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/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50 #include <linux/bitops.h>
52 #include <asm/byteorder.h>
54 #include <asm/uaccess.h>
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58 int uptodate, unsigned int nr_bytes, int dequeue)
64 error = uptodate ? uptodate : -EIO;
67 * if failfast is set on a request, override number of sectors and
68 * complete the whole request right now
70 if (blk_noretry_request(rq) && error)
71 nr_bytes = rq->hard_nr_sectors << 9;
73 if (!blk_fs_request(rq) && error && !rq->errors)
77 * decide whether to reenable DMA -- 3 is a random magic for now,
78 * if we DMA timeout more than 3 times, just stay in PIO
80 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
85 if (!__blk_end_request(rq, error, nr_bytes)) {
87 HWGROUP(drive)->rq = NULL;
95 * ide_end_request - complete an IDE I/O
96 * @drive: IDE device for the I/O
98 * @nr_sectors: number of sectors completed
100 * This is our end_request wrapper function. We complete the I/O
101 * update random number input and dequeue the request, which if
102 * it was tagged may be out of order.
105 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
107 unsigned int nr_bytes = nr_sectors << 9;
113 * room for locking improvements here, the calls below don't
114 * need the queue lock held at all
116 spin_lock_irqsave(&ide_lock, flags);
117 rq = HWGROUP(drive)->rq;
120 if (blk_pc_request(rq))
121 nr_bytes = rq->data_len;
123 nr_bytes = rq->hard_cur_sectors << 9;
126 ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
128 spin_unlock_irqrestore(&ide_lock, flags);
131 EXPORT_SYMBOL(ide_end_request);
134 * Power Management state machine. This one is rather trivial for now,
135 * we should probably add more, like switching back to PIO on suspend
136 * to help some BIOSes, re-do the door locking on resume, etc...
140 ide_pm_flush_cache = ide_pm_state_start_suspend,
143 idedisk_pm_restore_pio = ide_pm_state_start_resume,
148 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
150 struct request_pm_state *pm = rq->data;
152 if (drive->media != ide_disk)
155 switch (pm->pm_step) {
156 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) complete */
157 if (pm->pm_state == PM_EVENT_FREEZE)
158 pm->pm_step = ide_pm_state_completed;
160 pm->pm_step = idedisk_pm_standby;
162 case idedisk_pm_standby: /* Suspend step 2 (standby) complete */
163 pm->pm_step = ide_pm_state_completed;
165 case idedisk_pm_restore_pio: /* Resume step 1 complete */
166 pm->pm_step = idedisk_pm_idle;
168 case idedisk_pm_idle: /* Resume step 2 (idle) complete */
169 pm->pm_step = ide_pm_restore_dma;
174 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
176 struct request_pm_state *pm = rq->data;
177 ide_task_t *args = rq->special;
179 memset(args, 0, sizeof(*args));
181 switch (pm->pm_step) {
182 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) */
183 if (drive->media != ide_disk)
185 /* Not supported? Switch to next step now. */
186 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
187 ide_complete_power_step(drive, rq, 0, 0);
190 if (ide_id_has_flush_cache_ext(drive->id))
191 args->tf.command = ATA_CMD_FLUSH_EXT;
193 args->tf.command = ATA_CMD_FLUSH;
196 case idedisk_pm_standby: /* Suspend step 2 (standby) */
197 args->tf.command = ATA_CMD_STANDBYNOW1;
200 case idedisk_pm_restore_pio: /* Resume step 1 (restore PIO) */
201 ide_set_max_pio(drive);
203 * skip idedisk_pm_idle for ATAPI devices
205 if (drive->media != ide_disk)
206 pm->pm_step = ide_pm_restore_dma;
208 ide_complete_power_step(drive, rq, 0, 0);
211 case idedisk_pm_idle: /* Resume step 2 (idle) */
212 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
215 case ide_pm_restore_dma: /* Resume step 3 (restore DMA) */
217 * Right now, all we do is call ide_set_dma(drive),
218 * we could be smarter and check for current xfer_speed
219 * in struct drive etc...
221 if (drive->hwif->dma_ops == NULL)
224 * TODO: respect ->using_dma setting
229 pm->pm_step = ide_pm_state_completed;
233 args->tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
234 args->data_phase = TASKFILE_NO_DATA;
235 return do_rw_taskfile(drive, args);
239 * ide_end_dequeued_request - complete an IDE I/O
240 * @drive: IDE device for the I/O
242 * @nr_sectors: number of sectors completed
244 * Complete an I/O that is no longer on the request queue. This
245 * typically occurs when we pull the request and issue a REQUEST_SENSE.
