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/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>
51 #include <asm/byteorder.h>
53 #include <asm/uaccess.h>
55 #include <asm/bitops.h>
58 * ide_end_request - complete an IDE I/O
59 * @drive: IDE device for the I/O
61 * @nr_sectors: number of sectors completed
63 * This is our end_request wrapper function. We complete the I/O
64 * update random number input and dequeue the request, which if
65 * it was tagged may be out of order.
68 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
74 spin_lock_irqsave(&ide_lock, flags);
75 rq = HWGROUP(drive)->rq;
77 BUG_ON(!(rq->flags & REQ_STARTED));
80 nr_sectors = rq->hard_cur_sectors;
83 * if failfast is set on a request, override number of sectors and
84 * complete the whole request right now
86 if (blk_noretry_request(rq) && !uptodate)
87 nr_sectors = rq->hard_nr_sectors;
90 * decide whether to reenable DMA -- 3 is a random magic for now,
91 * if we DMA timeout more than 3 times, just stay in PIO
93 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
95 HWGROUP(drive)->hwif->ide_dma_on(drive);
98 if (!end_that_request_first(rq, uptodate, nr_sectors)) {
99 add_disk_randomness(rq->rq_disk);
100 if (!blk_rq_tagged(rq))
101 blkdev_dequeue_request(rq);
103 blk_queue_end_tag(drive->queue, rq);
104 HWGROUP(drive)->rq = NULL;
105 end_that_request_last(rq);
108 spin_unlock_irqrestore(&ide_lock, flags);
112 EXPORT_SYMBOL(ide_end_request);
115 * ide_complete_pm_request - end the current Power Management request
116 * @drive: target drive
119 * This function cleans up the current PM request and stops the queue
122 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
127 printk("%s: completing PM request, %s\n", drive->name,
128 blk_pm_suspend_request(rq) ? "suspend" : "resume");
130 spin_lock_irqsave(&ide_lock, flags);
131 if (blk_pm_suspend_request(rq)) {
132 blk_stop_queue(drive->queue);
135 blk_start_queue(drive->queue);
137 blkdev_dequeue_request(rq);
138 HWGROUP(drive)->rq = NULL;
139 end_that_request_last(rq);
140 spin_unlock_irqrestore(&ide_lock, flags);
144 * ide_end_drive_cmd - end an explicit drive command
149 * Clean up after success/failure of an explicit drive command.
150 * These get thrown onto the queue so they are synchronized with
151 * real I/O operations on the drive.
153 * In LBA48 mode we have to read the register set twice to get
154 * all the extra information out.
157 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
159 ide_hwif_t *hwif = HWIF(drive);
163 spin_lock_irqsave(&ide_lock, flags);
164 rq = HWGROUP(drive)->rq;
165 spin_unlock_irqrestore(&ide_lock, flags);
167 if (rq->flags & REQ_DRIVE_CMD) {
168 u8 *args = (u8 *) rq->buffer;
170 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
175 args[2] = hwif->INB(IDE_NSECTOR_REG);
177 } else if (rq->flags & REQ_DRIVE_TASK) {
178 u8 *args = (u8 *) rq->buffer;
180 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
185 args[2] = hwif->INB(IDE_NSECTOR_REG);
186 args[3] = hwif->INB(IDE_SECTOR_REG);
187 args[4] = hwif->INB(IDE_LCYL_REG);
188 args[5] = hwif->INB(IDE_HCYL_REG);
189 args[6] = hwif->INB(IDE_SELECT_REG);
191 } else if (rq->flags & REQ_DRIVE_TASKFILE) {
192 ide_task_t *args = (ide_task_t *) rq->special;
194 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
197 if (args->tf_in_flags.b.data) {
198 u16 data = hwif->INW(IDE_DATA_REG);
199 args->tfRegister[IDE_DATA_OFFSET] = (data) & 0xFF;
200 args->hobRegister[IDE_DATA_OFFSET_HOB] = (data >> 8) & 0xFF;
202 args->tfRegister[IDE_ERROR_OFFSET] = err;
203 args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
204 args->tfRegister[IDE_SECTOR_OFFSET] = hwif->INB(IDE_SECTOR_REG);
205 args->tfRegister[IDE_LCYL_OFFSET] = hwif->INB(IDE_LCYL_REG);
206 args->tfRegister[IDE_HCYL_OFFSET] = hwif->INB(IDE_HCYL_REG);
207 args->tfRegister[IDE_SELECT_OFFSET] = hwif->INB(IDE_SELECT_REG);
208 args->tfRegister[IDE_STATUS_OFFSET] = stat;
210 if (drive->addressing == 1) {
211 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG_HOB);
212 args->hobRegister[IDE_FEATURE_OFFSET_HOB] = hwif->INB(IDE_FEATURE_REG);
213 args->hobRegister[IDE_NSECTOR_OFFSET_HOB] = hwif->INB(IDE_NSECTOR_REG);
214 args->hobRegister[IDE_SECTOR_OFFSET_HOB] = hwif->INB(IDE_SECTOR_REG);
215 args->hobRegister[IDE_LCYL_OFFSET_HOB] = hwif->INB(IDE_LCYL_REG);
216 args->hobRegister[IDE_HCYL_OFFSET_HOB] = hwif->INB(IDE_HCYL_REG);
