Merge branch 'for-2.6.27' of git://linux-nfs.org/~bfields/linux
[linux-2.6] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
32 #include <linux/mm.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50 #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 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
362 {
363         if (rq->rq_disk) {
364                 ide_driver_t *drv;
365
366                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
367                 drv->end_request(drive, 0, 0);
368         } else
369                 ide_end_request(drive, 0, 0);
370 }
371
372 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
373 {
374         ide_hwif_t *hwif = drive->hwif;
375
376         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
377                 /* other bits are useless when BUSY */
378                 rq->errors |= ERROR_RESET;
379         } else if (stat & ERR_STAT) {
380                 /* err has different meaning on cdrom and tape */
381                 if (err == ABRT_ERR) {
382                         if (drive->select.b.lba &&
383                             /* some newer drives don't support WIN_SPECIFY */
384                             hwif->INB(hwif->io_ports.command_addr) ==
385                                 WIN_SPECIFY)
386                                 return ide_stopped;
387                 } else if ((err & BAD_CRC) == BAD_CRC) {
388                         /* UDMA crc error, just retry the operation */
389                         drive->crc_count++;
390                 } else if (err & (BBD_ERR | ECC_ERR)) {
391                         /* retries won't help these */
392                         rq->errors = ERROR_MAX;
393                 } else if (err & TRK0_ERR) {
394                         /* help it find track zero */
395                         rq->errors |= ERROR_RECAL;
396                 }
397         }
398
399         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ &&
400             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
401                 int nsect = drive->mult_count ? drive->mult_count : 1;
402
403                 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
404         }
405
406         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
407                 ide_kill_rq(drive, rq);
408                 return ide_stopped;
409         }
410
411         if (ide_read_status(drive) & (BUSY_STAT | DRQ_STAT))
412                 rq->errors |= ERROR_RESET;
413
414         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
415                 ++rq->errors;
416                 return ide_do_reset(drive);
417         }
418
419         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
420                 drive->special.b.recalibrate = 1;
421
422         ++rq->errors;
423
424         return ide_stopped;
425 }
426
427 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
428 {
429         ide_hwif_t *hwif = drive->hwif;
430
431         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
432                 /* other bits are useless when BUSY */
433                 rq->errors |= ERROR_RESET;
434         } else {
435                 /* add decoding error stuff */
436         }
437
438         if (ide_read_status(drive) & (BUSY_STAT | DRQ_STAT))
439                 /* force an abort */
440                 hwif->OUTBSYNC(hwif, WIN_IDLEIMMEDIATE,
441                                hwif->io_ports.command_addr);
442
443         if (rq->errors >= ERROR_MAX) {
444                 ide_kill_rq(drive, rq);
445         } else {
446                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
447                         ++rq->errors;
448                         return ide_do_reset(drive);
449                 }
450                 ++rq->errors;
451         }
452
453         return ide_stopped;
454 }
455
456 ide_startstop_t
457 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
458 {
459         if (drive->media == ide_disk)
460                 return ide_ata_error(drive, rq, stat, err);
461         return ide_atapi_error(drive, rq, stat, err);
462 }
463
464 EXPORT_SYMBOL_GPL(__ide_error);
465
466 /**
467  *      ide_error       -       handle an error on the IDE
468  *      @drive: drive the error occurred on
469  *      @msg: message to report
470  *      @stat: status bits
471  *
472  *      ide_error() takes action based on the error returned by the drive.
473  *      For normal I/O that may well include retries. We deal with
474  *      both new-style (taskfile) and old style command handling here.
