDEV_MASK; |
| 3810 | ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK; |
| 3811 | |
| 3812 | /* process hotplug request */ |
| 3813 | if (dev->flags & ATA_DFLAG_DETACH) |
| 3814 | ata_eh_detach_dev(dev); |
| 3815 | |
| 3816 | /* schedule probe if necessary */ |
| 3817 | if (!ata_dev_enabled(dev)) |
| 3818 | ata_eh_schedule_probe(dev); |
| 3819 | } |
| 3820 | } |
| 3821 | |
| 3822 | retry: |
| 3823 | rc = 0; |
| 3824 | |
| 3825 | /* if UNLOADING, finish immediately */ |
| 3826 | if (ap->pflags & ATA_PFLAG_UNLOADING) |
| 3827 | goto out; |
| 3828 | |
| 3829 | /* prep for EH */ |
| 3830 | ata_for_each_link(link, ap, EDGE) { |
| 3831 | struct ata_eh_context *ehc = &link->eh_context; |
| 3832 | |
| 3833 | /* skip EH if possible. */ |
| 3834 | if (ata_eh_skip_recovery(link)) |
| 3835 | ehc->i.action = 0; |
| 3836 | |
| 3837 | ata_for_each_dev(dev, link, ALL) |
| 3838 | ehc->classes[dev->devno] = ATA_DEV_UNKNOWN; |
| 3839 | } |
| 3840 | |
| 3841 | /* reset */ |
| 3842 | ata_for_each_link(link, ap, EDGE) { |
| 3843 | struct ata_eh_context *ehc = &link->eh_context; |
| 3844 | |
| 3845 | if (!(ehc->i.action & ATA_EH_RESET)) |
| 3846 | continue; |
| 3847 | |
| 3848 | rc = ata_eh_reset(link, ata_link_nr_vacant(link), |
| 3849 | prereset, softreset, hardreset, postreset); |
| 3850 | if (rc) { |
| 3851 | ata_link_err(link, "reset failed, giving up\n"); |
| 3852 | goto out; |
| 3853 | } |
| 3854 | } |
| 3855 | |
| 3856 | do { |
| 3857 | unsigned long now; |
| 3858 | |
| 3859 | /* |
| 3860 | * clears ATA_EH_PARK in eh_info and resets |
| 3861 | * ap->park_req_pending |
| 3862 | */ |
| 3863 | ata_eh_pull_park_action(ap); |
| 3864 | |
| 3865 | deadline = jiffies; |
| 3866 | ata_for_each_link(link, ap, EDGE) { |
| 3867 | ata_for_each_dev(dev, link, ALL) { |
| 3868 | struct ata_eh_context *ehc = &link->eh_context; |
| 3869 | unsigned long tmp; |
| 3870 | |
| 3871 | if (dev->class != ATA_DEV_ATA && |
| 3872 | dev->class != ATA_DEV_ZAC) |
| 3873 | continue; |
| 3874 | if (!(ehc->i.dev_action[dev->devno] & |
| 3875 | ATA_EH_PARK)) |
| 3876 | continue; |
| 3877 | tmp = dev->unpark_deadline; |
| 3878 | if (time_before(deadline, tmp)) |
| 3879 | deadline = tmp; |
| 3880 | else if (time_before_eq(tmp, jiffies)) |
| 3881 | continue; |
| 3882 | if (ehc->unloaded_mask & (1 << dev->devno)) |
| 3883 | continue; |
| 3884 | |
| 3885 | ata_eh_park_issue_cmd(dev, 1); |
| 3886 | } |
| 3887 | } |
| 3888 | |
| 3889 | now = jiffies; |
| 3890 | if (time_before_eq(deadline, now)) |
| 3891 | break; |
| 3892 | |
| 3893 | ata_eh_release(ap); |
| 3894 | deadline = wait_for_completion_timeout(&ap->park_req_pending, |
| 3895 | deadline - now); |
| 3896 | ata_eh_acquire(ap); |
| 3897 | } while (deadline); |
| 3898 | ata_for_each_link(link, ap, EDGE) { |
| 3899 | ata_for_each_dev(dev, link, ALL) { |
| 3900 | if (!(link->eh_context.unloaded_mask & |
| 3901 | (1 << dev->devno))) |
| 3902 | continue; |
| 3903 | |
| 3904 | ata_eh_park_issue_cmd(dev, 0); |
| 3905 | ata_eh_done(link, dev, ATA_EH_PARK); |
| 3906 | } |
| 3907 | } |
| 3908 | |
| 3909 | /* the rest */ |
| 3910 | nr_fails = 0; |
| 3911 | ata_for_each_link(link, ap, PMP_FIRST) { |
| 3912 | struct ata_eh_context *ehc = &link->eh_context; |
| 3913 | |
| 3914 | if (sata_pmp_attached(ap) && ata_is_host_link(link)) |