219 } else if (blk_pm_request(rq)) {
221 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
222 drive->name, rq->pm->pm_step, stat, err);
224 DRIVER(drive)->complete_power_step(drive, rq, stat, err);
225 if (rq->pm->pm_step == ide_pm_state_completed)
226 ide_complete_pm_request(drive, rq);
230 spin_lock_irqsave(&ide_lock, flags);
231 blkdev_dequeue_request(rq);
232 HWGROUP(drive)->rq = NULL;
233 end_that_request_last(rq);
234 spin_unlock_irqrestore(&ide_lock, flags);
237 EXPORT_SYMBOL(ide_end_drive_cmd);
240 * try_to_flush_leftover_data - flush junk
241 * @drive: drive to flush
243 * try_to_flush_leftover_data() is invoked in response to a drive
244 * unexpectedly having its DRQ_STAT bit set. As an alternative to
245 * resetting the drive, this routine tries to clear the condition
246 * by read a sector's worth of data from the drive. Of course,
247 * this may not help if the drive is *waiting* for data from *us*.
249 void try_to_flush_leftover_data (ide_drive_t *drive)
251 int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
253 if (drive->media != ide_disk)
257 u32 wcount = (i > 16) ? 16 : i;
260 HWIF(drive)->ata_input_data(drive, buffer, wcount);
264 EXPORT_SYMBOL(try_to_flush_leftover_data);
267 * FIXME Add an ATAPI error
271 * ide_error - handle an error on the IDE
272 * @drive: drive the error occurred on
273 * @msg: message to report
276 * ide_error() takes action based on the error returned by the drive.
277 * For normal I/O that may well include retries. We deal with
278 * both new-style (taskfile) and old style command handling here.
279 * In the case of taskfile command handling there is work left to
283 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
289 err = ide_dump_status(drive, msg, stat);
290 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
294 /* retry only "normal" I/O: */
295 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
297 ide_end_drive_cmd(drive, stat, err);
300 if (rq->flags & REQ_DRIVE_TASKFILE) {
302 ide_end_drive_cmd(drive, stat, err);
303 // ide_end_taskfile(drive, stat, err);
307 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
308 /* other bits are useless when BUSY */
309 rq->errors |= ERROR_RESET;
311 if (drive->media != ide_disk)
314 if (stat & ERR_STAT) {
315 /* err has different meaning on cdrom and tape */
316 if (err == ABRT_ERR) {
317 if (drive->select.b.lba &&
318 (hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY))
319 /* some newer drives don't
320 * support WIN_SPECIFY
323 } else if ((err & BAD_CRC) == BAD_CRC) {
325 /* UDMA crc error -- just retry the operation */
326 } else if (err & (BBD_ERR | ECC_ERR)) {
327 /* retries won't help these */
328 rq->errors = ERROR_MAX;
329 } else if (err & TRK0_ERR) {
330 /* help it find track zero */
331 rq->errors |= ERROR_RECAL;
335 if ((stat & DRQ_STAT) && rq_data_dir(rq) != WRITE)
336 try_to_flush_leftover_data(drive);
338 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) {
340 hwif->OUTB(WIN_IDLEIMMEDIATE,IDE_COMMAND_REG);
342 if (rq->errors >= ERROR_MAX) {
343 DRIVER(drive)->end_request(drive, 0, 0);
345 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
347 return ide_do_reset(drive);
349 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
350 drive->special.b.recalibrate = 1;
356 EXPORT_SYMBOL(ide_error);
359 * ide_abort - abort pending IDE operatins
360 * @drive: drive the error occurred on
361 * @msg: message to report
363 * ide_abort kills and cleans up when we are about to do a
364 * host initiated reset on active commands. Longer term we
365 * want handlers to have sensible abort handling themselves
367 * This differs fundamentally from ide_error because in
368 * this case the command is doing just fine when we
372 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
377 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
381 /* retry only "normal" I/O: */
382 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
384 ide_end_drive_cmd(drive, BUSY_STAT, 0);
387 if (rq->flags & REQ_DRIVE_TASKFILE) {
389 ide_end_drive_cmd(drive, BUSY_STAT, 0);
390 // ide_end_taskfile(drive, BUSY_STAT, 0);
394 rq->errors |= ERROR_RESET;
395 DRIVER(drive)->end_request(drive, 0, 0);
399 EXPORT_SYMBOL(ide_abort);
402 * ide_cmd - issue a simple drive command
403 * @drive: drive the command is for
405 * @nsect: sector byte
406 * @handler: handler for the command completion
408 * Issue a simple drive command with interrupts.