246 * We must still finish the old request but we must not tamper with the
247 * queue in the meantime.
249 * NOTE: This path does not handle barrier, but barrier is not supported
253 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
254 int uptodate, int nr_sectors)
259 spin_lock_irqsave(&ide_lock, flags);
260 BUG_ON(!blk_rq_started(rq));
261 ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
262 spin_unlock_irqrestore(&ide_lock, flags);
266 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
270 * ide_complete_pm_request - end the current Power Management request
271 * @drive: target drive
274 * This function cleans up the current PM request and stops the queue
277 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
282 printk("%s: completing PM request, %s\n", drive->name,
283 blk_pm_suspend_request(rq) ? "suspend" : "resume");
285 spin_lock_irqsave(&ide_lock, flags);
286 if (blk_pm_suspend_request(rq)) {
287 blk_stop_queue(drive->queue);
290 blk_start_queue(drive->queue);
292 HWGROUP(drive)->rq = NULL;
293 if (__blk_end_request(rq, 0, 0))
295 spin_unlock_irqrestore(&ide_lock, flags);
299 * ide_end_drive_cmd - end an explicit drive command
304 * Clean up after success/failure of an explicit drive command.
305 * These get thrown onto the queue so they are synchronized with
306 * real I/O operations on the drive.
308 * In LBA48 mode we have to read the register set twice to get
309 * all the extra information out.
312 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
317 spin_lock_irqsave(&ide_lock, flags);
318 rq = HWGROUP(drive)->rq;
319 spin_unlock_irqrestore(&ide_lock, flags);
321 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
322 ide_task_t *task = (ide_task_t *)rq->special;
325 rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
328 struct ide_taskfile *tf = &task->tf;
333 drive->hwif->tp_ops->tf_read(drive, task);
335 if (task->tf_flags & IDE_TFLAG_DYN)
338 } else if (blk_pm_request(rq)) {
339 struct request_pm_state *pm = rq->data;
341 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
342 drive->name, rq->pm->pm_step, stat, err);
344 ide_complete_power_step(drive, rq, stat, err);
345 if (pm->pm_step == ide_pm_state_completed)
346 ide_complete_pm_request(drive, rq);
350 spin_lock_irqsave(&ide_lock, flags);
351 HWGROUP(drive)->rq = NULL;
353 if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
356 spin_unlock_irqrestore(&ide_lock, flags);
359 EXPORT_SYMBOL(ide_end_drive_cmd);
361 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
366 drv = *(ide_driver_t **)rq->rq_disk->private_data;
367 drv->end_request(drive, 0, 0);
369 ide_end_request(drive, 0, 0);
372 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
374 ide_hwif_t *hwif = drive->hwif;
376 if ((stat & ATA_BUSY) || ((stat & ATA_DF) && !drive->nowerr)) {
377 /* other bits are useless when BUSY */
378 rq->errors |= ERROR_RESET;
379 } else if (stat & ATA_ERR) {
380 /* err has different meaning on cdrom and tape */
381 if (err == ATA_ABORTED) {
382 if (drive->select.b.lba &&
383 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
384 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
386 } else if ((err & BAD_CRC) == BAD_CRC) {
387 /* UDMA crc error, just retry the operation */
389 } else if (err & (ATA_BBK | ATA_UNC)) {
390 /* retries won't help these */
391 rq->errors = ERROR_MAX;
392 } else if (err & ATA_TRK0NF) {
393 /* help it find track zero */
394 rq->errors |= ERROR_RECAL;
398 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
399 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
400 int nsect = drive->mult_count ? drive->mult_count : 1;
402 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
405 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
406 ide_kill_rq(drive, rq);
410 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
411 rq->errors |= ERROR_RESET;
413 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
415 return ide_do_reset(drive);
418 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
419 drive->special.b.recalibrate = 1;
426 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
428 ide_hwif_t *hwif = drive->hwif;
430 if ((stat & ATA_BUSY) || ((stat & ATA_DF) && !drive->nowerr)) {
431 /* other bits are useless when BUSY */
432 rq->errors |= ERROR_RESET;
434 /* add decoding error stuff */
437 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
439 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
441 if (rq->errors >= ERROR_MAX) {
442 ide_kill_rq(drive, rq);
444 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
446 return ide_do_reset(drive);
455 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
457 if (drive->media == ide_disk)
458 return ide_ata_error(drive, rq, stat, err);
459 return ide_atapi_error(drive, rq, stat, err);
462 EXPORT_SYMBOL_GPL(__ide_error);
465 * ide_error - handle an error on the IDE
466 * @drive: drive the error occurred on
467 * @msg: message to report
470 * ide_error() takes action based on the error returned by the drive.
