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)) {
89 HWGROUP(drive)->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;
115 * room for locking improvements here, the calls below don't
116 * need the queue lock held at all
118 spin_lock_irqsave(&ide_lock, flags);
119 rq = HWGROUP(drive)->rq;
122 if (blk_pc_request(rq))
123 nr_bytes = rq->data_len;
125 nr_bytes = rq->hard_cur_sectors << 9;
128 ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
130 spin_unlock_irqrestore(&ide_lock, flags);
133 EXPORT_SYMBOL(ide_end_request);
135 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq)
137 struct request_pm_state *pm = rq->data;
140 printk(KERN_INFO "%s: complete_power_step(step: %d)\n",
141 drive->name, pm->pm_step);
143 if (drive->media != ide_disk)
146 switch (pm->pm_step) {
147 case IDE_PM_FLUSH_CACHE: /* Suspend step 1 (flush cache) */
148 if (pm->pm_state == PM_EVENT_FREEZE)
149 pm->pm_step = IDE_PM_COMPLETED;
151 pm->pm_step = IDE_PM_STANDBY;
153 case IDE_PM_STANDBY: /* Suspend step 2 (standby) */
154 pm->pm_step = IDE_PM_COMPLETED;
156 case IDE_PM_RESTORE_PIO: /* Resume step 1 (restore PIO) */
157 pm->pm_step = IDE_PM_IDLE;
159 case IDE_PM_IDLE: /* Resume step 2 (idle)*/
160 pm->pm_step = IDE_PM_RESTORE_DMA;
165 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
167 struct request_pm_state *pm = rq->data;
168 ide_task_t *args = rq->special;
170 memset(args, 0, sizeof(*args));
172 switch (pm->pm_step) {
173 case IDE_PM_FLUSH_CACHE: /* Suspend step 1 (flush cache) */
174 if (drive->media != ide_disk)
176 /* Not supported? Switch to next step now. */
177 if (ata_id_flush_enabled(drive->id) == 0 ||
178 (drive->dev_flags & IDE_DFLAG_WCACHE) == 0) {
179 ide_complete_power_step(drive, rq);
182 if (ata_id_flush_ext_enabled(drive->id))
183 args->tf.command = ATA_CMD_FLUSH_EXT;
185 args->tf.command = ATA_CMD_FLUSH;
187 case IDE_PM_STANDBY: /* Suspend step 2 (standby) */
188 args->tf.command = ATA_CMD_STANDBYNOW1;
190 case IDE_PM_RESTORE_PIO: /* Resume step 1 (restore PIO) */
191 ide_set_max_pio(drive);
193 * skip IDE_PM_IDLE for ATAPI devices
195 if (drive->media != ide_disk)
196 pm->pm_step = IDE_PM_RESTORE_DMA;
198 ide_complete_power_step(drive, rq);
200 case IDE_PM_IDLE: /* Resume step 2 (idle) */
201 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
203 case IDE_PM_RESTORE_DMA: /* Resume step 3 (restore DMA) */
205 * Right now, all we do is call ide_set_dma(drive),
206 * we could be smarter and check for current xfer_speed
207 * in struct drive etc...
209 if (drive->hwif->dma_ops == NULL)
211 if (drive->dev_flags & IDE_DFLAG_USING_DMA)
216 pm->pm_step = IDE_PM_COMPLETED;
220 args->tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
221 args->data_phase = TASKFILE_NO_DATA;
222 return do_rw_taskfile(drive, args);
226 * ide_end_dequeued_request - complete an IDE I/O
227 * @drive: IDE device for the I/O
229 * @nr_sectors: number of sectors completed
231 * Complete an I/O that is no longer on the request queue. This
232 * typically occurs when we pull the request and issue a REQUEST_SENSE.
233 * We must still finish the old request but we must not tamper with the
234 * queue in the meantime.
236 * NOTE: This path does not handle barrier, but barrier is not supported
240 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
241 int uptodate, int nr_sectors)
246 spin_lock_irqsave(&ide_lock, flags);
247 BUG_ON(!blk_rq_started(rq));
248 ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
249 spin_unlock_irqrestore(&ide_lock, flags);
253 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
257 * ide_complete_pm_request - end the current Power Management request
258 * @drive: target drive
261 * This function cleans up the current PM request and stops the queue
264 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
269 printk("%s: completing PM request, %s\n", drive->name,
270 blk_pm_suspend_request(rq) ? "suspend" : "resume");
272 spin_lock_irqsave(&ide_lock, flags);
273 if (blk_pm_suspend_request(rq)) {
274 blk_stop_queue(drive->queue);
276 drive->dev_flags &= ~IDE_DFLAG_BLOCKED;
277 blk_start_queue(drive->queue);
279 HWGROUP(drive)->rq = NULL;
280 if (__blk_end_request(rq, 0, 0))
282 spin_unlock_irqrestore(&ide_lock, flags);
286 * ide_end_drive_cmd - end an explicit drive command
291 * Clean up after success/failure of an explicit drive command.
