Merge branch 'locks' of git://linux-nfs.org/~bfields/linux
[linux-2.6] / drivers / ide / ide-io.c
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
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, unsigned int nr_bytes)
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_bytes = rq->hard_nr_sectors << 9;
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_chunk(rq, uptodate, nr_bytes)) {
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
104 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
105 {
106         unsigned int nr_bytes = nr_sectors << 9;
107         struct request *rq;
108         unsigned long flags;
109         int ret = 1;
110
111         /*
112          * room for locking improvements here, the calls below don't
113          * need the queue lock held at all
114          */
115         spin_lock_irqsave(&ide_lock, flags);
116         rq = HWGROUP(drive)->rq;
117
118         if (!nr_bytes) {
119                 if (blk_pc_request(rq))
120                         nr_bytes = rq->data_len;
121                 else
122                         nr_bytes = rq->hard_cur_sectors << 9;
123         }
124
125         ret = __ide_end_request(drive, rq, uptodate, nr_bytes);
126
127         spin_unlock_irqrestore(&ide_lock, flags);
128         return ret;
129 }
130 EXPORT_SYMBOL(ide_end_request);
131
132 /*
133  * Power Management state machine. This one is rather trivial for now,
134  * we should probably add more, like switching back to PIO on suspend
135  * to help some BIOSes, re-do the door locking on resume, etc...
136  */
137
138 enum {
139         ide_pm_flush_cache      = ide_pm_state_start_suspend,
140         idedisk_pm_standby,
141
142         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
143         idedisk_pm_idle,
144         ide_pm_restore_dma,
145 };
146
147 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
148 {
149         struct request_pm_state *pm = rq->data;
150
151         if (drive->media != ide_disk)
152                 return;
153
154         switch (pm->pm_step) {
155         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
156                 if (pm->pm_state == PM_EVENT_FREEZE)
157                         pm->pm_step = ide_pm_state_completed;
158                 else
159                         pm->pm_step = idedisk_pm_standby;
160                 break;
161         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
162                 pm->pm_step = ide_pm_state_completed;
163                 break;
164         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
165                 pm->pm_step = idedisk_pm_idle;
166                 break;
167         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
168                 pm->pm_step = ide_pm_restore_dma;
169                 break;
170         }
171 }
172
173 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
174 {
175         struct request_pm_state *pm = rq->data;
176         ide_task_t *args = rq->special;
177
178         memset(args, 0, sizeof(*args));
179
180         switch (pm->pm_step) {
181         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
182                 if (drive->media != ide_disk)
183                         break;
184                 /* Not supported? Switch to next step now. */
185                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
186                         ide_complete_power_step(drive, rq, 0, 0);
187                         return ide_stopped;
188                 }
189                 if (ide_id_has_flush_cache_ext(drive->id))
190                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
191                 else
192                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
193                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
194                 args->handler      = &task_no_data_intr;
195                 return do_rw_taskfile(drive, args);
196
197         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
198                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
199                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
200                 args->handler      = &task_no_data_intr;
201                 return do_rw_taskfile(drive, args);
202
203         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
204                 ide_set_max_pio(drive);
205                 /*
206                  * skip idedisk_pm_idle for ATAPI devices
207                  */
208                 if (drive->media != ide_disk)
209                         pm->pm_step = ide_pm_restore_dma;
210                 else
211                         ide_complete_power_step(drive, rq, 0, 0);
212                 return ide_stopped;
213
214         case idedisk_pm_idle:           /* Resume step 2 (idle) */
215                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
216                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
217                 args->handler = task_no_data_intr;
218                 return do_rw_taskfile(drive, args);
219
220         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
221                 /*
222                  * Right now, all we do is call hwif->ide_dma_check(drive),
223                  * we could be smarter and check for current xfer_speed
224                  * in struct drive etc...
225                  */
226                 if (drive->hwif->ide_dma_check == NULL)
227                         break;
228                 drive->hwif->dma_off_quietly(drive);
229                 /*
230                  * TODO: respect ->using_dma setting
231                  */
232                 ide_set_dma(drive);
233                 break;
234         }
235         pm->pm_step = ide_pm_state_completed;
236         return ide_stopped;
237 }
238
239 /**
240  *      ide_end_dequeued_request        -       complete an IDE I/O
241  *      @drive: IDE device for the I/O
242  *      @uptodate:
243  *      @nr_sectors: number of sectors completed
244  *
245  *      Complete an I/O that is no longer on the request queue. This
246  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
247  *      We must still finish the old request but we must not tamper with the
248  *      queue in the meantime.
249  *
250  *      NOTE: This path does not handle barrier, but barrier is not supported
251  *      on ide-cd anyway.
252  */
253
254 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
255                              int uptodate, int nr_sectors)
256 {
257         unsigned long flags;
258         int ret = 1;
259
260         spin_lock_irqsave(&ide_lock, flags);
261
262         BUG_ON(!blk_rq_started(rq));
263
264         /*
265          * if failfast is set on a request, override number of sectors and
266          * complete the whole request right now
267          */
268         if (blk_noretry_request(rq) && end_io_error(uptodate))
269                 nr_sectors = rq->hard_nr_sectors;
270
271         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
272                 rq->errors = -EIO;
273
274         /*
275          * decide whether to reenable DMA -- 3 is a random magic for now,
276          * if we DMA timeout more than 3 times, just stay in PIO
277          */
278         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
279                 drive->state = 0;
280                 HWGROUP(drive)->hwif->ide_dma_on(drive);
281         }
282
283         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
284                 add_disk_randomness(rq->rq_disk);
285                 if (blk_rq_tagged(rq))
286                         blk_queue_end_tag(drive->queue, rq);
287                 end_that_request_last(rq, uptodate);
288                 ret = 0;
289         }
290         spin_unlock_irqrestore(&ide_lock, flags);
291         return ret;
292 }
293 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
294
295
296 /**
297  *      ide_complete_pm_request - end the current Power Management request
298  *      @drive: target drive
299  *      @rq: request
300  *
301  *      This function cleans up the current PM request and stops the queue
302  *      if necessary.
