2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/config.h>
36 #include <linux/kernel.h>
37 #include <scsi/scsi.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_eh.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_cmnd.h>
42 #include "scsi_transport_api.h"
44 #include <linux/libata.h>
48 static void __ata_port_freeze(struct ata_port *ap);
50 static void ata_ering_record(struct ata_ering *ering, int is_io,
51 unsigned int err_mask)
53 struct ata_ering_entry *ent;
58 ering->cursor %= ATA_ERING_SIZE;
60 ent = &ering->ring[ering->cursor];
62 ent->err_mask = err_mask;
63 ent->timestamp = get_jiffies_64();
66 static struct ata_ering_entry * ata_ering_top(struct ata_ering *ering)
68 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
74 static int ata_ering_map(struct ata_ering *ering,
75 int (*map_fn)(struct ata_ering_entry *, void *),
79 struct ata_ering_entry *ent;
83 ent = &ering->ring[idx];
86 rc = map_fn(ent, arg);
89 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
90 } while (idx != ering->cursor);
96 * ata_scsi_timed_out - SCSI layer time out callback
97 * @cmd: timed out SCSI command
99 * Handles SCSI layer timeout. We race with normal completion of
100 * the qc for @cmd. If the qc is already gone, we lose and let
101 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
102 * timed out and EH should be invoked. Prevent ata_qc_complete()
103 * from finishing it by setting EH_SCHEDULED and return
106 * TODO: kill this function once old EH is gone.
109 * Called from timer context
112 * EH_HANDLED or EH_NOT_HANDLED
114 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
116 struct Scsi_Host *host = cmd->device->host;
117 struct ata_port *ap = ata_shost_to_port(host);
119 struct ata_queued_cmd *qc;
120 enum scsi_eh_timer_return ret;
124 if (ap->ops->error_handler) {
125 ret = EH_NOT_HANDLED;
130 spin_lock_irqsave(&ap->host_set->lock, flags);
131 qc = ata_qc_from_tag(ap, ap->active_tag);
133 WARN_ON(qc->scsicmd != cmd);
134 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
135 qc->err_mask |= AC_ERR_TIMEOUT;
136 ret = EH_NOT_HANDLED;
138 spin_unlock_irqrestore(&ap->host_set->lock, flags);
141 DPRINTK("EXIT, ret=%d\n", ret);
146 * ata_scsi_error - SCSI layer error handler callback
147 * @host: SCSI host on which error occurred
149 * Handles SCSI-layer-thrown error events.
152 * Inherited from SCSI layer (none, can sleep)
157 void ata_scsi_error(struct Scsi_Host *host)
159 struct ata_port *ap = ata_shost_to_port(host);
160 spinlock_t *hs_lock = &ap->host_set->lock;
161 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
166 /* synchronize with port task */
167 ata_port_flush_task(ap);
169 /* synchronize with host_set lock and sort out timeouts */
171 /* For new EH, all qcs are finished in one of three ways -
172 * normal completion, error completion, and SCSI timeout.
173 * Both cmpletions can race against SCSI timeout. When normal
174 * completion wins, the qc never reaches EH. When error
175 * completion wins, the qc has ATA_QCFLAG_FAILED set.
177 * When SCSI timeout wins, things are a bit more complex.
178 * Normal or error completion can occur after the timeout but
179 * before this point. In such cases, both types of
180 * completions are honored. A scmd is determined to have
181 * timed out iff its associated qc is active and not failed.
183 if (ap->ops->error_handler) {
184 struct scsi_cmnd *scmd, *tmp;
187 spin_lock_irqsave(hs_lock, flags);
189 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
190 struct ata_queued_cmd *qc;
192 for (i = 0; i < ATA_MAX_QUEUE; i++) {
193 qc = __ata_qc_from_tag(ap, i);
194 if (qc->flags & ATA_QCFLAG_ACTIVE &&
199 if (i < ATA_MAX_QUEUE) {
200 /* the scmd has an associated qc */
201 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
202 /* which hasn't failed yet, timeout */
203 qc->err_mask |= AC_ERR_TIMEOUT;
204 qc->flags |= ATA_QCFLAG_FAILED;
208 /* Normal completion occurred after
209 * SCSI timeout but before this point.
210 * Successfully complete it.
212 scmd->retries = scmd->allowed;
213 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
217 /* If we have timed out qcs. They belong to EH from
218 * this point but the state of the controller is
219 * unknown. Freeze the port to make sure the IRQ
220 * handler doesn't diddle with those qcs. This must
221 * be done atomically w.r.t. setting QCFLAG_FAILED.
