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->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
86 if (!__blk_end_request(rq, error, nr_bytes)) {
88 HWGROUP(drive)->rq = NULL;
96 * ide_end_request - complete an IDE I/O
97 * @drive: IDE device for the I/O
99 * @nr_sectors: number of sectors completed
101 * This is our end_request wrapper function. We complete the I/O
102 * update random number input and dequeue the request, which if
103 * it was tagged may be out of order.
106 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
108 unsigned int nr_bytes = nr_sectors << 9;
114 * room for locking improvements here, the calls below don't
115 * need the queue lock held at all
117 spin_lock_irqsave(&ide_lock, flags);
118 rq = HWGROUP(drive)->rq;
121 if (blk_pc_request(rq))
122 nr_bytes = rq->data_len;
124 nr_bytes = rq->hard_cur_sectors << 9;
127 ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
129 spin_unlock_irqrestore(&ide_lock, flags);
132 EXPORT_SYMBOL(ide_end_request);
135 * Power Management state machine. This one is rather trivial for now,
136 * we should probably add more, like switching back to PIO on suspend
137 * to help some BIOSes, re-do the door locking on resume, etc...
141 ide_pm_flush_cache = ide_pm_state_start_suspend,
144 idedisk_pm_restore_pio = ide_pm_state_start_resume,
149 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
151 struct request_pm_state *pm = rq->data;
153 if (drive->media != ide_disk)
156 switch (pm->pm_step) {
157 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) complete */
158 if (pm->pm_state == PM_EVENT_FREEZE)
159 pm->pm_step = ide_pm_state_completed;
161 pm->pm_step = idedisk_pm_standby;
163 case idedisk_pm_standby: /* Suspend step 2 (standby) complete */
164 pm->pm_step = ide_pm_state_completed;
166 case idedisk_pm_restore_pio: /* Resume step 1 complete */
167 pm->pm_step = idedisk_pm_idle;
169 case idedisk_pm_idle: /* Resume step 2 (idle) complete */
170 pm->pm_step = ide_pm_restore_dma;
175 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
177 struct request_pm_state *pm = rq->data;
178 ide_task_t *args = rq->special;
180 memset(args, 0, sizeof(*args));
182 switch (pm->pm_step) {
183 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) */
184 if (drive->media != ide_disk)
186 /* Not supported? Switch to next step now. */
187 if (ata_id_flush_enabled(drive->id) == 0 ||
188 (drive->dev_flags & IDE_DFLAG_WCACHE) == 0) {
189 ide_complete_power_step(drive, rq, 0, 0);
192 if (ata_id_flush_ext_enabled(drive->id))
193 args->tf.command = ATA_CMD_FLUSH_EXT;
195 args->tf.command = ATA_CMD_FLUSH;
198 case idedisk_pm_standby: /* Suspend step 2 (standby) */
199 args->tf.command = ATA_CMD_STANDBYNOW1;
202 case idedisk_pm_restore_pio: /* Resume step 1 (restore PIO) */
203 ide_set_max_pio(drive);
205 * skip idedisk_pm_idle for ATAPI devices
207 if (drive->media != ide_disk)
208 pm->pm_step = ide_pm_restore_dma;
210 ide_complete_power_step(drive, rq, 0, 0);
213 case idedisk_pm_idle: /* Resume step 2 (idle) */
214 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
217 case ide_pm_restore_dma: /* Resume step 3 (restore DMA) */
219 * Right now, all we do is call ide_set_dma(drive),
220 * we could be smarter and check for current xfer_speed
221 * in struct drive etc...
223 if (drive->hwif->dma_ops == NULL)
226 * TODO: respect IDE_DFLAG_USING_DMA
231 pm->pm_step = ide_pm_state_completed;
235 args->tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
236 args->data_phase = TASKFILE_NO_DATA;
237 return do_rw_taskfile(drive, args);
241 * ide_end_dequeued_request - complete an IDE I/O
242 * @drive: IDE device for the I/O
244 * @nr_sectors: number of sectors completed
246 * Complete an I/O that is no longer on the request queue. This
247 * typically occurs when we pull the request and issue a REQUEST_SENSE.
248 * We must still finish the old request but we must not tamper with the
249 * queue in the meantime.
