Merge branches 'timers/clocksource', 'timers/hpet', 'timers/hrtimers', 'timers/nohz...
[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/hdreg.h>
44 #include <linux/completion.h>
45 #include <linux/reboot.h>
46 #include <linux/cdrom.h>
47 #include <linux/seq_file.h>
48 #include <linux/device.h>
49 #include <linux/kmod.h>
50 #include <linux/scatterlist.h>
51 #include <linux/bitops.h>
52
53 #include <asm/byteorder.h>
54 #include <asm/irq.h>
55 #include <asm/uaccess.h>
56 #include <asm/io.h>
57
58 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
59                              int uptodate, unsigned int nr_bytes, int dequeue)
60 {
61         int ret = 1;
62         int error = 0;
63
64         if (uptodate <= 0)
65                 error = uptodate ? uptodate : -EIO;
66
67         /*
68          * if failfast is set on a request, override number of sectors and
69          * complete the whole request right now
70          */
71         if (blk_noretry_request(rq) && error)
72                 nr_bytes = rq->hard_nr_sectors << 9;
73
74         if (!blk_fs_request(rq) && error && !rq->errors)
75                 rq->errors = -EIO;
76
77         /*
78          * decide whether to reenable DMA -- 3 is a random magic for now,
79          * if we DMA timeout more than 3 times, just stay in PIO
80          */
81         if ((drive->dev_flags & IDE_DFLAG_DMA_PIO_RETRY) &&
82             drive->retry_pio <= 3) {
83                 drive->dev_flags &= ~IDE_DFLAG_DMA_PIO_RETRY;
84                 ide_dma_on(drive);
85         }
86
87         if (!__blk_end_request(rq, error, nr_bytes)) {
88                 if (dequeue)
89                         HWGROUP(drive)->rq = NULL;
90                 ret = 0;
91         }
92
93         return ret;
94 }
95
96 /**
97  *      ide_end_request         -       complete an IDE I/O
98  *      @drive: IDE device for the I/O
99  *      @uptodate:
100  *      @nr_sectors: number of sectors completed
101  *
102  *      This is our end_request wrapper function. We complete the I/O
103  *      update random number input and dequeue the request, which if
104  *      it was tagged may be out of order.
105  */
106
107 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
108 {
109         unsigned int nr_bytes = nr_sectors << 9;
110         struct request *rq;
111         unsigned long flags;
112         int ret = 1;
113
114         /*
115          * room for locking improvements here, the calls below don't
116          * need the queue lock held at all
117          */
118         spin_lock_irqsave(&ide_lock, flags);
119         rq = HWGROUP(drive)->rq;
120
121         if (!nr_bytes) {
122                 if (blk_pc_request(rq))
123                         nr_bytes = rq->data_len;
124                 else
125                         nr_bytes = rq->hard_cur_sectors << 9;
126         }
127
128         ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
129
130         spin_unlock_irqrestore(&ide_lock, flags);
131         return ret;
132 }
133 EXPORT_SYMBOL(ide_end_request);
134
135 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq)
136 {
137         struct request_pm_state *pm = rq->data;
138
139 #ifdef DEBUG_PM
140         printk(KERN_INFO "%s: complete_power_step(step: %d)\n",
141                 drive->name, pm->pm_step);
142 #endif
143         if (drive->media != ide_disk)
144                 return;
145
146         switch (pm->pm_step) {
147         case IDE_PM_FLUSH_CACHE:        /* Suspend step 1 (flush cache) */
148                 if (pm->pm_state == PM_EVENT_FREEZE)
149                         pm->pm_step = IDE_PM_COMPLETED;
150                 else
151                         pm->pm_step = IDE_PM_STANDBY;
152                 break;
153         case IDE_PM_STANDBY:            /* Suspend step 2 (standby) */
154                 pm->pm_step = IDE_PM_COMPLETED;
155                 break;
156         case IDE_PM_RESTORE_PIO:        /* Resume step 1 (restore PIO) */
157                 pm->pm_step = IDE_PM_IDLE;
158                 break;
159         case IDE_PM_IDLE:               /* Resume step 2 (idle)*/
160                 pm->pm_step = IDE_PM_RESTORE_DMA;
161                 break;
162         }
163 }
164
165 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
166 {
167         struct request_pm_state *pm = rq->data;
168         ide_task_t *args = rq->special;
169
170         memset(args, 0, sizeof(*args));
171
172         switch (pm->pm_step) {
173         case IDE_PM_FLUSH_CACHE:        /* Suspend step 1 (flush cache) */
174                 if (drive->media != ide_disk)
175                         break;
176                 /* Not supported? Switch to next step now. */
177                 if (ata_id_flush_enabled(drive->id) == 0 ||
178                     (drive->dev_flags & IDE_DFLAG_WCACHE) == 0) {
179                         ide_complete_power_step(drive, rq);
180                         return ide_stopped;
181                 }
182                 if (ata_id_flush_ext_enabled(drive->id))
183                         args->tf.command = ATA_CMD_FLUSH_EXT;
184                 else
185                         args->tf.command = ATA_CMD_FLUSH;
186                 goto out_do_tf;
187         case IDE_PM_STANDBY:            /* Suspend step 2 (standby) */
188                 args->tf.command = ATA_CMD_STANDBYNOW1;
189                 goto out_do_tf;
190         case IDE_PM_RESTORE_PIO:        /* Resume step 1 (restore PIO) */
191                 ide_set_max_pio(drive);
192                 /*
193                  * skip IDE_PM_IDLE for ATAPI devices
194                  */
195                 if (drive->media != ide_disk)
196                         pm->pm_step = IDE_PM_RESTORE_DMA;
197                 else
198                         ide_complete_power_step(drive, rq);
199                 return ide_stopped;
200         case IDE_PM_IDLE:               /* Resume step 2 (idle) */
201                 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
202                 goto out_do_tf;
203         case IDE_PM_RESTORE_DMA:        /* Resume step 3 (restore DMA) */
204                 /*
205                  * Right now, all we do is call ide_set_dma(drive),
206                  * we could be smarter and check for current xfer_speed
207                  * in struct drive etc...