224 __ata_port_freeze(ap);
226 spin_unlock_irqrestore(hs_lock, flags);
228 spin_unlock_wait(hs_lock);
231 /* invoke error handler */
232 if (ap->ops->error_handler) {
233 /* fetch & clear EH info */
234 spin_lock_irqsave(hs_lock, flags);
236 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
237 ap->eh_context.i = ap->eh_info;
238 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
240 ap->flags |= ATA_FLAG_EH_IN_PROGRESS;
241 ap->flags &= ~ATA_FLAG_EH_PENDING;
243 spin_unlock_irqrestore(hs_lock, flags);
246 ap->ops->error_handler(ap);
248 /* Exception might have happend after ->error_handler
249 * recovered the port but before this point. Repeat
252 spin_lock_irqsave(hs_lock, flags);
254 if (ap->flags & ATA_FLAG_EH_PENDING) {
256 ata_port_printk(ap, KERN_INFO,
257 "EH pending after completion, "
258 "repeating EH (cnt=%d)\n", repeat_cnt);
259 spin_unlock_irqrestore(hs_lock, flags);
262 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
263 "tries, giving up\n", ATA_EH_MAX_REPEAT);
266 /* this run is complete, make sure EH info is clear */
267 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
269 /* Clear host_eh_scheduled while holding hs_lock such
270 * that if exception occurs after this point but
271 * before EH completion, SCSI midlayer will
274 host->host_eh_scheduled = 0;
276 spin_unlock_irqrestore(hs_lock, flags);
278 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
279 ap->ops->eng_timeout(ap);
282 /* finish or retry handled scmd's and clean up */
283 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
285 scsi_eh_flush_done_q(&ap->eh_done_q);
288 spin_lock_irqsave(hs_lock, flags);
290 if (ap->flags & ATA_FLAG_RECOVERED)
291 ata_port_printk(ap, KERN_INFO, "EH complete\n");
292 ap->flags &= ~ATA_FLAG_RECOVERED;
294 /* tell wait_eh that we're done */
295 ap->flags &= ~ATA_FLAG_EH_IN_PROGRESS;
296 wake_up_all(&ap->eh_wait_q);
298 spin_unlock_irqrestore(hs_lock, flags);
304 * ata_port_wait_eh - Wait for the currently pending EH to complete
305 * @ap: Port to wait EH for
307 * Wait until the currently pending EH is complete.
310 * Kernel thread context (may sleep).
312 void ata_port_wait_eh(struct ata_port *ap)
318 spin_lock_irqsave(&ap->host_set->lock, flags);
320 while (ap->flags & (ATA_FLAG_EH_PENDING | ATA_FLAG_EH_IN_PROGRESS)) {
321 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
322 spin_unlock_irqrestore(&ap->host_set->lock, flags);
324 spin_lock_irqsave(&ap->host_set->lock, flags);
327 spin_unlock_irqrestore(&ap->host_set->lock, flags);
329 /* make sure SCSI EH is complete */
330 if (scsi_host_in_recovery(ap->host)) {
337 * ata_qc_timeout - Handle timeout of queued command
338 * @qc: Command that timed out
340 * Some part of the kernel (currently, only the SCSI layer)
341 * has noticed that the active command on port @ap has not
342 * completed after a specified length of time. Handle this
343 * condition by disabling DMA (if necessary) and completing
344 * transactions, with error if necessary.
346 * This also handles the case of the "lost interrupt", where
347 * for some reason (possibly hardware bug, possibly driver bug)
348 * an interrupt was not delivered to the driver, even though the
349 * transaction completed successfully.
351 * TODO: kill this function once old EH is gone.
354 * Inherited from SCSI layer (none, can sleep)
356 static void ata_qc_timeout(struct ata_queued_cmd *qc)
358 struct ata_port *ap = qc->ap;
359 struct ata_host_set *host_set = ap->host_set;
360 u8 host_stat = 0, drv_stat;
365 ap->hsm_task_state = HSM_ST_IDLE;
367 spin_lock_irqsave(&host_set->lock, flags);
369 switch (qc->tf.protocol) {
372 case ATA_PROT_ATAPI_DMA:
373 host_stat = ap->ops->bmdma_status(ap);
375 /* before we do anything else, clear DMA-Start bit */
376 ap->ops->bmdma_stop(qc);
382 drv_stat = ata_chk_status(ap);
384 /* ack bmdma irq events */
385 ap->ops->irq_clear(ap);
387 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
388 "stat 0x%x host_stat 0x%x\n",
389 qc->tf.command, drv_stat, host_stat);
391 /* complete taskfile transaction */
392 qc->err_mask |= AC_ERR_TIMEOUT;
396 spin_unlock_irqrestore(&host_set->lock, flags);
398 ata_eh_qc_complete(qc);
404 * ata_eng_timeout - Handle timeout of queued command
405 * @ap: Port on which timed-out command is active
407 * Some part of the kernel (currently, only the SCSI layer)
408 * has noticed that the active command on port @ap has not
409 * completed after a specified length of time. Handle this
410 * condition by disabling DMA (if necessary) and completing
411 * transactions, with error if necessary.
413 * This also handles the case of the "lost interrupt", where
414 * for some reason (possibly hardware bug, possibly driver bug)
415 * an interrupt was not delivered to the driver, even though the
416 * transaction completed successfully.
