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