| 3915 | goto config_lpm; |
| 3916 | |
| 3917 | /* revalidate existing devices and attach new ones */ |
| 3918 | rc = ata_eh_revalidate_and_attach(link, &dev); |
| 3919 | if (rc) |
| 3920 | goto rest_fail; |
| 3921 | |
| 3922 | /* if PMP got attached, return, pmp EH will take care of it */ |
| 3923 | if (link->device->class == ATA_DEV_PMP) { |
| 3924 | ehc->i.action = 0; |
| 3925 | return 0; |
| 3926 | } |
| 3927 | |
| 3928 | /* configure transfer mode if necessary */ |
| 3929 | if (ehc->i.flags & ATA_EHI_SETMODE) { |
| 3930 | rc = ata_set_mode(link, &dev); |
| 3931 | if (rc) |
| 3932 | goto rest_fail; |
| 3933 | ehc->i.flags &= ~ATA_EHI_SETMODE; |
| 3934 | } |
| 3935 | |
| 3936 | /* If reset has been issued, clear UA to avoid |
| 3937 | * disrupting the current users of the device. |
| 3938 | */ |
| 3939 | if (ehc->i.flags & ATA_EHI_DID_RESET) { |
| 3940 | ata_for_each_dev(dev, link, ALL) { |
| 3941 | if (dev->class != ATA_DEV_ATAPI) |
| 3942 | continue; |
| 3943 | rc = atapi_eh_clear_ua(dev); |
| 3944 | if (rc) |
| 3945 | goto rest_fail; |
| 3946 | if (zpodd_dev_enabled(dev)) |
| 3947 | zpodd_post_poweron(dev); |
| 3948 | } |
| 3949 | } |
| 3950 | |
| 3951 | /* retry flush if necessary */ |
| 3952 | ata_for_each_dev(dev, link, ALL) { |
| 3953 | if (dev->class != ATA_DEV_ATA && |
| 3954 | dev->class != ATA_DEV_ZAC) |
| 3955 | continue; |
| 3956 | rc = ata_eh_maybe_retry_flush(dev); |
| 3957 | if (rc) |
| 3958 | goto rest_fail; |
| 3959 | } |
| 3960 | |
| 3961 | config_lpm: |
| 3962 | /* configure link power saving */ |
| 3963 | if (link->lpm_policy != ap->target_lpm_policy) { |
| 3964 | rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev); |
| 3965 | if (rc) |
| 3966 | goto rest_fail; |
| 3967 | } |
| 3968 | |
| 3969 | /* this link is okay now */ |
| 3970 | ehc->i.flags = 0; |
| 3971 | continue; |
| 3972 | |
| 3973 | rest_fail: |
| 3974 | nr_fails++; |
| 3975 | if (dev) |
| 3976 | ata_eh_handle_dev_fail(dev, rc); |
| 3977 | |
| 3978 | if (ap->pflags & ATA_PFLAG_FROZEN) { |
| 3979 | /* PMP reset requires working host port. |
| 3980 | * Can't retry if it's frozen. |
| 3981 | */ |
| 3982 | if (sata_pmp_attached(ap)) |
| 3983 | goto out; |
| 3984 | break; |
| 3985 | } |
| 3986 | } |
| 3987 | |
| 3988 | if (nr_fails) |
| 3989 | goto retry; |
| 3990 | |
| 3991 | out: |
| 3992 | if (rc && r_failed_link) |
| 3993 | *r_failed_link = link; |
| 3994 | |
| 3995 | DPRINTK("EXIT, rc=%d\n", rc); |
| 3996 | return rc; |
| 3997 | } |
| 3998 | |
| 3999 | /** |
| 4000 | * ata_eh_finish - finish up EH |
| 4001 | * @ap: host port to finish EH for |
| 4002 | * |
| 4003 | * Recovery is complete. Clean up EH states and retry or finish |
| 4004 | * failed qcs. |
| 4005 | * |
| 4006 | * LOCKING: |
| 4007 | * None. |
| 4008 | */ |
| 4009 | void ata_eh_finish(struct ata_port *ap) |
| 4010 | { |
| 4011 | int tag; |
| 4012 | |
| 4013 | /* retry or finish qcs */ |
| 4014 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { |
| 4015 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
| 4016 | |
| 4017 | if (!(qc->flags & ATA_QCFLAG_FAILED)) |
| 4018 | continue; |
| 4019 | |
| 4020 | if (qc->err_mask) { |
| 4021 | /* FIXME: Once EH migration is complete, |
| 4022 | * generate sense data in this function, |
