1/*
2 *	IDE I/O functions
3 *
4 *	Basic PIO and command management functionality.
5 *
6 * This code was split off from ide.c. See ide.c for history and original
7 * copyrights.
8 *
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
12 * later version.
13 *
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.
18 *
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.
24 */
25
26
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>
32#include <linux/mm.h>
33#include <linux/interrupt.h>
34#include <linux/major.h>
35#include <linux/errno.h>
36#include <linux/genhd.h>
37#include <linux/blkpg.h>
38#include <linux/slab.h>
39#include <linux/init.h>
40#include <linux/pci.h>
41#include <linux/delay.h>
42#include <linux/ide.h>
43#include <linux/completion.h>
44#include <linux/reboot.h>
45#include <linux/cdrom.h>
46#include <linux/seq_file.h>
47#include <linux/device.h>
48#include <linux/kmod.h>
49#include <linux/scatterlist.h>
50
51#include <asm/byteorder.h>
52#include <asm/irq.h>
53#include <asm/uaccess.h>
54#include <asm/io.h>
55#include <asm/bitops.h>
56
57static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58			     int uptodate, int nr_sectors)
59{
60	int ret = 1;
61
62	/*
63	 * if failfast is set on a request, override number of sectors and
64	 * complete the whole request right now
65	 */
66	if (blk_noretry_request(rq) && end_io_error(uptodate))
67		nr_sectors = rq->hard_nr_sectors;
68
69	if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
70		rq->errors = -EIO;
71
72	/*
73	 * decide whether to reenable DMA -- 3 is a random magic for now,
74	 * if we DMA timeout more than 3 times, just stay in PIO
75	 */
76	if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
77		drive->state = 0;
78		HWGROUP(drive)->hwif->ide_dma_on(drive);
79	}
80
81	if (!end_that_request_first(rq, uptodate, nr_sectors)) {
82		add_disk_randomness(rq->rq_disk);
83		if (!list_empty(&rq->queuelist))
84			blkdev_dequeue_request(rq);
85		HWGROUP(drive)->rq = NULL;
86		end_that_request_last(rq, uptodate);
87		ret = 0;
88	}
89
90	return ret;
91}
92
93/**
94 *	ide_end_request		-	complete an IDE I/O
95 *	@drive: IDE device for the I/O
96 *	@uptodate:
97 *	@nr_sectors: number of sectors completed
98 *
99 *	This is our end_request wrapper function. We complete the I/O
100 *	update random number input and dequeue the request, which if
101 *	it was tagged may be out of order.
102 */
103
104int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
105{
106	struct request *rq;
107	unsigned long flags;
108	int ret = 1;
109
110	/*
111	 * room for locking improvements here, the calls below don't
112	 * need the queue lock held at all
113	 */
114	spin_lock_irqsave(&ide_lock, flags);
115	rq = HWGROUP(drive)->rq;
116
117	if (!nr_sectors)
118		nr_sectors = rq->hard_cur_sectors;
119
120	ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
121
122	spin_unlock_irqrestore(&ide_lock, flags);
123	return ret;
124}
125EXPORT_SYMBOL(ide_end_request);
126
127/*
128 * Power Management state machine. This one is rather trivial for now,
129 * we should probably add more, like switching back to PIO on suspend
130 * to help some BIOSes, re-do the door locking on resume, etc...
131 */
132
133enum {
134	ide_pm_flush_cache	= ide_pm_state_start_suspend,
135	idedisk_pm_standby,
136
137	idedisk_pm_restore_pio	= ide_pm_state_start_resume,
138	idedisk_pm_idle,
139	ide_pm_restore_dma,
140};
141
142static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
143{
144	struct request_pm_state *pm = rq->data;
145
146	if (drive->media != ide_disk)
147		return;
148
149	switch (pm->pm_step) {
150	case ide_pm_flush_cache:	/* Suspend step 1 (flush cache) complete */
151		if (pm->pm_state == PM_EVENT_FREEZE)
152			pm->pm_step = ide_pm_state_completed;
153		else
154			pm->pm_step = idedisk_pm_standby;
155		break;
156	case idedisk_pm_standby:	/* Suspend step 2 (standby) complete */
157		pm->pm_step = ide_pm_state_completed;
158		break;
159	case idedisk_pm_restore_pio:	/* Resume step 1 complete */
160		pm->pm_step = idedisk_pm_idle;
161		break;
162	case idedisk_pm_idle:		/* Resume step 2 (idle) complete */
163		pm->pm_step = ide_pm_restore_dma;
164		break;
165	}
166}
167
168static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
169{
170	struct request_pm_state *pm = rq->data;
171	ide_task_t *args = rq->special;
172
173	memset(args, 0, sizeof(*args));
174
175	switch (pm->pm_step) {
176	case ide_pm_flush_cache:	/* Suspend step 1 (flush cache) */
177		if (drive->media != ide_disk)
178			break;
179		/* Not supported? Switch to next step now. */
180		if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
181			ide_complete_power_step(drive, rq, 0, 0);
182			return ide_stopped;
183		}
184		if (ide_id_has_flush_cache_ext(drive->id))
185			args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
186		else
187			args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
188		args->command_type = IDE_DRIVE_TASK_NO_DATA;
189		args->handler	   = &task_no_data_intr;
190		return do_rw_taskfile(drive, args);
191
192	case idedisk_pm_standby:	/* Suspend step 2 (standby) */
193		args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
194		args->command_type = IDE_DRIVE_TASK_NO_DATA;
195		args->handler	   = &task_no_data_intr;
196		return do_rw_taskfile(drive, args);
197
198	case idedisk_pm_restore_pio:	/* Resume step 1 (restore PIO) */
199		if (drive->hwif->tuneproc != NULL)
200			drive->hwif->tuneproc(drive, 255);
201		/*
202		 * skip idedisk_pm_idle for ATAPI devices
203		 */
204		if (drive->media != ide_disk)
205			pm->pm_step = ide_pm_restore_dma;
206		else
207			ide_complete_power_step(drive, rq, 0, 0);
208		return ide_stopped;
209
210	case idedisk_pm_idle:		/* Resume step 2 (idle) */
211		args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
212		args->command_type = IDE_DRIVE_TASK_NO_DATA;
213		args->handler = task_no_data_intr;
214		return do_rw_taskfile(drive, args);
215
216	case ide_pm_restore_dma:	/* Resume step 3 (restore DMA) */
217		/*
218		 * Right now, all we do is call hwif->ide_dma_check(drive),
219		 * we could be smarter and check for current xfer_speed
220		 * in struct drive etc...
221		 */
222		if ((drive->id->capability & 1) == 0)
223			break;
224		if (drive->hwif->ide_dma_check == NULL)
225			break;
226		drive->hwif->dma_off_quietly(drive);
227		ide_set_dma(drive);
228		break;
229	}
230	pm->pm_step = ide_pm_state_completed;
231	return ide_stopped;
232}
233
234/**
235 *	ide_end_dequeued_request	-	complete an IDE I/O
236 *	@drive: IDE device for the I/O
237 *	@uptodate:
238 *	@nr_sectors: number of sectors completed
239 *
240 *	Complete an I/O that is no longer on the request queue. This
241 *	typically occurs when we pull the request and issue a REQUEST_SENSE.
242 *	We must still finish the old request but we must not tamper with the
243 *	queue in the meantime.
