1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * libata-eh.c - libata error handling 4 * 5 * Copyright 2006 Tejun Heo <htejun@gmail.com> 6 * 7 * libata documentation is available via 'make {ps|pdf}docs', 8 * as Documentation/driver-api/libata.rst 9 * 10 * Hardware documentation available from http://www.t13.org/ and 11 * http://www.sata-io.org/ 12 */ 13 14 #include <linux/kernel.h> 15 #include <linux/blkdev.h> 16 #include <linux/export.h> 17 #include <linux/pci.h> 18 #include <scsi/scsi.h> 19 #include <scsi/scsi_host.h> 20 #include <scsi/scsi_eh.h> 21 #include <scsi/scsi_device.h> 22 #include <scsi/scsi_cmnd.h> 23 #include <scsi/scsi_dbg.h> 24 #include "../scsi/scsi_transport_api.h" 25 26 #include <linux/libata.h> 27 28 #include <trace/events/libata.h> 29 #include "libata.h" 30 31 enum { 32 /* speed down verdicts */ 33 ATA_EH_SPDN_NCQ_OFF = (1 << 0), 34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1), 35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2), 36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3), 37 38 /* error flags */ 39 ATA_EFLAG_IS_IO = (1 << 0), 40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1), 41 ATA_EFLAG_OLD_ER = (1 << 31), 42 43 /* error categories */ 44 ATA_ECAT_NONE = 0, 45 ATA_ECAT_ATA_BUS = 1, 46 ATA_ECAT_TOUT_HSM = 2, 47 ATA_ECAT_UNK_DEV = 3, 48 ATA_ECAT_DUBIOUS_NONE = 4, 49 ATA_ECAT_DUBIOUS_ATA_BUS = 5, 50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6, 51 ATA_ECAT_DUBIOUS_UNK_DEV = 7, 52 ATA_ECAT_NR = 8, 53 54 ATA_EH_CMD_DFL_TIMEOUT = 5000, 55 56 /* always put at least this amount of time between resets */ 57 ATA_EH_RESET_COOL_DOWN = 5000, 58 59 /* Waiting in ->prereset can never be reliable. It's 60 * sometimes nice to wait there but it can't be depended upon; 61 * otherwise, we wouldn't be resetting. Just give it enough 62 * time for most drives to spin up. 63 */ 64 ATA_EH_PRERESET_TIMEOUT = 10000, 65 ATA_EH_FASTDRAIN_INTERVAL = 3000, 66 67 ATA_EH_UA_TRIES = 5, 68 69 /* probe speed down parameters, see ata_eh_schedule_probe() */ 70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */ 71 ATA_EH_PROBE_TRIALS = 2, 72 }; 73 74 /* The following table determines how we sequence resets. Each entry 75 * represents timeout for that try. The first try can be soft or 76 * hardreset. All others are hardreset if available. In most cases 77 * the first reset w/ 10sec timeout should succeed. Following entries 78 * are mostly for error handling, hotplug and those outlier devices that 79 * take an exceptionally long time to recover from reset. 80 */ 81 static const unsigned int ata_eh_reset_timeouts[] = { 82 10000, /* most drives spin up by 10sec */ 83 10000, /* > 99% working drives spin up before 20sec */ 84 35000, /* give > 30 secs of idleness for outlier devices */ 85 5000, /* and sweet one last chance */ 86 UINT_MAX, /* > 1 min has elapsed, give up */ 87 }; 88 89 static const unsigned int ata_eh_identify_timeouts[] = { 90 5000, /* covers > 99% of successes and not too boring on failures */ 91 10000, /* combined time till here is enough even for media access */ 92 30000, /* for true idiots */ 93 UINT_MAX, 94 }; 95 96 static const unsigned int ata_eh_revalidate_timeouts[] = { 97 15000, /* Some drives are slow to read log pages when waking-up */ 98 15000, /* combined time till here is enough even for media access */ 99 UINT_MAX, 100 }; 101 102 static const unsigned int ata_eh_flush_timeouts[] = { 103 15000, /* be generous with flush */ 104 15000, /* ditto */ 105 30000, /* and even more generous */ 106 UINT_MAX, 107 }; 108 109 static const unsigned int ata_eh_other_timeouts[] = { 110 5000, /* same rationale as identify timeout */ 111 10000, /* ditto */ 112 /* but no merciful 30sec for other commands, it just isn't worth it */ 113 UINT_MAX, 114 }; 115 116 struct ata_eh_cmd_timeout_ent { 117 const u8 *commands; 118 const unsigned int *timeouts; 119 }; 120 121 /* The following table determines timeouts to use for EH internal 122 * commands. Each table entry is a command class and matches the 123 * commands the entry applies to and the timeout table to use. 124 * 125 * On the retry after a command timed out, the next timeout value from 126 * the table is used. If the table doesn't contain further entries, 127 * the last value is used. 128 * 129 * ehc->cmd_timeout_idx keeps track of which timeout to use per 130 * command class, so if SET_FEATURES times out on the first try, the 131 * next try will use the second timeout value only for that class. 132 */ 133 #define CMDS(cmds...) (const u8 []){ cmds, 0 } 134 static const struct ata_eh_cmd_timeout_ent 135 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = { 136 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI), 137 .timeouts = ata_eh_identify_timeouts, }, 138 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT), 139 .timeouts = ata_eh_revalidate_timeouts, }, 140 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT), 141 .timeouts = ata_eh_other_timeouts, }, 142 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT), 143 .timeouts = ata_eh_other_timeouts, }, 144 { .commands = CMDS(ATA_CMD_SET_FEATURES), 145 .timeouts = ata_eh_other_timeouts, }, 146 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS), 147 .timeouts = ata_eh_other_timeouts, }, 148 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT), 149 .timeouts = ata_eh_flush_timeouts }, 150 { .commands = CMDS(ATA_CMD_VERIFY), 151 .timeouts = ata_eh_reset_timeouts }, 152 }; 153 #undef CMDS 154 155 static void __ata_port_freeze(struct ata_port *ap); 156 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy, 157 struct ata_device **r_failed_dev); 158 #ifdef CONFIG_PM 159 static void ata_eh_handle_port_suspend(struct ata_port *ap); 160 static void ata_eh_handle_port_resume(struct ata_port *ap); 161 #else /* CONFIG_PM */ 162 static void ata_eh_handle_port_suspend(struct ata_port *ap) 163 { } 164 165 static void ata_eh_handle_port_resume(struct ata_port *ap) 166 { } 167 #endif /* CONFIG_PM */ 168 169 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, 170 const char *fmt, va_list args) 171 { 172 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len, 173 ATA_EH_DESC_LEN - ehi->desc_len, 174 fmt, args); 175 } 176 177 /** 178 * __ata_ehi_push_desc - push error description without adding separator 179 * @ehi: target EHI 180 * @fmt: printf format string 181 * 182 * Format string according to @fmt and append it to @ehi->desc. 183 * 184 * LOCKING: 185 * spin_lock_irqsave(host lock) 186 */ 187 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...) 188 { 189 va_list args; 190 191 va_start(args, fmt); 192 __ata_ehi_pushv_desc(ehi, fmt, args); 193 va_end(args); 194 } 195 EXPORT_SYMBOL_GPL(__ata_ehi_push_desc); 196 197 /** 198 * ata_ehi_push_desc - push error description with separator 199 * @ehi: target EHI 200 * @fmt: printf format string 201 * 202 * Format string according to @fmt and append it to @ehi->desc. 203 * If @ehi->desc is not empty, ", " is added in-between. 204 * 205 * LOCKING: 206 * spin_lock_irqsave(host lock) 207 */ 208 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...) 209 { 210 va_list args; 211 212 if (ehi->desc_len) 213 __ata_ehi_push_desc(ehi, ", "); 214 215 va_start(args, fmt); 216 __ata_ehi_pushv_desc(ehi, fmt, args); 217 va_end(args); 218 } 219 EXPORT_SYMBOL_GPL(ata_ehi_push_desc); 220 221 /** 222 * ata_ehi_clear_desc - clean error description 223 * @ehi: target EHI 224 * 225 * Clear @ehi->desc. 226 * 227 * LOCKING: 228 * spin_lock_irqsave(host lock) 229 */ 230 void ata_ehi_clear_desc(struct ata_eh_info *ehi) 231 { 232 ehi->desc[0] = '\0'; 233 ehi->desc_len = 0; 234 } 235 EXPORT_SYMBOL_GPL(ata_ehi_clear_desc); 236 237 /** 238 * ata_port_desc - append port description 239 * @ap: target ATA port 240 * @fmt: printf format string 241 * 242 * Format string according to @fmt and append it to port 243 * description. If port description is not empty, " " is added 244 * in-between. This function is to be used while initializing 245 * ata_host. The description is printed on host registration. 246 * 247 * LOCKING: 248 * None. 249 */ 250 void ata_port_desc(struct ata_port *ap, const char *fmt, ...) 251 { 252 va_list args; 253 254 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING)); 255 256 if (ap->link.eh_info.desc_len) 257 __ata_ehi_push_desc(&ap->link.eh_info, " "); 258 259 va_start(args, fmt); 260 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args); 261 va_end(args); 262 } 263 EXPORT_SYMBOL_GPL(ata_port_desc); 264 265 #ifdef CONFIG_PCI 266 /** 267 * ata_port_pbar_desc - append PCI BAR description 268 * @ap: target ATA port 269 * @bar: target PCI BAR 270 * @offset: offset into PCI BAR 271 * @name: name of the area 272 * 273 * If @offset is negative, this function formats a string which 274 * contains the name, address, size and type of the BAR and 275 * appends it to the port description. If @offset is zero or 276 * positive, only name and offsetted address is appended. 277 * 278 * LOCKING: 279 * None. 280 */ 281 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset, 282 const char *name) 283 { 284 struct pci_dev *pdev = to_pci_dev(ap->host->dev); 285 char *type = ""; 286 unsigned long long start, len; 287 288 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) 289 type = "m"; 290 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) 291 type = "i"; 292 293 start = (unsigned long long)pci_resource_start(pdev, bar); 294 len = (unsigned long long)pci_resource_len(pdev, bar); 295 296 if (offset < 0) 297 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start); 298 else 299 ata_port_desc(ap, "%s 0x%llx", name, 300 start + (unsigned long long)offset); 301 } 302 EXPORT_SYMBOL_GPL(ata_port_pbar_desc); 303 #endif /* CONFIG_PCI */ 304 305 static int ata_lookup_timeout_table(u8 cmd) 306 { 307 int i; 308 309 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) { 310 const u8 *cur; 311 312 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++) 313 if (*cur == cmd) 314 return i; 315 } 316 317 return -1; 318 } 319 320 /** 321 * ata_internal_cmd_timeout - determine timeout for an internal command 322 * @dev: target device 323 * @cmd: internal command to be issued 324 * 325 * Determine timeout for internal command @cmd for @dev. 326 * 327 * LOCKING: 328 * EH context. 329 * 330 * RETURNS: 331 * Determined timeout. 332 */ 333 unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd) 334 { 335 struct ata_eh_context *ehc = &dev->link->eh_context; 336 int ent = ata_lookup_timeout_table(cmd); 337 int idx; 338 339 if (ent < 0) 340 return ATA_EH_CMD_DFL_TIMEOUT; 341 342 idx = ehc->cmd_timeout_idx[dev->devno][ent]; 343 return ata_eh_cmd_timeout_table[ent].timeouts[idx]; 344 } 345 346 /** 347 * ata_internal_cmd_timed_out - notification for internal command timeout 348 * @dev: target device 349 * @cmd: internal command which timed out 350 * 351 * Notify EH that internal command @cmd for @dev timed out. This 352 * function should be called only for commands whose timeouts are 353 * determined using ata_internal_cmd_timeout(). 354 * 355 * LOCKING: 356 * EH context. 357 */ 358 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd) 359 { 360 struct ata_eh_context *ehc = &dev->link->eh_context; 361 int ent = ata_lookup_timeout_table(cmd); 362 int idx; 363 364 if (ent < 0) 365 return; 366 367 idx = ehc->cmd_timeout_idx[dev->devno][ent]; 368 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX) 369 ehc->cmd_timeout_idx[dev->devno][ent]++; 370 } 371 372 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags, 373 unsigned int err_mask) 374 { 375 struct ata_ering_entry *ent; 376 377 WARN_ON(!err_mask); 378 379 ering->cursor++; 380 ering->cursor %= ATA_ERING_SIZE; 381 382 ent = &ering->ring[ering->cursor]; 383 ent->eflags = eflags; 384 ent->err_mask = err_mask; 385 ent->timestamp = get_jiffies_64(); 386 } 387 388 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering) 389 { 390 struct ata_ering_entry *ent = &ering->ring[ering->cursor]; 391 392 if (ent->err_mask) 393 return ent; 394 return NULL; 395 } 396 397 int ata_ering_map(struct ata_ering *ering, 398 int (*map_fn)(struct ata_ering_entry *, void *), 399 void *arg) 400 { 401 int idx, rc = 0; 402 struct ata_ering_entry *ent; 403 404 idx = ering->cursor; 405 do { 406 ent = &ering->ring[idx]; 407 if (!ent->err_mask) 408 break; 409 rc = map_fn(ent, arg); 410 if (rc) 411 break; 412 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE; 413 } while (idx != ering->cursor); 414 415 return rc; 416 } 417 418 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg) 419 { 420 ent->eflags |= ATA_EFLAG_OLD_ER; 421 return 0; 422 } 423 424 static void ata_ering_clear(struct ata_ering *ering) 425 { 426 ata_ering_map(ering, ata_ering_clear_cb, NULL); 427 } 428 429 static unsigned int ata_eh_dev_action(struct ata_device *dev) 430 { 431 struct ata_eh_context *ehc = &dev->link->eh_context; 432 433 return ehc->i.action | ehc->i.dev_action[dev->devno]; 434 } 435 436 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev, 437 struct ata_eh_info *ehi, unsigned int action) 438 { 439 struct ata_device *tdev; 440 441 if (!dev) { 442 ehi->action &= ~action; 443 ata_for_each_dev(tdev, link, ALL) 444 ehi->dev_action[tdev->devno] &= ~action; 445 } else { 446 /* doesn't make sense for port-wide EH actions */ 447 WARN_ON(!(action & ATA_EH_PERDEV_MASK)); 448 449 /* break ehi->action into ehi->dev_action */ 450 if (ehi->action & action) { 451 ata_for_each_dev(tdev, link, ALL) 452 ehi->dev_action[tdev->devno] |= 453 ehi->action & action; 454 ehi->action &= ~action; 455 } 456 457 /* turn off the specified per-dev action */ 458 ehi->dev_action[dev->devno] &= ~action; 459 } 460 } 461 462 /** 463 * ata_eh_acquire - acquire EH ownership 464 * @ap: ATA port to acquire EH ownership for 465 * 466 * Acquire EH ownership for @ap. This is the basic exclusion 467 * mechanism for ports sharing a host. Only one port hanging off 468 * the same host can claim the ownership of EH. 469 * 470 * LOCKING: 471 * EH context. 472 */ 473 void ata_eh_acquire(struct ata_port *ap) 474 { 475 mutex_lock(&ap->host->eh_mutex); 476 WARN_ON_ONCE(ap->host->eh_owner); 477 ap->host->eh_owner = current; 478 } 479 480 /** 481 * ata_eh_release - release EH ownership 482 * @ap: ATA port to release EH ownership for 483 * 484 * Release EH ownership for @ap if the caller. The caller must 485 * have acquired EH ownership using ata_eh_acquire() previously. 486 * 487 * LOCKING: 488 * EH context. 489 */ 490 void ata_eh_release(struct ata_port *ap) 491 { 492 WARN_ON_ONCE(ap->host->eh_owner != current); 493 ap->host->eh_owner = NULL; 494 mutex_unlock(&ap->host->eh_mutex); 495 } 496 497 static void ata_eh_unload(struct ata_port *ap) 498 { 499 struct ata_link *link; 500 struct ata_device *dev; 501 unsigned long flags; 502 503 /* 504 * Unless we are restarting, transition all enabled devices to 505 * standby power mode. 506 */ 507 if (system_state != SYSTEM_RESTART) { 508 ata_for_each_link(link, ap, PMP_FIRST) { 509 ata_for_each_dev(dev, link, ENABLED) 510 ata_dev_power_set_standby(dev); 511 } 512 } 513 514 /* 515 * Restore SControl IPM and SPD for the next driver and 516 * disable attached devices. 517 */ 518 ata_for_each_link(link, ap, PMP_FIRST) { 519 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0); 520 ata_for_each_dev(dev, link, ALL) 521 ata_dev_disable(dev); 522 } 523 524 /* freeze and set UNLOADED */ 525 spin_lock_irqsave(ap->lock, flags); 526 527 ata_port_freeze(ap); /* won't be thawed */ 528 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */ 529 ap->pflags |= ATA_PFLAG_UNLOADED; 530 531 spin_unlock_irqrestore(ap->lock, flags); 532 } 533 534 /** 535 * ata_scsi_error - SCSI layer error handler callback 536 * @host: SCSI host on which error occurred 537 * 538 * Handles SCSI-layer-thrown error events. 