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/config.h>
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/string.h>
31 #include <linux/kernel.h>
32 #include <linux/timer.h>
34 #include <linux/interrupt.h>
35 #include <linux/major.h>
36 #include <linux/errno.h>
37 #include <linux/genhd.h>
38 #include <linux/blkpg.h>
39 #include <linux/slab.h>
40 #include <linux/init.h>
41 #include <linux/pci.h>
42 #include <linux/delay.h>
43 #include <linux/ide.h>
44 #include <linux/devfs_fs_kernel.h>
45 #include <linux/completion.h>
46 #include <linux/reboot.h>
47 #include <linux/cdrom.h>
48 #include <linux/seq_file.h>
49 #include <linux/kmod.h>
51 #include <asm/byteorder.h>
53 #include <asm/uaccess.h>
55 #include <asm/bitops.h>
57 #include "ide_modes.h"
59 #if (DISK_RECOVERY_TIME > 0)
61 Error So the User Has To Fix the Compilation And Stop Hacking Port 0x43
62 Does anyone ever use this anyway ??
65 * For really screwy hardware (hey, at least it *can* be used with Linux)
66 * we can enforce a minimum delay time between successive operations.
68 static unsigned long read_timer (ide_hwif_t *hwif)
70 unsigned long t, flags;
73 /* FIXME this is completely unsafe! */
74 local_irq_save(flags);
78 i |= inb_p(0x40) << 8;
79 local_irq_restore(flags);
82 #endif /* DISK_RECOVERY_TIME */
84 static inline void set_recovery_timer (ide_hwif_t *hwif)
86 #if (DISK_RECOVERY_TIME > 0)
87 hwif->last_time = read_timer(hwif);
88 #endif /* DISK_RECOVERY_TIME */
92 * ide_end_request - complete an IDE I/O
93 * @drive: IDE device for the I/O
96 * This is our end_request wrapper function. We complete the I/O
97 * update random number input and dequeue the request.
100 int ide_end_request (ide_drive_t *drive, int uptodate)
106 spin_lock_irqsave(&io_request_lock, flags);
107 rq = HWGROUP(drive)->rq;
110 * decide whether to reenable DMA -- 3 is a random magic for now,
111 * if we DMA timeout more than 3 times, just stay in PIO
113 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
115 HWGROUP(drive)->hwif->ide_dma_on(drive);
118 if (!end_that_request_first(rq, uptodate, drive->name)) {
119 add_blkdev_randomness(MAJOR(rq->rq_dev));
120 blkdev_dequeue_request(rq);
121 HWGROUP(drive)->rq = NULL;
122 end_that_request_last(rq);
126 spin_unlock_irqrestore(&io_request_lock, flags);
130 EXPORT_SYMBOL(ide_end_request);
133 * ide_end_drive_cmd - end an explicit drive command
138 * Clean up after success/failure of an explicit drive command.
139 * These get thrown onto the queue so they are synchronized with
140 * real I/O operations on the drive.
142 * In LBA48 mode we have to read the register set twice to get
143 * all the extra information out.
146 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
148 ide_hwif_t *hwif = HWIF(drive);
152 spin_lock_irqsave(&io_request_lock, flags);
153 rq = HWGROUP(drive)->rq;
154 spin_unlock_irqrestore(&io_request_lock, flags);
159 u8 *args = (u8 *) rq->buffer;
161 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
166 args[2] = hwif->INB(IDE_NSECTOR_REG);
172 u8 *args = (u8 *) rq->buffer;
174 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
179 args[2] = hwif->INB(IDE_NSECTOR_REG);
180 args[3] = hwif->INB(IDE_SECTOR_REG);
181 args[4] = hwif->INB(IDE_LCYL_REG);
182 args[5] = hwif->INB(IDE_HCYL_REG);
183 args[6] = hwif->INB(IDE_SELECT_REG);
187 case IDE_DRIVE_TASKFILE:
189 ide_task_t *args = (ide_task_t *) rq->special;
191 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
194 if (args->tf_in_flags.b.data) {
195 u16 data = hwif->INW(IDE_DATA_REG);
196 args->tfRegister[IDE_DATA_OFFSET] = (data) & 0xFF;
197 args->hobRegister[IDE_DATA_OFFSET_HOB] = (data >> 8) & 0xFF;
199 args->tfRegister[IDE_ERROR_OFFSET] = err;
200 args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
201 args->tfRegister[IDE_SECTOR_OFFSET] = hwif->INB(IDE_SECTOR_REG);
202 args->tfRegister[IDE_LCYL_OFFSET] = hwif->INB(IDE_LCYL_REG);
203 args->tfRegister[IDE_HCYL_OFFSET] = hwif->INB(IDE_HCYL_REG);