303  */
304 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
305 {
306         unsigned long flags;
307
308 #ifdef DEBUG_PM
309         printk("%s: completing PM request, %s\n", drive->name,
310                blk_pm_suspend_request(rq) ? "suspend" : "resume");
311 #endif
312         spin_lock_irqsave(&ide_lock, flags);
313         if (blk_pm_suspend_request(rq)) {
314                 blk_stop_queue(drive->queue);
315         } else {
316                 drive->blocked = 0;
317                 blk_start_queue(drive->queue);
318         }
319         blkdev_dequeue_request(rq);
320         HWGROUP(drive)->rq = NULL;
321         end_that_request_last(rq, 1);
322         spin_unlock_irqrestore(&ide_lock, flags);
323 }
324
325 /*
326  * FIXME: probably move this somewhere else, name is bad too :)
327  */
328 u64 ide_get_error_location(ide_drive_t *drive, char *args)
329 {
330         u32 high, low;
331         u8 hcyl, lcyl, sect;
332         u64 sector;
333
334         high = 0;
335         hcyl = args[5];
336         lcyl = args[4];
337         sect = args[3];
338
339         if (ide_id_has_flush_cache_ext(drive->id)) {
340                 low = (hcyl << 16) | (lcyl << 8) | sect;
341                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
342                 high = ide_read_24(drive);
343         } else {
344                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
345                 if (cur & 0x40) {
346                         high = cur & 0xf;
347                         low = (hcyl << 16) | (lcyl << 8) | sect;
348                 } else {
349                         low = hcyl * drive->head * drive->sect;
350                         low += lcyl * drive->sect;
351                         low += sect - 1;
352                 }
353         }
354
355         sector = ((u64) high << 24) | low;
356         return sector;
357 }
358 EXPORT_SYMBOL(ide_get_error_location);
359
360 /**
361  *      ide_end_drive_cmd       -       end an explicit drive command
362  *      @drive: command 
363  *      @stat: status bits
364  *      @err: error bits
365  *
366  *      Clean up after success/failure of an explicit drive command.
367  *      These get thrown onto the queue so they are synchronized with
368  *      real I/O operations on the drive.
369  *
370  *      In LBA48 mode we have to read the register set twice to get
371  *      all the extra information out.
372  */
373  
374 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
375 {
376         ide_hwif_t *hwif = HWIF(drive);
377         unsigned long flags;
378         struct request *rq;
379
380         spin_lock_irqsave(&ide_lock, flags);
381         rq = HWGROUP(drive)->rq;
382         spin_unlock_irqrestore(&ide_lock, flags);
383
384         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
385                 u8 *args = (u8 *) rq->buffer;
386                 if (rq->errors == 0)
387                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
388
389                 if (args) {
390                         args[0] = stat;
391                         args[1] = err;
392                         args[2] = hwif->INB(IDE_NSECTOR_REG);
393                 }
394         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
395                 u8 *args = (u8 *) rq->buffer;
396                 if (rq->errors == 0)
397                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
398
399                 if (args) {
400                         args[0] = stat;
401                         args[1] = err;
402                         args[2] = hwif->INB(IDE_NSECTOR_REG);
403                         args[3] = hwif->INB(IDE_SECTOR_REG);
404                         args[4] = hwif->INB(IDE_LCYL_REG);
405                         args[5] = hwif->INB(IDE_HCYL_REG);
406                         args[6] = hwif->INB(IDE_SELECT_REG);
407                 }
408         } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
409                 ide_task_t *args = (ide_task_t *) rq->special;
410                 if (rq->errors == 0)
411                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
412                         
413                 if (args) {
414                         if (args->tf_in_flags.b.data) {
415                                 u16 data                                = hwif->INW(IDE_DATA_REG);
416                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
417                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
418                         }
419                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
420                         /* be sure we're looking at the low order bits */
421                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
422                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
423                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
424                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
425                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
426                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
427                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
428
429                         if (drive->addressing == 1) {
430                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
431                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
432                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
433                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
434                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
435                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
436                         }
437                 }
438         } else if (blk_pm_request(rq)) {
439                 struct request_pm_state *pm = rq->data;
440 #ifdef DEBUG_PM
441                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
442                         drive->name, rq->pm->pm_step, stat, err);
443 #endif
444                 ide_complete_power_step(drive, rq, stat, err);
445                 if (pm->pm_step == ide_pm_state_completed)
446                         ide_complete_pm_request(drive, rq);
447                 return;
448         }
449
450         spin_lock_irqsave(&ide_lock, flags);
451         blkdev_dequeue_request(rq);
452         HWGROUP(drive)->rq = NULL;
453         rq->errors = err;
454         end_that_request_last(rq, !rq->errors);
455         spin_unlock_irqrestore(&ide_lock, flags);
456 }
457
458 EXPORT_SYMBOL(ide_end_drive_cmd);
459
460 /**
461  *      try_to_flush_leftover_data      -       flush junk
462  *      @drive: drive to flush
463  *
464  *      try_to_flush_leftover_data() is invoked in response to a drive
465  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
466  *      resetting the drive, this routine tries to clear the condition
467  *      by read a sector's worth of data from the drive.  Of course,
468  *      this may not help if the drive is *waiting* for data from *us*.
469  */
470 static void try_to_flush_leftover_data (ide_drive_t *drive)
471 {
472         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
473
474         if (drive->media != ide_disk)
475                 return;
476         while (i > 0) {
477                 u32 buffer[16];
478                 u32 wcount = (i > 16) ? 16 : i;
479
480                 i -= wcount;
481                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
482         }
483 }
484
485 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
486 {
487         if (rq->rq_disk) {
488                 ide_driver_t *drv;
489
490                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
491                 drv->end_request(drive, 0, 0);
492         } else
493                 ide_end_request(drive, 0, 0);
494 }
495
496 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
497 {
498         ide_hwif_t *hwif = drive->hwif;
499
500         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
501                 /* other bits are useless when BUSY */
502                 rq->errors |= ERROR_RESET;
503         } else if (stat & ERR_STAT) {
504                 /* err has different meaning on cdrom and tape */
505                 if (err == ABRT_ERR) {
506                         if (drive->select.b.lba &&
507                             /* some newer drives don't support WIN_SPECIFY */
508                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
509                                 return ide_stopped;
510                 } else if ((err & BAD_CRC) == BAD_CRC) {
511                         /* UDMA crc error, just retry the operation */
512                         drive->crc_count++;
513                 } else if (err & (BBD_ERR | ECC_ERR)) {
514                         /* retries won't help these */
515                         rq->errors = ERROR_MAX;
516                 } else if (err & TRK0_ERR) {
517                         /* help it find track zero */
518                         rq->errors |= ERROR_RECAL;
519                 }
520         }
521
522         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && hwif->err_stops_fifo == 0)
523                 try_to_flush_leftover_data(drive);
524
525         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
526                 ide_kill_rq(drive, rq);
527                 return ide_stopped;
528         }
529
530         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
531                 rq->errors |= ERROR_RESET;
532
533         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
534                 ++rq->errors;
535                 return ide_do_reset(drive);
536         }
537
538         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
539                 drive->special.b.recalibrate = 1;
540
541         ++rq->errors;
542
543         return ide_stopped;
544 }
545
546 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
547 {
548         ide_hwif_t *hwif = drive->hwif;
549
550         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
551                 /* other bits are useless when BUSY */
552                 rq->errors |= ERROR_RESET;
553         } else {
554                 /* add decoding error stuff */
555         }
556
557         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
558                 /* force an abort */
559                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
560
561         if (rq->errors >= ERROR_MAX) {
562                 ide_kill_rq(drive, rq);
563         } else {
564                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
565                         ++rq->errors;
566                         return ide_do_reset(drive);
567                 }
568                 ++rq->errors;
569         }
570
571         return ide_stopped;
572 }
573
574 ide_startstop_t
575 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
576 {
577         if (drive->media == ide_disk)
578                 return ide_ata_error(drive, rq, stat, err);
579         return ide_atapi_error(drive, rq, stat, err);
580 }
581
582 EXPORT_SYMBOL_GPL(__ide_error);
583
584 /**
585  *      ide_error       -       handle an error on the IDE
586  *      @drive: drive the error occurred on
587  *      @msg: message to report
588  *      @stat: status bits
589  *
590  *      ide_error() takes action based on the error returned by the drive.