539 * 540 * LOCKING: 541 * Inherited from SCSI layer (none, can sleep) 542 * 543 * RETURNS: 544 * Zero. 545 */ 546 void ata_scsi_error(struct Scsi_Host *host) 547 { 548 struct ata_port *ap = ata_shost_to_port(host); 549 unsigned long flags; 550 LIST_HEAD(eh_work_q); 551 552 spin_lock_irqsave(host->host_lock, flags); 553 list_splice_init(&host->eh_cmd_q, &eh_work_q); 554 spin_unlock_irqrestore(host->host_lock, flags); 555 556 ata_scsi_cmd_error_handler(host, ap, &eh_work_q); 557 558 /* If we timed raced normal completion and there is nothing to 559 recover nr_timedout == 0 why exactly are we doing error recovery ? */ 560 ata_scsi_port_error_handler(host, ap); 561 562 /* finish or retry handled scmd's and clean up */ 563 WARN_ON(!list_empty(&eh_work_q)); 564 565 } 566 567 /** 568 * ata_scsi_cmd_error_handler - error callback for a list of commands 569 * @host: scsi host containing the port 570 * @ap: ATA port within the host 571 * @eh_work_q: list of commands to process 572 * 573 * process the given list of commands and return those finished to the 574 * ap->eh_done_q. This function is the first part of the libata error 575 * handler which processes a given list of failed commands. 576 */ 577 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap, 578 struct list_head *eh_work_q) 579 { 580 int i; 581 unsigned long flags; 582 struct scsi_cmnd *scmd, *tmp; 583 int nr_timedout = 0; 584 585 /* make sure sff pio task is not running */ 586 ata_sff_flush_pio_task(ap); 587 588 /* synchronize with host lock and sort out timeouts */ 589 590 /* 591 * For EH, all qcs are finished in one of three ways - 592 * normal completion, error completion, and SCSI timeout. 593 * Both completions can race against SCSI timeout. When normal 594 * completion wins, the qc never reaches EH. When error 595 * completion wins, the qc has ATA_QCFLAG_EH set. 596 * 597 * When SCSI timeout wins, things are a bit more complex. 598 * Normal or error completion can occur after the timeout but 599 * before this point. In such cases, both types of 600 * completions are honored. A scmd is determined to have 601 * timed out iff its associated qc is active and not failed. 602 */ 603 spin_lock_irqsave(ap->lock, flags); 604 605 /* 606 * This must occur under the ap->lock as we don't want 607 * a polled recovery to race the real interrupt handler 608 * 609 * The lost_interrupt handler checks for any completed but 610 * non-notified command and completes much like an IRQ handler. 611 * 612 * We then fall into the error recovery code which will treat 613 * this as if normal completion won the race 614 */ 615 if (ap->ops->lost_interrupt) 616 ap->ops->lost_interrupt(ap); 617 618 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) { 619 struct ata_queued_cmd *qc; 620 621 /* 622 * If the scmd was added to EH, via ata_qc_schedule_eh() -> 623 * scsi_timeout() -> scsi_eh_scmd_add(), scsi_timeout() will 624 * have set DID_TIME_OUT (since libata does not have an abort 625 * handler). Thus, to clear DID_TIME_OUT, clear the host byte. 626 */ 627 set_host_byte(scmd, DID_OK); 628 629 ata_qc_for_each_raw(ap, qc, i) { 630 if (qc->flags & ATA_QCFLAG_ACTIVE && 631 qc->scsicmd == scmd) 632 break; 633 } 634 635 if (i < ATA_MAX_QUEUE) { 636 /* the scmd has an associated qc */ 637 if (!(qc->flags & ATA_QCFLAG_EH)) { 638 /* which hasn't failed yet, timeout */ 639 set_host_byte(scmd, DID_TIME_OUT); 640 qc->err_mask |= AC_ERR_TIMEOUT; 641 qc->flags |= ATA_QCFLAG_EH; 642 nr_timedout++; 643 } 644 } else { 645 /* Normal completion occurred after 646 * SCSI timeout but before this point. 647 * Successfully complete it. 648 */ 649 scmd->retries = scmd->allowed; 650 scsi_eh_finish_cmd(scmd, &ap->eh_done_q); 651 } 652 } 653 654 /* 655 * If we have timed out qcs. They belong to EH from 656 * this point but the state of the controller is 657 * unknown. Freeze the port to make sure the IRQ 658 * handler doesn't diddle with those qcs. This must 659 * be done atomically w.r.t. setting ATA_QCFLAG_EH. 660 */ 661 if (nr_timedout) 662 __ata_port_freeze(ap); 663 664 /* initialize eh_tries */ 665 ap->eh_tries = ATA_EH_MAX_TRIES; 666 667 spin_unlock_irqrestore(ap->lock, flags); 668 } 669 EXPORT_SYMBOL(ata_scsi_cmd_error_handler); 670 671 /** 672 * ata_scsi_port_error_handler - recover the port after the commands 673 * @host: SCSI host containing the port 674 * @ap: the ATA port 675 * 676 * Handle the recovery of the port @ap after all the commands 677 * have been recovered. 678 */ 679 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap) 680 { 681 unsigned long flags; 682 struct ata_link *link; 683 684 /* acquire EH ownership */ 685 ata_eh_acquire(ap); 686 repeat: 687 /* kill fast drain timer */ 688 del_timer_sync(&ap->fastdrain_timer); 689 690 /* process port resume request */ 691 ata_eh_handle_port_resume(ap); 692 693 /* fetch & clear EH info */ 694 spin_lock_irqsave(ap->lock, flags); 695 696 ata_for_each_link(link, ap, HOST_FIRST) { 697 struct ata_eh_context *ehc = &link->eh_context; 698 struct ata_device *dev; 699 700 memset(&link->eh_context, 0, sizeof(link->eh_context)); 701 link->eh_context.i = link->eh_info; 702 memset(&link->eh_info, 0, sizeof(link->eh_info)); 703 704 ata_for_each_dev(dev, link, ENABLED) { 705 int devno = dev->devno; 706 707 ehc->saved_xfer_mode[devno] = dev->xfer_mode; 708 if (ata_ncq_enabled(dev)) 709 ehc->saved_ncq_enabled |= 1 << devno; 710 711 /* If we are resuming, wake up the device */ 712 if (ap->pflags & ATA_PFLAG_RESUMING) { 713 dev->flags |= ATA_DFLAG_RESUMING; 714 ehc->i.dev_action[devno] |= ATA_EH_SET_ACTIVE; 715 } 716 } 717 } 718 719 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS; 720 ap->pflags &= ~ATA_PFLAG_EH_PENDING; 721 ap->excl_link = NULL; /* don't maintain exclusion over EH */ 722 723 spin_unlock_irqrestore(ap->lock, flags); 724 725 /* invoke EH, skip if unloading or suspended */ 726 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED))) 727 ap->ops->error_handler(ap); 728 else { 729 /* if unloading, commence suicide */ 730 if ((ap->pflags & ATA_PFLAG_UNLOADING) && 731 !(ap->pflags & ATA_PFLAG_UNLOADED)) 732 ata_eh_unload(ap); 733 ata_eh_finish(ap); 734 } 735 736 /* process port suspend request */ 737 ata_eh_handle_port_suspend(ap); 738 739 /* 740 * Exception might have happened after ->error_handler recovered the 741 * port but before this point. Repeat EH in such case. 742 */ 743 spin_lock_irqsave(ap->lock, flags); 744 745 if (ap->pflags & ATA_PFLAG_EH_PENDING) { 746 if (--ap->eh_tries) { 747 spin_unlock_irqrestore(ap->lock, flags); 748 goto repeat; 749 } 750 ata_port_err(ap, 751 "EH pending after %d tries, giving up\n", 752 ATA_EH_MAX_TRIES); 753 ap->pflags &= ~ATA_PFLAG_EH_PENDING; 754 } 755 756 /* this run is complete, make sure EH info is clear */ 757 ata_for_each_link(link, ap, HOST_FIRST) 758 memset(&link->eh_info, 0, sizeof(link->eh_info)); 759 760 /* 761 * end eh (clear host_eh_scheduled) while holding ap->lock such that if 762 * exception occurs after this point but before EH completion, SCSI 763 * midlayer will re-initiate EH. 764 */ 765 ap->ops->end_eh(ap); 766 767 spin_unlock_irqrestore(ap->lock, flags); 768 ata_eh_release(ap); 769 770 scsi_eh_flush_done_q(&ap->eh_done_q); 771 772 /* clean up */ 773 spin_lock_irqsave(ap->lock, flags); 774 775 ap->pflags &= ~ATA_PFLAG_RESUMING; 776 777 if (ap->pflags & ATA_PFLAG_LOADING) 778 ap->pflags &= ~ATA_PFLAG_LOADING; 779 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) && 780 !(ap->flags & ATA_FLAG_SAS_HOST)) 781 schedule_delayed_work(&ap->hotplug_task, 0); 782 783 if (ap->pflags & ATA_PFLAG_RECOVERED) 784 ata_port_info(ap, "EH complete\n"); 785 786 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED); 787 788 /* tell wait_eh that we're done */ 789 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS; 790 wake_up_all(&ap->eh_wait_q); 791 792 spin_unlock_irqrestore(ap->lock, flags); 793 } 794 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler); 795 796 /** 797 * ata_port_wait_eh - Wait for the currently pending EH to complete 798 * @ap: Port to wait EH for 799 * 800 * Wait until the currently pending EH is complete. 801 * 802 * LOCKING: 803 * Kernel thread context (may sleep). 804 */ 805 void ata_port_wait_eh(struct ata_port *ap) 806 { 807 unsigned long flags; 808 DEFINE_WAIT(wait); 809 810 retry: 811 spin_lock_irqsave(ap->lock, flags); 812 813 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) { 814 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE); 815 spin_unlock_irqrestore(ap->lock, flags); 816 schedule(); 817 spin_lock_irqsave(ap->lock, flags); 818 } 819 finish_wait(&ap->eh_wait_q, &wait); 820 821 spin_unlock_irqrestore(ap->lock, flags); 822 823 /* make sure SCSI EH is complete */ 824 if (scsi_host_in_recovery(ap->scsi_host)) { 825 ata_msleep(ap, 10); 826 goto retry; 827 } 828 } 829 EXPORT_SYMBOL_GPL(ata_port_wait_eh); 830 831 static unsigned int ata_eh_nr_in_flight(struct ata_port *ap) 832 { 833 struct ata_queued_cmd *qc; 834 unsigned int tag; 835 unsigned int nr = 0; 836 837 /* count only non-internal commands */ 838 ata_qc_for_each(ap, qc, tag) { 839 if (qc) 840 nr++; 841 } 842 843 return nr; 844 } 845 846 void ata_eh_fastdrain_timerfn(struct timer_list *t) 847 { 848 struct ata_port *ap = from_timer(ap, t, fastdrain_timer); 849 unsigned long flags; 850 unsigned int cnt; 851 852 spin_lock_irqsave(ap->lock, flags); 853 854 cnt = ata_eh_nr_in_flight(ap); 855 856 /* are we done? */ 857 if (!cnt) 858 goto out_unlock; 859 860 if (cnt == ap->fastdrain_cnt) { 861 struct ata_queued_cmd *qc; 862 unsigned int tag; 863 864 /* No progress during the last interval, tag all 865 * in-flight qcs as timed out and freeze the port. 866 */ 867 ata_qc_for_each(ap, qc, tag) { 868 if (qc) 869 qc->err_mask |= AC_ERR_TIMEOUT; 870 } 871 872 ata_port_freeze(ap); 873 } else { 874 /* some qcs have finished, give it another chance */ 875 ap->fastdrain_cnt = cnt; 876 ap->fastdrain_timer.expires = 877 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL); 878 add_timer(&ap->fastdrain_timer); 879 } 880 881 out_unlock: 882 spin_unlock_irqrestore(ap->lock, flags); 883 } 884 885 /** 886 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain 887 * @ap: target ATA port 888 * @fastdrain: activate fast drain 889 * 890 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain 891 * is non-zero and EH wasn't pending before. Fast drain ensures 892 * that EH kicks in in timely manner. 893 * 894 * LOCKING: 895 * spin_lock_irqsave(host lock) 896 */ 897 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain) 898 { 899 unsigned int cnt; 900 901 /* already scheduled? */ 902 if (ap->pflags & ATA_PFLAG_EH_PENDING) 903 return; 904 905 ap->pflags |= ATA_PFLAG_EH_PENDING; 906 907 if (!fastdrain) 908 return; 909 910 /* do we have in-flight qcs? */ 911 cnt = ata_eh_nr_in_flight(ap); 912 if (!cnt) 913 return; 914 915 /* activate fast drain */ 916 ap->fastdrain_cnt = cnt; 917 ap->fastdrain_timer.expires = 918 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL); 919 add_timer(&ap->fastdrain_timer); 920 } 921 922 /** 923 * ata_qc_schedule_eh - schedule qc for error handling 924 * @qc: command to schedule error handling for 925 * 926 * Schedule error handling for @qc. EH will kick in as soon as 927 * other commands are drained. 928 * 929 * LOCKING: 930 * spin_lock_irqsave(host lock) 931 */ 932 void ata_qc_schedule_eh(struct ata_queued_cmd *qc) 933 { 934 struct ata_port *ap = qc->ap; 935 936 qc->flags |= ATA_QCFLAG_EH; 937 ata_eh_set_pending(ap, 1); 938 939 /* The following will fail if timeout has already expired. 940 * ata_scsi_error() takes care of such scmds on EH entry. 941 * Note that ATA_QCFLAG_EH is unconditionally set after 942 * this function completes. 943 */ 944 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd)); 945 } 946 947 /** 948 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine 949 * @ap: ATA port to schedule EH for 950 * 951 * LOCKING: inherited from ata_port_schedule_eh 952 * spin_lock_irqsave(host lock) 953 */ 954 void ata_std_sched_eh(struct ata_port *ap) 955 { 956 if (ap->pflags & ATA_PFLAG_INITIALIZING) 957 return; 958 959 ata_eh_set_pending(ap, 1); 960 scsi_schedule_eh(ap->scsi_host); 961 962 trace_ata_std_sched_eh(ap); 963 } 964 EXPORT_SYMBOL_GPL(ata_std_sched_eh); 965 966 /** 967 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine 968 * @ap: ATA port to end EH for 969 * 970 * In the libata object model there is a 1:1 mapping of ata_port to 971 * shost, so host fields can be directly manipulated under ap->lock, in 972 * the libsas case we need to hold a lock at the ha->level to coordinate 973 * these events. 974 * 975 * LOCKING: 976 * spin_lock_irqsave(host lock) 977 */ 978 void ata_std_end_eh(struct ata_port *ap) 979 { 980 struct Scsi_Host *host = ap->scsi_host; 981 982 host->host_eh_scheduled = 0; 983 } 984 EXPORT_SYMBOL(ata_std_end_eh); 985 986 987 /** 988 * ata_port_schedule_eh - schedule error handling without a qc 989 * @ap: ATA port to schedule EH for 990 * 991 * Schedule error handling for @ap. EH will kick in as soon as 992 * all commands are drained. 993 * 994 * LOCKING: 995 * spin_lock_irqsave(host lock) 996 */ 997 void ata_port_schedule_eh(struct ata_port *ap) 998 { 999 /* see: ata_std_sched_eh, unless you know better */ 1000 ap->ops->sched_eh(ap); 1001 } 1002 EXPORT_SYMBOL_GPL(ata_port_schedule_eh); 1003 1004 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link) 1005 { 1006 struct ata_queued_cmd *qc; 1007 int tag, nr_aborted = 0; 1008 1009 /* we're gonna abort all commands, no need for fast drain */ 1010 ata_eh_set_pending(ap, 0); 1011 1012 /* include internal tag in iteration */ 1013 ata_qc_for_each_with_internal(ap, qc, tag) { 1014 if (qc && (!link || qc->dev->link == link)) { 1015 qc->flags |= ATA_QCFLAG_EH; 1016 ata_qc_complete(qc); 1017 nr_aborted++; 1018 } 1019 } 1020 1021 if (!nr_aborted) 1022 ata_port_schedule_eh(ap); 1023 1024 return nr_aborted; 1025 } 1026 1027 /** 1028 * ata_link_abort - abort all qc's on the link 1029 * @link: ATA link to abort qc's for 1030 * 1031 * Abort all active qc's active on @link and schedule EH. 1032 * 1033 * LOCKING: 1034 * spin_lock_irqsave(host lock) 1035 * 1036 * RETURNS: 1037 * Number of aborted qc's. 1038 */ 1039 int ata_link_abort(struct ata_link *link) 1040 { 1041 return ata_do_link_abort(link->ap, link); 1042 } 1043 EXPORT_SYMBOL_GPL(ata_link_abort); 1044 1045 /** 1046 * ata_port_abort - abort all qc's on the port 1047 * @ap: ATA port to abort qc's for 1048 * 1049 * Abort all active qc's of @ap and schedule EH. 1050 * 1051 * LOCKING: 1052 * spin_lock_irqsave(host_set lock) 1053 * 1054 * RETURNS: 1055 * Number of aborted qc's. 1056 */ 1057 int ata_port_abort(struct ata_port *ap) 1058 { 1059 return ata_do_link_abort(ap, NULL); 1060 } 1061 EXPORT_SYMBOL_GPL(ata_port_abort); 1062 1063 /** 1064 * __ata_port_freeze - freeze port 1065 * @ap: ATA port to freeze 1066 * 1067 * This function is called when HSM violation or some other 1068 * condition disrupts normal operation of the port. Frozen port 1069 * is not allowed to perform any operation until the port is 1070 * thawed, which usually follows a successful reset. 1071 * 1072 * ap->ops->freeze() callback can be used for freezing the port 1073 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a 1074 * port cannot be frozen hardware-wise, the interrupt handler 1075 * must ack and clear interrupts unconditionally while the port 1076 * is frozen. 1077 * 1078 * LOCKING: 1079 * spin_lock_irqsave(host lock) 1080 */ 1081 static void __ata_port_freeze(struct ata_port *ap) 1082 { 1083 if (ap->ops->freeze) 1084 ap->ops->freeze(ap); 1085 1086 ap->pflags |= ATA_PFLAG_FROZEN; 1087 1088 trace_ata_port_freeze(ap); 1089 } 1090 1091 /** 1092 * ata_port_freeze - abort & freeze port 1093 * @ap: ATA port to freeze 1094 * 1095 * Abort and freeze @ap. The freeze operation must be called 1096 * first, because some hardware requires special operations 1097 * before the taskfile registers are accessible. 