244 *
245 *	NOTE: This path does not handle barrier, but barrier is not supported
246 *	on ide-cd anyway.
247 */
248
249int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
250			     int uptodate, int nr_sectors)
251{
252	unsigned long flags;
253	int ret = 1;
254
255	spin_lock_irqsave(&ide_lock, flags);
256
257	BUG_ON(!blk_rq_started(rq));
258
259	/*
260	 * if failfast is set on a request, override number of sectors and
261	 * complete the whole request right now
262	 */
263	if (blk_noretry_request(rq) && end_io_error(uptodate))
264		nr_sectors = rq->hard_nr_sectors;
265
266	if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
267		rq->errors = -EIO;
268
269	/*
270	 * decide whether to reenable DMA -- 3 is a random magic for now,
271	 * if we DMA timeout more than 3 times, just stay in PIO
272	 */
273	if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
274		drive->state = 0;
275		HWGROUP(drive)->hwif->ide_dma_on(drive);
276	}
277
278	if (!end_that_request_first(rq, uptodate, nr_sectors)) {
279		add_disk_randomness(rq->rq_disk);
280		if (blk_rq_tagged(rq))
281			blk_queue_end_tag(drive->queue, rq);
282		end_that_request_last(rq, uptodate);
283		ret = 0;
284	}
285	spin_unlock_irqrestore(&ide_lock, flags);
286	return ret;
287}
288EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
289
290
291/**
292 *	ide_complete_pm_request - end the current Power Management request
293 *	@drive: target drive
294 *	@rq: request
295 *
296 *	This function cleans up the current PM request and stops the queue
297 *	if necessary.
298 */
299static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
300{
301	unsigned long flags;
302
303#ifdef DEBUG_PM
304	printk("%s: completing PM request, %s\n", drive->name,
305	       blk_pm_suspend_request(rq) ? "suspend" : "resume");
306#endif
307	spin_lock_irqsave(&ide_lock, flags);
308	if (blk_pm_suspend_request(rq)) {
309		blk_stop_queue(drive->queue);
310	} else {
311		drive->blocked = 0;
312		blk_start_queue(drive->queue);
313	}
314	blkdev_dequeue_request(rq);
315	HWGROUP(drive)->rq = NULL;
316	end_that_request_last(rq, 1);
317	spin_unlock_irqrestore(&ide_lock, flags);
318}
319
320u64 ide_get_error_location(ide_drive_t *drive, char *args)
321{
322	u32 high, low;
323	u8 hcyl, lcyl, sect;
324	u64 sector;
325
326	high = 0;
327	hcyl = args[5];
328	lcyl = args[4];
329	sect = args[3];
330
331	if (ide_id_has_flush_cache_ext(drive->id)) {
332		low = (hcyl << 16) | (lcyl << 8) | sect;
333		HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
334		high = ide_read_24(drive);
335	} else {
336		u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
337		if (cur & 0x40) {
338			high = cur & 0xf;
339			low = (hcyl << 16) | (lcyl << 8) | sect;
340		} else {
341			low = hcyl * drive->head * drive->sect;
342			low += lcyl * drive->sect;
343			low += sect - 1;
344		}
345	}
346
347	sector = ((u64) high << 24) | low;
348	return sector;
349}
350EXPORT_SYMBOL(ide_get_error_location);
351
352/**
353 *	ide_end_drive_cmd	-	end an explicit drive command
354 *	@drive: command
355 *	@stat: status bits
356 *	@err: error bits
357 *
358 *	Clean up after success/failure of an explicit drive command.
359 *	These get thrown onto the queue so they are synchronized with
360 *	real I/O operations on the drive.
361 *
362 *	In LBA48 mode we have to read the register set twice to get
363 *	all the extra information out.
364 */
365
366void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
367{
368	ide_hwif_t *hwif = HWIF(drive);
369	unsigned long flags;
370	struct request *rq;
371
372	spin_lock_irqsave(&ide_lock, flags);
373	rq = HWGROUP(drive)->rq;
374	spin_unlock_irqrestore(&ide_lock, flags);
375
376	if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
377		u8 *args = (u8 *) rq->buffer;
378		if (rq->errors == 0)
379			rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
380
381		if (args) {
382			args[0] = stat;
383			args[1] = err;
384			args[2] = hwif->INB(IDE_NSECTOR_REG);
385		}
386	} else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
387		u8 *args = (u8 *) rq->buffer;
388		if (rq->errors == 0)
389			rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
390
391		if (args) {
392			args[0] = stat;
393			args[1] = err;
394			args[2] = hwif->INB(IDE_NSECTOR_REG);
395			args[3] = hwif->INB(IDE_SECTOR_REG);
396			args[4] = hwif->INB(IDE_LCYL_REG);
397			args[5] = hwif->INB(IDE_HCYL_REG);
398			args[6] = hwif->INB(IDE_SELECT_REG);
399		}
400	} else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
401		ide_task_t *args = (ide_task_t *) rq->special;
402		if (rq->errors == 0)
403			rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
404
405		if (args) {
406			if (args->tf_in_flags.b.data) {
407				u16 data				= hwif->INW(IDE_DATA_REG);
408				args->tfRegister[IDE_DATA_OFFSET]	= (data) & 0xFF;
409				args->hobRegister[IDE_DATA_OFFSET]	= (data >> 8) & 0xFF;
410			}
411			args->tfRegister[IDE_ERROR_OFFSET]   = err;
412			/* be sure we're looking at the low order bits */
413			hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
414			args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
415			args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
416			args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
417			args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
418			args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
419			args->tfRegister[IDE_STATUS_OFFSET]  = stat;
420
421			if (drive->addressing == 1) {
422				hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
423				args->hobRegister[IDE_FEATURE_OFFSET]	= hwif->INB(IDE_FEATURE_REG);
424				args->hobRegister[IDE_NSECTOR_OFFSET]	= hwif->INB(IDE_NSECTOR_REG);
425				args->hobRegister[IDE_SECTOR_OFFSET]	= hwif->INB(IDE_SECTOR_REG);
426				args->hobRegister[IDE_LCYL_OFFSET]	= hwif->INB(IDE_LCYL_REG);
427				args->hobRegister[IDE_HCYL_OFFSET]	= hwif->INB(IDE_HCYL_REG);
428			}
429		}
430	} else if (blk_pm_request(rq)) {
431		struct request_pm_state *pm = rq->data;
432#ifdef DEBUG_PM
433		printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
434			drive->name, rq->pm->pm_step, stat, err);
435#endif
436		ide_complete_power_step(drive, rq, stat, err);
437		if (pm->pm_step == ide_pm_state_completed)
438			ide_complete_pm_request(drive, rq);
439		return;
440	}
441
442	spin_lock_irqsave(&ide_lock, flags);
443	blkdev_dequeue_request(rq);
444	HWGROUP(drive)->rq = NULL;
445	rq->errors = err;
446	end_that_request_last(rq, !rq->errors);
447	spin_unlock_irqrestore(&ide_lock, flags);
448}
449
450EXPORT_SYMBOL(ide_end_drive_cmd);
451
452/**
453 *	try_to_flush_leftover_data	-	flush junk
454 *	@drive: drive to flush
455 *
456 *	try_to_flush_leftover_data() is invoked in response to a drive
457 *	unexpectedly having its DRQ_STAT bit set.  As an alternative to
458 *	resetting the drive, this routine tries to clear the condition
459 *	by read a sector's worth of data from the drive.  Of course,
460 *	this may not help if the drive is *waiting* for data from *us*.