591  *      For normal I/O that may well include retries. We deal with
592  *      both new-style (taskfile) and old style command handling here.
593  *      In the case of taskfile command handling there is work left to
594  *      do
595  */
596  
597 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
598 {
599         struct request *rq;
600         u8 err;
601
602         err = ide_dump_status(drive, msg, stat);
603
604         if ((rq = HWGROUP(drive)->rq) == NULL)
605                 return ide_stopped;
606
607         /* retry only "normal" I/O: */
608         if (!blk_fs_request(rq)) {
609                 rq->errors = 1;
610                 ide_end_drive_cmd(drive, stat, err);
611                 return ide_stopped;
612         }
613
614         if (rq->rq_disk) {
615                 ide_driver_t *drv;
616
617                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
618                 return drv->error(drive, rq, stat, err);
619         } else
620                 return __ide_error(drive, rq, stat, err);
621 }
622
623 EXPORT_SYMBOL_GPL(ide_error);
624
625 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
626 {
627         if (drive->media != ide_disk)
628                 rq->errors |= ERROR_RESET;
629
630         ide_kill_rq(drive, rq);
631
632         return ide_stopped;
633 }
634
635 EXPORT_SYMBOL_GPL(__ide_abort);
636
637 /**
638  *      ide_abort       -       abort pending IDE operations
639  *      @drive: drive the error occurred on
640  *      @msg: message to report
641  *
642  *      ide_abort kills and cleans up when we are about to do a 
643  *      host initiated reset on active commands. Longer term we
644  *      want handlers to have sensible abort handling themselves
645  *
646  *      This differs fundamentally from ide_error because in 
647  *      this case the command is doing just fine when we
648  *      blow it away.
649  */
650  
651 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
652 {
653         struct request *rq;
654
655         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
656                 return ide_stopped;
657
658         /* retry only "normal" I/O: */
659         if (!blk_fs_request(rq)) {
660                 rq->errors = 1;
661                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
662                 return ide_stopped;
663         }
664
665         if (rq->rq_disk) {
666                 ide_driver_t *drv;
667
668                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
669                 return drv->abort(drive, rq);
670         } else
671                 return __ide_abort(drive, rq);
672 }
673
674 /**
675  *      ide_cmd         -       issue a simple drive command
676  *      @drive: drive the command is for
677  *      @cmd: command byte
678  *      @nsect: sector byte
679  *      @handler: handler for the command completion
680  *
681  *      Issue a simple drive command with interrupts.
682  *      The drive must be selected beforehand.
683  */
684
685 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
686                 ide_handler_t *handler)
687 {
688         ide_hwif_t *hwif = HWIF(drive);
689         if (IDE_CONTROL_REG)
690                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
691         SELECT_MASK(drive,0);
692         hwif->OUTB(nsect,IDE_NSECTOR_REG);
693         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
694 }
695
696 /**
697  *      drive_cmd_intr          -       drive command completion interrupt
698  *      @drive: drive the completion interrupt occurred on
699  *
700  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
701  *      We do any necessary data reading and then wait for the drive to
702  *      go non busy. At that point we may read the error data and complete
703  *      the request
704  */
705  
706 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
707 {
708         struct request *rq = HWGROUP(drive)->rq;
709         ide_hwif_t *hwif = HWIF(drive);
710         u8 *args = (u8 *) rq->buffer;
711         u8 stat = hwif->INB(IDE_STATUS_REG);
712         int retries = 10;
713
714         local_irq_enable_in_hardirq();
715         if ((stat & DRQ_STAT) && args && args[3]) {
716                 u8 io_32bit = drive->io_32bit;
717                 drive->io_32bit = 0;
718                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
719                 drive->io_32bit = io_32bit;
720                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
721                         udelay(100);
722         }
723
724         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
725                 return ide_error(drive, "drive_cmd", stat);
726                 /* calls ide_end_drive_cmd */
727         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
728         return ide_stopped;
729 }
730
731 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
732 {
733         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
734         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
735         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
736         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
737         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
738         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
739
740         task->handler = &set_geometry_intr;
741 }
742
743 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
744 {
745         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
746         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
747
748         task->handler = &recal_intr;
749 }
750
751 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
752 {
753         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
754         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
755
756         task->handler = &set_multmode_intr;
757 }
758
759 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
760 {
761         special_t *s = &drive->special;
762         ide_task_t args;
763
764         memset(&args, 0, sizeof(ide_task_t));
765         args.command_type = IDE_DRIVE_TASK_NO_DATA;
766
767         if (s->b.set_geometry) {
768                 s->b.set_geometry = 0;
769                 ide_init_specify_cmd(drive, &args);
770         } else if (s->b.recalibrate) {
771                 s->b.recalibrate = 0;
772                 ide_init_restore_cmd(drive, &args);
773         } else if (s->b.set_multmode) {
774                 s->b.set_multmode = 0;
775                 if (drive->mult_req > drive->id->max_multsect)
776                         drive->mult_req = drive->id->max_multsect;
777                 ide_init_setmult_cmd(drive, &args);
778         } else if (s->all) {
779                 int special = s->all;
780                 s->all = 0;
781                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
782                 return ide_stopped;
783         }
784
785         do_rw_taskfile(drive, &args);
786
787         return ide_started;
788 }
789
790 /*
791  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
792  */
793 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
794 {
795         switch (req_pio) {
796         case 202:
797         case 201:
798         case 200:
799         case 102:
800         case 101:
801         case 100:
802                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
803         case 9:
804         case 8:
805                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
806         case 7:
807         case 6:
808                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
809         default:
810                 return 0;
811         }
812 }
813
814 /**
815  *      do_special              -       issue some special commands
816  *      @drive: drive the command is for
817  *
818  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
819  *      commands to a drive.  It used to do much more, but has been scaled
820  *      back.