204 args->tfRegister[IDE_SELECT_OFFSET] = hwif->INB(IDE_SELECT_REG);
205 args->tfRegister[IDE_STATUS_OFFSET] = stat;
207 if (drive->addressing == 1) {
208 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG_HOB);
209 args->hobRegister[IDE_FEATURE_OFFSET_HOB] = hwif->INB(IDE_FEATURE_REG);
210 args->hobRegister[IDE_NSECTOR_OFFSET_HOB] = hwif->INB(IDE_NSECTOR_REG);
211 args->hobRegister[IDE_SECTOR_OFFSET_HOB] = hwif->INB(IDE_SECTOR_REG);
212 args->hobRegister[IDE_LCYL_OFFSET_HOB] = hwif->INB(IDE_LCYL_REG);
213 args->hobRegister[IDE_HCYL_OFFSET_HOB] = hwif->INB(IDE_HCYL_REG);
221 spin_lock_irqsave(&io_request_lock, flags);
222 blkdev_dequeue_request(rq);
223 HWGROUP(drive)->rq = NULL;
224 end_that_request_last(rq);
225 spin_unlock_irqrestore(&io_request_lock, flags);
228 EXPORT_SYMBOL(ide_end_drive_cmd);
231 * try_to_flush_leftover_data - flush junk
232 * @drive: drive to flush
234 * try_to_flush_leftover_data() is invoked in response to a drive
235 * unexpectedly having its DRQ_STAT bit set. As an alternative to
236 * resetting the drive, this routine tries to clear the condition
237 * by read a sector's worth of data from the drive. Of course,
238 * this may not help if the drive is *waiting* for data from *us*.
240 void try_to_flush_leftover_data (ide_drive_t *drive)
242 int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
244 if (drive->media != ide_disk)
248 u32 wcount = (i > 16) ? 16 : i;
251 HWIF(drive)->ata_input_data(drive, buffer, wcount);
255 EXPORT_SYMBOL(try_to_flush_leftover_data);
258 * FIXME Add an ATAPI error
262 * ide_error - handle an error on the IDE
263 * @drive: drive the error occurred on
264 * @msg: message to report
267 * ide_error() takes action based on the error returned by the drive.
268 * For normal I/O that may well include retries. We deal with
269 * both new-style (taskfile) and old style command handling here.
270 * In the case of taskfile command handling there is work left to
274 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
280 err = ide_dump_status(drive, msg, stat);
281 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
285 /* retry only "normal" I/O: */
286 if (rq->cmd == IDE_DRIVE_CMD || rq->cmd == IDE_DRIVE_TASK) {
288 ide_end_drive_cmd(drive, stat, err);
291 if (rq->cmd == IDE_DRIVE_TASKFILE) {
293 ide_end_drive_cmd(drive, stat, err);
294 // ide_end_taskfile(drive, stat, err);
298 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
299 /* other bits are useless when BUSY */
300 rq->errors |= ERROR_RESET;
302 if (drive->media != ide_disk)
305 if (stat & ERR_STAT) {
306 /* err has different meaning on cdrom and tape */
307 if (err == ABRT_ERR) {
308 if (drive->select.b.lba &&
309 (hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY))
310 /* some newer drives don't
311 * support WIN_SPECIFY
314 } else if ((err & BAD_CRC) == BAD_CRC) {
316 /* UDMA crc error -- just retry the operation */
317 } else if (err & (BBD_ERR | ECC_ERR)) {
318 /* retries won't help these */
319 rq->errors = ERROR_MAX;
320 } else if (err & TRK0_ERR) {
321 /* help it find track zero */
322 rq->errors |= ERROR_RECAL;
326 if ((stat & DRQ_STAT) && rq->cmd != WRITE)
327 try_to_flush_leftover_data(drive);
329 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) {
331 hwif->OUTB(WIN_IDLEIMMEDIATE,IDE_COMMAND_REG);
333 if (rq->errors >= ERROR_MAX) {
334 if (drive->driver != NULL)
335 DRIVER(drive)->end_request(drive, 0);
337 ide_end_request(drive, 0);
339 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
341 return ide_do_reset(drive);
343 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
344 drive->special.b.recalibrate = 1;
350 EXPORT_SYMBOL(ide_error);
353 * ide_cmd - issue a simple drive command
354 * @drive: drive the command is for
356 * @nsect: sector byte
357 * @handler: handler for the command completion
359 * Issue a simple drive command with interrupts.
360 * The drive must be selected beforehand.
