Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[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 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
508 {
509         if (drive->media != ide_disk)
510                 rq->errors |= ERROR_RESET;
511
512         ide_kill_rq(drive, rq);
513
514         return ide_stopped;
515 }
516
517 EXPORT_SYMBOL_GPL(__ide_abort);
518
519 /**
520  *      ide_abort       -       abort pending IDE operations
521  *      @drive: drive the error occurred on
522  *      @msg: message to report
523  *
524  *      ide_abort kills and cleans up when we are about to do a 
525  *      host initiated reset on active commands. Longer term we
526  *      want handlers to have sensible abort handling themselves
527  *
528  *      This differs fundamentally from ide_error because in 
529  *      this case the command is doing just fine when we
530  *      blow it away.
531  */
532  
533 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
534 {
535         struct request *rq;
536
537         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
538                 return ide_stopped;
539
540         /* retry only "normal" I/O: */
541         if (!blk_fs_request(rq)) {
542                 rq->errors = 1;
543                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
544                 return ide_stopped;
545         }
546
547         if (rq->rq_disk) {
548                 ide_driver_t *drv;
549
550                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
551                 return drv->abort(drive, rq);
552         } else
553                 return __ide_abort(drive, rq);
554 }
555
556 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
557 {
558         tf->nsect   = drive->sect;
559         tf->lbal    = drive->sect;
560         tf->lbam    = drive->cyl;
561         tf->lbah    = drive->cyl >> 8;
562         tf->device  = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
563         tf->command = WIN_SPECIFY;
564 }
565
566 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
567 {
568         tf->nsect   = drive->sect;
569         tf->command = WIN_RESTORE;
570 }
571
572 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
573 {
574         tf->nsect   = drive->mult_req;
575         tf->command = WIN_SETMULT;
576 }
577
578 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
579 {
580         special_t *s = &drive->special;
581         ide_task_t args;
582
583         memset(&args, 0, sizeof(ide_task_t));
584         args.data_phase = TASKFILE_NO_DATA;
585
586         if (s->b.set_geometry) {
587                 s->b.set_geometry = 0;
588                 ide_tf_set_specify_cmd(drive, &args.tf);
589         } else if (s->b.recalibrate) {
590                 s->b.recalibrate = 0;
591                 ide_tf_set_restore_cmd(drive, &args.tf);
592         } else if (s->b.set_multmode) {
593                 s->b.set_multmode = 0;
594                 if (drive->mult_req > drive->id->max_multsect)
595                         drive->mult_req = drive->id->max_multsect;
596                 ide_tf_set_setmult_cmd(drive, &args.tf);
597         } else if (s->all) {
598                 int special = s->all;
599                 s->all = 0;
600                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
601                 return ide_stopped;
602         }
603
604         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
605                         IDE_TFLAG_CUSTOM_HANDLER;
606
607         do_rw_taskfile(drive, &args);
608
609         return ide_started;
610 }
611
612 /*
613  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
614  */
615 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
616 {
617         switch (req_pio) {
618         case 202:
619         case 201:
620         case 200:
621         case 102:
622         case 101:
623         case 100:
624                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
625         case 9:
626         case 8:
627                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
628         case 7:
629         case 6:
630                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
631         default:
632                 return 0;
633         }
634 }
635
636 /**
637  *      do_special              -       issue some special commands
638  *      @drive: drive the command is for
639  *
640  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
641  *      commands to a drive.  It used to do much more, but has been scaled
642  *      back.