821  */
822
823 static ide_startstop_t do_special (ide_drive_t *drive)
824 {
825         special_t *s = &drive->special;
826
827 #ifdef DEBUG
828         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
829 #endif
830         if (s->b.set_tune) {
831                 ide_hwif_t *hwif = drive->hwif;
832                 u8 req_pio = drive->tune_req;
833
834                 s->b.set_tune = 0;
835
836                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
837                         if (hwif->set_pio_mode)
838                                 hwif->set_pio_mode(drive, req_pio);
839                 } else {
840                         int keep_dma = drive->using_dma;
841
842                         ide_set_pio(drive, req_pio);
843
844                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
845                                 if (keep_dma)
846                                         hwif->ide_dma_on(drive);
847                         }
848                 }
849
850                 return ide_stopped;
851         } else {
852                 if (drive->media == ide_disk)
853                         return ide_disk_special(drive);
854
855                 s->all = 0;
856                 drive->mult_req = 0;
857                 return ide_stopped;
858         }
859 }
860
861 void ide_map_sg(ide_drive_t *drive, struct request *rq)
862 {
863         ide_hwif_t *hwif = drive->hwif;
864         struct scatterlist *sg = hwif->sg_table;
865
866         if (hwif->sg_mapped)    /* needed by ide-scsi */
867                 return;
868
869         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
870                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
871         } else {
872                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
873                 hwif->sg_nents = 1;
874         }
875 }
876
877 EXPORT_SYMBOL_GPL(ide_map_sg);
878
879 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
880 {
881         ide_hwif_t *hwif = drive->hwif;
882
883         hwif->nsect = hwif->nleft = rq->nr_sectors;
884         hwif->cursg = hwif->cursg_ofs = 0;
885 }
886
887 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
888
889 /**
890  *      execute_drive_command   -       issue special drive command
891  *      @drive: the drive to issue the command on
892  *      @rq: the request structure holding the command
893  *
894  *      execute_drive_cmd() issues a special drive command,  usually 
895  *      initiated by ioctl() from the external hdparm program. The
896  *      command can be a drive command, drive task or taskfile 
897  *      operation. Weirdly you can call it with NULL to wait for
898  *      all commands to finish. Don't do this as that is due to change
899  */
900
901 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
902                 struct request *rq)
903 {
904         ide_hwif_t *hwif = HWIF(drive);
905         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
906                 ide_task_t *args = rq->special;
907  
908                 if (!args)
909                         goto done;
910
911                 hwif->data_phase = args->data_phase;
912
913                 switch (hwif->data_phase) {
914                 case TASKFILE_MULTI_OUT:
915                 case TASKFILE_OUT:
916                 case TASKFILE_MULTI_IN:
917                 case TASKFILE_IN:
918                         ide_init_sg_cmd(drive, rq);
919                         ide_map_sg(drive, rq);
920                 default:
921                         break;
922                 }
923
924                 if (args->tf_out_flags.all != 0) 
925                         return flagged_taskfile(drive, args);
926                 return do_rw_taskfile(drive, args);
927         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
928                 u8 *args = rq->buffer;
929                 u8 sel;
930  
931                 if (!args)
932                         goto done;
933 #ifdef DEBUG
934                 printk("%s: DRIVE_TASK_CMD ", drive->name);
935                 printk("cmd=0x%02x ", args[0]);
936                 printk("fr=0x%02x ", args[1]);
937                 printk("ns=0x%02x ", args[2]);
938                 printk("sc=0x%02x ", args[3]);
939                 printk("lcyl=0x%02x ", args[4]);
940                 printk("hcyl=0x%02x ", args[5]);
941                 printk("sel=0x%02x\n", args[6]);
942 #endif
943                 hwif->OUTB(args[1], IDE_FEATURE_REG);
944                 hwif->OUTB(args[3], IDE_SECTOR_REG);
945                 hwif->OUTB(args[4], IDE_LCYL_REG);
946                 hwif->OUTB(args[5], IDE_HCYL_REG);
947                 sel = (args[6] & ~0x10);
948                 if (drive->select.b.unit)
949                         sel |= 0x10;
950                 hwif->OUTB(sel, IDE_SELECT_REG);
951                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
952                 return ide_started;
953         } else if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
954                 u8 *args = rq->buffer;
955
956                 if (!args)
957                         goto done;
958 #ifdef DEBUG
959                 printk("%s: DRIVE_CMD ", drive->name);
960                 printk("cmd=0x%02x ", args[0]);
961                 printk("sc=0x%02x ", args[1]);
962                 printk("fr=0x%02x ", args[2]);
963                 printk("xx=0x%02x\n", args[3]);
964 #endif
965                 if (args[0] == WIN_SMART) {
966                         hwif->OUTB(0x4f, IDE_LCYL_REG);
967                         hwif->OUTB(0xc2, IDE_HCYL_REG);
968                         hwif->OUTB(args[2],IDE_FEATURE_REG);
969                         hwif->OUTB(args[1],IDE_SECTOR_REG);
970                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
971                         return ide_started;
972                 }
973                 hwif->OUTB(args[2],IDE_FEATURE_REG);
974                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
975                 return ide_started;
976         }
977
978 done:
979         /*
980          * NULL is actually a valid way of waiting for
981          * all current requests to be flushed from the queue.
982          */
983 #ifdef DEBUG
984         printk("%s: DRIVE_CMD (null)\n", drive->name);
985 #endif
986         ide_end_drive_cmd(drive,
987                         hwif->INB(IDE_STATUS_REG),
988                         hwif->INB(IDE_ERROR_REG));
989         return ide_stopped;
990 }
991
992 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
993 {
994         struct request_pm_state *pm = rq->data;
995
996         if (blk_pm_suspend_request(rq) &&
997             pm->pm_step == ide_pm_state_start_suspend)
998                 /* Mark drive blocked when starting the suspend sequence. */
999                 drive->blocked = 1;
1000         else if (blk_pm_resume_request(rq) &&
1001                  pm->pm_step == ide_pm_state_start_resume) {
1002                 /* 
1003                  * The first thing we do on wakeup is to wait for BSY bit to
1004                  * go away (with a looong timeout) as a drive on this hwif may
1005                  * just be POSTing itself.
1006                  * We do that before even selecting as the "other" device on
1007                  * the bus may be broken enough to walk on our toes at this
1008                  * point.
1009                  */
1010                 int rc;
1011 #ifdef DEBUG_PM
1012                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
1013 #endif
1014                 rc = ide_wait_not_busy(HWIF(drive), 35000);
1015                 if (rc)
1016                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
1017                 SELECT_DRIVE(drive);
1018                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
1019                 rc = ide_wait_not_busy(HWIF(drive), 100000);
1020                 if (rc)
1021                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
1022         }
1023 }
1024
1025 /**
1026  *      start_request   -       start of I/O and command issuing for IDE
1027  *
1028  *      start_request() initiates handling of a new I/O request. It
1029  *      accepts commands and I/O (read/write) requests. It also does
1030  *      the final remapping for weird stuff like EZDrive. Once 
1031  *      device mapper can work sector level the EZDrive stuff can go away
1032  *
1033  *      FIXME: this function needs a rename
1034  */
1035  
1036 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
1037 {
1038         ide_startstop_t startstop;
1039         sector_t block;
1040
1041         BUG_ON(!blk_rq_started(rq));
1042
1043 #ifdef DEBUG
1044         printk("%s: start_request: current=0x%08lx\n",
1045                 HWIF(drive)->name, (unsigned long) rq);
1046 #endif
1047
1048         /* bail early if we've exceeded max_failures */
1049         if (drive->max_failures && (drive->failures > drive->max_failures)) {
1050                 goto kill_rq;
1051         }
1052
1053         block    = rq->sector;
1054         if (blk_fs_request(rq) &&
1055             (drive->media == ide_disk || drive->media == ide_floppy)) {
1056                 block += drive->sect0;
1057         }
1058         /* Yecch - this will shift the entire interval,
1059            possibly killing some innocent following sector */
1060         if (block == 0 && drive->remap_0_to_1 == 1)
1061                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1062
1063         if (blk_pm_request(rq))
1064                 ide_check_pm_state(drive, rq);
1065
1066         SELECT_DRIVE(drive);
1067         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1068                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1069                 return startstop;
1070         }
1071         if (!drive->special.all) {
1072                 ide_driver_t *drv;
1073
1074                 /*
1075                  * We reset the drive so we need to issue a SETFEATURES.