418 * TODO: kill this function once old EH is gone.
421 * Inherited from SCSI layer (none, can sleep)
423 void ata_eng_timeout(struct ata_port *ap)
427 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
433 * ata_qc_schedule_eh - schedule qc for error handling
434 * @qc: command to schedule error handling for
436 * Schedule error handling for @qc. EH will kick in as soon as
437 * other commands are drained.
440 * spin_lock_irqsave(host_set lock)
442 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
444 struct ata_port *ap = qc->ap;
446 WARN_ON(!ap->ops->error_handler);
448 qc->flags |= ATA_QCFLAG_FAILED;
449 qc->ap->flags |= ATA_FLAG_EH_PENDING;
451 /* The following will fail if timeout has already expired.
452 * ata_scsi_error() takes care of such scmds on EH entry.
453 * Note that ATA_QCFLAG_FAILED is unconditionally set after
454 * this function completes.
456 scsi_req_abort_cmd(qc->scsicmd);
460 * ata_port_schedule_eh - schedule error handling without a qc
461 * @ap: ATA port to schedule EH for
463 * Schedule error handling for @ap. EH will kick in as soon as
464 * all commands are drained.
467 * spin_lock_irqsave(host_set lock)
469 void ata_port_schedule_eh(struct ata_port *ap)
471 WARN_ON(!ap->ops->error_handler);
473 ap->flags |= ATA_FLAG_EH_PENDING;
474 scsi_schedule_eh(ap->host);
476 DPRINTK("port EH scheduled\n");
480 * ata_port_abort - abort all qc's on the port
481 * @ap: ATA port to abort qc's for
483 * Abort all active qc's of @ap and schedule EH.
486 * spin_lock_irqsave(host_set lock)
489 * Number of aborted qc's.
491 int ata_port_abort(struct ata_port *ap)
493 int tag, nr_aborted = 0;
495 WARN_ON(!ap->ops->error_handler);
497 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
498 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
501 qc->flags |= ATA_QCFLAG_FAILED;
508 ata_port_schedule_eh(ap);
514 * __ata_port_freeze - freeze port
515 * @ap: ATA port to freeze
517 * This function is called when HSM violation or some other
518 * condition disrupts normal operation of the port. Frozen port
519 * is not allowed to perform any operation until the port is
520 * thawed, which usually follows a successful reset.
522 * ap->ops->freeze() callback can be used for freezing the port
523 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
524 * port cannot be frozen hardware-wise, the interrupt handler
525 * must ack and clear interrupts unconditionally while the port
529 * spin_lock_irqsave(host_set lock)
531 static void __ata_port_freeze(struct ata_port *ap)
533 WARN_ON(!ap->ops->error_handler);
538 ap->flags |= ATA_FLAG_FROZEN;
540 DPRINTK("ata%u port frozen\n", ap->id);
544 * ata_port_freeze - abort & freeze port
545 * @ap: ATA port to freeze
547 * Abort and freeze @ap.
550 * spin_lock_irqsave(host_set lock)
553 * Number of aborted commands.
555 int ata_port_freeze(struct ata_port *ap)
559 WARN_ON(!ap->ops->error_handler);
561 nr_aborted = ata_port_abort(ap);
562 __ata_port_freeze(ap);
568 * ata_eh_freeze_port - EH helper to freeze port
569 * @ap: ATA port to freeze
576 void ata_eh_freeze_port(struct ata_port *ap)
580 if (!ap->ops->error_handler)
583 spin_lock_irqsave(&ap->host_set->lock, flags);
584 __ata_port_freeze(ap);
585 spin_unlock_irqrestore(&ap->host_set->lock, flags);
589 * ata_port_thaw_port - EH helper to thaw port
590 * @ap: ATA port to thaw
592 * Thaw frozen port @ap.
597 void ata_eh_thaw_port(struct ata_port *ap)
601 if (!ap->ops->error_handler)
604 spin_lock_irqsave(&ap->host_set->lock, flags);
606 ap->flags &= ~ATA_FLAG_FROZEN;
611 spin_unlock_irqrestore(&ap->host_set->lock, flags);
613 DPRINTK("ata%u port thawed\n", ap->id);
616 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
621 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
623 struct ata_port *ap = qc->ap;
624 struct scsi_cmnd *scmd = qc->scsicmd;
627 spin_lock_irqsave(&ap->host_set->lock, flags);
628 qc->scsidone = ata_eh_scsidone;
629 __ata_qc_complete(qc);
630 WARN_ON(ata_tag_valid(qc->tag));
631 spin_unlock_irqrestore(&ap->host_set->lock, flags);
633 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
637 * ata_eh_qc_complete - Complete an active ATA command from EH
638 * @qc: Command to complete
640 * Indicate to the mid and upper layers that an ATA command has
641 * completed. To be used from EH.
643 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
645 struct scsi_cmnd *scmd = qc->scsicmd;
646 scmd->retries = scmd->allowed;
647 __ata_eh_qc_complete(qc);
651 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
652 * @qc: Command to retry
654 * Indicate to the mid and upper layers that an ATA command
655 * should be retried. To be used from EH.