475  *      In the case of taskfile command handling there is work left to
476  *      do
477  */
478  
479 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
480 {
481         struct request *rq;
482         u8 err;
483
484         err = ide_dump_status(drive, msg, stat);
485
486         if ((rq = HWGROUP(drive)->rq) == NULL)
487                 return ide_stopped;
488
489         /* retry only "normal" I/O: */
490         if (!blk_fs_request(rq)) {
491                 rq->errors = 1;
492                 ide_end_drive_cmd(drive, stat, err);
493                 return ide_stopped;
494         }
495
496         if (rq->rq_disk) {
497                 ide_driver_t *drv;
498
499                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
500                 return drv->error(drive, rq, stat, err);
501         } else
502                 return __ide_error(drive, rq, stat, err);
503 }
504
505 EXPORT_SYMBOL_GPL(ide_error);
506
507 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
508 {
509         tf->nsect   = drive->sect;
510         tf->lbal    = drive->sect;
511         tf->lbam    = drive->cyl;
512         tf->lbah    = drive->cyl >> 8;
513         tf->device  = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
514         tf->command = WIN_SPECIFY;
515 }
516
517 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
518 {
519         tf->nsect   = drive->sect;
520         tf->command = WIN_RESTORE;
521 }
522
523 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
524 {
525         tf->nsect   = drive->mult_req;
526         tf->command = WIN_SETMULT;
527 }
528
529 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
530 {
531         special_t *s = &drive->special;
532         ide_task_t args;
533
534         memset(&args, 0, sizeof(ide_task_t));
535         args.data_phase = TASKFILE_NO_DATA;
536
537         if (s->b.set_geometry) {
538                 s->b.set_geometry = 0;
539                 ide_tf_set_specify_cmd(drive, &args.tf);
540         } else if (s->b.recalibrate) {
541                 s->b.recalibrate = 0;
542                 ide_tf_set_restore_cmd(drive, &args.tf);
543         } else if (s->b.set_multmode) {
544                 s->b.set_multmode = 0;
545                 if (drive->mult_req > drive->id->max_multsect)
546                         drive->mult_req = drive->id->max_multsect;
547                 ide_tf_set_setmult_cmd(drive, &args.tf);
548         } else if (s->all) {
549                 int special = s->all;
550                 s->all = 0;
551                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
552                 return ide_stopped;
553         }
554
555         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
556                         IDE_TFLAG_CUSTOM_HANDLER;
557
558         do_rw_taskfile(drive, &args);
559
560         return ide_started;
561 }
562
563 /*
564  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
565  */
566 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
567 {
568         switch (req_pio) {
569         case 202:
570         case 201:
571         case 200:
572         case 102:
573         case 101:
574         case 100:
575                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
576         case 9:
577         case 8:
578                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
579         case 7:
580         case 6:
581                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
582         default:
583                 return 0;
584         }
585 }
586
587 /**
588  *      do_special              -       issue some special commands
589  *      @drive: drive the command is for
590  *
591  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
592  *      commands to a drive.  It used to do much more, but has been scaled
593  *      back.
594  */
595
596 static ide_startstop_t do_special (ide_drive_t *drive)
597 {
598         special_t *s = &drive->special;
599
600 #ifdef DEBUG
601         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
602 #endif
603         if (s->b.set_tune) {
604                 ide_hwif_t *hwif = drive->hwif;
605                 const struct ide_port_ops *port_ops = hwif->port_ops;
606                 u8 req_pio = drive->tune_req;
607
608                 s->b.set_tune = 0;
609
610                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
611                         /*
612                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
613                          */
614                         if (req_pio == 8 || req_pio == 9) {
615                                 unsigned long flags;
616
617                                 spin_lock_irqsave(&ide_lock, flags);
618                                 port_ops->set_pio_mode(drive, req_pio);
619                                 spin_unlock_irqrestore(&ide_lock, flags);
620                         } else
621                                 port_ops->set_pio_mode(drive, req_pio);
622                 } else {
623                         int keep_dma = drive->using_dma;
624
625                         ide_set_pio(drive, req_pio);
626
627                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
628                                 if (keep_dma)
629                                         ide_dma_on(drive);
630                         }
631                 }
632
633                 return ide_stopped;
634         } else {
635                 if (drive->media == ide_disk)
636                         return ide_disk_special(drive);
637
638                 s->all = 0;
639                 drive->mult_req = 0;
640                 return ide_stopped;
641         }
642 }
643
644 void ide_map_sg(ide_drive_t *drive, struct request *rq)
645 {
646         ide_hwif_t *hwif = drive->hwif;
647         struct scatterlist *sg = hwif->sg_table;
648
649         if (hwif->sg_mapped)    /* needed by ide-scsi */
650                 return;
651
652         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
653                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
654         } else {
655                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
656                 hwif->sg_nents = 1;
657         }
658 }
659
660 EXPORT_SYMBOL_GPL(ide_map_sg);
661
662 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
663 {
664         ide_hwif_t *hwif = drive->hwif;
665
666         hwif->nsect = hwif->nleft = rq->nr_sectors;
667         hwif->cursg_ofs = 0;
668         hwif->cursg = NULL;
669 }
670
671 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
672
673 /**
674  *      execute_drive_command   -       issue special drive command
675  *      @drive: the drive to issue the command on
676  *      @rq: the request structure holding the command
677  *
678  *      execute_drive_cmd() issues a special drive command,  usually 
679  *      initiated by ioctl() from the external hdparm program. The
680  *      command can be a drive command, drive task or taskfile 
681  *      operation. Weirdly you can call it with NULL to wait for
682  *      all commands to finish. Don't do this as that is due to change
683  */
684
685 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
686                 struct request *rq)
687 {
688         ide_hwif_t *hwif = HWIF(drive);
689         ide_task_t *task = rq->special;
690
691         if (task) {
692                 hwif->data_phase = task->data_phase;
693
694                 switch (hwif->data_phase) {
695                 case TASKFILE_MULTI_OUT:
696                 case TASKFILE_OUT:
697                 case TASKFILE_MULTI_IN:
698                 case TASKFILE_IN:
699                         ide_init_sg_cmd(drive, rq);
700                         ide_map_sg(drive, rq);
701                 default:
702                         break;
703                 }
704
705                 return do_rw_taskfile(drive, task);
706         }
707
708         /*
709          * NULL is actually a valid way of waiting for
710          * all current requests to be flushed from the queue.