643  */
644
645 static ide_startstop_t do_special (ide_drive_t *drive)
646 {
647         special_t *s = &drive->special;
648
649 #ifdef DEBUG
650         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
651 #endif
652         if (s->b.set_tune) {
653                 ide_hwif_t *hwif = drive->hwif;
654                 const struct ide_port_ops *port_ops = hwif->port_ops;
655                 u8 req_pio = drive->tune_req;
656
657                 s->b.set_tune = 0;
658
659                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
660                         /*
661                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
662                          */
663                         if (req_pio == 8 || req_pio == 9) {
664                                 unsigned long flags;
665
666                                 spin_lock_irqsave(&ide_lock, flags);
667                                 port_ops->set_pio_mode(drive, req_pio);
668                                 spin_unlock_irqrestore(&ide_lock, flags);
669                         } else
670                                 port_ops->set_pio_mode(drive, req_pio);
671                 } else {
672                         int keep_dma = drive->using_dma;
673
674                         ide_set_pio(drive, req_pio);
675
676                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
677                                 if (keep_dma)
678                                         ide_dma_on(drive);
679                         }
680                 }
681
682                 return ide_stopped;
683         } else {
684                 if (drive->media == ide_disk)
685                         return ide_disk_special(drive);
686
687                 s->all = 0;
688                 drive->mult_req = 0;
689                 return ide_stopped;
690         }
691 }
692
693 void ide_map_sg(ide_drive_t *drive, struct request *rq)
694 {
695         ide_hwif_t *hwif = drive->hwif;
696         struct scatterlist *sg = hwif->sg_table;
697
698         if (hwif->sg_mapped)    /* needed by ide-scsi */
699                 return;
700
701         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
702                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
703         } else {
704                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
705                 hwif->sg_nents = 1;
706         }
707 }
708
709 EXPORT_SYMBOL_GPL(ide_map_sg);
710
711 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
712 {
713         ide_hwif_t *hwif = drive->hwif;
714
715         hwif->nsect = hwif->nleft = rq->nr_sectors;
716         hwif->cursg_ofs = 0;
717         hwif->cursg = NULL;
718 }
719
720 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
721
722 /**
723  *      execute_drive_command   -       issue special drive command
724  *      @drive: the drive to issue the command on
725  *      @rq: the request structure holding the command
726  *
727  *      execute_drive_cmd() issues a special drive command,  usually 
728  *      initiated by ioctl() from the external hdparm program. The
729  *      command can be a drive command, drive task or taskfile 
730  *      operation. Weirdly you can call it with NULL to wait for
731  *      all commands to finish. Don't do this as that is due to change
732  */
733
734 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
735                 struct request *rq)
736 {
737         ide_hwif_t *hwif = HWIF(drive);
738         ide_task_t *task = rq->special;
739
740         if (task) {
741                 hwif->data_phase = task->data_phase;
742
743                 switch (hwif->data_phase) {
744                 case TASKFILE_MULTI_OUT:
745                 case TASKFILE_OUT:
746                 case TASKFILE_MULTI_IN:
747                 case TASKFILE_IN:
748                         ide_init_sg_cmd(drive, rq);
749                         ide_map_sg(drive, rq);
750                 default:
751                         break;
752                 }
753
754                 return do_rw_taskfile(drive, task);
755         }
756
757         /*
758          * NULL is actually a valid way of waiting for
759          * all current requests to be flushed from the queue.
760          */
761 #ifdef DEBUG
762         printk("%s: DRIVE_CMD (null)\n", drive->name);
763 #endif
764         ide_end_drive_cmd(drive, ide_read_status(drive), ide_read_error(drive));
765
766         return ide_stopped;
767 }
768
769 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
770 {
771         struct request_pm_state *pm = rq->data;
772
773         if (blk_pm_suspend_request(rq) &&
774             pm->pm_step == ide_pm_state_start_suspend)
775                 /* Mark drive blocked when starting the suspend sequence. */
776                 drive->blocked = 1;
777         else if (blk_pm_resume_request(rq) &&
778                  pm->pm_step == ide_pm_state_start_resume) {
779                 /* 
780                  * The first thing we do on wakeup is to wait for BSY bit to
781                  * go away (with a looong timeout) as a drive on this hwif may
782                  * just be POSTing itself.
783                  * We do that before even selecting as the "other" device on
784                  * the bus may be broken enough to walk on our toes at this
785                  * point.
786                  */
787                 int rc;
788 #ifdef DEBUG_PM
789                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
790 #endif
791                 rc = ide_wait_not_busy(HWIF(drive), 35000);
792                 if (rc)
793                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
794                 SELECT_DRIVE(drive);
795                 ide_set_irq(drive, 1);
796                 rc = ide_wait_not_busy(HWIF(drive), 100000);
797                 if (rc)
798                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
799         }
800 }
801
802 /**
803  *      start_request   -       start of I/O and command issuing for IDE
804  *
805  *      start_request() initiates handling of a new I/O request. It
806  *      accepts commands and I/O (read/write) requests. It also does
807  *      the final remapping for weird stuff like EZDrive. Once 
808  *      device mapper can work sector level the EZDrive stuff can go away
809  *
810  *      FIXME: this function needs a rename
811  */
812  
813 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
814 {
815         ide_startstop_t startstop;
816         sector_t block;
817
818         BUG_ON(!blk_rq_started(rq));
819
820 #ifdef DEBUG
821         printk("%s: start_request: current=0x%08lx\n",
822                 HWIF(drive)->name, (unsigned long) rq);
823 #endif
824
825         /* bail early if we've exceeded max_failures */
826         if (drive->max_failures && (drive->failures > drive->max_failures)) {
827                 rq->cmd_flags |= REQ_FAILED;
828                 goto kill_rq;
829         }
830
831         block    = rq->sector;
832         if (blk_fs_request(rq) &&
833             (drive->media == ide_disk || drive->media == ide_floppy)) {
834                 block += drive->sect0;
835         }
836         /* Yecch - this will shift the entire interval,
837            possibly killing some innocent following sector */
838         if (block == 0 && drive->remap_0_to_1 == 1)
839                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
840
841         if (blk_pm_request(rq))
842                 ide_check_pm_state(drive, rq);
843
844         SELECT_DRIVE(drive);
845         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
846                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
847                 return startstop;
848         }
849         if (!drive->special.all) {
850                 ide_driver_t *drv;
851
852                 /*
853                  * We reset the drive so we need to issue a SETFEATURES.