657 * SCSI midlayer limits the number of retries to scmd->allowed.
658 * scmd->retries is decremented for commands which get retried
659 * due to unrelated failures (qc->err_mask is zero).
661 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
663 struct scsi_cmnd *scmd = qc->scsicmd;
664 if (!qc->err_mask && scmd->retries)
666 __ata_eh_qc_complete(qc);
670 * ata_eh_about_to_do - about to perform eh_action
671 * @ap: target ATA port
672 * @action: action about to be performed
674 * Called just before performing EH actions to clear related bits
675 * in @ap->eh_info such that eh actions are not unnecessarily
681 static void ata_eh_about_to_do(struct ata_port *ap, unsigned int action)
685 spin_lock_irqsave(&ap->host_set->lock, flags);
686 ap->eh_info.action &= ~action;
687 ap->flags |= ATA_FLAG_RECOVERED;
688 spin_unlock_irqrestore(&ap->host_set->lock, flags);
692 * ata_err_string - convert err_mask to descriptive string
693 * @err_mask: error mask to convert to string
695 * Convert @err_mask to descriptive string. Errors are
696 * prioritized according to severity and only the most severe
703 * Descriptive string for @err_mask
705 static const char * ata_err_string(unsigned int err_mask)
707 if (err_mask & AC_ERR_HOST_BUS)
708 return "host bus error";
709 if (err_mask & AC_ERR_ATA_BUS)
710 return "ATA bus error";
711 if (err_mask & AC_ERR_TIMEOUT)
713 if (err_mask & AC_ERR_HSM)
714 return "HSM violation";
715 if (err_mask & AC_ERR_SYSTEM)
716 return "internal error";
717 if (err_mask & AC_ERR_MEDIA)
718 return "media error";
719 if (err_mask & AC_ERR_INVALID)
720 return "invalid argument";
721 if (err_mask & AC_ERR_DEV)
722 return "device error";
723 return "unknown error";
727 * ata_read_log_page - read a specific log page
728 * @dev: target device
729 * @page: page to read
730 * @buf: buffer to store read page
731 * @sectors: number of sectors to read
733 * Read log page using READ_LOG_EXT command.
736 * Kernel thread context (may sleep).
739 * 0 on success, AC_ERR_* mask otherwise.
741 static unsigned int ata_read_log_page(struct ata_device *dev,
742 u8 page, void *buf, unsigned int sectors)
744 struct ata_taskfile tf;
745 unsigned int err_mask;
747 DPRINTK("read log page - page %d\n", page);
749 ata_tf_init(dev, &tf);
750 tf.command = ATA_CMD_READ_LOG_EXT;
753 tf.hob_nsect = sectors >> 8;
754 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
755 tf.protocol = ATA_PROT_PIO;
757 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
758 buf, sectors * ATA_SECT_SIZE);
760 DPRINTK("EXIT, err_mask=%x\n", err_mask);
765 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
766 * @dev: Device to read log page 10h from
767 * @tag: Resulting tag of the failed command
768 * @tf: Resulting taskfile registers of the failed command
770 * Read log page 10h to obtain NCQ error details and clear error
774 * Kernel thread context (may sleep).
777 * 0 on success, -errno otherwise.
779 static int ata_eh_read_log_10h(struct ata_device *dev,
780 int *tag, struct ata_taskfile *tf)
782 u8 *buf = dev->ap->sector_buf;
783 unsigned int err_mask;
787 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
792 for (i = 0; i < ATA_SECT_SIZE; i++)
795 ata_dev_printk(dev, KERN_WARNING,
796 "invalid checksum 0x%x on log page 10h\n", csum);
801 *tag = buf[0] & 0x1f;
803 tf->command = buf[2];
804 tf->feature = buf[3];
809 tf->hob_lbal = buf[8];
810 tf->hob_lbam = buf[9];
811 tf->hob_lbah = buf[10];
813 tf->hob_nsect = buf[13];
819 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
820 * @dev: device to perform REQUEST_SENSE to
821 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
823 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
824 * SENSE. This function is EH helper.
827 * Kernel thread context (may sleep).