711          */
712 #ifdef DEBUG
713         printk("%s: DRIVE_CMD (null)\n", drive->name);
714 #endif
715         ide_end_drive_cmd(drive, ide_read_status(drive), ide_read_error(drive));
716
717         return ide_stopped;
718 }
719
720 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
721 {
722         switch (rq->cmd[0]) {
723         case REQ_DRIVE_RESET:
724                 return ide_do_reset(drive);
725         default:
726                 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
727                 ide_end_request(drive, 0, 0);
728                 return ide_stopped;
729         }
730 }
731
732 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
733 {
734         struct request_pm_state *pm = rq->data;
735
736         if (blk_pm_suspend_request(rq) &&
737             pm->pm_step == ide_pm_state_start_suspend)
738                 /* Mark drive blocked when starting the suspend sequence. */
739                 drive->blocked = 1;
740         else if (blk_pm_resume_request(rq) &&
741                  pm->pm_step == ide_pm_state_start_resume) {
742                 /* 
743                  * The first thing we do on wakeup is to wait for BSY bit to
744                  * go away (with a looong timeout) as a drive on this hwif may
745                  * just be POSTing itself.
746                  * We do that before even selecting as the "other" device on
747                  * the bus may be broken enough to walk on our toes at this
748                  * point.
749                  */
750                 int rc;
751 #ifdef DEBUG_PM
752                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
753 #endif
754                 rc = ide_wait_not_busy(HWIF(drive), 35000);
755                 if (rc)
756                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
757                 SELECT_DRIVE(drive);
758                 ide_set_irq(drive, 1);
759                 rc = ide_wait_not_busy(HWIF(drive), 100000);
760                 if (rc)
761                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
762         }
763 }
764
765 /**
766  *      start_request   -       start of I/O and command issuing for IDE
767  *
768  *      start_request() initiates handling of a new I/O request. It
769  *      accepts commands and I/O (read/write) requests. It also does
770  *      the final remapping for weird stuff like EZDrive. Once 
771  *      device mapper can work sector level the EZDrive stuff can go away
772  *
773  *      FIXME: this function needs a rename
774  */
775  
776 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
777 {
778         ide_startstop_t startstop;
779         sector_t block;
780
781         BUG_ON(!blk_rq_started(rq));
782
783 #ifdef DEBUG
784         printk("%s: start_request: current=0x%08lx\n",
785                 HWIF(drive)->name, (unsigned long) rq);
786 #endif
787
788         /* bail early if we've exceeded max_failures */
789         if (drive->max_failures && (drive->failures > drive->max_failures)) {
790                 rq->cmd_flags |= REQ_FAILED;
791                 goto kill_rq;
792         }
793
794         block    = rq->sector;
795         if (blk_fs_request(rq) &&
796             (drive->media == ide_disk || drive->media == ide_floppy)) {
797                 block += drive->sect0;
798         }
799         /* Yecch - this will shift the entire interval,
800            possibly killing some innocent following sector */
801         if (block == 0 && drive->remap_0_to_1 == 1)
802                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
803
804         if (blk_pm_request(rq))
805                 ide_check_pm_state(drive, rq);
806
807         SELECT_DRIVE(drive);
808         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
809                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
810                 return startstop;
811         }
812         if (!drive->special.all) {
813                 ide_driver_t *drv;
814
815                 /*
816                  * We reset the drive so we need to issue a SETFEATURES.
817                  * Do it _after_ do_special() restored device parameters.
818                  */
819                 if (drive->current_speed == 0xff)
820                         ide_config_drive_speed(drive, drive->desired_speed);
821
822                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
823                         return execute_drive_cmd(drive, rq);
824                 else if (blk_pm_request(rq)) {
825                         struct request_pm_state *pm = rq->data;
826 #ifdef DEBUG_PM
827                         printk("%s: start_power_step(step: %d)\n",
828                                 drive->name, rq->pm->pm_step);
829 #endif
830                         startstop = ide_start_power_step(drive, rq);
831                         if (startstop == ide_stopped &&
832                             pm->pm_step == ide_pm_state_completed)
833                                 ide_complete_pm_request(drive, rq);
834                         return startstop;
835                 } else if (!rq->rq_disk && blk_special_request(rq))
836                         /*
837                          * TODO: Once all ULDs have been modified to
838                          * check for specific op codes rather than
839                          * blindly accepting any special request, the
840                          * check for ->rq_disk above may be replaced
841                          * by a more suitable mechanism or even
842                          * dropped entirely.
843                          */
844                         return ide_special_rq(drive, rq);
845
846                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
847                 return drv->do_request(drive, rq, block);
848         }
849         return do_special(drive);
850 kill_rq:
851         ide_kill_rq(drive, rq);
852         return ide_stopped;
853 }
854
855 /**
856  *      ide_stall_queue         -       pause an IDE device
857  *      @drive: drive to stall
858  *      @timeout: time to stall for (jiffies)
859  *
860  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
861  *      to the hwgroup by sleeping for timeout jiffies.