292 * These get thrown onto the queue so they are synchronized with
293 * real I/O operations on the drive.
295 * In LBA48 mode we have to read the register set twice to get
296 * all the extra information out.
299 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
304 spin_lock_irqsave(&ide_lock, flags);
305 rq = HWGROUP(drive)->rq;
306 spin_unlock_irqrestore(&ide_lock, flags);
308 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
309 ide_task_t *task = (ide_task_t *)rq->special;
312 rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
315 struct ide_taskfile *tf = &task->tf;
320 drive->hwif->tp_ops->tf_read(drive, task);
322 if (task->tf_flags & IDE_TFLAG_DYN)
325 } else if (blk_pm_request(rq)) {
326 struct request_pm_state *pm = rq->data;
328 ide_complete_power_step(drive, rq);
329 if (pm->pm_step == IDE_PM_COMPLETED)
330 ide_complete_pm_request(drive, rq);
334 spin_lock_irqsave(&ide_lock, flags);
335 HWGROUP(drive)->rq = NULL;
337 if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
340 spin_unlock_irqrestore(&ide_lock, flags);
343 EXPORT_SYMBOL(ide_end_drive_cmd);
345 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
350 drv = *(ide_driver_t **)rq->rq_disk->private_data;
351 drv->end_request(drive, 0, 0);
353 ide_end_request(drive, 0, 0);
356 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
358 ide_hwif_t *hwif = drive->hwif;
360 if ((stat & ATA_BUSY) ||
361 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
362 /* other bits are useless when BUSY */
363 rq->errors |= ERROR_RESET;
364 } else if (stat & ATA_ERR) {
365 /* err has different meaning on cdrom and tape */
366 if (err == ATA_ABORTED) {
367 if ((drive->dev_flags & IDE_DFLAG_LBA) &&
368 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
369 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
371 } else if ((err & BAD_CRC) == BAD_CRC) {
372 /* UDMA crc error, just retry the operation */
374 } else if (err & (ATA_BBK | ATA_UNC)) {
375 /* retries won't help these */
376 rq->errors = ERROR_MAX;
377 } else if (err & ATA_TRK0NF) {
378 /* help it find track zero */
379 rq->errors |= ERROR_RECAL;
383 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
384 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
385 int nsect = drive->mult_count ? drive->mult_count : 1;
387 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
390 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
391 ide_kill_rq(drive, rq);
395 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
396 rq->errors |= ERROR_RESET;
398 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
400 return ide_do_reset(drive);
403 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
404 drive->special.b.recalibrate = 1;
411 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
413 ide_hwif_t *hwif = drive->hwif;
415 if ((stat & ATA_BUSY) ||
416 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
417 /* other bits are useless when BUSY */
418 rq->errors |= ERROR_RESET;
420 /* add decoding error stuff */
423 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
425 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
427 if (rq->errors >= ERROR_MAX) {
428 ide_kill_rq(drive, rq);
430 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
432 return ide_do_reset(drive);
441 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
443 if (drive->media == ide_disk)
444 return ide_ata_error(drive, rq, stat, err);
445 return ide_atapi_error(drive, rq, stat, err);
448 EXPORT_SYMBOL_GPL(__ide_error);
451 * ide_error - handle an error on the IDE
452 * @drive: drive the error occurred on
453 * @msg: message to report
456 * ide_error() takes action based on the error returned by the drive.
457 * For normal I/O that may well include retries. We deal with
458 * both new-style (taskfile) and old style command handling here.