471 * For normal I/O that may well include retries. We deal with
472 * both new-style (taskfile) and old style command handling here.
473 * In the case of taskfile command handling there is work left to
477 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
482 err = ide_dump_status(drive, msg, stat);
484 if ((rq = HWGROUP(drive)->rq) == NULL)
487 /* retry only "normal" I/O: */
488 if (!blk_fs_request(rq)) {
490 ide_end_drive_cmd(drive, stat, err);
497 drv = *(ide_driver_t **)rq->rq_disk->private_data;
498 return drv->error(drive, rq, stat, err);
500 return __ide_error(drive, rq, stat, err);
503 EXPORT_SYMBOL_GPL(ide_error);
505 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
507 tf->nsect = drive->sect;
508 tf->lbal = drive->sect;
509 tf->lbam = drive->cyl;
510 tf->lbah = drive->cyl >> 8;
511 tf->device = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
512 tf->command = ATA_CMD_INIT_DEV_PARAMS;
515 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
517 tf->nsect = drive->sect;
518 tf->command = ATA_CMD_RESTORE;
521 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
523 tf->nsect = drive->mult_req;
524 tf->command = ATA_CMD_SET_MULTI;
527 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
529 special_t *s = &drive->special;
532 memset(&args, 0, sizeof(ide_task_t));
533 args.data_phase = TASKFILE_NO_DATA;
535 if (s->b.set_geometry) {
536 s->b.set_geometry = 0;
537 ide_tf_set_specify_cmd(drive, &args.tf);
538 } else if (s->b.recalibrate) {
539 s->b.recalibrate = 0;
540 ide_tf_set_restore_cmd(drive, &args.tf);
541 } else if (s->b.set_multmode) {
542 s->b.set_multmode = 0;
543 ide_tf_set_setmult_cmd(drive, &args.tf);
545 int special = s->all;
547 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
551 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
552 IDE_TFLAG_CUSTOM_HANDLER;
554 do_rw_taskfile(drive, &args);
560 * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
562 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
571 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
574 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
577 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
584 * do_special - issue some special commands
585 * @drive: drive the command is for
587 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
588 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
590 * It used to do much more, but has been scaled back.
593 static ide_startstop_t do_special (ide_drive_t *drive)
595 special_t *s = &drive->special;
598 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
601 ide_hwif_t *hwif = drive->hwif;
602 const struct ide_port_ops *port_ops = hwif->port_ops;
603 u8 req_pio = drive->tune_req;
607 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
609 * take ide_lock for drive->[no_]unmask/[no_]io_32bit
611 if (req_pio == 8 || req_pio == 9) {
614 spin_lock_irqsave(&ide_lock, flags);
615 port_ops->set_pio_mode(drive, req_pio);
616 spin_unlock_irqrestore(&ide_lock, flags);
618 port_ops->set_pio_mode(drive, req_pio);
620 int keep_dma = drive->using_dma;
622 ide_set_pio(drive, req_pio);
624 if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
632 if (drive->media == ide_disk)
633 return ide_disk_special(drive);
641 void ide_map_sg(ide_drive_t *drive, struct request *rq)
643 ide_hwif_t *hwif = drive->hwif;
644 struct scatterlist *sg = hwif->sg_table;
646 if (hwif->sg_mapped) /* needed by ide-scsi */
649 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
650 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
652 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
657 EXPORT_SYMBOL_GPL(ide_map_sg);
659 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
661 ide_hwif_t *hwif = drive->hwif;
663 hwif->nsect = hwif->nleft = rq->nr_sectors;
668 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
671 * execute_drive_command - issue special drive command
672 * @drive: the drive to issue the command on
673 * @rq: the request structure holding the command
675 * execute_drive_cmd() issues a special drive command, usually
676 * initiated by ioctl() from the external hdparm program. The
677 * command can be a drive command, drive task or taskfile
678 * operation. Weirdly you can call it with NULL to wait for
679 * all commands to finish. Don't do this as that is due to change
682 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
685 ide_hwif_t *hwif = HWIF(drive);
686 ide_task_t *task = rq->special;
689 hwif->data_phase = task->data_phase;
691 switch (hwif->data_phase) {
692 case TASKFILE_MULTI_OUT:
694 case TASKFILE_MULTI_IN:
696 ide_init_sg_cmd(drive, rq);
697 ide_map_sg(drive, rq);
702 return do_rw_taskfile(drive, task);
706 * NULL is actually a valid way of waiting for
707 * all current requests to be flushed from the queue.