251 * NOTE: This path does not handle barrier, but barrier is not supported
255 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
256 int uptodate, int nr_sectors)
261 spin_lock_irqsave(&ide_lock, flags);
262 BUG_ON(!blk_rq_started(rq));
263 ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
264 spin_unlock_irqrestore(&ide_lock, flags);
268 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
272 * ide_complete_pm_request - end the current Power Management request
273 * @drive: target drive
276 * This function cleans up the current PM request and stops the queue
279 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
284 printk("%s: completing PM request, %s\n", drive->name,
285 blk_pm_suspend_request(rq) ? "suspend" : "resume");
287 spin_lock_irqsave(&ide_lock, flags);
288 if (blk_pm_suspend_request(rq)) {
289 blk_stop_queue(drive->queue);
291 drive->dev_flags &= ~IDE_DFLAG_BLOCKED;
292 blk_start_queue(drive->queue);
294 HWGROUP(drive)->rq = NULL;
295 if (__blk_end_request(rq, 0, 0))
297 spin_unlock_irqrestore(&ide_lock, flags);
301 * ide_end_drive_cmd - end an explicit drive command
306 * Clean up after success/failure of an explicit drive command.
307 * These get thrown onto the queue so they are synchronized with
308 * real I/O operations on the drive.
310 * In LBA48 mode we have to read the register set twice to get
311 * all the extra information out.
314 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
319 spin_lock_irqsave(&ide_lock, flags);
320 rq = HWGROUP(drive)->rq;
321 spin_unlock_irqrestore(&ide_lock, flags);
323 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
324 ide_task_t *task = (ide_task_t *)rq->special;
327 rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
330 struct ide_taskfile *tf = &task->tf;
335 drive->hwif->tp_ops->tf_read(drive, task);
337 if (task->tf_flags & IDE_TFLAG_DYN)
340 } else if (blk_pm_request(rq)) {
341 struct request_pm_state *pm = rq->data;
343 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
344 drive->name, rq->pm->pm_step, stat, err);
346 ide_complete_power_step(drive, rq, stat, err);
347 if (pm->pm_step == ide_pm_state_completed)
348 ide_complete_pm_request(drive, rq);
352 spin_lock_irqsave(&ide_lock, flags);
353 HWGROUP(drive)->rq = NULL;
355 if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
358 spin_unlock_irqrestore(&ide_lock, flags);
361 EXPORT_SYMBOL(ide_end_drive_cmd);
363 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
368 drv = *(ide_driver_t **)rq->rq_disk->private_data;
369 drv->end_request(drive, 0, 0);
371 ide_end_request(drive, 0, 0);
374 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
376 ide_hwif_t *hwif = drive->hwif;
378 if ((stat & ATA_BUSY) ||
379 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
380 /* other bits are useless when BUSY */
381 rq->errors |= ERROR_RESET;
382 } else if (stat & ATA_ERR) {
383 /* err has different meaning on cdrom and tape */
384 if (err == ATA_ABORTED) {
385 if (drive->select.b.lba &&
386 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
387 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
389 } else if ((err & BAD_CRC) == BAD_CRC) {
390 /* UDMA crc error, just retry the operation */
392 } else if (err & (ATA_BBK | ATA_UNC)) {
393 /* retries won't help these */
394 rq->errors = ERROR_MAX;
395 } else if (err & ATA_TRK0NF) {
396 /* help it find track zero */
397 rq->errors |= ERROR_RECAL;
401 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
402 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
403 int nsect = drive->mult_count ? drive->mult_count : 1;
405 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
408 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
409 ide_kill_rq(drive, rq);
413 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
414 rq->errors |= ERROR_RESET;
416 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
418 return ide_do_reset(drive);
421 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
422 drive->special.b.recalibrate = 1;
429 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
431 ide_hwif_t *hwif = drive->hwif;
433 if ((stat & ATA_BUSY) ||
434 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
435 /* other bits are useless when BUSY */
436 rq->errors |= ERROR_RESET;
438 /* add decoding error stuff */
441 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
443 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
445 if (rq->errors >= ERROR_MAX) {
446 ide_kill_rq(drive, rq);
448 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
450 return ide_do_reset(drive);
459 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
461 if (drive->media == ide_disk)
462 return ide_ata_error(drive, rq, stat, err);
463 return ide_atapi_error(drive, rq, stat, err);
466 EXPORT_SYMBOL_GPL(__ide_error);
469 * ide_error - handle an error on the IDE
470 * @drive: drive the error occurred on
471 * @msg: message to report
474 * ide_error() takes action based on the error returned by the drive.
475 * For normal I/O that may well include retries. We deal with
476 * both new-style (taskfile) and old style command handling here.