409 * The drive must be selected beforehand.
412 void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, ide_handler_t *handler)
414 ide_hwif_t *hwif = HWIF(drive);
416 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
417 SELECT_MASK(drive,0);
418 hwif->OUTB(nsect,IDE_NSECTOR_REG);
419 ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
422 EXPORT_SYMBOL(ide_cmd);
425 * drive_cmd_intr - drive command completion interrupt
426 * @drive: drive the completion interrupt occurred on
428 * drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
429 * We do any necessary daya reading and then wait for the drive to
430 * go non busy. At that point we may read the error data and complete
434 ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
436 struct request *rq = HWGROUP(drive)->rq;
437 ide_hwif_t *hwif = HWIF(drive);
438 u8 *args = (u8 *) rq->buffer;
439 u8 stat = hwif->INB(IDE_STATUS_REG);
443 if ((stat & DRQ_STAT) && args && args[3]) {
444 u8 io_32bit = drive->io_32bit;
446 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
447 drive->io_32bit = io_32bit;
448 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
452 if (!OK_STAT(stat, READY_STAT, BAD_STAT) && DRIVER(drive) != NULL)
453 return DRIVER(drive)->error(drive, "drive_cmd", stat);
454 /* calls ide_end_drive_cmd */
455 ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
459 EXPORT_SYMBOL(drive_cmd_intr);
462 * do_special - issue some special commands
463 * @drive: drive the command is for
465 * do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
466 * commands to a drive. It used to do much more, but has been scaled
470 ide_startstop_t do_special (ide_drive_t *drive)
472 special_t *s = &drive->special;
475 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
479 if (HWIF(drive)->tuneproc != NULL)
480 HWIF(drive)->tuneproc(drive, drive->tune_req);
484 return DRIVER(drive)->special(drive);
487 EXPORT_SYMBOL(do_special);
490 * execute_drive_command - issue special drive command
491 * @drive: the drive to issue th command on
492 * @rq: the request structure holding the command
494 * execute_drive_cmd() issues a special drive command, usually
495 * initiated by ioctl() from the external hdparm program. The
496 * command can be a drive command, drive task or taskfile
497 * operation. Weirdly you can call it with NULL to wait for
498 * all commands to finish. Don't do this as that is due to change
501 ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
503 ide_hwif_t *hwif = HWIF(drive);
504 if (rq->flags & REQ_DRIVE_TASKFILE) {
505 ide_task_t *args = rq->special;
510 if (args->tf_out_flags.all != 0)
511 return flagged_taskfile(drive, args);
512 return do_rw_taskfile(drive, args);
513 } else if (rq->flags & REQ_DRIVE_TASK) {
514 u8 *args = rq->buffer;
520 printk("%s: DRIVE_TASK_CMD ", drive->name);
521 printk("cmd=0x%02x ", args[0]);
522 printk("fr=0x%02x ", args[1]);
523 printk("ns=0x%02x ", args[2]);
524 printk("sc=0x%02x ", args[3]);
525 printk("lcyl=0x%02x ", args[4]);
526 printk("hcyl=0x%02x ", args[5]);
527 printk("sel=0x%02x\n", args[6]);
529 hwif->OUTB(args[1], IDE_FEATURE_REG);
530 hwif->OUTB(args[3], IDE_SECTOR_REG);
531 hwif->OUTB(args[4], IDE_LCYL_REG);
532 hwif->OUTB(args[5], IDE_HCYL_REG);
533 sel = (args[6] & ~0x10);
534 if (drive->select.b.unit)
536 hwif->OUTB(sel, IDE_SELECT_REG);
537 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
539 } else if (rq->flags & REQ_DRIVE_CMD) {
540 u8 *args = rq->buffer;
545 printk("%s: DRIVE_CMD ", drive->name);
546 printk("cmd=0x%02x ", args[0]);
547 printk("sc=0x%02x ", args[1]);
548 printk("fr=0x%02x ", args[2]);
549 printk("xx=0x%02x\n", args[3]);
551 if (args[0] == WIN_SMART) {
552 hwif->OUTB(0x4f, IDE_LCYL_REG);
553 hwif->OUTB(0xc2, IDE_HCYL_REG);
554 hwif->OUTB(args[2],IDE_FEATURE_REG);
555 hwif->OUTB(args[1],IDE_SECTOR_REG);
556 ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
559 hwif->OUTB(args[2],IDE_FEATURE_REG);
560 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
566 * NULL is actually a valid way of waiting for
567 * all current requests to be flushed from the queue.