830 * 0 on success, AC_ERR_* mask on failure
832 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
833 unsigned char *sense_buf)
835 struct ata_port *ap = dev->ap;
836 struct ata_taskfile tf;
837 u8 cdb[ATAPI_CDB_LEN];
839 DPRINTK("ATAPI request sense\n");
841 ata_tf_init(dev, &tf);
843 /* FIXME: is this needed? */
844 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
846 /* XXX: why tf_read here? */
847 ap->ops->tf_read(ap, &tf);
849 /* fill these in, for the case where they are -not- overwritten */
851 sense_buf[2] = tf.feature >> 4;
853 memset(cdb, 0, ATAPI_CDB_LEN);
854 cdb[0] = REQUEST_SENSE;
855 cdb[4] = SCSI_SENSE_BUFFERSIZE;
857 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
858 tf.command = ATA_CMD_PACKET;
860 /* is it pointless to prefer PIO for "safety reasons"? */
861 if (ap->flags & ATA_FLAG_PIO_DMA) {
862 tf.protocol = ATA_PROT_ATAPI_DMA;
863 tf.feature |= ATAPI_PKT_DMA;
865 tf.protocol = ATA_PROT_ATAPI;
866 tf.lbam = (8 * 1024) & 0xff;
867 tf.lbah = (8 * 1024) >> 8;
870 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
871 sense_buf, SCSI_SENSE_BUFFERSIZE);
875 * ata_eh_analyze_serror - analyze SError for a failed port
876 * @ap: ATA port to analyze SError for
878 * Analyze SError if available and further determine cause of
884 static void ata_eh_analyze_serror(struct ata_port *ap)
886 struct ata_eh_context *ehc = &ap->eh_context;
887 u32 serror = ehc->i.serror;
888 unsigned int err_mask = 0, action = 0;
890 if (serror & SERR_PERSISTENT) {
891 err_mask |= AC_ERR_ATA_BUS;
892 action |= ATA_EH_HARDRESET;
895 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
896 err_mask |= AC_ERR_ATA_BUS;
897 action |= ATA_EH_SOFTRESET;
899 if (serror & SERR_PROTOCOL) {
900 err_mask |= AC_ERR_HSM;
901 action |= ATA_EH_SOFTRESET;
903 if (serror & SERR_INTERNAL) {
904 err_mask |= AC_ERR_SYSTEM;
905 action |= ATA_EH_SOFTRESET;
907 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG)) {
908 err_mask |= AC_ERR_ATA_BUS;
909 action |= ATA_EH_HARDRESET;
912 ehc->i.err_mask |= err_mask;
913 ehc->i.action |= action;
917 * ata_eh_analyze_ncq_error - analyze NCQ error
918 * @ap: ATA port to analyze NCQ error for
920 * Read log page 10h, determine the offending qc and acquire
921 * error status TF. For NCQ device errors, all LLDDs have to do
922 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
926 * Kernel thread context (may sleep).
928 static void ata_eh_analyze_ncq_error(struct ata_port *ap)
930 struct ata_eh_context *ehc = &ap->eh_context;
931 struct ata_device *dev = ap->device;
932 struct ata_queued_cmd *qc;
933 struct ata_taskfile tf;
936 /* if frozen, we can't do much */
937 if (ap->flags & ATA_FLAG_FROZEN)
940 /* is it NCQ device error? */
941 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
944 /* has LLDD analyzed already? */
945 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
946 qc = __ata_qc_from_tag(ap, tag);
948 if (!(qc->flags & ATA_QCFLAG_FAILED))
955 /* okay, this error is ours */
956 rc = ata_eh_read_log_10h(dev, &tag, &tf);
958 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
963 if (!(ap->sactive & (1 << tag))) {
964 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
965 "inactive tag %d\n", tag);
969 /* we've got the perpetrator, condemn it */
970 qc = __ata_qc_from_tag(ap, tag);
971 memcpy(&qc->result_tf, &tf, sizeof(tf));
972 qc->err_mask |= AC_ERR_DEV;
973 ehc->i.err_mask &= ~AC_ERR_DEV;
977 * ata_eh_analyze_tf - analyze taskfile of a failed qc
979 * @tf: Taskfile registers to analyze
981 * Analyze taskfile of @qc and further determine cause of
982 * failure. This function also requests ATAPI sense data if
986 * Kernel thread context (may sleep).
989 * Determined recovery action
991 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
992 const struct ata_taskfile *tf)
994 unsigned int tmp, action = 0;
995 u8 stat = tf->command, err = tf->feature;
997 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
998 qc->err_mask |= AC_ERR_HSM;
999 return ATA_EH_SOFTRESET;
1002 if (!(qc->err_mask & AC_ERR_DEV))
1005 switch (qc->dev->class) {
1008 qc->err_mask |= AC_ERR_ATA_BUS;
1010 qc->err_mask |= AC_ERR_MEDIA;
1012 qc->err_mask |= AC_ERR_INVALID;
1016 tmp = atapi_eh_request_sense(qc->dev,
1017 qc->scsicmd->sense_buffer);
1019 /* ATA_QCFLAG_SENSE_VALID is used to tell
1020 * atapi_qc_complete() that sense data is
1023 * TODO: interpret sense data and set
1024 * appropriate err_mask.