477 * In the case of taskfile command handling there is work left to
481 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
486 err = ide_dump_status(drive, msg, stat);
488 if ((rq = HWGROUP(drive)->rq) == NULL)
491 /* retry only "normal" I/O: */
492 if (!blk_fs_request(rq)) {
494 ide_end_drive_cmd(drive, stat, err);
501 drv = *(ide_driver_t **)rq->rq_disk->private_data;
502 return drv->error(drive, rq, stat, err);
504 return __ide_error(drive, rq, stat, err);
507 EXPORT_SYMBOL_GPL(ide_error);
509 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
511 tf->nsect = drive->sect;
512 tf->lbal = drive->sect;
513 tf->lbam = drive->cyl;
514 tf->lbah = drive->cyl >> 8;
515 tf->device = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
516 tf->command = ATA_CMD_INIT_DEV_PARAMS;
519 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
521 tf->nsect = drive->sect;
522 tf->command = ATA_CMD_RESTORE;
525 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
527 tf->nsect = drive->mult_req;
528 tf->command = ATA_CMD_SET_MULTI;
531 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
533 special_t *s = &drive->special;
536 memset(&args, 0, sizeof(ide_task_t));
537 args.data_phase = TASKFILE_NO_DATA;
539 if (s->b.set_geometry) {
540 s->b.set_geometry = 0;
541 ide_tf_set_specify_cmd(drive, &args.tf);
542 } else if (s->b.recalibrate) {
543 s->b.recalibrate = 0;
544 ide_tf_set_restore_cmd(drive, &args.tf);
545 } else if (s->b.set_multmode) {
546 s->b.set_multmode = 0;
547 ide_tf_set_setmult_cmd(drive, &args.tf);
549 int special = s->all;
551 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
555 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
556 IDE_TFLAG_CUSTOM_HANDLER;
558 do_rw_taskfile(drive, &args);
564 * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
566 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
575 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
578 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
581 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
588 * do_special - issue some special commands
589 * @drive: drive the command is for
591 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
592 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
594 * It used to do much more, but has been scaled back.
597 static ide_startstop_t do_special (ide_drive_t *drive)
599 special_t *s = &drive->special;
602 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
605 ide_hwif_t *hwif = drive->hwif;
606 const struct ide_port_ops *port_ops = hwif->port_ops;
607 u8 req_pio = drive->tune_req;
611 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
613 * take ide_lock for IDE_DFLAG_[NO_]UNMASK/[NO_]IO_32BIT
615 if (req_pio == 8 || req_pio == 9) {
618 spin_lock_irqsave(&ide_lock, flags);
619 port_ops->set_pio_mode(drive, req_pio);
620 spin_unlock_irqrestore(&ide_lock, flags);
622 port_ops->set_pio_mode(drive, req_pio);
625 !!(drive->dev_flags & IDE_DFLAG_USING_DMA);
627 ide_set_pio(drive, req_pio);
629 if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
637 if (drive->media == ide_disk)
638 return ide_disk_special(drive);
646 void ide_map_sg(ide_drive_t *drive, struct request *rq)
648 ide_hwif_t *hwif = drive->hwif;
649 struct scatterlist *sg = hwif->sg_table;
651 if (hwif->sg_mapped) /* needed by ide-scsi */
654 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
655 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
657 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
662 EXPORT_SYMBOL_GPL(ide_map_sg);
664 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
666 ide_hwif_t *hwif = drive->hwif;
668 hwif->nsect = hwif->nleft = rq->nr_sectors;
673 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
676 * execute_drive_command - issue special drive command
677 * @drive: the drive to issue the command on
678 * @rq: the request structure holding the command
680 * execute_drive_cmd() issues a special drive command, usually
681 * initiated by ioctl() from the external hdparm program. The
682 * command can be a drive command, drive task or taskfile
683 * operation. Weirdly you can call it with NULL to wait for
684 * all commands to finish. Don't do this as that is due to change
687 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
690 ide_hwif_t *hwif = HWIF(drive);
691 ide_task_t *task = rq->special;
694 hwif->data_phase = task->data_phase;
696 switch (hwif->data_phase) {
697 case TASKFILE_MULTI_OUT:
699 case TASKFILE_MULTI_IN:
701 ide_init_sg_cmd(drive, rq);
702 ide_map_sg(drive, rq);
707 return do_rw_taskfile(drive, task);
711 * NULL is actually a valid way of waiting for
712 * all current requests to be flushed from the queue.