854                  * Do it _after_ do_special() restored device parameters.
855                  */
856                 if (drive->current_speed == 0xff)
857                         ide_config_drive_speed(drive, drive->desired_speed);
858
859                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
860                         return execute_drive_cmd(drive, rq);
861                 else if (blk_pm_request(rq)) {
862                         struct request_pm_state *pm = rq->data;
863 #ifdef DEBUG_PM
864                         printk("%s: start_power_step(step: %d)\n",
865                                 drive->name, rq->pm->pm_step);
866 #endif
867                         startstop = ide_start_power_step(drive, rq);
868                         if (startstop == ide_stopped &&
869                             pm->pm_step == ide_pm_state_completed)
870                                 ide_complete_pm_request(drive, rq);
871                         return startstop;
872                 }
873
874                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
875                 return drv->do_request(drive, rq, block);
876         }
877         return do_special(drive);
878 kill_rq:
879         ide_kill_rq(drive, rq);
880         return ide_stopped;
881 }
882
883 /**
884  *      ide_stall_queue         -       pause an IDE device
885  *      @drive: drive to stall
886  *      @timeout: time to stall for (jiffies)
887  *
888  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
889  *      to the hwgroup by sleeping for timeout jiffies.
890  */
891  
892 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
893 {
894         if (timeout > WAIT_WORSTCASE)
895                 timeout = WAIT_WORSTCASE;
896         drive->sleep = timeout + jiffies;
897         drive->sleeping = 1;
898 }
899
900 EXPORT_SYMBOL(ide_stall_queue);
901
902 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
903
904 /**
905  *      choose_drive            -       select a drive to service
906  *      @hwgroup: hardware group to select on
907  *
908  *      choose_drive() selects the next drive which will be serviced.
909  *      This is necessary because the IDE layer can't issue commands
910  *      to both drives on the same cable, unlike SCSI.
911  */
912  
913 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
914 {
915         ide_drive_t *drive, *best;
916
917 repeat: 
918         best = NULL;
919         drive = hwgroup->drive;
920
921         /*
922          * drive is doing pre-flush, ordered write, post-flush sequence. even
923          * though that is 3 requests, it must be seen as a single transaction.
924          * we must not preempt this drive until that is complete
925          */
926         if (blk_queue_flushing(drive->queue)) {
927                 /*
928                  * small race where queue could get replugged during
929                  * the 3-request flush cycle, just yank the plug since
930                  * we want it to finish asap
931                  */
932                 blk_remove_plug(drive->queue);
933                 return drive;
934         }
935
936         do {
937                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
938                     && !elv_queue_empty(drive->queue)) {
939                         if (!best
940                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
941                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
942                         {
943                                 if (!blk_queue_plugged(drive->queue))
944                                         best = drive;
945                         }
946                 }
947         } while ((drive = drive->next) != hwgroup->drive);
948         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
949                 long t = (signed long)(WAKEUP(best) - jiffies);
950                 if (t >= WAIT_MIN_SLEEP) {
951                 /*
952                  * We *may* have some time to spare, but first let's see if
953                  * someone can potentially benefit from our nice mood today..
954                  */
955                         drive = best->next;
956                         do {
957                                 if (!drive->sleeping
958                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
959                                  && time_before(WAKEUP(drive), jiffies + t))
960                                 {
961                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
962                                         goto repeat;
963                                 }
964                         } while ((drive = drive->next) != best);
965                 }
966         }
967         return best;
968 }
969
970 /*
971  * Issue a new request to a drive from hwgroup
972  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
973  *
974  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
975  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
976  * may have both interfaces in a single hwgroup to "serialize" access.