363 void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, ide_handler_t *handler)
365 ide_hwif_t *hwif = HWIF(drive);
366 if (HWGROUP(drive)->handler != NULL)
368 ide_set_handler(drive, handler, WAIT_CMD, NULL);
370 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
371 SELECT_MASK(drive,0);
372 hwif->OUTB(nsect,IDE_NSECTOR_REG);
373 hwif->OUTB(cmd,IDE_COMMAND_REG);
376 EXPORT_SYMBOL(ide_cmd);
379 * drive_cmd_intr - drive command completion interrupt
380 * @drive: drive the completion interrupt occurred on
382 * drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
383 * We do any neccessary daya reading and then wait for the drive to
384 * go non busy. At that point we may read the error data and complete
388 ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
390 struct request *rq = HWGROUP(drive)->rq;
391 ide_hwif_t *hwif = HWIF(drive);
392 u8 *args = (u8 *) rq->buffer;
393 u8 stat = hwif->INB(IDE_STATUS_REG);
397 if ((stat & DRQ_STAT) && args && args[3]) {
398 u8 io_32bit = drive->io_32bit;
400 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
401 drive->io_32bit = io_32bit;
402 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
406 if (!OK_STAT(stat, READY_STAT, BAD_STAT))
407 return DRIVER(drive)->error(drive, "drive_cmd", stat);
408 /* calls ide_end_drive_cmd */
409 ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
413 EXPORT_SYMBOL(drive_cmd_intr);
416 * do_special - issue some special commands
417 * @drive: drive the command is for
419 * do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
420 * commands to a drive. It used to do much more, but has been scaled
424 ide_startstop_t do_special (ide_drive_t *drive)
426 special_t *s = &drive->special;
429 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
433 if (HWIF(drive)->tuneproc != NULL)
434 HWIF(drive)->tuneproc(drive, drive->tune_req);
435 } else if (drive->driver != NULL) {
436 return DRIVER(drive)->special(drive);
438 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, s->all);
444 EXPORT_SYMBOL(do_special);
447 * execute_drive_command - issue special drive command
448 * @drive: the drive to issue th command on
449 * @rq: the request structure holding the command
451 * execute_drive_cmd() issues a special drive command, usually
452 * initiated by ioctl() from the external hdparm program. The
453 * command can be a drive command, drive task or taskfile
454 * operation. Weirdly you can call it with NULL to wait for
455 * all commands to finish. Don't do this as that is due to change
458 ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
460 ide_hwif_t *hwif = HWIF(drive);
462 case IDE_DRIVE_TASKFILE:
464 ide_task_t *args = rq->special;
468 if (args->tf_out_flags.all != 0)
469 return flagged_taskfile(drive, args);
470 return do_rw_taskfile(drive, args);
474 u8 *args = rq->buffer;
479 printk("%s: DRIVE_TASK_CMD ", drive->name);
480 printk("cmd=0x%02x ", args[0]);
481 printk("fr=0x%02x ", args[1]);
482 printk("ns=0x%02x ", args[2]);
483 printk("sc=0x%02x ", args[3]);
484 printk("lcyl=0x%02x ", args[4]);
485 printk("hcyl=0x%02x ", args[5]);
486 printk("sel=0x%02x\n", args[6]);
488 hwif->OUTB(args[1], IDE_FEATURE_REG);
489 hwif->OUTB(args[3], IDE_SECTOR_REG);
490 hwif->OUTB(args[4], IDE_LCYL_REG);
491 hwif->OUTB(args[5], IDE_HCYL_REG);
492 sel = (args[6] & ~0x10);
493 if (drive->select.b.unit)
495 hwif->OUTB(sel, IDE_SELECT_REG);
496 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
501 u8 *args = rq->buffer;
505 printk("%s: DRIVE_CMD ", drive->name);
506 printk("cmd=0x%02x ", args[0]);
507 printk("sc=0x%02x ", args[1]);
508 printk("fr=0x%02x ", args[2]);
509 printk("xx=0x%02x\n", args[3]);
511 if (args[0] == WIN_SMART) {
512 hwif->OUTB(0x4f, IDE_LCYL_REG);
513 hwif->OUTB(0xc2, IDE_HCYL_REG);
514 hwif->OUTB(args[2],IDE_FEATURE_REG);
515 hwif->OUTB(args[1],IDE_SECTOR_REG);
516 ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
519 hwif->OUTB(args[2],IDE_FEATURE_REG);
520 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
527 * NULL is actually a valid way of waiting for
528 * all current requests to be flushed from the queue.
531 printk("%s: DRIVE_CMD (null)\n", drive->name);
533 ide_end_drive_cmd(drive,
534 hwif->INB(IDE_STATUS_REG),
535 hwif->INB(IDE_ERROR_REG));
539 EXPORT_SYMBOL(execute_drive_cmd);
542 * start_request - start of I/O and command issuing for IDE
544 * start_request() initiates handling of a new I/O request. It
545 * accepts commands and I/O (read/write) requests. It also does
546 * the final remapping for weird stuff like EZDrive. Once
547 * device mapper can work sector level the EZDrive stuff can go away
549 * FIXME: this function needs a rename
552 ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
554 ide_startstop_t startstop;
555 unsigned long block, blockend;
556 unsigned int minor = MINOR(rq->rq_dev), unit = minor >> PARTN_BITS;
557 ide_hwif_t *hwif = HWIF(drive);
560 printk("%s: start_request: current=0x%08lx\n",
561 hwif->name, (unsigned long) rq);
564 /* bail early if we've exceeded max_failures */
565 if (drive->max_failures && (drive->failures > drive->max_failures)) {
570 * bail early if we've sent a device to sleep, however how to wake
571 * this needs to be a masked flag. FIXME for proper operations.