715 printk("%s: DRIVE_CMD (null)\n", drive->name);
717 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
718 ide_read_error(drive));
723 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
726 struct request_queue *q = drive->queue;
730 if (!(setting->flags & DS_SYNC))
731 return setting->set(drive, arg);
733 rq = blk_get_request(q, READ, GFP_KERNEL);
737 rq->cmd_type = REQ_TYPE_SPECIAL;
739 rq->cmd[0] = REQ_DEVSET_EXEC;
740 *(int *)&rq->cmd[1] = arg;
741 rq->special = setting->set;
743 if (blk_execute_rq(q, NULL, rq, 0))
749 EXPORT_SYMBOL_GPL(ide_devset_execute);
751 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
753 switch (rq->cmd[0]) {
754 case REQ_DEVSET_EXEC:
756 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
758 err = setfunc(drive, *(int *)&rq->cmd[1]);
763 ide_end_request(drive, err, 0);
766 case REQ_DRIVE_RESET:
767 return ide_do_reset(drive);
769 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
770 ide_end_request(drive, 0, 0);
775 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
777 struct request_pm_state *pm = rq->data;
779 if (blk_pm_suspend_request(rq) &&
780 pm->pm_step == ide_pm_state_start_suspend)
781 /* Mark drive blocked when starting the suspend sequence. */
782 drive->dev_flags |= IDE_DFLAG_BLOCKED;
783 else if (blk_pm_resume_request(rq) &&
784 pm->pm_step == ide_pm_state_start_resume) {
786 * The first thing we do on wakeup is to wait for BSY bit to
787 * go away (with a looong timeout) as a drive on this hwif may
788 * just be POSTing itself.
789 * We do that before even selecting as the "other" device on
790 * the bus may be broken enough to walk on our toes at this
793 ide_hwif_t *hwif = drive->hwif;
796 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
798 rc = ide_wait_not_busy(hwif, 35000);
800 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
802 hwif->tp_ops->set_irq(hwif, 1);
803 rc = ide_wait_not_busy(hwif, 100000);
805 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
810 * start_request - start of I/O and command issuing for IDE
812 * start_request() initiates handling of a new I/O request. It
813 * accepts commands and I/O (read/write) requests.
815 * FIXME: this function needs a rename
818 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
820 ide_startstop_t startstop;
822 BUG_ON(!blk_rq_started(rq));
825 printk("%s: start_request: current=0x%08lx\n",
826 HWIF(drive)->name, (unsigned long) rq);
829 /* bail early if we've exceeded max_failures */
830 if (drive->max_failures && (drive->failures > drive->max_failures)) {
831 rq->cmd_flags |= REQ_FAILED;
835 if (blk_pm_request(rq))
836 ide_check_pm_state(drive, rq);
839 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
840 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
841 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
844 if (!drive->special.all) {
848 * We reset the drive so we need to issue a SETFEATURES.
849 * Do it _after_ do_special() restored device parameters.
851 if (drive->current_speed == 0xff)
852 ide_config_drive_speed(drive, drive->desired_speed);
854 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
855 return execute_drive_cmd(drive, rq);
856 else if (blk_pm_request(rq)) {
857 struct request_pm_state *pm = rq->data;
859 printk("%s: start_power_step(step: %d)\n",
860 drive->name, rq->pm->pm_step);
862 startstop = ide_start_power_step(drive, rq);
863 if (startstop == ide_stopped &&
864 pm->pm_step == ide_pm_state_completed)
865 ide_complete_pm_request(drive, rq);
867 } else if (!rq->rq_disk && blk_special_request(rq))
869 * TODO: Once all ULDs have been modified to
870 * check for specific op codes rather than
871 * blindly accepting any special request, the
872 * check for ->rq_disk above may be replaced
873 * by a more suitable mechanism or even
876 return ide_special_rq(drive, rq);
878 drv = *(ide_driver_t **)rq->rq_disk->private_data;
880 return drv->do_request(drive, rq, rq->sector);
882 return do_special(drive);
884 ide_kill_rq(drive, rq);
889 * ide_stall_queue - pause an IDE device
890 * @drive: drive to stall
891 * @timeout: time to stall for (jiffies)
893 * ide_stall_queue() can be used by a drive to give excess bandwidth back
894 * to the hwgroup by sleeping for timeout jiffies.
897 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
899 if (timeout > WAIT_WORSTCASE)
900 timeout = WAIT_WORSTCASE;
901 drive->sleep = timeout + jiffies;
902 drive->dev_flags |= IDE_DFLAG_SLEEPING;
905 EXPORT_SYMBOL(ide_stall_queue);
907 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
910 * choose_drive - select a drive to service
911 * @hwgroup: hardware group to select on
913 * choose_drive() selects the next drive which will be serviced.
914 * This is necessary because the IDE layer can't issue commands
915 * to both drives on the same cable, unlike SCSI.
918 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
920 ide_drive_t *drive, *best;
924 drive = hwgroup->drive;
927 * drive is doing pre-flush, ordered write, post-flush sequence. even
928 * though that is 3 requests, it must be seen as a single transaction.
929 * we must not preempt this drive until that is complete
931 if (blk_queue_flushing(drive->queue)) {
933 * small race where queue could get replugged during
934 * the 3-request flush cycle, just yank the plug since
935 * we want it to finish asap
937 blk_remove_plug(drive->queue);
942 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
943 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
945 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
946 !elv_queue_empty(drive->queue)) {
948 (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
949 (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
950 if (!blk_queue_plugged(drive->queue))
954 } while ((drive = drive->next) != hwgroup->drive);
956 if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
957 (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
958 best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
959 long t = (signed long)(WAKEUP(best) - jiffies);
960 if (t >= WAIT_MIN_SLEEP) {
962 * We *may* have some time to spare, but first let's see if
963 * someone can potentially benefit from our nice mood today..