570 printk("%s: DRIVE_CMD (null)\n", drive->name);
572 ide_end_drive_cmd(drive,
573 hwif->INB(IDE_STATUS_REG),
574 hwif->INB(IDE_ERROR_REG));
578 EXPORT_SYMBOL(execute_drive_cmd);
581 * start_request - start of I/O and command issuing for IDE
583 * start_request() initiates handling of a new I/O request. It
584 * accepts commands and I/O (read/write) requests. It also does
585 * the final remapping for weird stuff like EZDrive. Once
586 * device mapper can work sector level the EZDrive stuff can go away
588 * FIXME: this function needs a rename
591 ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
593 ide_startstop_t startstop;
596 BUG_ON(!(rq->flags & REQ_STARTED));
599 printk("%s: start_request: current=0x%08lx\n",
600 HWIF(drive)->name, (unsigned long) rq);
603 /* bail early if we've exceeded max_failures */
604 if (drive->max_failures && (drive->failures > drive->max_failures)) {
609 * bail early if we've sent a device to sleep, however how to wake
610 * this needs to be a masked flag. FIXME for proper operations.
612 if (drive->suspend_reset)
616 if (blk_fs_request(rq) &&
617 (drive->media == ide_disk || drive->media == ide_floppy)) {
618 block += drive->sect0;
620 /* Yecch - this will shift the entire interval,
621 possibly killing some innocent following sector */
622 if (block == 0 && drive->remap_0_to_1 == 1)
623 block = 1; /* redirect MBR access to EZ-Drive partn table */
625 if (blk_pm_suspend_request(rq) &&
626 rq->pm->pm_step == ide_pm_state_start_suspend)
627 /* Mark drive blocked when starting the suspend sequence. */
629 else if (blk_pm_resume_request(rq) &&
630 rq->pm->pm_step == ide_pm_state_start_resume) {
632 * The first thing we do on wakeup is to wait for BSY bit to
633 * go away (with a looong timeout) as a drive on this hwif may
634 * just be POSTing itself.
635 * We do that before even selecting as the "other" device on
636 * the bus may be broken enough to walk on our toes at this
641 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
643 rc = ide_wait_not_busy(HWIF(drive), 35000);
645 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
647 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
648 rc = ide_wait_not_busy(HWIF(drive), 10000);
650 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
654 if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
655 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
658 if (!drive->special.all) {
659 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
660 return execute_drive_cmd(drive, rq);
661 else if (rq->flags & REQ_DRIVE_TASKFILE)
662 return execute_drive_cmd(drive, rq);
663 else if (blk_pm_request(rq)) {
665 printk("%s: start_power_step(step: %d)\n",
666 drive->name, rq->pm->pm_step);
668 startstop = DRIVER(drive)->start_power_step(drive, rq);
669 if (startstop == ide_stopped &&
670 rq->pm->pm_step == ide_pm_state_completed)
671 ide_complete_pm_request(drive, rq);
674 return (DRIVER(drive)->do_request(drive, rq, block));
676 return do_special(drive);
678 DRIVER(drive)->end_request(drive, 0, 0);
682 EXPORT_SYMBOL(start_request);
685 * ide_stall_queue - pause an IDE device
686 * @drive: drive to stall
687 * @timeout: time to stall for (jiffies)
689 * ide_stall_queue() can be used by a drive to give excess bandwidth back
690 * to the hwgroup by sleeping for timeout jiffies.
693 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
695 if (timeout > WAIT_WORSTCASE)
696 timeout = WAIT_WORSTCASE;
697 drive->sleep = timeout + jiffies;
700 EXPORT_SYMBOL(ide_stall_queue);
702 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
705 * choose_drive - select a drive to service
706 * @hwgroup: hardware group to select on
708 * choose_drive() selects the next drive which will be serviced.
709 * This is necessary because the IDE layer can't issue commands
710 * to both drives on the same cable, unlike SCSI.
713 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
715 ide_drive_t *drive, *best;
719 drive = hwgroup->drive;
721 if ((!drive->sleep || time_after_eq(jiffies, drive->sleep))
722 && !elv_queue_empty(drive->queue)) {
724 || (drive->sleep && (!best->sleep || 0 < (signed long)(best->sleep - drive->sleep)))
725 || (!best->sleep && 0 < (signed long)(WAKEUP(best) - WAKEUP(drive))))
727 if (!blk_queue_plugged(drive->queue))
731 } while ((drive = drive->next) != hwgroup->drive);
732 if (best && best->nice1 && !best->sleep && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
733 long t = (signed long)(WAKEUP(best) - jiffies);
734 if (t >= WAIT_MIN_SLEEP) {
736 * We *may* have some time to spare, but first let's see if
737 * someone can potentially benefit from our nice mood today..
742 /* FIXME: use time_before */
743 && 0 < (signed long)(WAKEUP(drive) - (jiffies - best->service_time))
744 && 0 < (signed long)((jiffies + t) - WAKEUP(drive)))
746 ide_stall_queue(best, IDE_MIN(t, 10 * WAIT_MIN_SLEEP));
749 } while ((drive = drive->next) != best);
756 * Issue a new request to a drive from hwgroup
757 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
759 * A hwgroup is a serialized group of IDE interfaces. Usually there is
760 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
761 * may have both interfaces in a single hwgroup to "serialize" access.
