ide-disk: fix flush requests (take 2)
[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 #include <linux/bitops.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, unsigned int nr_bytes, int dequeue)
59 {
60         int ret = 1;
61         int error = 0;
62
63         if (uptodate <= 0)
64                 error = uptodate ? uptodate : -EIO;
65
66         /*
67          * if failfast is set on a request, override number of sectors and
68          * complete the whole request right now
69          */
70         if (blk_noretry_request(rq) && error)
71                 nr_bytes = rq->hard_nr_sectors << 9;
72
73         if (!blk_fs_request(rq) && error && !rq->errors)
74                 rq->errors = -EIO;
75
76         /*
77          * decide whether to reenable DMA -- 3 is a random magic for now,
78          * if we DMA timeout more than 3 times, just stay in PIO
79          */
80         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
81                 drive->state = 0;
82                 ide_dma_on(drive);
83         }
84
85         if (!__blk_end_request(rq, error, nr_bytes)) {
86                 if (dequeue)
87                         HWGROUP(drive)->rq = NULL;
88                 ret = 0;
89         }
90
91         return ret;
92 }
93
94 /**
95  *      ide_end_request         -       complete an IDE I/O
96  *      @drive: IDE device for the I/O
97  *      @uptodate:
98  *      @nr_sectors: number of sectors completed
99  *
100  *      This is our end_request wrapper function. We complete the I/O
101  *      update random number input and dequeue the request, which if
102  *      it was tagged may be out of order.
103  */
104
105 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
106 {
107         unsigned int nr_bytes = nr_sectors << 9;
108         struct request *rq;
109         unsigned long flags;
110         int ret = 1;
111
112         /*
113          * room for locking improvements here, the calls below don't
114          * need the queue lock held at all
115          */
116         spin_lock_irqsave(&ide_lock, flags);
117         rq = HWGROUP(drive)->rq;
118
119         if (!nr_bytes) {
120                 if (blk_pc_request(rq))
121                         nr_bytes = rq->data_len;
122                 else
123                         nr_bytes = rq->hard_cur_sectors << 9;
124         }
125
126         ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
127
128         spin_unlock_irqrestore(&ide_lock, flags);
129         return ret;
130 }
131 EXPORT_SYMBOL(ide_end_request);
132
133 /*
134  * Power Management state machine. This one is rather trivial for now,
135  * we should probably add more, like switching back to PIO on suspend
136  * to help some BIOSes, re-do the door locking on resume, etc...
137  */
138
139 enum {
140         ide_pm_flush_cache      = ide_pm_state_start_suspend,
141         idedisk_pm_standby,
142
143         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
144         idedisk_pm_idle,
145         ide_pm_restore_dma,
146 };
147
148 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
149 {
150         struct request_pm_state *pm = rq->data;
151
152         if (drive->media != ide_disk)
153                 return;
154
155         switch (pm->pm_step) {
156         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
157                 if (pm->pm_state == PM_EVENT_FREEZE)
158                         pm->pm_step = ide_pm_state_completed;
159                 else
160                         pm->pm_step = idedisk_pm_standby;
161                 break;
162         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
163                 pm->pm_step = ide_pm_state_completed;
164                 break;
165         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
166                 pm->pm_step = idedisk_pm_idle;
167                 break;
168         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
169                 pm->pm_step = ide_pm_restore_dma;
170                 break;
171         }
172 }
173
174 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
175 {
176         struct request_pm_state *pm = rq->data;
177         ide_task_t *args = rq->special;
178
179         memset(args, 0, sizeof(*args));
180
181         switch (pm->pm_step) {
182         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
183                 if (drive->media != ide_disk)
184                         break;
185                 /* Not supported? Switch to next step now. */
186                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
187                         ide_complete_power_step(drive, rq, 0, 0);
188                         return ide_stopped;
189                 }
190                 if (ide_id_has_flush_cache_ext(drive->id))
191                         args->tf.command = WIN_FLUSH_CACHE_EXT;
192                 else
193                         args->tf.command = WIN_FLUSH_CACHE;
194                 goto out_do_tf;
195
196         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
197                 args->tf.command = WIN_STANDBYNOW1;
198                 goto out_do_tf;
199
200         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
201                 ide_set_max_pio(drive);
202                 /*
203                  * skip idedisk_pm_idle for ATAPI devices
204                  */
205                 if (drive->media != ide_disk)
206                         pm->pm_step = ide_pm_restore_dma;
207                 else
208                         ide_complete_power_step(drive, rq, 0, 0);
209                 return ide_stopped;
210
211         case idedisk_pm_idle:           /* Resume step 2 (idle) */
212                 args->tf.command = WIN_IDLEIMMEDIATE;
213                 goto out_do_tf;
214
215         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
216                 /*
217                  * Right now, all we do is call ide_set_dma(drive),
218                  * we could be smarter and check for current xfer_speed
219                  * in struct drive etc...
220                  */
221                 if (drive->hwif->dma_host_set == NULL)
222                         break;
223                 /*
224                  * TODO: respect ->using_dma setting
225                  */
226                 ide_set_dma(drive);
227                 break;
228         }
229         pm->pm_step = ide_pm_state_completed;
230         return ide_stopped;
231
232 out_do_tf:
233         args->tf_flags   = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
234         args->data_phase = TASKFILE_NO_DATA;
235         return do_rw_taskfile(drive, args);
236 }
237
238 /**
239  *      ide_end_dequeued_request        -       complete an IDE I/O
240  *      @drive: IDE device for the I/O
241  *      @uptodate:
242  *      @nr_sectors: number of sectors completed
243  *
244  *      Complete an I/O that is no longer on the request queue. This
245  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
246  *      We must still finish the old request but we must not tamper with the
247  *      queue in the meantime.
248  *
249  *      NOTE: This path does not handle barrier, but barrier is not supported
250  *      on ide-cd anyway.
251  */
252
253 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
254                              int uptodate, int nr_sectors)
255 {
256         unsigned long flags;
257         int ret;
258
259         spin_lock_irqsave(&ide_lock, flags);
260         BUG_ON(!blk_rq_started(rq));
261         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
262         spin_unlock_irqrestore(&ide_lock, flags);
263
264         return ret;
265 }
266 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
267
268
269 /**
270  *      ide_complete_pm_request - end the current Power Management request
271  *      @drive: target drive
272  *      @rq: request
273  *
274  *      This function cleans up the current PM request and stops the queue
275  *      if necessary.