967 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
968 && time_before(jiffies - best->service_time, WAKEUP(drive))
969 && time_before(WAKEUP(drive), jiffies + t))
971 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
974 } while ((drive = drive->next) != best);
981 * Issue a new request to a drive from hwgroup
982 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
984 * A hwgroup is a serialized group of IDE interfaces. Usually there is
985 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
986 * may have both interfaces in a single hwgroup to "serialize" access.
987 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
988 * together into one hwgroup for serialized access.
990 * Note also that several hwgroups can end up sharing a single IRQ,
991 * possibly along with many other devices. This is especially common in
992 * PCI-based systems with off-board IDE controller cards.
994 * The IDE driver uses the single global ide_lock spinlock to protect
995 * access to the request queues, and to protect the hwgroup->busy flag.
997 * The first thread into the driver for a particular hwgroup sets the
998 * hwgroup->busy flag to indicate that this hwgroup is now active,
999 * and then initiates processing of the top request from the request queue.
1001 * Other threads attempting entry notice the busy setting, and will simply
1002 * queue their new requests and exit immediately. Note that hwgroup->busy
1003 * remains set even when the driver is merely awaiting the next interrupt.
1004 * Thus, the meaning is "this hwgroup is busy processing a request".
1006 * When processing of a request completes, the completing thread or IRQ-handler
1007 * will start the next request from the queue. If no more work remains,
1008 * the driver will clear the hwgroup->busy flag and exit.
1010 * The ide_lock (spinlock) is used to protect all access to the
1011 * hwgroup->busy flag, but is otherwise not needed for most processing in
1012 * the driver. This makes the driver much more friendlier to shared IRQs
1013 * than previous designs, while remaining 100% (?) SMP safe and capable.
1015 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1020 ide_startstop_t startstop;
1023 /* for atari only: POSSIBLY BROKEN HERE(?) */
1024 ide_get_lock(ide_intr, hwgroup);
1026 /* caller must own ide_lock */
1027 BUG_ON(!irqs_disabled());
1029 while (!hwgroup->busy) {
1031 drive = choose_drive(hwgroup);
1032 if (drive == NULL) {
1034 unsigned long sleep = 0; /* shut up, gcc */
1036 drive = hwgroup->drive;
1038 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
1040 time_before(drive->sleep, sleep))) {
1042 sleep = drive->sleep;
1044 } while ((drive = drive->next) != hwgroup->drive);
1047 * Take a short snooze, and then wake up this hwgroup again.
1048 * This gives other hwgroups on the same a chance to
1049 * play fairly with us, just in case there are big differences
1050 * in relative throughputs.. don't want to hog the cpu too much.
1052 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1053 sleep = jiffies + WAIT_MIN_SLEEP;
1055 if (timer_pending(&hwgroup->timer))
1056 printk(KERN_CRIT "ide_set_handler: timer already active\n");
1058 /* so that ide_timer_expiry knows what to do */
1059 hwgroup->sleeping = 1;
1060 hwgroup->req_gen_timer = hwgroup->req_gen;
1061 mod_timer(&hwgroup->timer, sleep);
1062 /* we purposely leave hwgroup->busy==1
1065 /* Ugly, but how can we sleep for the lock
1066 * otherwise? perhaps from tq_disk?
1069 /* for atari only */
1074 /* no more work for this hwgroup (for now) */
1079 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1081 * set nIEN for previous hwif, drives in the
1082 * quirk_list may not like intr setups/cleanups
1084 if (drive->quirk_list != 1)
1085 hwif->tp_ops->set_irq(hwif, 0);
1087 hwgroup->hwif = hwif;
1088 hwgroup->drive = drive;
1089 drive->dev_flags &= ~IDE_DFLAG_SLEEPING;
1090 drive->service_start = jiffies;
1092 if (blk_queue_plugged(drive->queue)) {
1093 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1098 * we know that the queue isn't empty, but this can happen
1099 * if the q->prep_rq_fn() decides to kill a request
1101 rq = elv_next_request(drive->queue);
1108 * Sanity: don't accept a request that isn't a PM request
1109 * if we are currently power managed. This is very important as
1110 * blk_stop_queue() doesn't prevent the elv_next_request()
1111 * above to return us whatever is in the queue. Since we call
1112 * ide_do_request() ourselves, we end up taking requests while
1113 * the queue is blocked...