459 * In the case of taskfile command handling there is work left to
463 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
468 err = ide_dump_status(drive, msg, stat);
470 if ((rq = HWGROUP(drive)->rq) == NULL)
473 /* retry only "normal" I/O: */
474 if (!blk_fs_request(rq)) {
476 ide_end_drive_cmd(drive, stat, err);
483 drv = *(ide_driver_t **)rq->rq_disk->private_data;
484 return drv->error(drive, rq, stat, err);
486 return __ide_error(drive, rq, stat, err);
489 EXPORT_SYMBOL_GPL(ide_error);
491 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
493 tf->nsect = drive->sect;
494 tf->lbal = drive->sect;
495 tf->lbam = drive->cyl;
496 tf->lbah = drive->cyl >> 8;
497 tf->device = (drive->head - 1) | drive->select;
498 tf->command = ATA_CMD_INIT_DEV_PARAMS;
501 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
503 tf->nsect = drive->sect;
504 tf->command = ATA_CMD_RESTORE;
507 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
509 tf->nsect = drive->mult_req;
510 tf->command = ATA_CMD_SET_MULTI;
513 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
515 special_t *s = &drive->special;
518 memset(&args, 0, sizeof(ide_task_t));
519 args.data_phase = TASKFILE_NO_DATA;
521 if (s->b.set_geometry) {
522 s->b.set_geometry = 0;
523 ide_tf_set_specify_cmd(drive, &args.tf);
524 } else if (s->b.recalibrate) {
525 s->b.recalibrate = 0;
526 ide_tf_set_restore_cmd(drive, &args.tf);
527 } else if (s->b.set_multmode) {
528 s->b.set_multmode = 0;
529 ide_tf_set_setmult_cmd(drive, &args.tf);
531 int special = s->all;
533 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
537 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
538 IDE_TFLAG_CUSTOM_HANDLER;
540 do_rw_taskfile(drive, &args);
546 * do_special - issue some special commands
547 * @drive: drive the command is for
549 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
550 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
552 * It used to do much more, but has been scaled back.
555 static ide_startstop_t do_special (ide_drive_t *drive)
557 special_t *s = &drive->special;
560 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
562 if (drive->media == ide_disk)
563 return ide_disk_special(drive);
570 void ide_map_sg(ide_drive_t *drive, struct request *rq)
572 ide_hwif_t *hwif = drive->hwif;
573 struct scatterlist *sg = hwif->sg_table;
575 if (hwif->sg_mapped) /* needed by ide-scsi */
578 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
579 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
581 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
586 EXPORT_SYMBOL_GPL(ide_map_sg);
588 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
590 ide_hwif_t *hwif = drive->hwif;
592 hwif->nsect = hwif->nleft = rq->nr_sectors;
597 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
600 * execute_drive_command - issue special drive command
601 * @drive: the drive to issue the command on
602 * @rq: the request structure holding the command
604 * execute_drive_cmd() issues a special drive command, usually
605 * initiated by ioctl() from the external hdparm program. The
606 * command can be a drive command, drive task or taskfile
607 * operation. Weirdly you can call it with NULL to wait for
608 * all commands to finish. Don't do this as that is due to change
611 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
614 ide_hwif_t *hwif = HWIF(drive);
615 ide_task_t *task = rq->special;
618 hwif->data_phase = task->data_phase;
620 switch (hwif->data_phase) {
621 case TASKFILE_MULTI_OUT:
623 case TASKFILE_MULTI_IN:
625 ide_init_sg_cmd(drive, rq);
626 ide_map_sg(drive, rq);
631 return do_rw_taskfile(drive, task);
635 * NULL is actually a valid way of waiting for
636 * all current requests to be flushed from the queue.
639 printk("%s: DRIVE_CMD (null)\n", drive->name);
641 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
642 ide_read_error(drive));
647 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
650 struct request_queue *q = drive->queue;
654 if (!(setting->flags & DS_SYNC))
655 return setting->set(drive, arg);
657 rq = blk_get_request(q, READ, __GFP_WAIT);
658 rq->cmd_type = REQ_TYPE_SPECIAL;
660 rq->cmd[0] = REQ_DEVSET_EXEC;
661 *(int *)&rq->cmd[1] = arg;
662 rq->special = setting->set;
664 if (blk_execute_rq(q, NULL, rq, 0))
670 EXPORT_SYMBOL_GPL(ide_devset_execute);
672 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
676 if (cmd == REQ_PARK_HEADS || cmd == REQ_UNPARK_HEADS) {
678 struct ide_taskfile *tf = &task.tf;
680 memset(&task, 0, sizeof(task));
681 if (cmd == REQ_PARK_HEADS) {
682 drive->sleep = *(unsigned long *)rq->special;
683 drive->dev_flags |= IDE_DFLAG_SLEEPING;
684 tf->command = ATA_CMD_IDLEIMMEDIATE;
689 task.tf_flags |= IDE_TFLAG_CUSTOM_HANDLER;
690 } else /* cmd == REQ_UNPARK_HEADS */
691 tf->command = ATA_CMD_CHK_POWER;
693 task.tf_flags |= IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
695 drive->hwif->data_phase = task.data_phase = TASKFILE_NO_DATA;
696 return do_rw_taskfile(drive, &task);
700 case REQ_DEVSET_EXEC:
702 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
704 err = setfunc(drive, *(int *)&rq->cmd[1]);
709 ide_end_request(drive, err, 0);
712 case REQ_DRIVE_RESET:
713 return ide_do_reset(drive);
715 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
716 ide_end_request(drive, 0, 0);
721 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
723 struct request_pm_state *pm = rq->data;
725 if (blk_pm_suspend_request(rq) &&
726 pm->pm_step == IDE_PM_START_SUSPEND)
727 /* Mark drive blocked when starting the suspend sequence. */
728 drive->dev_flags |= IDE_DFLAG_BLOCKED;
729 else if (blk_pm_resume_request(rq) &&
730 pm->pm_step == IDE_PM_START_RESUME) {
732 * The first thing we do on wakeup is to wait for BSY bit to
733 * go away (with a looong timeout) as a drive on this hwif may
734 * just be POSTing itself.
