4 * Basic PIO and command management functionality.
6 * This code was split off from ide.c. See ide.c for history and original
9 * This program is free software; you can redistribute it and/or modify it
10 * under the terms of the GNU General Public License as published by the
11 * Free Software Foundation; either version 2, or (at your option) any
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
19 * For the avoidance of doubt the "preferred form" of this code is one which
20 * is in an open non patent encumbered format. Where cryptographic key signing
21 * forms part of the process of creating an executable the information
22 * including keys needed to generate an equivalently functional executable
23 * are deemed to be part of the source code.
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/hdreg.h>
44 #include <linux/completion.h>
45 #include <linux/reboot.h>
46 #include <linux/cdrom.h>
47 #include <linux/seq_file.h>
48 #include <linux/device.h>
49 #include <linux/kmod.h>
50 #include <linux/scatterlist.h>
51 #include <linux/bitops.h>
53 #include <asm/byteorder.h>
55 #include <asm/uaccess.h>
58 static int __ide_end_request(ide_drive_t
*drive
, struct request
*rq
,
59 int uptodate
, unsigned int nr_bytes
, int dequeue
)
65 error
= uptodate
? uptodate
: -EIO
;
68 * if failfast is set on a request, override number of sectors and
69 * complete the whole request right now
71 if (blk_noretry_request(rq
) && error
)
72 nr_bytes
= rq
->hard_nr_sectors
<< 9;
74 if (!blk_fs_request(rq
) && error
&& !rq
->errors
)
78 * decide whether to reenable DMA -- 3 is a random magic for now,
79 * if we DMA timeout more than 3 times, just stay in PIO
81 if ((drive
->dev_flags
& IDE_DFLAG_DMA_PIO_RETRY
) &&
82 drive
->retry_pio
<= 3) {
83 drive
->dev_flags
&= ~IDE_DFLAG_DMA_PIO_RETRY
;
87 if (!blk_end_request(rq
, error
, nr_bytes
))
90 if (ret
== 0 && dequeue
)
91 drive
->hwif
->hwgroup
->rq
= NULL
;
97 * ide_end_request - complete an IDE I/O
98 * @drive: IDE device for the I/O
100 * @nr_sectors: number of sectors completed
102 * This is our end_request wrapper function. We complete the I/O
103 * update random number input and dequeue the request, which if
104 * it was tagged may be out of order.
107 int ide_end_request (ide_drive_t
*drive
, int uptodate
, int nr_sectors
)
109 unsigned int nr_bytes
= nr_sectors
<< 9;
110 struct request
*rq
= drive
->hwif
->hwgroup
->rq
;
113 if (blk_pc_request(rq
))
114 nr_bytes
= rq
->data_len
;
116 nr_bytes
= rq
->hard_cur_sectors
<< 9;
119 return __ide_end_request(drive
, rq
, uptodate
, nr_bytes
, 1);
121 EXPORT_SYMBOL(ide_end_request
);
124 * ide_end_dequeued_request - complete an IDE I/O
125 * @drive: IDE device for the I/O
127 * @nr_sectors: number of sectors completed
129 * Complete an I/O that is no longer on the request queue. This
130 * typically occurs when we pull the request and issue a REQUEST_SENSE.
131 * We must still finish the old request but we must not tamper with the
132 * queue in the meantime.
134 * NOTE: This path does not handle barrier, but barrier is not supported
138 int ide_end_dequeued_request(ide_drive_t
*drive
, struct request
*rq
,
139 int uptodate
, int nr_sectors
)
141 BUG_ON(!blk_rq_started(rq
));
143 return __ide_end_request(drive
, rq
, uptodate
, nr_sectors
<< 9, 0);
145 EXPORT_SYMBOL_GPL(ide_end_dequeued_request
);
148 * ide_end_drive_cmd - end an explicit drive command
153 * Clean up after success/failure of an explicit drive command.
154 * These get thrown onto the queue so they are synchronized with
155 * real I/O operations on the drive.
157 * In LBA48 mode we have to read the register set twice to get
158 * all the extra information out.