276  */
277 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
278 {
279         unsigned long flags;
280
281 #ifdef DEBUG_PM
282         printk("%s: completing PM request, %s\n", drive->name,
283                blk_pm_suspend_request(rq) ? "suspend" : "resume");
284 #endif
285         spin_lock_irqsave(&ide_lock, flags);
286         if (blk_pm_suspend_request(rq)) {
287                 blk_stop_queue(drive->queue);
288         } else {
289                 drive->blocked = 0;
290                 blk_start_queue(drive->queue);
291         }
292         HWGROUP(drive)->rq = NULL;
293         if (__blk_end_request(rq, 0, 0))
294                 BUG();
295         spin_unlock_irqrestore(&ide_lock, flags);
296 }
297
298 void ide_tf_read(ide_drive_t *drive, ide_task_t *task)
299 {
300         ide_hwif_t *hwif = drive->hwif;
301         struct ide_taskfile *tf = &task->tf;
302
303         if (task->tf_flags & IDE_TFLAG_IN_DATA) {
304                 u16 data = hwif->INW(IDE_DATA_REG);
305
306                 tf->data = data & 0xff;
307                 tf->hob_data = (data >> 8) & 0xff;
308         }
309
310         /* be sure we're looking at the low order bits */
311         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
312
313         if (task->tf_flags & IDE_TFLAG_IN_NSECT)
314                 tf->nsect  = hwif->INB(IDE_NSECTOR_REG);
315         if (task->tf_flags & IDE_TFLAG_IN_LBAL)
316                 tf->lbal   = hwif->INB(IDE_SECTOR_REG);
317         if (task->tf_flags & IDE_TFLAG_IN_LBAM)
318                 tf->lbam   = hwif->INB(IDE_LCYL_REG);
319         if (task->tf_flags & IDE_TFLAG_IN_LBAH)
320                 tf->lbah   = hwif->INB(IDE_HCYL_REG);
321         if (task->tf_flags & IDE_TFLAG_IN_DEVICE)
322                 tf->device = hwif->INB(IDE_SELECT_REG);
323
324         if (task->tf_flags & IDE_TFLAG_LBA48) {
325                 hwif->OUTB(drive->ctl | 0x80, IDE_CONTROL_REG);
326
327                 if (task->tf_flags & IDE_TFLAG_IN_HOB_FEATURE)
328                         tf->hob_feature = hwif->INB(IDE_FEATURE_REG);
329                 if (task->tf_flags & IDE_TFLAG_IN_HOB_NSECT)
330                         tf->hob_nsect   = hwif->INB(IDE_NSECTOR_REG);
331                 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAL)
332                         tf->hob_lbal    = hwif->INB(IDE_SECTOR_REG);
333                 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAM)
334                         tf->hob_lbam    = hwif->INB(IDE_LCYL_REG);
335                 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAH)
336                         tf->hob_lbah    = hwif->INB(IDE_HCYL_REG);
337         }
338 }
339
340 /**
341  *      ide_end_drive_cmd       -       end an explicit drive command
342  *      @drive: command 
343  *      @stat: status bits
344  *      @err: error bits
345  *
346  *      Clean up after success/failure of an explicit drive command.
347  *      These get thrown onto the queue so they are synchronized with
348  *      real I/O operations on the drive.
349  *
350  *      In LBA48 mode we have to read the register set twice to get
351  *      all the extra information out.
352  */
353  
354 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
355 {
356         unsigned long flags;
357         struct request *rq;
358
359         spin_lock_irqsave(&ide_lock, flags);
360         rq = HWGROUP(drive)->rq;
361         spin_unlock_irqrestore(&ide_lock, flags);
362
363         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
364                 ide_task_t *task = (ide_task_t *)rq->special;
365
366                 if (rq->errors == 0)
367                         rq->errors = !OK_STAT(stat, READY_STAT, BAD_STAT);
368
369                 if (task) {
370                         struct ide_taskfile *tf = &task->tf;
371
372                         tf->error = err;
373                         tf->status = stat;
374
375                         ide_tf_read(drive, task);
376
377                         if (task->tf_flags & IDE_TFLAG_DYN)
378                                 kfree(task);
379                 }
380         } else if (blk_pm_request(rq)) {
381                 struct request_pm_state *pm = rq->data;
382 #ifdef DEBUG_PM
383                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
384                         drive->name, rq->pm->pm_step, stat, err);
385 #endif
386                 ide_complete_power_step(drive, rq, stat, err);
387                 if (pm->pm_step == ide_pm_state_completed)
388                         ide_complete_pm_request(drive, rq);
389                 return;
390         }
391
392         spin_lock_irqsave(&ide_lock, flags);
393         HWGROUP(drive)->rq = NULL;
394         rq->errors = err;
395         if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
396                                        blk_rq_bytes(rq))))
397                 BUG();
398         spin_unlock_irqrestore(&ide_lock, flags);
399 }
400
401 EXPORT_SYMBOL(ide_end_drive_cmd);
402
403 /**
404  *      try_to_flush_leftover_data      -       flush junk
405  *      @drive: drive to flush
406  *
407  *      try_to_flush_leftover_data() is invoked in response to a drive
408  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
409  *      resetting the drive, this routine tries to clear the condition
410  *      by read a sector's worth of data from the drive.  Of course,
411  *      this may not help if the drive is *waiting* for data from *us*.