1026 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1028 qc->err_mask |= tmp;
1031 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1032 action |= ATA_EH_SOFTRESET;
1037 static int ata_eh_categorize_ering_entry(struct ata_ering_entry *ent)
1039 if (ent->err_mask & (AC_ERR_ATA_BUS | AC_ERR_TIMEOUT))
1043 if (ent->err_mask & AC_ERR_HSM)
1045 if ((ent->err_mask &
1046 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1053 struct speed_down_needed_arg {
1058 static int speed_down_needed_cb(struct ata_ering_entry *ent, void *void_arg)
1060 struct speed_down_needed_arg *arg = void_arg;
1062 if (ent->timestamp < arg->since)
1065 arg->nr_errors[ata_eh_categorize_ering_entry(ent)]++;
1070 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1071 * @dev: Device of interest
1073 * This function examines error ring of @dev and determines
1074 * whether speed down is necessary. Speed down is necessary if
1075 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1076 * errors during last 15 minutes.
1078 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1079 * violation for known supported commands.
1081 * Cat-2 errors are unclassified DEV error for known supported
1085 * Inherited from caller.
1088 * 1 if speed down is necessary, 0 otherwise
1090 static int ata_eh_speed_down_needed(struct ata_device *dev)
1092 const u64 interval = 15LLU * 60 * HZ;
1093 static const int err_limits[3] = { -1, 3, 10 };
1094 struct speed_down_needed_arg arg;
1095 struct ata_ering_entry *ent;
1099 ent = ata_ering_top(&dev->ering);
1103 err_cat = ata_eh_categorize_ering_entry(ent);
1107 memset(&arg, 0, sizeof(arg));
1109 j64 = get_jiffies_64();
1110 if (j64 >= interval)
1111 arg.since = j64 - interval;
1115 ata_ering_map(&dev->ering, speed_down_needed_cb, &arg);
1117 return arg.nr_errors[err_cat] > err_limits[err_cat];
1121 * ata_eh_speed_down - record error and speed down if necessary
1122 * @dev: Failed device
1123 * @is_io: Did the device fail during normal IO?
1124 * @err_mask: err_mask of the error
1126 * Record error and examine error history to determine whether
1127 * adjusting transmission speed is necessary. It also sets
1128 * transmission limits appropriately if such adjustment is
1132 * Kernel thread context (may sleep).
1135 * 0 on success, -errno otherwise
1137 static int ata_eh_speed_down(struct ata_device *dev, int is_io,
1138 unsigned int err_mask)
1143 /* record error and determine whether speed down is necessary */
1144 ata_ering_record(&dev->ering, is_io, err_mask);
1146 if (!ata_eh_speed_down_needed(dev))
1149 /* speed down SATA link speed if possible */
1150 if (sata_down_spd_limit(dev->ap) == 0)
1151 return ATA_EH_HARDRESET;
1153 /* lower transfer mode */
1154 if (ata_down_xfermask_limit(dev, 0) == 0)
1155 return ATA_EH_SOFTRESET;
1157 ata_dev_printk(dev, KERN_ERR,
1158 "speed down requested but no transfer mode left\n");
1163 * ata_eh_autopsy - analyze error and determine recovery action
1164 * @ap: ATA port to perform autopsy on
1166 * Analyze why @ap failed and determine which recovery action is
1167 * needed. This function also sets more detailed AC_ERR_* values
1168 * and fills sense data for ATAPI CHECK SENSE.
1171 * Kernel thread context (may sleep).
1173 static void ata_eh_autopsy(struct ata_port *ap)
1175 struct ata_eh_context *ehc = &ap->eh_context;
1176 unsigned int action = ehc->i.action;
1177 struct ata_device *failed_dev = NULL;
1178 unsigned int all_err_mask = 0;
1185 /* obtain and analyze SError */
1186 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1188 ehc->i.serror |= serror;
1189 ata_eh_analyze_serror(ap);
1190 } else if (rc != -EOPNOTSUPP)
1191 action |= ATA_EH_HARDRESET;
1193 /* analyze NCQ failure */
1194 ata_eh_analyze_ncq_error(ap);
1196 /* any real error trumps AC_ERR_OTHER */
1197 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1198 ehc->i.err_mask &= ~AC_ERR_OTHER;
1200 all_err_mask |= ehc->i.err_mask;
1202 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1203 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1205 if (!(qc->flags & ATA_QCFLAG_FAILED))
1208 /* inherit upper level err_mask */
1209 qc->err_mask |= ehc->i.err_mask;
1212 action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1214 /* DEV errors are probably spurious in case of ATA_BUS error */
1215 if (qc->err_mask & AC_ERR_ATA_BUS)
1216 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1219 /* any real error trumps unknown error */
1220 if (qc->err_mask & ~AC_ERR_OTHER)
1221 qc->err_mask &= ~AC_ERR_OTHER;
1223 /* SENSE_VALID trumps dev/unknown error and revalidation */
1224 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1225 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1226 action &= ~ATA_EH_REVALIDATE;
1229 /* accumulate error info */
1230 failed_dev = qc->dev;
1231 all_err_mask |= qc->err_mask;
1232 if (qc->flags & ATA_QCFLAG_IO)
1236 /* speed down iff command was in progress */
1238 action |= ata_eh_speed_down(failed_dev, is_io, all_err_mask);
1240 /* enforce default EH actions */
1241 if (ap->flags & ATA_FLAG_FROZEN ||
1242 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1243 action |= ATA_EH_SOFTRESET;
1244 else if (all_err_mask)
1245 action |= ATA_EH_REVALIDATE;
1247 /* record autopsy result */
1248 ehc->i.dev = failed_dev;
1249 ehc->i.action = action;
1255 * ata_eh_report - report error handling to user
1256 * @ap: ATA port EH is going on
1258 * Report EH to user.