161 void ide_end_drive_cmd (ide_drive_t
*drive
, u8 stat
, u8 err
)
163 ide_hwgroup_t
*hwgroup
= drive
->hwif
->hwgroup
;
164 struct request
*rq
= hwgroup
->rq
;
166 if (rq
->cmd_type
== REQ_TYPE_ATA_TASKFILE
) {
167 ide_task_t
*task
= (ide_task_t
*)rq
->special
;
170 struct ide_taskfile
*tf
= &task
->tf
;
175 drive
->hwif
->tp_ops
->tf_read(drive
, task
);
177 if (task
->tf_flags
& IDE_TFLAG_DYN
)
180 } else if (blk_pm_request(rq
)) {
181 struct request_pm_state
*pm
= rq
->data
;
183 ide_complete_power_step(drive
, rq
);
184 if (pm
->pm_step
== IDE_PM_COMPLETED
)
185 ide_complete_pm_request(drive
, rq
);
193 if (unlikely(blk_end_request(rq
, (rq
->errors
? -EIO
: 0),
197 EXPORT_SYMBOL(ide_end_drive_cmd
);
199 static void ide_kill_rq(ide_drive_t
*drive
, struct request
*rq
)
204 drv
= *(ide_driver_t
**)rq
->rq_disk
->private_data
;
205 drv
->end_request(drive
, 0, 0);
207 ide_end_request(drive
, 0, 0);
210 static ide_startstop_t
ide_ata_error(ide_drive_t
*drive
, struct request
*rq
, u8 stat
, u8 err
)
212 ide_hwif_t
*hwif
= drive
->hwif
;
214 if ((stat
& ATA_BUSY
) ||
215 ((stat
& ATA_DF
) && (drive
->dev_flags
& IDE_DFLAG_NOWERR
) == 0)) {
216 /* other bits are useless when BUSY */
217 rq
->errors
|= ERROR_RESET
;
218 } else if (stat
& ATA_ERR
) {
219 /* err has different meaning on cdrom and tape */
220 if (err
== ATA_ABORTED
) {
221 if ((drive
->dev_flags
& IDE_DFLAG_LBA
) &&
222 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
223 hwif
->tp_ops
->read_status(hwif
) == ATA_CMD_INIT_DEV_PARAMS
)
225 } else if ((err
& BAD_CRC
) == BAD_CRC
) {
226 /* UDMA crc error, just retry the operation */
228 } else if (err
& (ATA_BBK
| ATA_UNC
)) {
229 /* retries won't help these */
230 rq
->errors
= ERROR_MAX
;
231 } else if (err
& ATA_TRK0NF
) {
232 /* help it find track zero */
233 rq
->errors
|= ERROR_RECAL
;
237 if ((stat
& ATA_DRQ
) && rq_data_dir(rq
) == READ
&&
238 (hwif
->host_flags
& IDE_HFLAG_ERROR_STOPS_FIFO
) == 0) {
239 int nsect
= drive
->mult_count
? drive
->mult_count
: 1;
241 ide_pad_transfer(drive
, READ
, nsect
* SECTOR_SIZE
);
244 if (rq
->errors
>= ERROR_MAX
|| blk_noretry_request(rq
)) {
245 ide_kill_rq(drive
, rq
);
249 if (hwif
->tp_ops
->read_status(hwif
) & (ATA_BUSY
| ATA_DRQ
))
250 rq
->errors
|= ERROR_RESET
;
252 if ((rq
->errors
& ERROR_RESET
) == ERROR_RESET
) {
254 return ide_do_reset(drive
);
257 if ((rq
->errors
& ERROR_RECAL
) == ERROR_RECAL
)
258 drive
->special
.b
.recalibrate
= 1;
265 static ide_startstop_t
ide_atapi_error(ide_drive_t
*drive
, struct request
*rq
, u8 stat
, u8 err
)
267 ide_hwif_t
*hwif
= drive
->hwif
;
269 if ((stat
& ATA_BUSY
) ||
270 ((stat
& ATA_DF
) && (drive
->dev_flags
& IDE_DFLAG_NOWERR
) == 0)) {
271 /* other bits are useless when BUSY */
272 rq
->errors
|= ERROR_RESET
;
274 /* add decoding error stuff */
277 if (hwif
->tp_ops
->read_status(hwif
) & (ATA_BUSY
| ATA_DRQ
))
279 hwif
->tp_ops
->exec_command(hwif
, ATA_CMD_IDLEIMMEDIATE
);
281 if (rq
->errors
>= ERROR_MAX
) {
282 ide_kill_rq(drive
, rq
);
284 if ((rq
->errors
& ERROR_RESET
) == ERROR_RESET
) {
286 return ide_do_reset(drive
);
295 __ide_error(ide_drive_t
*drive
, struct request
*rq
, u8 stat
, u8 err
)
297 if (drive
->media
== ide_disk
)
298 return ide_ata_error(drive
, rq
, stat
, err
);
299 return ide_atapi_error(drive
, rq
, stat
, err
);
302 EXPORT_SYMBOL_GPL(__ide_error
);
305 * ide_error - handle an error on the IDE
306 * @drive: drive the error occurred on
307 * @msg: message to report
310 * ide_error() takes action based on the error returned by the drive.
311 * For normal I/O that may well include retries. We deal with
312 * both new-style (taskfile) and old style command handling here.