1115 * We let requests forced at head of queue with ide-preempt
1116 * though. I hope that doesn't happen too much, hopefully not
1117 * unless the subdriver triggers such a thing in its own PM
1120 * We count how many times we loop here to make sure we service
1121 * all drives in the hwgroup without looping for ever
1123 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
1124 blk_pm_request(rq) == 0 &&
1125 (rq->cmd_flags & REQ_PREEMPT) == 0) {
1126 drive = drive->next ? drive->next : hwgroup->drive;
1127 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1129 /* We clear busy, there should be no pending ATA command at this point. */
1137 * Some systems have trouble with IDE IRQs arriving while
1138 * the driver is still setting things up. So, here we disable
1139 * the IRQ used by this interface while the request is being started.
1140 * This may look bad at first, but pretty much the same thing
1141 * happens anyway when any interrupt comes in, IDE or otherwise
1142 * -- the kernel masks the IRQ while it is being handled.
1144 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1145 disable_irq_nosync(hwif->irq);
1146 spin_unlock(&ide_lock);
1147 local_irq_enable_in_hardirq();
1148 /* allow other IRQs while we start this request */
1149 startstop = start_request(drive, rq);
1150 spin_lock_irq(&ide_lock);
1151 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1152 enable_irq(hwif->irq);
1153 if (startstop == ide_stopped)
1159 * Passes the stuff to ide_do_request
1161 void do_ide_request(struct request_queue *q)
1163 ide_drive_t *drive = q->queuedata;
1165 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1169 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1170 * retry the current request in pio mode instead of risking tossing it
1173 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1175 ide_hwif_t *hwif = HWIF(drive);
1177 ide_startstop_t ret = ide_stopped;
1180 * end current dma transaction
1184 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1185 (void)hwif->dma_ops->dma_end(drive);
1186 ret = ide_error(drive, "dma timeout error",
1187 hwif->tp_ops->read_status(hwif));
1189 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1190 hwif->dma_ops->dma_timeout(drive);
1194 * disable dma for now, but remember that we did so because of
1195 * a timeout -- we'll reenable after we finish this next request
1196 * (or rather the first chunk of it) in pio.
1199 drive->state = DMA_PIO_RETRY;
1200 ide_dma_off_quietly(drive);
1203 * un-busy drive etc (hwgroup->busy is cleared on return) and
1204 * make sure request is sane
1206 rq = HWGROUP(drive)->rq;
1211 HWGROUP(drive)->rq = NULL;
1218 rq->sector = rq->bio->bi_sector;
1219 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1220 rq->hard_cur_sectors = rq->current_nr_sectors;
1221 rq->buffer = bio_data(rq->bio);
1227 * ide_timer_expiry - handle lack of an IDE interrupt
1228 * @data: timer callback magic (hwgroup)
1230 * An IDE command has timed out before the expected drive return
1231 * occurred. At this point we attempt to clean up the current
1232 * mess. If the current handler includes an expiry handler then
1233 * we invoke the expiry handler, and providing it is happy the
1234 * work is done. If that fails we apply generic recovery rules
1235 * invoking the handler and checking the drive DMA status. We
1236 * have an excessively incestuous relationship with the DMA
1237 * logic that wants cleaning up.
1240 void ide_timer_expiry (unsigned long data)
1242 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1243 ide_handler_t *handler;
1244 ide_expiry_t *expiry;
1245 unsigned long flags;
1246 unsigned long wait = -1;
1248 spin_lock_irqsave(&ide_lock, flags);
1250 if (((handler = hwgroup->handler) == NULL) ||
1251 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1253 * Either a marginal timeout occurred
1254 * (got the interrupt just as timer expired),
1255 * or we were "sleeping" to give other devices a chance.
1256 * Either way, we don't really want to complain about anything.