862  */
863  
864 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
865 {
866         if (timeout > WAIT_WORSTCASE)
867                 timeout = WAIT_WORSTCASE;
868         drive->sleep = timeout + jiffies;
869         drive->sleeping = 1;
870 }
871
872 EXPORT_SYMBOL(ide_stall_queue);
873
874 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
875
876 /**
877  *      choose_drive            -       select a drive to service
878  *      @hwgroup: hardware group to select on
879  *
880  *      choose_drive() selects the next drive which will be serviced.
881  *      This is necessary because the IDE layer can't issue commands
882  *      to both drives on the same cable, unlike SCSI.
883  */
884  
885 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
886 {
887         ide_drive_t *drive, *best;
888
889 repeat: 
890         best = NULL;
891         drive = hwgroup->drive;
892
893         /*
894          * drive is doing pre-flush, ordered write, post-flush sequence. even
895          * though that is 3 requests, it must be seen as a single transaction.
896          * we must not preempt this drive until that is complete
897          */
898         if (blk_queue_flushing(drive->queue)) {
899                 /*
900                  * small race where queue could get replugged during
901                  * the 3-request flush cycle, just yank the plug since
902                  * we want it to finish asap
903                  */
904                 blk_remove_plug(drive->queue);
905                 return drive;
906         }
907
908         do {
909                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
910                     && !elv_queue_empty(drive->queue)) {
911                         if (!best
912                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
913                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
914                         {
915                                 if (!blk_queue_plugged(drive->queue))
916                                         best = drive;
917                         }
918                 }
919         } while ((drive = drive->next) != hwgroup->drive);
920         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
921                 long t = (signed long)(WAKEUP(best) - jiffies);
922                 if (t >= WAIT_MIN_SLEEP) {
923                 /*
924                  * We *may* have some time to spare, but first let's see if
925                  * someone can potentially benefit from our nice mood today..
926                  */
927                         drive = best->next;
928                         do {
929                                 if (!drive->sleeping
930                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
931                                  && time_before(WAKEUP(drive), jiffies + t))
932                                 {
933                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
934                                         goto repeat;
935                                 }
936                         } while ((drive = drive->next) != best);
937                 }
938         }
939         return best;
940 }
941
942 /*
943  * Issue a new request to a drive from hwgroup
944  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
945  *
946  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
947  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
948  * may have both interfaces in a single hwgroup to "serialize" access.
949  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
950  * together into one hwgroup for serialized access.
951  *
952  * Note also that several hwgroups can end up sharing a single IRQ,
953  * possibly along with many other devices.  This is especially common in
954  * PCI-based systems with off-board IDE controller cards.
955  *
956  * The IDE driver uses the single global ide_lock spinlock to protect
957  * access to the request queues, and to protect the hwgroup->busy flag.
958  *
959  * The first thread into the driver for a particular hwgroup sets the
960  * hwgroup->busy flag to indicate that this hwgroup is now active,
961  * and then initiates processing of the top request from the request queue.
962  *
963  * Other threads attempting entry notice the busy setting, and will simply
964  * queue their new requests and exit immediately.  Note that hwgroup->busy
965  * remains set even when the driver is merely awaiting the next interrupt.
966  * Thus, the meaning is "this hwgroup is busy processing a request".
967  *
968  * When processing of a request completes, the completing thread or IRQ-handler
969  * will start the next request from the queue.  If no more work remains,
970  * the driver will clear the hwgroup->busy flag and exit.
971  *
972  * The ide_lock (spinlock) is used to protect all access to the
973  * hwgroup->busy flag, but is otherwise not needed for most processing in
974  * the driver.  This makes the driver much more friendlier to shared IRQs
975  * than previous designs, while remaining 100% (?) SMP safe and capable.
976  */
977 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
978 {
979         ide_drive_t     *drive;
980         ide_hwif_t      *hwif;
981         struct request  *rq;
982         ide_startstop_t startstop;
983         int             loops = 0;
984
985         /* for atari only: POSSIBLY BROKEN HERE(?) */
986         ide_get_lock(ide_intr, hwgroup);
987
988         /* caller must own ide_lock */
989         BUG_ON(!irqs_disabled());
990
991         while (!hwgroup->busy) {
992                 hwgroup->busy = 1;
993                 drive = choose_drive(hwgroup);
994                 if (drive == NULL) {
995                         int sleeping = 0;
996                         unsigned long sleep = 0; /* shut up, gcc */
997                         hwgroup->rq = NULL;
998                         drive = hwgroup->drive;
999                         do {
1000                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1001                                         sleeping = 1;
1002                                         sleep = drive->sleep;
1003                                 }
1004                         } while ((drive = drive->next) != hwgroup->drive);
1005                         if (sleeping) {
1006                 /*
1007                  * Take a short snooze, and then wake up this hwgroup again.