461 */
462static void try_to_flush_leftover_data (ide_drive_t *drive)
463{
464	int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
465
466	if (drive->media != ide_disk)
467		return;
468	while (i > 0) {
469		u32 buffer[16];
470		u32 wcount = (i > 16) ? 16 : i;
471
472		i -= wcount;
473		HWIF(drive)->ata_input_data(drive, buffer, wcount);
474	}
475}
476
477static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
478{
479	if (rq->rq_disk) {
480		ide_driver_t *drv;
481
482		drv = *(ide_driver_t **)rq->rq_disk->private_data;
483		drv->end_request(drive, 0, 0);
484	} else
485		ide_end_request(drive, 0, 0);
486}
487
488static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
489{
490	ide_hwif_t *hwif = drive->hwif;
491
492	if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
493		/* other bits are useless when BUSY */
494		rq->errors |= ERROR_RESET;
495	} else if (stat & ERR_STAT) {
496		/* err has different meaning on cdrom and tape */
497		if (err == ABRT_ERR) {
498			if (drive->select.b.lba &&
499			    /* some newer drives don't support WIN_SPECIFY */
500			    hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
501				return ide_stopped;
502		} else if ((err & BAD_CRC) == BAD_CRC) {
503			/* UDMA crc error, just retry the operation */
504			drive->crc_count++;
505		} else if (err & (BBD_ERR | ECC_ERR)) {
506			/* retries won't help these */
507			rq->errors = ERROR_MAX;
508		} else if (err & TRK0_ERR) {
509			/* help it find track zero */
510			rq->errors |= ERROR_RECAL;
511		}
512	}
513
514	if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && hwif->err_stops_fifo == 0)
515		try_to_flush_leftover_data(drive);
516
517	if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
518		ide_kill_rq(drive, rq);
519		return ide_stopped;
520	}
521
522	if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
523		rq->errors |= ERROR_RESET;
524
525	if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
526		++rq->errors;
527		return ide_do_reset(drive);
528	}
529
530	if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
531		drive->special.b.recalibrate = 1;
532
533	++rq->errors;
534
535	return ide_stopped;
536}
537
538static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
539{
540	ide_hwif_t *hwif = drive->hwif;
541
542	if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
543		/* other bits are useless when BUSY */
544		rq->errors |= ERROR_RESET;
545	} else {
546		/* add decoding error stuff */
547	}
548
549	if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
550		/* force an abort */
551		hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
552
553	if (rq->errors >= ERROR_MAX) {
554		ide_kill_rq(drive, rq);
555	} else {
556		if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
557			++rq->errors;
558			return ide_do_reset(drive);
559		}
560		++rq->errors;
561	}
562
563	return ide_stopped;
564}
565
566ide_startstop_t
567__ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
568{
569	if (drive->media == ide_disk)
570		return ide_ata_error(drive, rq, stat, err);
571	return ide_atapi_error(drive, rq, stat, err);
572}
573
574EXPORT_SYMBOL_GPL(__ide_error);
575
576/**
577 *	ide_error	-	handle an error on the IDE
578 *	@drive: drive the error occurred on
579 *	@msg: message to report
580 *	@stat: status bits
581 *
582 *	ide_error() takes action based on the error returned by the drive.
583 *	For normal I/O that may well include retries. We deal with
584 *	both new-style (taskfile) and old style command handling here.
585 *	In the case of taskfile command handling there is work left to
586 *	do
587 */
588
589ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
590{
591	struct request *rq;
592	u8 err;
593
594	err = ide_dump_status(drive, msg, stat);
595
596	if ((rq = HWGROUP(drive)->rq) == NULL)
597		return ide_stopped;
598
599	/* retry only "normal" I/O: */
600	if (!blk_fs_request(rq)) {
601		rq->errors = 1;
602		ide_end_drive_cmd(drive, stat, err);
603		return ide_stopped;
604	}
605
606	if (rq->rq_disk) {
607		ide_driver_t *drv;
608
609		drv = *(ide_driver_t **)rq->rq_disk->private_data;
610		return drv->error(drive, rq, stat, err);
611	} else
612		return __ide_error(drive, rq, stat, err);
613}
614
615EXPORT_SYMBOL_GPL(ide_error);
616
617ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
618{
619	if (drive->media != ide_disk)
620		rq->errors |= ERROR_RESET;
621
622	ide_kill_rq(drive, rq);
623
624	return ide_stopped;
625}
626
627EXPORT_SYMBOL_GPL(__ide_abort);
628
629/**
630 *	ide_abort	-	abort pending IDE operations
631 *	@drive: drive the error occurred on
632 *	@msg: message to report
633 *
634 *	ide_abort kills and cleans up when we are about to do a
635 *	host initiated reset on active commands. Longer term we
636 *	want handlers to have sensible abort handling themselves
637 *
638 *	This differs fundamentally from ide_error because in
639 *	this case the command is doing just fine when we
640 *	blow it away.
641 */
642
643ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
644{
645	struct request *rq;
646
647	if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
648		return ide_stopped;
649
650	/* retry only "normal" I/O: */
651	if (!blk_fs_request(rq)) {
652		rq->errors = 1;
653		ide_end_drive_cmd(drive, BUSY_STAT, 0);
654		return ide_stopped;
655	}
656
657	if (rq->rq_disk) {
658		ide_driver_t *drv;
659
660		drv = *(ide_driver_t **)rq->rq_disk->private_data;
661		return drv->abort(drive, rq);
662	} else
663		return __ide_abort(drive, rq);
664}
665
666/**
667 *	ide_cmd		-	issue a simple drive command
668 *	@drive: drive the command is for
669 *	@cmd: command byte
670 *	@nsect: sector byte
671 *	@handler: handler for the command completion
672 *
673 *	Issue a simple drive command with interrupts.
674 *	The drive must be selected beforehand.