735 * We do that before even selecting as the "other" device on
736 * the bus may be broken enough to walk on our toes at this
739 ide_hwif_t *hwif = drive->hwif;
742 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
744 rc = ide_wait_not_busy(hwif, 35000);
746 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
748 hwif->tp_ops->set_irq(hwif, 1);
749 rc = ide_wait_not_busy(hwif, 100000);
751 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
756 * start_request - start of I/O and command issuing for IDE
758 * start_request() initiates handling of a new I/O request. It
759 * accepts commands and I/O (read/write) requests.
761 * FIXME: this function needs a rename
764 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
766 ide_startstop_t startstop;
768 BUG_ON(!blk_rq_started(rq));
771 printk("%s: start_request: current=0x%08lx\n",
772 HWIF(drive)->name, (unsigned long) rq);
775 /* bail early if we've exceeded max_failures */
776 if (drive->max_failures && (drive->failures > drive->max_failures)) {
777 rq->cmd_flags |= REQ_FAILED;
781 if (blk_pm_request(rq))
782 ide_check_pm_state(drive, rq);
785 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
786 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
787 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
790 if (!drive->special.all) {
794 * We reset the drive so we need to issue a SETFEATURES.
795 * Do it _after_ do_special() restored device parameters.
797 if (drive->current_speed == 0xff)
798 ide_config_drive_speed(drive, drive->desired_speed);
800 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
801 return execute_drive_cmd(drive, rq);
802 else if (blk_pm_request(rq)) {
803 struct request_pm_state *pm = rq->data;
805 printk("%s: start_power_step(step: %d)\n",
806 drive->name, pm->pm_step);
808 startstop = ide_start_power_step(drive, rq);
809 if (startstop == ide_stopped &&
810 pm->pm_step == IDE_PM_COMPLETED)
811 ide_complete_pm_request(drive, rq);
813 } else if (!rq->rq_disk && blk_special_request(rq))
815 * TODO: Once all ULDs have been modified to
816 * check for specific op codes rather than
817 * blindly accepting any special request, the
818 * check for ->rq_disk above may be replaced
819 * by a more suitable mechanism or even
822 return ide_special_rq(drive, rq);
824 drv = *(ide_driver_t **)rq->rq_disk->private_data;
826 return drv->do_request(drive, rq, rq->sector);
828 return do_special(drive);
830 ide_kill_rq(drive, rq);
835 * ide_stall_queue - pause an IDE device
836 * @drive: drive to stall
837 * @timeout: time to stall for (jiffies)
839 * ide_stall_queue() can be used by a drive to give excess bandwidth back
840 * to the hwgroup by sleeping for timeout jiffies.
843 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
845 if (timeout > WAIT_WORSTCASE)
846 timeout = WAIT_WORSTCASE;
847 drive->sleep = timeout + jiffies;
848 drive->dev_flags |= IDE_DFLAG_SLEEPING;
851 EXPORT_SYMBOL(ide_stall_queue);
853 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
856 * choose_drive - select a drive to service
857 * @hwgroup: hardware group to select on
859 * choose_drive() selects the next drive which will be serviced.
860 * This is necessary because the IDE layer can't issue commands
861 * to both drives on the same cable, unlike SCSI.
864 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
866 ide_drive_t *drive, *best;
870 drive = hwgroup->drive;
873 * drive is doing pre-flush, ordered write, post-flush sequence. even
874 * though that is 3 requests, it must be seen as a single transaction.