412  */
413 static void try_to_flush_leftover_data (ide_drive_t *drive)
414 {
415         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
416
417         if (drive->media != ide_disk)
418                 return;
419         while (i > 0) {
420                 u32 buffer[16];
421                 u32 wcount = (i > 16) ? 16 : i;
422
423                 i -= wcount;
424                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
425         }
426 }
427
428 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
429 {
430         if (rq->rq_disk) {
431                 ide_driver_t *drv;
432
433                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
434                 drv->end_request(drive, 0, 0);
435         } else
436                 ide_end_request(drive, 0, 0);
437 }
438
439 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
440 {
441         ide_hwif_t *hwif = drive->hwif;
442
443         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
444                 /* other bits are useless when BUSY */
445                 rq->errors |= ERROR_RESET;
446         } else if (stat & ERR_STAT) {
447                 /* err has different meaning on cdrom and tape */
448                 if (err == ABRT_ERR) {
449                         if (drive->select.b.lba &&
450                             /* some newer drives don't support WIN_SPECIFY */
451                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
452                                 return ide_stopped;
453                 } else if ((err & BAD_CRC) == BAD_CRC) {
454                         /* UDMA crc error, just retry the operation */
455                         drive->crc_count++;
456                 } else if (err & (BBD_ERR | ECC_ERR)) {
457                         /* retries won't help these */
458                         rq->errors = ERROR_MAX;
459                 } else if (err & TRK0_ERR) {
460                         /* help it find track zero */
461                         rq->errors |= ERROR_RECAL;
462                 }
463         }
464
465         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ &&
466             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0)
467                 try_to_flush_leftover_data(drive);
468
469         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
470                 ide_kill_rq(drive, rq);
471                 return ide_stopped;
472         }
473
474         if (ide_read_status(drive) & (BUSY_STAT | DRQ_STAT))
475                 rq->errors |= ERROR_RESET;
476
477         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
478                 ++rq->errors;
479                 return ide_do_reset(drive);
480         }
481
482         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
483                 drive->special.b.recalibrate = 1;
484
485         ++rq->errors;
486
487         return ide_stopped;
488 }
489
490 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
491 {
492         ide_hwif_t *hwif = drive->hwif;
493
494         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
495                 /* other bits are useless when BUSY */
496                 rq->errors |= ERROR_RESET;
497         } else {
498                 /* add decoding error stuff */
499         }
500
501         if (ide_read_status(drive) & (BUSY_STAT | DRQ_STAT))
502                 /* force an abort */
503                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
504
505         if (rq->errors >= ERROR_MAX) {
506                 ide_kill_rq(drive, rq);
507         } else {
508                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
509                         ++rq->errors;
510                         return ide_do_reset(drive);
511                 }
512                 ++rq->errors;
513         }
514
515         return ide_stopped;
516 }
517
518 ide_startstop_t
519 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
520 {
521         if (drive->media == ide_disk)
522                 return ide_ata_error(drive, rq, stat, err);
523         return ide_atapi_error(drive, rq, stat, err);
524 }
525
526 EXPORT_SYMBOL_GPL(__ide_error);
527
528 /**
529  *      ide_error       -       handle an error on the IDE
530  *      @drive: drive the error occurred on
531  *      @msg: message to report
532  *      @stat: status bits
533  *
534  *      ide_error() takes action based on the error returned by the drive.
535  *      For normal I/O that may well include retries. We deal with
536  *      both new-style (taskfile) and old style command handling here.
537  *      In the case of taskfile command handling there is work left to
538  *      do
539  */
540  
541 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
542 {
543         struct request *rq;
544         u8 err;
545
546         err = ide_dump_status(drive, msg, stat);
547
548         if ((rq = HWGROUP(drive)->rq) == NULL)
549                 return ide_stopped;
550
551         /* retry only "normal" I/O: */
552         if (!blk_fs_request(rq)) {
553                 rq->errors = 1;
554                 ide_end_drive_cmd(drive, stat, err);
555                 return ide_stopped;
556         }
557
558         if (rq->rq_disk) {
559                 ide_driver_t *drv;
560
561                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
562                 return drv->error(drive, rq, stat, err);
563         } else
564                 return __ide_error(drive, rq, stat, err);
565 }
566
567 EXPORT_SYMBOL_GPL(ide_error);
568
569 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
570 {
571         if (drive->media != ide_disk)
572                 rq->errors |= ERROR_RESET;
573
574         ide_kill_rq(drive, rq);
575
576         return ide_stopped;
577 }
578
579 EXPORT_SYMBOL_GPL(__ide_abort);
580
581 /**
582  *      ide_abort       -       abort pending IDE operations
583  *      @drive: drive the error occurred on
584  *      @msg: message to report
585  *
586  *      ide_abort kills and cleans up when we are about to do a 
587  *      host initiated reset on active commands. Longer term we
588  *      want handlers to have sensible abort handling themselves
589  *
590  *      This differs fundamentally from ide_error because in 
591  *      this case the command is doing just fine when we
592  *      blow it away.
593  */
594  
595 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
596 {
597         struct request *rq;
598
599         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
600                 return ide_stopped;
601
602         /* retry only "normal" I/O: */
603         if (!blk_fs_request(rq)) {
604                 rq->errors = 1;
605                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
606                 return ide_stopped;
607         }
608
609         if (rq->rq_disk) {
610                 ide_driver_t *drv;
611
612                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
613                 return drv->abort(drive, rq);
614         } else
615                 return __ide_abort(drive, rq);
616 }
617
618 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
619 {
620         tf->nsect   = drive->sect;
621         tf->lbal    = drive->sect;
622         tf->lbam    = drive->cyl;
623         tf->lbah    = drive->cyl >> 8;
624         tf->device  = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
625         tf->command = WIN_SPECIFY;
626 }
627
628 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
629 {
630         tf->nsect   = drive->sect;
631         tf->command = WIN_RESTORE;
632 }
633
634 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
635 {
636         tf->nsect   = drive->mult_req;
637         tf->command = WIN_SETMULT;
638 }
639
640 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
641 {
642         special_t *s = &drive->special;
643         ide_task_t args;
644
645         memset(&args, 0, sizeof(ide_task_t));
646         args.data_phase = TASKFILE_NO_DATA;
647
648         if (s->b.set_geometry) {
649                 s->b.set_geometry = 0;
650                 ide_tf_set_specify_cmd(drive, &args.tf);
651         } else if (s->b.recalibrate) {
652                 s->b.recalibrate = 0;
653                 ide_tf_set_restore_cmd(drive, &args.tf);
654         } else if (s->b.set_multmode) {
655                 s->b.set_multmode = 0;
656                 if (drive->mult_req > drive->id->max_multsect)
657                         drive->mult_req = drive->id->max_multsect;
658                 ide_tf_set_setmult_cmd(drive, &args.tf);
659         } else if (s->all) {
660                 int special = s->all;
661                 s->all = 0;
662                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
663                 return ide_stopped;
664         }
665
666         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
667                         IDE_TFLAG_CUSTOM_HANDLER;
668
669         do_rw_taskfile(drive, &args);
670
671         return ide_started;
672 }
673
674 /*
675  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
676  */
677 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
678 {
679         switch (req_pio) {
680         case 202:
681         case 201:
682         case 200:
683         case 102:
684         case 101:
685         case 100:
686                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
687         case 9:
688         case 8:
689                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
690         case 7:
691         case 6:
692                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
693         default:
694                 return 0;
695         }
696 }
697
698 /**
699  *      do_special              -       issue some special commands
700  *      @drive: drive the command is for
701  *
702  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
703  *      commands to a drive.  It used to do much more, but has been scaled
704  *      back.