1098 * 1099 * LOCKING: 1100 * spin_lock_irqsave(host lock) 1101 * 1102 * RETURNS: 1103 * Number of aborted commands. 1104 */ 1105 int ata_port_freeze(struct ata_port *ap) 1106 { 1107 __ata_port_freeze(ap); 1108 1109 return ata_port_abort(ap); 1110 } 1111 EXPORT_SYMBOL_GPL(ata_port_freeze); 1112 1113 /** 1114 * ata_eh_freeze_port - EH helper to freeze port 1115 * @ap: ATA port to freeze 1116 * 1117 * Freeze @ap. 1118 * 1119 * LOCKING: 1120 * None. 1121 */ 1122 void ata_eh_freeze_port(struct ata_port *ap) 1123 { 1124 unsigned long flags; 1125 1126 spin_lock_irqsave(ap->lock, flags); 1127 __ata_port_freeze(ap); 1128 spin_unlock_irqrestore(ap->lock, flags); 1129 } 1130 EXPORT_SYMBOL_GPL(ata_eh_freeze_port); 1131 1132 /** 1133 * ata_eh_thaw_port - EH helper to thaw port 1134 * @ap: ATA port to thaw 1135 * 1136 * Thaw frozen port @ap. 1137 * 1138 * LOCKING: 1139 * None. 1140 */ 1141 void ata_eh_thaw_port(struct ata_port *ap) 1142 { 1143 unsigned long flags; 1144 1145 spin_lock_irqsave(ap->lock, flags); 1146 1147 ap->pflags &= ~ATA_PFLAG_FROZEN; 1148 1149 if (ap->ops->thaw) 1150 ap->ops->thaw(ap); 1151 1152 spin_unlock_irqrestore(ap->lock, flags); 1153 1154 trace_ata_port_thaw(ap); 1155 } 1156 1157 static void ata_eh_scsidone(struct scsi_cmnd *scmd) 1158 { 1159 /* nada */ 1160 } 1161 1162 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc) 1163 { 1164 struct ata_port *ap = qc->ap; 1165 struct scsi_cmnd *scmd = qc->scsicmd; 1166 unsigned long flags; 1167 1168 spin_lock_irqsave(ap->lock, flags); 1169 qc->scsidone = ata_eh_scsidone; 1170 __ata_qc_complete(qc); 1171 WARN_ON(ata_tag_valid(qc->tag)); 1172 spin_unlock_irqrestore(ap->lock, flags); 1173 1174 scsi_eh_finish_cmd(scmd, &ap->eh_done_q); 1175 } 1176 1177 /** 1178 * ata_eh_qc_complete - Complete an active ATA command from EH 1179 * @qc: Command to complete 1180 * 1181 * Indicate to the mid and upper layers that an ATA command has 1182 * completed. To be used from EH. 1183 */ 1184 void ata_eh_qc_complete(struct ata_queued_cmd *qc) 1185 { 1186 struct scsi_cmnd *scmd = qc->scsicmd; 1187 scmd->retries = scmd->allowed; 1188 __ata_eh_qc_complete(qc); 1189 } 1190 1191 /** 1192 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH 1193 * @qc: Command to retry 1194 * 1195 * Indicate to the mid and upper layers that an ATA command 1196 * should be retried. To be used from EH. 1197 * 1198 * SCSI midlayer limits the number of retries to scmd->allowed. 1199 * scmd->allowed is incremented for commands which get retried 1200 * due to unrelated failures (qc->err_mask is zero). 1201 */ 1202 void ata_eh_qc_retry(struct ata_queued_cmd *qc) 1203 { 1204 struct scsi_cmnd *scmd = qc->scsicmd; 1205 if (!qc->err_mask) 1206 scmd->allowed++; 1207 __ata_eh_qc_complete(qc); 1208 } 1209 1210 /** 1211 * ata_dev_disable - disable ATA device 1212 * @dev: ATA device to disable 1213 * 1214 * Disable @dev. 1215 * 1216 * Locking: 1217 * EH context. 1218 */ 1219 void ata_dev_disable(struct ata_device *dev) 1220 { 1221 if (!ata_dev_enabled(dev)) 1222 return; 1223 1224 ata_dev_warn(dev, "disable device\n"); 1225 ata_acpi_on_disable(dev); 1226 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET); 1227 dev->class++; 1228 1229 /* From now till the next successful probe, ering is used to 1230 * track probe failures. Clear accumulated device error info. 1231 */ 1232 ata_ering_clear(&dev->ering); 1233 } 1234 EXPORT_SYMBOL_GPL(ata_dev_disable); 1235 1236 /** 1237 * ata_eh_detach_dev - detach ATA device 1238 * @dev: ATA device to detach 1239 * 1240 * Detach @dev. 1241 * 1242 * LOCKING: 1243 * None. 1244 */ 1245 void ata_eh_detach_dev(struct ata_device *dev) 1246 { 1247 struct ata_link *link = dev->link; 1248 struct ata_port *ap = link->ap; 1249 struct ata_eh_context *ehc = &link->eh_context; 1250 unsigned long flags; 1251 1252 /* 1253 * If the device is still enabled, transition it to standby power mode 1254 * (i.e. spin down HDDs). 1255 */ 1256 if (ata_dev_enabled(dev)) 1257 ata_dev_power_set_standby(dev); 1258 1259 ata_dev_disable(dev); 1260 1261 spin_lock_irqsave(ap->lock, flags); 1262 1263 dev->flags &= ~ATA_DFLAG_DETACH; 1264 1265 if (ata_scsi_offline_dev(dev)) { 1266 dev->flags |= ATA_DFLAG_DETACHED; 1267 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; 1268 } 1269 1270 /* clear per-dev EH info */ 1271 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK); 1272 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK); 1273 ehc->saved_xfer_mode[dev->devno] = 0; 1274 ehc->saved_ncq_enabled &= ~(1 << dev->devno); 1275 1276 spin_unlock_irqrestore(ap->lock, flags); 1277 } 1278 1279 /** 1280 * ata_eh_about_to_do - about to perform eh_action 1281 * @link: target ATA link 1282 * @dev: target ATA dev for per-dev action (can be NULL) 1283 * @action: action about to be performed 1284 * 1285 * Called just before performing EH actions to clear related bits 1286 * in @link->eh_info such that eh actions are not unnecessarily 1287 * repeated. 1288 * 1289 * LOCKING: 1290 * None. 1291 */ 1292 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev, 1293 unsigned int action) 1294 { 1295 struct ata_port *ap = link->ap; 1296 struct ata_eh_info *ehi = &link->eh_info; 1297 struct ata_eh_context *ehc = &link->eh_context; 1298 unsigned long flags; 1299 1300 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action); 1301 1302 spin_lock_irqsave(ap->lock, flags); 1303 1304 ata_eh_clear_action(link, dev, ehi, action); 1305 1306 /* About to take EH action, set RECOVERED. Ignore actions on 1307 * slave links as master will do them again. 1308 */ 1309 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link) 1310 ap->pflags |= ATA_PFLAG_RECOVERED; 1311 1312 spin_unlock_irqrestore(ap->lock, flags); 1313 } 1314 1315 /** 1316 * ata_eh_done - EH action complete 1317 * @link: ATA link for which EH actions are complete 1318 * @dev: target ATA dev for per-dev action (can be NULL) 1319 * @action: action just completed 1320 * 1321 * Called right after performing EH actions to clear related bits 1322 * in @link->eh_context. 1323 * 1324 * LOCKING: 1325 * None. 1326 */ 1327 void ata_eh_done(struct ata_link *link, struct ata_device *dev, 1328 unsigned int action) 1329 { 1330 struct ata_eh_context *ehc = &link->eh_context; 1331 1332 trace_ata_eh_done(link, dev ? dev->devno : 0, action); 1333 1334 ata_eh_clear_action(link, dev, &ehc->i, action); 1335 } 1336 1337 /** 1338 * ata_err_string - convert err_mask to descriptive string 1339 * @err_mask: error mask to convert to string 1340 * 1341 * Convert @err_mask to descriptive string. Errors are 1342 * prioritized according to severity and only the most severe 1343 * error is reported. 1344 * 1345 * LOCKING: 1346 * None. 1347 * 1348 * RETURNS: 1349 * Descriptive string for @err_mask 1350 */ 1351 static const char *ata_err_string(unsigned int err_mask) 1352 { 1353 if (err_mask & AC_ERR_HOST_BUS) 1354 return "host bus error"; 1355 if (err_mask & AC_ERR_ATA_BUS) 1356 return "ATA bus error"; 1357 if (err_mask & AC_ERR_TIMEOUT) 1358 return "timeout"; 1359 if (err_mask & AC_ERR_HSM) 1360 return "HSM violation"; 1361 if (err_mask & AC_ERR_SYSTEM) 1362 return "internal error"; 1363 if (err_mask & AC_ERR_MEDIA) 1364 return "media error"; 1365 if (err_mask & AC_ERR_INVALID) 1366 return "invalid argument"; 1367 if (err_mask & AC_ERR_DEV) 1368 return "device error"; 1369 if (err_mask & AC_ERR_NCQ) 1370 return "NCQ error"; 1371 if (err_mask & AC_ERR_NODEV_HINT) 1372 return "Polling detection error"; 1373 return "unknown error"; 1374 } 1375 1376 /** 1377 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY 1378 * @dev: target ATAPI device 1379 * @r_sense_key: out parameter for sense_key 1380 * 1381 * Perform ATAPI TEST_UNIT_READY. 1382 * 1383 * LOCKING: 1384 * EH context (may sleep). 1385 * 1386 * RETURNS: 1387 * 0 on success, AC_ERR_* mask on failure. 1388 */ 1389 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key) 1390 { 1391 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 }; 1392 struct ata_taskfile tf; 1393 unsigned int err_mask; 1394 1395 ata_tf_init(dev, &tf); 1396 1397 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; 1398 tf.command = ATA_CMD_PACKET; 1399 tf.protocol = ATAPI_PROT_NODATA; 1400 1401 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0); 1402 if (err_mask == AC_ERR_DEV) 1403 *r_sense_key = tf.error >> 4; 1404 return err_mask; 1405 } 1406 1407 /** 1408 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT 1409 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to 1410 * 1411 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK 1412 * SENSE. This function is an EH helper. 1413 * 1414 * LOCKING: 1415 * Kernel thread context (may sleep). 1416 * 1417 * RETURNS: 1418 * true if sense data could be fetched, false otherwise. 1419 */ 1420 static bool ata_eh_request_sense(struct ata_queued_cmd *qc) 1421 { 1422 struct scsi_cmnd *cmd = qc->scsicmd; 1423 struct ata_device *dev = qc->dev; 1424 struct ata_taskfile tf; 1425 unsigned int err_mask; 1426 1427 if (ata_port_is_frozen(qc->ap)) { 1428 ata_dev_warn(dev, "sense data available but port frozen\n"); 1429 return false; 1430 } 1431 1432 if (!ata_id_sense_reporting_enabled(dev->id)) { 1433 ata_dev_warn(qc->dev, "sense data reporting disabled\n"); 1434 return false; 1435 } 1436 1437 ata_tf_init(dev, &tf); 1438 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; 1439 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48; 1440 tf.command = ATA_CMD_REQ_SENSE_DATA; 1441 tf.protocol = ATA_PROT_NODATA; 1442 1443 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0); 1444 /* Ignore err_mask; ATA_ERR might be set */ 1445 if (tf.status & ATA_SENSE) { 1446 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) { 1447 /* Set sense without also setting scsicmd->result */ 1448 scsi_build_sense_buffer(dev->flags & ATA_DFLAG_D_SENSE, 1449 cmd->sense_buffer, tf.lbah, 1450 tf.lbam, tf.lbal); 1451 qc->flags |= ATA_QCFLAG_SENSE_VALID; 1452 return true; 1453 } 1454 } else { 1455 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n", 1456 tf.status, err_mask); 1457 } 1458 1459 return false; 1460 } 1461 1462 /** 1463 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE 1464 * @dev: device to perform REQUEST_SENSE to 1465 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long) 1466 * @dfl_sense_key: default sense key to use 1467 * 1468 * Perform ATAPI REQUEST_SENSE after the device reported CHECK 1469 * SENSE. This function is EH helper. 1470 * 1471 * LOCKING: 1472 * Kernel thread context (may sleep). 1473 * 1474 * RETURNS: 1475 * 0 on success, AC_ERR_* mask on failure 1476 */ 1477 unsigned int atapi_eh_request_sense(struct ata_device *dev, 1478 u8 *sense_buf, u8 dfl_sense_key) 1479 { 1480 u8 cdb[ATAPI_CDB_LEN] = 1481 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 }; 1482 struct ata_port *ap = dev->link->ap; 1483 struct ata_taskfile tf; 1484 1485 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE); 1486 1487 /* initialize sense_buf with the error register, 1488 * for the case where they are -not- overwritten 1489 */ 1490 sense_buf[0] = 0x70; 1491 sense_buf[2] = dfl_sense_key; 1492 1493 /* some devices time out if garbage left in tf */ 1494 ata_tf_init(dev, &tf); 1495 1496 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; 1497 tf.command = ATA_CMD_PACKET; 1498 1499 /* is it pointless to prefer PIO for "safety reasons"? */ 1500 if (ap->flags & ATA_FLAG_PIO_DMA) { 1501 tf.protocol = ATAPI_PROT_DMA; 1502 tf.feature |= ATAPI_PKT_DMA; 1503 } else { 1504 tf.protocol = ATAPI_PROT_PIO; 1505 tf.lbam = SCSI_SENSE_BUFFERSIZE; 1506 tf.lbah = 0; 1507 } 1508 1509 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE, 1510 sense_buf, SCSI_SENSE_BUFFERSIZE, 0); 1511 } 1512 1513 /** 1514 * ata_eh_analyze_serror - analyze SError for a failed port 1515 * @link: ATA link to analyze SError for 1516 * 1517 * Analyze SError if available and further determine cause of 1518 * failure. 1519 * 1520 * LOCKING: 1521 * None. 1522 */ 1523 static void ata_eh_analyze_serror(struct ata_link *link) 1524 { 1525 struct ata_eh_context *ehc = &link->eh_context; 1526 u32 serror = ehc->i.serror; 1527 unsigned int err_mask = 0, action = 0; 1528 u32 hotplug_mask; 1529 1530 if (serror & (SERR_PERSISTENT | SERR_DATA)) { 1531 err_mask |= AC_ERR_ATA_BUS; 1532 action |= ATA_EH_RESET; 1533 } 1534 if (serror & SERR_PROTOCOL) { 1535 err_mask |= AC_ERR_HSM; 1536 action |= ATA_EH_RESET; 1537 } 1538 if (serror & SERR_INTERNAL) { 1539 err_mask |= AC_ERR_SYSTEM; 1540 action |= ATA_EH_RESET; 1541 } 1542 1543 /* Determine whether a hotplug event has occurred. Both 1544 * SError.N/X are considered hotplug events for enabled or 1545 * host links. For disabled PMP links, only N bit is 1546 * considered as X bit is left at 1 for link plugging. 1547 */ 1548 if (link->lpm_policy > ATA_LPM_MAX_POWER) 1549 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */ 1550 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link)) 1551 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG; 1552 else 1553 hotplug_mask = SERR_PHYRDY_CHG; 1554 1555 if (serror & hotplug_mask) 1556 ata_ehi_hotplugged(&ehc->i); 1557 1558 ehc->i.err_mask |= err_mask; 1559 ehc->i.action |= action; 1560 } 1561 1562 /** 1563 * ata_eh_analyze_tf - analyze taskfile of a failed qc 1564 * @qc: qc to analyze 1565 * 1566 * Analyze taskfile of @qc and further determine cause of 1567 * failure. This function also requests ATAPI sense data if 1568 * available. 1569 * 1570 * LOCKING: 1571 * Kernel thread context (may sleep). 1572 * 1573 * RETURNS: 1574 * Determined recovery action 1575 */ 1576 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc) 1577 { 1578 const struct ata_taskfile *tf = &qc->result_tf; 1579 unsigned int tmp, action = 0; 1580 u8 stat = tf->status, err = tf->error; 1581 1582 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) { 1583 qc->err_mask |= AC_ERR_HSM; 1584 return ATA_EH_RESET; 1585 } 1586 1587 if (stat & (ATA_ERR | ATA_DF)) { 1588 qc->err_mask |= AC_ERR_DEV; 1589 /* 1590 * Sense data reporting does not work if the 1591 * device fault bit is set. 1592 */ 1593 if (stat & ATA_DF) 1594 stat &= ~ATA_SENSE; 1595 } else { 1596 return 0; 1597 } 1598 1599 switch (qc->dev->class) { 1600 case ATA_DEV_ATA: 1601 case ATA_DEV_ZAC: 1602 /* 1603 * Fetch the sense data explicitly if: 1604 * -It was a non-NCQ command that failed, or 1605 * -It was a NCQ command that failed, but the sense data 1606 * was not included in the NCQ command error log 1607 * (i.e. NCQ autosense is not supported by the device). 1608 */ 1609 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) && 1610 (stat & ATA_SENSE) && ata_eh_request_sense(qc)) 1611 set_status_byte(qc->scsicmd, SAM_STAT_CHECK_CONDITION); 1612 if (err & ATA_ICRC) 1613 qc->err_mask |= AC_ERR_ATA_BUS; 1614 if (err & (ATA_UNC | ATA_AMNF)) 1615 qc->err_mask |= AC_ERR_MEDIA; 1616 if (err & ATA_IDNF) 1617 qc->err_mask |= AC_ERR_INVALID; 1618 break; 1619 1620 case ATA_DEV_ATAPI: 1621 if (!ata_port_is_frozen(qc->ap)) { 1622 tmp = atapi_eh_request_sense(qc->dev, 1623 qc->scsicmd->sense_buffer, 1624 qc->result_tf.error >> 4); 1625 if (!tmp) 1626 qc->flags |= ATA_QCFLAG_SENSE_VALID; 1627 else 1628 qc->err_mask |= tmp; 1629 } 1630 } 1631 1632 if (qc->flags & ATA_QCFLAG_SENSE_VALID) { 1633 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd); 1634 /* 1635 * SUCCESS here means that the sense code could be 1636 * evaluated and should be passed to the upper layers 1637 * for correct evaluation. 1638 * FAILED means the sense code could not be interpreted 1639 * and the device would need to be reset. 1640 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the 1641 * command would need to be retried. 