977  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
978  * together into one hwgroup for serialized access.
979  *
980  * Note also that several hwgroups can end up sharing a single IRQ,
981  * possibly along with many other devices.  This is especially common in
982  * PCI-based systems with off-board IDE controller cards.
983  *
984  * The IDE driver uses the single global ide_lock spinlock to protect
985  * access to the request queues, and to protect the hwgroup->busy flag.
986  *
987  * The first thread into the driver for a particular hwgroup sets the
988  * hwgroup->busy flag to indicate that this hwgroup is now active,
989  * and then initiates processing of the top request from the request queue.
990  *
991  * Other threads attempting entry notice the busy setting, and will simply
992  * queue their new requests and exit immediately.  Note that hwgroup->busy
993  * remains set even when the driver is merely awaiting the next interrupt.
994  * Thus, the meaning is "this hwgroup is busy processing a request".
995  *
996  * When processing of a request completes, the completing thread or IRQ-handler
997  * will start the next request from the queue.  If no more work remains,
998  * the driver will clear the hwgroup->busy flag and exit.
999  *
1000  * The ide_lock (spinlock) is used to protect all access to the
1001  * hwgroup->busy flag, but is otherwise not needed for most processing in
1002  * the driver.  This makes the driver much more friendlier to shared IRQs
1003  * than previous designs, while remaining 100% (?) SMP safe and capable.
1004  */
1005 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1006 {
1007         ide_drive_t     *drive;
1008         ide_hwif_t      *hwif;
1009         struct request  *rq;
1010         ide_startstop_t startstop;
1011         int             loops = 0;
1012
1013         /* for atari only: POSSIBLY BROKEN HERE(?) */
1014         ide_get_lock(ide_intr, hwgroup);
1015
1016         /* caller must own ide_lock */
1017         BUG_ON(!irqs_disabled());
1018
1019         while (!hwgroup->busy) {
1020                 hwgroup->busy = 1;
1021                 drive = choose_drive(hwgroup);
1022                 if (drive == NULL) {
1023                         int sleeping = 0;
1024                         unsigned long sleep = 0; /* shut up, gcc */
1025                         hwgroup->rq = NULL;
1026                         drive = hwgroup->drive;
1027                         do {
1028                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1029                                         sleeping = 1;
1030                                         sleep = drive->sleep;
1031                                 }
1032                         } while ((drive = drive->next) != hwgroup->drive);
1033                         if (sleeping) {
1034                 /*
1035                  * Take a short snooze, and then wake up this hwgroup again.
1036                  * This gives other hwgroups on the same a chance to
1037                  * play fairly with us, just in case there are big differences
1038                  * in relative throughputs.. don't want to hog the cpu too much.
1039                  */
1040                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1041                                         sleep = jiffies + WAIT_MIN_SLEEP;
1042 #if 1
1043                                 if (timer_pending(&hwgroup->timer))
1044                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1045 #endif
1046                                 /* so that ide_timer_expiry knows what to do */
1047                                 hwgroup->sleeping = 1;
1048                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1049                                 mod_timer(&hwgroup->timer, sleep);
1050                                 /* we purposely leave hwgroup->busy==1
1051                                  * while sleeping */
1052                         } else {
1053                                 /* Ugly, but how can we sleep for the lock
1054                                  * otherwise? perhaps from tq_disk?
1055                                  */
1056
1057                                 /* for atari only */
1058                                 ide_release_lock();
1059                                 hwgroup->busy = 0;
1060                         }
1061
1062                         /* no more work for this hwgroup (for now) */
1063                         return;
1064                 }
1065         again:
1066                 hwif = HWIF(drive);
1067                 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1068                         /*
1069                          * set nIEN for previous hwif, drives in the
1070                          * quirk_list may not like intr setups/cleanups
1071                          */
1072                         if (drive->quirk_list != 1)
1073                                 ide_set_irq(drive, 0);
1074                 }
1075                 hwgroup->hwif = hwif;
1076                 hwgroup->drive = drive;
1077                 drive->sleeping = 0;
1078                 drive->service_start = jiffies;
1079
1080                 if (blk_queue_plugged(drive->queue)) {
1081                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1082                         break;
1083                 }
1084
1085                 /*
1086                  * we know that the queue isn't empty, but this can happen
1087                  * if the q->prep_rq_fn() decides to kill a request
1088                  */
1089                 rq = elv_next_request(drive->queue);
1090                 if (!rq) {
1091                         hwgroup->busy = 0;
1092                         break;
1093                 }
1094
1095                 /*
1096                  * Sanity: don't accept a request that isn't a PM request
1097                  * if we are currently power managed. This is very important as
1098                  * blk_stop_queue() doesn't prevent the elv_next_request()
1099                  * above to return us whatever is in the queue. Since we call
1100                  * ide_do_request() ourselves, we end up taking requests while
1101                  * the queue is blocked...