705  */
706
707 static ide_startstop_t do_special (ide_drive_t *drive)
708 {
709         special_t *s = &drive->special;
710
711 #ifdef DEBUG
712         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
713 #endif
714         if (s->b.set_tune) {
715                 ide_hwif_t *hwif = drive->hwif;
716                 u8 req_pio = drive->tune_req;
717
718                 s->b.set_tune = 0;
719
720                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
721
722                         if (hwif->set_pio_mode == NULL)
723                                 return ide_stopped;
724
725                         /*
726                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
727                          */
728                         if (req_pio == 8 || req_pio == 9) {
729                                 unsigned long flags;
730
731                                 spin_lock_irqsave(&ide_lock, flags);
732                                 hwif->set_pio_mode(drive, req_pio);
733                                 spin_unlock_irqrestore(&ide_lock, flags);
734                         } else
735                                 hwif->set_pio_mode(drive, req_pio);
736                 } else {
737                         int keep_dma = drive->using_dma;
738
739                         ide_set_pio(drive, req_pio);
740
741                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
742                                 if (keep_dma)
743                                         ide_dma_on(drive);
744                         }
745                 }
746
747                 return ide_stopped;
748         } else {
749                 if (drive->media == ide_disk)
750                         return ide_disk_special(drive);
751
752                 s->all = 0;
753                 drive->mult_req = 0;
754                 return ide_stopped;
755         }
756 }
757
758 void ide_map_sg(ide_drive_t *drive, struct request *rq)
759 {
760         ide_hwif_t *hwif = drive->hwif;
761         struct scatterlist *sg = hwif->sg_table;
762
763         if (hwif->sg_mapped)    /* needed by ide-scsi */
764                 return;
765
766         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
767                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
768         } else {
769                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
770                 hwif->sg_nents = 1;
771         }
772 }
773
774 EXPORT_SYMBOL_GPL(ide_map_sg);
775
776 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
777 {
778         ide_hwif_t *hwif = drive->hwif;
779
780         hwif->nsect = hwif->nleft = rq->nr_sectors;
781         hwif->cursg_ofs = 0;
782         hwif->cursg = NULL;
783 }
784
785 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
786
787 /**
788  *      execute_drive_command   -       issue special drive command
789  *      @drive: the drive to issue the command on
790  *      @rq: the request structure holding the command
791  *
792  *      execute_drive_cmd() issues a special drive command,  usually 
793  *      initiated by ioctl() from the external hdparm program. The
794  *      command can be a drive command, drive task or taskfile 
795  *      operation. Weirdly you can call it with NULL to wait for
796  *      all commands to finish. Don't do this as that is due to change
797  */
798
799 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
800                 struct request *rq)
801 {
802         ide_hwif_t *hwif = HWIF(drive);
803         ide_task_t *task = rq->special;
804
805         if (task) {
806                 hwif->data_phase = task->data_phase;
807
808                 switch (hwif->data_phase) {
809                 case TASKFILE_MULTI_OUT:
810                 case TASKFILE_OUT:
811                 case TASKFILE_MULTI_IN:
812                 case TASKFILE_IN:
813                         ide_init_sg_cmd(drive, rq);
814                         ide_map_sg(drive, rq);
815                 default:
816                         break;
817                 }
818
819                 return do_rw_taskfile(drive, task);
820         }
821
822         /*
823          * NULL is actually a valid way of waiting for
824          * all current requests to be flushed from the queue.
825          */
826 #ifdef DEBUG
827         printk("%s: DRIVE_CMD (null)\n", drive->name);
828 #endif
829         ide_end_drive_cmd(drive, ide_read_status(drive), ide_read_error(drive));
830
831         return ide_stopped;
832 }
833
834 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
835 {
836         struct request_pm_state *pm = rq->data;
837
838         if (blk_pm_suspend_request(rq) &&
839             pm->pm_step == ide_pm_state_start_suspend)
840                 /* Mark drive blocked when starting the suspend sequence. */
841                 drive->blocked = 1;
842         else if (blk_pm_resume_request(rq) &&
843                  pm->pm_step == ide_pm_state_start_resume) {
844                 /* 
845                  * The first thing we do on wakeup is to wait for BSY bit to
846                  * go away (with a looong timeout) as a drive on this hwif may
847                  * just be POSTing itself.
848                  * We do that before even selecting as the "other" device on
849                  * the bus may be broken enough to walk on our toes at this
850                  * point.
851                  */
852                 int rc;
853 #ifdef DEBUG_PM
854                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
855 #endif
856                 rc = ide_wait_not_busy(HWIF(drive), 35000);
857                 if (rc)
858                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
859                 SELECT_DRIVE(drive);
860                 ide_set_irq(drive, 1);
861                 rc = ide_wait_not_busy(HWIF(drive), 100000);
862                 if (rc)
863                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
864         }
865 }
866
867 /**
868  *      start_request   -       start of I/O and command issuing for IDE
869  *
870  *      start_request() initiates handling of a new I/O request. It
871  *      accepts commands and I/O (read/write) requests. It also does
872  *      the final remapping for weird stuff like EZDrive. Once 
873  *      device mapper can work sector level the EZDrive stuff can go away
874  *
875  *      FIXME: this function needs a rename
876  */
877  
878 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
879 {
880         ide_startstop_t startstop;
881         sector_t block;
882
883         BUG_ON(!blk_rq_started(rq));
884
885 #ifdef DEBUG
886         printk("%s: start_request: current=0x%08lx\n",
887                 HWIF(drive)->name, (unsigned long) rq);
888 #endif
889
890         /* bail early if we've exceeded max_failures */
891         if (drive->max_failures && (drive->failures > drive->max_failures)) {
892                 rq->cmd_flags |= REQ_FAILED;
893                 goto kill_rq;
894         }
895
896         block    = rq->sector;
897         if (blk_fs_request(rq) &&
898             (drive->media == ide_disk || drive->media == ide_floppy)) {
899                 block += drive->sect0;
900         }
901         /* Yecch - this will shift the entire interval,
902            possibly killing some innocent following sector */
903         if (block == 0 && drive->remap_0_to_1 == 1)
904                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
905
906         if (blk_pm_request(rq))
907                 ide_check_pm_state(drive, rq);
908
909         SELECT_DRIVE(drive);
910         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
911                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
912                 return startstop;
913         }
914         if (!drive->special.all) {
915                 ide_driver_t *drv;
916
917                 /*
918                  * We reset the drive so we need to issue a SETFEATURES.
919                  * Do it _after_ do_special() restored device parameters.