675 */
676
677static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
678		ide_handler_t *handler)
679{
680	ide_hwif_t *hwif = HWIF(drive);
681	if (IDE_CONTROL_REG)
682		hwif->OUTB(drive->ctl,IDE_CONTROL_REG);	/* clear nIEN */
683	SELECT_MASK(drive,0);
684	hwif->OUTB(nsect,IDE_NSECTOR_REG);
685	ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
686}
687
688/**
689 *	drive_cmd_intr		- 	drive command completion interrupt
690 *	@drive: drive the completion interrupt occurred on
691 *
692 *	drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
693 *	We do any necessary data reading and then wait for the drive to
694 *	go non busy. At that point we may read the error data and complete
695 *	the request
696 */
697
698static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
699{
700	struct request *rq = HWGROUP(drive)->rq;
701	ide_hwif_t *hwif = HWIF(drive);
702	u8 *args = (u8 *) rq->buffer;
703	u8 stat = hwif->INB(IDE_STATUS_REG);
704	int retries = 10;
705
706	local_irq_enable_in_hardirq();
707	if ((stat & DRQ_STAT) && args && args[3]) {
708		u8 io_32bit = drive->io_32bit;
709		drive->io_32bit = 0;
710		hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
711		drive->io_32bit = io_32bit;
712		while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
713			udelay(100);
714	}
715
716	if (!OK_STAT(stat, READY_STAT, BAD_STAT))
717		return ide_error(drive, "drive_cmd", stat);
718		/* calls ide_end_drive_cmd */
719	ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
720	return ide_stopped;
721}
722
723static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
724{
725	task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
726	task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
727	task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
728	task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
729	task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
730	task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
731
732	task->handler = &set_geometry_intr;
733}
734
735static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
736{
737	task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
738	task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
739
740	task->handler = &recal_intr;
741}
742
743static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
744{
745	task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
746	task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
747
748	task->handler = &set_multmode_intr;
749}
750
751static ide_startstop_t ide_disk_special(ide_drive_t *drive)
752{
753	special_t *s = &drive->special;
754	ide_task_t args;
755
756	memset(&args, 0, sizeof(ide_task_t));
757	args.command_type = IDE_DRIVE_TASK_NO_DATA;
758
759	if (s->b.set_geometry) {
760		s->b.set_geometry = 0;
761		ide_init_specify_cmd(drive, &args);
762	} else if (s->b.recalibrate) {
763		s->b.recalibrate = 0;
764		ide_init_restore_cmd(drive, &args);
765	} else if (s->b.set_multmode) {
766		s->b.set_multmode = 0;
767		if (drive->mult_req > drive->id->max_multsect)
768			drive->mult_req = drive->id->max_multsect;
769		ide_init_setmult_cmd(drive, &args);
770	} else if (s->all) {
771		int special = s->all;
772		s->all = 0;
773		printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
774		return ide_stopped;
775	}
776
777	do_rw_taskfile(drive, &args);
778
779	return ide_started;
780}
781
782/**
783 *	do_special		-	issue some special commands
784 *	@drive: drive the command is for
785 *
786 *	do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
787 *	commands to a drive.  It used to do much more, but has been scaled
788 *	back.
789 */
790
791static ide_startstop_t do_special (ide_drive_t *drive)
792{
793	special_t *s = &drive->special;
794
795#ifdef DEBUG
796	printk("%s: do_special: 0x%02x\n", drive->name, s->all);
797#endif
798	if (s->b.set_tune) {
799		s->b.set_tune = 0;
800		if (HWIF(drive)->tuneproc != NULL)
801			HWIF(drive)->tuneproc(drive, drive->tune_req);
802		return ide_stopped;
803	} else {
804		if (drive->media == ide_disk)
805			return ide_disk_special(drive);
806
807		s->all = 0;
808		drive->mult_req = 0;
809		return ide_stopped;
810	}
811}
812
813void ide_map_sg(ide_drive_t *drive, struct request *rq)
814{
815	ide_hwif_t *hwif = drive->hwif;
816	struct scatterlist *sg = hwif->sg_table;
817
818	if (hwif->sg_mapped)	/* needed by ide-scsi */
819		return;
820
821	if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
822		hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
823	} else {
824		sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
825		hwif->sg_nents = 1;
826	}
827}
828
829EXPORT_SYMBOL_GPL(ide_map_sg);
830
831void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
832{
833	ide_hwif_t *hwif = drive->hwif;
834
835	hwif->nsect = hwif->nleft = rq->nr_sectors;
836	hwif->cursg = hwif->cursg_ofs = 0;
837}
838
839EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
840
841/**
842 *	execute_drive_command	-	issue special drive command
843 *	@drive: the drive to issue the command on
844 *	@rq: the request structure holding the command
845 *
846 *	execute_drive_cmd() issues a special drive command,  usually
847 *	initiated by ioctl() from the external hdparm program. The
848 *	command can be a drive command, drive task or taskfile
849 *	operation. Weirdly you can call it with NULL to wait for
850 *	all commands to finish. Don't do this as that is due to change
851 */
852
853static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
854		struct request *rq)
855{
856	ide_hwif_t *hwif = HWIF(drive);
857	if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
858 		ide_task_t *args = rq->special;
859
860		if (!args)
861			goto done;
862
863		hwif->data_phase = args->data_phase;
864
865		switch (hwif->data_phase) {
866		case TASKFILE_MULTI_OUT:
867		case TASKFILE_OUT:
868		case TASKFILE_MULTI_IN:
869		case TASKFILE_IN:
870			ide_init_sg_cmd(drive, rq);
871			ide_map_sg(drive, rq);
872		default:
873			break;
874		}
875
876		if (args->tf_out_flags.all != 0)
877			return flagged_taskfile(drive, args);
878		return do_rw_taskfile(drive, args);
879	} else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
880		u8 *args = rq->buffer;
881		u8 sel;
882
883		if (!args)
884			goto done;
885#ifdef DEBUG
886 		printk("%s: DRIVE_TASK_CMD ", drive->name);
887 		printk("cmd=0x%02x ", args[0]);
888 		printk("fr=0x%02x ", args[1]);
889 		printk("ns=0x%02x ", args[2]);
890 		printk("sc=0x%02x ", args[3]);
891 		printk("lcyl=0x%02x ", args[4]);
892 		printk("hcyl=0x%02x ", args[5]);
893 		printk("sel=0x%02x\n", args[6]);
894#endif
895 		hwif->OUTB(args[1], IDE_FEATURE_REG);
896 		hwif->OUTB(args[3], IDE_SECTOR_REG);
897 		hwif->OUTB(args[4], IDE_LCYL_REG);
898 		hwif->OUTB(args[5], IDE_HCYL_REG);
899 		sel = (args[6] & ~0x10);
900 		if (drive->select.b.unit)
901 			sel |= 0x10;
902 		hwif->OUTB(sel, IDE_SELECT_REG);
903 		ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
904 		return ide_started;
905 	} else if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
906 		u8 *args = rq->buffer;
907
908		if (!args)
909			goto done;
910#ifdef DEBUG
911 		printk("%s: DRIVE_CMD ", drive->name);
912 		printk("cmd=0x%02x ", args[0]);
913 		printk("sc=0x%02x ", args[1]);
914 		printk("fr=0x%02x ", args[2]);
915 		printk("xx=0x%02x\n", args[3]);
916#endif
917 		if (args[0] == WIN_SMART) {
918 			hwif->OUTB(0x4f, IDE_LCYL_REG);
919 			hwif->OUTB(0xc2, IDE_HCYL_REG);
920 			hwif->OUTB(args[2],IDE_FEATURE_REG);
921 			hwif->OUTB(args[1],IDE_SECTOR_REG);
922 			ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
923 			return ide_started;
924 		}
925 		hwif->OUTB(args[2],IDE_FEATURE_REG);
926 		ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
927 		return ide_started;
928 	}
929
930done:
931 	/*
932 	 * NULL is actually a valid way of waiting for
933 	 * all current requests to be flushed from the queue.
934 	 */
935#ifdef DEBUG
936 	printk("%s: DRIVE_CMD (null)\n", drive->name);
937#endif
938 	ide_end_drive_cmd(drive,
939			hwif->INB(IDE_STATUS_REG),
940			hwif->INB(IDE_ERROR_REG));
941 	return ide_stopped;
942}
943
944static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
945{
946	struct request_pm_state *pm = rq->data;
947
948	if (blk_pm_suspend_request(rq) &&
949	    pm->pm_step == ide_pm_state_start_suspend)
950		/* Mark drive blocked when starting the suspend sequence. */
951		drive->blocked = 1;
952	else if (blk_pm_resume_request(rq) &&
953		 pm->pm_step == ide_pm_state_start_resume) {
954		/*
955		 * The first thing we do on wakeup is to wait for BSY bit to
956		 * go away (with a looong timeout) as a drive on this hwif may
957		 * just be POSTing itself.