1008                  * This gives other hwgroups on the same a chance to
1009                  * play fairly with us, just in case there are big differences
1010                  * in relative throughputs.. don't want to hog the cpu too much.
1011                  */
1012                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1013                                         sleep = jiffies + WAIT_MIN_SLEEP;
1014 #if 1
1015                                 if (timer_pending(&hwgroup->timer))
1016                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1017 #endif
1018                                 /* so that ide_timer_expiry knows what to do */
1019                                 hwgroup->sleeping = 1;
1020                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1021                                 mod_timer(&hwgroup->timer, sleep);
1022                                 /* we purposely leave hwgroup->busy==1
1023                                  * while sleeping */
1024                         } else {
1025                                 /* Ugly, but how can we sleep for the lock
1026                                  * otherwise? perhaps from tq_disk?
1027                                  */
1028
1029                                 /* for atari only */
1030                                 ide_release_lock();
1031                                 hwgroup->busy = 0;
1032                         }
1033
1034                         /* no more work for this hwgroup (for now) */
1035                         return;
1036                 }
1037         again:
1038                 hwif = HWIF(drive);
1039                 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1040                         /*
1041                          * set nIEN for previous hwif, drives in the
1042                          * quirk_list may not like intr setups/cleanups
1043                          */
1044                         if (drive->quirk_list != 1)
1045                                 ide_set_irq(drive, 0);
1046                 }
1047                 hwgroup->hwif = hwif;
1048                 hwgroup->drive = drive;
1049                 drive->sleeping = 0;
1050                 drive->service_start = jiffies;
1051
1052                 if (blk_queue_plugged(drive->queue)) {
1053                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1054                         break;
1055                 }
1056
1057                 /*
1058                  * we know that the queue isn't empty, but this can happen
1059                  * if the q->prep_rq_fn() decides to kill a request
1060                  */
1061                 rq = elv_next_request(drive->queue);
1062                 if (!rq) {
1063                         hwgroup->busy = 0;
1064                         break;
1065                 }
1066
1067                 /*
1068                  * Sanity: don't accept a request that isn't a PM request
1069                  * if we are currently power managed. This is very important as
1070                  * blk_stop_queue() doesn't prevent the elv_next_request()
1071                  * above to return us whatever is in the queue. Since we call
1072                  * ide_do_request() ourselves, we end up taking requests while
1073                  * the queue is blocked...
1074                  * 
1075                  * We let requests forced at head of queue with ide-preempt
1076                  * though. I hope that doesn't happen too much, hopefully not
1077                  * unless the subdriver triggers such a thing in its own PM
1078                  * state machine.
1079                  *
1080                  * We count how many times we loop here to make sure we service
1081                  * all drives in the hwgroup without looping for ever
1082                  */
1083                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1084                         drive = drive->next ? drive->next : hwgroup->drive;
1085                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1086                                 goto again;
1087                         /* We clear busy, there should be no pending ATA command at this point. */
1088                         hwgroup->busy = 0;
1089                         break;
1090                 }
1091
1092                 hwgroup->rq = rq;
1093
1094                 /*
1095                  * Some systems have trouble with IDE IRQs arriving while
1096                  * the driver is still setting things up.  So, here we disable
1097                  * the IRQ used by this interface while the request is being started.
1098                  * This may look bad at first, but pretty much the same thing
1099                  * happens anyway when any interrupt comes in, IDE or otherwise
1100                  *  -- the kernel masks the IRQ while it is being handled.
1101                  */
1102                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1103                         disable_irq_nosync(hwif->irq);
1104                 spin_unlock(&ide_lock);
1105                 local_irq_enable_in_hardirq();
1106                         /* allow other IRQs while we start this request */
1107                 startstop = start_request(drive, rq);
1108                 spin_lock_irq(&ide_lock);
1109                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1110                         enable_irq(hwif->irq);
1111                 if (startstop == ide_stopped)
1112                         hwgroup->busy = 0;
1113         }
1114 }
1115
1116 /*
1117  * Passes the stuff to ide_do_request
1118  */
1119 void do_ide_request(struct request_queue *q)
1120 {
1121         ide_drive_t *drive = q->queuedata;
1122
1123         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1124 }
1125
1126 /*
1127  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1128  * retry the current request in pio mode instead of risking tossing it
1129  * all away
1130  */
1131 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1132 {
1133         ide_hwif_t *hwif = HWIF(drive);
1134         struct request *rq;
1135         ide_startstop_t ret = ide_stopped;
1136
1137         /*
1138          * end current dma transaction
1139          */
1140
1141         if (error < 0) {
1142                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1143                 (void)hwif->dma_ops->dma_end(drive);
1144                 ret = ide_error(drive, "dma timeout error",
1145                                 ide_read_status(drive));
1146         } else {
1147                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1148                 hwif->dma_ops->dma_timeout(drive);
1149         }
1150
1151         /*
1152          * disable dma for now, but remember that we did so because of
1153          * a timeout -- we'll reenable after we finish this next request
1154          * (or rather the first chunk of it) in pio.