920                  */
921                 if (drive->current_speed == 0xff)
922                         ide_config_drive_speed(drive, drive->desired_speed);
923
924                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
925                         return execute_drive_cmd(drive, rq);
926                 else if (blk_pm_request(rq)) {
927                         struct request_pm_state *pm = rq->data;
928 #ifdef DEBUG_PM
929                         printk("%s: start_power_step(step: %d)\n",
930                                 drive->name, rq->pm->pm_step);
931 #endif
932                         startstop = ide_start_power_step(drive, rq);
933                         if (startstop == ide_stopped &&
934                             pm->pm_step == ide_pm_state_completed)
935                                 ide_complete_pm_request(drive, rq);
936                         return startstop;
937                 }
938
939                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
940                 return drv->do_request(drive, rq, block);
941         }
942         return do_special(drive);
943 kill_rq:
944         ide_kill_rq(drive, rq);
945         return ide_stopped;
946 }
947
948 /**
949  *      ide_stall_queue         -       pause an IDE device
950  *      @drive: drive to stall
951  *      @timeout: time to stall for (jiffies)
952  *
953  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
954  *      to the hwgroup by sleeping for timeout jiffies.
955  */
956  
957 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
958 {
959         if (timeout > WAIT_WORSTCASE)
960                 timeout = WAIT_WORSTCASE;
961         drive->sleep = timeout + jiffies;
962         drive->sleeping = 1;
963 }
964
965 EXPORT_SYMBOL(ide_stall_queue);
966
967 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
968
969 /**
970  *      choose_drive            -       select a drive to service
971  *      @hwgroup: hardware group to select on
972  *
973  *      choose_drive() selects the next drive which will be serviced.
974  *      This is necessary because the IDE layer can't issue commands
975  *      to both drives on the same cable, unlike SCSI.
976  */
977  
978 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
979 {
980         ide_drive_t *drive, *best;
981
982 repeat: 
983         best = NULL;
984         drive = hwgroup->drive;
985
986         /*
987          * drive is doing pre-flush, ordered write, post-flush sequence. even
988          * though that is 3 requests, it must be seen as a single transaction.
989          * we must not preempt this drive until that is complete
990          */
991         if (blk_queue_flushing(drive->queue)) {
992                 /*
993                  * small race where queue could get replugged during
994                  * the 3-request flush cycle, just yank the plug since
995                  * we want it to finish asap
996                  */
997                 blk_remove_plug(drive->queue);
998                 return drive;
999         }
1000
1001         do {
1002                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1003                     && !elv_queue_empty(drive->queue)) {
1004                         if (!best
1005                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1006                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1007                         {
1008                                 if (!blk_queue_plugged(drive->queue))
1009                                         best = drive;
1010                         }
1011                 }
1012         } while ((drive = drive->next) != hwgroup->drive);
1013         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1014                 long t = (signed long)(WAKEUP(best) - jiffies);
1015                 if (t >= WAIT_MIN_SLEEP) {
1016                 /*
1017                  * We *may* have some time to spare, but first let's see if
1018                  * someone can potentially benefit from our nice mood today..
1019                  */
1020                         drive = best->next;
1021                         do {
1022                                 if (!drive->sleeping
1023                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1024                                  && time_before(WAKEUP(drive), jiffies + t))
1025                                 {
1026                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1027                                         goto repeat;
1028                                 }
1029                         } while ((drive = drive->next) != best);
1030                 }
1031         }
1032         return best;
1033 }
1034
1035 /*
1036  * Issue a new request to a drive from hwgroup
1037  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1038  *
1039  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1040  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1041  * may have both interfaces in a single hwgroup to "serialize" access.
1042  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1043  * together into one hwgroup for serialized access.
1044  *
1045  * Note also that several hwgroups can end up sharing a single IRQ,
1046  * possibly along with many other devices.  This is especially common in
1047  * PCI-based systems with off-board IDE controller cards.
1048  *
1049  * The IDE driver uses the single global ide_lock spinlock to protect
1050  * access to the request queues, and to protect the hwgroup->busy flag.
1051  *
1052  * The first thread into the driver for a particular hwgroup sets the
1053  * hwgroup->busy flag to indicate that this hwgroup is now active,
1054  * and then initiates processing of the top request from the request queue.
1055  *
1056  * Other threads attempting entry notice the busy setting, and will simply
1057  * queue their new requests and exit immediately.  Note that hwgroup->busy
1058  * remains set even when the driver is merely awaiting the next interrupt.
1059  * Thus, the meaning is "this hwgroup is busy processing a request".
1060  *
1061  * When processing of a request completes, the completing thread or IRQ-handler
1062  * will start the next request from the queue.  If no more work remains,
1063  * the driver will clear the hwgroup->busy flag and exit.
1064  *
1065  * The ide_lock (spinlock) is used to protect all access to the
1066  * hwgroup->busy flag, but is otherwise not needed for most processing in
1067  * the driver.  This makes the driver much more friendlier to shared IRQs
1068  * than previous designs, while remaining 100% (?) SMP safe and capable.
1069  */
1070 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1071 {
1072         ide_drive_t     *drive;
1073         ide_hwif_t      *hwif;
1074         struct request  *rq;
1075         ide_startstop_t startstop;
1076         int             loops = 0;
1077
1078         /* for atari only: POSSIBLY BROKEN HERE(?) */
1079         ide_get_lock(ide_intr, hwgroup);
1080
1081         /* caller must own ide_lock */
1082         BUG_ON(!irqs_disabled());
1083
1084         while (!hwgroup->busy) {
1085                 hwgroup->busy = 1;
1086                 drive = choose_drive(hwgroup);
1087                 if (drive == NULL) {
1088                         int sleeping = 0;
1089                         unsigned long sleep = 0; /* shut up, gcc */
1090                         hwgroup->rq = NULL;
1091                         drive = hwgroup->drive;
1092                         do {
1093                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1094                                         sleeping = 1;
1095                                         sleep = drive->sleep;
1096                                 }
1097                         } while ((drive = drive->next) != hwgroup->drive);
1098                         if (sleeping) {
1099                 /*
1100                  * Take a short snooze, and then wake up this hwgroup again.
1101                  * This gives other hwgroups on the same a chance to
1102                  * play fairly with us, just in case there are big differences
1103                  * in relative throughputs.. don't want to hog the cpu too much.
1104                  */
1105                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1106                                         sleep = jiffies + WAIT_MIN_SLEEP;
1107 #if 1
1108                                 if (timer_pending(&hwgroup->timer))
1109                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1110 #endif
1111                                 /* so that ide_timer_expiry knows what to do */
1112                                 hwgroup->sleeping = 1;
1113                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1114                                 mod_timer(&hwgroup->timer, sleep);
1115                                 /* we purposely leave hwgroup->busy==1
1116                                  * while sleeping */
1117                         } else {
1118                                 /* Ugly, but how can we sleep for the lock
1119                                  * otherwise? perhaps from tq_disk?