710 printk("%s: DRIVE_CMD (null)\n", drive->name);
712 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
713 ide_read_error(drive));
718 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
720 switch (rq->cmd[0]) {
721 case REQ_DRIVE_RESET:
722 return ide_do_reset(drive);
724 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
725 ide_end_request(drive, 0, 0);
730 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
732 struct request_pm_state *pm = rq->data;
734 if (blk_pm_suspend_request(rq) &&
735 pm->pm_step == ide_pm_state_start_suspend)
736 /* Mark drive blocked when starting the suspend sequence. */
738 else if (blk_pm_resume_request(rq) &&
739 pm->pm_step == ide_pm_state_start_resume) {
741 * The first thing we do on wakeup is to wait for BSY bit to
742 * go away (with a looong timeout) as a drive on this hwif may
743 * just be POSTing itself.
744 * We do that before even selecting as the "other" device on
745 * the bus may be broken enough to walk on our toes at this
748 ide_hwif_t *hwif = drive->hwif;
751 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
753 rc = ide_wait_not_busy(hwif, 35000);
755 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
757 hwif->tp_ops->set_irq(hwif, 1);
758 rc = ide_wait_not_busy(hwif, 100000);
760 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
765 * start_request - start of I/O and command issuing for IDE
767 * start_request() initiates handling of a new I/O request. It
768 * accepts commands and I/O (read/write) requests.
770 * FIXME: this function needs a rename
773 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
775 ide_startstop_t startstop;
777 BUG_ON(!blk_rq_started(rq));
780 printk("%s: start_request: current=0x%08lx\n",
781 HWIF(drive)->name, (unsigned long) rq);
784 /* bail early if we've exceeded max_failures */
785 if (drive->max_failures && (drive->failures > drive->max_failures)) {
786 rq->cmd_flags |= REQ_FAILED;
790 if (blk_pm_request(rq))
791 ide_check_pm_state(drive, rq);
794 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
795 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
796 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
799 if (!drive->special.all) {
803 * We reset the drive so we need to issue a SETFEATURES.
804 * Do it _after_ do_special() restored device parameters.
806 if (drive->current_speed == 0xff)
807 ide_config_drive_speed(drive, drive->desired_speed);
809 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
810 return execute_drive_cmd(drive, rq);
811 else if (blk_pm_request(rq)) {
812 struct request_pm_state *pm = rq->data;
814 printk("%s: start_power_step(step: %d)\n",
815 drive->name, rq->pm->pm_step);
817 startstop = ide_start_power_step(drive, rq);
818 if (startstop == ide_stopped &&
819 pm->pm_step == ide_pm_state_completed)
820 ide_complete_pm_request(drive, rq);
822 } else if (!rq->rq_disk && blk_special_request(rq))
824 * TODO: Once all ULDs have been modified to
825 * check for specific op codes rather than
826 * blindly accepting any special request, the
827 * check for ->rq_disk above may be replaced
828 * by a more suitable mechanism or even
831 return ide_special_rq(drive, rq);
833 drv = *(ide_driver_t **)rq->rq_disk->private_data;
835 return drv->do_request(drive, rq, rq->sector);
837 return do_special(drive);
839 ide_kill_rq(drive, rq);
844 * ide_stall_queue - pause an IDE device
845 * @drive: drive to stall
846 * @timeout: time to stall for (jiffies)
848 * ide_stall_queue() can be used by a drive to give excess bandwidth back
849 * to the hwgroup by sleeping for timeout jiffies.
852 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
854 if (timeout > WAIT_WORSTCASE)
855 timeout = WAIT_WORSTCASE;
856 drive->sleep = timeout + jiffies;
860 EXPORT_SYMBOL(ide_stall_queue);
862 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
865 * choose_drive - select a drive to service
866 * @hwgroup: hardware group to select on
868 * choose_drive() selects the next drive which will be serviced.
869 * This is necessary because the IDE layer can't issue commands
870 * to both drives on the same cable, unlike SCSI.
873 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
875 ide_drive_t *drive, *best;
879 drive = hwgroup->drive;
882 * drive is doing pre-flush, ordered write, post-flush sequence. even
883 * though that is 3 requests, it must be seen as a single transaction.
884 * we must not preempt this drive until that is complete
886 if (blk_queue_flushing(drive->queue)) {
888 * small race where queue could get replugged during
889 * the 3-request flush cycle, just yank the plug since
890 * we want it to finish asap
892 blk_remove_plug(drive->queue);
897 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
898 && !elv_queue_empty(drive->queue)) {
900 || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
901 || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
903 if (!blk_queue_plugged(drive->queue))
907 } while ((drive = drive->next) != hwgroup->drive);
908 if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
909 long t = (signed long)(WAKEUP(best) - jiffies);
910 if (t >= WAIT_MIN_SLEEP) {
912 * We *may* have some time to spare, but first let's see if
913 * someone can potentially benefit from our nice mood today..