1102                  * 
1103                  * We let requests forced at head of queue with ide-preempt
1104                  * though. I hope that doesn't happen too much, hopefully not
1105                  * unless the subdriver triggers such a thing in its own PM
1106                  * state machine.
1107                  *
1108                  * We count how many times we loop here to make sure we service
1109                  * all drives in the hwgroup without looping for ever
1110                  */
1111                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1112                         drive = drive->next ? drive->next : hwgroup->drive;
1113                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1114                                 goto again;
1115                         /* We clear busy, there should be no pending ATA command at this point. */
1116                         hwgroup->busy = 0;
1117                         break;
1118                 }
1119
1120                 hwgroup->rq = rq;
1121
1122                 /*
1123                  * Some systems have trouble with IDE IRQs arriving while
1124                  * the driver is still setting things up.  So, here we disable
1125                  * the IRQ used by this interface while the request is being started.
1126                  * This may look bad at first, but pretty much the same thing
1127                  * happens anyway when any interrupt comes in, IDE or otherwise
1128                  *  -- the kernel masks the IRQ while it is being handled.
1129                  */
1130                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1131                         disable_irq_nosync(hwif->irq);
1132                 spin_unlock(&ide_lock);
1133                 local_irq_enable_in_hardirq();
1134                         /* allow other IRQs while we start this request */
1135                 startstop = start_request(drive, rq);
1136                 spin_lock_irq(&ide_lock);
1137                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1138                         enable_irq(hwif->irq);
1139                 if (startstop == ide_stopped)
1140                         hwgroup->busy = 0;
1141         }
1142 }
1143
1144 /*
1145  * Passes the stuff to ide_do_request
1146  */
1147 void do_ide_request(struct request_queue *q)
1148 {
1149         ide_drive_t *drive = q->queuedata;
1150
1151         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1152 }
1153
1154 /*
1155  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1156  * retry the current request in pio mode instead of risking tossing it
1157  * all away
1158  */
1159 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1160 {
1161         ide_hwif_t *hwif = HWIF(drive);
1162         struct request *rq;
1163         ide_startstop_t ret = ide_stopped;
1164
1165         /*
1166          * end current dma transaction
1167          */
1168
1169         if (error < 0) {
1170                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1171                 (void)hwif->dma_ops->dma_end(drive);
1172                 ret = ide_error(drive, "dma timeout error",
1173                                 ide_read_status(drive));
1174         } else {
1175                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1176                 hwif->dma_ops->dma_timeout(drive);
1177         }
1178
1179         /*
1180          * disable dma for now, but remember that we did so because of
1181          * a timeout -- we'll reenable after we finish this next request
1182          * (or rather the first chunk of it) in pio.
1183          */
1184         drive->retry_pio++;
1185         drive->state = DMA_PIO_RETRY;
1186         ide_dma_off_quietly(drive);
1187
1188         /*
1189          * un-busy drive etc (hwgroup->busy is cleared on return) and
1190          * make sure request is sane
1191          */
1192         rq = HWGROUP(drive)->rq;
1193
1194         if (!rq)
1195                 goto out;
1196
1197         HWGROUP(drive)->rq = NULL;
1198
1199         rq->errors = 0;
1200
1201         if (!rq->bio)
1202                 goto out;
1203
1204         rq->sector = rq->bio->bi_sector;
1205         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1206         rq->hard_cur_sectors = rq->current_nr_sectors;
1207         rq->buffer = bio_data(rq->bio);
1208 out:
1209         return ret;
1210 }
1211
1212 /**
1213  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1214  *      @data: timer callback magic (hwgroup)
1215  *
1216  *      An IDE command has timed out before the expected drive return
1217  *      occurred. At this point we attempt to clean up the current
1218  *      mess. If the current handler includes an expiry handler then
1219  *      we invoke the expiry handler, and providing it is happy the
1220  *      work is done. If that fails we apply generic recovery rules
1221  *      invoking the handler and checking the drive DMA status. We
1222  *      have an excessively incestuous relationship with the DMA
1223  *      logic that wants cleaning up.