1076                  * Do it _after_ do_special() restored device parameters.
1077                  */
1078                 if (drive->current_speed == 0xff)
1079                         ide_config_drive_speed(drive, drive->desired_speed);
1080
1081                 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1082                     rq->cmd_type == REQ_TYPE_ATA_TASK ||
1083                     rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1084                         return execute_drive_cmd(drive, rq);
1085                 else if (blk_pm_request(rq)) {
1086                         struct request_pm_state *pm = rq->data;
1087 #ifdef DEBUG_PM
1088                         printk("%s: start_power_step(step: %d)\n",
1089                                 drive->name, rq->pm->pm_step);
1090 #endif
1091                         startstop = ide_start_power_step(drive, rq);
1092                         if (startstop == ide_stopped &&
1093                             pm->pm_step == ide_pm_state_completed)
1094                                 ide_complete_pm_request(drive, rq);
1095                         return startstop;
1096                 }
1097
1098                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1099                 return drv->do_request(drive, rq, block);
1100         }
1101         return do_special(drive);
1102 kill_rq:
1103         ide_kill_rq(drive, rq);
1104         return ide_stopped;
1105 }
1106
1107 /**
1108  *      ide_stall_queue         -       pause an IDE device
1109  *      @drive: drive to stall
1110  *      @timeout: time to stall for (jiffies)
1111  *
1112  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1113  *      to the hwgroup by sleeping for timeout jiffies.
1114  */
1115  
1116 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1117 {
1118         if (timeout > WAIT_WORSTCASE)
1119                 timeout = WAIT_WORSTCASE;
1120         drive->sleep = timeout + jiffies;
1121         drive->sleeping = 1;
1122 }
1123
1124 EXPORT_SYMBOL(ide_stall_queue);
1125
1126 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1127
1128 /**
1129  *      choose_drive            -       select a drive to service
1130  *      @hwgroup: hardware group to select on
1131  *
1132  *      choose_drive() selects the next drive which will be serviced.
1133  *      This is necessary because the IDE layer can't issue commands
1134  *      to both drives on the same cable, unlike SCSI.
1135  */
1136  
1137 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1138 {
1139         ide_drive_t *drive, *best;
1140
1141 repeat: 
1142         best = NULL;
1143         drive = hwgroup->drive;
1144
1145         /*
1146          * drive is doing pre-flush, ordered write, post-flush sequence. even
1147          * though that is 3 requests, it must be seen as a single transaction.
1148          * we must not preempt this drive until that is complete
1149          */
1150         if (blk_queue_flushing(drive->queue)) {
1151                 /*
1152                  * small race where queue could get replugged during
1153                  * the 3-request flush cycle, just yank the plug since
1154                  * we want it to finish asap
1155                  */
1156                 blk_remove_plug(drive->queue);
1157                 return drive;
1158         }
1159
1160         do {
1161                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1162                     && !elv_queue_empty(drive->queue)) {
1163                         if (!best
1164                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1165                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1166                         {
1167                                 if (!blk_queue_plugged(drive->queue))
1168                                         best = drive;
1169                         }
1170                 }
1171         } while ((drive = drive->next) != hwgroup->drive);
1172         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1173                 long t = (signed long)(WAKEUP(best) - jiffies);
1174                 if (t >= WAIT_MIN_SLEEP) {
1175                 /*
1176                  * We *may* have some time to spare, but first let's see if
1177                  * someone can potentially benefit from our nice mood today..
1178                  */
1179                         drive = best->next;
1180                         do {
1181                                 if (!drive->sleeping
1182                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1183                                  && time_before(WAKEUP(drive), jiffies + t))
1184                                 {
1185                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1186                                         goto repeat;
1187                                 }
1188                         } while ((drive = drive->next) != best);
1189                 }
1190         }
1191         return best;
1192 }
1193
1194 /*
1195  * Issue a new request to a drive from hwgroup
1196  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1197  *
1198  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1199  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1200  * may have both interfaces in a single hwgroup to "serialize" access.
1201  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1202  * together into one hwgroup for serialized access.
1203  *
1204  * Note also that several hwgroups can end up sharing a single IRQ,
1205  * possibly along with many other devices.  This is especially common in
1206  * PCI-based systems with off-board IDE controller cards.
1207  *
1208  * The IDE driver uses the single global ide_lock spinlock to protect
1209  * access to the request queues, and to protect the hwgroup->busy flag.
1210  *
1211  * The first thread into the driver for a particular hwgroup sets the
1212  * hwgroup->busy flag to indicate that this hwgroup is now active,
1213  * and then initiates processing of the top request from the request queue.
1214  *
1215  * Other threads attempting entry notice the busy setting, and will simply
1216  * queue their new requests and exit immediately.  Note that hwgroup->busy
1217  * remains set even when the driver is merely awaiting the next interrupt.
1218  * Thus, the meaning is "this hwgroup is busy processing a request".
1219  *
1220  * When processing of a request completes, the completing thread or IRQ-handler
1221  * will start the next request from the queue.  If no more work remains,
1222  * the driver will clear the hwgroup->busy flag and exit.
1223  *
1224  * The ide_lock (spinlock) is used to protect all access to the
1225  * hwgroup->busy flag, but is otherwise not needed for most processing in
1226  * the driver.  This makes the driver much more friendlier to shared IRQs
1227  * than previous designs, while remaining 100% (?) SMP safe and capable.
1228  */
1229 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1230 {
1231         ide_drive_t     *drive;
1232         ide_hwif_t      *hwif;
1233         struct request  *rq;
1234         ide_startstop_t startstop;
1235         int             loops = 0;
1236
1237         /* for atari only: POSSIBLY BROKEN HERE(?) */
1238         ide_get_lock(ide_intr, hwgroup);
1239
1240         /* caller must own ide_lock */
1241         BUG_ON(!irqs_disabled());
1242
1243         while (!hwgroup->busy) {
1244                 hwgroup->busy = 1;
1245                 drive = choose_drive(hwgroup);
1246                 if (drive == NULL) {
1247                         int sleeping = 0;
1248                         unsigned long sleep = 0; /* shut up, gcc */
1249                         hwgroup->rq = NULL;
1250                         drive = hwgroup->drive;
1251                         do {
1252                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1253                                         sleeping = 1;
1254                                         sleep = drive->sleep;
1255                                 }
1256                         } while ((drive = drive->next) != hwgroup->drive);
1257                         if (sleeping) {
1258                 /*
1259                  * Take a short snooze, and then wake up this hwgroup again.