313 * In the case of taskfile command handling there is work left to
317 ide_startstop_t
ide_error (ide_drive_t
*drive
, const char *msg
, u8 stat
)
322 err
= ide_dump_status(drive
, msg
, stat
);
324 if ((rq
= HWGROUP(drive
)->rq
) == NULL
)
327 /* retry only "normal" I/O: */
328 if (!blk_fs_request(rq
)) {
330 ide_end_drive_cmd(drive
, stat
, err
);
337 drv
= *(ide_driver_t
**)rq
->rq_disk
->private_data
;
338 return drv
->error(drive
, rq
, stat
, err
);
340 return __ide_error(drive
, rq
, stat
, err
);
343 EXPORT_SYMBOL_GPL(ide_error
);
345 static void ide_tf_set_specify_cmd(ide_drive_t
*drive
, struct ide_taskfile
*tf
)
347 tf
->nsect
= drive
->sect
;
348 tf
->lbal
= drive
->sect
;
349 tf
->lbam
= drive
->cyl
;
350 tf
->lbah
= drive
->cyl
>> 8;
351 tf
->device
= (drive
->head
- 1) | drive
->select
;
352 tf
->command
= ATA_CMD_INIT_DEV_PARAMS
;
355 static void ide_tf_set_restore_cmd(ide_drive_t
*drive
, struct ide_taskfile
*tf
)
357 tf
->nsect
= drive
->sect
;
358 tf
->command
= ATA_CMD_RESTORE
;
361 static void ide_tf_set_setmult_cmd(ide_drive_t
*drive
, struct ide_taskfile
*tf
)
363 tf
->nsect
= drive
->mult_req
;
364 tf
->command
= ATA_CMD_SET_MULTI
;
367 static ide_startstop_t
ide_disk_special(ide_drive_t
*drive
)
369 special_t
*s
= &drive
->special
;
372 memset(&args
, 0, sizeof(ide_task_t
));
373 args
.data_phase
= TASKFILE_NO_DATA
;
375 if (s
->b
.set_geometry
) {
376 s
->b
.set_geometry
= 0;
377 ide_tf_set_specify_cmd(drive
, &args
.tf
);
378 } else if (s
->b
.recalibrate
) {
379 s
->b
.recalibrate
= 0;
380 ide_tf_set_restore_cmd(drive
, &args
.tf
);
381 } else if (s
->b
.set_multmode
) {
382 s
->b
.set_multmode
= 0;
383 ide_tf_set_setmult_cmd(drive
, &args
.tf
);
385 int special
= s
->all
;
387 printk(KERN_ERR
"%s: bad special flag: 0x%02x\n", drive
->name
, special
);
391 args
.tf_flags
= IDE_TFLAG_TF
| IDE_TFLAG_DEVICE
|
392 IDE_TFLAG_CUSTOM_HANDLER
;
394 do_rw_taskfile(drive
, &args
);
400 * do_special - issue some special commands
401 * @drive: drive the command is for
403 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
404 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
406 * It used to do much more, but has been scaled back.
409 static ide_startstop_t
do_special (ide_drive_t
*drive
)
411 special_t
*s
= &drive
->special
;
414 printk("%s: do_special: 0x%02x\n", drive
->name
, s
->all
);
416 if (drive
->media
== ide_disk
)
417 return ide_disk_special(drive
);
424 void ide_map_sg(ide_drive_t
*drive
, struct request
*rq
)
426 ide_hwif_t
*hwif
= drive
->hwif
;
427 struct scatterlist
*sg
= hwif
->sg_table
;
429 if (hwif
->sg_mapped
) /* needed by ide-scsi */
432 if (rq
->cmd_type
!= REQ_TYPE_ATA_TASKFILE
) {
433 hwif
->sg_nents
= blk_rq_map_sg(drive
->queue
, rq
, sg
);
435 sg_init_one(sg
, rq
->buffer
, rq
->nr_sectors
* SECTOR_SIZE
);
440 EXPORT_SYMBOL_GPL(ide_map_sg
);
442 void ide_init_sg_cmd(ide_drive_t
*drive
, struct request
*rq
)
444 ide_hwif_t
*hwif
= drive
->hwif
;
446 hwif
->nsect
= hwif
->nleft
= rq
->nr_sectors
;
451 EXPORT_SYMBOL_GPL(ide_init_sg_cmd
);
454 * execute_drive_command - issue special drive command
455 * @drive: the drive to issue the command on
456 * @rq: the request structure holding the command
458 * execute_drive_cmd() issues a special drive command, usually
459 * initiated by ioctl() from the external hdparm program. The
460 * command can be a drive command, drive task or taskfile
461 * operation. Weirdly you can call it with NULL to wait for
462 * all commands to finish. Don't do this as that is due to change
465 static ide_startstop_t
execute_drive_cmd (ide_drive_t
*drive
,
468 ide_hwif_t
*hwif
= HWIF(drive
);
469 ide_task_t
*task
= rq
->special
;
472 hwif
->data_phase
= task
->data_phase
;
474 switch (hwif
->data_phase
) {
475 case TASKFILE_MULTI_OUT
:
477 case TASKFILE_MULTI_IN
:
479 ide_init_sg_cmd(drive
, rq
);
480 ide_map_sg(drive
, rq
);
485 return do_rw_taskfile(drive
, task
);
489 * NULL is actually a valid way of waiting for
490 * all current requests to be flushed from the queue.
493 printk("%s: DRIVE_CMD (null)\n", drive
->name
);
495 ide_end_drive_cmd(drive
, hwif
->tp_ops
->read_status(hwif
),
496 ide_read_error(drive
));
501 int ide_devset_execute(ide_drive_t
*drive
, const struct ide_devset
*setting
,
504 struct request_queue
*q
= drive
->queue
;
508 if (!(setting
->flags
& DS_SYNC
))
509 return setting
->set(drive
, arg
);
511 rq
= blk_get_request(q
, READ
, __GFP_WAIT
);
512 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
514 rq
->cmd
[0] = REQ_DEVSET_EXEC
;
515 *(int *)&rq
->cmd
[1] = arg
;
516 rq
->special
= setting
->set
;
518 if (blk_execute_rq(q
, NULL
, rq
, 0))
524 EXPORT_SYMBOL_GPL(ide_devset_execute
);
526 static ide_startstop_t
ide_special_rq(ide_drive_t
*drive
, struct request
*rq
)
530 if (cmd
== REQ_PARK_HEADS
|| cmd
== REQ_UNPARK_HEADS
) {
532 struct ide_taskfile
*tf
= &task
.tf
;
534 memset(&task
, 0, sizeof(task
));
535 if (cmd
== REQ_PARK_HEADS
) {
536 drive
->sleep
= *(unsigned long *)rq
->special
;
537 drive
->dev_flags
|= IDE_DFLAG_SLEEPING
;
538 tf
->command
= ATA_CMD_IDLEIMMEDIATE
;
543 task
.tf_flags
|= IDE_TFLAG_CUSTOM_HANDLER
;
544 } else /* cmd == REQ_UNPARK_HEADS */
545 tf
->command
= ATA_CMD_CHK_POWER
;
547 task
.tf_flags
|= IDE_TFLAG_TF
| IDE_TFLAG_DEVICE
;
549 drive
->hwif
->data_phase
= task
.data_phase
= TASKFILE_NO_DATA
;
550 return do_rw_taskfile(drive
, &task
);
554 case REQ_DEVSET_EXEC
:
556 int err
, (*setfunc
)(ide_drive_t
*, int) = rq
->special
;
558 err
= setfunc(drive
, *(int *)&rq
->cmd
[1]);
563 ide_end_request(drive
, err
, 0);
566 case REQ_DRIVE_RESET
:
567 return ide_do_reset(drive
);
569 blk_dump_rq_flags(rq
, "ide_special_rq - bad request");
570 ide_end_request(drive
, 0, 0);
576 * start_request - start of I/O and command issuing for IDE
578 * start_request() initiates handling of a new I/O request. It
579 * accepts commands and I/O (read/write) requests.