918 && time_before(jiffies - best->service_time, WAKEUP(drive))
919 && time_before(WAKEUP(drive), jiffies + t))
921 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
924 } while ((drive = drive->next) != best);
931 * Issue a new request to a drive from hwgroup
932 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
934 * A hwgroup is a serialized group of IDE interfaces. Usually there is
935 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
936 * may have both interfaces in a single hwgroup to "serialize" access.
937 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
938 * together into one hwgroup for serialized access.
940 * Note also that several hwgroups can end up sharing a single IRQ,
941 * possibly along with many other devices. This is especially common in
942 * PCI-based systems with off-board IDE controller cards.
944 * The IDE driver uses the single global ide_lock spinlock to protect
945 * access to the request queues, and to protect the hwgroup->busy flag.
947 * The first thread into the driver for a particular hwgroup sets the
948 * hwgroup->busy flag to indicate that this hwgroup is now active,
949 * and then initiates processing of the top request from the request queue.
951 * Other threads attempting entry notice the busy setting, and will simply
952 * queue their new requests and exit immediately. Note that hwgroup->busy
953 * remains set even when the driver is merely awaiting the next interrupt.
954 * Thus, the meaning is "this hwgroup is busy processing a request".
956 * When processing of a request completes, the completing thread or IRQ-handler
957 * will start the next request from the queue. If no more work remains,
958 * the driver will clear the hwgroup->busy flag and exit.
960 * The ide_lock (spinlock) is used to protect all access to the
961 * hwgroup->busy flag, but is otherwise not needed for most processing in
962 * the driver. This makes the driver much more friendlier to shared IRQs
963 * than previous designs, while remaining 100% (?) SMP safe and capable.
965 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
970 ide_startstop_t startstop;
973 /* for atari only: POSSIBLY BROKEN HERE(?) */
974 ide_get_lock(ide_intr, hwgroup);
976 /* caller must own ide_lock */
977 BUG_ON(!irqs_disabled());
979 while (!hwgroup->busy) {
981 drive = choose_drive(hwgroup);
984 unsigned long sleep = 0; /* shut up, gcc */
986 drive = hwgroup->drive;
988 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
990 sleep = drive->sleep;
992 } while ((drive = drive->next) != hwgroup->drive);
995 * Take a short snooze, and then wake up this hwgroup again.
996 * This gives other hwgroups on the same a chance to
997 * play fairly with us, just in case there are big differences
998 * in relative throughputs.. don't want to hog the cpu too much.
1000 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1001 sleep = jiffies + WAIT_MIN_SLEEP;
1003 if (timer_pending(&hwgroup->timer))
1004 printk(KERN_CRIT "ide_set_handler: timer already active\n");
1006 /* so that ide_timer_expiry knows what to do */
1007 hwgroup->sleeping = 1;
1008 hwgroup->req_gen_timer = hwgroup->req_gen;
1009 mod_timer(&hwgroup->timer, sleep);
1010 /* we purposely leave hwgroup->busy==1
1013 /* Ugly, but how can we sleep for the lock
1014 * otherwise? perhaps from tq_disk?
1017 /* for atari only */
1022 /* no more work for this hwgroup (for now) */
1027 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1029 * set nIEN for previous hwif, drives in the
1030 * quirk_list may not like intr setups/cleanups
1032 if (drive->quirk_list != 1)
1033 hwif->tp_ops->set_irq(hwif, 0);
1035 hwgroup->hwif = hwif;
1036 hwgroup->drive = drive;
1037 drive->sleeping = 0;
1038 drive->service_start = jiffies;
1040 if (blk_queue_plugged(drive->queue)) {
1041 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1046 * we know that the queue isn't empty, but this can happen
1047 * if the q->prep_rq_fn() decides to kill a request
1049 rq = elv_next_request(drive->queue);
1056 * Sanity: don't accept a request that isn't a PM request
1057 * if we are currently power managed. This is very important as
1058 * blk_stop_queue() doesn't prevent the elv_next_request()
1059 * above to return us whatever is in the queue. Since we call
1060 * ide_do_request() ourselves, we end up taking requests while
1061 * the queue is blocked...