1155          */
1156         drive->retry_pio++;
1157         drive->state = DMA_PIO_RETRY;
1158         ide_dma_off_quietly(drive);
1159
1160         /*
1161          * un-busy drive etc (hwgroup->busy is cleared on return) and
1162          * make sure request is sane
1163          */
1164         rq = HWGROUP(drive)->rq;
1165
1166         if (!rq)
1167                 goto out;
1168
1169         HWGROUP(drive)->rq = NULL;
1170
1171         rq->errors = 0;
1172
1173         if (!rq->bio)
1174                 goto out;
1175
1176         rq->sector = rq->bio->bi_sector;
1177         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1178         rq->hard_cur_sectors = rq->current_nr_sectors;
1179         rq->buffer = bio_data(rq->bio);
1180 out:
1181         return ret;
1182 }
1183
1184 /**
1185  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1186  *      @data: timer callback magic (hwgroup)
1187  *
1188  *      An IDE command has timed out before the expected drive return
1189  *      occurred. At this point we attempt to clean up the current
1190  *      mess. If the current handler includes an expiry handler then
1191  *      we invoke the expiry handler, and providing it is happy the
1192  *      work is done. If that fails we apply generic recovery rules
1193  *      invoking the handler and checking the drive DMA status. We
1194  *      have an excessively incestuous relationship with the DMA
1195  *      logic that wants cleaning up.
1196  */
1197  
1198 void ide_timer_expiry (unsigned long data)
1199 {
1200         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1201         ide_handler_t   *handler;
1202         ide_expiry_t    *expiry;
1203         unsigned long   flags;
1204         unsigned long   wait = -1;
1205
1206         spin_lock_irqsave(&ide_lock, flags);
1207
1208         if (((handler = hwgroup->handler) == NULL) ||
1209             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1210                 /*
1211                  * Either a marginal timeout occurred
1212                  * (got the interrupt just as timer expired),
1213                  * or we were "sleeping" to give other devices a chance.
1214                  * Either way, we don't really want to complain about anything.
1215                  */
1216                 if (hwgroup->sleeping) {
1217                         hwgroup->sleeping = 0;
1218                         hwgroup->busy = 0;
1219                 }
1220         } else {
1221                 ide_drive_t *drive = hwgroup->drive;
1222                 if (!drive) {
1223                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1224                         hwgroup->handler = NULL;
1225                 } else {
1226                         ide_hwif_t *hwif;
1227                         ide_startstop_t startstop = ide_stopped;
1228                         if (!hwgroup->busy) {
1229                                 hwgroup->busy = 1;      /* paranoia */
1230                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1231                         }
1232                         if ((expiry = hwgroup->expiry) != NULL) {
1233                                 /* continue */
1234                                 if ((wait = expiry(drive)) > 0) {
1235                                         /* reset timer */
1236                                         hwgroup->timer.expires  = jiffies + wait;
1237                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1238                                         add_timer(&hwgroup->timer);
1239                                         spin_unlock_irqrestore(&ide_lock, flags);
1240                                         return;
1241                                 }
1242                         }
1243                         hwgroup->handler = NULL;
1244                         /*
1245                          * We need to simulate a real interrupt when invoking
1246                          * the handler() function, which means we need to
1247                          * globally mask the specific IRQ:
1248                          */
1249                         spin_unlock(&ide_lock);
1250                         hwif  = HWIF(drive);
1251                         /* disable_irq_nosync ?? */
1252                         disable_irq(hwif->irq);
1253                         /* local CPU only,
1254                          * as if we were handling an interrupt */
1255                         local_irq_disable();
1256                         if (hwgroup->polling) {
1257                                 startstop = handler(drive);
1258                         } else if (drive_is_ready(drive)) {
1259                                 if (drive->waiting_for_dma)
1260                                         hwif->dma_ops->dma_lost_irq(drive);
1261                                 (void)ide_ack_intr(hwif);
1262                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1263                                 startstop = handler(drive);
1264                         } else {
1265                                 if (drive->waiting_for_dma) {
1266                                         startstop = ide_dma_timeout_retry(drive, wait);
1267                                 } else
1268                                         startstop =
1269                                         ide_error(drive, "irq timeout",
1270                                                   ide_read_status(drive));
1271                         }
1272                         drive->service_time = jiffies - drive->service_start;
1273                         spin_lock_irq(&ide_lock);
1274                         enable_irq(hwif->irq);
1275                         if (startstop == ide_stopped)
1276                                 hwgroup->busy = 0;
1277                 }
1278         }
1279         ide_do_request(hwgroup, IDE_NO_IRQ);
1280         spin_unlock_irqrestore(&ide_lock, flags);
1281 }
1282
1283 /**
1284  *      unexpected_intr         -       handle an unexpected IDE interrupt
1285  *      @irq: interrupt line
1286  *      @hwgroup: hwgroup being processed
1287  *
1288  *      There's nothing really useful we can do with an unexpected interrupt,
1289  *      other than reading the status register (to clear it), and logging it.