1642 */ 1643 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) { 1644 qc->flags |= ATA_QCFLAG_RETRY; 1645 qc->err_mask |= AC_ERR_OTHER; 1646 } else if (ret != SUCCESS) { 1647 qc->err_mask |= AC_ERR_HSM; 1648 } 1649 } 1650 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS)) 1651 action |= ATA_EH_RESET; 1652 1653 return action; 1654 } 1655 1656 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask, 1657 int *xfer_ok) 1658 { 1659 int base = 0; 1660 1661 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER)) 1662 *xfer_ok = 1; 1663 1664 if (!*xfer_ok) 1665 base = ATA_ECAT_DUBIOUS_NONE; 1666 1667 if (err_mask & AC_ERR_ATA_BUS) 1668 return base + ATA_ECAT_ATA_BUS; 1669 1670 if (err_mask & AC_ERR_TIMEOUT) 1671 return base + ATA_ECAT_TOUT_HSM; 1672 1673 if (eflags & ATA_EFLAG_IS_IO) { 1674 if (err_mask & AC_ERR_HSM) 1675 return base + ATA_ECAT_TOUT_HSM; 1676 if ((err_mask & 1677 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV) 1678 return base + ATA_ECAT_UNK_DEV; 1679 } 1680 1681 return 0; 1682 } 1683 1684 struct speed_down_verdict_arg { 1685 u64 since; 1686 int xfer_ok; 1687 int nr_errors[ATA_ECAT_NR]; 1688 }; 1689 1690 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg) 1691 { 1692 struct speed_down_verdict_arg *arg = void_arg; 1693 int cat; 1694 1695 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since)) 1696 return -1; 1697 1698 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask, 1699 &arg->xfer_ok); 1700 arg->nr_errors[cat]++; 1701 1702 return 0; 1703 } 1704 1705 /** 1706 * ata_eh_speed_down_verdict - Determine speed down verdict 1707 * @dev: Device of interest 1708 * 1709 * This function examines error ring of @dev and determines 1710 * whether NCQ needs to be turned off, transfer speed should be 1711 * stepped down, or falling back to PIO is necessary. 1712 * 1713 * ECAT_ATA_BUS : ATA_BUS error for any command 1714 * 1715 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for 1716 * IO commands 1717 * 1718 * ECAT_UNK_DEV : Unknown DEV error for IO commands 1719 * 1720 * ECAT_DUBIOUS_* : Identical to above three but occurred while 1721 * data transfer hasn't been verified. 1722 * 1723 * Verdicts are 1724 * 1725 * NCQ_OFF : Turn off NCQ. 1726 * 1727 * SPEED_DOWN : Speed down transfer speed but don't fall back 1728 * to PIO. 1729 * 1730 * FALLBACK_TO_PIO : Fall back to PIO. 1731 * 1732 * Even if multiple verdicts are returned, only one action is 1733 * taken per error. An action triggered by non-DUBIOUS errors 1734 * clears ering, while one triggered by DUBIOUS_* errors doesn't. 1735 * This is to expedite speed down decisions right after device is 1736 * initially configured. 1737 * 1738 * The following are speed down rules. #1 and #2 deal with 1739 * DUBIOUS errors. 1740 * 1741 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors 1742 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO. 1743 * 1744 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors 1745 * occurred during last 5 mins, NCQ_OFF. 1746 * 1747 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors 1748 * occurred during last 5 mins, FALLBACK_TO_PIO 1749 * 1750 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred 1751 * during last 10 mins, NCQ_OFF. 1752 * 1753 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6 1754 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN. 1755 * 1756 * LOCKING: 1757 * Inherited from caller. 1758 * 1759 * RETURNS: 1760 * OR of ATA_EH_SPDN_* flags. 1761 */ 1762 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev) 1763 { 1764 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ; 1765 u64 j64 = get_jiffies_64(); 1766 struct speed_down_verdict_arg arg; 1767 unsigned int verdict = 0; 1768 1769 /* scan past 5 mins of error history */ 1770 memset(&arg, 0, sizeof(arg)); 1771 arg.since = j64 - min(j64, j5mins); 1772 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg); 1773 1774 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] + 1775 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1) 1776 verdict |= ATA_EH_SPDN_SPEED_DOWN | 1777 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS; 1778 1779 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] + 1780 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1) 1781 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS; 1782 1783 if (arg.nr_errors[ATA_ECAT_ATA_BUS] + 1784 arg.nr_errors[ATA_ECAT_TOUT_HSM] + 1785 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6) 1786 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO; 1787 1788 /* scan past 10 mins of error history */ 1789 memset(&arg, 0, sizeof(arg)); 1790 arg.since = j64 - min(j64, j10mins); 1791 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg); 1792 1793 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] + 1794 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3) 1795 verdict |= ATA_EH_SPDN_NCQ_OFF; 1796 1797 if (arg.nr_errors[ATA_ECAT_ATA_BUS] + 1798 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 || 1799 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6) 1800 verdict |= ATA_EH_SPDN_SPEED_DOWN; 1801 1802 return verdict; 1803 } 1804 1805 /** 1806 * ata_eh_speed_down - record error and speed down if necessary 1807 * @dev: Failed device 1808 * @eflags: mask of ATA_EFLAG_* flags 1809 * @err_mask: err_mask of the error 1810 * 1811 * Record error and examine error history to determine whether 1812 * adjusting transmission speed is necessary. It also sets 1813 * transmission limits appropriately if such adjustment is 1814 * necessary. 1815 * 1816 * LOCKING: 1817 * Kernel thread context (may sleep). 1818 * 1819 * RETURNS: 1820 * Determined recovery action. 1821 */ 1822 static unsigned int ata_eh_speed_down(struct ata_device *dev, 1823 unsigned int eflags, unsigned int err_mask) 1824 { 1825 struct ata_link *link = ata_dev_phys_link(dev); 1826 int xfer_ok = 0; 1827 unsigned int verdict; 1828 unsigned int action = 0; 1829 1830 /* don't bother if Cat-0 error */ 1831 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0) 1832 return 0; 1833 1834 /* record error and determine whether speed down is necessary */ 1835 ata_ering_record(&dev->ering, eflags, err_mask); 1836 verdict = ata_eh_speed_down_verdict(dev); 1837 1838 /* turn off NCQ? */ 1839 if ((verdict & ATA_EH_SPDN_NCQ_OFF) && ata_ncq_enabled(dev)) { 1840 dev->flags |= ATA_DFLAG_NCQ_OFF; 1841 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n"); 1842 goto done; 1843 } 1844 1845 /* speed down? */ 1846 if (verdict & ATA_EH_SPDN_SPEED_DOWN) { 1847 /* speed down SATA link speed if possible */ 1848 if (sata_down_spd_limit(link, 0) == 0) { 1849 action |= ATA_EH_RESET; 1850 goto done; 1851 } 1852 1853 /* lower transfer mode */ 1854 if (dev->spdn_cnt < 2) { 1855 static const int dma_dnxfer_sel[] = 1856 { ATA_DNXFER_DMA, ATA_DNXFER_40C }; 1857 static const int pio_dnxfer_sel[] = 1858 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 }; 1859 int sel; 1860 1861 if (dev->xfer_shift != ATA_SHIFT_PIO) 1862 sel = dma_dnxfer_sel[dev->spdn_cnt]; 1863 else 1864 sel = pio_dnxfer_sel[dev->spdn_cnt]; 1865 1866 dev->spdn_cnt++; 1867 1868 if (ata_down_xfermask_limit(dev, sel) == 0) { 1869 action |= ATA_EH_RESET; 1870 goto done; 1871 } 1872 } 1873 } 1874 1875 /* Fall back to PIO? Slowing down to PIO is meaningless for 1876 * SATA ATA devices. Consider it only for PATA and SATAPI. 1877 */ 1878 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) && 1879 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) && 1880 (dev->xfer_shift != ATA_SHIFT_PIO)) { 1881 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) { 1882 dev->spdn_cnt = 0; 1883 action |= ATA_EH_RESET; 1884 goto done; 1885 } 1886 } 1887 1888 return 0; 1889 done: 1890 /* device has been slowed down, blow error history */ 1891 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS)) 1892 ata_ering_clear(&dev->ering); 1893 return action; 1894 } 1895 1896 /** 1897 * ata_eh_worth_retry - analyze error and decide whether to retry 1898 * @qc: qc to possibly retry 1899 * 1900 * Look at the cause of the error and decide if a retry 1901 * might be useful or not. We don't want to retry media errors 1902 * because the drive itself has probably already taken 10-30 seconds 1903 * doing its own internal retries before reporting the failure. 1904 */ 1905 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc) 1906 { 1907 if (qc->err_mask & AC_ERR_MEDIA) 1908 return 0; /* don't retry media errors */ 1909 if (qc->flags & ATA_QCFLAG_IO) 1910 return 1; /* otherwise retry anything from fs stack */ 1911 if (qc->err_mask & AC_ERR_INVALID) 1912 return 0; /* don't retry these */ 1913 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */ 1914 } 1915 1916 /** 1917 * ata_eh_quiet - check if we need to be quiet about a command error 1918 * @qc: qc to check 1919 * 1920 * Look at the qc flags anbd its scsi command request flags to determine 1921 * if we need to be quiet about the command failure. 1922 */ 1923 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc) 1924 { 1925 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET) 1926 qc->flags |= ATA_QCFLAG_QUIET; 1927 return qc->flags & ATA_QCFLAG_QUIET; 1928 } 1929 1930 static int ata_eh_read_sense_success_non_ncq(struct ata_link *link) 1931 { 1932 struct ata_port *ap = link->ap; 1933 struct ata_queued_cmd *qc; 1934 1935 qc = __ata_qc_from_tag(ap, link->active_tag); 1936 if (!qc) 1937 return -EIO; 1938 1939 if (!(qc->flags & ATA_QCFLAG_EH) || 1940 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) || 1941 qc->err_mask) 1942 return -EIO; 1943 1944 if (!ata_eh_request_sense(qc)) 1945 return -EIO; 1946 1947 /* 1948 * If we have sense data, call scsi_check_sense() in order to set the 1949 * correct SCSI ML byte (if any). No point in checking the return value, 1950 * since the command has already completed successfully. 1951 */ 1952 scsi_check_sense(qc->scsicmd); 1953 1954 return 0; 1955 } 1956 1957 static void ata_eh_get_success_sense(struct ata_link *link) 1958 { 1959 struct ata_eh_context *ehc = &link->eh_context; 1960 struct ata_device *dev = link->device; 1961 struct ata_port *ap = link->ap; 1962 struct ata_queued_cmd *qc; 1963 int tag, ret = 0; 1964 1965 if (!(ehc->i.dev_action[dev->devno] & ATA_EH_GET_SUCCESS_SENSE)) 1966 return; 1967 1968 /* if frozen, we can't do much */ 1969 if (ata_port_is_frozen(ap)) { 1970 ata_dev_warn(dev, 1971 "successful sense data available but port frozen\n"); 1972 goto out; 1973 } 1974 1975 /* 1976 * If the link has sactive set, then we have outstanding NCQ commands 1977 * and have to read the Successful NCQ Commands log to get the sense 1978 * data. Otherwise, we are dealing with a non-NCQ command and use 1979 * request sense ext command to retrieve the sense data. 1980 */ 1981 if (link->sactive) 1982 ret = ata_eh_read_sense_success_ncq_log(link); 1983 else 1984 ret = ata_eh_read_sense_success_non_ncq(link); 1985 if (ret) 1986 goto out; 1987 1988 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE); 1989 return; 1990 1991 out: 1992 /* 1993 * If we failed to get sense data for a successful command that ought to 1994 * have sense data, we cannot simply return BLK_STS_OK to user space. 1995 * This is because we can't know if the sense data that we couldn't get 1996 * was actually "DATA CURRENTLY UNAVAILABLE". Reporting such a command 1997 * as success to user space would result in a silent data corruption. 1998 * Thus, add a bogus ABORTED_COMMAND sense data to such commands, such 1999 * that SCSI will report these commands as BLK_STS_IOERR to user space. 2000 */ 2001 ata_qc_for_each_raw(ap, qc, tag) { 2002 if (!(qc->flags & ATA_QCFLAG_EH) || 2003 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) || 2004 qc->err_mask || 2005 ata_dev_phys_link(qc->dev) != link) 2006 continue; 2007 2008 /* We managed to get sense for this success command, skip. */ 2009 if (qc->flags & ATA_QCFLAG_SENSE_VALID) 2010 continue; 2011 2012 /* This success command did not have any sense data, skip. */ 2013 if (!(qc->result_tf.status & ATA_SENSE)) 2014 continue; 2015 2016 /* This success command had sense data, but we failed to get. */ 2017 ata_scsi_set_sense(dev, qc->scsicmd, ABORTED_COMMAND, 0, 0); 2018 qc->flags |= ATA_QCFLAG_SENSE_VALID; 2019 } 2020 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE); 2021 } 2022 2023 /** 2024 * ata_eh_link_autopsy - analyze error and determine recovery action 2025 * @link: host link to perform autopsy on 2026 * 2027 * Analyze why @link failed and determine which recovery actions 2028 * are needed. This function also sets more detailed AC_ERR_* 2029 * values and fills sense data for ATAPI CHECK SENSE. 2030 * 2031 * LOCKING: 2032 * Kernel thread context (may sleep). 2033 */ 2034 static void ata_eh_link_autopsy(struct ata_link *link) 2035 { 2036 struct ata_port *ap = link->ap; 2037 struct ata_eh_context *ehc = &link->eh_context; 2038 struct ata_queued_cmd *qc; 2039 struct ata_device *dev; 2040 unsigned int all_err_mask = 0, eflags = 0; 2041 int tag, nr_failed = 0, nr_quiet = 0; 2042 u32 serror; 2043 int rc; 2044 2045 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY) 2046 return; 2047 2048 /* obtain and analyze SError */ 2049 rc = sata_scr_read(link, SCR_ERROR, &serror); 2050 if (rc == 0) { 2051 ehc->i.serror |= serror; 2052 ata_eh_analyze_serror(link); 2053 } else if (rc != -EOPNOTSUPP) { 2054 /* SError read failed, force reset and probing */ 2055 ehc->i.probe_mask |= ATA_ALL_DEVICES; 2056 ehc->i.action |= ATA_EH_RESET; 2057 ehc->i.err_mask |= AC_ERR_OTHER; 2058 } 2059 2060 /* analyze NCQ failure */ 2061 ata_eh_analyze_ncq_error(link); 2062 2063 /* 2064 * Check if this was a successful command that simply needs sense data. 2065 * Since the sense data is not part of the completion, we need to fetch 2066 * it using an additional command. Since this can't be done from irq 2067 * context, the sense data for successful commands are fetched by EH. 2068 */ 2069 ata_eh_get_success_sense(link); 2070 2071 /* any real error trumps AC_ERR_OTHER */ 2072 if (ehc->i.err_mask & ~AC_ERR_OTHER) 2073 ehc->i.err_mask &= ~AC_ERR_OTHER; 2074 2075 all_err_mask |= ehc->i.err_mask; 2076 2077 ata_qc_for_each_raw(ap, qc, tag) { 2078 if (!(qc->flags & ATA_QCFLAG_EH) || 2079 qc->flags & ATA_QCFLAG_RETRY || 2080 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD || 2081 ata_dev_phys_link(qc->dev) != link) 2082 continue; 2083 2084 /* inherit upper level err_mask */ 2085 qc->err_mask |= ehc->i.err_mask; 2086 2087 /* analyze TF */ 2088 ehc->i.action |= ata_eh_analyze_tf(qc); 2089 2090 /* DEV errors are probably spurious in case of ATA_BUS error */ 2091 if (qc->err_mask & AC_ERR_ATA_BUS) 2092 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA | 2093 AC_ERR_INVALID); 2094 2095 /* any real error trumps unknown error */ 2096 if (qc->err_mask & ~AC_ERR_OTHER) 2097 qc->err_mask &= ~AC_ERR_OTHER; 2098 2099 /* 2100 * SENSE_VALID trumps dev/unknown error and revalidation. Upper 2101 * layers will determine whether the command is worth retrying 2102 * based on the sense data and device class/type. Otherwise, 2103 * determine directly if the command is worth retrying using its 2104 * error mask and flags. 2105 */ 2106 if (qc->flags & ATA_QCFLAG_SENSE_VALID) 2107 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER); 2108 else if (ata_eh_worth_retry(qc)) 2109 qc->flags |= ATA_QCFLAG_RETRY; 2110 2111 /* accumulate error info */ 2112 ehc->i.dev = qc->dev; 2113 all_err_mask |= qc->err_mask; 2114 if (qc->flags & ATA_QCFLAG_IO) 2115 eflags |= ATA_EFLAG_IS_IO; 2116 trace_ata_eh_link_autopsy_qc(qc); 2117 2118 /* Count quiet errors */ 2119 if (ata_eh_quiet(qc)) 2120 nr_quiet++; 2121 nr_failed++; 2122 } 2123 2124 /* If all failed commands requested silence, then be quiet */ 2125 if (nr_quiet == nr_failed) 2126 ehc->i.flags |= ATA_EHI_QUIET; 2127 2128 /* enforce default EH actions */ 2129 if (ata_port_is_frozen(ap) || 2130 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT)) 2131 ehc->i.action |= ATA_EH_RESET; 2132 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) || 2133 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV))) 2134 ehc->i.action |= ATA_EH_REVALIDATE; 2135 2136 /* If we have offending qcs and the associated failed device, 2137 * perform per-dev EH action only on the offending device. 