958		 * We do that before even selecting as the "other" device on
959		 * the bus may be broken enough to walk on our toes at this
960		 * point.
961		 */
962		int rc;
963#ifdef DEBUG_PM
964		printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
965#endif
966		rc = ide_wait_not_busy(HWIF(drive), 35000);
967		if (rc)
968			printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
969		SELECT_DRIVE(drive);
970		HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
971		rc = ide_wait_not_busy(HWIF(drive), 100000);
972		if (rc)
973			printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
974	}
975}
976
977
978static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
979{
980	ide_startstop_t startstop;
981	sector_t block;
982
983	BUG_ON(!blk_rq_started(rq));
984
985#ifdef DEBUG
986	printk("%s: start_request: current=0x%08lx\n",
987		HWIF(drive)->name, (unsigned long) rq);
988#endif
989
990	/* bail early if we've exceeded max_failures */
991	if (drive->max_failures && (drive->failures > drive->max_failures)) {
992		goto kill_rq;
993	}
994
995	block    = rq->sector;
996	if (blk_fs_request(rq) &&
997	    (drive->media == ide_disk || drive->media == ide_floppy)) {
998		block += drive->sect0;
999	}
1000	/* Yecch - this will shift the entire interval,
1001	   possibly killing some innocent following sector */
1002	if (block == 0 && drive->remap_0_to_1 == 1)
1003		block = 1;  /* redirect MBR access to EZ-Drive partn table */
1004
1005	if (blk_pm_request(rq))
1006		ide_check_pm_state(drive, rq);
1007
1008	SELECT_DRIVE(drive);
1009	if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1010		printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1011		return startstop;
1012	}
1013	if (!drive->special.all) {
1014		ide_driver_t *drv;
1015
1016		/*
1017		 * We reset the drive so we need to issue a SETFEATURES.
1018		 * Do it _after_ do_special() restored device parameters.
1019		 */
1020		if (drive->current_speed == 0xff)
1021			ide_config_drive_speed(drive, drive->desired_speed);
1022
1023		if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1024		    rq->cmd_type == REQ_TYPE_ATA_TASK ||
1025		    rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1026			return execute_drive_cmd(drive, rq);
1027		else if (blk_pm_request(rq)) {
1028			struct request_pm_state *pm = rq->data;
1029#ifdef DEBUG_PM
1030			printk("%s: start_power_step(step: %d)\n",
1031				drive->name, rq->pm->pm_step);
1032#endif
1033			startstop = ide_start_power_step(drive, rq);
1034			if (startstop == ide_stopped &&
1035			    pm->pm_step == ide_pm_state_completed)
1036				ide_complete_pm_request(drive, rq);
1037			return startstop;
1038		}
1039
1040		drv = *(ide_driver_t **)rq->rq_disk->private_data;
1041		return drv->do_request(drive, rq, block);
1042	}
1043	return do_special(drive);
1044kill_rq:
1045	ide_kill_rq(drive, rq);
1046	return ide_stopped;
1047}
1048
1049/**
1050 *	ide_stall_queue		-	pause an IDE device
1051 *	@drive: drive to stall
1052 *	@timeout: time to stall for (jiffies)
1053 *
1054 *	ide_stall_queue() can be used by a drive to give excess bandwidth back
1055 *	to the hwgroup by sleeping for timeout jiffies.
1056 */
1057
1058void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1059{
1060	if (timeout > WAIT_WORSTCASE)
1061		timeout = WAIT_WORSTCASE;
1062	drive->sleep = timeout + jiffies;
1063	drive->sleeping = 1;
1064}
1065
1066EXPORT_SYMBOL(ide_stall_queue);
1067
1068#define WAKEUP(drive)	((drive)->service_start + 2 * (drive)->service_time)
1069
1070/**
1071 *	choose_drive		-	select a drive to service
1072 *	@hwgroup: hardware group to select on
1073 *
1074 *	choose_drive() selects the next drive which will be serviced.
1075 *	This is necessary because the IDE layer can't issue commands
1076 *	to both drives on the same cable, unlike SCSI.
1077 */
1078
1079static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1080{
1081	ide_drive_t *drive, *best;
1082
1083repeat:
1084	best = NULL;
1085	drive = hwgroup->drive;
1086
1087	/*
1088	 * drive is doing pre-flush, ordered write, post-flush sequence. even
1089	 * though that is 3 requests, it must be seen as a single transaction.
1090	 * we must not preempt this drive until that is complete
1091	 */
1092	if (blk_queue_flushing(drive->queue)) {
1093		/*
1094		 * small race where queue could get replugged during
1095		 * the 3-request flush cycle, just yank the plug since
1096		 * we want it to finish asap
1097		 */
1098		blk_remove_plug(drive->queue);
1099		return drive;
1100	}
1101
1102	do {
1103		if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1104		    && !elv_queue_empty(drive->queue)) {
1105			if (!best
1106			 || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1107			 || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1108			{
1109				if (!blk_queue_plugged(drive->queue))
1110					best = drive;
1111			}
1112		}
1113	} while ((drive = drive->next) != hwgroup->drive);
1114	if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1115		long t = (signed long)(WAKEUP(best) - jiffies);
1116		if (t >= WAIT_MIN_SLEEP) {
1117		/*
1118		 * We *may* have some time to spare, but first let's see if
1119		 * someone can potentially benefit from our nice mood today..
1120		 */
1121			drive = best->next;
1122			do {
1123				if (!drive->sleeping
1124				 && time_before(jiffies - best->service_time, WAKEUP(drive))
1125				 && time_before(WAKEUP(drive), jiffies + t))
1126				{
1127					ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1128					goto repeat;
1129				}
1130			} while ((drive = drive->next) != best);
1131		}
1132	}
1133	return best;
1134}
1135
1136/*
1137 * Issue a new request to a drive from hwgroup
1138 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1139 *
1140 * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1141 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1142 * may have both interfaces in a single hwgroup to "serialize" access.
1143 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1144 * together into one hwgroup for serialized access.
1145 *
1146 * Note also that several hwgroups can end up sharing a single IRQ,
1147 * possibly along with many other devices.  This is especially common in
1148 * PCI-based systems with off-board IDE controller cards.
1149 *
1150 * The IDE driver uses the single global ide_lock spinlock to protect
1151 * access to the request queues, and to protect the hwgroup->busy flag.
1152 *
1153 * The first thread into the driver for a particular hwgroup sets the
1154 * hwgroup->busy flag to indicate that this hwgroup is now active,
1155 * and then initiates processing of the top request from the request queue.
1156 *
1157 * Other threads attempting entry notice the busy setting, and will simply
1158 * queue their new requests and exit immediately.  Note that hwgroup->busy
1159 * remains set even when the driver is merely awaiting the next interrupt.
1160 * Thus, the meaning is "this hwgroup is busy processing a request".
1161 *
1162 * When processing of a request completes, the completing thread or IRQ-handler
1163 * will start the next request from the queue.  If no more work remains,
1164 * the driver will clear the hwgroup->busy flag and exit.