1290  *      There should be no way that an irq can happen before we're ready for it,
1291  *      so we needn't worry much about losing an "important" interrupt here.
1292  *
1293  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1294  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1295  *      looks "good", we just ignore the interrupt completely.
1296  *
1297  *      This routine assumes __cli() is in effect when called.
1298  *
1299  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1300  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1301  *      we could screw up by interfering with a new request being set up for 
1302  *      irq15.
1303  *
1304  *      In reality, this is a non-issue.  The new command is not sent unless 
1305  *      the drive is ready to accept one, in which case we know the drive is
1306  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1307  *      before completing the issuance of any new drive command, so we will not
1308  *      be accidentally invoked as a result of any valid command completion
1309  *      interrupt.
1310  *
1311  *      Note that we must walk the entire hwgroup here. We know which hwif
1312  *      is doing the current command, but we don't know which hwif burped
1313  *      mysteriously.
1314  */
1315  
1316 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1317 {
1318         u8 stat;
1319         ide_hwif_t *hwif = hwgroup->hwif;
1320
1321         /*
1322          * handle the unexpected interrupt
1323          */
1324         do {
1325                 if (hwif->irq == irq) {
1326                         stat = hwif->INB(hwif->io_ports.status_addr);
1327                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1328                                 /* Try to not flood the console with msgs */
1329                                 static unsigned long last_msgtime, count;
1330                                 ++count;
1331                                 if (time_after(jiffies, last_msgtime + HZ)) {
1332                                         last_msgtime = jiffies;
1333                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1334                                                 "status=0x%02x, count=%ld\n",
1335                                                 hwif->name,
1336                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1337                                 }
1338                         }
1339                 }
1340         } while ((hwif = hwif->next) != hwgroup->hwif);
1341 }
1342
1343 /**
1344  *      ide_intr        -       default IDE interrupt handler
1345  *      @irq: interrupt number
1346  *      @dev_id: hwif group
1347  *      @regs: unused weirdness from the kernel irq layer
1348  *
1349  *      This is the default IRQ handler for the IDE layer. You should
1350  *      not need to override it. If you do be aware it is subtle in
1351  *      places
1352  *
1353  *      hwgroup->hwif is the interface in the group currently performing
1354  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1355  *      the IRQ handler to call. As we issue a command the handlers
1356  *      step through multiple states, reassigning the handler to the
1357  *      next step in the process. Unlike a smart SCSI controller IDE
1358  *      expects the main processor to sequence the various transfer
1359  *      stages. We also manage a poll timer to catch up with most
1360  *      timeout situations. There are still a few where the handlers
1361  *      don't ever decide to give up.
1362  *
1363  *      The handler eventually returns ide_stopped to indicate the
1364  *      request completed. At this point we issue the next request
1365  *      on the hwgroup and the process begins again.
1366  */
1367  
1368 irqreturn_t ide_intr (int irq, void *dev_id)
1369 {
1370         unsigned long flags;
1371         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1372         ide_hwif_t *hwif;
1373         ide_drive_t *drive;
1374         ide_handler_t *handler;
1375         ide_startstop_t startstop;
1376
1377         spin_lock_irqsave(&ide_lock, flags);
1378         hwif = hwgroup->hwif;
1379
1380         if (!ide_ack_intr(hwif)) {
1381                 spin_unlock_irqrestore(&ide_lock, flags);
1382                 return IRQ_NONE;
1383         }
1384
1385         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1386                 /*
1387                  * Not expecting an interrupt from this drive.
1388                  * That means this could be:
1389                  *      (1) an interrupt from another PCI device
1390                  *      sharing the same PCI INT# as us.
1391                  * or   (2) a drive just entered sleep or standby mode,
1392                  *      and is interrupting to let us know.
1393                  * or   (3) a spurious interrupt of unknown origin.
1394                  *
1395                  * For PCI, we cannot tell the difference,
1396                  * so in that case we just ignore it and hope it goes away.
1397                  *
1398                  * FIXME: unexpected_intr should be hwif-> then we can
1399                  * remove all the ifdef PCI crap
1400                  */
1401 #ifdef CONFIG_BLK_DEV_IDEPCI
1402                 if (hwif->chipset != ide_pci)
1403 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1404                 {
1405                         /*
1406                          * Probably not a shared PCI interrupt,
1407                          * so we can safely try to do something about it:
1408                          */
1409                         unexpected_intr(irq, hwgroup);
1410 #ifdef CONFIG_BLK_DEV_IDEPCI
1411                 } else {
1412                         /*
1413                          * Whack the status register, just in case
1414                          * we have a leftover pending IRQ.