762 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
763 * together into one hwgroup for serialized access.
765 * Note also that several hwgroups can end up sharing a single IRQ,
766 * possibly along with many other devices. This is especially common in
767 * PCI-based systems with off-board IDE controller cards.
769 * The IDE driver uses the single global ide_lock spinlock to protect
770 * access to the request queues, and to protect the hwgroup->busy flag.
772 * The first thread into the driver for a particular hwgroup sets the
773 * hwgroup->busy flag to indicate that this hwgroup is now active,
774 * and then initiates processing of the top request from the request queue.
776 * Other threads attempting entry notice the busy setting, and will simply
777 * queue their new requests and exit immediately. Note that hwgroup->busy
778 * remains set even when the driver is merely awaiting the next interrupt.
779 * Thus, the meaning is "this hwgroup is busy processing a request".
781 * When processing of a request completes, the completing thread or IRQ-handler
782 * will start the next request from the queue. If no more work remains,
783 * the driver will clear the hwgroup->busy flag and exit.
785 * The ide_lock (spinlock) is used to protect all access to the
786 * hwgroup->busy flag, but is otherwise not needed for most processing in
787 * the driver. This makes the driver much more friendlier to shared IRQs
788 * than previous designs, while remaining 100% (?) SMP safe and capable.
790 /* --BenH: made non-static as ide-pmac.c uses it to kick the hwgroup back
791 * into life on wakeup from machine sleep.
793 void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
798 ide_startstop_t startstop;
800 /* for atari only: POSSIBLY BROKEN HERE(?) */
801 ide_get_lock(ide_intr, hwgroup);
803 /* caller must own ide_lock */
804 BUG_ON(!irqs_disabled());
806 while (!hwgroup->busy) {
808 drive = choose_drive(hwgroup);
810 unsigned long sleep = 0;
812 drive = hwgroup->drive;
814 if (drive->sleep && (!sleep || 0 < (signed long)(sleep - drive->sleep)))
815 sleep = drive->sleep;
816 } while ((drive = drive->next) != hwgroup->drive);
819 * Take a short snooze, and then wake up this hwgroup again.
820 * This gives other hwgroups on the same a chance to
821 * play fairly with us, just in case there are big differences
822 * in relative throughputs.. don't want to hog the cpu too much.
824 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
825 sleep = jiffies + WAIT_MIN_SLEEP;
827 if (timer_pending(&hwgroup->timer))
828 printk(KERN_CRIT "ide_set_handler: timer already active\n");
830 /* so that ide_timer_expiry knows what to do */
831 hwgroup->sleeping = 1;
832 mod_timer(&hwgroup->timer, sleep);
833 /* we purposely leave hwgroup->busy==1
836 /* Ugly, but how can we sleep for the lock
837 * otherwise? perhaps from tq_disk?
845 /* no more work for this hwgroup (for now) */
849 if (hwgroup->hwif->sharing_irq &&
850 hwif != hwgroup->hwif &&
851 hwif->io_ports[IDE_CONTROL_OFFSET]) {
852 /* set nIEN for previous hwif */
853 SELECT_INTERRUPT(drive);
855 hwgroup->hwif = hwif;
856 hwgroup->drive = drive;
858 drive->service_start = jiffies;
861 if (!ata_can_queue(drive)) {
862 if (!ata_pending_commands(drive))
868 if (blk_queue_plugged(drive->queue)) {
869 if (drive->using_tcq)
872 printk(KERN_ERR "ide: huh? queue was plugged!\n");
877 * we know that the queue isn't empty, but this can happen
878 * if the q->prep_rq_fn() decides to kill a request
880 rq = elv_next_request(drive->queue);
882 hwgroup->busy = !!ata_pending_commands(drive);
887 * Sanity: don't accept a request that isn't a PM request
888 * if we are currently power managed. This is very important as
889 * blk_stop_queue() doesn't prevent the elv_next_request()
890 * above to return us whatever is in the queue. Since we call
891 * ide_do_request() ourselves, we end up taking requests while
892 * the queue is blocked...
894 * We let requests forced at head of queue with ide-preempt
895 * though. I hope that doesn't happen too much, hopefully not
896 * unless the subdriver triggers such a thing in its own PM
899 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
900 /* We clear busy, there should be no pending ATA command at this point. */
905 if (!rq->bio && ata_pending_commands(drive))
911 * Some systems have trouble with IDE IRQs arriving while
912 * the driver is still setting things up. So, here we disable
913 * the IRQ used by this interface while the request is being started.