1258 if (hwgroup->sleeping) {
1259 hwgroup->sleeping = 0;
1263 ide_drive_t *drive = hwgroup->drive;
1265 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1266 hwgroup->handler = NULL;
1269 ide_startstop_t startstop = ide_stopped;
1270 if (!hwgroup->busy) {
1271 hwgroup->busy = 1; /* paranoia */
1272 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1274 if ((expiry = hwgroup->expiry) != NULL) {
1276 if ((wait = expiry(drive)) > 0) {
1278 hwgroup->timer.expires = jiffies + wait;
1279 hwgroup->req_gen_timer = hwgroup->req_gen;
1280 add_timer(&hwgroup->timer);
1281 spin_unlock_irqrestore(&ide_lock, flags);
1285 hwgroup->handler = NULL;
1287 * We need to simulate a real interrupt when invoking
1288 * the handler() function, which means we need to
1289 * globally mask the specific IRQ:
1291 spin_unlock(&ide_lock);
1293 /* disable_irq_nosync ?? */
1294 disable_irq(hwif->irq);
1296 * as if we were handling an interrupt */
1297 local_irq_disable();
1298 if (hwgroup->polling) {
1299 startstop = handler(drive);
1300 } else if (drive_is_ready(drive)) {
1301 if (drive->waiting_for_dma)
1302 hwif->dma_ops->dma_lost_irq(drive);
1303 (void)ide_ack_intr(hwif);
1304 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1305 startstop = handler(drive);
1307 if (drive->waiting_for_dma) {
1308 startstop = ide_dma_timeout_retry(drive, wait);
1311 ide_error(drive, "irq timeout",
1312 hwif->tp_ops->read_status(hwif));
1314 drive->service_time = jiffies - drive->service_start;
1315 spin_lock_irq(&ide_lock);
1316 enable_irq(hwif->irq);
1317 if (startstop == ide_stopped)
1321 ide_do_request(hwgroup, IDE_NO_IRQ);
1322 spin_unlock_irqrestore(&ide_lock, flags);
1326 * unexpected_intr - handle an unexpected IDE interrupt
1327 * @irq: interrupt line
1328 * @hwgroup: hwgroup being processed
1330 * There's nothing really useful we can do with an unexpected interrupt,
1331 * other than reading the status register (to clear it), and logging it.
1332 * There should be no way that an irq can happen before we're ready for it,
1333 * so we needn't worry much about losing an "important" interrupt here.
1335 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1336 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1337 * looks "good", we just ignore the interrupt completely.
1339 * This routine assumes __cli() is in effect when called.
1341 * If an unexpected interrupt happens on irq15 while we are handling irq14
1342 * and if the two interfaces are "serialized" (CMD640), then it looks like
1343 * we could screw up by interfering with a new request being set up for
1346 * In reality, this is a non-issue. The new command is not sent unless
1347 * the drive is ready to accept one, in which case we know the drive is
1348 * not trying to interrupt us. And ide_set_handler() is always invoked
1349 * before completing the issuance of any new drive command, so we will not
1350 * be accidentally invoked as a result of any valid command completion
1353 * Note that we must walk the entire hwgroup here. We know which hwif
1354 * is doing the current command, but we don't know which hwif burped
1358 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1361 ide_hwif_t *hwif = hwgroup->hwif;
1364 * handle the unexpected interrupt
1367 if (hwif->irq == irq) {
1368 stat = hwif->tp_ops->read_status(hwif);
1370 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1371 /* Try to not flood the console with msgs */
1372 static unsigned long last_msgtime, count;
1374 if (time_after(jiffies, last_msgtime + HZ)) {
1375 last_msgtime = jiffies;
1376 printk(KERN_ERR "%s%s: unexpected interrupt, "
1377 "status=0x%02x, count=%ld\n",
1379 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1383 } while ((hwif = hwif->next) != hwgroup->hwif);
1387 * ide_intr - default IDE interrupt handler
1388 * @irq: interrupt number
1389 * @dev_id: hwif group
1390 * @regs: unused weirdness from the kernel irq layer
1392 * This is the default IRQ handler for the IDE layer. You should
1393 * not need to override it. If you do be aware it is subtle in
1396 * hwgroup->hwif is the interface in the group currently performing
1397 * a command. hwgroup->drive is the drive and hwgroup->handler is
1398 * the IRQ handler to call. As we issue a command the handlers
1399 * step through multiple states, reassigning the handler to the
1400 * next step in the process. Unlike a smart SCSI controller IDE
1401 * expects the main processor to sequence the various transfer
1402 * stages. We also manage a poll timer to catch up with most
1403 * timeout situations. There are still a few where the handlers
1404 * don't ever decide to give up.
1406 * The handler eventually returns ide_stopped to indicate the
1407 * request completed. At this point we issue the next request
1408 * on the hwgroup and the process begins again.
1411 irqreturn_t ide_intr (int irq, void *dev_id)
1413 unsigned long flags;
1414 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1417 ide_handler_t *handler;
1418 ide_startstop_t startstop;
1420 spin_lock_irqsave(&ide_lock, flags);
1421 hwif = hwgroup->hwif;
1423 if (!ide_ack_intr(hwif)) {
1424 spin_unlock_irqrestore(&ide_lock, flags);
1428 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1430 * Not expecting an interrupt from this drive.
1431 * That means this could be:
1432 * (1) an interrupt from another PCI device
1433 * sharing the same PCI INT# as us.
1434 * or (2) a drive just entered sleep or standby mode,
1435 * and is interrupting to let us know.
1436 * or (3) a spurious interrupt of unknown origin.