581 * FIXME: this function needs a rename
584 static ide_startstop_t
start_request (ide_drive_t
*drive
, struct request
*rq
)
586 ide_startstop_t startstop
;
588 BUG_ON(!blk_rq_started(rq
));
591 printk("%s: start_request: current=0x%08lx\n",
592 HWIF(drive
)->name
, (unsigned long) rq
);
595 /* bail early if we've exceeded max_failures */
596 if (drive
->max_failures
&& (drive
->failures
> drive
->max_failures
)) {
597 rq
->cmd_flags
|= REQ_FAILED
;
601 if (blk_pm_request(rq
))
602 ide_check_pm_state(drive
, rq
);
605 if (ide_wait_stat(&startstop
, drive
, drive
->ready_stat
,
606 ATA_BUSY
| ATA_DRQ
, WAIT_READY
)) {
607 printk(KERN_ERR
"%s: drive not ready for command\n", drive
->name
);
610 if (!drive
->special
.all
) {
614 * We reset the drive so we need to issue a SETFEATURES.
615 * Do it _after_ do_special() restored device parameters.
617 if (drive
->current_speed
== 0xff)
618 ide_config_drive_speed(drive
, drive
->desired_speed
);
620 if (rq
->cmd_type
== REQ_TYPE_ATA_TASKFILE
)
621 return execute_drive_cmd(drive
, rq
);
622 else if (blk_pm_request(rq
)) {
623 struct request_pm_state
*pm
= rq
->data
;
625 printk("%s: start_power_step(step: %d)\n",
626 drive
->name
, pm
->pm_step
);
628 startstop
= ide_start_power_step(drive
, rq
);
629 if (startstop
== ide_stopped
&&
630 pm
->pm_step
== IDE_PM_COMPLETED
)
631 ide_complete_pm_request(drive
, rq
);
633 } else if (!rq
->rq_disk
&& blk_special_request(rq
))
635 * TODO: Once all ULDs have been modified to
636 * check for specific op codes rather than
637 * blindly accepting any special request, the
638 * check for ->rq_disk above may be replaced
639 * by a more suitable mechanism or even
642 return ide_special_rq(drive
, rq
);
644 drv
= *(ide_driver_t
**)rq
->rq_disk
->private_data
;
646 return drv
->do_request(drive
, rq
, rq
->sector
);
648 return do_special(drive
);
650 ide_kill_rq(drive
, rq
);
655 * ide_stall_queue - pause an IDE device
656 * @drive: drive to stall
657 * @timeout: time to stall for (jiffies)
659 * ide_stall_queue() can be used by a drive to give excess bandwidth back
660 * to the hwgroup by sleeping for timeout jiffies.
663 void ide_stall_queue (ide_drive_t
*drive
, unsigned long timeout
)
665 if (timeout
> WAIT_WORSTCASE
)
666 timeout
= WAIT_WORSTCASE
;
667 drive
->sleep
= timeout
+ jiffies
;
668 drive
->dev_flags
|= IDE_DFLAG_SLEEPING
;
671 EXPORT_SYMBOL(ide_stall_queue
);
673 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
676 * choose_drive - select a drive to service
677 * @hwgroup: hardware group to select on
679 * choose_drive() selects the next drive which will be serviced.
680 * This is necessary because the IDE layer can't issue commands
681 * to both drives on the same cable, unlike SCSI.
684 static inline ide_drive_t
*choose_drive (ide_hwgroup_t
*hwgroup
)
686 ide_drive_t
*drive
, *best
;
690 drive
= hwgroup
->drive
;
693 * drive is doing pre-flush, ordered write, post-flush sequence. even
694 * though that is 3 requests, it must be seen as a single transaction.