875 * we must not preempt this drive until that is complete
877 if (blk_queue_flushing(drive->queue)) {
879 * small race where queue could get replugged during
880 * the 3-request flush cycle, just yank the plug since
881 * we want it to finish asap
883 blk_remove_plug(drive->queue);
888 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
889 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
891 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
892 !elv_queue_empty(drive->queue)) {
894 (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
895 (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
896 if (!blk_queue_plugged(drive->queue))
900 } while ((drive = drive->next) != hwgroup->drive);
902 if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
903 (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
904 best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
905 long t = (signed long)(WAKEUP(best) - jiffies);
906 if (t >= WAIT_MIN_SLEEP) {
908 * We *may* have some time to spare, but first let's see if
909 * someone can potentially benefit from our nice mood today..
913 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
914 && time_before(jiffies - best->service_time, WAKEUP(drive))
915 && time_before(WAKEUP(drive), jiffies + t))
917 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
920 } while ((drive = drive->next) != best);
927 * Issue a new request to a drive from hwgroup
928 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
930 * A hwgroup is a serialized group of IDE interfaces. Usually there is
931 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
932 * may have both interfaces in a single hwgroup to "serialize" access.
933 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
934 * together into one hwgroup for serialized access.
936 * Note also that several hwgroups can end up sharing a single IRQ,
937 * possibly along with many other devices. This is especially common in
938 * PCI-based systems with off-board IDE controller cards.
940 * The IDE driver uses the single global ide_lock spinlock to protect
941 * access to the request queues, and to protect the hwgroup->busy flag.
943 * The first thread into the driver for a particular hwgroup sets the
944 * hwgroup->busy flag to indicate that this hwgroup is now active,
945 * and then initiates processing of the top request from the request queue.
947 * Other threads attempting entry notice the busy setting, and will simply
948 * queue their new requests and exit immediately. Note that hwgroup->busy
949 * remains set even when the driver is merely awaiting the next interrupt.
950 * Thus, the meaning is "this hwgroup is busy processing a request".
952 * When processing of a request completes, the completing thread or IRQ-handler
953 * will start the next request from the queue. If no more work remains,
954 * the driver will clear the hwgroup->busy flag and exit.
956 * The ide_lock (spinlock) is used to protect all access to the
957 * hwgroup->busy flag, but is otherwise not needed for most processing in
958 * the driver. This makes the driver much more friendlier to shared IRQs
959 * than previous designs, while remaining 100% (?) SMP safe and capable.
961 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
966 ide_startstop_t startstop;
969 /* caller must own ide_lock */
970 BUG_ON(!irqs_disabled());
972 while (!hwgroup->busy) {
975 ide_get_lock(ide_intr, hwgroup);
976 drive = choose_drive(hwgroup);
979 unsigned long sleep = 0; /* shut up, gcc */
981 drive = hwgroup->drive;
983 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
985 time_before(drive->sleep, sleep))) {
987 sleep = drive->sleep;
989 } while ((drive = drive->next) != hwgroup->drive);
992 * Take a short snooze, and then wake up this hwgroup again.
993 * This gives other hwgroups on the same a chance to
994 * play fairly with us, just in case there are big differences
995 * in relative throughputs.. don't want to hog the cpu too much.
997 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
998 sleep = jiffies + WAIT_MIN_SLEEP;
1000 if (timer_pending(&hwgroup->timer))
1001 printk(KERN_CRIT "ide_set_handler: timer already active\n");
1003 /* so that ide_timer_expiry knows what to do */
1004 hwgroup->sleeping = 1;
1005 hwgroup->req_gen_timer = hwgroup->req_gen;
1006 mod_timer(&hwgroup->timer, sleep);
1007 /* we purposely leave hwgroup->busy==1
1010 /* Ugly, but how can we sleep for the lock
1011 * otherwise? perhaps from tq_disk?
1014 /* for atari only */
1019 /* no more work for this hwgroup (for now) */
1024 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1026 * set nIEN for previous hwif, drives in the
1027 * quirk_list may not like intr setups/cleanups
1029 if (drive->quirk_list != 1)
1030 hwif->tp_ops->set_irq(hwif, 0);
1032 hwgroup->hwif = hwif;
1033 hwgroup->drive = drive;
1034 drive->dev_flags &= ~(IDE_DFLAG_SLEEPING | IDE_DFLAG_PARKED);
1035 drive->service_start = jiffies;
1037 if (blk_queue_plugged(drive->queue)) {
1038 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1043 * we know that the queue isn't empty, but this can happen
1044 * if the q->prep_rq_fn() decides to kill a request
1046 rq = elv_next_request(drive->queue);
1053 * Sanity: don't accept a request that isn't a PM request
1054 * if we are currently power managed. This is very important as
1055 * blk_stop_queue() doesn't prevent the elv_next_request()
1056 * above to return us whatever is in the queue. Since we call
1057 * ide_do_request() ourselves, we end up taking requests while
1058 * the queue is blocked...