1224  */
1225  
1226 void ide_timer_expiry (unsigned long data)
1227 {
1228         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1229         ide_handler_t   *handler;
1230         ide_expiry_t    *expiry;
1231         unsigned long   flags;
1232         unsigned long   wait = -1;
1233
1234         spin_lock_irqsave(&ide_lock, flags);
1235
1236         if (((handler = hwgroup->handler) == NULL) ||
1237             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1238                 /*
1239                  * Either a marginal timeout occurred
1240                  * (got the interrupt just as timer expired),
1241                  * or we were "sleeping" to give other devices a chance.
1242                  * Either way, we don't really want to complain about anything.
1243                  */
1244                 if (hwgroup->sleeping) {
1245                         hwgroup->sleeping = 0;
1246                         hwgroup->busy = 0;
1247                 }
1248         } else {
1249                 ide_drive_t *drive = hwgroup->drive;
1250                 if (!drive) {
1251                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1252                         hwgroup->handler = NULL;
1253                 } else {
1254                         ide_hwif_t *hwif;
1255                         ide_startstop_t startstop = ide_stopped;
1256                         if (!hwgroup->busy) {
1257                                 hwgroup->busy = 1;      /* paranoia */
1258                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1259                         }
1260                         if ((expiry = hwgroup->expiry) != NULL) {
1261                                 /* continue */
1262                                 if ((wait = expiry(drive)) > 0) {
1263                                         /* reset timer */
1264                                         hwgroup->timer.expires  = jiffies + wait;
1265                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1266                                         add_timer(&hwgroup->timer);
1267                                         spin_unlock_irqrestore(&ide_lock, flags);
1268                                         return;
1269                                 }
1270                         }
1271                         hwgroup->handler = NULL;
1272                         /*
1273                          * We need to simulate a real interrupt when invoking
1274                          * the handler() function, which means we need to
1275                          * globally mask the specific IRQ:
1276                          */
1277                         spin_unlock(&ide_lock);
1278                         hwif  = HWIF(drive);
1279                         /* disable_irq_nosync ?? */
1280                         disable_irq(hwif->irq);
1281                         /* local CPU only,
1282                          * as if we were handling an interrupt */
1283                         local_irq_disable();
1284                         if (hwgroup->polling) {
1285                                 startstop = handler(drive);
1286                         } else if (drive_is_ready(drive)) {
1287                                 if (drive->waiting_for_dma)
1288                                         hwif->dma_ops->dma_lost_irq(drive);
1289                                 (void)ide_ack_intr(hwif);
1290                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1291                                 startstop = handler(drive);
1292                         } else {
1293                                 if (drive->waiting_for_dma) {
1294                                         startstop = ide_dma_timeout_retry(drive, wait);
1295                                 } else
1296                                         startstop =
1297                                         ide_error(drive, "irq timeout",
1298                                                   ide_read_status(drive));
1299                         }
1300                         drive->service_time = jiffies - drive->service_start;
1301                         spin_lock_irq(&ide_lock);
1302                         enable_irq(hwif->irq);
1303                         if (startstop == ide_stopped)
1304                                 hwgroup->busy = 0;
1305                 }
1306         }
1307         ide_do_request(hwgroup, IDE_NO_IRQ);
1308         spin_unlock_irqrestore(&ide_lock, flags);
1309 }
1310
1311 /**
1312  *      unexpected_intr         -       handle an unexpected IDE interrupt
1313  *      @irq: interrupt line
1314  *      @hwgroup: hwgroup being processed
1315  *
1316  *      There's nothing really useful we can do with an unexpected interrupt,
1317  *      other than reading the status register (to clear it), and logging it.
1318  *      There should be no way that an irq can happen before we're ready for it,
1319  *      so we needn't worry much about losing an "important" interrupt here.
1320  *
1321  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1322  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1323  *      looks "good", we just ignore the interrupt completely.
1324  *
1325  *      This routine assumes __cli() is in effect when called.
1326  *
1327  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1328  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1329  *      we could screw up by interfering with a new request being set up for 
1330  *      irq15.
1331  *
1332  *      In reality, this is a non-issue.  The new command is not sent unless 
1333  *      the drive is ready to accept one, in which case we know the drive is
1334  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1335  *      before completing the issuance of any new drive command, so we will not
1336  *      be accidentally invoked as a result of any valid command completion
1337  *      interrupt.