1260                  * This gives other hwgroups on the same a chance to
1261                  * play fairly with us, just in case there are big differences
1262                  * in relative throughputs.. don't want to hog the cpu too much.
1263                  */
1264                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1265                                         sleep = jiffies + WAIT_MIN_SLEEP;
1266 #if 1
1267                                 if (timer_pending(&hwgroup->timer))
1268                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1269 #endif
1270                                 /* so that ide_timer_expiry knows what to do */
1271                                 hwgroup->sleeping = 1;
1272                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1273                                 mod_timer(&hwgroup->timer, sleep);
1274                                 /* we purposely leave hwgroup->busy==1
1275                                  * while sleeping */
1276                         } else {
1277                                 /* Ugly, but how can we sleep for the lock
1278                                  * otherwise? perhaps from tq_disk?
1279                                  */
1280
1281                                 /* for atari only */
1282                                 ide_release_lock();
1283                                 hwgroup->busy = 0;
1284                         }
1285
1286                         /* no more work for this hwgroup (for now) */
1287                         return;
1288                 }
1289         again:
1290                 hwif = HWIF(drive);
1291                 if (hwgroup->hwif->sharing_irq &&
1292                     hwif != hwgroup->hwif &&
1293                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1294                         /* set nIEN for previous hwif */
1295                         SELECT_INTERRUPT(drive);
1296                 }
1297                 hwgroup->hwif = hwif;
1298                 hwgroup->drive = drive;
1299                 drive->sleeping = 0;
1300                 drive->service_start = jiffies;
1301
1302                 if (blk_queue_plugged(drive->queue)) {
1303                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1304                         break;
1305                 }
1306
1307                 /*
1308                  * we know that the queue isn't empty, but this can happen
1309                  * if the q->prep_rq_fn() decides to kill a request
1310                  */
1311                 rq = elv_next_request(drive->queue);
1312                 if (!rq) {
1313                         hwgroup->busy = 0;
1314                         break;
1315                 }
1316
1317                 /*
1318                  * Sanity: don't accept a request that isn't a PM request
1319                  * if we are currently power managed. This is very important as
1320                  * blk_stop_queue() doesn't prevent the elv_next_request()
1321                  * above to return us whatever is in the queue. Since we call
1322                  * ide_do_request() ourselves, we end up taking requests while
1323                  * the queue is blocked...
1324                  * 
1325                  * We let requests forced at head of queue with ide-preempt
1326                  * though. I hope that doesn't happen too much, hopefully not
1327                  * unless the subdriver triggers such a thing in its own PM
1328                  * state machine.
1329                  *
1330                  * We count how many times we loop here to make sure we service
1331                  * all drives in the hwgroup without looping for ever
1332                  */
1333                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1334                         drive = drive->next ? drive->next : hwgroup->drive;
1335                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1336                                 goto again;
1337                         /* We clear busy, there should be no pending ATA command at this point. */
1338                         hwgroup->busy = 0;
1339                         break;
1340                 }
1341
1342                 hwgroup->rq = rq;
1343
1344                 /*
1345                  * Some systems have trouble with IDE IRQs arriving while
1346                  * the driver is still setting things up.  So, here we disable
1347                  * the IRQ used by this interface while the request is being started.
1348                  * This may look bad at first, but pretty much the same thing
1349                  * happens anyway when any interrupt comes in, IDE or otherwise
1350                  *  -- the kernel masks the IRQ while it is being handled.
1351                  */
1352                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1353                         disable_irq_nosync(hwif->irq);
1354                 spin_unlock(&ide_lock);
1355                 local_irq_enable_in_hardirq();
1356                         /* allow other IRQs while we start this request */
1357                 startstop = start_request(drive, rq);
1358                 spin_lock_irq(&ide_lock);
1359                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1360                         enable_irq(hwif->irq);
1361                 if (startstop == ide_stopped)
1362                         hwgroup->busy = 0;
1363         }
1364 }
1365
1366 /*
1367  * Passes the stuff to ide_do_request
1368  */
1369 void do_ide_request(struct request_queue *q)
1370 {
1371         ide_drive_t *drive = q->queuedata;
1372
1373         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1374 }
1375
1376 /*
1377  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1378  * retry the current request in pio mode instead of risking tossing it
1379  * all away
1380  */
1381 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1382 {
1383         ide_hwif_t *hwif = HWIF(drive);
1384         struct request *rq;
1385         ide_startstop_t ret = ide_stopped;
1386
1387         /*
1388          * end current dma transaction
1389          */
1390
1391         if (error < 0) {
1392                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1393                 (void)HWIF(drive)->ide_dma_end(drive);
1394                 ret = ide_error(drive, "dma timeout error",
1395                                                 hwif->INB(IDE_STATUS_REG));
1396         } else {
1397                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1398                 hwif->dma_timeout(drive);
1399         }
1400
1401         /*
1402          * disable dma for now, but remember that we did so because of
1403          * a timeout -- we'll reenable after we finish this next request
1404          * (or rather the first chunk of it) in pio.
1405          */
1406         drive->retry_pio++;
1407         drive->state = DMA_PIO_RETRY;
1408         hwif->dma_off_quietly(drive);
1409
1410         /*
1411          * un-busy drive etc (hwgroup->busy is cleared on return) and
1412          * make sure request is sane
1413          */
1414         rq = HWGROUP(drive)->rq;
1415
1416         if (!rq)
1417                 goto out;
1418
1419         HWGROUP(drive)->rq = NULL;
1420
1421         rq->errors = 0;
1422
1423         if (!rq->bio)
1424                 goto out;
1425
1426         rq->sector = rq->bio->bi_sector;
1427         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1428         rq->hard_cur_sectors = rq->current_nr_sectors;
1429         rq->buffer = bio_data(rq->bio);
1430 out:
1431         return ret;
1432 }
1433
1434 /**
1435  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1436  *      @data: timer callback magic (hwgroup)
1437  *
1438  *      An IDE command has timed out before the expected drive return
1439  *      occurred. At this point we attempt to clean up the current
1440  *      mess. If the current handler includes an expiry handler then
1441  *      we invoke the expiry handler, and providing it is happy the
1442  *      work is done. If that fails we apply generic recovery rules
1443  *      invoking the handler and checking the drive DMA status. We
1444  *      have an excessively incestuous relationship with the DMA
1445  *      logic that wants cleaning up.
1446  */
1447  
1448 void ide_timer_expiry (unsigned long data)
1449 {
1450         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1451         ide_handler_t   *handler;
1452         ide_expiry_t    *expiry;
1453         unsigned long   flags;
1454         unsigned long   wait = -1;
1455
1456         spin_lock_irqsave(&ide_lock, flags);
1457
1458         if (((handler = hwgroup->handler) == NULL) ||
1459             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1460                 /*
1461                  * Either a marginal timeout occurred
1462                  * (got the interrupt just as timer expired),
1463                  * or we were "sleeping" to give other devices a chance.