573 if (drive->suspend_reset) {
577 if (unit >= MAX_DRIVES) {
578 printk(KERN_ERR "%s: bad device number: %s\n",
579 hwif->name, kdevname(rq->rq_dev));
583 if (rq->bh && !buffer_locked(rq->bh)) {
584 printk(KERN_ERR "%s: block not locked\n", drive->name);
589 blockend = block + rq->nr_sectors;
591 if (blk_fs_request(rq) &&
592 (drive->media == ide_disk || drive->media == ide_floppy)) {
593 if ((blockend < block) || (blockend > drive->part[minor&PARTN_MASK].nr_sects)) {
594 printk(KERN_ERR "%s%c: bad access: block=%ld, count=%ld\n", drive->name,
595 (minor&PARTN_MASK)?'0'+(minor&PARTN_MASK):' ', block, rq->nr_sectors);
598 block += drive->part[minor&PARTN_MASK].start_sect + drive->sect0;
600 /* Yecch - this will shift the entire interval,
601 possibly killing some innocent following sector */
602 if (block == 0 && drive->remap_0_to_1 == 1)
603 block = 1; /* redirect MBR access to EZ-Drive partn table */
605 #if (DISK_RECOVERY_TIME > 0)
606 while ((read_timer() - hwif->last_time) < DISK_RECOVERY_TIME);
610 if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
611 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
614 if (!drive->special.all) {
618 return execute_drive_cmd(drive, rq);
619 case IDE_DRIVE_TASKFILE:
620 return execute_drive_cmd(drive, rq);
624 if (drive->driver != NULL) {
625 return (DRIVER(drive)->do_request(drive, rq, block));
627 printk(KERN_WARNING "%s: media type %d not supported\n", drive->name, drive->media);
630 return do_special(drive);
632 if (drive->driver != NULL)
633 DRIVER(drive)->end_request(drive, 0);
635 ide_end_request(drive, 0);
639 EXPORT_SYMBOL(start_request);
641 int restart_request (ide_drive_t *drive, struct request *rq)
643 (void) start_request(drive, rq);
647 EXPORT_SYMBOL(restart_request);
650 * ide_stall_queue - pause an IDE device
651 * @drive: drive to stall
652 * @timeout: time to stall for (jiffies)
654 * ide_stall_queue() can be used by a drive to give excess bandwidth back
655 * to the hwgroup by sleeping for timeout jiffies.
658 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
660 if (timeout > WAIT_WORSTCASE)
661 timeout = WAIT_WORSTCASE;
662 drive->sleep = timeout + jiffies;
665 EXPORT_SYMBOL(ide_stall_queue);
667 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
670 * choose_drive - select a drive to service
671 * @hwgroup: hardware group to select on
673 * choose_drive() selects the next drive which will be serviced.
674 * This is neccessary because the IDE layer can't issue commands
675 * to both drives on the same cable, unlike SCSI.
678 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
680 ide_drive_t *drive, *best;
684 drive = hwgroup->drive;
686 if (!blk_queue_empty(&drive->queue) && (!drive->sleep || time_after_eq(jiffies, drive->sleep))) {
688 || (drive->sleep && (!best->sleep || 0 < (signed long)(best->sleep - drive->sleep)))
689 || (!best->sleep && 0 < (signed long)(WAKEUP(best) - WAKEUP(drive))))
691 if (!blk_queue_plugged(&drive->queue))
695 } while ((drive = drive->next) != hwgroup->drive);
696 if (best && best->nice1 && !best->sleep && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
697 long t = (signed long)(WAKEUP(best) - jiffies);
698 if (t >= WAIT_MIN_SLEEP) {
700 * We *may* have some time to spare, but first let's see if
701 * someone can potentially benefit from our nice mood today..
706 && 0 < (signed long)(WAKEUP(drive) - (jiffies - best->service_time))
707 && 0 < (signed long)((jiffies + t) - WAKEUP(drive)))
709 ide_stall_queue(best, IDE_MIN(t, 10 * WAIT_MIN_SLEEP));
712 } while ((drive = drive->next) != best);
719 * Issue a new request to a drive from hwgroup
720 * Caller must have already done spin_lock_irqsave(&io_request_lock, ..);
722 * A hwgroup is a serialized group of IDE interfaces. Usually there is
723 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
724 * may have both interfaces in a single hwgroup to "serialize" access.
725 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
726 * together into one hwgroup for serialized access.