| 4023 | * considering both err_mask and tf. |
| 4024 | */ |
| 4025 | if (qc->flags & ATA_QCFLAG_RETRY) |
| 4026 | ata_eh_qc_retry(qc); |
| 4027 | else |
| 4028 | ata_eh_qc_complete(qc); |
| 4029 | } else { |
| 4030 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) { |
| 4031 | ata_eh_qc_complete(qc); |
| 4032 | } else { |
| 4033 | /* feed zero TF to sense generation */ |
| 4034 | memset(&qc->result_tf, 0, sizeof(qc->result_tf)); |
| 4035 | ata_eh_qc_retry(qc); |
| 4036 | } |
| 4037 | } |
| 4038 | } |
| 4039 | |
| 4040 | /* make sure nr_active_links is zero after EH */ |
| 4041 | WARN_ON(ap->nr_active_links); |
| 4042 | ap->nr_active_links = 0; |
| 4043 | } |
| 4044 | |
| 4045 | /** |
| 4046 | * ata_do_eh - do standard error handling |
| 4047 | * @ap: host port to handle error for |
| 4048 | * |
| 4049 | * @prereset: prereset method (can be NULL) |
| 4050 | * @softreset: softreset method (can be NULL) |
| 4051 | * @hardreset: hardreset method (can be NULL) |
| 4052 | * @postreset: postreset method (can be NULL) |
| 4053 | * |
| 4054 | * Perform standard error handling sequence. |
| 4055 | * |
| 4056 | * LOCKING: |
| 4057 | * Kernel thread context (may sleep). |
| 4058 | */ |
| 4059 | void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset, |
| 4060 | ata_reset_fn_t softreset, ata_reset_fn_t hardreset, |
| 4061 | ata_postreset_fn_t postreset) |
| 4062 | { |
| 4063 | struct ata_device *dev; |
| 4064 | int rc; |
| 4065 | |
| 4066 | ata_eh_autopsy(ap); |
| 4067 | ata_eh_report(ap); |
| 4068 | |
| 4069 | rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset, |
| 4070 | NULL); |
| 4071 | if (rc) { |
| 4072 | ata_for_each_dev(dev, &ap->link, ALL) |
| 4073 | ata_dev_disable(dev); |
| 4074 | } |
| 4075 | |
| 4076 | ata_eh_finish(ap); |
| 4077 | } |
| 4078 | |
| 4079 | /** |
| 4080 | * ata_std_error_handler - standard error handler |
| 4081 | * @ap: host port to handle error for |
| 4082 | * |
| 4083 | * Standard error handler |
| 4084 | * |
| 4085 | * LOCKING: |
| 4086 | * Kernel thread context (may sleep). |
| 4087 | */ |
| 4088 | void ata_std_error_handler(struct ata_port *ap) |
| 4089 | { |
| 4090 | struct ata_port_operations *ops = ap->ops; |
| 4091 | ata_reset_fn_t hardreset = ops->hardreset; |
| 4092 | |
| 4093 | /* ignore built-in hardreset if SCR access is not available */ |
| 4094 | if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link)) |
| 4095 | hardreset = NULL; |
| 4096 | |
| 4097 | ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset); |
| 4098 | } |
| 4099 | |
| 4100 | #ifdef CONFIG_PM |
| 4101 | /** |
| 4102 | * ata_eh_handle_port_suspend - perform port suspend operation |
| 4103 | * @ap: port to suspend |
| 4104 | * |
| 4105 | * Suspend @ap. |
| 4106 | * |
| 4107 | * LOCKING: |
| 4108 | * Kernel thread context (may sleep). |
| 4109 | */ |
| 4110 | static void ata_eh_handle_port_suspend(struct ata_port *ap) |
| 4111 | { |
| 4112 | unsigned long flags; |
| 4113 | int rc = 0; |
| 4114 | struct ata_device *dev; |
| 4115 | |
| 4116 | /* are we suspending? */ |
| 4117 | spin_lock_irqsave(ap->lock, flags); |
| 4118 | if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || |
| 4119 | ap->pm_mesg.event & PM_EVENT_RESUME) { |