1120                                  */
1121
1122                                 /* for atari only */
1123                                 ide_release_lock();
1124                                 hwgroup->busy = 0;
1125                         }
1126
1127                         /* no more work for this hwgroup (for now) */
1128                         return;
1129                 }
1130         again:
1131                 hwif = HWIF(drive);
1132                 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1133                         /*
1134                          * set nIEN for previous hwif, drives in the
1135                          * quirk_list may not like intr setups/cleanups
1136                          */
1137                         if (drive->quirk_list != 1)
1138                                 ide_set_irq(drive, 0);
1139                 }
1140                 hwgroup->hwif = hwif;
1141                 hwgroup->drive = drive;
1142                 drive->sleeping = 0;
1143                 drive->service_start = jiffies;
1144
1145                 if (blk_queue_plugged(drive->queue)) {
1146                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1147                         break;
1148                 }
1149
1150                 /*
1151                  * we know that the queue isn't empty, but this can happen
1152                  * if the q->prep_rq_fn() decides to kill a request
1153                  */
1154                 rq = elv_next_request(drive->queue);
1155                 if (!rq) {
1156                         hwgroup->busy = 0;
1157                         break;
1158                 }
1159
1160                 /*
1161                  * Sanity: don't accept a request that isn't a PM request
1162                  * if we are currently power managed. This is very important as
1163                  * blk_stop_queue() doesn't prevent the elv_next_request()
1164                  * above to return us whatever is in the queue. Since we call
1165                  * ide_do_request() ourselves, we end up taking requests while
1166                  * the queue is blocked...
1167                  * 
1168                  * We let requests forced at head of queue with ide-preempt
1169                  * though. I hope that doesn't happen too much, hopefully not
1170                  * unless the subdriver triggers such a thing in its own PM
1171                  * state machine.
1172                  *
1173                  * We count how many times we loop here to make sure we service
1174                  * all drives in the hwgroup without looping for ever
1175                  */
1176                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1177                         drive = drive->next ? drive->next : hwgroup->drive;
1178                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1179                                 goto again;
1180                         /* We clear busy, there should be no pending ATA command at this point. */
1181                         hwgroup->busy = 0;
1182                         break;
1183                 }
1184
1185                 hwgroup->rq = rq;
1186
1187                 /*
1188                  * Some systems have trouble with IDE IRQs arriving while
1189                  * the driver is still setting things up.  So, here we disable
1190                  * the IRQ used by this interface while the request is being started.
1191                  * This may look bad at first, but pretty much the same thing
1192                  * happens anyway when any interrupt comes in, IDE or otherwise
1193                  *  -- the kernel masks the IRQ while it is being handled.
1194                  */
1195                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1196                         disable_irq_nosync(hwif->irq);
1197                 spin_unlock(&ide_lock);
1198                 local_irq_enable_in_hardirq();
1199                         /* allow other IRQs while we start this request */
1200                 startstop = start_request(drive, rq);
1201                 spin_lock_irq(&ide_lock);
1202                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1203                         enable_irq(hwif->irq);
1204                 if (startstop == ide_stopped)
1205                         hwgroup->busy = 0;
1206         }
1207 }
1208
1209 /*
1210  * Passes the stuff to ide_do_request
1211  */
1212 void do_ide_request(struct request_queue *q)
1213 {
1214         ide_drive_t *drive = q->queuedata;
1215
1216         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1217 }
1218
1219 /*
1220  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1221  * retry the current request in pio mode instead of risking tossing it
1222  * all away
1223  */
1224 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1225 {
1226         ide_hwif_t *hwif = HWIF(drive);
1227         struct request *rq;
1228         ide_startstop_t ret = ide_stopped;
1229
1230         /*
1231          * end current dma transaction
1232          */
1233
1234         if (error < 0) {
1235                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1236                 (void)HWIF(drive)->ide_dma_end(drive);
1237                 ret = ide_error(drive, "dma timeout error",
1238                                 ide_read_status(drive));
1239         } else {
1240                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1241                 hwif->dma_timeout(drive);
1242         }
1243
1244         /*
1245          * disable dma for now, but remember that we did so because of
1246          * a timeout -- we'll reenable after we finish this next request
1247          * (or rather the first chunk of it) in pio.
1248          */
1249         drive->retry_pio++;
1250         drive->state = DMA_PIO_RETRY;
1251         ide_dma_off_quietly(drive);
1252
1253         /*
1254          * un-busy drive etc (hwgroup->busy is cleared on return) and
1255          * make sure request is sane
1256          */
1257         rq = HWGROUP(drive)->rq;
1258
1259         if (!rq)
1260                 goto out;
1261
1262         HWGROUP(drive)->rq = NULL;
1263
1264         rq->errors = 0;
1265
1266         if (!rq->bio)
1267                 goto out;
1268
1269         rq->sector = rq->bio->bi_sector;
1270         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1271         rq->hard_cur_sectors = rq->current_nr_sectors;
1272         rq->buffer = bio_data(rq->bio);
1273 out:
1274         return ret;
1275 }
1276
1277 /**
1278  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1279  *      @data: timer callback magic (hwgroup)
1280  *
1281  *      An IDE command has timed out before the expected drive return
1282  *      occurred. At this point we attempt to clean up the current
1283  *      mess. If the current handler includes an expiry handler then
1284  *      we invoke the expiry handler, and providing it is happy the
1285  *      work is done. If that fails we apply generic recovery rules
1286  *      invoking the handler and checking the drive DMA status. We
1287  *      have an excessively incestuous relationship with the DMA
1288  *      logic that wants cleaning up.
1289  */
1290  
1291 void ide_timer_expiry (unsigned long data)
1292 {
1293         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1294         ide_handler_t   *handler;
1295         ide_expiry_t    *expiry;
1296         unsigned long   flags;
1297         unsigned long   wait = -1;
1298
1299         spin_lock_irqsave(&ide_lock, flags);
1300
1301         if (((handler = hwgroup->handler) == NULL) ||
1302             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1303                 /*
1304                  * Either a marginal timeout occurred
1305                  * (got the interrupt just as timer expired),
1306                  * or we were "sleeping" to give other devices a chance.
1307                  * Either way, we don't really want to complain about anything.