914 * This may look bad at first, but pretty much the same thing
915 * happens anyway when any interrupt comes in, IDE or otherwise
916 * -- the kernel masks the IRQ while it is being handled.
918 if (hwif->irq != masked_irq)
919 disable_irq_nosync(hwif->irq);
920 spin_unlock(&ide_lock);
922 /* allow other IRQs while we start this request */
923 startstop = start_request(drive, rq);
924 spin_lock_irq(&ide_lock);
925 if (hwif->irq != masked_irq)
926 enable_irq(hwif->irq);
927 if (startstop == ide_released)
929 if (startstop == ide_stopped)
934 EXPORT_SYMBOL(ide_do_request);
937 * Passes the stuff to ide_do_request
939 void do_ide_request(request_queue_t *q)
941 ide_do_request(q->queuedata, IDE_NO_IRQ);
945 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
946 * retry the current request in pio mode instead of risking tossing it
949 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
951 ide_hwif_t *hwif = HWIF(drive);
953 ide_startstop_t ret = ide_stopped;
956 * end current dma transaction
960 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
961 (void)HWIF(drive)->ide_dma_end(drive);
962 ret = DRIVER(drive)->error(drive, "dma timeout error",
963 hwif->INB(IDE_STATUS_REG));
965 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
966 (void) hwif->ide_dma_timeout(drive);
970 * disable dma for now, but remember that we did so because of
971 * a timeout -- we'll reenable after we finish this next request
972 * (or rather the first chunk of it) in pio.
975 drive->state = DMA_PIO_RETRY;
976 (void) hwif->ide_dma_off_quietly(drive);
979 * un-busy drive etc (hwgroup->busy is cleared on return) and
980 * make sure request is sane
982 rq = HWGROUP(drive)->rq;
983 HWGROUP(drive)->rq = NULL;
986 rq->sector = rq->bio->bi_sector;
987 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
988 rq->hard_cur_sectors = rq->current_nr_sectors;
996 * ide_timer_expiry - handle lack of an IDE interrupt
997 * @data: timer callback magic (hwgroup)
999 * An IDE command has timed out before the expected drive return
1000 * occurred. At this point we attempt to clean up the current
1001 * mess. If the current handler includes an expiry handler then
1002 * we invoke the expiry handler, and providing it is happy the
1003 * work is done. If that fails we apply generic recovery rules
1004 * invoking the handler and checking the drive DMA status. We
1005 * have an excessively incestuous relationship with the DMA
1006 * logic that wants cleaning up.
1009 void ide_timer_expiry (unsigned long data)
1011 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1012 ide_handler_t *handler;
1013 ide_expiry_t *expiry;
1014 unsigned long flags;
1015 unsigned long wait = -1;
1017 spin_lock_irqsave(&ide_lock, flags);
1019 if ((handler = hwgroup->handler) == NULL) {
1021 * Either a marginal timeout occurred
1022 * (got the interrupt just as timer expired),
1023 * or we were "sleeping" to give other devices a chance.
1024 * Either way, we don't really want to complain about anything.
1026 if (hwgroup->sleeping) {
1027 hwgroup->sleeping = 0;
1031 ide_drive_t *drive = hwgroup->drive;
1033 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1034 hwgroup->handler = NULL;
1037 ide_startstop_t startstop = ide_stopped;
1038 if (!hwgroup->busy) {
1039 hwgroup->busy = 1; /* paranoia */
1040 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1042 if ((expiry = hwgroup->expiry) != NULL) {
1044 if ((wait = expiry(drive)) > 0) {
1046 hwgroup->timer.expires = jiffies + wait;
1047 add_timer(&hwgroup->timer);
1048 spin_unlock_irqrestore(&ide_lock, flags);
1052 hwgroup->handler = NULL;
1054 * We need to simulate a real interrupt when invoking
1055 * the handler() function, which means we need to
1056 * globally mask the specific IRQ:
1058 spin_unlock(&ide_lock);
1060 #if DISABLE_IRQ_NOSYNC
1061 disable_irq_nosync(hwif->irq);
1063 /* disable_irq_nosync ?? */
1064 disable_irq(hwif->irq);
1065 #endif /* DISABLE_IRQ_NOSYNC */
1067 * as if we were handling an interrupt */
1068 local_irq_disable();
1069 if (hwgroup->poll_timeout != 0) {
1070 startstop = handler(drive);
1071 } else if (drive_is_ready(drive)) {
1072 if (drive->waiting_for_dma)
1073 (void) hwgroup->hwif->ide_dma_lostirq(drive);
1074 (void)ide_ack_intr(hwif);
1075 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1076 startstop = handler(drive);
1078 if (drive->waiting_for_dma) {
1079 startstop = ide_dma_timeout_retry(drive, wait);
1082 DRIVER(drive)->error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1084 drive->service_time = jiffies - drive->service_start;
1085 spin_lock_irq(&ide_lock);
1086 enable_irq(hwif->irq);
1087 if (startstop == ide_stopped)
1091 ide_do_request(hwgroup, IDE_NO_IRQ);
1092 spin_unlock_irqrestore(&ide_lock, flags);
1095 EXPORT_SYMBOL(ide_timer_expiry);
1098 * unexpected_intr - handle an unexpected IDE interrupt
1099 * @irq: interrupt line
1100 * @hwgroup: hwgroup being processed
1102 * There's nothing really useful we can do with an unexpected interrupt,
1103 * other than reading the status register (to clear it), and logging it.