695 * we must not preempt this drive until that is complete
697 if (blk_queue_flushing(drive
->queue
)) {
699 * small race where queue could get replugged during
700 * the 3-request flush cycle, just yank the plug since
701 * we want it to finish asap
703 blk_remove_plug(drive
->queue
);
708 u8 dev_s
= !!(drive
->dev_flags
& IDE_DFLAG_SLEEPING
);
709 u8 best_s
= (best
&& !!(best
->dev_flags
& IDE_DFLAG_SLEEPING
));
711 if ((dev_s
== 0 || time_after_eq(jiffies
, drive
->sleep
)) &&
712 !elv_queue_empty(drive
->queue
)) {
714 (dev_s
&& (best_s
== 0 || time_before(drive
->sleep
, best
->sleep
))) ||
715 (best_s
== 0 && time_before(WAKEUP(drive
), WAKEUP(best
)))) {
716 if (!blk_queue_plugged(drive
->queue
))
720 } while ((drive
= drive
->next
) != hwgroup
->drive
);
722 if (best
&& (best
->dev_flags
& IDE_DFLAG_NICE1
) &&
723 (best
->dev_flags
& IDE_DFLAG_SLEEPING
) == 0 &&
724 best
!= hwgroup
->drive
&& best
->service_time
> WAIT_MIN_SLEEP
) {
725 long t
= (signed long)(WAKEUP(best
) - jiffies
);
726 if (t
>= WAIT_MIN_SLEEP
) {
728 * We *may* have some time to spare, but first let's see if
729 * someone can potentially benefit from our nice mood today..
733 if ((drive
->dev_flags
& IDE_DFLAG_SLEEPING
) == 0
734 && time_before(jiffies
- best
->service_time
, WAKEUP(drive
))
735 && time_before(WAKEUP(drive
), jiffies
+ t
))
737 ide_stall_queue(best
, min_t(long, t
, 10 * WAIT_MIN_SLEEP
));
740 } while ((drive
= drive
->next
) != best
);
747 * Issue a new request to a drive from hwgroup
748 * Caller must have already done spin_lock_irqsave(&hwgroup->lock, ..);
750 * A hwgroup is a serialized group of IDE interfaces. Usually there is
751 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
752 * may have both interfaces in a single hwgroup to "serialize" access.
753 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
754 * together into one hwgroup for serialized access.
756 * Note also that several hwgroups can end up sharing a single IRQ,
757 * possibly along with many other devices. This is especially common in
758 * PCI-based systems with off-board IDE controller cards.
760 * The IDE driver uses a per-hwgroup spinlock to protect
761 * access to the request queues, and to protect the hwgroup->busy flag.
763 * The first thread into the driver for a particular hwgroup sets the
764 * hwgroup->busy flag to indicate that this hwgroup is now active,
765 * and then initiates processing of the top request from the request queue.
767 * Other threads attempting entry notice the busy setting, and will simply
768 * queue their new requests and exit immediately. Note that hwgroup->busy
769 * remains set even when the driver is merely awaiting the next interrupt.
770 * Thus, the meaning is "this hwgroup is busy processing a request".
772 * When processing of a request completes, the completing thread or IRQ-handler
773 * will start the next request from the queue. If no more work remains,
774 * the driver will clear the hwgroup->busy flag and exit.
776 * The per-hwgroup spinlock is used to protect all access to the
777 * hwgroup->busy flag, but is otherwise not needed for most processing in
778 * the driver. This makes the driver much more friendlier to shared IRQs
779 * than previous designs, while remaining 100% (?) SMP safe and capable.
781 void do_ide_request(struct request_queue
*q
)
783 ide_drive_t
*orig_drive
= q
->queuedata
;
784 ide_hwgroup_t
*hwgroup
= orig_drive
->hwif
->hwgroup
;
788 ide_startstop_t startstop
;
790 /* caller must own hwgroup->lock */
791 BUG_ON(!irqs_disabled());
793 while (!ide_lock_hwgroup(hwgroup
)) {
794 drive
= choose_drive(hwgroup
);
797 unsigned long sleep
= 0; /* shut up, gcc */
799 drive
= hwgroup
->drive
;
801 if ((drive
->dev_flags
& IDE_DFLAG_SLEEPING
) &&
803 time_before(drive
->sleep
, sleep
))) {
805 sleep
= drive
->sleep
;
807 } while ((drive
= drive
->next
) != hwgroup
->drive
);
810 * Take a short snooze, and then wake up this hwgroup again.
811 * This gives other hwgroups on the same a chance to
812 * play fairly with us, just in case there are big differences
813 * in relative throughputs.. don't want to hog the cpu too much.
815 if (time_before(sleep
, jiffies
+ WAIT_MIN_SLEEP
))
816 sleep
= jiffies
+ WAIT_MIN_SLEEP
;
818 if (timer_pending(&hwgroup
->timer
))
819 printk(KERN_CRIT
"ide_set_handler: timer already active\n");
821 /* so that ide_timer_expiry knows what to do */
822 hwgroup
->sleeping
= 1;
823 hwgroup
->req_gen_timer
= hwgroup
->req_gen
;
824 mod_timer(&hwgroup
->timer
, sleep
);
825 /* we purposely leave hwgroup locked
828 ide_unlock_hwgroup(hwgroup
);
830 /* no more work for this hwgroup (for now) */
834 if (drive
!= orig_drive
)
839 if (hwif
!= hwgroup
->hwif
) {
841 * set nIEN for previous hwif, drives in the
842 * quirk_list may not like intr setups/cleanups
844 if (drive
->quirk_list
== 0)
845 hwif
->tp_ops
->set_irq(hwif
, 0);
847 hwgroup
->hwif
= hwif
;
848 hwgroup
->drive
= drive
;
849 drive
->dev_flags
&= ~(IDE_DFLAG_SLEEPING
| IDE_DFLAG_PARKED
);
850 drive
->service_start
= jiffies
;
853 * we know that the queue isn't empty, but this can happen
854 * if the q->prep_rq_fn() decides to kill a request
856 rq
= elv_next_request(drive
->queue
);
858 ide_unlock_hwgroup(hwgroup
);
863 * Sanity: don't accept a request that isn't a PM request
864 * if we are currently power managed. This is very important as
865 * blk_stop_queue() doesn't prevent the elv_next_request()
866 * above to return us whatever is in the queue. Since we call
867 * ide_do_request() ourselves, we end up taking requests while
868 * the queue is blocked...