1415                          */
1416                         (void) hwif->INB(hwif->io_ports.status_addr);
1417 #endif /* CONFIG_BLK_DEV_IDEPCI */
1418                 }
1419                 spin_unlock_irqrestore(&ide_lock, flags);
1420                 return IRQ_NONE;
1421         }
1422         drive = hwgroup->drive;
1423         if (!drive) {
1424                 /*
1425                  * This should NEVER happen, and there isn't much
1426                  * we could do about it here.
1427                  *
1428                  * [Note - this can occur if the drive is hot unplugged]
1429                  */
1430                 spin_unlock_irqrestore(&ide_lock, flags);
1431                 return IRQ_HANDLED;
1432         }
1433         if (!drive_is_ready(drive)) {
1434                 /*
1435                  * This happens regularly when we share a PCI IRQ with
1436                  * another device.  Unfortunately, it can also happen
1437                  * with some buggy drives that trigger the IRQ before
1438                  * their status register is up to date.  Hopefully we have
1439                  * enough advance overhead that the latter isn't a problem.
1440                  */
1441                 spin_unlock_irqrestore(&ide_lock, flags);
1442                 return IRQ_NONE;
1443         }
1444         if (!hwgroup->busy) {
1445                 hwgroup->busy = 1;      /* paranoia */
1446                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1447         }
1448         hwgroup->handler = NULL;
1449         hwgroup->req_gen++;
1450         del_timer(&hwgroup->timer);
1451         spin_unlock(&ide_lock);
1452
1453         /* Some controllers might set DMA INTR no matter DMA or PIO;
1454          * bmdma status might need to be cleared even for
1455          * PIO interrupts to prevent spurious/lost irq.
1456          */
1457         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1458                 /* ide_dma_end() needs bmdma status for error checking.
1459                  * So, skip clearing bmdma status here and leave it
1460                  * to ide_dma_end() if this is dma interrupt.
1461                  */
1462                 hwif->ide_dma_clear_irq(drive);
1463
1464         if (drive->unmask)
1465                 local_irq_enable_in_hardirq();
1466         /* service this interrupt, may set handler for next interrupt */
1467         startstop = handler(drive);
1468         spin_lock_irq(&ide_lock);
1469
1470         /*
1471          * Note that handler() may have set things up for another
1472          * interrupt to occur soon, but it cannot happen until
1473          * we exit from this routine, because it will be the
1474          * same irq as is currently being serviced here, and Linux
1475          * won't allow another of the same (on any CPU) until we return.
1476          */
1477         drive->service_time = jiffies - drive->service_start;
1478         if (startstop == ide_stopped) {
1479                 if (hwgroup->handler == NULL) { /* paranoia */
1480                         hwgroup->busy = 0;
1481                         ide_do_request(hwgroup, hwif->irq);
1482                 } else {
1483                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1484                                 "on exit\n", drive->name);
1485                 }
1486         }
1487         spin_unlock_irqrestore(&ide_lock, flags);
1488         return IRQ_HANDLED;
1489 }
1490
1491 /**
1492  *      ide_do_drive_cmd        -       issue IDE special command
1493  *      @drive: device to issue command
1494  *      @rq: request to issue
1495  *
1496  *      This function issues a special IDE device request
1497  *      onto the request queue.
1498  *
1499  *      the rq is queued at the head of the request queue, displacing
1500  *      the currently-being-processed request and this function
1501  *      returns immediately without waiting for the new rq to be
1502  *      completed.  This is VERY DANGEROUS, and is intended for
1503  *      careful use by the ATAPI tape/cdrom driver code.
1504  */
1505
1506 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1507 {
1508         unsigned long flags;
1509         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1510
1511         spin_lock_irqsave(&ide_lock, flags);
1512         hwgroup->rq = NULL;
1513         __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1514         __generic_unplug_device(drive->queue);
1515         spin_unlock_irqrestore(&ide_lock, flags);
1516 }
1517
1518 EXPORT_SYMBOL(ide_do_drive_cmd);
1519
1520 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1521 {
1522         ide_task_t task;
1523
1524         memset(&task, 0, sizeof(task));
1525         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1526                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1527         task.tf.feature = dma;          /* Use PIO/DMA */
1528         task.tf.lbam    = bcount & 0xff;
1529         task.tf.lbah    = (bcount >> 8) & 0xff;
1530
1531         ide_tf_dump(drive->name, &task.tf);
1532         ide_set_irq(drive, 1);
1533         SELECT_MASK(drive, 0);
1534         drive->hwif->tf_load(drive, &task);
1535 }
1536
1537 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1538
1539 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1540 {
1541         ide_hwif_t *hwif = drive->hwif;
1542         u8 buf[4] = { 0 };
1543
1544         while (len > 0) {
1545                 if (write)
1546                         hwif->output_data(drive, NULL, buf, min(4, len));
1547                 else
1548                         hwif->input_data(drive, NULL, buf, min(4, len));
1549                 len -= 4;
1550         }
1551 }
1552 EXPORT_SYMBOL_GPL(ide_pad_transfer);