1104 * There should be no way that an irq can happen before we're ready for it,
1105 * so we needn't worry much about losing an "important" interrupt here.
1107 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1108 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1109 * looks "good", we just ignore the interrupt completely.
1111 * This routine assumes __cli() is in effect when called.
1113 * If an unexpected interrupt happens on irq15 while we are handling irq14
1114 * and if the two interfaces are "serialized" (CMD640), then it looks like
1115 * we could screw up by interfering with a new request being set up for
1118 * In reality, this is a non-issue. The new command is not sent unless
1119 * the drive is ready to accept one, in which case we know the drive is
1120 * not trying to interrupt us. And ide_set_handler() is always invoked
1121 * before completing the issuance of any new drive command, so we will not
1122 * be accidentally invoked as a result of any valid command completion
1125 * Note that we must walk the entire hwgroup here. We know which hwif
1126 * is doing the current command, but we don't know which hwif burped
1130 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1133 ide_hwif_t *hwif = hwgroup->hwif;
1136 * handle the unexpected interrupt
1139 if (hwif->irq == irq) {
1140 stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1141 if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1142 /* Try to not flood the console with msgs */
1143 static unsigned long last_msgtime, count;
1145 if (time_after(jiffies, last_msgtime + HZ)) {
1146 last_msgtime = jiffies;
1147 printk(KERN_ERR "%s%s: unexpected interrupt, "
1148 "status=0x%02x, count=%ld\n",
1150 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1154 } while ((hwif = hwif->next) != hwgroup->hwif);
1158 * ide_intr - default IDE interrupt handler
1159 * @irq: interrupt number
1160 * @dev_id: hwif group
1161 * @regs: unused weirdness from the kernel irq layer
1163 * This is the default IRQ handler for the IDE layer. You should
1164 * not need to override it. If you do be aware it is subtle in
1167 * hwgroup->hwif is the interface in the group currently performing
1168 * a command. hwgroup->drive is the drive and hwgroup->handler is
1169 * the IRQ handler to call. As we issue a command the handlers
1170 * step through multiple states, reassigning the handler to the
1171 * next step in the process. Unlike a smart SCSI controller IDE
1172 * expects the main processor to sequence the various transfer
1173 * stages. We also manage a poll timer to catch up with most
1174 * timeout situations. There are still a few where the handlers
1175 * don't ever decide to give up.
1177 * The handler eventually returns ide_stopped to indicate the
1178 * request completed. At this point we issue the next request
1179 * on the hwgroup and the process begins again.
1182 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1184 unsigned long flags;
1185 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1188 ide_handler_t *handler;
1189 ide_startstop_t startstop;
1191 spin_lock_irqsave(&ide_lock, flags);
1192 hwif = hwgroup->hwif;
1194 if (!ide_ack_intr(hwif)) {
1195 spin_unlock_irqrestore(&ide_lock, flags);
1199 if ((handler = hwgroup->handler) == NULL ||
1200 hwgroup->poll_timeout != 0) {
1202 * Not expecting an interrupt from this drive.
1203 * That means this could be:
1204 * (1) an interrupt from another PCI device
1205 * sharing the same PCI INT# as us.
1206 * or (2) a drive just entered sleep or standby mode,
1207 * and is interrupting to let us know.
1208 * or (3) a spurious interrupt of unknown origin.
1210 * For PCI, we cannot tell the difference,
1211 * so in that case we just ignore it and hope it goes away.
1213 * FIXME: unexpected_intr should be hwif-> then we can
1214 * remove all the ifdef PCI crap
1216 #ifdef CONFIG_BLK_DEV_IDEPCI
1217 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1218 #endif /* CONFIG_BLK_DEV_IDEPCI */
1221 * Probably not a shared PCI interrupt,
1222 * so we can safely try to do something about it:
1224 unexpected_intr(irq, hwgroup);
1225 #ifdef CONFIG_BLK_DEV_IDEPCI
1228 * Whack the status register, just in case
1229 * we have a leftover pending IRQ.
1231 (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1232 #endif /* CONFIG_BLK_DEV_IDEPCI */
1234 spin_unlock_irqrestore(&ide_lock, flags);
1237 drive = hwgroup->drive;
1240 * This should NEVER happen, and there isn't much
1241 * we could do about it here.