728 * Note also that several hwgroups can end up sharing a single IRQ,
729 * possibly along with many other devices. This is especially common in
730 * PCI-based systems with off-board IDE controller cards.
732 * The IDE driver uses the single global io_request_lock spinlock to protect
733 * access to the request queues, and to protect the hwgroup->busy flag.
735 * The first thread into the driver for a particular hwgroup sets the
736 * hwgroup->busy flag to indicate that this hwgroup is now active,
737 * and then initiates processing of the top request from the request queue.
739 * Other threads attempting entry notice the busy setting, and will simply
740 * queue their new requests and exit immediately. Note that hwgroup->busy
741 * remains set even when the driver is merely awaiting the next interrupt.
742 * Thus, the meaning is "this hwgroup is busy processing a request".
744 * When processing of a request completes, the completing thread or IRQ-handler
745 * will start the next request from the queue. If no more work remains,
746 * the driver will clear the hwgroup->busy flag and exit.
748 * The io_request_lock (spinlock) is used to protect all access to the
749 * hwgroup->busy flag, but is otherwise not needed for most processing in
750 * the driver. This makes the driver much more friendlier to shared IRQs
751 * than previous designs, while remaining 100% (?) SMP safe and capable.
753 /* --BenH: made non-static as ide-pmac.c uses it to kick the hwgroup back
754 * into life on wakeup from machine sleep.
756 void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
761 ide_startstop_t startstop;
763 /* for atari only: POSSIBLY BROKEN HERE(?) */
764 ide_get_lock(&ide_intr_lock, ide_intr, hwgroup);
766 /* necessary paranoia: ensure IRQs are masked on local CPU */
769 while (!hwgroup->busy) {
771 drive = choose_drive(hwgroup);
773 unsigned long sleep = 0;
775 drive = hwgroup->drive;
777 if (drive->sleep && (!sleep || 0 < (signed long)(sleep - drive->sleep)))
778 sleep = drive->sleep;
779 } while ((drive = drive->next) != hwgroup->drive);
782 * Take a short snooze, and then wake up this hwgroup again.
783 * This gives other hwgroups on the same a chance to
784 * play fairly with us, just in case there are big differences
785 * in relative throughputs.. don't want to hog the cpu too much.
787 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
788 sleep = jiffies + WAIT_MIN_SLEEP;
790 if (timer_pending(&hwgroup->timer))
791 printk(KERN_ERR "ide_set_handler: timer already active\n");
793 /* so that ide_timer_expiry knows what to do */
794 hwgroup->sleeping = 1;
795 mod_timer(&hwgroup->timer, sleep);
796 /* we purposely leave hwgroup->busy==1
799 /* Ugly, but how can we sleep for the lock
800 * otherwise? perhaps from tq_disk?
804 ide_release_lock(&ide_intr_lock);
807 /* no more work for this hwgroup (for now) */
811 if (hwgroup->hwif->sharing_irq &&
812 hwif != hwgroup->hwif &&
813 hwif->io_ports[IDE_CONTROL_OFFSET]) {
814 /* set nIEN for previous hwif */
815 SELECT_INTERRUPT(drive);
817 hwgroup->hwif = hwif;
818 hwgroup->drive = drive;
820 drive->service_start = jiffies;
823 if (blk_queue_plugged(&drive->queue))
824 printk(KERN_ERR "%s: Huh? nuking plugged queue\n", drive->name);
826 rq = blkdev_entry_next_request(&drive->queue.queue_head);
829 * Some systems have trouble with IDE IRQs arriving while
830 * the driver is still setting things up. So, here we disable
831 * the IRQ used by this interface while the request is being started.
832 * This may look bad at first, but pretty much the same thing
833 * happens anyway when any interrupt comes in, IDE or otherwise
834 * -- the kernel masks the IRQ while it is being handled.
836 if (masked_irq && hwif->irq != masked_irq)
837 disable_irq_nosync(hwif->irq);
838 spin_unlock(&io_request_lock);
840 /* allow other IRQs while we start this request */
841 startstop = start_request(drive, rq);
842 spin_lock_irq(&io_request_lock);
843 if (masked_irq && hwif->irq != masked_irq)
844 enable_irq(hwif->irq);
845 if (startstop == ide_stopped)
850 EXPORT_SYMBOL(ide_do_request);
853 * ide_get_queue() returns the queue which corresponds to a given device.