1060 * We let requests forced at head of queue with ide-preempt
1061 * though. I hope that doesn't happen too much, hopefully not
1062 * unless the subdriver triggers such a thing in its own PM
1065 * We count how many times we loop here to make sure we service
1066 * all drives in the hwgroup without looping for ever
1068 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
1069 blk_pm_request(rq) == 0 &&
1070 (rq->cmd_flags & REQ_PREEMPT) == 0) {
1071 drive = drive->next ? drive->next : hwgroup->drive;
1072 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1074 /* We clear busy, there should be no pending ATA command at this point. */
1082 * Some systems have trouble with IDE IRQs arriving while
1083 * the driver is still setting things up. So, here we disable
1084 * the IRQ used by this interface while the request is being started.
1085 * This may look bad at first, but pretty much the same thing
1086 * happens anyway when any interrupt comes in, IDE or otherwise
1087 * -- the kernel masks the IRQ while it is being handled.
1089 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1090 disable_irq_nosync(hwif->irq);
1091 spin_unlock(&ide_lock);
1092 local_irq_enable_in_hardirq();
1093 /* allow other IRQs while we start this request */
1094 startstop = start_request(drive, rq);
1095 spin_lock_irq(&ide_lock);
1096 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1097 enable_irq(hwif->irq);
1098 if (startstop == ide_stopped)
1104 * Passes the stuff to ide_do_request
1106 void do_ide_request(struct request_queue *q)
1108 ide_drive_t *drive = q->queuedata;
1110 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1114 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1115 * retry the current request in pio mode instead of risking tossing it
1118 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1120 ide_hwif_t *hwif = HWIF(drive);
1122 ide_startstop_t ret = ide_stopped;
1125 * end current dma transaction
1129 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1130 (void)hwif->dma_ops->dma_end(drive);
1131 ret = ide_error(drive, "dma timeout error",
1132 hwif->tp_ops->read_status(hwif));
1134 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1135 hwif->dma_ops->dma_timeout(drive);
1139 * disable dma for now, but remember that we did so because of
1140 * a timeout -- we'll reenable after we finish this next request
1141 * (or rather the first chunk of it) in pio.
1143 drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
1145 ide_dma_off_quietly(drive);
1148 * un-busy drive etc (hwgroup->busy is cleared on return) and
1149 * make sure request is sane
1151 rq = HWGROUP(drive)->rq;
1156 HWGROUP(drive)->rq = NULL;
1163 rq->sector = rq->bio->bi_sector;
1164 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1165 rq->hard_cur_sectors = rq->current_nr_sectors;
1166 rq->buffer = bio_data(rq->bio);
1172 * ide_timer_expiry - handle lack of an IDE interrupt
1173 * @data: timer callback magic (hwgroup)
1175 * An IDE command has timed out before the expected drive return
1176 * occurred. At this point we attempt to clean up the current
1177 * mess. If the current handler includes an expiry handler then
1178 * we invoke the expiry handler, and providing it is happy the
1179 * work is done. If that fails we apply generic recovery rules
1180 * invoking the handler and checking the drive DMA status. We
1181 * have an excessively incestuous relationship with the DMA
1182 * logic that wants cleaning up.
1185 void ide_timer_expiry (unsigned long data)
1187 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1188 ide_handler_t *handler;
1189 ide_expiry_t *expiry;
1190 unsigned long flags;
1191 unsigned long wait = -1;
1193 spin_lock_irqsave(&ide_lock, flags);
1195 if (((handler = hwgroup->handler) == NULL) ||
1196 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1198 * Either a marginal timeout occurred
1199 * (got the interrupt just as timer expired),
1200 * or we were "sleeping" to give other devices a chance.
1201 * Either way, we don't really want to complain about anything.