1263 static void ata_eh_report(struct ata_port *ap)
1265 struct ata_eh_context *ehc = &ap->eh_context;
1266 const char *frozen, *desc;
1267 int tag, nr_failed = 0;
1270 if (ehc->i.desc[0] != '\0')
1273 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1274 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1276 if (!(qc->flags & ATA_QCFLAG_FAILED))
1278 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1284 if (!nr_failed && !ehc->i.err_mask)
1288 if (ap->flags & ATA_FLAG_FROZEN)
1292 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1293 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1294 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1295 ehc->i.action, frozen);
1297 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1299 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1300 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1301 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1302 ehc->i.action, frozen);
1304 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1307 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1308 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1310 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1313 ata_dev_printk(qc->dev, KERN_ERR, "tag %d cmd 0x%x "
1314 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1315 qc->tag, qc->tf.command, qc->err_mask,
1316 qc->result_tf.command, qc->result_tf.feature,
1317 ata_err_string(qc->err_mask));
1321 static int ata_eh_followup_srst_needed(int rc, int classify,
1322 const unsigned int *classes)
1328 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1333 static int ata_eh_reset(struct ata_port *ap, int classify,
1334 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1335 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1337 struct ata_eh_context *ehc = &ap->eh_context;
1338 unsigned int *classes = ehc->classes;
1339 int tries = ATA_EH_RESET_TRIES;
1340 unsigned int action;
1341 ata_reset_fn_t reset;
1342 int i, did_followup_srst, rc;
1344 /* Determine which reset to use and record in ehc->i.action.
1345 * prereset() may examine and modify it.
1347 action = ehc->i.action;
1348 ehc->i.action &= ~ATA_EH_RESET_MASK;
1349 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
1350 !(action & ATA_EH_HARDRESET))))
1351 ehc->i.action |= ATA_EH_SOFTRESET;
1353 ehc->i.action |= ATA_EH_HARDRESET;
1358 ata_port_printk(ap, KERN_ERR,
1359 "prereset failed (errno=%d)\n", rc);
1364 /* prereset() might have modified ehc->i.action */
1365 if (ehc->i.action & ATA_EH_HARDRESET)
1367 else if (ehc->i.action & ATA_EH_SOFTRESET)
1370 /* prereset told us not to reset, bang classes and return */
1371 for (i = 0; i < ATA_MAX_DEVICES; i++)
1372 classes[i] = ATA_DEV_NONE;
1376 /* did prereset() screw up? if so, fix up to avoid oopsing */
1378 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1379 "invalid reset type\n");
1387 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1388 reset == softreset ? "soft" : "hard");
1391 ata_eh_about_to_do(ap, ATA_EH_RESET_MASK);
1392 ehc->i.flags |= ATA_EHI_DID_RESET;
1394 rc = ata_do_reset(ap, reset, classes);
1396 did_followup_srst = 0;
1397 if (reset == hardreset &&
1398 ata_eh_followup_srst_needed(rc, classify, classes)) {
1399 /* okay, let's do follow-up softreset */
1400 did_followup_srst = 1;
1404 ata_port_printk(ap, KERN_ERR,
1405 "follow-up softreset required "
1406 "but no softreset avaliable\n");
1410 ata_eh_about_to_do(ap, ATA_EH_RESET_MASK);
1411 rc = ata_do_reset(ap, reset, classes);
1413 if (rc == 0 && classify &&
1414 classes[0] == ATA_DEV_UNKNOWN) {
1415 ata_port_printk(ap, KERN_ERR,
1416 "classification failed\n");
1421 if (rc && --tries) {
1424 if (reset == softreset) {
1425 if (did_followup_srst)
1426 type = "follow-up soft";
1432 ata_port_printk(ap, KERN_WARNING,
1433 "%sreset failed, retrying in 5 secs\n", type);
1436 if (reset == hardreset)
1437 sata_down_spd_limit(ap);
1444 /* After the reset, the device state is PIO 0 and the
1445 * controller state is undefined. Record the mode.