855 request_queue_t *ide_get_queue (kdev_t dev)
857 ide_hwif_t *hwif = (ide_hwif_t *)blk_dev[MAJOR(dev)].data;
859 return &hwif->drives[DEVICE_NR(dev) & 1].queue;
862 EXPORT_SYMBOL(ide_get_queue);
865 * Passes the stuff to ide_do_request
867 void do_ide_request(request_queue_t *q)
869 ide_do_request(q->queuedata, 0);
873 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
874 * retry the current request in pio mode instead of risking tossing it
877 void ide_dma_timeout_retry(ide_drive_t *drive)
879 ide_hwif_t *hwif = HWIF(drive);
883 * end current dma transaction
885 (void) hwif->ide_dma_end(drive);
888 * complain a little, later we might remove some of this verbosity
890 printk(KERN_ERR "%s: timeout waiting for DMA\n", drive->name);
891 (void) hwif->ide_dma_timeout(drive);
894 * disable dma for now, but remember that we did so because of
895 * a timeout -- we'll reenable after we finish this next request
896 * (or rather the first chunk of it) in pio.
899 drive->state = DMA_PIO_RETRY;
900 (void) hwif->ide_dma_off_quietly(drive);
903 * un-busy drive etc (hwgroup->busy is cleared on return) and
904 * make sure request is sane
906 rq = HWGROUP(drive)->rq;
907 HWGROUP(drive)->rq = NULL;
910 rq->sector = rq->bh->b_rsector;
911 rq->current_nr_sectors = rq->bh->b_size >> 9;
912 rq->hard_cur_sectors = rq->current_nr_sectors;
913 rq->buffer = rq->bh->b_data;
916 EXPORT_SYMBOL(ide_dma_timeout_retry);
919 * ide_timer_expiry - handle lack of an IDE interrupt
920 * @data: timer callback magic (hwgroup)
922 * An IDE command has timed out before the expected drive return
923 * occurred. At this point we attempt to clean up the current
924 * mess. If the current handler includes an expiry handler then
925 * we invoke the expiry handler, and providing it is happy the
926 * work is done. If that fails we apply generic recovery rules
927 * invoking the handler and checking the drive DMA status. We
928 * have an excessively incestuous relationship with the DMA
929 * logic that wants cleaning up.
932 void ide_timer_expiry (unsigned long data)
934 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
935 ide_handler_t *handler;
936 ide_expiry_t *expiry;
940 spin_lock_irqsave(&io_request_lock, flags);
941 del_timer(&hwgroup->timer);
943 if ((handler = hwgroup->handler) == NULL) {
945 * Either a marginal timeout occurred
946 * (got the interrupt just as timer expired),
947 * or we were "sleeping" to give other devices a chance.
948 * Either way, we don't really want to complain about anything.
950 if (hwgroup->sleeping) {
951 hwgroup->sleeping = 0;
955 ide_drive_t *drive = hwgroup->drive;
957 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
958 hwgroup->handler = NULL;
961 ide_startstop_t startstop = ide_stopped;
962 if (!hwgroup->busy) {
963 hwgroup->busy = 1; /* paranoia */
964 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
966 if ((expiry = hwgroup->expiry) != NULL) {
968 if ((wait = expiry(drive)) != 0) {
970 hwgroup->timer.expires = jiffies + wait;
971 add_timer(&hwgroup->timer);
972 spin_unlock_irqrestore(&io_request_lock, flags);
976 hwgroup->handler = NULL;
978 * We need to simulate a real interrupt when invoking
979 * the handler() function, which means we need to
980 * globally mask the specific IRQ:
982 spin_unlock(&io_request_lock);
984 #if DISABLE_IRQ_NOSYNC
985 disable_irq_nosync(hwif->irq);
987 /* disable_irq_nosync ?? */
988 disable_irq(hwif->irq);
989 #endif /* DISABLE_IRQ_NOSYNC */
992 * as if we were handling an interrupt */
994 if (hwgroup->poll_timeout != 0) {
995 startstop = handler(drive);
996 } else if (drive_is_ready(drive)) {
997 if (drive->waiting_for_dma)
998 (void) hwgroup->hwif->ide_dma_lostirq(drive);
999 (void)ide_ack_intr(hwif);
1000 printk(KERN_ERR "%s: lost interrupt\n", drive->name);
1001 startstop = handler(drive);
1003 if (drive->waiting_for_dma) {
1004 startstop = ide_stopped;
1005 ide_dma_timeout_retry(drive);
1007 startstop = DRIVER(drive)->error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1009 set_recovery_timer(hwif);
1010 drive->service_time = jiffies - drive->service_start;
1011 enable_irq(hwif->irq);
1012 spin_lock_irq(&io_request_lock);
1013 if (startstop == ide_stopped)
1017 ide_do_request(hwgroup, 0);
1018 spin_unlock_irqrestore(&io_request_lock, flags);
1021 EXPORT_SYMBOL(ide_timer_expiry);
1024 * unexpected_intr - handle an unexpected IDE interrupt
1025 * @irq: interrupt line
1026 * @hwgroup: hwgroup being processed
1028 * There's nothing really useful we can do with an unexpected interrupt,
1029 * other than reading the status register (to clear it), and logging it.
1030 * There should be no way that an irq can happen before we're ready for it,
1031 * so we needn't worry much about losing an "important" interrupt here.