| 4120 | spin_unlock_irqrestore(ap->lock, flags); |
| 4121 | return; |
| 4122 | } |
| 4123 | spin_unlock_irqrestore(ap->lock, flags); |
| 4124 | |
| 4125 | WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED); |
| 4126 | |
| 4127 | /* |
| 4128 | * If we have a ZPODD attached, check its zero |
| 4129 | * power ready status before the port is frozen. |
| 4130 | * Only needed for runtime suspend. |
| 4131 | */ |
| 4132 | if (PMSG_IS_AUTO(ap->pm_mesg)) { |
| 4133 | ata_for_each_dev(dev, &ap->link, ENABLED) { |
| 4134 | if (zpodd_dev_enabled(dev)) |
| 4135 | zpodd_on_suspend(dev); |
| 4136 | } |
| 4137 | } |
| 4138 | |
| 4139 | /* tell ACPI we're suspending */ |
| 4140 | rc = ata_acpi_on_suspend(ap); |
| 4141 | if (rc) |
| 4142 | goto out; |
| 4143 | |
| 4144 | /* suspend */ |
| 4145 | ata_eh_freeze_port(ap); |
| 4146 | |
| 4147 | if (ap->ops->port_suspend) |
| 4148 | rc = ap->ops->port_suspend(ap, ap->pm_mesg); |
| 4149 | |
| 4150 | ata_acpi_set_state(ap, ap->pm_mesg); |
| 4151 | out: |
| 4152 | /* update the flags */ |
| 4153 | spin_lock_irqsave(ap->lock, flags); |
| 4154 | |
| 4155 | ap->pflags &= ~ATA_PFLAG_PM_PENDING; |
| 4156 | if (rc == 0) |
| 4157 | ap->pflags |= ATA_PFLAG_SUSPENDED; |
| 4158 | else if (ap->pflags & ATA_PFLAG_FROZEN) |
| 4159 | ata_port_schedule_eh(ap); |
| 4160 | |
| 4161 | spin_unlock_irqrestore(ap->lock, flags); |
| 4162 | |
| 4163 | return; |
| 4164 | } |
| 4165 | |
| 4166 | /** |
| 4167 | * ata_eh_handle_port_resume - perform port resume operation |
| 4168 | * @ap: port to resume |
| 4169 | * |
| 4170 | * Resume @ap. |
| 4171 | * |
| 4172 | * LOCKING: |
| 4173 | * Kernel thread context (may sleep). |
| 4174 | */ |
| 4175 | static void ata_eh_handle_port_resume(struct ata_port *ap) |
| 4176 | { |
| 4177 | struct ata_link *link; |
| 4178 | struct ata_device *dev; |
| 4179 | unsigned long flags; |
| 4180 | |
| 4181 | /* are we resuming? */ |
| 4182 | spin_lock_irqsave(ap->lock, flags); |
| 4183 | if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || |
| 4184 | !(ap->pm_mesg.event & PM_EVENT_RESUME)) { |
| 4185 | spin_unlock_irqrestore(ap->lock, flags); |
| 4186 | return; |
| 4187 | } |
| 4188 | spin_unlock_irqrestore(ap->lock, flags); |
| 4189 | |
| 4190 | WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED)); |
| 4191 | |
| 4192 | /* |
| 4193 | * Error timestamps are in jiffies which doesn't run while |
| 4194 | * suspended and PHY events during resume isn't too uncommon. |
| 4195 | * When the two are combined, it can lead to unnecessary speed |
| 4196 | * downs if the machine is suspended and resumed repeatedly. |
| 4197 | * Clear error history. |
| 4198 | */ |
| 4199 | ata_for_each_link(link, ap, HOST_FIRST) |
| 4200 | ata_for_each_dev(dev, link, ALL) |
| 4201 | ata_ering_clear(&dev->ering); |
| 4202 | |
| 4203 | ata_acpi_set_state(ap, ap->pm_mesg); |
| 4204 | |
| 4205 | if (ap->ops->port_resume) |
| 4206 | ap->ops->port_resume(ap); |
| 4207 | |
| 4208 | /* tell ACPI that we're resuming */ |
| 4209 | ata_acpi_on_resume(ap); |
| 4210 | |
| 4211 | /* update the flags */ |
| 4212 | spin_lock_irqsave(ap->lock, flags); |
| 4213 | ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED); |
| 4214 | spin_unlock_irqrestore(ap->lock, flags); |
| 4215 | } |
| 4216 | #endif /* CONFIG_PM */ |