1447 for (i = 0; i < ATA_MAX_DEVICES; i++)
1448 ap->device[i].pio_mode = XFER_PIO_0;
1451 postreset(ap, classes);
1453 /* reset successful, schedule revalidation */
1455 ehc->i.action &= ~ATA_EH_RESET_MASK;
1456 ehc->i.action |= ATA_EH_REVALIDATE;
1462 static int ata_eh_revalidate(struct ata_port *ap,
1463 struct ata_device **r_failed_dev)
1465 struct ata_eh_context *ehc = &ap->eh_context;
1466 struct ata_device *dev;
1471 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1472 dev = &ap->device[i];
1474 if (ehc->i.action & ATA_EH_REVALIDATE && ata_dev_enabled(dev) &&
1475 (!ehc->i.dev || ehc->i.dev == dev)) {
1476 if (ata_port_offline(ap)) {
1481 ata_eh_about_to_do(ap, ATA_EH_REVALIDATE);
1482 rc = ata_dev_revalidate(dev,
1483 ehc->i.flags & ATA_EHI_DID_RESET);
1487 ehc->i.action &= ~ATA_EH_REVALIDATE;
1492 *r_failed_dev = dev;
1498 static int ata_port_nr_enabled(struct ata_port *ap)
1502 for (i = 0; i < ATA_MAX_DEVICES; i++)
1503 if (ata_dev_enabled(&ap->device[i]))
1509 * ata_eh_recover - recover host port after error
1510 * @ap: host port to recover
1511 * @prereset: prereset method (can be NULL)
1512 * @softreset: softreset method (can be NULL)
1513 * @hardreset: hardreset method (can be NULL)
1514 * @postreset: postreset method (can be NULL)
1516 * This is the alpha and omega, eum and yang, heart and soul of
1517 * libata exception handling. On entry, actions required to
1518 * recover each devices are recorded in eh_context. This
1519 * function executes all the operations with appropriate retrials
1520 * and fallbacks to resurrect failed devices.
1523 * Kernel thread context (may sleep).
1526 * 0 on success, -errno on failure.
1528 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
1529 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
1530 ata_postreset_fn_t postreset)
1532 struct ata_eh_context *ehc = &ap->eh_context;
1533 struct ata_device *dev;
1534 int down_xfermask, i, rc;
1538 /* prep for recovery */
1539 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1540 dev = &ap->device[i];
1542 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
1549 /* skip EH if possible. */
1550 if (!ata_port_nr_enabled(ap) && !(ap->flags & ATA_FLAG_FROZEN))
1554 if (ehc->i.action & ATA_EH_RESET_MASK) {
1555 ata_eh_freeze_port(ap);
1557 rc = ata_eh_reset(ap, 0, prereset, softreset, hardreset,
1560 ata_port_printk(ap, KERN_ERR,
1561 "reset failed, giving up\n");
1565 ata_eh_thaw_port(ap);
1568 /* revalidate existing devices */
1569 rc = ata_eh_revalidate(ap, &dev);
1573 /* configure transfer mode if the port has been reset */
1574 if (ehc->i.flags & ATA_EHI_DID_RESET) {
1575 rc = ata_set_mode(ap, &dev);
1588 ehc->tries[dev->devno] = 0;
1591 sata_down_spd_limit(ap);
1593 ehc->tries[dev->devno]--;
1594 if (down_xfermask &&
1595 ata_down_xfermask_limit(dev, ehc->tries[dev->devno] == 1))
1596 ehc->tries[dev->devno] = 0;
1599 /* disable device if it has used up all its chances */
1600 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno])
1601 ata_dev_disable(dev);
1603 /* soft didn't work? be haaaaard */
1604 if (ehc->i.flags & ATA_EHI_DID_RESET)
1605 ehc->i.action |= ATA_EH_HARDRESET;
1607 ehc->i.action |= ATA_EH_SOFTRESET;
1609 if (ata_port_nr_enabled(ap)) {
1610 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
1611 "devices, retrying in 5 secs\n");
1614 /* no device left, repeat fast */
1622 for (i = 0; i < ATA_MAX_DEVICES; i++)
1623 ata_dev_disable(&ap->device[i]);
1626 DPRINTK("EXIT, rc=%d\n", rc);
1631 * ata_eh_finish - finish up EH
1632 * @ap: host port to finish EH for
1634 * Recovery is complete. Clean up EH states and retry or finish
1640 static void ata_eh_finish(struct ata_port *ap)
1644 /* retry or finish qcs */
1645 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1646 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1648 if (!(qc->flags & ATA_QCFLAG_FAILED))
1652 /* FIXME: Once EH migration is complete,
1653 * generate sense data in this function,
1654 * considering both err_mask and tf.
1656 if (qc->err_mask & AC_ERR_INVALID)
1657 ata_eh_qc_complete(qc);
1659 ata_eh_qc_retry(qc);
1661 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1662 ata_eh_qc_complete(qc);
1664 /* feed zero TF to sense generation */
1665 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
1666 ata_eh_qc_retry(qc);
1673 * ata_do_eh - do standard error handling
1674 * @ap: host port to handle error for
1675 * @prereset: prereset method (can be NULL)
1676 * @softreset: softreset method (can be NULL)
1677 * @hardreset: hardreset method (can be NULL)
1678 * @postreset: postreset method (can be NULL)
1680 * Perform standard error handling sequence.
1683 * Kernel thread context (may sleep).
1685 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
1686 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
1687 ata_postreset_fn_t postreset)
1691 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);