1033 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1034 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1035 * looks "good", we just ignore the interrupt completely.
1037 * This routine assumes __cli() is in effect when called.
1039 * If an unexpected interrupt happens on irq15 while we are handling irq14
1040 * and if the two interfaces are "serialized" (CMD640), then it looks like
1041 * we could screw up by interfering with a new request being set up for
1044 * In reality, this is a non-issue. The new command is not sent unless
1045 * the drive is ready to accept one, in which case we know the drive is
1046 * not trying to interrupt us. And ide_set_handler() is always invoked
1047 * before completing the issuance of any new drive command, so we will not
1048 * be accidentally invoked as a result of any valid command completion
1051 * Note that we must walk the entire hwgroup here. We know which hwif
1052 * is doing the current command, but we don't know which hwif burped
1056 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1059 ide_hwif_t *hwif = hwgroup->hwif;
1062 * handle the unexpected interrupt
1065 if (hwif->irq == irq) {
1066 stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1067 if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1068 /* Try to not flood the console with msgs */
1069 static unsigned long last_msgtime, count;
1071 if (time_after(jiffies, last_msgtime + HZ)) {
1072 last_msgtime = jiffies;
1073 printk(KERN_ERR "%s%s: unexpected interrupt, "
1074 "status=0x%02x, count=%ld\n",
1076 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1080 } while ((hwif = hwif->next) != hwgroup->hwif);
1084 * ide_intr - default IDE interrupt handler
1085 * @irq: interrupt number
1086 * @dev_id: hwif group
1087 * @regs: unused weirdness from the kernel irq layer
1089 * This is the default IRQ handler for the IDE layer. You should
1090 * not need to override it. If you do be aware it is subtle in
1093 * hwgroup->hwif is the interface in the group currently performing
1094 * a command. hwgroup->drive is the drive and hwgroup->handler is
1095 * the IRQ handler to call. As we issue a command the handlers
1096 * step through multiple states, reassigning the handler to the
1097 * next step in the process. Unlike a smart SCSI controller IDE
1098 * expects the main processor to sequence the various transfer
1099 * stages. We also manage a poll timer to catch up with most
1100 * timeout situations. There are still a few where the handlers
1101 * don't ever decide to give up.
1103 * The handler eventually returns ide_stopped to indicate the
1104 * request completed. At this point we issue the next request
1105 * on the hwgroup and the process begins again.
1108 void ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1110 unsigned long flags;
1111 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1114 ide_handler_t *handler;
1115 ide_startstop_t startstop;
1117 spin_lock_irqsave(&io_request_lock, flags);
1118 hwif = hwgroup->hwif;
1120 if (!ide_ack_intr(hwif)) {
1121 spin_unlock_irqrestore(&io_request_lock, flags);
1125 if ((handler = hwgroup->handler) == NULL ||
1126 hwgroup->poll_timeout != 0) {
1128 * Not expecting an interrupt from this drive.
1129 * That means this could be:
1130 * (1) an interrupt from another PCI device
1131 * sharing the same PCI INT# as us.
1132 * or (2) a drive just entered sleep or standby mode,
1133 * and is interrupting to let us know.
1134 * or (3) a spurious interrupt of unknown origin.
1136 * For PCI, we cannot tell the difference,
1137 * so in that case we just ignore it and hope it goes away.
1139 #ifdef CONFIG_BLK_DEV_IDEPCI
1140 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1141 #endif /* CONFIG_BLK_DEV_IDEPCI */
1144 * Probably not a shared PCI interrupt,
1145 * so we can safely try to do something about it:
1147 unexpected_intr(irq, hwgroup);
1148 #ifdef CONFIG_BLK_DEV_IDEPCI
1151 * Whack the status register, just in case
1152 * we have a leftover pending IRQ.
1154 (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1155 #endif /* CONFIG_BLK_DEV_IDEPCI */
1157 spin_unlock_irqrestore(&io_request_lock, flags);
1160 drive = hwgroup->drive;
1163 * This should NEVER happen, and there isn't much
1164 * we could do about it here.
1166 spin_unlock_irqrestore(&io_request_lock, flags);
1169 if (!drive_is_ready(drive)) {
1171 * This happens regularly when we share a PCI IRQ with
1172 * another device. Unfortunately, it can also happen
1173 * with some buggy drives that trigger the IRQ before
1174 * their status register is up to date. Hopefully we have
1175 * enough advance overhead that the latter isn't a problem.