1464                  * Either way, we don't really want to complain about anything.
1465                  */
1466                 if (hwgroup->sleeping) {
1467                         hwgroup->sleeping = 0;
1468                         hwgroup->busy = 0;
1469                 }
1470         } else {
1471                 ide_drive_t *drive = hwgroup->drive;
1472                 if (!drive) {
1473                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1474                         hwgroup->handler = NULL;
1475                 } else {
1476                         ide_hwif_t *hwif;
1477                         ide_startstop_t startstop = ide_stopped;
1478                         if (!hwgroup->busy) {
1479                                 hwgroup->busy = 1;      /* paranoia */
1480                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1481                         }
1482                         if ((expiry = hwgroup->expiry) != NULL) {
1483                                 /* continue */
1484                                 if ((wait = expiry(drive)) > 0) {
1485                                         /* reset timer */
1486                                         hwgroup->timer.expires  = jiffies + wait;
1487                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1488                                         add_timer(&hwgroup->timer);
1489                                         spin_unlock_irqrestore(&ide_lock, flags);
1490                                         return;
1491                                 }
1492                         }
1493                         hwgroup->handler = NULL;
1494                         /*
1495                          * We need to simulate a real interrupt when invoking
1496                          * the handler() function, which means we need to
1497                          * globally mask the specific IRQ:
1498                          */
1499                         spin_unlock(&ide_lock);
1500                         hwif  = HWIF(drive);
1501 #if DISABLE_IRQ_NOSYNC
1502                         disable_irq_nosync(hwif->irq);
1503 #else
1504                         /* disable_irq_nosync ?? */
1505                         disable_irq(hwif->irq);
1506 #endif /* DISABLE_IRQ_NOSYNC */
1507                         /* local CPU only,
1508                          * as if we were handling an interrupt */
1509                         local_irq_disable();
1510                         if (hwgroup->polling) {
1511                                 startstop = handler(drive);
1512                         } else if (drive_is_ready(drive)) {
1513                                 if (drive->waiting_for_dma)
1514                                         hwgroup->hwif->dma_lost_irq(drive);
1515                                 (void)ide_ack_intr(hwif);
1516                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1517                                 startstop = handler(drive);
1518                         } else {
1519                                 if (drive->waiting_for_dma) {
1520                                         startstop = ide_dma_timeout_retry(drive, wait);
1521                                 } else
1522                                         startstop =
1523                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1524                         }
1525                         drive->service_time = jiffies - drive->service_start;
1526                         spin_lock_irq(&ide_lock);
1527                         enable_irq(hwif->irq);
1528                         if (startstop == ide_stopped)
1529                                 hwgroup->busy = 0;
1530                 }
1531         }
1532         ide_do_request(hwgroup, IDE_NO_IRQ);
1533         spin_unlock_irqrestore(&ide_lock, flags);
1534 }
1535
1536 /**
1537  *      unexpected_intr         -       handle an unexpected IDE interrupt
1538  *      @irq: interrupt line
1539  *      @hwgroup: hwgroup being processed
1540  *
1541  *      There's nothing really useful we can do with an unexpected interrupt,
1542  *      other than reading the status register (to clear it), and logging it.
1543  *      There should be no way that an irq can happen before we're ready for it,
1544  *      so we needn't worry much about losing an "important" interrupt here.
1545  *
1546  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1547  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1548  *      looks "good", we just ignore the interrupt completely.
1549  *
1550  *      This routine assumes __cli() is in effect when called.
1551  *
1552  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1553  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1554  *      we could screw up by interfering with a new request being set up for 
1555  *      irq15.
1556  *
1557  *      In reality, this is a non-issue.  The new command is not sent unless 
1558  *      the drive is ready to accept one, in which case we know the drive is
1559  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1560  *      before completing the issuance of any new drive command, so we will not
1561  *      be accidentally invoked as a result of any valid command completion
1562  *      interrupt.
1563  *
1564  *      Note that we must walk the entire hwgroup here. We know which hwif
1565  *      is doing the current command, but we don't know which hwif burped
1566  *      mysteriously.
1567  */
1568  
1569 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1570 {
1571         u8 stat;
1572         ide_hwif_t *hwif = hwgroup->hwif;
1573
1574         /*
1575          * handle the unexpected interrupt
1576          */
1577         do {
1578                 if (hwif->irq == irq) {
1579                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1580                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1581                                 /* Try to not flood the console with msgs */
1582                                 static unsigned long last_msgtime, count;
1583                                 ++count;
1584                                 if (time_after(jiffies, last_msgtime + HZ)) {
1585                                         last_msgtime = jiffies;
1586                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1587                                                 "status=0x%02x, count=%ld\n",
1588                                                 hwif->name,
1589                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1590                                 }
1591                         }
1592                 }
1593         } while ((hwif = hwif->next) != hwgroup->hwif);
1594 }
1595
1596 /**
1597  *      ide_intr        -       default IDE interrupt handler
1598  *      @irq: interrupt number
1599  *      @dev_id: hwif group
1600  *      @regs: unused weirdness from the kernel irq layer
1601  *
1602  *      This is the default IRQ handler for the IDE layer. You should
1603  *      not need to override it. If you do be aware it is subtle in
1604  *      places
1605  *
1606  *      hwgroup->hwif is the interface in the group currently performing
1607  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1608  *      the IRQ handler to call. As we issue a command the handlers
1609  *      step through multiple states, reassigning the handler to the
1610  *      next step in the process. Unlike a smart SCSI controller IDE
1611  *      expects the main processor to sequence the various transfer
1612  *      stages. We also manage a poll timer to catch up with most
1613  *      timeout situations. There are still a few where the handlers
1614  *      don't ever decide to give up.
1615  *
1616  *      The handler eventually returns ide_stopped to indicate the
1617  *      request completed. At this point we issue the next request
1618  *      on the hwgroup and the process begins again.
1619  */
1620  
1621 irqreturn_t ide_intr (int irq, void *dev_id)
1622 {
1623         unsigned long flags;
1624         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1625         ide_hwif_t *hwif;
1626         ide_drive_t *drive;
1627         ide_handler_t *handler;
1628         ide_startstop_t startstop;
1629
1630         spin_lock_irqsave(&ide_lock, flags);
1631         hwif = hwgroup->hwif;
1632
1633         if (!ide_ack_intr(hwif)) {
1634                 spin_unlock_irqrestore(&ide_lock, flags);
1635                 return IRQ_NONE;
1636         }
1637
1638         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1639                 /*
1640                  * Not expecting an interrupt from this drive.
1641                  * That means this could be:
1642                  *      (1) an interrupt from another PCI device
1643                  *      sharing the same PCI INT# as us.