1063 * We let requests forced at head of queue with ide-preempt
1064 * though. I hope that doesn't happen too much, hopefully not
1065 * unless the subdriver triggers such a thing in its own PM
1068 * We count how many times we loop here to make sure we service
1069 * all drives in the hwgroup without looping for ever
1071 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1072 drive = drive->next ? drive->next : hwgroup->drive;
1073 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1075 /* We clear busy, there should be no pending ATA command at this point. */
1083 * Some systems have trouble with IDE IRQs arriving while
1084 * the driver is still setting things up. So, here we disable
1085 * the IRQ used by this interface while the request is being started.
1086 * This may look bad at first, but pretty much the same thing
1087 * happens anyway when any interrupt comes in, IDE or otherwise
1088 * -- the kernel masks the IRQ while it is being handled.
1090 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1091 disable_irq_nosync(hwif->irq);
1092 spin_unlock(&ide_lock);
1093 local_irq_enable_in_hardirq();
1094 /* allow other IRQs while we start this request */
1095 startstop = start_request(drive, rq);
1096 spin_lock_irq(&ide_lock);
1097 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1098 enable_irq(hwif->irq);
1099 if (startstop == ide_stopped)
1105 * Passes the stuff to ide_do_request
1107 void do_ide_request(struct request_queue *q)
1109 ide_drive_t *drive = q->queuedata;
1111 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1115 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1116 * retry the current request in pio mode instead of risking tossing it
1119 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1121 ide_hwif_t *hwif = HWIF(drive);
1123 ide_startstop_t ret = ide_stopped;
1126 * end current dma transaction
1130 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1131 (void)hwif->dma_ops->dma_end(drive);
1132 ret = ide_error(drive, "dma timeout error",
1133 hwif->tp_ops->read_status(hwif));
1135 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1136 hwif->dma_ops->dma_timeout(drive);
1140 * disable dma for now, but remember that we did so because of
1141 * a timeout -- we'll reenable after we finish this next request
1142 * (or rather the first chunk of it) in pio.
1145 drive->state = DMA_PIO_RETRY;
1146 ide_dma_off_quietly(drive);
1149 * un-busy drive etc (hwgroup->busy is cleared on return) and
1150 * make sure request is sane
1152 rq = HWGROUP(drive)->rq;
1157 HWGROUP(drive)->rq = NULL;
1164 rq->sector = rq->bio->bi_sector;
1165 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1166 rq->hard_cur_sectors = rq->current_nr_sectors;
1167 rq->buffer = bio_data(rq->bio);
1173 * ide_timer_expiry - handle lack of an IDE interrupt
1174 * @data: timer callback magic (hwgroup)
1176 * An IDE command has timed out before the expected drive return
1177 * occurred. At this point we attempt to clean up the current
1178 * mess. If the current handler includes an expiry handler then
1179 * we invoke the expiry handler, and providing it is happy the
1180 * work is done. If that fails we apply generic recovery rules
1181 * invoking the handler and checking the drive DMA status. We
1182 * have an excessively incestuous relationship with the DMA
1183 * logic that wants cleaning up.
1186 void ide_timer_expiry (unsigned long data)
1188 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1189 ide_handler_t *handler;
1190 ide_expiry_t *expiry;
1191 unsigned long flags;
1192 unsigned long wait = -1;
1194 spin_lock_irqsave(&ide_lock, flags);
1196 if (((handler = hwgroup->handler) == NULL) ||
1197 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1199 * Either a marginal timeout occurred
1200 * (got the interrupt just as timer expired),
1201 * or we were "sleeping" to give other devices a chance.
1202 * Either way, we don't really want to complain about anything.