208                  */
209                 if (drive->hwif->dma_ops == NULL)
210                         break;
211                 /*
212                  * TODO: respect IDE_DFLAG_USING_DMA
213                  */
214                 ide_set_dma(drive);
215                 break;
216         }
217
218         pm->pm_step = IDE_PM_COMPLETED;
219         return ide_stopped;
220
221 out_do_tf:
222         args->tf_flags   = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
223         args->data_phase = TASKFILE_NO_DATA;
224         return do_rw_taskfile(drive, args);
225 }
226
227 /**
228  *      ide_end_dequeued_request        -       complete an IDE I/O
229  *      @drive: IDE device for the I/O
230  *      @uptodate:
231  *      @nr_sectors: number of sectors completed
232  *
233  *      Complete an I/O that is no longer on the request queue. This
234  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
235  *      We must still finish the old request but we must not tamper with the
236  *      queue in the meantime.
237  *
238  *      NOTE: This path does not handle barrier, but barrier is not supported
239  *      on ide-cd anyway.
240  */
241
242 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
243                              int uptodate, int nr_sectors)
244 {
245         unsigned long flags;
246         int ret;
247
248         spin_lock_irqsave(&ide_lock, flags);
249         BUG_ON(!blk_rq_started(rq));
250         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
251         spin_unlock_irqrestore(&ide_lock, flags);
252
253         return ret;
254 }
255 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
256
257
258 /**
259  *      ide_complete_pm_request - end the current Power Management request
260  *      @drive: target drive
261  *      @rq: request
262  *
263  *      This function cleans up the current PM request and stops the queue
264  *      if necessary.
265  */
266 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
267 {
268         unsigned long flags;
269
270 #ifdef DEBUG_PM
271         printk("%s: completing PM request, %s\n", drive->name,
272                blk_pm_suspend_request(rq) ? "suspend" : "resume");
273 #endif
274         spin_lock_irqsave(&ide_lock, flags);
275         if (blk_pm_suspend_request(rq)) {
276                 blk_stop_queue(drive->queue);
277         } else {
278                 drive->dev_flags &= ~IDE_DFLAG_BLOCKED;
279                 blk_start_queue(drive->queue);
280         }
281         HWGROUP(drive)->rq = NULL;
282         if (__blk_end_request(rq, 0, 0))
283                 BUG();
284         spin_unlock_irqrestore(&ide_lock, flags);
285 }
286
287 /**
288  *      ide_end_drive_cmd       -       end an explicit drive command
289  *      @drive: command 
290  *      @stat: status bits
291  *      @err: error bits
292  *
293  *      Clean up after success/failure of an explicit drive command.
294  *      These get thrown onto the queue so they are synchronized with
295  *      real I/O operations on the drive.
296  *
297  *      In LBA48 mode we have to read the register set twice to get
298  *      all the extra information out.