1177 spin_unlock_irqrestore(&io_request_lock, flags);
1180 if (!hwgroup->busy) {
1181 hwgroup->busy = 1; /* paranoia */
1182 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1184 hwgroup->handler = NULL;
1185 del_timer(&hwgroup->timer);
1186 spin_unlock(&io_request_lock);
1191 /* service this interrupt, may set handler for next interrupt */
1192 startstop = handler(drive);
1193 spin_lock_irq(&io_request_lock);
1196 * Note that handler() may have set things up for another
1197 * interrupt to occur soon, but it cannot happen until
1198 * we exit from this routine, because it will be the
1199 * same irq as is currently being serviced here, and Linux
1200 * won't allow another of the same (on any CPU) until we return.
1202 set_recovery_timer(HWIF(drive));
1203 drive->service_time = jiffies - drive->service_start;
1204 if (startstop == ide_stopped) {
1205 if (hwgroup->handler == NULL) { /* paranoia */
1207 ide_do_request(hwgroup, hwif->irq);
1209 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1210 "on exit\n", drive->name);
1213 spin_unlock_irqrestore(&io_request_lock, flags);
1216 EXPORT_SYMBOL(ide_intr);
1219 * get_info_ptr() returns the (ide_drive_t *) for a given device number.
1220 * It returns NULL if the given device number does not match any present drives.
1222 ide_drive_t *get_info_ptr (kdev_t i_rdev)
1224 int major = MAJOR(i_rdev);
1227 for (h = 0; h < MAX_HWIFS; ++h) {
1228 ide_hwif_t *hwif = &ide_hwifs[h];
1229 if (hwif->present && major == hwif->major) {
1230 unsigned unit = DEVICE_NR(i_rdev);
1231 if (unit < MAX_DRIVES) {
1232 ide_drive_t *drive = &hwif->drives[unit];
1242 EXPORT_SYMBOL(get_info_ptr);
1245 * ide_init_drive_cmd - initialize a drive command request
1246 * @rq: request object
1248 * Initialize a request before we fill it in and send it down to
1249 * ide_do_drive_cmd. Commands must be set up by this function. Right
1250 * now it doesn't do a lot, but if that changes abusers will have a
1254 void ide_init_drive_cmd (struct request *rq)
1256 memset(rq, 0, sizeof(*rq));
1257 rq->cmd = IDE_DRIVE_CMD;
1260 EXPORT_SYMBOL(ide_init_drive_cmd);
1263 * ide_do_drive_cmd - issue IDE special command
1264 * @drive: device to issue command
1265 * @rq: request to issue
1266 * @action: action for processing
1268 * This function issues a special IDE device request
1269 * onto the request queue.
1271 * If action is ide_wait, then the rq is queued at the end of the
1272 * request queue, and the function sleeps until it has been processed.
1273 * This is for use when invoked from an ioctl handler.
1275 * If action is ide_preempt, then the rq is queued at the head of
1276 * the request queue, displacing the currently-being-processed
1277 * request and this function returns immediately without waiting
1278 * for the new rq to be completed. This is VERY DANGEROUS, and is
1279 * intended for careful use by the ATAPI tape/cdrom driver code.
1281 * If action is ide_next, then the rq is queued immediately after
1282 * the currently-being-processed-request (if any), and the function
1283 * returns without waiting for the new rq to be completed. As above,
1284 * This is VERY DANGEROUS, and is intended for careful use by the
1285 * ATAPI tape/cdrom driver code.
1287 * If action is ide_end, then the rq is queued at the end of the
1288 * request queue, and the function returns immediately without waiting
1289 * for the new rq to be completed. This is again intended for careful
1290 * use by the ATAPI tape/cdrom driver code.
1293 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1295 unsigned long flags;
1296 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1297 unsigned int major = HWIF(drive)->major;
1298 request_queue_t *q = &drive->queue;
1299 struct list_head *queue_head = &q->queue_head;
1300 DECLARE_COMPLETION(wait);
1302 #ifdef CONFIG_BLK_DEV_PDC4030
1303 if (HWIF(drive)->chipset == ide_pdc4030 && rq->buffer != NULL)
1304 return -ENOSYS; /* special drive cmds not supported */
1307 rq->rq_status = RQ_ACTIVE;
1308 rq->rq_dev = MKDEV(major,(drive->select.b.unit)<<PARTN_BITS);
1309 if (action == ide_wait)
1310 rq->waiting = &wait;
1311 spin_lock_irqsave(&io_request_lock, flags);
1312 if (blk_queue_empty(q) || action == ide_preempt) {
1313 if (action == ide_preempt)
1316 if (action == ide_wait || action == ide_end) {
1317 queue_head = queue_head->prev;
1319 queue_head = queue_head->next;
1321 list_add(&rq->queue, queue_head);
1322 ide_do_request(hwgroup, 0);
1323 spin_unlock_irqrestore(&io_request_lock, flags);
1324 if (action == ide_wait) {
1325 /* wait for it to be serviced */
1326 wait_for_completion(&wait);
1327 /* return -EIO if errors */
1328 return rq->errors ? -EIO : 0;
1334 EXPORT_SYMBOL(ide_do_drive_cmd);