1308                  */
1309                 if (hwgroup->sleeping) {
1310                         hwgroup->sleeping = 0;
1311                         hwgroup->busy = 0;
1312                 }
1313         } else {
1314                 ide_drive_t *drive = hwgroup->drive;
1315                 if (!drive) {
1316                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1317                         hwgroup->handler = NULL;
1318                 } else {
1319                         ide_hwif_t *hwif;
1320                         ide_startstop_t startstop = ide_stopped;
1321                         if (!hwgroup->busy) {
1322                                 hwgroup->busy = 1;      /* paranoia */
1323                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1324                         }
1325                         if ((expiry = hwgroup->expiry) != NULL) {
1326                                 /* continue */
1327                                 if ((wait = expiry(drive)) > 0) {
1328                                         /* reset timer */
1329                                         hwgroup->timer.expires  = jiffies + wait;
1330                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1331                                         add_timer(&hwgroup->timer);
1332                                         spin_unlock_irqrestore(&ide_lock, flags);
1333                                         return;
1334                                 }
1335                         }
1336                         hwgroup->handler = NULL;
1337                         /*
1338                          * We need to simulate a real interrupt when invoking
1339                          * the handler() function, which means we need to
1340                          * globally mask the specific IRQ:
1341                          */
1342                         spin_unlock(&ide_lock);
1343                         hwif  = HWIF(drive);
1344                         /* disable_irq_nosync ?? */
1345                         disable_irq(hwif->irq);
1346                         /* local CPU only,
1347                          * as if we were handling an interrupt */
1348                         local_irq_disable();
1349                         if (hwgroup->polling) {
1350                                 startstop = handler(drive);
1351                         } else if (drive_is_ready(drive)) {
1352                                 if (drive->waiting_for_dma)
1353                                         hwgroup->hwif->dma_lost_irq(drive);
1354                                 (void)ide_ack_intr(hwif);
1355                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1356                                 startstop = handler(drive);
1357                         } else {
1358                                 if (drive->waiting_for_dma) {
1359                                         startstop = ide_dma_timeout_retry(drive, wait);
1360                                 } else
1361                                         startstop =
1362                                         ide_error(drive, "irq timeout",
1363                                                   ide_read_status(drive));
1364                         }
1365                         drive->service_time = jiffies - drive->service_start;
1366                         spin_lock_irq(&ide_lock);
1367                         enable_irq(hwif->irq);
1368                         if (startstop == ide_stopped)
1369                                 hwgroup->busy = 0;
1370                 }
1371         }
1372         ide_do_request(hwgroup, IDE_NO_IRQ);
1373         spin_unlock_irqrestore(&ide_lock, flags);
1374 }
1375
1376 /**
1377  *      unexpected_intr         -       handle an unexpected IDE interrupt
1378  *      @irq: interrupt line
1379  *      @hwgroup: hwgroup being processed
1380  *
1381  *      There's nothing really useful we can do with an unexpected interrupt,
1382  *      other than reading the status register (to clear it), and logging it.
1383  *      There should be no way that an irq can happen before we're ready for it,
1384  *      so we needn't worry much about losing an "important" interrupt here.
1385  *
1386  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1387  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1388  *      looks "good", we just ignore the interrupt completely.
1389  *
1390  *      This routine assumes __cli() is in effect when called.
1391  *
1392  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1393  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1394  *      we could screw up by interfering with a new request being set up for 
1395  *      irq15.
1396  *
1397  *      In reality, this is a non-issue.  The new command is not sent unless 
1398  *      the drive is ready to accept one, in which case we know the drive is
1399  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1400  *      before completing the issuance of any new drive command, so we will not
1401  *      be accidentally invoked as a result of any valid command completion
1402  *      interrupt.
1403  *
1404  *      Note that we must walk the entire hwgroup here. We know which hwif
1405  *      is doing the current command, but we don't know which hwif burped
1406  *      mysteriously.
1407  */
1408  
1409 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1410 {
1411         u8 stat;
1412         ide_hwif_t *hwif = hwgroup->hwif;
1413
1414         /*
1415          * handle the unexpected interrupt
1416          */
1417         do {
1418                 if (hwif->irq == irq) {
1419                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1420                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1421                                 /* Try to not flood the console with msgs */
1422                                 static unsigned long last_msgtime, count;
1423                                 ++count;
1424                                 if (time_after(jiffies, last_msgtime + HZ)) {
1425                                         last_msgtime = jiffies;
1426                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1427                                                 "status=0x%02x, count=%ld\n",
1428                                                 hwif->name,
1429                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1430                                 }
1431                         }
1432                 }
1433         } while ((hwif = hwif->next) != hwgroup->hwif);
1434 }
1435
1436 /**
1437  *      ide_intr        -       default IDE interrupt handler
1438  *      @irq: interrupt number
1439  *      @dev_id: hwif group
1440  *      @regs: unused weirdness from the kernel irq layer
1441  *
1442  *      This is the default IRQ handler for the IDE layer. You should
1443  *      not need to override it. If you do be aware it is subtle in
1444  *      places
1445  *
1446  *      hwgroup->hwif is the interface in the group currently performing
1447  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1448  *      the IRQ handler to call. As we issue a command the handlers
1449  *      step through multiple states, reassigning the handler to the
1450  *      next step in the process. Unlike a smart SCSI controller IDE
1451  *      expects the main processor to sequence the various transfer
1452  *      stages. We also manage a poll timer to catch up with most
1453  *      timeout situations. There are still a few where the handlers
1454  *      don't ever decide to give up.
1455  *
1456  *      The handler eventually returns ide_stopped to indicate the
1457  *      request completed. At this point we issue the next request
1458  *      on the hwgroup and the process begins again.
1459  */
1460  
1461 irqreturn_t ide_intr (int irq, void *dev_id)
1462 {
1463         unsigned long flags;
1464         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1465         ide_hwif_t *hwif;
1466         ide_drive_t *drive;
1467         ide_handler_t *handler;
1468         ide_startstop_t startstop;
1469
1470         spin_lock_irqsave(&ide_lock, flags);
1471         hwif = hwgroup->hwif;
1472
1473         if (!ide_ack_intr(hwif)) {
1474                 spin_unlock_irqrestore(&ide_lock, flags);
1475                 return IRQ_NONE;
1476         }
1477
1478         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1479                 /*
1480                  * Not expecting an interrupt from this drive.
1481                  * That means this could be:
1482                  *      (1) an interrupt from another PCI device
1483                  *      sharing the same PCI INT# as us.
1484                  * or   (2) a drive just entered sleep or standby mode,
1485                  *      and is interrupting to let us know.
1486                  * or   (3) a spurious interrupt of unknown origin.
1487                  *
1488                  * For PCI, we cannot tell the difference,
1489                  * so in that case we just ignore it and hope it goes away.
1490                  *
1491                  * FIXME: unexpected_intr should be hwif-> then we can
1492                  * remove all the ifdef PCI crap
1493                  */
1494 #ifdef CONFIG_BLK_DEV_IDEPCI
1495                 if (hwif->chipset != ide_pci)
1496 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1497                 {
1498                         /*
1499                          * Probably not a shared PCI interrupt,
1500                          * so we can safely try to do something about it:
1501                          */
1502                         unexpected_intr(irq, hwgroup);
1503 #ifdef CONFIG_BLK_DEV_IDEPCI
1504                 } else {
1505                         /*
1506                          * Whack the status register, just in case
1507                          * we have a leftover pending IRQ.