1243 * [Note - this can occur if the drive is hot unplugged]
1245 spin_unlock_irqrestore(&ide_lock, flags);
1248 if (!drive_is_ready(drive)) {
1250 * This happens regularly when we share a PCI IRQ with
1251 * another device. Unfortunately, it can also happen
1252 * with some buggy drives that trigger the IRQ before
1253 * their status register is up to date. Hopefully we have
1254 * enough advance overhead that the latter isn't a problem.
1256 spin_unlock_irqrestore(&ide_lock, flags);
1259 if (!hwgroup->busy) {
1260 hwgroup->busy = 1; /* paranoia */
1261 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1263 hwgroup->handler = NULL;
1264 del_timer(&hwgroup->timer);
1265 spin_unlock(&ide_lock);
1269 /* service this interrupt, may set handler for next interrupt */
1270 startstop = handler(drive);
1271 spin_lock_irq(&ide_lock);
1274 * Note that handler() may have set things up for another
1275 * interrupt to occur soon, but it cannot happen until
1276 * we exit from this routine, because it will be the
1277 * same irq as is currently being serviced here, and Linux
1278 * won't allow another of the same (on any CPU) until we return.
1280 drive->service_time = jiffies - drive->service_start;
1281 if (startstop == ide_stopped) {
1282 if (hwgroup->handler == NULL) { /* paranoia */
1284 ide_do_request(hwgroup, hwif->irq);
1286 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1287 "on exit\n", drive->name);
1290 spin_unlock_irqrestore(&ide_lock, flags);
1294 EXPORT_SYMBOL(ide_intr);
1297 * ide_init_drive_cmd - initialize a drive command request
1298 * @rq: request object
1300 * Initialize a request before we fill it in and send it down to
1301 * ide_do_drive_cmd. Commands must be set up by this function. Right
1302 * now it doesn't do a lot, but if that changes abusers will have a
1306 void ide_init_drive_cmd (struct request *rq)
1308 memset(rq, 0, sizeof(*rq));
1309 rq->flags = REQ_DRIVE_CMD;
1312 EXPORT_SYMBOL(ide_init_drive_cmd);
1315 * ide_do_drive_cmd - issue IDE special command
1316 * @drive: device to issue command
1317 * @rq: request to issue
1318 * @action: action for processing
1320 * This function issues a special IDE device request
1321 * onto the request queue.
1323 * If action is ide_wait, then the rq is queued at the end of the
1324 * request queue, and the function sleeps until it has been processed.
1325 * This is for use when invoked from an ioctl handler.
1327 * If action is ide_preempt, then the rq is queued at the head of
1328 * the request queue, displacing the currently-being-processed
1329 * request and this function returns immediately without waiting
1330 * for the new rq to be completed. This is VERY DANGEROUS, and is
1331 * intended for careful use by the ATAPI tape/cdrom driver code.
1333 * If action is ide_next, then the rq is queued immediately after
1334 * the currently-being-processed-request (if any), and the function
1335 * returns without waiting for the new rq to be completed. As above,
1336 * This is VERY DANGEROUS, and is intended for careful use by the
1337 * ATAPI tape/cdrom driver code.
1339 * If action is ide_end, then the rq is queued at the end of the
1340 * request queue, and the function returns immediately without waiting
1341 * for the new rq to be completed. This is again intended for careful
1342 * use by the ATAPI tape/cdrom driver code.
1345 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1347 unsigned long flags;
1348 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1349 DECLARE_COMPLETION(wait);
1350 int where = ELEVATOR_INSERT_BACK, err;
1351 int must_wait = (action == ide_wait || action == ide_head_wait);
1353 #ifdef CONFIG_BLK_DEV_PDC4030
1355 * FIXME: there should be a drive or hwif->special
1356 * handler that points here by default, not hacks
1357 * in the ide-io.c code
1359 * FIXME2: That code breaks power management if used with
1360 * this chipset, that really doesn't belong here !
1362 if (HWIF(drive)->chipset == ide_pdc4030 && rq->buffer != NULL)
1363 return -ENOSYS; /* special drive cmds not supported */
1366 rq->rq_status = RQ_ACTIVE;
1368 rq->rq_disk = drive->disk;
1371 * we need to hold an extra reference to request for safe inspection
1376 rq->waiting = &wait;
1379 spin_lock_irqsave(&ide_lock, flags);
1380 if (action == ide_preempt)
1382 if (action == ide_preempt || action == ide_head_wait) {
1383 where = ELEVATOR_INSERT_FRONT;
1384 rq->flags |= REQ_PREEMPT;
1386 __elv_add_request(drive->queue, rq, where, 0);
1387 ide_do_request(hwgroup, IDE_NO_IRQ);
1388 spin_unlock_irqrestore(&ide_lock, flags);
1392 wait_for_completion(&wait);
1396 blk_put_request(rq);
1402 EXPORT_SYMBOL(ide_do_drive_cmd);