2138 */ 2139 if (ehc->i.dev) { 2140 ehc->i.dev_action[ehc->i.dev->devno] |= 2141 ehc->i.action & ATA_EH_PERDEV_MASK; 2142 ehc->i.action &= ~ATA_EH_PERDEV_MASK; 2143 } 2144 2145 /* propagate timeout to host link */ 2146 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link)) 2147 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT; 2148 2149 /* record error and consider speeding down */ 2150 dev = ehc->i.dev; 2151 if (!dev && ((ata_link_max_devices(link) == 1 && 2152 ata_dev_enabled(link->device)))) 2153 dev = link->device; 2154 2155 if (dev) { 2156 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER) 2157 eflags |= ATA_EFLAG_DUBIOUS_XFER; 2158 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask); 2159 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask); 2160 } 2161 } 2162 2163 /** 2164 * ata_eh_autopsy - analyze error and determine recovery action 2165 * @ap: host port to perform autopsy on 2166 * 2167 * Analyze all links of @ap and determine why they failed and 2168 * which recovery actions are needed. 2169 * 2170 * LOCKING: 2171 * Kernel thread context (may sleep). 2172 */ 2173 void ata_eh_autopsy(struct ata_port *ap) 2174 { 2175 struct ata_link *link; 2176 2177 ata_for_each_link(link, ap, EDGE) 2178 ata_eh_link_autopsy(link); 2179 2180 /* Handle the frigging slave link. Autopsy is done similarly 2181 * but actions and flags are transferred over to the master 2182 * link and handled from there. 2183 */ 2184 if (ap->slave_link) { 2185 struct ata_eh_context *mehc = &ap->link.eh_context; 2186 struct ata_eh_context *sehc = &ap->slave_link->eh_context; 2187 2188 /* transfer control flags from master to slave */ 2189 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK; 2190 2191 /* perform autopsy on the slave link */ 2192 ata_eh_link_autopsy(ap->slave_link); 2193 2194 /* transfer actions from slave to master and clear slave */ 2195 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS); 2196 mehc->i.action |= sehc->i.action; 2197 mehc->i.dev_action[1] |= sehc->i.dev_action[1]; 2198 mehc->i.flags |= sehc->i.flags; 2199 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS); 2200 } 2201 2202 /* Autopsy of fanout ports can affect host link autopsy. 2203 * Perform host link autopsy last. 2204 */ 2205 if (sata_pmp_attached(ap)) 2206 ata_eh_link_autopsy(&ap->link); 2207 } 2208 2209 /** 2210 * ata_get_cmd_name - get name for ATA command 2211 * @command: ATA command code to get name for 2212 * 2213 * Return a textual name of the given command or "unknown" 2214 * 2215 * LOCKING: 2216 * None 2217 */ 2218 const char *ata_get_cmd_name(u8 command) 2219 { 2220 #ifdef CONFIG_ATA_VERBOSE_ERROR 2221 static const struct 2222 { 2223 u8 command; 2224 const char *text; 2225 } cmd_descr[] = { 2226 { ATA_CMD_DEV_RESET, "DEVICE RESET" }, 2227 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" }, 2228 { ATA_CMD_STANDBY, "STANDBY" }, 2229 { ATA_CMD_IDLE, "IDLE" }, 2230 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" }, 2231 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" }, 2232 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" }, 2233 { ATA_CMD_NOP, "NOP" }, 2234 { ATA_CMD_FLUSH, "FLUSH CACHE" }, 2235 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" }, 2236 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" }, 2237 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" }, 2238 { ATA_CMD_SERVICE, "SERVICE" }, 2239 { ATA_CMD_READ, "READ DMA" }, 2240 { ATA_CMD_READ_EXT, "READ DMA EXT" }, 2241 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" }, 2242 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" }, 2243 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" }, 2244 { ATA_CMD_WRITE, "WRITE DMA" }, 2245 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" }, 2246 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" }, 2247 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" }, 2248 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" }, 2249 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" }, 2250 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" }, 2251 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" }, 2252 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" }, 2253 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" }, 2254 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" }, 2255 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" }, 2256 { ATA_CMD_PIO_READ, "READ SECTOR(S)" }, 2257 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" }, 2258 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" }, 2259 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" }, 2260 { ATA_CMD_READ_MULTI, "READ MULTIPLE" }, 2261 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" }, 2262 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" }, 2263 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" }, 2264 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" }, 2265 { ATA_CMD_SET_FEATURES, "SET FEATURES" }, 2266 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" }, 2267 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" }, 2268 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" }, 2269 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" }, 2270 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" }, 2271 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" }, 2272 { ATA_CMD_SLEEP, "SLEEP" }, 2273 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" }, 2274 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" }, 2275 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" }, 2276 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" }, 2277 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" }, 2278 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" }, 2279 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" }, 2280 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" }, 2281 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" }, 2282 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" }, 2283 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" }, 2284 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" }, 2285 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" }, 2286 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" }, 2287 { ATA_CMD_PMP_READ, "READ BUFFER" }, 2288 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" }, 2289 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" }, 2290 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" }, 2291 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" }, 2292 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" }, 2293 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" }, 2294 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" }, 2295 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" }, 2296 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" }, 2297 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" }, 2298 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" }, 2299 { ATA_CMD_SMART, "SMART" }, 2300 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" }, 2301 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" }, 2302 { ATA_CMD_DSM, "DATA SET MANAGEMENT" }, 2303 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" }, 2304 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" }, 2305 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" }, 2306 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" }, 2307 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" }, 2308 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" }, 2309 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" }, 2310 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" }, 2311 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" }, 2312 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" }, 2313 { ATA_CMD_READ_LONG, "READ LONG (with retries)" }, 2314 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" }, 2315 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" }, 2316 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" }, 2317 { ATA_CMD_RESTORE, "RECALIBRATE" }, 2318 { 0, NULL } /* terminate list */ 2319 }; 2320 2321 unsigned int i; 2322 for (i = 0; cmd_descr[i].text; i++) 2323 if (cmd_descr[i].command == command) 2324 return cmd_descr[i].text; 2325 #endif 2326 2327 return "unknown"; 2328 } 2329 EXPORT_SYMBOL_GPL(ata_get_cmd_name); 2330 2331 /** 2332 * ata_eh_link_report - report error handling to user 2333 * @link: ATA link EH is going on 2334 * 2335 * Report EH to user. 2336 * 2337 * LOCKING: 2338 * None. 2339 */ 2340 static void ata_eh_link_report(struct ata_link *link) 2341 { 2342 struct ata_port *ap = link->ap; 2343 struct ata_eh_context *ehc = &link->eh_context; 2344 struct ata_queued_cmd *qc; 2345 const char *frozen, *desc; 2346 char tries_buf[16] = ""; 2347 int tag, nr_failed = 0; 2348 2349 if (ehc->i.flags & ATA_EHI_QUIET) 2350 return; 2351 2352 desc = NULL; 2353 if (ehc->i.desc[0] != '\0') 2354 desc = ehc->i.desc; 2355 2356 ata_qc_for_each_raw(ap, qc, tag) { 2357 if (!(qc->flags & ATA_QCFLAG_EH) || 2358 ata_dev_phys_link(qc->dev) != link || 2359 ((qc->flags & ATA_QCFLAG_QUIET) && 2360 qc->err_mask == AC_ERR_DEV)) 2361 continue; 2362 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask) 2363 continue; 2364 2365 nr_failed++; 2366 } 2367 2368 if (!nr_failed && !ehc->i.err_mask) 2369 return; 2370 2371 frozen = ""; 2372 if (ata_port_is_frozen(ap)) 2373 frozen = " frozen"; 2374 2375 if (ap->eh_tries < ATA_EH_MAX_TRIES) 2376 snprintf(tries_buf, sizeof(tries_buf), " t%d", 2377 ap->eh_tries); 2378 2379 if (ehc->i.dev) { 2380 ata_dev_err(ehc->i.dev, "exception Emask 0x%x " 2381 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n", 2382 ehc->i.err_mask, link->sactive, ehc->i.serror, 2383 ehc->i.action, frozen, tries_buf); 2384 if (desc) 2385 ata_dev_err(ehc->i.dev, "%s\n", desc); 2386 } else { 2387 ata_link_err(link, "exception Emask 0x%x " 2388 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n", 2389 ehc->i.err_mask, link->sactive, ehc->i.serror, 2390 ehc->i.action, frozen, tries_buf); 2391 if (desc) 2392 ata_link_err(link, "%s\n", desc); 2393 } 2394 2395 #ifdef CONFIG_ATA_VERBOSE_ERROR 2396 if (ehc->i.serror) 2397 ata_link_err(link, 2398 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n", 2399 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "", 2400 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "", 2401 ehc->i.serror & SERR_DATA ? "UnrecovData " : "", 2402 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "", 2403 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "", 2404 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "", 2405 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "", 2406 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "", 2407 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "", 2408 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "", 2409 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "", 2410 ehc->i.serror & SERR_CRC ? "BadCRC " : "", 2411 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "", 2412 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "", 2413 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "", 2414 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "", 2415 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : ""); 2416 #endif 2417 2418 ata_qc_for_each_raw(ap, qc, tag) { 2419 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf; 2420 char data_buf[20] = ""; 2421 char cdb_buf[70] = ""; 2422 2423 if (!(qc->flags & ATA_QCFLAG_EH) || 2424 ata_dev_phys_link(qc->dev) != link || !qc->err_mask) 2425 continue; 2426 2427 if (qc->dma_dir != DMA_NONE) { 2428 static const char *dma_str[] = { 2429 [DMA_BIDIRECTIONAL] = "bidi", 2430 [DMA_TO_DEVICE] = "out", 2431 [DMA_FROM_DEVICE] = "in", 2432 }; 2433 const char *prot_str = NULL; 2434 2435 switch (qc->tf.protocol) { 2436 case ATA_PROT_UNKNOWN: 2437 prot_str = "unknown"; 2438 break; 2439 case ATA_PROT_NODATA: 2440 prot_str = "nodata"; 2441 break; 2442 case ATA_PROT_PIO: 2443 prot_str = "pio"; 2444 break; 2445 case ATA_PROT_DMA: 2446 prot_str = "dma"; 2447 break; 2448 case ATA_PROT_NCQ: 2449 prot_str = "ncq dma"; 2450 break; 2451 case ATA_PROT_NCQ_NODATA: 2452 prot_str = "ncq nodata"; 2453 break; 2454 case ATAPI_PROT_NODATA: 2455 prot_str = "nodata"; 2456 break; 2457 case ATAPI_PROT_PIO: 2458 prot_str = "pio"; 2459 break; 2460 case ATAPI_PROT_DMA: 2461 prot_str = "dma"; 2462 break; 2463 } 2464 snprintf(data_buf, sizeof(data_buf), " %s %u %s", 2465 prot_str, qc->nbytes, dma_str[qc->dma_dir]); 2466 } 2467 2468 if (ata_is_atapi(qc->tf.protocol)) { 2469 const u8 *cdb = qc->cdb; 2470 size_t cdb_len = qc->dev->cdb_len; 2471 2472 if (qc->scsicmd) { 2473 cdb = qc->scsicmd->cmnd; 2474 cdb_len = qc->scsicmd->cmd_len; 2475 } 2476 __scsi_format_command(cdb_buf, sizeof(cdb_buf), 2477 cdb, cdb_len); 2478 } else 2479 ata_dev_err(qc->dev, "failed command: %s\n", 2480 ata_get_cmd_name(cmd->command)); 2481 2482 ata_dev_err(qc->dev, 2483 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " 2484 "tag %d%s\n %s" 2485 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " 2486 "Emask 0x%x (%s)%s\n", 2487 cmd->command, cmd->feature, cmd->nsect, 2488 cmd->lbal, cmd->lbam, cmd->lbah, 2489 cmd->hob_feature, cmd->hob_nsect, 2490 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah, 2491 cmd->device, qc->tag, data_buf, cdb_buf, 2492 res->status, res->error, res->nsect, 2493 res->lbal, res->lbam, res->lbah, 2494 res->hob_feature, res->hob_nsect, 2495 res->hob_lbal, res->hob_lbam, res->hob_lbah, 2496 res->device, qc->err_mask, ata_err_string(qc->err_mask), 2497 qc->err_mask & AC_ERR_NCQ ? " <F>" : ""); 2498 2499 #ifdef CONFIG_ATA_VERBOSE_ERROR 2500 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ | 2501 ATA_SENSE | ATA_ERR)) { 2502 if (res->status & ATA_BUSY) 2503 ata_dev_err(qc->dev, "status: { Busy }\n"); 2504 else 2505 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n", 2506 res->status & ATA_DRDY ? "DRDY " : "", 2507 res->status & ATA_DF ? "DF " : "", 2508 res->status & ATA_DRQ ? "DRQ " : "", 2509 res->status & ATA_SENSE ? "SENSE " : "", 2510 res->status & ATA_ERR ? "ERR " : ""); 2511 } 2512 2513 if (cmd->command != ATA_CMD_PACKET && 2514 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF | 2515 ATA_ABORTED))) 2516 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n", 2517 res->error & ATA_ICRC ? "ICRC " : "", 2518 res->error & ATA_UNC ? "UNC " : "", 2519 res->error & ATA_AMNF ? "AMNF " : "", 2520 res->error & ATA_IDNF ? "IDNF " : "", 2521 res->error & ATA_ABORTED ? "ABRT " : ""); 2522 #endif 2523 } 2524 } 2525 2526 /** 2527 * ata_eh_report - report error handling to user 2528 * @ap: ATA port to report EH about 2529 * 2530 * Report EH to user. 2531 * 2532 * LOCKING: 2533 * None. 2534 */ 2535 void ata_eh_report(struct ata_port *ap) 2536 { 2537 struct ata_link *link; 2538 2539 ata_for_each_link(link, ap, HOST_FIRST) 2540 ata_eh_link_report(link); 2541 } 2542 2543 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset, 2544 unsigned int *classes, unsigned long deadline, 2545 bool clear_classes) 2546 { 2547 struct ata_device *dev; 2548 2549 if (clear_classes) 2550 ata_for_each_dev(dev, link, ALL) 2551 classes[dev->devno] = ATA_DEV_UNKNOWN; 2552 2553 return reset(link, classes, deadline); 2554 } 2555 2556 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc) 2557 { 2558 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link)) 2559 return 0; 2560 if (rc == -EAGAIN) 2561 return 1; 2562 if (sata_pmp_supported(link->ap) && ata_is_host_link(link)) 2563 return 1; 2564 return 0; 2565 } 2566 2567 int ata_eh_reset(struct ata_link *link, int classify, 2568 ata_prereset_fn_t prereset, ata_reset_fn_t softreset, 2569 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset) 2570 { 2571 struct ata_port *ap = link->ap; 2572 struct ata_link *slave = ap->slave_link; 2573 struct ata_eh_context *ehc = &link->eh_context; 2574 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL; 2575 unsigned int *classes = ehc->classes; 2576 unsigned int lflags = link->flags; 2577 int verbose = !