1508                          */
1509                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1510 #endif /* CONFIG_BLK_DEV_IDEPCI */
1511                 }
1512                 spin_unlock_irqrestore(&ide_lock, flags);
1513                 return IRQ_NONE;
1514         }
1515         drive = hwgroup->drive;
1516         if (!drive) {
1517                 /*
1518                  * This should NEVER happen, and there isn't much
1519                  * we could do about it here.
1520                  *
1521                  * [Note - this can occur if the drive is hot unplugged]
1522                  */
1523                 spin_unlock_irqrestore(&ide_lock, flags);
1524                 return IRQ_HANDLED;
1525         }
1526         if (!drive_is_ready(drive)) {
1527                 /*
1528                  * This happens regularly when we share a PCI IRQ with
1529                  * another device.  Unfortunately, it can also happen
1530                  * with some buggy drives that trigger the IRQ before
1531                  * their status register is up to date.  Hopefully we have
1532                  * enough advance overhead that the latter isn't a problem.
1533                  */
1534                 spin_unlock_irqrestore(&ide_lock, flags);
1535                 return IRQ_NONE;
1536         }
1537         if (!hwgroup->busy) {
1538                 hwgroup->busy = 1;      /* paranoia */
1539                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1540         }
1541         hwgroup->handler = NULL;
1542         hwgroup->req_gen++;
1543         del_timer(&hwgroup->timer);
1544         spin_unlock(&ide_lock);
1545
1546         /* Some controllers might set DMA INTR no matter DMA or PIO;
1547          * bmdma status might need to be cleared even for
1548          * PIO interrupts to prevent spurious/lost irq.
1549          */
1550         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1551                 /* ide_dma_end() needs bmdma status for error checking.
1552                  * So, skip clearing bmdma status here and leave it
1553                  * to ide_dma_end() if this is dma interrupt.
1554                  */
1555                 hwif->ide_dma_clear_irq(drive);
1556
1557         if (drive->unmask)
1558                 local_irq_enable_in_hardirq();
1559         /* service this interrupt, may set handler for next interrupt */
1560         startstop = handler(drive);
1561         spin_lock_irq(&ide_lock);
1562
1563         /*
1564          * Note that handler() may have set things up for another
1565          * interrupt to occur soon, but it cannot happen until
1566          * we exit from this routine, because it will be the
1567          * same irq as is currently being serviced here, and Linux
1568          * won't allow another of the same (on any CPU) until we return.
1569          */
1570         drive->service_time = jiffies - drive->service_start;
1571         if (startstop == ide_stopped) {
1572                 if (hwgroup->handler == NULL) { /* paranoia */
1573                         hwgroup->busy = 0;
1574                         ide_do_request(hwgroup, hwif->irq);
1575                 } else {
1576                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1577                                 "on exit\n", drive->name);
1578                 }
1579         }
1580         spin_unlock_irqrestore(&ide_lock, flags);
1581         return IRQ_HANDLED;
1582 }
1583
1584 /**
1585  *      ide_init_drive_cmd      -       initialize a drive command request
1586  *      @rq: request object
1587  *
1588  *      Initialize a request before we fill it in and send it down to
1589  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1590  *      now it doesn't do a lot, but if that changes abusers will have a
1591  *      nasty surprise.
1592  */
1593
1594 void ide_init_drive_cmd (struct request *rq)
1595 {
1596         memset(rq, 0, sizeof(*rq));
1597         rq->ref_count = 1;
1598 }
1599
1600 EXPORT_SYMBOL(ide_init_drive_cmd);
1601
1602 /**
1603  *      ide_do_drive_cmd        -       issue IDE special command
1604  *      @drive: device to issue command
1605  *      @rq: request to issue
1606  *      @action: action for processing
1607  *
1608  *      This function issues a special IDE device request
1609  *      onto the request queue.
1610  *
1611  *      If action is ide_wait, then the rq is queued at the end of the
1612  *      request queue, and the function sleeps until it has been processed.
1613  *      This is for use when invoked from an ioctl handler.
1614  *
1615  *      If action is ide_preempt, then the rq is queued at the head of
1616  *      the request queue, displacing the currently-being-processed
1617  *      request and this function returns immediately without waiting
1618  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1619  *      intended for careful use by the ATAPI tape/cdrom driver code.
1620  *
1621  *      If action is ide_end, then the rq is queued at the end of the
1622  *      request queue, and the function returns immediately without waiting
1623  *      for the new rq to be completed. This is again intended for careful
1624  *      use by the ATAPI tape/cdrom driver code.
1625  */
1626  
1627 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1628 {
1629         unsigned long flags;
1630         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1631         DECLARE_COMPLETION_ONSTACK(wait);
1632         int where = ELEVATOR_INSERT_BACK, err;
1633         int must_wait = (action == ide_wait || action == ide_head_wait);
1634
1635         rq->errors = 0;
1636
1637         /*
1638          * we need to hold an extra reference to request for safe inspection
1639          * after completion
1640          */
1641         if (must_wait) {
1642                 rq->ref_count++;
1643                 rq->end_io_data = &wait;
1644                 rq->end_io = blk_end_sync_rq;
1645         }
1646
1647         spin_lock_irqsave(&ide_lock, flags);
1648         if (action == ide_preempt)
1649                 hwgroup->rq = NULL;
1650         if (action == ide_preempt || action == ide_head_wait) {
1651                 where = ELEVATOR_INSERT_FRONT;
1652                 rq->cmd_flags |= REQ_PREEMPT;
1653         }
1654         __elv_add_request(drive->queue, rq, where, 0);
1655         ide_do_request(hwgroup, IDE_NO_IRQ);
1656         spin_unlock_irqrestore(&ide_lock, flags);
1657
1658         err = 0;
1659         if (must_wait) {
1660                 wait_for_completion(&wait);
1661                 if (rq->errors)
1662                         err = -EIO;
1663
1664                 blk_put_request(rq);
1665         }
1666
1667         return err;
1668 }
1669
1670 EXPORT_SYMBOL(ide_do_drive_cmd);
1671
1672 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1673 {
1674         ide_task_t task;
1675
1676         memset(&task, 0, sizeof(task));
1677         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1678                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1679         task.tf.feature = dma;          /* Use PIO/DMA */
1680         task.tf.lbam    = bcount & 0xff;
1681         task.tf.lbah    = (bcount >> 8) & 0xff;
1682
1683         ide_tf_load(drive, &task);
1684 }
1685
1686 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);