1203 if (hwgroup->sleeping) {
1204 hwgroup->sleeping = 0;
1208 ide_drive_t *drive = hwgroup->drive;
1210 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1211 hwgroup->handler = NULL;
1214 ide_startstop_t startstop = ide_stopped;
1215 if (!hwgroup->busy) {
1216 hwgroup->busy = 1; /* paranoia */
1217 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1219 if ((expiry = hwgroup->expiry) != NULL) {
1221 if ((wait = expiry(drive)) > 0) {
1223 hwgroup->timer.expires = jiffies + wait;
1224 hwgroup->req_gen_timer = hwgroup->req_gen;
1225 add_timer(&hwgroup->timer);
1226 spin_unlock_irqrestore(&ide_lock, flags);
1230 hwgroup->handler = NULL;
1232 * We need to simulate a real interrupt when invoking
1233 * the handler() function, which means we need to
1234 * globally mask the specific IRQ:
1236 spin_unlock(&ide_lock);
1238 /* disable_irq_nosync ?? */
1239 disable_irq(hwif->irq);
1241 * as if we were handling an interrupt */
1242 local_irq_disable();
1243 if (hwgroup->polling) {
1244 startstop = handler(drive);
1245 } else if (drive_is_ready(drive)) {
1246 if (drive->waiting_for_dma)
1247 hwif->dma_ops->dma_lost_irq(drive);
1248 (void)ide_ack_intr(hwif);
1249 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1250 startstop = handler(drive);
1252 if (drive->waiting_for_dma) {
1253 startstop = ide_dma_timeout_retry(drive, wait);
1256 ide_error(drive, "irq timeout",
1257 hwif->tp_ops->read_status(hwif));
1259 drive->service_time = jiffies - drive->service_start;
1260 spin_lock_irq(&ide_lock);
1261 enable_irq(hwif->irq);
1262 if (startstop == ide_stopped)
1266 ide_do_request(hwgroup, IDE_NO_IRQ);
1267 spin_unlock_irqrestore(&ide_lock, flags);
1271 * unexpected_intr - handle an unexpected IDE interrupt
1272 * @irq: interrupt line
1273 * @hwgroup: hwgroup being processed
1275 * There's nothing really useful we can do with an unexpected interrupt,
1276 * other than reading the status register (to clear it), and logging it.
1277 * There should be no way that an irq can happen before we're ready for it,
1278 * so we needn't worry much about losing an "important" interrupt here.
1280 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1281 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1282 * looks "good", we just ignore the interrupt completely.
1284 * This routine assumes __cli() is in effect when called.
1286 * If an unexpected interrupt happens on irq15 while we are handling irq14
1287 * and if the two interfaces are "serialized" (CMD640), then it looks like
1288 * we could screw up by interfering with a new request being set up for
1291 * In reality, this is a non-issue. The new command is not sent unless
1292 * the drive is ready to accept one, in which case we know the drive is
1293 * not trying to interrupt us. And ide_set_handler() is always invoked
1294 * before completing the issuance of any new drive command, so we will not
1295 * be accidentally invoked as a result of any valid command completion
1298 * Note that we must walk the entire hwgroup here. We know which hwif
1299 * is doing the current command, but we don't know which hwif burped
1303 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1306 ide_hwif_t *hwif = hwgroup->hwif;
1309 * handle the unexpected interrupt
1312 if (hwif->irq == irq) {
1313 stat = hwif->tp_ops->read_status(hwif);
1315 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1316 /* Try to not flood the console with msgs */
1317 static unsigned long last_msgtime, count;
1319 if (time_after(jiffies, last_msgtime + HZ)) {
1320 last_msgtime = jiffies;
1321 printk(KERN_ERR "%s%s: unexpected interrupt, "
1322 "status=0x%02x, count=%ld\n",
1324 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1328 } while ((hwif = hwif->next) != hwgroup->hwif);
1332 * ide_intr - default IDE interrupt handler
1333 * @irq: interrupt number
1334 * @dev_id: hwif group
1335 * @regs: unused weirdness from the kernel irq layer
1337 * This is the default IRQ handler for the IDE layer. You should
1338 * not need to override it. If you do be aware it is subtle in
1341 * hwgroup->hwif is the interface in the group currently performing
1342 * a command. hwgroup->drive is the drive and hwgroup->handler is
1343 * the IRQ handler to call. As we issue a command the handlers
1344 * step through multiple states, reassigning the handler to the
1345 * next step in the process. Unlike a smart SCSI controller IDE
1346 * expects the main processor to sequence the various transfer
1347 * stages. We also manage a poll timer to catch up with most
1348 * timeout situations. There are still a few where the handlers
1349 * don't ever decide to give up.
1351 * The handler eventually returns ide_stopped to indicate the
1352 * request completed. At this point we issue the next request
1353 * on the hwgroup and the process begins again.
1356 irqreturn_t ide_intr (int irq, void *dev_id)
1358 unsigned long flags;
1359 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1362 ide_handler_t *handler;
1363 ide_startstop_t startstop;
1365 spin_lock_irqsave(&ide_lock, flags);
1366 hwif = hwgroup->hwif;
1368 if (!ide_ack_intr(hwif)) {
1369 spin_unlock_irqrestore(&ide_lock, flags);
1373 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1375 * Not expecting an interrupt from this drive.
1376 * That means this could be:
1377 * (1) an interrupt from another PCI device
1378 * sharing the same PCI INT# as us.