1438 * For PCI, we cannot tell the difference,
1439 * so in that case we just ignore it and hope it goes away.
1441 * FIXME: unexpected_intr should be hwif-> then we can
1442 * remove all the ifdef PCI crap
1444 #ifdef CONFIG_BLK_DEV_IDEPCI
1445 if (hwif->chipset != ide_pci)
1446 #endif /* CONFIG_BLK_DEV_IDEPCI */
1449 * Probably not a shared PCI interrupt,
1450 * so we can safely try to do something about it:
1452 unexpected_intr(irq, hwgroup);
1453 #ifdef CONFIG_BLK_DEV_IDEPCI
1456 * Whack the status register, just in case
1457 * we have a leftover pending IRQ.
1459 (void)hwif->tp_ops->read_status(hwif);
1460 #endif /* CONFIG_BLK_DEV_IDEPCI */
1462 spin_unlock_irqrestore(&ide_lock, flags);
1465 drive = hwgroup->drive;
1468 * This should NEVER happen, and there isn't much
1469 * we could do about it here.
1471 * [Note - this can occur if the drive is hot unplugged]
1473 spin_unlock_irqrestore(&ide_lock, flags);
1476 if (!drive_is_ready(drive)) {
1478 * This happens regularly when we share a PCI IRQ with
1479 * another device. Unfortunately, it can also happen
1480 * with some buggy drives that trigger the IRQ before
1481 * their status register is up to date. Hopefully we have
1482 * enough advance overhead that the latter isn't a problem.
1484 spin_unlock_irqrestore(&ide_lock, flags);
1487 if (!hwgroup->busy) {
1488 hwgroup->busy = 1; /* paranoia */
1489 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1491 hwgroup->handler = NULL;
1493 del_timer(&hwgroup->timer);
1494 spin_unlock(&ide_lock);
1496 /* Some controllers might set DMA INTR no matter DMA or PIO;
1497 * bmdma status might need to be cleared even for
1498 * PIO interrupts to prevent spurious/lost irq.
1500 if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1501 /* ide_dma_end() needs bmdma status for error checking.
1502 * So, skip clearing bmdma status here and leave it
1503 * to ide_dma_end() if this is dma interrupt.
1505 hwif->ide_dma_clear_irq(drive);
1507 if (drive->dev_flags & IDE_DFLAG_UNMASK)
1508 local_irq_enable_in_hardirq();
1509 /* service this interrupt, may set handler for next interrupt */
1510 startstop = handler(drive);
1511 spin_lock_irq(&ide_lock);
1514 * Note that handler() may have set things up for another
1515 * interrupt to occur soon, but it cannot happen until
1516 * we exit from this routine, because it will be the
1517 * same irq as is currently being serviced here, and Linux
1518 * won't allow another of the same (on any CPU) until we return.
1520 drive->service_time = jiffies - drive->service_start;
1521 if (startstop == ide_stopped) {
1522 if (hwgroup->handler == NULL) { /* paranoia */
1524 ide_do_request(hwgroup, hwif->irq);
1526 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1527 "on exit\n", drive->name);
1530 spin_unlock_irqrestore(&ide_lock, flags);
1535 * ide_do_drive_cmd - issue IDE special command
1536 * @drive: device to issue command
1537 * @rq: request to issue
1539 * This function issues a special IDE device request
1540 * onto the request queue.
1542 * the rq is queued at the head of the request queue, displacing
1543 * the currently-being-processed request and this function
1544 * returns immediately without waiting for the new rq to be
1545 * completed. This is VERY DANGEROUS, and is intended for
1546 * careful use by the ATAPI tape/cdrom driver code.
1549 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1551 unsigned long flags;
1552 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1554 spin_lock_irqsave(&ide_lock, flags);
1556 __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1557 __generic_unplug_device(drive->queue);
1558 spin_unlock_irqrestore(&ide_lock, flags);
1561 EXPORT_SYMBOL(ide_do_drive_cmd);
1563 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1565 ide_hwif_t *hwif = drive->hwif;
1568 memset(&task, 0, sizeof(task));
1569 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1570 IDE_TFLAG_OUT_FEATURE | tf_flags;
1571 task.tf.feature = dma; /* Use PIO/DMA */
1572 task.tf.lbam = bcount & 0xff;
1573 task.tf.lbah = (bcount >> 8) & 0xff;
1575 ide_tf_dump(drive->name, &task.tf);
1576 hwif->tp_ops->set_irq(hwif, 1);
1577 SELECT_MASK(drive, 0);
1578 hwif->tp_ops->tf_load(drive, &task);
1581 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1583 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1585 ide_hwif_t *hwif = drive->hwif;
1590 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1592 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1596 EXPORT_SYMBOL_GPL(ide_pad_transfer);