870 * We let requests forced at head of queue with ide-preempt
871 * though. I hope that doesn't happen too much, hopefully not
872 * unless the subdriver triggers such a thing in its own PM
875 if ((drive
->dev_flags
& IDE_DFLAG_BLOCKED
) &&
876 blk_pm_request(rq
) == 0 &&
877 (rq
->cmd_flags
& REQ_PREEMPT
) == 0) {
878 /* there should be no pending command at this point */
879 ide_unlock_hwgroup(hwgroup
);
885 spin_unlock_irq(&hwgroup
->lock
);
886 startstop
= start_request(drive
, rq
);
887 spin_lock_irq(&hwgroup
->lock
);
889 if (startstop
== ide_stopped
) {
890 ide_unlock_hwgroup(hwgroup
);
891 if (!elv_queue_empty(orig_drive
->queue
))
892 blk_plug_device(orig_drive
->queue
);
898 if (!elv_queue_empty(orig_drive
->queue
))
899 blk_plug_device(orig_drive
->queue
);
903 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
904 * retry the current request in pio mode instead of risking tossing it
907 static ide_startstop_t
ide_dma_timeout_retry(ide_drive_t
*drive
, int error
)
909 ide_hwif_t
*hwif
= HWIF(drive
);
911 ide_startstop_t ret
= ide_stopped
;
914 * end current dma transaction
918 printk(KERN_WARNING
"%s: DMA timeout error\n", drive
->name
);
919 (void)hwif
->dma_ops
->dma_end(drive
);
920 ret
= ide_error(drive
, "dma timeout error",
921 hwif
->tp_ops
->read_status(hwif
));
923 printk(KERN_WARNING
"%s: DMA timeout retry\n", drive
->name
);
924 hwif
->dma_ops
->dma_timeout(drive
);
928 * disable dma for now, but remember that we did so because of
929 * a timeout -- we'll reenable after we finish this next request
930 * (or rather the first chunk of it) in pio.
932 drive
->dev_flags
|= IDE_DFLAG_DMA_PIO_RETRY
;
934 ide_dma_off_quietly(drive
);
937 * un-busy drive etc (hwgroup->busy is cleared on return) and
938 * make sure request is sane
940 rq
= HWGROUP(drive
)->rq
;
945 HWGROUP(drive
)->rq
= NULL
;
952 rq
->sector
= rq
->bio
->bi_sector
;
953 rq
->current_nr_sectors
= bio_iovec(rq
->bio
)->bv_len
>> 9;
954 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
955 rq
->buffer
= bio_data(rq
->bio
);
961 * ide_timer_expiry - handle lack of an IDE interrupt
962 * @data: timer callback magic (hwgroup)
964 * An IDE command has timed out before the expected drive return
965 * occurred. At this point we attempt to clean up the current
966 * mess. If the current handler includes an expiry handler then
967 * we invoke the expiry handler, and providing it is happy the
968 * work is done. If that fails we apply generic recovery rules
969 * invoking the handler and checking the drive DMA status. We
970 * have an excessively incestuous relationship with the DMA
971 * logic that wants cleaning up.
974 void ide_timer_expiry (unsigned long data
)
976 ide_hwgroup_t
*hwgroup
= (ide_hwgroup_t
*) data
;
977 ide_handler_t
*handler
;
978 ide_expiry_t
*expiry
;
980 unsigned long wait
= -1;
982 spin_lock_irqsave(&hwgroup
->lock
, flags
);
984 if (((handler
= hwgroup
->handler
) == NULL
) ||
985 (hwgroup
->req_gen
!= hwgroup
->req_gen_timer
)) {
987 * Either a marginal timeout occurred
988 * (got the interrupt just as timer expired),
989 * or we were "sleeping" to give other devices a chance.
990 * Either way, we don't really want to complain about anything.