1204 if (hwgroup->sleeping) {
1205 hwgroup->sleeping = 0;
1209 ide_drive_t *drive = hwgroup->drive;
1211 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1212 hwgroup->handler = NULL;
1215 ide_startstop_t startstop = ide_stopped;
1216 if (!hwgroup->busy) {
1217 hwgroup->busy = 1; /* paranoia */
1218 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1220 if ((expiry = hwgroup->expiry) != NULL) {
1222 if ((wait = expiry(drive)) > 0) {
1224 hwgroup->timer.expires = jiffies + wait;
1225 hwgroup->req_gen_timer = hwgroup->req_gen;
1226 add_timer(&hwgroup->timer);
1227 spin_unlock_irqrestore(&ide_lock, flags);
1231 hwgroup->handler = NULL;
1233 * We need to simulate a real interrupt when invoking
1234 * the handler() function, which means we need to
1235 * globally mask the specific IRQ:
1237 spin_unlock(&ide_lock);
1239 /* disable_irq_nosync ?? */
1240 disable_irq(hwif->irq);
1242 * as if we were handling an interrupt */
1243 local_irq_disable();
1244 if (hwgroup->polling) {
1245 startstop = handler(drive);
1246 } else if (drive_is_ready(drive)) {
1247 if (drive->waiting_for_dma)
1248 hwif->dma_ops->dma_lost_irq(drive);
1249 (void)ide_ack_intr(hwif);
1250 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1251 startstop = handler(drive);
1253 if (drive->waiting_for_dma) {
1254 startstop = ide_dma_timeout_retry(drive, wait);
1257 ide_error(drive, "irq timeout",
1258 hwif->tp_ops->read_status(hwif));
1260 drive->service_time = jiffies - drive->service_start;
1261 spin_lock_irq(&ide_lock);
1262 enable_irq(hwif->irq);
1263 if (startstop == ide_stopped)
1267 ide_do_request(hwgroup, IDE_NO_IRQ);
1268 spin_unlock_irqrestore(&ide_lock, flags);
1272 * unexpected_intr - handle an unexpected IDE interrupt
1273 * @irq: interrupt line
1274 * @hwgroup: hwgroup being processed
1276 * There's nothing really useful we can do with an unexpected interrupt,
1277 * other than reading the status register (to clear it), and logging it.
1278 * There should be no way that an irq can happen before we're ready for it,
1279 * so we needn't worry much about losing an "important" interrupt here.
1281 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1282 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1283 * looks "good", we just ignore the interrupt completely.
1285 * This routine assumes __cli() is in effect when called.
1287 * If an unexpected interrupt happens on irq15 while we are handling irq14
1288 * and if the two interfaces are "serialized" (CMD640), then it looks like
1289 * we could screw up by interfering with a new request being set up for
1292 * In reality, this is a non-issue. The new command is not sent unless
1293 * the drive is ready to accept one, in which case we know the drive is
1294 * not trying to interrupt us. And ide_set_handler() is always invoked
1295 * before completing the issuance of any new drive command, so we will not
1296 * be accidentally invoked as a result of any valid command completion
1299 * Note that we must walk the entire hwgroup here. We know which hwif
1300 * is doing the current command, but we don't know which hwif burped
1304 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1307 ide_hwif_t *hwif = hwgroup->hwif;
1310 * handle the unexpected interrupt
1313 if (hwif->irq == irq) {
1314 stat = hwif->tp_ops->read_status(hwif);
1316 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1317 /* Try to not flood the console with msgs */
1318 static unsigned long last_msgtime, count;
1320 if (time_after(jiffies, last_msgtime + HZ)) {
1321 last_msgtime = jiffies;
1322 printk(KERN_ERR "%s%s: unexpected interrupt, "
1323 "status=0x%02x, count=%ld\n",
1325 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1329 } while ((hwif = hwif->next) != hwgroup->hwif);
1333 * ide_intr - default IDE interrupt handler
1334 * @irq: interrupt number
1335 * @dev_id: hwif group
1336 * @regs: unused weirdness from the kernel irq layer
1338 * This is the default IRQ handler for the IDE layer. You should
1339 * not need to override it. If you do be aware it is subtle in
1342 * hwgroup->hwif is the interface in the group currently performing
1343 * a command. hwgroup->drive is the drive and hwgroup->handler is
1344 * the IRQ handler to call. As we issue a command the handlers
1345 * step through multiple states, reassigning the handler to the
1346 * next step in the process. Unlike a smart SCSI controller IDE
1347 * expects the main processor to sequence the various transfer
1348 * stages. We also manage a poll timer to catch up with most
1349 * timeout situations. There are still a few where the handlers
1350 * don't ever decide to give up.
1352 * The handler eventually returns ide_stopped to indicate the
1353 * request completed. At this point we issue the next request
1354 * on the hwgroup and the process begins again.
1357 irqreturn_t ide_intr (int irq, void *dev_id)
1359 unsigned long flags;
1360 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1363 ide_handler_t *handler;
1364 ide_startstop_t startstop;
1366 spin_lock_irqsave(&ide_lock, flags);
1367 hwif = hwgroup->hwif;
1369 if (!ide_ack_intr(hwif)) {
1370 spin_unlock_irqrestore(&ide_lock, flags);
1374 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1376 * Not expecting an interrupt from this drive.
1377 * That means this could be:
1378 * (1) an interrupt from another PCI device
1379 * sharing the same PCI INT# as us.
1380 * or (2) a drive just entered sleep or standby mode,
1381 * and is interrupting to let us know.
1382 * or (3) a spurious interrupt of unknown origin.
1384 * For PCI, we cannot tell the difference,
1385 * so in that case we just ignore it and hope it goes away.