1165 *
1166 * The ide_lock (spinlock) is used to protect all access to the
1167 * hwgroup->busy flag, but is otherwise not needed for most processing in
1168 * the driver.  This makes the driver much more friendlier to shared IRQs
1169 * than previous designs, while remaining 100% (?) SMP safe and capable.
1170 */
1171static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1172{
1173	ide_drive_t	*drive;
1174	ide_hwif_t	*hwif;
1175	struct request	*rq;
1176	ide_startstop_t	startstop;
1177	int             loops = 0;
1178
1179	/* for atari only: POSSIBLY BROKEN HERE(?) */
1180	ide_get_lock(ide_intr, hwgroup);
1181
1182	/* caller must own ide_lock */
1183	BUG_ON(!irqs_disabled());
1184
1185	while (!hwgroup->busy) {
1186		hwgroup->busy = 1;
1187		drive = choose_drive(hwgroup);
1188		if (drive == NULL) {
1189			int sleeping = 0;
1190			unsigned long sleep = 0; /* shut up, gcc */
1191			hwgroup->rq = NULL;
1192			drive = hwgroup->drive;
1193			do {
1194				if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1195					sleeping = 1;
1196					sleep = drive->sleep;
1197				}
1198			} while ((drive = drive->next) != hwgroup->drive);
1199			if (sleeping) {
1200		/*
1201		 * Take a short snooze, and then wake up this hwgroup again.
1202		 * This gives other hwgroups on the same a chance to
1203		 * play fairly with us, just in case there are big differences
1204		 * in relative throughputs.. don't want to hog the cpu too much.
1205		 */
1206				if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1207					sleep = jiffies + WAIT_MIN_SLEEP;
1208				if (timer_pending(&hwgroup->timer))
1209					printk(KERN_CRIT "ide_set_handler: timer already active\n");
1210				/* so that ide_timer_expiry knows what to do */
1211				hwgroup->sleeping = 1;
1212				hwgroup->req_gen_timer = hwgroup->req_gen;
1213				mod_timer(&hwgroup->timer, sleep);
1214				/* we purposely leave hwgroup->busy==1
1215				 * while sleeping */
1216			} else {
1217				/* Ugly, but how can we sleep for the lock
1218				 * otherwise? perhaps from tq_disk?
1219				 */
1220
1221				/* for atari only */
1222				ide_release_lock();
1223				hwgroup->busy = 0;
1224			}
1225
1226			/* no more work for this hwgroup (for now) */
1227			return;
1228		}
1229	again:
1230		hwif = HWIF(drive);
1231		if (hwgroup->hwif->sharing_irq &&
1232		    hwif != hwgroup->hwif &&
1233		    hwif->io_ports[IDE_CONTROL_OFFSET]) {
1234			/* set nIEN for previous hwif */
1235			SELECT_INTERRUPT(drive);
1236		}
1237		hwgroup->hwif = hwif;
1238		hwgroup->drive = drive;
1239		drive->sleeping = 0;
1240		drive->service_start = jiffies;
1241
1242		if (blk_queue_plugged(drive->queue)) {
1243			printk(KERN_ERR "ide: huh? queue was plugged!\n");
1244			break;
1245		}
1246
1247		/*
1248		 * we know that the queue isn't empty, but this can happen
1249		 * if the q->prep_rq_fn() decides to kill a request
1250		 */
1251		rq = elv_next_request(drive->queue);
1252		if (!rq) {
1253			hwgroup->busy = 0;
1254			break;
1255		}
1256
1257		/*
1258		 * Sanity: don't accept a request that isn't a PM request
1259		 * if we are currently power managed. This is very important as
1260		 * blk_stop_queue() doesn't prevent the elv_next_request()
1261		 * above to return us whatever is in the queue. Since we call
1262		 * ide_do_request() ourselves, we end up taking requests while
1263		 * the queue is blocked...
1264		 *
1265		 * We let requests forced at head of queue with ide-preempt
1266		 * though. I hope that doesn't happen too much, hopefully not
1267		 * unless the subdriver triggers such a thing in its own PM
1268		 * state machine.
1269		 *
1270		 * We count how many times we loop here to make sure we service
1271		 * all drives in the hwgroup without looping for ever
1272		 */
1273		if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1274			drive = drive->next ? drive->next : hwgroup->drive;
1275			if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1276				goto again;
1277			/* We clear busy, there should be no pending ATA command at this point. */
1278			hwgroup->busy = 0;
1279			break;
1280		}
1281
1282		hwgroup->rq = rq;
1283
1284		/*
1285		 * Some systems have trouble with IDE IRQs arriving while
1286		 * the driver is still setting things up.  So, here we disable
1287		 * the IRQ used by this interface while the request is being started.
1288		 * This may look bad at first, but pretty much the same thing
1289		 * happens anyway when any interrupt comes in, IDE or otherwise
1290		 *  -- the kernel masks the IRQ while it is being handled.
1291		 */
1292		if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1293			disable_irq_nosync(hwif->irq);
1294		spin_unlock(&ide_lock);
1295		local_irq_enable_in_hardirq();
1296			/* allow other IRQs while we start this request */
1297		startstop = start_request(drive, rq);
1298		spin_lock_irq(&ide_lock);
1299		if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1300			enable_irq(hwif->irq);
1301		if (startstop == ide_stopped)
1302			hwgroup->busy = 0;
1303	}
1304}
1305
1306/*
1307 * Passes the stuff to ide_do_request
1308 */
1309void do_ide_request(request_queue_t *q)
1310{
1311	ide_drive_t *drive = q->queuedata;
1312
1313	ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1314}
1315
1316/*
1317 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1318 * retry the current request in pio mode instead of risking tossing it
1319 * all away
1320 */
1321static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1322{
1323	ide_hwif_t *hwif = HWIF(drive);
1324	struct request *rq;
1325	ide_startstop_t ret = ide_stopped;
1326
1327	/*
1328	 * end current dma transaction
1329	 */
1330
1331	if (error < 0) {
1332		printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1333		(void)HWIF(drive)->ide_dma_end(drive);
1334		ret = ide_error(drive, "dma timeout error",
1335						hwif->INB(IDE_STATUS_REG));
1336	} else {
1337		printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1338		(void) hwif->ide_dma_timeout(drive);
1339	}
1340
1341	/*
1342	 * disable dma for now, but remember that we did so because of
1343	 * a timeout -- we'll reenable after we finish this next request
1344	 * (or rather the first chunk of it) in pio.
1345	 */
1346	drive->retry_pio++;
1347	drive->state = DMA_PIO_RETRY;
1348	hwif->dma_off_quietly(drive);
1349
1350	/*
1351	 * un-busy drive etc (hwgroup->busy is cleared on return) and
1352	 * make sure request is sane
1353	 */
1354	rq = HWGROUP(drive)->rq;
1355
1356	if (!rq)
1357		goto out;
1358
1359	HWGROUP(drive)->rq = NULL;
1360
1361	rq->errors = 0;
1362
1363	if (!rq->bio)
1364		goto out;
1365
1366	rq->sector = rq->bio->bi_sector;
1367	rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1368	rq->hard_cur_sectors = rq->current_nr_sectors;
1369	rq->buffer = bio_data(rq->bio);
1370out:
1371	return ret;
1372}
1373
1374/**
1375 *	ide_timer_expiry	-	handle lack of an IDE interrupt
1376 *	@data: timer callback magic (hwgroup)
1377 *
1378 *	An IDE command has timed out before the expected drive return
1379 *	occurred. At this point we attempt to clean up the current
1380 *	mess. If the current handler includes an expiry handler then
1381 *	we invoke the expiry handler, and providing it is happy the
1382 *	work is done. If that fails we apply generic recovery rules
1383 *	invoking the handler and checking the drive DMA status. We
1384 *	have an excessively incestuous relationship with the DMA
1385 *	logic that wants cleaning up.