992 if (hwgroup
->sleeping
) {
993 hwgroup
->sleeping
= 0;
994 ide_unlock_hwgroup(hwgroup
);
997 ide_drive_t
*drive
= hwgroup
->drive
;
999 printk(KERN_ERR
"ide_timer_expiry: hwgroup->drive was NULL\n");
1000 hwgroup
->handler
= NULL
;
1003 ide_startstop_t startstop
= ide_stopped
;
1005 if ((expiry
= hwgroup
->expiry
) != NULL
) {
1007 if ((wait
= expiry(drive
)) > 0) {
1009 hwgroup
->timer
.expires
= jiffies
+ wait
;
1010 hwgroup
->req_gen_timer
= hwgroup
->req_gen
;
1011 add_timer(&hwgroup
->timer
);
1012 spin_unlock_irqrestore(&hwgroup
->lock
, flags
);
1016 hwgroup
->handler
= NULL
;
1018 * We need to simulate a real interrupt when invoking
1019 * the handler() function, which means we need to
1020 * globally mask the specific IRQ:
1022 spin_unlock(&hwgroup
->lock
);
1024 /* disable_irq_nosync ?? */
1025 disable_irq(hwif
->irq
);
1027 * as if we were handling an interrupt */
1028 local_irq_disable();
1029 if (hwgroup
->polling
) {
1030 startstop
= handler(drive
);
1031 } else if (drive_is_ready(drive
)) {
1032 if (drive
->waiting_for_dma
)
1033 hwif
->dma_ops
->dma_lost_irq(drive
);
1034 (void)ide_ack_intr(hwif
);
1035 printk(KERN_WARNING
"%s: lost interrupt\n", drive
->name
);
1036 startstop
= handler(drive
);
1038 if (drive
->waiting_for_dma
) {
1039 startstop
= ide_dma_timeout_retry(drive
, wait
);
1042 ide_error(drive
, "irq timeout",
1043 hwif
->tp_ops
->read_status(hwif
));
1045 drive
->service_time
= jiffies
- drive
->service_start
;
1046 spin_lock_irq(&hwgroup
->lock
);
1047 enable_irq(hwif
->irq
);
1048 if (startstop
== ide_stopped
) {
1049 ide_unlock_hwgroup(hwgroup
);
1050 if (!elv_queue_empty(drive
->queue
))
1051 blk_plug_device(drive
->queue
);
1055 spin_unlock_irqrestore(&hwgroup
->lock
, flags
);
1059 * unexpected_intr - handle an unexpected IDE interrupt
1060 * @irq: interrupt line
1061 * @hwgroup: hwgroup being processed
1063 * There's nothing really useful we can do with an unexpected interrupt,
1064 * other than reading the status register (to clear it), and logging it.
1065 * There should be no way that an irq can happen before we're ready for it,
1066 * so we needn't worry much about losing an "important" interrupt here.
1068 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1069 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1070 * looks "good", we just ignore the interrupt completely.
1072 * This routine assumes __cli() is in effect when called.
1074 * If an unexpected interrupt happens on irq15 while we are handling irq14
1075 * and if the two interfaces are "serialized" (CMD640), then it looks like
1076 * we could screw up by interfering with a new request being set up for
1079 * In reality, this is a non-issue. The new command is not sent unless
1080 * the drive is ready to accept one, in which case we know the drive is
1081 * not trying to interrupt us. And ide_set_handler() is always invoked
1082 * before completing the issuance of any new drive command, so we will not
1083 * be accidentally invoked as a result of any valid command completion
1086 * Note that we must walk the entire hwgroup here. We know which hwif
1087 * is doing the current command, but we don't know which hwif burped
1091 static void unexpected_intr (int irq
, ide_hwgroup_t
*hwgroup
)
1094 ide_hwif_t
*hwif
= hwgroup
->hwif
;
1097 * handle the unexpected interrupt
1100 if (hwif
->irq
== irq
) {
1101 stat
= hwif
->tp_ops
->read_status(hwif
);
1103 if (!OK_STAT(stat
, ATA_DRDY
, BAD_STAT
)) {
1104 /* Try to not flood the console with msgs */
1105 static unsigned long last_msgtime
, count
;
1107 if (time_after(jiffies
, last_msgtime
+ HZ
)) {
1108 last_msgtime
= jiffies
;
1109 printk(KERN_ERR
"%s%s: unexpected interrupt, "
1110 "status=0x%02x, count=%ld\n",
1112 (hwif
->next
==hwgroup
->hwif
) ? "" : "(?)", stat
, count
);
1116 } while ((hwif
= hwif
->next
) != hwgroup
->hwif
);
1120 * ide_intr - default IDE interrupt handler
1121 * @irq: interrupt number
1122 * @dev_id: hwif group
1123 * @regs: unused weirdness from the kernel irq layer
1125 * This is the default IRQ handler for the IDE layer. You should
1126 * not need to override it. If you do be aware it is subtle in
1129 * hwgroup->hwif is the interface in the group currently performing
1130 * a command. hwgroup->drive is the drive and hwgroup->handler is
1131 * the IRQ handler to call. As we issue a command the handlers
1132 * step through multiple states, reassigning the handler to the
1133 * next step in the process. Unlike a smart SCSI controller IDE
1134 * expects the main processor to sequence the various transfer
1135 * stages. We also manage a poll timer to catch up with most
1136 * timeout situations. There are still a few where the handlers
1137 * don't ever decide to give up.
1139 * The handler eventually returns ide_stopped to indicate the
1140 * request completed. At this point we issue the next request
1141 * on the hwgroup and the process begins again.
1144 irqreturn_t
ide_intr (int irq
, void *dev_id
)
1146 unsigned long flags
;
1147 ide_hwgroup_t
*hwgroup
= (ide_hwgroup_t
*)dev_id
;
1148 ide_hwif_t
*hwif
= hwgroup
->hwif
;
1150 ide_handler_t
*handler
;
1151 ide_startstop_t startstop
;
1152 irqreturn_t irq_ret
= IRQ_NONE
;
1154 spin_lock_irqsave(&hwgroup
->lock
, flags
);
1156 if (!ide_ack_intr(hwif
))
1159 if ((handler
= hwgroup
->handler
) == NULL
|| hwgroup
->polling
) {
1161 * Not expecting an interrupt from this drive.
1162 * That means this could be:
1163 * (1) an interrupt from another PCI device
1164 * sharing the same PCI INT# as us.
1165 * or (2) a drive just entered sleep or standby mode,
1166 * and is interrupting to let us know.