1338  *
1339  *      Note that we must walk the entire hwgroup here. We know which hwif
1340  *      is doing the current command, but we don't know which hwif burped
1341  *      mysteriously.
1342  */
1343  
1344 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1345 {
1346         u8 stat;
1347         ide_hwif_t *hwif = hwgroup->hwif;
1348
1349         /*
1350          * handle the unexpected interrupt
1351          */
1352         do {
1353                 if (hwif->irq == irq) {
1354                         stat = hwif->INB(hwif->io_ports.status_addr);
1355                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1356                                 /* Try to not flood the console with msgs */
1357                                 static unsigned long last_msgtime, count;
1358                                 ++count;
1359                                 if (time_after(jiffies, last_msgtime + HZ)) {
1360                                         last_msgtime = jiffies;
1361                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1362                                                 "status=0x%02x, count=%ld\n",
1363                                                 hwif->name,
1364                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1365                                 }
1366                         }
1367                 }
1368         } while ((hwif = hwif->next) != hwgroup->hwif);
1369 }
1370
1371 /**
1372  *      ide_intr        -       default IDE interrupt handler
1373  *      @irq: interrupt number
1374  *      @dev_id: hwif group
1375  *      @regs: unused weirdness from the kernel irq layer
1376  *
1377  *      This is the default IRQ handler for the IDE layer. You should
1378  *      not need to override it. If you do be aware it is subtle in
1379  *      places
1380  *
1381  *      hwgroup->hwif is the interface in the group currently performing
1382  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1383  *      the IRQ handler to call. As we issue a command the handlers
1384  *      step through multiple states, reassigning the handler to the
1385  *      next step in the process. Unlike a smart SCSI controller IDE
1386  *      expects the main processor to sequence the various transfer
1387  *      stages. We also manage a poll timer to catch up with most
1388  *      timeout situations. There are still a few where the handlers
1389  *      don't ever decide to give up.
1390  *
1391  *      The handler eventually returns ide_stopped to indicate the
1392  *      request completed. At this point we issue the next request
1393  *      on the hwgroup and the process begins again.
1394  */
1395  
1396 irqreturn_t ide_intr (int irq, void *dev_id)
1397 {
1398         unsigned long flags;
1399         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1400         ide_hwif_t *hwif;
1401         ide_drive_t *drive;
1402         ide_handler_t *handler;
1403         ide_startstop_t startstop;
1404
1405         spin_lock_irqsave(&ide_lock, flags);
1406         hwif = hwgroup->hwif;
1407
1408         if (!ide_ack_intr(hwif)) {
1409                 spin_unlock_irqrestore(&ide_lock, flags);
1410                 return IRQ_NONE;
1411         }
1412
1413         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1414                 /*
1415                  * Not expecting an interrupt from this drive.
1416                  * That means this could be:
1417                  *      (1) an interrupt from another PCI device
1418                  *      sharing the same PCI INT# as us.
1419                  * or   (2) a drive just entered sleep or standby mode,
1420                  *      and is interrupting to let us know.
1421                  * or   (3) a spurious interrupt of unknown origin.
1422                  *
1423                  * For PCI, we cannot tell the difference,
1424                  * so in that case we just ignore it and hope it goes away.
1425                  *
1426                  * FIXME: unexpected_intr should be hwif-> then we can
1427                  * remove all the ifdef PCI crap
1428                  */
1429 #ifdef CONFIG_BLK_DEV_IDEPCI
1430                 if (hwif->chipset != ide_pci)
1431 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1432                 {
1433                         /*
1434                          * Probably not a shared PCI interrupt,
1435                          * so we can safely try to do something about it:
1436                          */
1437                         unexpected_intr(irq, hwgroup);
1438 #ifdef CONFIG_BLK_DEV_IDEPCI
1439                 } else {
1440                         /*
1441                          * Whack the status register, just in case
1442                          * we have a leftover pending IRQ.
1443                          */
1444                         (void) hwif->INB(hwif->io_ports.status_addr);
1445 #endif /* CONFIG_BLK_DEV_IDEPCI */
1446                 }
1447                 spin_unlock_irqrestore(&ide_lock, flags);
1448                 return IRQ_NONE;
1449         }
1450         drive = hwgroup->drive;
1451         if (!drive) {
1452                 /*
1453                  * This should NEVER happen, and there isn't much
1454                  * we could do about it here.