1644                  * or   (2) a drive just entered sleep or standby mode,
1645                  *      and is interrupting to let us know.
1646                  * or   (3) a spurious interrupt of unknown origin.
1647                  *
1648                  * For PCI, we cannot tell the difference,
1649                  * so in that case we just ignore it and hope it goes away.
1650                  *
1651                  * FIXME: unexpected_intr should be hwif-> then we can
1652                  * remove all the ifdef PCI crap
1653                  */
1654 #ifdef CONFIG_BLK_DEV_IDEPCI
1655                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1656 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1657                 {
1658                         /*
1659                          * Probably not a shared PCI interrupt,
1660                          * so we can safely try to do something about it:
1661                          */
1662                         unexpected_intr(irq, hwgroup);
1663 #ifdef CONFIG_BLK_DEV_IDEPCI
1664                 } else {
1665                         /*
1666                          * Whack the status register, just in case
1667                          * we have a leftover pending IRQ.
1668                          */
1669                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1670 #endif /* CONFIG_BLK_DEV_IDEPCI */
1671                 }
1672                 spin_unlock_irqrestore(&ide_lock, flags);
1673                 return IRQ_NONE;
1674         }
1675         drive = hwgroup->drive;
1676         if (!drive) {
1677                 /*
1678                  * This should NEVER happen, and there isn't much
1679                  * we could do about it here.
1680                  *
1681                  * [Note - this can occur if the drive is hot unplugged]
1682                  */
1683                 spin_unlock_irqrestore(&ide_lock, flags);
1684                 return IRQ_HANDLED;
1685         }
1686         if (!drive_is_ready(drive)) {
1687                 /*
1688                  * This happens regularly when we share a PCI IRQ with
1689                  * another device.  Unfortunately, it can also happen
1690                  * with some buggy drives that trigger the IRQ before
1691                  * their status register is up to date.  Hopefully we have
1692                  * enough advance overhead that the latter isn't a problem.
1693                  */
1694                 spin_unlock_irqrestore(&ide_lock, flags);
1695                 return IRQ_NONE;
1696         }
1697         if (!hwgroup->busy) {
1698                 hwgroup->busy = 1;      /* paranoia */
1699                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1700         }
1701         hwgroup->handler = NULL;
1702         hwgroup->req_gen++;
1703         del_timer(&hwgroup->timer);
1704         spin_unlock(&ide_lock);
1705
1706         /* Some controllers might set DMA INTR no matter DMA or PIO;
1707          * bmdma status might need to be cleared even for
1708          * PIO interrupts to prevent spurious/lost irq.
1709          */
1710         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1711                 /* ide_dma_end() needs bmdma status for error checking.
1712                  * So, skip clearing bmdma status here and leave it
1713                  * to ide_dma_end() if this is dma interrupt.
1714                  */
1715                 hwif->ide_dma_clear_irq(drive);
1716
1717         if (drive->unmask)
1718                 local_irq_enable_in_hardirq();
1719         /* service this interrupt, may set handler for next interrupt */
1720         startstop = handler(drive);
1721         spin_lock_irq(&ide_lock);
1722
1723         /*
1724          * Note that handler() may have set things up for another
1725          * interrupt to occur soon, but it cannot happen until
1726          * we exit from this routine, because it will be the
1727          * same irq as is currently being serviced here, and Linux
1728          * won't allow another of the same (on any CPU) until we return.
1729          */
1730         drive->service_time = jiffies - drive->service_start;
1731         if (startstop == ide_stopped) {
1732                 if (hwgroup->handler == NULL) { /* paranoia */
1733                         hwgroup->busy = 0;
1734                         ide_do_request(hwgroup, hwif->irq);
1735                 } else {
1736                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1737                                 "on exit\n", drive->name);
1738                 }
1739         }
1740         spin_unlock_irqrestore(&ide_lock, flags);
1741         return IRQ_HANDLED;
1742 }
1743
1744 /**
1745  *      ide_init_drive_cmd      -       initialize a drive command request
1746  *      @rq: request object
1747  *
1748  *      Initialize a request before we fill it in and send it down to
1749  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1750  *      now it doesn't do a lot, but if that changes abusers will have a
1751  *      nasty surprise.
1752  */
1753
1754 void ide_init_drive_cmd (struct request *rq)
1755 {
1756         memset(rq, 0, sizeof(*rq));
1757         rq->cmd_type = REQ_TYPE_ATA_CMD;
1758         rq->ref_count = 1;
1759 }
1760
1761 EXPORT_SYMBOL(ide_init_drive_cmd);
1762
1763 /**
1764  *      ide_do_drive_cmd        -       issue IDE special command
1765  *      @drive: device to issue command
1766  *      @rq: request to issue
1767  *      @action: action for processing
1768  *
1769  *      This function issues a special IDE device request
1770  *      onto the request queue.
1771  *
1772  *      If action is ide_wait, then the rq is queued at the end of the
1773  *      request queue, and the function sleeps until it has been processed.
1774  *      This is for use when invoked from an ioctl handler.
1775  *
1776  *      If action is ide_preempt, then the rq is queued at the head of
1777  *      the request queue, displacing the currently-being-processed
1778  *      request and this function returns immediately without waiting
1779  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1780  *      intended for careful use by the ATAPI tape/cdrom driver code.
1781  *
1782  *      If action is ide_end, then the rq is queued at the end of the
1783  *      request queue, and the function returns immediately without waiting
1784  *      for the new rq to be completed. This is again intended for careful
1785  *      use by the ATAPI tape/cdrom driver code.
1786  */
1787  
1788 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1789 {
1790         unsigned long flags;
1791         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1792         DECLARE_COMPLETION_ONSTACK(wait);
1793         int where = ELEVATOR_INSERT_BACK, err;
1794         int must_wait = (action == ide_wait || action == ide_head_wait);
1795
1796         rq->errors = 0;
1797
1798         /*
1799          * we need to hold an extra reference to request for safe inspection
1800          * after completion
1801          */
1802         if (must_wait) {
1803                 rq->ref_count++;
1804                 rq->end_io_data = &wait;
1805                 rq->end_io = blk_end_sync_rq;
1806         }
1807
1808         spin_lock_irqsave(&ide_lock, flags);
1809         if (action == ide_preempt)
1810                 hwgroup->rq = NULL;
1811         if (action == ide_preempt || action == ide_head_wait) {
1812                 where = ELEVATOR_INSERT_FRONT;
1813                 rq->cmd_flags |= REQ_PREEMPT;
1814         }
1815         __elv_add_request(drive->queue, rq, where, 0);
1816         ide_do_request(hwgroup, IDE_NO_IRQ);
1817         spin_unlock_irqrestore(&ide_lock, flags);
1818
1819         err = 0;
1820         if (must_wait) {
1821                 wait_for_completion(&wait);
1822                 if (rq->errors)
1823                         err = -EIO;
1824
1825                 blk_put_request(rq);
1826         }
1827
1828         return err;
1829 }
1830
1831 EXPORT_SYMBOL(ide_do_drive_cmd);