(ehc->i.flags & ATA_EHI_QUIET); 2578 int max_tries = 0, try = 0; 2579 struct ata_link *failed_link; 2580 struct ata_device *dev; 2581 unsigned long deadline, now; 2582 ata_reset_fn_t reset; 2583 unsigned long flags; 2584 u32 sstatus; 2585 int nr_unknown, rc; 2586 2587 /* 2588 * Prepare to reset 2589 */ 2590 while (ata_eh_reset_timeouts[max_tries] != UINT_MAX) 2591 max_tries++; 2592 if (link->flags & ATA_LFLAG_RST_ONCE) 2593 max_tries = 1; 2594 if (link->flags & ATA_LFLAG_NO_HRST) 2595 hardreset = NULL; 2596 if (link->flags & ATA_LFLAG_NO_SRST) 2597 softreset = NULL; 2598 2599 /* make sure each reset attempt is at least COOL_DOWN apart */ 2600 if (ehc->i.flags & ATA_EHI_DID_RESET) { 2601 now = jiffies; 2602 WARN_ON(time_after(ehc->last_reset, now)); 2603 deadline = ata_deadline(ehc->last_reset, 2604 ATA_EH_RESET_COOL_DOWN); 2605 if (time_before(now, deadline)) 2606 schedule_timeout_uninterruptible(deadline - now); 2607 } 2608 2609 spin_lock_irqsave(ap->lock, flags); 2610 ap->pflags |= ATA_PFLAG_RESETTING; 2611 spin_unlock_irqrestore(ap->lock, flags); 2612 2613 ata_eh_about_to_do(link, NULL, ATA_EH_RESET); 2614 2615 ata_for_each_dev(dev, link, ALL) { 2616 /* If we issue an SRST then an ATA drive (not ATAPI) 2617 * may change configuration and be in PIO0 timing. If 2618 * we do a hard reset (or are coming from power on) 2619 * this is true for ATA or ATAPI. Until we've set a 2620 * suitable controller mode we should not touch the 2621 * bus as we may be talking too fast. 2622 */ 2623 dev->pio_mode = XFER_PIO_0; 2624 dev->dma_mode = 0xff; 2625 2626 /* If the controller has a pio mode setup function 2627 * then use it to set the chipset to rights. Don't 2628 * touch the DMA setup as that will be dealt with when 2629 * configuring devices. 2630 */ 2631 if (ap->ops->set_piomode) 2632 ap->ops->set_piomode(ap, dev); 2633 } 2634 2635 /* prefer hardreset */ 2636 reset = NULL; 2637 ehc->i.action &= ~ATA_EH_RESET; 2638 if (hardreset) { 2639 reset = hardreset; 2640 ehc->i.action |= ATA_EH_HARDRESET; 2641 } else if (softreset) { 2642 reset = softreset; 2643 ehc->i.action |= ATA_EH_SOFTRESET; 2644 } 2645 2646 if (prereset) { 2647 unsigned long deadline = ata_deadline(jiffies, 2648 ATA_EH_PRERESET_TIMEOUT); 2649 2650 if (slave) { 2651 sehc->i.action &= ~ATA_EH_RESET; 2652 sehc->i.action |= ehc->i.action; 2653 } 2654 2655 rc = prereset(link, deadline); 2656 2657 /* If present, do prereset on slave link too. Reset 2658 * is skipped iff both master and slave links report 2659 * -ENOENT or clear ATA_EH_RESET. 2660 */ 2661 if (slave && (rc == 0 || rc == -ENOENT)) { 2662 int tmp; 2663 2664 tmp = prereset(slave, deadline); 2665 if (tmp != -ENOENT) 2666 rc = tmp; 2667 2668 ehc->i.action |= sehc->i.action; 2669 } 2670 2671 if (rc) { 2672 if (rc == -ENOENT) { 2673 ata_link_dbg(link, "port disabled--ignoring\n"); 2674 ehc->i.action &= ~ATA_EH_RESET; 2675 2676 ata_for_each_dev(dev, link, ALL) 2677 classes[dev->devno] = ATA_DEV_NONE; 2678 2679 rc = 0; 2680 } else 2681 ata_link_err(link, 2682 "prereset failed (errno=%d)\n", 2683 rc); 2684 goto out; 2685 } 2686 2687 /* prereset() might have cleared ATA_EH_RESET. If so, 2688 * bang classes, thaw and return. 2689 */ 2690 if (reset && !(ehc->i.action & ATA_EH_RESET)) { 2691 ata_for_each_dev(dev, link, ALL) 2692 classes[dev->devno] = ATA_DEV_NONE; 2693 if (ata_port_is_frozen(ap) && ata_is_host_link(link)) 2694 ata_eh_thaw_port(ap); 2695 rc = 0; 2696 goto out; 2697 } 2698 } 2699 2700 retry: 2701 /* 2702 * Perform reset 2703 */ 2704 if (ata_is_host_link(link)) 2705 ata_eh_freeze_port(ap); 2706 2707 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]); 2708 2709 if (reset) { 2710 if (verbose) 2711 ata_link_info(link, "%s resetting link\n", 2712 reset == softreset ? "soft" : "hard"); 2713 2714 /* mark that this EH session started with reset */ 2715 ehc->last_reset = jiffies; 2716 if (reset == hardreset) { 2717 ehc->i.flags |= ATA_EHI_DID_HARDRESET; 2718 trace_ata_link_hardreset_begin(link, classes, deadline); 2719 } else { 2720 ehc->i.flags |= ATA_EHI_DID_SOFTRESET; 2721 trace_ata_link_softreset_begin(link, classes, deadline); 2722 } 2723 2724 rc = ata_do_reset(link, reset, classes, deadline, true); 2725 if (reset == hardreset) 2726 trace_ata_link_hardreset_end(link, classes, rc); 2727 else 2728 trace_ata_link_softreset_end(link, classes, rc); 2729 if (rc && rc != -EAGAIN) { 2730 failed_link = link; 2731 goto fail; 2732 } 2733 2734 /* hardreset slave link if existent */ 2735 if (slave && reset == hardreset) { 2736 int tmp; 2737 2738 if (verbose) 2739 ata_link_info(slave, "hard resetting link\n"); 2740 2741 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET); 2742 trace_ata_slave_hardreset_begin(slave, classes, 2743 deadline); 2744 tmp = ata_do_reset(slave, reset, classes, deadline, 2745 false); 2746 trace_ata_slave_hardreset_end(slave, classes, tmp); 2747 switch (tmp) { 2748 case -EAGAIN: 2749 rc = -EAGAIN; 2750 break; 2751 case 0: 2752 break; 2753 default: 2754 failed_link = slave; 2755 rc = tmp; 2756 goto fail; 2757 } 2758 } 2759 2760 /* perform follow-up SRST if necessary */ 2761 if (reset == hardreset && 2762 ata_eh_followup_srst_needed(link, rc)) { 2763 reset = softreset; 2764 2765 if (!reset) { 2766 ata_link_err(link, 2767 "follow-up softreset required but no softreset available\n"); 2768 failed_link = link; 2769 rc = -EINVAL; 2770 goto fail; 2771 } 2772 2773 ata_eh_about_to_do(link, NULL, ATA_EH_RESET); 2774 trace_ata_link_softreset_begin(link, classes, deadline); 2775 rc = ata_do_reset(link, reset, classes, deadline, true); 2776 trace_ata_link_softreset_end(link, classes, rc); 2777 if (rc) { 2778 failed_link = link; 2779 goto fail; 2780 } 2781 } 2782 } else { 2783 if (verbose) 2784 ata_link_info(link, 2785 "no reset method available, skipping reset\n"); 2786 if (!(lflags & ATA_LFLAG_ASSUME_CLASS)) 2787 lflags |= ATA_LFLAG_ASSUME_ATA; 2788 } 2789 2790 /* 2791 * Post-reset processing 2792 */ 2793 ata_for_each_dev(dev, link, ALL) { 2794 /* After the reset, the device state is PIO 0 and the 2795 * controller state is undefined. Reset also wakes up 2796 * drives from sleeping mode. 2797 */ 2798 dev->pio_mode = XFER_PIO_0; 2799 dev->flags &= ~ATA_DFLAG_SLEEPING; 2800 2801 if (ata_phys_link_offline(ata_dev_phys_link(dev))) 2802 continue; 2803 2804 /* apply class override */ 2805 if (lflags & ATA_LFLAG_ASSUME_ATA) 2806 classes[dev->devno] = ATA_DEV_ATA; 2807 else if (lflags & ATA_LFLAG_ASSUME_SEMB) 2808 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; 2809 } 2810 2811 /* record current link speed */ 2812 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0) 2813 link->sata_spd = (sstatus >> 4) & 0xf; 2814 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0) 2815 slave->sata_spd = (sstatus >> 4) & 0xf; 2816 2817 /* thaw the port */ 2818 if (ata_is_host_link(link)) 2819 ata_eh_thaw_port(ap); 2820 2821 /* postreset() should clear hardware SError. Although SError 2822 * is cleared during link resume, clearing SError here is 2823 * necessary as some PHYs raise hotplug events after SRST. 2824 * This introduces race condition where hotplug occurs between 2825 * reset and here. This race is mediated by cross checking 2826 * link onlineness and classification result later. 2827 */ 2828 if (postreset) { 2829 postreset(link, classes); 2830 trace_ata_link_postreset(link, classes, rc); 2831 if (slave) { 2832 postreset(slave, classes); 2833 trace_ata_slave_postreset(slave, classes, rc); 2834 } 2835 } 2836 2837 /* clear cached SError */ 2838 spin_lock_irqsave(link->ap->lock, flags); 2839 link->eh_info.serror = 0; 2840 if (slave) 2841 slave->eh_info.serror = 0; 2842 spin_unlock_irqrestore(link->ap->lock, flags); 2843 2844 /* 2845 * Make sure onlineness and classification result correspond. 2846 * Hotplug could have happened during reset and some 2847 * controllers fail to wait while a drive is spinning up after 2848 * being hotplugged causing misdetection. By cross checking 2849 * link on/offlineness and classification result, those 2850 * conditions can be reliably detected and retried. 2851 */ 2852 nr_unknown = 0; 2853 ata_for_each_dev(dev, link, ALL) { 2854 if (ata_phys_link_online(ata_dev_phys_link(dev))) { 2855 if (classes[dev->devno] == ATA_DEV_UNKNOWN) { 2856 ata_dev_dbg(dev, "link online but device misclassified\n"); 2857 classes[dev->devno] = ATA_DEV_NONE; 2858 nr_unknown++; 2859 } 2860 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) { 2861 if (ata_class_enabled(classes[dev->devno])) 2862 ata_dev_dbg(dev, 2863 "link offline, clearing class %d to NONE\n", 2864 classes[dev->devno]); 2865 classes[dev->devno] = ATA_DEV_NONE; 2866 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) { 2867 ata_dev_dbg(dev, 2868 "link status unknown, clearing UNKNOWN to NONE\n"); 2869 classes[dev->devno] = ATA_DEV_NONE; 2870 } 2871 } 2872 2873 if (classify && nr_unknown) { 2874 if (try < max_tries) { 2875 ata_link_warn(link, 2876 "link online but %d devices misclassified, retrying\n", 2877 nr_unknown); 2878 failed_link = link; 2879 rc = -EAGAIN; 2880 goto fail; 2881 } 2882 ata_link_warn(link, 2883 "link online but %d devices misclassified, " 2884 "device detection might fail\n", nr_unknown); 2885 } 2886 2887 /* reset successful, schedule revalidation */ 2888 ata_eh_done(link, NULL, ATA_EH_RESET); 2889 if (slave) 2890 ata_eh_done(slave, NULL, ATA_EH_RESET); 2891 ehc->last_reset = jiffies; /* update to completion time */ 2892 ehc->i.action |= ATA_EH_REVALIDATE; 2893 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */ 2894 2895 rc = 0; 2896 out: 2897 /* clear hotplug flag */ 2898 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED; 2899 if (slave) 2900 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED; 2901 2902 spin_lock_irqsave(ap->lock, flags); 2903 ap->pflags &= ~ATA_PFLAG_RESETTING; 2904 spin_unlock_irqrestore(ap->lock, flags); 2905 2906 return rc; 2907 2908 fail: 2909 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */ 2910 if (!ata_is_host_link(link) && 2911 sata_scr_read(link, SCR_STATUS, &sstatus)) 2912 rc = -ERESTART; 2913 2914 if (try >= max_tries) { 2915 /* 2916 * Thaw host port even if reset failed, so that the port 2917 * can be retried on the next phy event. This risks 2918 * repeated EH runs but seems to be a better tradeoff than 2919 * shutting down a port after a botched hotplug attempt. 2920 */ 2921 if (ata_is_host_link(link)) 2922 ata_eh_thaw_port(ap); 2923 goto out; 2924 } 2925 2926 now = jiffies; 2927 if (time_before(now, deadline)) { 2928 unsigned long delta = deadline - now; 2929 2930 ata_link_warn(failed_link, 2931 "reset failed (errno=%d), retrying in %u secs\n", 2932 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000)); 2933 2934 ata_eh_release(ap); 2935 while (delta) 2936 delta = schedule_timeout_uninterruptible(delta); 2937 ata_eh_acquire(ap); 2938 } 2939 2940 /* 2941 * While disks spinup behind PMP, some controllers fail sending SRST. 2942 * They need to be reset - as well as the PMP - before retrying. 2943 */ 2944 if (rc == -ERESTART) { 2945 if (ata_is_host_link(link)) 2946 ata_eh_thaw_port(ap); 2947 goto out; 2948 } 2949 2950 if (try == max_tries - 1) { 2951 sata_down_spd_limit(link, 0); 2952 if (slave) 2953 sata_down_spd_limit(slave, 0); 2954 } else if (rc == -EPIPE) 2955 sata_down_spd_limit(failed_link, 0); 2956 2957 if (hardreset) 2958 reset = hardreset; 2959 goto retry; 2960 } 2961 2962 static inline void ata_eh_pull_park_action(struct ata_port *ap) 2963 { 2964 struct ata_link *link; 2965 struct ata_device *dev; 2966 unsigned long flags; 2967 2968 /* 2969 * This function can be thought of as an extended version of 2970 * ata_eh_about_to_do() specially crafted to accommodate the 2971 * requirements of ATA_EH_PARK handling. Since the EH thread 2972 * does not leave the do {} while () loop in ata_eh_recover as 2973 * long as the timeout for a park request to *one* device on 2974 * the port has not expired, and since we still want to pick 2975 * up park requests to other devices on the same port or 2976 * timeout updates for the same device, we have to pull 2977 * ATA_EH_PARK actions from eh_info into eh_context.i 2978 * ourselves at the beginning of each pass over the loop. 2979 * 2980 * Additionally, all write accesses to &ap->park_req_pending 2981 * through reinit_completion() (see below) or complete_all() 2982 * (see ata_scsi_park_store()) are protected by the host lock. 2983 * As a result we have that park_req_pending.done is zero on 2984 * exit from this function, i.e. when ATA_EH_PARK actions for 2985 * *all* devices on port ap have been pulled into the 2986 * respective eh_context structs. If, and only if, 2987 * park_req_pending.done is non-zero by the time we reach 2988 * wait_for_completion_timeout(), another ATA_EH_PARK action 2989 * has been scheduled for at least one of the devices on port 2990 * ap and we have to cycle over the do {} while () loop in 2991 * ata_eh_recover() again. 2992 */ 2993 2994 spin_lock_irqsave(ap->lock, flags); 2995 reinit_completion(&ap->park_req_pending); 2996 ata_for_each_link(link, ap, EDGE) { 2997 ata_for_each_dev(dev, link, ALL) { 2998 struct ata_eh_info *ehi = &link->eh_info; 2999 3000 link->eh_context.i.dev_action[dev->devno] |= 3001 ehi->dev_action[dev->devno] & ATA_EH_PARK; 3002 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK); 3003 } 3004 } 3005 spin_unlock_irqrestore(ap->lock, flags); 3006 } 3007 3008 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park) 3009 { 3010 struct ata_eh_context *ehc = &dev->link->eh_context; 3011 struct ata_taskfile tf; 3012 unsigned int err_mask; 3013 3014 ata_tf_init(dev, &tf); 3015 if (park) { 3016 ehc->unloaded_mask |= 1 << dev->devno; 3017 tf.command = ATA_CMD_IDLEIMMEDIATE; 3018 tf.feature = 0x44; 3019 tf.lbal = 0x4c; 3020 tf.lbam = 0x4e; 3021 tf.lbah = 0x55; 3022 } else { 3023 ehc->unloaded_mask &= ~(1 << dev->devno); 3024 tf.command = ATA_CMD_CHK_POWER; 3025 } 3026 3027 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR; 3028 tf.protocol = ATA_PROT_NODATA; 3029 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0); 3030 if (park && (err_mask || tf.lbal != 0xc4)) { 3031 ata_dev_err(dev, "head unload failed!\n"); 3032 ehc->unloaded_mask &= ~(1 << dev->devno); 3033 } 3034 } 3035 3036 static int ata_eh_revalidate_and_attach(struct ata_link *link, 3037 struct ata_device **r_failed_dev) 3038 { 3039 struct ata_port *ap = link->ap; 3040 struct ata_eh_context *ehc = &link->eh_context; 3041 struct ata_device *dev; 3042 unsigned int new_mask = 0; 3043 unsigned long flags; 3044 int rc = 0; 3045 3046 /* For PATA drive side cable detection to work, IDENTIFY must 3047 * be done backwards such that PDIAG- is released by the slave 3048 * device before the master device is identified. 