1167 * or (3) a spurious interrupt of unknown origin.
1169 * For PCI, we cannot tell the difference,
1170 * so in that case we just ignore it and hope it goes away.
1172 * FIXME: unexpected_intr should be hwif-> then we can
1173 * remove all the ifdef PCI crap
1175 #ifdef CONFIG_BLK_DEV_IDEPCI
1176 if (hwif
->chipset
!= ide_pci
)
1177 #endif /* CONFIG_BLK_DEV_IDEPCI */
1180 * Probably not a shared PCI interrupt,
1181 * so we can safely try to do something about it:
1183 unexpected_intr(irq
, hwgroup
);
1184 #ifdef CONFIG_BLK_DEV_IDEPCI
1187 * Whack the status register, just in case
1188 * we have a leftover pending IRQ.
1190 (void)hwif
->tp_ops
->read_status(hwif
);
1191 #endif /* CONFIG_BLK_DEV_IDEPCI */
1196 drive
= hwgroup
->drive
;
1199 * This should NEVER happen, and there isn't much
1200 * we could do about it here.
1202 * [Note - this can occur if the drive is hot unplugged]
1207 if (!drive_is_ready(drive
))
1209 * This happens regularly when we share a PCI IRQ with
1210 * another device. Unfortunately, it can also happen
1211 * with some buggy drives that trigger the IRQ before
1212 * their status register is up to date. Hopefully we have
1213 * enough advance overhead that the latter isn't a problem.
1217 hwgroup
->handler
= NULL
;
1219 del_timer(&hwgroup
->timer
);
1220 spin_unlock(&hwgroup
->lock
);
1222 if (hwif
->port_ops
&& hwif
->port_ops
->clear_irq
)
1223 hwif
->port_ops
->clear_irq(drive
);
1225 if (drive
->dev_flags
& IDE_DFLAG_UNMASK
)
1226 local_irq_enable_in_hardirq();
1228 /* service this interrupt, may set handler for next interrupt */
1229 startstop
= handler(drive
);
1231 spin_lock_irq(&hwgroup
->lock
);
1233 * Note that handler() may have set things up for another
1234 * interrupt to occur soon, but it cannot happen until
1235 * we exit from this routine, because it will be the
1236 * same irq as is currently being serviced here, and Linux
1237 * won't allow another of the same (on any CPU) until we return.
1239 drive
->service_time
= jiffies
- drive
->service_start
;
1240 if (startstop
== ide_stopped
) {
1241 if (hwgroup
->handler
== NULL
) { /* paranoia */
1242 ide_unlock_hwgroup(hwgroup
);
1243 if (!elv_queue_empty(drive
->queue
))
1244 blk_plug_device(drive
->queue
);
1246 printk(KERN_ERR
"%s: %s: huh? expected NULL handler "
1247 "on exit\n", __func__
, drive
->name
);
1250 irq_ret
= IRQ_HANDLED
;
1252 spin_unlock_irqrestore(&hwgroup
->lock
, flags
);
1257 * ide_do_drive_cmd - issue IDE special command
1258 * @drive: device to issue command
1259 * @rq: request to issue
1261 * This function issues a special IDE device request
1262 * onto the request queue.
1264 * the rq is queued at the head of the request queue, displacing
1265 * the currently-being-processed request and this function
1266 * returns immediately without waiting for the new rq to be
1267 * completed. This is VERY DANGEROUS, and is intended for
1268 * careful use by the ATAPI tape/cdrom driver code.
1271 void ide_do_drive_cmd(ide_drive_t
*drive
, struct request
*rq
)
1273 ide_hwgroup_t
*hwgroup
= drive
->hwif
->hwgroup
;
1274 struct request_queue
*q
= drive
->queue
;
1275 unsigned long flags
;
1279 spin_lock_irqsave(q
->queue_lock
, flags
);
1280 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FRONT
, 0);
1281 blk_start_queueing(q
);
1282 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1284 EXPORT_SYMBOL(ide_do_drive_cmd
);
1286 void ide_pktcmd_tf_load(ide_drive_t
*drive
, u32 tf_flags
, u16 bcount
, u8 dma
)
1288 ide_hwif_t
*hwif
= drive
->hwif
;
1291 memset(&task
, 0, sizeof(task
));
1292 task
.tf_flags
= IDE_TFLAG_OUT_LBAH
| IDE_TFLAG_OUT_LBAM
|
1293 IDE_TFLAG_OUT_FEATURE
| tf_flags
;
1294 task
.tf
.feature
= dma
; /* Use PIO/DMA */
1295 task
.tf
.lbam
= bcount
& 0xff;
1296 task
.tf
.lbah
= (bcount
>> 8) & 0xff;
1298 ide_tf_dump(drive
->name
, &task
.tf
);
1299 hwif
->tp_ops
->set_irq(hwif
, 1);
1300 SELECT_MASK(drive
, 0);
1301 hwif
->tp_ops
->tf_load(drive
, &task
);
1304 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load
);
1306 void ide_pad_transfer(ide_drive_t
*drive
, int write
, int len
)
1308 ide_hwif_t
*hwif
= drive
->hwif
;
1313 hwif
->tp_ops
->output_data(drive
, NULL
, buf
, min(4, len
));
1315 hwif
->tp_ops
->input_data(drive
, NULL
, buf
, min(4, len
));
1319 EXPORT_SYMBOL_GPL(ide_pad_transfer
);