1455                  *
1456                  * [Note - this can occur if the drive is hot unplugged]
1457                  */
1458                 spin_unlock_irqrestore(&ide_lock, flags);
1459                 return IRQ_HANDLED;
1460         }
1461         if (!drive_is_ready(drive)) {
1462                 /*
1463                  * This happens regularly when we share a PCI IRQ with
1464                  * another device.  Unfortunately, it can also happen
1465                  * with some buggy drives that trigger the IRQ before
1466                  * their status register is up to date.  Hopefully we have
1467                  * enough advance overhead that the latter isn't a problem.
1468                  */
1469                 spin_unlock_irqrestore(&ide_lock, flags);
1470                 return IRQ_NONE;
1471         }
1472         if (!hwgroup->busy) {
1473                 hwgroup->busy = 1;      /* paranoia */
1474                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1475         }
1476         hwgroup->handler = NULL;
1477         hwgroup->req_gen++;
1478         del_timer(&hwgroup->timer);
1479         spin_unlock(&ide_lock);
1480
1481         /* Some controllers might set DMA INTR no matter DMA or PIO;
1482          * bmdma status might need to be cleared even for
1483          * PIO interrupts to prevent spurious/lost irq.
1484          */
1485         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1486                 /* ide_dma_end() needs bmdma status for error checking.
1487                  * So, skip clearing bmdma status here and leave it
1488                  * to ide_dma_end() if this is dma interrupt.
1489                  */
1490                 hwif->ide_dma_clear_irq(drive);
1491
1492         if (drive->unmask)
1493                 local_irq_enable_in_hardirq();
1494         /* service this interrupt, may set handler for next interrupt */
1495         startstop = handler(drive);
1496         spin_lock_irq(&ide_lock);
1497
1498         /*
1499          * Note that handler() may have set things up for another
1500          * interrupt to occur soon, but it cannot happen until
1501          * we exit from this routine, because it will be the
1502          * same irq as is currently being serviced here, and Linux
1503          * won't allow another of the same (on any CPU) until we return.
1504          */
1505         drive->service_time = jiffies - drive->service_start;
1506         if (startstop == ide_stopped) {
1507                 if (hwgroup->handler == NULL) { /* paranoia */
1508                         hwgroup->busy = 0;
1509                         ide_do_request(hwgroup, hwif->irq);
1510                 } else {
1511                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1512                                 "on exit\n", drive->name);
1513                 }
1514         }
1515         spin_unlock_irqrestore(&ide_lock, flags);
1516         return IRQ_HANDLED;
1517 }
1518
1519 /**
1520  *      ide_do_drive_cmd        -       issue IDE special command
1521  *      @drive: device to issue command
1522  *      @rq: request to issue
1523  *
1524  *      This function issues a special IDE device request
1525  *      onto the request queue.
1526  *
1527  *      the rq is queued at the head of the request queue, displacing
1528  *      the currently-being-processed request and this function
1529  *      returns immediately without waiting for the new rq to be
1530  *      completed.  This is VERY DANGEROUS, and is intended for
1531  *      careful use by the ATAPI tape/cdrom driver code.
1532  */
1533
1534 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1535 {
1536         unsigned long flags;
1537         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1538
1539         spin_lock_irqsave(&ide_lock, flags);
1540         hwgroup->rq = NULL;
1541         __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1542         __generic_unplug_device(drive->queue);
1543         spin_unlock_irqrestore(&ide_lock, flags);
1544 }
1545
1546 EXPORT_SYMBOL(ide_do_drive_cmd);
1547
1548 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1549 {
1550         ide_task_t task;
1551
1552         memset(&task, 0, sizeof(task));
1553         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1554                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1555         task.tf.feature = dma;          /* Use PIO/DMA */
1556         task.tf.lbam    = bcount & 0xff;
1557         task.tf.lbah    = (bcount >> 8) & 0xff;
1558
1559         ide_tf_dump(drive->name, &task.tf);
1560         ide_set_irq(drive, 1);
1561         SELECT_MASK(drive, 0);
1562         drive->hwif->tf_load(drive, &task);
1563 }
1564
1565 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1566
1567 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1568 {
1569         ide_hwif_t *hwif = drive->hwif;
1570         u8 buf[4] = { 0 };
1571
1572         while (len > 0) {
1573                 if (write)
1574                         hwif->output_data(drive, NULL, buf, min(4, len));
1575                 else
1576                         hwif->input_data(drive, NULL, buf, min(4, len));
1577                 len -= 4;
1578         }
1579 }
1580 EXPORT_SYMBOL_GPL(ide_pad_transfer);