1386 */
1387
1388void ide_timer_expiry (unsigned long data)
1389{
1390	ide_hwgroup_t	*hwgroup = (ide_hwgroup_t *) data;
1391	ide_handler_t	*handler;
1392	ide_expiry_t	*expiry;
1393	unsigned long	flags;
1394	unsigned long	wait = -1;
1395
1396	spin_lock_irqsave(&ide_lock, flags);
1397
1398	if (((handler = hwgroup->handler) == NULL) ||
1399	    (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1400		/*
1401		 * Either a marginal timeout occurred
1402		 * (got the interrupt just as timer expired),
1403		 * or we were "sleeping" to give other devices a chance.
1404		 * Either way, we don't really want to complain about anything.
1405		 */
1406		if (hwgroup->sleeping) {
1407			hwgroup->sleeping = 0;
1408			hwgroup->busy = 0;
1409		}
1410	} else {
1411		ide_drive_t *drive = hwgroup->drive;
1412		if (!drive) {
1413			printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1414			hwgroup->handler = NULL;
1415		} else {
1416			ide_hwif_t *hwif;
1417			ide_startstop_t startstop = ide_stopped;
1418			if (!hwgroup->busy) {
1419				hwgroup->busy = 1;	/* paranoia */
1420				printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1421			}
1422			if ((expiry = hwgroup->expiry) != NULL) {
1423				/* continue */
1424				if ((wait = expiry(drive)) > 0) {
1425					/* reset timer */
1426					hwgroup->timer.expires  = jiffies + wait;
1427					hwgroup->req_gen_timer = hwgroup->req_gen;
1428					add_timer(&hwgroup->timer);
1429					spin_unlock_irqrestore(&ide_lock, flags);
1430					return;
1431				}
1432			}
1433			hwgroup->handler = NULL;
1434			/*
1435			 * We need to simulate a real interrupt when invoking
1436			 * the handler() function, which means we need to
1437			 * globally mask the specific IRQ:
1438			 */
1439			spin_unlock(&ide_lock);
1440			hwif  = HWIF(drive);
1441#if DISABLE_IRQ_NOSYNC
1442			disable_irq_nosync(hwif->irq);
1443#else
1444			/* disable_irq_nosync ?? */
1445			disable_irq(hwif->irq);
1446#endif /* DISABLE_IRQ_NOSYNC */
1447			/* local CPU only,
1448			 * as if we were handling an interrupt */
1449			local_irq_disable();
1450			if (hwgroup->polling) {
1451				startstop = handler(drive);
1452			} else if (drive_is_ready(drive)) {
1453				if (drive->waiting_for_dma)
1454					(void) hwgroup->hwif->ide_dma_lostirq(drive);
1455				(void)ide_ack_intr(hwif);
1456				printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1457				startstop = handler(drive);
1458			} else {
1459				if (drive->waiting_for_dma) {
1460					startstop = ide_dma_timeout_retry(drive, wait);
1461				} else
1462					startstop =
1463					ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1464			}
1465			drive->service_time = jiffies - drive->service_start;
1466			spin_lock_irq(&ide_lock);
1467			enable_irq(hwif->irq);
1468			if (startstop == ide_stopped)
1469				hwgroup->busy = 0;
1470		}
1471	}
1472	ide_do_request(hwgroup, IDE_NO_IRQ);
1473	spin_unlock_irqrestore(&ide_lock, flags);
1474}
1475
1476/**
1477 *	unexpected_intr		-	handle an unexpected IDE interrupt
1478 *	@irq: interrupt line
1479 *	@hwgroup: hwgroup being processed
1480 *
1481 *	There's nothing really useful we can do with an unexpected interrupt,
1482 *	other than reading the status register (to clear it), and logging it.
1483 *	There should be no way that an irq can happen before we're ready for it,
1484 *	so we needn't worry much about losing an "important" interrupt here.
1485 *
1486 *	On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1487 *	the drive enters "idle", "standby", or "sleep" mode, so if the status
1488 *	looks "good", we just ignore the interrupt completely.
1489 *
1490 *	This routine assumes __cli() is in effect when called.
1491 *
1492 *	If an unexpected interrupt happens on irq15 while we are handling irq14
1493 *	and if the two interfaces are "serialized" (CMD640), then it looks like
1494 *	we could screw up by interfering with a new request being set up for
1495 *	irq15.
1496 *
1497 *	In reality, this is a non-issue.  The new command is not sent unless
1498 *	the drive is ready to accept one, in which case we know the drive is
1499 *	not trying to interrupt us.  And ide_set_handler() is always invoked
1500 *	before completing the issuance of any new drive command, so we will not
1501 *	be accidentally invoked as a result of any valid command completion
1502 *	interrupt.
1503 *
1504 *	Note that we must walk the entire hwgroup here. We know which hwif
1505 *	is doing the current command, but we don't know which hwif burped
1506 *	mysteriously.
1507 */
1508
1509static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1510{
1511	u8 stat;
1512	ide_hwif_t *hwif = hwgroup->hwif;
1513
1514	/*
1515	 * handle the unexpected interrupt
1516	 */
1517	do {
1518		if (hwif->irq == irq) {
1519			stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1520			if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1521				/* Try to not flood the console with msgs */
1522				static unsigned long last_msgtime, count;
1523				++count;
1524				if (time_after(jiffies, last_msgtime + HZ)) {
1525					last_msgtime = jiffies;
1526					printk(KERN_ERR "%s%s: unexpected interrupt, "
1527						"status=0x%02x, count=%ld\n",
1528						hwif->name,
1529						(hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1530				}
1531			}
1532		}
1533	} while ((hwif = hwif->next) != hwgroup->hwif);
1534}
1535
1536/**
1537 *	ide_intr	-	default IDE interrupt handler
1538 *	@irq: interrupt number
1539 *	@dev_id: hwif group
1540 *	@regs: unused weirdness from the kernel irq layer
1541 *
1542 *	This is the default IRQ handler for the IDE layer. You should
1543 *	not need to override it. If you do be aware it is subtle in
1544 *	places
1545 *
1546 *	hwgroup->hwif is the interface in the group currently performing
1547 *	a command. hwgroup->drive is the drive and hwgroup->handler is
1548 *	the IRQ handler to call. As we issue a command the handlers
1549 *	step through multiple states, reassigning the handler to the
1550 *	next step in the process. Unlike a smart SCSI controller IDE
1551 *	expects the main processor to sequence the various transfer
1552 *	stages. We also manage a poll timer to catch up with most
1553 *	timeout situations. There are still a few where the handlers
1554 *	don't ever decide to give up.
1555 *
1556 *	The handler eventually returns ide_stopped to indicate the
1557 *	request completed. At this point we issue the next request
1558 *	on the hwgroup and the process begins again.