1387 * FIXME: unexpected_intr should be hwif-> then we can
1388 * remove all the ifdef PCI crap
1390 #ifdef CONFIG_BLK_DEV_IDEPCI
1391 if (hwif->chipset != ide_pci)
1392 #endif /* CONFIG_BLK_DEV_IDEPCI */
1395 * Probably not a shared PCI interrupt,
1396 * so we can safely try to do something about it:
1398 unexpected_intr(irq, hwgroup);
1399 #ifdef CONFIG_BLK_DEV_IDEPCI
1402 * Whack the status register, just in case
1403 * we have a leftover pending IRQ.
1405 (void)hwif->tp_ops->read_status(hwif);
1406 #endif /* CONFIG_BLK_DEV_IDEPCI */
1408 spin_unlock_irqrestore(&ide_lock, flags);
1411 drive = hwgroup->drive;
1414 * This should NEVER happen, and there isn't much
1415 * we could do about it here.
1417 * [Note - this can occur if the drive is hot unplugged]
1419 spin_unlock_irqrestore(&ide_lock, flags);
1422 if (!drive_is_ready(drive)) {
1424 * This happens regularly when we share a PCI IRQ with
1425 * another device. Unfortunately, it can also happen
1426 * with some buggy drives that trigger the IRQ before
1427 * their status register is up to date. Hopefully we have
1428 * enough advance overhead that the latter isn't a problem.
1430 spin_unlock_irqrestore(&ide_lock, flags);
1433 if (!hwgroup->busy) {
1434 hwgroup->busy = 1; /* paranoia */
1435 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1437 hwgroup->handler = NULL;
1439 del_timer(&hwgroup->timer);
1440 spin_unlock(&ide_lock);
1442 /* Some controllers might set DMA INTR no matter DMA or PIO;
1443 * bmdma status might need to be cleared even for
1444 * PIO interrupts to prevent spurious/lost irq.
1446 if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1447 /* ide_dma_end() needs bmdma status for error checking.
1448 * So, skip clearing bmdma status here and leave it
1449 * to ide_dma_end() if this is dma interrupt.
1451 hwif->ide_dma_clear_irq(drive);
1454 local_irq_enable_in_hardirq();
1455 /* service this interrupt, may set handler for next interrupt */
1456 startstop = handler(drive);
1457 spin_lock_irq(&ide_lock);
1460 * Note that handler() may have set things up for another
1461 * interrupt to occur soon, but it cannot happen until
1462 * we exit from this routine, because it will be the
1463 * same irq as is currently being serviced here, and Linux
1464 * won't allow another of the same (on any CPU) until we return.
1466 drive->service_time = jiffies - drive->service_start;
1467 if (startstop == ide_stopped) {
1468 if (hwgroup->handler == NULL) { /* paranoia */
1470 ide_do_request(hwgroup, hwif->irq);
1472 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1473 "on exit\n", drive->name);
1476 spin_unlock_irqrestore(&ide_lock, flags);
1481 * ide_do_drive_cmd - issue IDE special command
1482 * @drive: device to issue command
1483 * @rq: request to issue
1485 * This function issues a special IDE device request
1486 * onto the request queue.
1488 * the rq is queued at the head of the request queue, displacing
1489 * the currently-being-processed request and this function
1490 * returns immediately without waiting for the new rq to be
1491 * completed. This is VERY DANGEROUS, and is intended for
1492 * careful use by the ATAPI tape/cdrom driver code.
1495 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1497 unsigned long flags;
1498 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1500 spin_lock_irqsave(&ide_lock, flags);
1502 __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1503 __generic_unplug_device(drive->queue);
1504 spin_unlock_irqrestore(&ide_lock, flags);
1507 EXPORT_SYMBOL(ide_do_drive_cmd);
1509 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1511 ide_hwif_t *hwif = drive->hwif;
1514 memset(&task, 0, sizeof(task));
1515 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1516 IDE_TFLAG_OUT_FEATURE | tf_flags;
1517 task.tf.feature = dma; /* Use PIO/DMA */
1518 task.tf.lbam = bcount & 0xff;
1519 task.tf.lbah = (bcount >> 8) & 0xff;
1521 ide_tf_dump(drive->name, &task.tf);
1522 hwif->tp_ops->set_irq(hwif, 1);
1523 SELECT_MASK(drive, 0);
1524 hwif->tp_ops->tf_load(drive, &task);
1527 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1529 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1531 ide_hwif_t *hwif = drive->hwif;
1536 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1538 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1542 EXPORT_SYMBOL_GPL(ide_pad_transfer);