1379 * or (2) a drive just entered sleep or standby mode,
1380 * and is interrupting to let us know.
1381 * or (3) a spurious interrupt of unknown origin.
1383 * For PCI, we cannot tell the difference,
1384 * so in that case we just ignore it and hope it goes away.
1386 * FIXME: unexpected_intr should be hwif-> then we can
1387 * remove all the ifdef PCI crap
1389 #ifdef CONFIG_BLK_DEV_IDEPCI
1390 if (hwif->chipset != ide_pci)
1391 #endif /* CONFIG_BLK_DEV_IDEPCI */
1394 * Probably not a shared PCI interrupt,
1395 * so we can safely try to do something about it:
1397 unexpected_intr(irq, hwgroup);
1398 #ifdef CONFIG_BLK_DEV_IDEPCI
1401 * Whack the status register, just in case
1402 * we have a leftover pending IRQ.
1404 (void)hwif->tp_ops->read_status(hwif);
1405 #endif /* CONFIG_BLK_DEV_IDEPCI */
1407 spin_unlock_irqrestore(&ide_lock, flags);
1410 drive = hwgroup->drive;
1413 * This should NEVER happen, and there isn't much
1414 * we could do about it here.
1416 * [Note - this can occur if the drive is hot unplugged]
1418 spin_unlock_irqrestore(&ide_lock, flags);
1421 if (!drive_is_ready(drive)) {
1423 * This happens regularly when we share a PCI IRQ with
1424 * another device. Unfortunately, it can also happen
1425 * with some buggy drives that trigger the IRQ before
1426 * their status register is up to date. Hopefully we have
1427 * enough advance overhead that the latter isn't a problem.
1429 spin_unlock_irqrestore(&ide_lock, flags);
1432 if (!hwgroup->busy) {
1433 hwgroup->busy = 1; /* paranoia */
1434 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1436 hwgroup->handler = NULL;
1438 del_timer(&hwgroup->timer);
1439 spin_unlock(&ide_lock);
1441 if (hwif->port_ops && hwif->port_ops->clear_irq)
1442 hwif->port_ops->clear_irq(drive);
1444 if (drive->dev_flags & IDE_DFLAG_UNMASK)
1445 local_irq_enable_in_hardirq();
1447 /* service this interrupt, may set handler for next interrupt */
1448 startstop = handler(drive);
1450 spin_lock_irq(&ide_lock);
1452 * Note that handler() may have set things up for another
1453 * interrupt to occur soon, but it cannot happen until
1454 * we exit from this routine, because it will be the
1455 * same irq as is currently being serviced here, and Linux
1456 * won't allow another of the same (on any CPU) until we return.
1458 drive->service_time = jiffies - drive->service_start;
1459 if (startstop == ide_stopped) {
1460 if (hwgroup->handler == NULL) { /* paranoia */
1462 ide_do_request(hwgroup, hwif->irq);
1464 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1465 "on exit\n", drive->name);
1468 spin_unlock_irqrestore(&ide_lock, flags);
1473 * ide_do_drive_cmd - issue IDE special command
1474 * @drive: device to issue command
1475 * @rq: request to issue
1477 * This function issues a special IDE device request
1478 * onto the request queue.
1480 * the rq is queued at the head of the request queue, displacing
1481 * the currently-being-processed request and this function
1482 * returns immediately without waiting for the new rq to be
1483 * completed. This is VERY DANGEROUS, and is intended for
1484 * careful use by the ATAPI tape/cdrom driver code.
1487 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1489 unsigned long flags;
1490 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1492 spin_lock_irqsave(&ide_lock, flags);
1494 __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 0);
1495 blk_start_queueing(drive->queue);
1496 spin_unlock_irqrestore(&ide_lock, flags);
1499 EXPORT_SYMBOL(ide_do_drive_cmd);
1501 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1503 ide_hwif_t *hwif = drive->hwif;
1506 memset(&task, 0, sizeof(task));
1507 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1508 IDE_TFLAG_OUT_FEATURE | tf_flags;
1509 task.tf.feature = dma; /* Use PIO/DMA */
1510 task.tf.lbam = bcount & 0xff;
1511 task.tf.lbah = (bcount >> 8) & 0xff;
1513 ide_tf_dump(drive->name, &task.tf);
1514 hwif->tp_ops->set_irq(hwif, 1);
1515 SELECT_MASK(drive, 0);
1516 hwif->tp_ops->tf_load(drive, &task);
1519 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1521 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1523 ide_hwif_t *hwif = drive->hwif;
1528 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1530 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1534 EXPORT_SYMBOL_GPL(ide_pad_transfer);