3049 */ 3050 ata_for_each_dev(dev, link, ALL_REVERSE) { 3051 unsigned int action = ata_eh_dev_action(dev); 3052 unsigned int readid_flags = 0; 3053 3054 if (ehc->i.flags & ATA_EHI_DID_RESET) 3055 readid_flags |= ATA_READID_POSTRESET; 3056 3057 /* 3058 * When resuming, before executing any command, make sure to 3059 * transition the device to the active power mode. 3060 */ 3061 if ((action & ATA_EH_SET_ACTIVE) && ata_dev_enabled(dev)) { 3062 ata_dev_power_set_active(dev); 3063 ata_eh_done(link, dev, ATA_EH_SET_ACTIVE); 3064 } 3065 3066 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) { 3067 WARN_ON(dev->class == ATA_DEV_PMP); 3068 3069 /* 3070 * The link may be in a deep sleep, wake it up. 3071 * 3072 * If the link is in deep sleep, ata_phys_link_offline() 3073 * will return true, causing the revalidation to fail, 3074 * which leads to a (potentially) needless hard reset. 3075 * 3076 * ata_eh_recover() will later restore the link policy 3077 * to ap->target_lpm_policy after revalidation is done. 3078 */ 3079 if (link->lpm_policy > ATA_LPM_MAX_POWER) { 3080 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER, 3081 r_failed_dev); 3082 if (rc) 3083 goto err; 3084 } 3085 3086 if (ata_phys_link_offline(ata_dev_phys_link(dev))) { 3087 rc = -EIO; 3088 goto err; 3089 } 3090 3091 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE); 3092 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno], 3093 readid_flags); 3094 if (rc) 3095 goto err; 3096 3097 ata_eh_done(link, dev, ATA_EH_REVALIDATE); 3098 3099 /* Configuration may have changed, reconfigure 3100 * transfer mode. 3101 */ 3102 ehc->i.flags |= ATA_EHI_SETMODE; 3103 3104 /* schedule the scsi_rescan_device() here */ 3105 schedule_delayed_work(&ap->scsi_rescan_task, 0); 3106 } else if (dev->class == ATA_DEV_UNKNOWN && 3107 ehc->tries[dev->devno] && 3108 ata_class_enabled(ehc->classes[dev->devno])) { 3109 /* Temporarily set dev->class, it will be 3110 * permanently set once all configurations are 3111 * complete. This is necessary because new 3112 * device configuration is done in two 3113 * separate loops. 3114 */ 3115 dev->class = ehc->classes[dev->devno]; 3116 3117 if (dev->class == ATA_DEV_PMP) 3118 rc = sata_pmp_attach(dev); 3119 else 3120 rc = ata_dev_read_id(dev, &dev->class, 3121 readid_flags, dev->id); 3122 3123 /* read_id might have changed class, store and reset */ 3124 ehc->classes[dev->devno] = dev->class; 3125 dev->class = ATA_DEV_UNKNOWN; 3126 3127 switch (rc) { 3128 case 0: 3129 /* clear error info accumulated during probe */ 3130 ata_ering_clear(&dev->ering); 3131 new_mask |= 1 << dev->devno; 3132 break; 3133 case -ENOENT: 3134 /* IDENTIFY was issued to non-existent 3135 * device. No need to reset. Just 3136 * thaw and ignore the device. 3137 */ 3138 ata_eh_thaw_port(ap); 3139 break; 3140 default: 3141 goto err; 3142 } 3143 } 3144 } 3145 3146 /* PDIAG- should have been released, ask cable type if post-reset */ 3147 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) { 3148 if (ap->ops->cable_detect) 3149 ap->cbl = ap->ops->cable_detect(ap); 3150 ata_force_cbl(ap); 3151 } 3152 3153 /* Configure new devices forward such that user doesn't see 3154 * device detection messages backwards. 3155 */ 3156 ata_for_each_dev(dev, link, ALL) { 3157 if (!(new_mask & (1 << dev->devno))) 3158 continue; 3159 3160 dev->class = ehc->classes[dev->devno]; 3161 3162 if (dev->class == ATA_DEV_PMP) 3163 continue; 3164 3165 ehc->i.flags |= ATA_EHI_PRINTINFO; 3166 rc = ata_dev_configure(dev); 3167 ehc->i.flags &= ~ATA_EHI_PRINTINFO; 3168 if (rc) { 3169 dev->class = ATA_DEV_UNKNOWN; 3170 goto err; 3171 } 3172 3173 spin_lock_irqsave(ap->lock, flags); 3174 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; 3175 spin_unlock_irqrestore(ap->lock, flags); 3176 3177 /* new device discovered, configure xfermode */ 3178 ehc->i.flags |= ATA_EHI_SETMODE; 3179 } 3180 3181 return 0; 3182 3183 err: 3184 dev->flags &= ~ATA_DFLAG_RESUMING; 3185 *r_failed_dev = dev; 3186 return rc; 3187 } 3188 3189 /** 3190 * ata_set_mode - Program timings and issue SET FEATURES - XFER 3191 * @link: link on which timings will be programmed 3192 * @r_failed_dev: out parameter for failed device 3193 * 3194 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If 3195 * ata_set_mode() fails, pointer to the failing device is 3196 * returned in @r_failed_dev. 3197 * 3198 * LOCKING: 3199 * PCI/etc. bus probe sem. 3200 * 3201 * RETURNS: 3202 * 0 on success, negative errno otherwise 3203 */ 3204 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev) 3205 { 3206 struct ata_port *ap = link->ap; 3207 struct ata_device *dev; 3208 int rc; 3209 3210 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */ 3211 ata_for_each_dev(dev, link, ENABLED) { 3212 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) { 3213 struct ata_ering_entry *ent; 3214 3215 ent = ata_ering_top(&dev->ering); 3216 if (ent) 3217 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER; 3218 } 3219 } 3220 3221 /* has private set_mode? */ 3222 if (ap->ops->set_mode) 3223 rc = ap->ops->set_mode(link, r_failed_dev); 3224 else 3225 rc = ata_do_set_mode(link, r_failed_dev); 3226 3227 /* if transfer mode has changed, set DUBIOUS_XFER on device */ 3228 ata_for_each_dev(dev, link, ENABLED) { 3229 struct ata_eh_context *ehc = &link->eh_context; 3230 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno]; 3231 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno)); 3232 3233 if (dev->xfer_mode != saved_xfer_mode || 3234 ata_ncq_enabled(dev) != saved_ncq) 3235 dev->flags |= ATA_DFLAG_DUBIOUS_XFER; 3236 } 3237 3238 return rc; 3239 } 3240 3241 /** 3242 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset 3243 * @dev: ATAPI device to clear UA for 3244 * 3245 * Resets and other operations can make an ATAPI device raise 3246 * UNIT ATTENTION which causes the next operation to fail. This 3247 * function clears UA. 3248 * 3249 * LOCKING: 3250 * EH context (may sleep). 3251 * 3252 * RETURNS: 3253 * 0 on success, -errno on failure. 3254 */ 3255 static int atapi_eh_clear_ua(struct ata_device *dev) 3256 { 3257 int i; 3258 3259 for (i = 0; i < ATA_EH_UA_TRIES; i++) { 3260 u8 *sense_buffer = dev->link->ap->sector_buf; 3261 u8 sense_key = 0; 3262 unsigned int err_mask; 3263 3264 err_mask = atapi_eh_tur(dev, &sense_key); 3265 if (err_mask != 0 && err_mask != AC_ERR_DEV) { 3266 ata_dev_warn(dev, 3267 "TEST_UNIT_READY failed (err_mask=0x%x)\n", 3268 err_mask); 3269 return -EIO; 3270 } 3271 3272 if (!err_mask || sense_key != UNIT_ATTENTION) 3273 return 0; 3274 3275 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key); 3276 if (err_mask) { 3277 ata_dev_warn(dev, "failed to clear " 3278 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask); 3279 return -EIO; 3280 } 3281 } 3282 3283 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n", 3284 ATA_EH_UA_TRIES); 3285 3286 return 0; 3287 } 3288 3289 /** 3290 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary 3291 * @dev: ATA device which may need FLUSH retry 3292 * 3293 * If @dev failed FLUSH, it needs to be reported upper layer 3294 * immediately as it means that @dev failed to remap and already 3295 * lost at least a sector and further FLUSH retrials won't make 3296 * any difference to the lost sector. However, if FLUSH failed 3297 * for other reasons, for example transmission error, FLUSH needs 3298 * to be retried. 3299 * 3300 * This function determines whether FLUSH failure retry is 3301 * necessary and performs it if so. 3302 * 3303 * RETURNS: 3304 * 0 if EH can continue, -errno if EH needs to be repeated. 3305 */ 3306 static int ata_eh_maybe_retry_flush(struct ata_device *dev) 3307 { 3308 struct ata_link *link = dev->link; 3309 struct ata_port *ap = link->ap; 3310 struct ata_queued_cmd *qc; 3311 struct ata_taskfile tf; 3312 unsigned int err_mask; 3313 int rc = 0; 3314 3315 /* did flush fail for this device? */ 3316 if (!ata_tag_valid(link->active_tag)) 3317 return 0; 3318 3319 qc = __ata_qc_from_tag(ap, link->active_tag); 3320 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT && 3321 qc->tf.command != ATA_CMD_FLUSH)) 3322 return 0; 3323 3324 /* if the device failed it, it should be reported to upper layers */ 3325 if (qc->err_mask & AC_ERR_DEV) 3326 return 0; 3327 3328 /* flush failed for some other reason, give it another shot */ 3329 ata_tf_init(dev, &tf); 3330 3331 tf.command = qc->tf.command; 3332 tf.flags |= ATA_TFLAG_DEVICE; 3333 tf.protocol = ATA_PROT_NODATA; 3334 3335 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n", 3336 tf.command, qc->err_mask); 3337 3338 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0); 3339 if (!err_mask) { 3340 /* 3341 * FLUSH is complete but there's no way to 3342 * successfully complete a failed command from EH. 3343 * Making sure retry is allowed at least once and 3344 * retrying it should do the trick - whatever was in 3345 * the cache is already on the platter and this won't 3346 * cause infinite loop. 3347 */ 3348 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1); 3349 } else { 3350 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n", 3351 err_mask); 3352 rc = -EIO; 3353 3354 /* if device failed it, report it to upper layers */ 3355 if (err_mask & AC_ERR_DEV) { 3356 qc->err_mask |= AC_ERR_DEV; 3357 qc->result_tf = tf; 3358 if (!ata_port_is_frozen(ap)) 3359 rc = 0; 3360 } 3361 } 3362 return rc; 3363 } 3364 3365 /** 3366 * ata_eh_set_lpm - configure SATA interface power management 3367 * @link: link to configure power management 3368 * @policy: the link power management policy 3369 * @r_failed_dev: out parameter for failed device 3370 * 3371 * Enable SATA Interface power management. This will enable 3372 * Device Interface Power Management (DIPM) for min_power and 3373 * medium_power_with_dipm policies, and then call driver specific 3374 * callbacks for enabling Host Initiated Power management. 3375 * 3376 * LOCKING: 3377 * EH context. 3378 * 3379 * RETURNS: 3380 * 0 on success, -errno on failure. 3381 */ 3382 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy, 3383 struct ata_device **r_failed_dev) 3384 { 3385 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL; 3386 struct ata_eh_context *ehc = &link->eh_context; 3387 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL; 3388 enum ata_lpm_policy old_policy = link->lpm_policy; 3389 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM; 3390 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM; 3391 unsigned int err_mask; 3392 int rc; 3393 3394 /* if the link or host doesn't do LPM, noop */ 3395 if (!IS_ENABLED(CONFIG_SATA_HOST) || 3396 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm)) 3397 return 0; 3398 3399 /* 3400 * DIPM is enabled only for MIN_POWER as some devices 3401 * misbehave when the host NACKs transition to SLUMBER. Order 3402 * device and link configurations such that the host always 3403 * allows DIPM requests. 3404 */ 3405 ata_for_each_dev(dev, link, ENABLED) { 3406 bool hipm = ata_id_has_hipm(dev->id); 3407 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm; 3408 3409 /* find the first enabled and LPM enabled devices */ 3410 if (!link_dev) 3411 link_dev = dev; 3412 3413 if (!lpm_dev && (hipm || dipm)) 3414 lpm_dev = dev; 3415 3416 hints &= ~ATA_LPM_EMPTY; 3417 if (!hipm) 3418 hints &= ~ATA_LPM_HIPM; 3419 3420 /* disable DIPM before changing link config */ 3421 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) { 3422 err_mask = ata_dev_set_feature(dev, 3423 SETFEATURES_SATA_DISABLE, SATA_DIPM); 3424 if (err_mask && err_mask != AC_ERR_DEV) { 3425 ata_dev_warn(dev, 3426 "failed to disable DIPM, Emask 0x%x\n", 3427 err_mask); 3428 rc = -EIO; 3429 goto fail; 3430 } 3431 } 3432 } 3433 3434 if (ap) { 3435 rc = ap->ops->set_lpm(link, policy, hints); 3436 if (!rc && ap->slave_link) 3437 rc = ap->ops->set_lpm(ap->slave_link, policy, hints); 3438 } else 3439 rc = sata_pmp_set_lpm(link, policy, hints); 3440 3441 /* 3442 * Attribute link config failure to the first (LPM) enabled 3443 * device on the link. 3444 */ 3445 if (rc) { 3446 if (rc == -EOPNOTSUPP) { 3447 link->flags |= ATA_LFLAG_NO_LPM; 3448 return 0; 3449 } 3450 dev = lpm_dev ? lpm_dev : link_dev; 3451 goto fail; 3452 } 3453 3454 /* 3455 * Low level driver acked the transition. Issue DIPM command 3456 * with the new policy set. 3457 */ 3458 link->lpm_policy = policy; 3459 if (ap && ap->slave_link) 3460 ap->slave_link->lpm_policy = policy; 3461 3462 /* host config updated, enable DIPM if transitioning to MIN_POWER */ 3463 ata_for_each_dev(dev, link, ENABLED) { 3464 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm && 3465 ata_id_has_dipm(dev->id)) { 3466 err_mask = ata_dev_set_feature(dev, 3467 SETFEATURES_SATA_ENABLE, SATA_DIPM); 3468 if (err_mask && err_mask != AC_ERR_DEV) { 3469 ata_dev_warn(dev, 3470 "failed to enable DIPM, Emask 0x%x\n", 3471 err_mask); 3472 rc = -EIO; 3473 goto fail; 3474 } 3475 } 3476 } 3477 3478 link->last_lpm_change = jiffies; 3479 link->flags |= ATA_LFLAG_CHANGED; 3480 3481 return 0; 3482 3483 fail: 3484 /* restore the old policy */ 3485 link->lpm_policy = old_policy; 3486 if (ap && ap->slave_link) 3487 ap->slave_link->lpm_policy = old_policy; 3488 3489 /* if no device or only one more chance is left, disable LPM */ 3490 if (!dev || ehc->tries[dev->devno] <= 2) { 3491 ata_link_warn(link, "disabling LPM on the link\n"); 3492 link->flags |= ATA_LFLAG_NO_LPM; 3493 } 3494 if (r_failed_dev) 3495 *r_failed_dev = dev; 3496 return rc; 3497 } 3498 3499 int ata_link_nr_enabled(struct ata_link *link) 3500 { 3501 struct ata_device *dev; 3502 int cnt = 0; 3503 3504 ata_for_each_dev(dev, link, ENABLED) 3505 cnt++; 3506 return cnt; 3507 } 3508 3509 static int ata_link_nr_vacant(struct ata_link *link) 3510 { 3511 struct ata_device *dev; 3512 int cnt = 0; 3513 3514 ata_for_each_dev(dev, link, ALL) 3515 if (dev->class == ATA_DEV_UNKNOWN) 3516 cnt++; 3517 return cnt; 3518 } 3519 3520 static int ata_eh_skip_recovery(struct ata_link *link) 3521 { 3522 struct ata_port *ap = link->ap; 3523 struct ata_eh_context *ehc = &link->eh_context; 3524 struct ata_device *dev; 3525 3526 /* skip disabled links */ 3527 if (link->flags & ATA_LFLAG_DISABLED) 3528 return 1; 3529 3530 /* skip if explicitly requested */ 3531 if (ehc->i.flags & ATA_EHI_NO_RECOVERY) 3532 return 1; 3533 3534 /* thaw frozen port and recover failed devices */ 3535 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link)) 3536 return 0; 3537 3538 /* reset at least once if reset is requested */ 3539 if ((ehc->i.action & ATA_EH_RESET) && 3540 !(ehc->i.flags & ATA_EHI_DID_RESET)) 3541 return 0; 3542 3543 /* skip if class codes for all vacant slots are ATA_DEV_NONE */ 3544 ata_for_each_dev(dev, link, ALL) { 3545 if (dev->class == ATA_DEV_UNKNOWN && 3546 ehc->classes[dev->devno] != ATA_DEV_NONE) 3547 return 0; 3548 } 3549 3550 return 1; 3551 } 3552 3553 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg) 3554 { 3555 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL); 3556 u64 now = get_jiffies_64(); 3557 int *trials = void_arg; 3558 3559 if ((ent->eflags & ATA_EFLAG_OLD_ER) || 3560 (ent->timestamp < now - min(now, interval))) 3561 return -1; 3562 3563 (*trials)++; 3564 return 0; 3565 } 3566 3567 static int ata_eh_schedule_probe(struct ata_device *dev) 3568 { 3569 struct ata_eh_context *ehc = &dev->link->eh_context; 3570 struct ata_link *link = ata_dev_phys_link(dev); 3571 int trials = 0; 3572 3573 if (!(ehc->i.probe_mask & (1 << dev->devno)) || 3574 (ehc->did_probe_mask & (1 << dev->devno))) 3575 return 0; 3576 3577 ata_eh_detach_dev(dev); 3578 ata_dev_init(dev); 3579 ehc->did_probe_mask |= (1 << dev->devno); 3580 ehc->i.action |= ATA_EH_RESET; 3581 ehc->saved_xfer_mode[dev->devno] = 0; 3582 ehc->saved_ncq_enabled &= ~(1 << dev->devno); 3583 3584 /* the link maybe in a deep sleep, wake it up */ 3585 if (link->lpm_policy > ATA_LPM_MAX_POWER) { 3586 if (ata_is_host_link(link)) 3587 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER, 3588 ATA_LPM_EMPTY); 3589 else 3590 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER, 3591 ATA_LPM_EMPTY); 3592 } 3593 3594 /* Record and count probe trials on the ering. The specific 3595 * error mask used is irrelevant. Because a successful device 3596 * detection clears the ering, this count accumulates only if 3597 * there are consecutive failed probes. 