1559 */
1560
1561irqreturn_t ide_intr (int irq, void *dev_id)
1562{
1563	unsigned long flags;
1564	ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1565	ide_hwif_t *hwif;
1566	ide_drive_t *drive;
1567	ide_handler_t *handler;
1568	ide_startstop_t startstop;
1569
1570	spin_lock_irqsave(&ide_lock, flags);
1571	hwif = hwgroup->hwif;
1572
1573	if (!ide_ack_intr(hwif)) {
1574		spin_unlock_irqrestore(&ide_lock, flags);
1575		return IRQ_NONE;
1576	}
1577
1578	if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1579#ifdef CONFIG_BLK_DEV_IDEPCI
1580		if (hwif->pci_dev && !hwif->pci_dev->vendor)
1581#endif	/* CONFIG_BLK_DEV_IDEPCI */
1582		{
1583			/*
1584			 * Probably not a shared PCI interrupt,
1585			 * so we can safely try to do something about it:
1586			 */
1587			unexpected_intr(irq, hwgroup);
1588#ifdef CONFIG_BLK_DEV_IDEPCI
1589		} else {
1590			/*
1591			 * Whack the status register, just in case
1592			 * we have a leftover pending IRQ.
1593			 */
1594			(void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1595#endif /* CONFIG_BLK_DEV_IDEPCI */
1596		}
1597		spin_unlock_irqrestore(&ide_lock, flags);
1598		return IRQ_NONE;
1599	}
1600	drive = hwgroup->drive;
1601	if (!drive) {
1602		/*
1603		 * This should NEVER happen, and there isn't much
1604		 * we could do about it here.
1605		 *
1606		 * [Note - this can occur if the drive is hot unplugged]
1607		 */
1608		spin_unlock_irqrestore(&ide_lock, flags);
1609		return IRQ_HANDLED;
1610	}
1611	if (!drive_is_ready(drive)) {
1612		/*
1613		 * This happens regularly when we share a PCI IRQ with
1614		 * another device.  Unfortunately, it can also happen
1615		 * with some buggy drives that trigger the IRQ before
1616		 * their status register is up to date.  Hopefully we have
1617		 * enough advance overhead that the latter isn't a problem.
1618		 */
1619		spin_unlock_irqrestore(&ide_lock, flags);
1620		return IRQ_NONE;
1621	}
1622	if (!hwgroup->busy) {
1623		hwgroup->busy = 1;	/* paranoia */
1624		printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1625	}
1626	hwgroup->handler = NULL;
1627	hwgroup->req_gen++;
1628	del_timer(&hwgroup->timer);
1629	spin_unlock(&ide_lock);
1630
1631	/* Some controllers might set DMA INTR no matter DMA or PIO;
1632	 * bmdma status might need to be cleared even for
1633	 * PIO interrupts to prevent spurious/lost irq.
1634	 */
1635	if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1636		/* ide_dma_end() needs bmdma status for error checking.
1637		 * So, skip clearing bmdma status here and leave it
1638		 * to ide_dma_end() if this is dma interrupt.
1639		 */
1640		hwif->ide_dma_clear_irq(drive);
1641
1642	if (drive->unmask)
1643		local_irq_enable_in_hardirq();
1644	/* service this interrupt, may set handler for next interrupt */
1645	startstop = handler(drive);
1646	spin_lock_irq(&ide_lock);
1647
1648	/*
1649	 * Note that handler() may have set things up for another
1650	 * interrupt to occur soon, but it cannot happen until
1651	 * we exit from this routine, because it will be the
1652	 * same irq as is currently being serviced here, and Linux
1653	 * won't allow another of the same (on any CPU) until we return.
1654	 */
1655	drive->service_time = jiffies - drive->service_start;
1656	if (startstop == ide_stopped) {
1657		if (hwgroup->handler == NULL) {	/* paranoia */
1658			hwgroup->busy = 0;
1659			ide_do_request(hwgroup, hwif->irq);
1660		} else {
1661			printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1662				"on exit\n", drive->name);
1663		}
1664	}
1665	spin_unlock_irqrestore(&ide_lock, flags);
1666	return IRQ_HANDLED;
1667}
1668
1669/**
1670 *	ide_init_drive_cmd	-	initialize a drive command request
1671 *	@rq: request object
1672 *
1673 *	Initialize a request before we fill it in and send it down to
1674 *	ide_do_drive_cmd. Commands must be set up by this function. Right
1675 *	now it doesn't do a lot, but if that changes abusers will have a
1676 *	nasty surprise.
1677 */
1678
1679void ide_init_drive_cmd (struct request *rq)
1680{
1681	memset(rq, 0, sizeof(*rq));
1682	rq->cmd_type = REQ_TYPE_ATA_CMD;
1683	rq->ref_count = 1;
1684}
1685
1686EXPORT_SYMBOL(ide_init_drive_cmd);
1687
1688/**
1689 *	ide_do_drive_cmd	-	issue IDE special command
1690 *	@drive: device to issue command
1691 *	@rq: request to issue
1692 *	@action: action for processing
1693 *
1694 *	This function issues a special IDE device request
1695 *	onto the request queue.
1696 *
1697 *	If action is ide_wait, then the rq is queued at the end of the
1698 *	request queue, and the function sleeps until it has been processed.
1699 *	This is for use when invoked from an ioctl handler.
1700 *
1701 *	If action is ide_preempt, then the rq is queued at the head of
1702 *	the request queue, displacing the currently-being-processed
1703 *	request and this function returns immediately without waiting
1704 *	for the new rq to be completed.  This is VERY DANGEROUS, and is
1705 *	intended for careful use by the ATAPI tape/cdrom driver code.
1706 *
1707 *	If action is ide_end, then the rq is queued at the end of the
1708 *	request queue, and the function returns immediately without waiting
1709 *	for the new rq to be completed. This is again intended for careful
1710 *	use by the ATAPI tape/cdrom driver code.
1711 */
1712
1713int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1714{
1715	unsigned long flags;
1716	ide_hwgroup_t *hwgroup = HWGROUP(drive);
1717	DECLARE_COMPLETION_ONSTACK(wait);
1718	int where = ELEVATOR_INSERT_BACK, err;
1719	int must_wait = (action == ide_wait || action == ide_head_wait);
1720
1721	rq->errors = 0;
1722
1723	/*
1724	 * we need to hold an extra reference to request for safe inspection
1725	 * after completion
1726	 */
1727	if (must_wait) {
1728		rq->ref_count++;
1729		rq->end_io_data = &wait;
1730		rq->end_io = blk_end_sync_rq;
1731	}
1732
1733	spin_lock_irqsave(&ide_lock, flags);
1734	if (action == ide_preempt)
1735		hwgroup->rq = NULL;
1736	if (action == ide_preempt || action == ide_head_wait) {
1737		where = ELEVATOR_INSERT_FRONT;
1738		rq->cmd_flags |= REQ_PREEMPT;
1739	}
1740	__elv_add_request(drive->queue, rq, where, 0);
1741	ide_do_request(hwgroup, IDE_NO_IRQ);
1742	spin_unlock_irqrestore(&ide_lock, flags);
1743
1744	err = 0;
1745	if (must_wait) {
1746		wait_for_completion(&wait);
1747		if (rq->errors)
1748			err = -EIO;
1749
1750		blk_put_request(rq);
1751	}
1752
1753	return err;
1754}
1755
1756EXPORT_SYMBOL(ide_do_drive_cmd);
1757