3598 * 3599 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS 3600 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is 3601 * forced to 1.5Gbps. 3602 * 3603 * This is to work around cases where failed link speed 3604 * negotiation results in device misdetection leading to 3605 * infinite DEVXCHG or PHRDY CHG events. 3606 */ 3607 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER); 3608 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials); 3609 3610 if (trials > ATA_EH_PROBE_TRIALS) 3611 sata_down_spd_limit(link, 1); 3612 3613 return 1; 3614 } 3615 3616 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err) 3617 { 3618 struct ata_eh_context *ehc = &dev->link->eh_context; 3619 3620 /* -EAGAIN from EH routine indicates retry without prejudice. 3621 * The requester is responsible for ensuring forward progress. 3622 */ 3623 if (err != -EAGAIN) 3624 ehc->tries[dev->devno]--; 3625 3626 switch (err) { 3627 case -ENODEV: 3628 /* device missing or wrong IDENTIFY data, schedule probing */ 3629 ehc->i.probe_mask |= (1 << dev->devno); 3630 fallthrough; 3631 case -EINVAL: 3632 /* give it just one more chance */ 3633 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1); 3634 fallthrough; 3635 case -EIO: 3636 if (ehc->tries[dev->devno] == 1) { 3637 /* This is the last chance, better to slow 3638 * down than lose it. 3639 */ 3640 sata_down_spd_limit(ata_dev_phys_link(dev), 0); 3641 if (dev->pio_mode > XFER_PIO_0) 3642 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO); 3643 } 3644 } 3645 3646 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) { 3647 /* disable device if it has used up all its chances */ 3648 ata_dev_disable(dev); 3649 3650 /* detach if offline */ 3651 if (ata_phys_link_offline(ata_dev_phys_link(dev))) 3652 ata_eh_detach_dev(dev); 3653 3654 /* schedule probe if necessary */ 3655 if (ata_eh_schedule_probe(dev)) { 3656 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; 3657 memset(ehc->cmd_timeout_idx[dev->devno], 0, 3658 sizeof(ehc->cmd_timeout_idx[dev->devno])); 3659 } 3660 3661 return 1; 3662 } else { 3663 ehc->i.action |= ATA_EH_RESET; 3664 return 0; 3665 } 3666 } 3667 3668 /** 3669 * ata_eh_recover - recover host port after error 3670 * @ap: host port to recover 3671 * @prereset: prereset method (can be NULL) 3672 * @softreset: softreset method (can be NULL) 3673 * @hardreset: hardreset method (can be NULL) 3674 * @postreset: postreset method (can be NULL) 3675 * @r_failed_link: out parameter for failed link 3676 * 3677 * This is the alpha and omega, eum and yang, heart and soul of 3678 * libata exception handling. On entry, actions required to 3679 * recover each link and hotplug requests are recorded in the 3680 * link's eh_context. This function executes all the operations 3681 * with appropriate retrials and fallbacks to resurrect failed 3682 * devices, detach goners and greet newcomers. 3683 * 3684 * LOCKING: 3685 * Kernel thread context (may sleep). 3686 * 3687 * RETURNS: 3688 * 0 on success, -errno on failure. 3689 */ 3690 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset, 3691 ata_reset_fn_t softreset, ata_reset_fn_t hardreset, 3692 ata_postreset_fn_t postreset, 3693 struct ata_link **r_failed_link) 3694 { 3695 struct ata_link *link; 3696 struct ata_device *dev; 3697 int rc, nr_fails; 3698 unsigned long flags, deadline; 3699 3700 /* prep for recovery */ 3701 ata_for_each_link(link, ap, EDGE) { 3702 struct ata_eh_context *ehc = &link->eh_context; 3703 3704 /* re-enable link? */ 3705 if (ehc->i.action & ATA_EH_ENABLE_LINK) { 3706 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK); 3707 spin_lock_irqsave(ap->lock, flags); 3708 link->flags &= ~ATA_LFLAG_DISABLED; 3709 spin_unlock_irqrestore(ap->lock, flags); 3710 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK); 3711 } 3712 3713 ata_for_each_dev(dev, link, ALL) { 3714 if (link->flags & ATA_LFLAG_NO_RETRY) 3715 ehc->tries[dev->devno] = 1; 3716 else 3717 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; 3718 3719 /* collect port action mask recorded in dev actions */ 3720 ehc->i.action |= ehc->i.dev_action[dev->devno] & 3721 ~ATA_EH_PERDEV_MASK; 3722 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK; 3723 3724 /* process hotplug request */ 3725 if (dev->flags & ATA_DFLAG_DETACH) 3726 ata_eh_detach_dev(dev); 3727 3728 /* schedule probe if necessary */ 3729 if (!ata_dev_enabled(dev)) 3730 ata_eh_schedule_probe(dev); 3731 } 3732 } 3733 3734 retry: 3735 rc = 0; 3736 3737 /* if UNLOADING, finish immediately */ 3738 if (ap->pflags & ATA_PFLAG_UNLOADING) 3739 goto out; 3740 3741 /* prep for EH */ 3742 ata_for_each_link(link, ap, EDGE) { 3743 struct ata_eh_context *ehc = &link->eh_context; 3744 3745 /* skip EH if possible. */ 3746 if (ata_eh_skip_recovery(link)) 3747 ehc->i.action = 0; 3748 3749 ata_for_each_dev(dev, link, ALL) 3750 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN; 3751 } 3752 3753 /* reset */ 3754 ata_for_each_link(link, ap, EDGE) { 3755 struct ata_eh_context *ehc = &link->eh_context; 3756 3757 if (!(ehc->i.action & ATA_EH_RESET)) 3758 continue; 3759 3760 rc = ata_eh_reset(link, ata_link_nr_vacant(link), 3761 prereset, softreset, hardreset, postreset); 3762 if (rc) { 3763 ata_link_err(link, "reset failed, giving up\n"); 3764 goto out; 3765 } 3766 } 3767 3768 do { 3769 unsigned long now; 3770 3771 /* 3772 * clears ATA_EH_PARK in eh_info and resets 3773 * ap->park_req_pending 3774 */ 3775 ata_eh_pull_park_action(ap); 3776 3777 deadline = jiffies; 3778 ata_for_each_link(link, ap, EDGE) { 3779 ata_for_each_dev(dev, link, ALL) { 3780 struct ata_eh_context *ehc = &link->eh_context; 3781 unsigned long tmp; 3782 3783 if (dev->class != ATA_DEV_ATA && 3784 dev->class != ATA_DEV_ZAC) 3785 continue; 3786 if (!(ehc->i.dev_action[dev->devno] & 3787 ATA_EH_PARK)) 3788 continue; 3789 tmp = dev->unpark_deadline; 3790 if (time_before(deadline, tmp)) 3791 deadline = tmp; 3792 else if (time_before_eq(tmp, jiffies)) 3793 continue; 3794 if (ehc->unloaded_mask & (1 << dev->devno)) 3795 continue; 3796 3797 ata_eh_park_issue_cmd(dev, 1); 3798 } 3799 } 3800 3801 now = jiffies; 3802 if (time_before_eq(deadline, now)) 3803 break; 3804 3805 ata_eh_release(ap); 3806 deadline = wait_for_completion_timeout(&ap->park_req_pending, 3807 deadline - now); 3808 ata_eh_acquire(ap); 3809 } while (deadline); 3810 ata_for_each_link(link, ap, EDGE) { 3811 ata_for_each_dev(dev, link, ALL) { 3812 if (!(link->eh_context.unloaded_mask & 3813 (1 << dev->devno))) 3814 continue; 3815 3816 ata_eh_park_issue_cmd(dev, 0); 3817 ata_eh_done(link, dev, ATA_EH_PARK); 3818 } 3819 } 3820 3821 /* the rest */ 3822 nr_fails = 0; 3823 ata_for_each_link(link, ap, PMP_FIRST) { 3824 struct ata_eh_context *ehc = &link->eh_context; 3825 3826 if (sata_pmp_attached(ap) && ata_is_host_link(link)) 3827 goto config_lpm; 3828 3829 /* revalidate existing devices and attach new ones */ 3830 rc = ata_eh_revalidate_and_attach(link, &dev); 3831 if (rc) 3832 goto rest_fail; 3833 3834 /* if PMP got attached, return, pmp EH will take care of it */ 3835 if (link->device->class == ATA_DEV_PMP) { 3836 ehc->i.action = 0; 3837 return 0; 3838 } 3839 3840 /* configure transfer mode if necessary */ 3841 if (ehc->i.flags & ATA_EHI_SETMODE) { 3842 rc = ata_set_mode(link, &dev); 3843 if (rc) 3844 goto rest_fail; 3845 ehc->i.flags &= ~ATA_EHI_SETMODE; 3846 } 3847 3848 /* If reset has been issued, clear UA to avoid 3849 * disrupting the current users of the device. 3850 */ 3851 if (ehc->i.flags & ATA_EHI_DID_RESET) { 3852 ata_for_each_dev(dev, link, ALL) { 3853 if (dev->class != ATA_DEV_ATAPI) 3854 continue; 3855 rc = atapi_eh_clear_ua(dev); 3856 if (rc) 3857 goto rest_fail; 3858 if (zpodd_dev_enabled(dev)) 3859 zpodd_post_poweron(dev); 3860 } 3861 } 3862 3863 /* retry flush if necessary */ 3864 ata_for_each_dev(dev, link, ALL) { 3865 if (dev->class != ATA_DEV_ATA && 3866 dev->class != ATA_DEV_ZAC) 3867 continue; 3868 rc = ata_eh_maybe_retry_flush(dev); 3869 if (rc) 3870 goto rest_fail; 3871 } 3872 3873 config_lpm: 3874 /* configure link power saving */ 3875 if (link->lpm_policy != ap->target_lpm_policy) { 3876 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev); 3877 if (rc) 3878 goto rest_fail; 3879 } 3880 3881 /* this link is okay now */ 3882 ehc->i.flags = 0; 3883 continue; 3884 3885 rest_fail: 3886 nr_fails++; 3887 if (dev) 3888 ata_eh_handle_dev_fail(dev, rc); 3889 3890 if (ata_port_is_frozen(ap)) { 3891 /* PMP reset requires working host port. 3892 * Can't retry if it's frozen. 3893 */ 3894 if (sata_pmp_attached(ap)) 3895 goto out; 3896 break; 3897 } 3898 } 3899 3900 if (nr_fails) 3901 goto retry; 3902 3903 out: 3904 if (rc && r_failed_link) 3905 *r_failed_link = link; 3906 3907 return rc; 3908 } 3909 3910 /** 3911 * ata_eh_finish - finish up EH 3912 * @ap: host port to finish EH for 3913 * 3914 * Recovery is complete. Clean up EH states and retry or finish 3915 * failed qcs. 3916 * 3917 * LOCKING: 3918 * None. 3919 */ 3920 void ata_eh_finish(struct ata_port *ap) 3921 { 3922 struct ata_queued_cmd *qc; 3923 int tag; 3924 3925 /* retry or finish qcs */ 3926 ata_qc_for_each_raw(ap, qc, tag) { 3927 if (!(qc->flags & ATA_QCFLAG_EH)) 3928 continue; 3929 3930 if (qc->err_mask) { 3931 /* FIXME: Once EH migration is complete, 3932 * generate sense data in this function, 3933 * considering both err_mask and tf. 3934 */ 3935 if (qc->flags & ATA_QCFLAG_RETRY) { 3936 /* 3937 * Since qc->err_mask is set, ata_eh_qc_retry() 3938 * will not increment scmd->allowed, so upper 3939 * layer will only retry the command if it has 3940 * not already been retried too many times. 3941 */ 3942 ata_eh_qc_retry(qc); 3943 } else { 3944 ata_eh_qc_complete(qc); 3945 } 3946 } else { 3947 if (qc->flags & ATA_QCFLAG_SENSE_VALID || 3948 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) { 3949 ata_eh_qc_complete(qc); 3950 } else { 3951 /* feed zero TF to sense generation */ 3952 memset(&qc->result_tf, 0, sizeof(qc->result_tf)); 3953 /* 3954 * Since qc->err_mask is not set, 3955 * ata_eh_qc_retry() will increment 3956 * scmd->allowed, so upper layer is guaranteed 3957 * to retry the command. 3958 */ 3959 ata_eh_qc_retry(qc); 3960 } 3961 } 3962 } 3963 3964 /* make sure nr_active_links is zero after EH */ 3965 WARN_ON(ap->nr_active_links); 3966 ap->nr_active_links = 0; 3967 } 3968 3969 /** 3970 * ata_do_eh - do standard error handling 3971 * @ap: host port to handle error for 3972 * 3973 * @prereset: prereset method (can be NULL) 3974 * @softreset: softreset method (can be NULL) 3975 * @hardreset: hardreset method (can be NULL) 3976 * @postreset: postreset method (can be NULL) 3977 * 3978 * Perform standard error handling sequence. 3979 * 3980 * LOCKING: 3981 * Kernel thread context (may sleep). 3982 */ 3983 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset, 3984 ata_reset_fn_t softreset, ata_reset_fn_t hardreset, 3985 ata_postreset_fn_t postreset) 3986 { 3987 struct ata_device *dev; 3988 int rc; 3989 3990 ata_eh_autopsy(ap); 3991 ata_eh_report(ap); 3992 3993 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset, 3994 NULL); 3995 if (rc) { 3996 ata_for_each_dev(dev, &ap->link, ALL) 3997 ata_dev_disable(dev); 3998 } 3999 4000 ata_eh_finish(ap); 4001 } 4002 4003 /** 4004 * ata_std_error_handler - standard error handler 4005 * @ap: host port to handle error for 4006 * 4007 * Standard error handler 4008 * 4009 * LOCKING: 4010 * Kernel thread context (may sleep). 4011 */ 4012 void ata_std_error_handler(struct ata_port *ap) 4013 { 4014 struct ata_port_operations *ops = ap->ops; 4015 ata_reset_fn_t hardreset = ops->hardreset; 4016 4017 /* ignore built-in hardreset if SCR access is not available */ 4018 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link)) 4019 hardreset = NULL; 4020 4021 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset); 4022 } 4023 EXPORT_SYMBOL_GPL(ata_std_error_handler); 4024 4025 #ifdef CONFIG_PM 4026 /** 4027 * ata_eh_handle_port_suspend - perform port suspend operation 4028 * @ap: port to suspend 4029 * 4030 * Suspend @ap. 4031 * 4032 * LOCKING: 4033 * Kernel thread context (may sleep). 4034 */ 4035 static void ata_eh_handle_port_suspend(struct ata_port *ap) 4036 { 4037 unsigned long flags; 4038 int rc = 0; 4039 struct ata_device *dev; 4040 struct ata_link *link; 4041 4042 /* are we suspending? */ 4043 spin_lock_irqsave(ap->lock, flags); 4044 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || 4045 ap->pm_mesg.event & PM_EVENT_RESUME) { 4046 spin_unlock_irqrestore(ap->lock, flags); 4047 return; 4048 } 4049 spin_unlock_irqrestore(ap->lock, flags); 4050 4051 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED); 4052 4053 /* 4054 * We will reach this point for all of the PM events: 4055 * PM_EVENT_SUSPEND (if runtime pm, PM_EVENT_AUTO will also be set) 4056 * PM_EVENT_FREEZE, and PM_EVENT_HIBERNATE. 4057 * 4058 * We do not want to perform disk spin down for PM_EVENT_FREEZE. 4059 * (Spin down will be performed by the subsequent PM_EVENT_HIBERNATE.) 4060 */ 4061 if (!(ap->pm_mesg.event & PM_EVENT_FREEZE)) { 4062 /* Set all devices attached to the port in standby mode */ 4063 ata_for_each_link(link, ap, HOST_FIRST) { 4064 ata_for_each_dev(dev, link, ENABLED) 4065 ata_dev_power_set_standby(dev); 4066 } 4067 } 4068 4069 /* 4070 * If we have a ZPODD attached, check its zero 4071 * power ready status before the port is frozen. 4072 * Only needed for runtime suspend. 4073 */ 4074 if (PMSG_IS_AUTO(ap->pm_mesg)) { 4075 ata_for_each_dev(dev, &ap->link, ENABLED) { 4076 if (zpodd_dev_enabled(dev)) 4077 zpodd_on_suspend(dev); 4078 } 4079 } 4080 4081 /* suspend */ 4082 ata_eh_freeze_port(ap); 4083 4084 if (ap->ops->port_suspend) 4085 rc = ap->ops->port_suspend(ap, ap->pm_mesg); 4086 4087 ata_acpi_set_state(ap, ap->pm_mesg); 4088 4089 /* update the flags */ 4090 spin_lock_irqsave(ap->lock, flags); 4091 4092 ap->pflags &= ~ATA_PFLAG_PM_PENDING; 4093 if (rc == 0) 4094 ap->pflags |= ATA_PFLAG_SUSPENDED; 4095 else if (ata_port_is_frozen(ap)) 4096 ata_port_schedule_eh(ap); 4097 4098 spin_unlock_irqrestore(ap->lock, flags); 4099 4100 return; 4101 } 4102 4103 /** 4104 * ata_eh_handle_port_resume - perform port resume operation 4105 * @ap: port to resume 4106 * 4107 * Resume @ap. 4108 * 4109 * LOCKING: 4110 * Kernel thread context (may sleep). 4111 */ 4112 static void ata_eh_handle_port_resume(struct ata_port *ap) 4113 { 4114 struct ata_link *link; 4115 struct ata_device *dev; 4116 unsigned long flags; 4117 4118 /* are we resuming? */ 4119 spin_lock_irqsave(ap->lock, flags); 4120 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || 4121 !(ap->pm_mesg.event & PM_EVENT_RESUME)) { 4122 spin_unlock_irqrestore(ap->lock, flags); 4123 return; 4124 } 4125 spin_unlock_irqrestore(ap->lock, flags); 4126 4127 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED)); 4128 4129 /* 4130 * Error timestamps are in jiffies which doesn't run while 4131 * suspended and PHY events during resume isn't too uncommon. 4132 * When the two are combined, it can lead to unnecessary speed 4133 * downs if the machine is suspended and resumed repeatedly. 4134 * Clear error history. 4135 */ 4136 ata_for_each_link(link, ap, HOST_FIRST) 4137 ata_for_each_dev(dev, link, ALL) 4138 ata_ering_clear(&dev->ering); 4139 4140 ata_acpi_set_state(ap, ap->pm_mesg); 4141 4142 if (ap->ops->port_resume) 4143 ap->ops->port_resume(ap); 4144 4145 /* tell ACPI that we're resuming */ 4146 ata_acpi_on_resume(ap); 4147 4148 /* update the flags */ 4149 spin_lock_irqsave(ap->lock, flags); 4150 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED); 4151 ap->pflags |= ATA_PFLAG_RESUMING; 4152 spin_unlock_irqrestore(ap->lock, flags); 4153 } 4154 #endif /* CONFIG_PM */ 4155