299  */
300  
301 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
302 {
303         unsigned long flags;
304         struct request *rq;
305
306         spin_lock_irqsave(&ide_lock, flags);
307         rq = HWGROUP(drive)->rq;
308         spin_unlock_irqrestore(&ide_lock, flags);
309
310         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
311                 ide_task_t *task = (ide_task_t *)rq->special;
312
313                 if (rq->errors == 0)
314                         rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
315
316                 if (task) {
317                         struct ide_taskfile *tf = &task->tf;
318
319                         tf->error = err;
320                         tf->status = stat;
321
322                         drive->hwif->tp_ops->tf_read(drive, task);
323
324                         if (task->tf_flags & IDE_TFLAG_DYN)
325                                 kfree(task);
326                 }
327         } else if (blk_pm_request(rq)) {
328                 struct request_pm_state *pm = rq->data;
329
330                 ide_complete_power_step(drive, rq);
331                 if (pm->pm_step == IDE_PM_COMPLETED)
332                         ide_complete_pm_request(drive, rq);
333                 return;
334         }
335
336         spin_lock_irqsave(&ide_lock, flags);
337         HWGROUP(drive)->rq = NULL;
338         rq->errors = err;
339         if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
340                                        blk_rq_bytes(rq))))
341                 BUG();
342         spin_unlock_irqrestore(&ide_lock, flags);
343 }
344
345 EXPORT_SYMBOL(ide_end_drive_cmd);
346
347 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
348 {
349         if (rq->rq_disk) {
350                 ide_driver_t *drv;
351
352                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
353                 drv->end_request(drive, 0, 0);
354         } else
355                 ide_end_request(drive, 0, 0);
356 }
357
358 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
359 {
360         ide_hwif_t *hwif = drive->hwif;
361
362         if ((stat & ATA_BUSY) ||
363             ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
364                 /* other bits are useless when BUSY */
365                 rq->errors |= ERROR_RESET;
366         } else if (stat & ATA_ERR) {
367                 /* err has different meaning on cdrom and tape */
368                 if (err == ATA_ABORTED) {
369                         if ((drive->dev_flags & IDE_DFLAG_LBA) &&
370                             /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
371                             hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
372                                 return ide_stopped;
373                 } else if ((err & BAD_CRC) == BAD_CRC) {
374                         /* UDMA crc error, just retry the operation */
375                         drive->crc_count++;
376                 } else if (err & (ATA_BBK | ATA_UNC)) {
377                         /* retries won't help these */
378                         rq->errors = ERROR_MAX;
379                 } else if (err & ATA_TRK0NF) {
380                         /* help it find track zero */
381                         rq->errors |= ERROR_RECAL;
382                 }
383         }
384
385         if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
386             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
387                 int nsect = drive->mult_count ? drive->mult_count : 1;
388
389                 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
390         }
391
392         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
393                 ide_kill_rq(drive, rq);
394                 return ide_stopped;
395         }
396
397         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
398                 rq->errors |= ERROR_RESET;
399
400         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
401                 ++rq->errors;
402                 return ide_do_reset(drive);
403         }
404
405         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
406                 drive->special.b.recalibrate = 1;
407
408         ++rq->errors;
409
410         return ide_stopped;
411 }
412
413 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
414 {
415         ide_hwif_t *hwif = drive->hwif;
416
417         if ((stat & ATA_BUSY) ||
418             ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
419                 /* other bits are useless when BUSY */
420                 rq->errors |= ERROR_RESET;
421         } else {
422                 /* add decoding error stuff */
423         }
424
425         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
426                 /* force an abort */
427                 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
428
429         if (rq->errors >= ERROR_MAX) {
430                 ide_kill_rq(drive, rq);
431         } else {
432                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
433                         ++rq->errors;
434                         return ide_do_reset(drive);
435                 }
436                 ++rq->errors;
437         }
438
439         return ide_stopped;
440 }
441
442 ide_startstop_t
443 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
444 {
445         if (drive->media == ide_disk)
446                 return ide_ata_error(drive, rq, stat, err);
447         return ide_atapi_error(drive, rq, stat, err);
448 }
449
450 EXPORT_SYMBOL_GPL(__ide_error);
451
452 /**
453  *      ide_error       -       handle an error on the IDE
454  *      @drive: drive the error occurred on
455  *      @msg: message to report
456  *      @stat: status bits
457  *
458  *      ide_error() takes action based on the error returned by the drive.
459  *      For normal I/O that may well include retries. We deal with
460  *      both new-style (taskfile) and old style command handling here.
461  *      In the case of taskfile command handling there is work left to
462  *      do
463  */
464  
465 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
466 {
467         struct request *rq;
468         u8 err;
469
470         err = ide_dump_status(drive, msg, stat);
471
472         if ((rq = HWGROUP(drive)->rq) == NULL)
473                 return ide_stopped;
474
475         /* retry only "normal" I/O: */
476         if (!blk_fs_request(rq)) {
477                 rq->errors = 1;
478                 ide_end_drive_cmd(drive, stat, err);
479                 return ide_stopped;
480         }
481
482         if (rq->rq_disk) {
483                 ide_driver_t *drv;
484
485                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
486                 return drv->error(drive, rq, stat, err);
487         } else
488                 return __ide_error(drive, rq, stat, err);
489 }
490
491 EXPORT_SYMBOL_GPL(ide_error);
492
493 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
494 {
495         tf->nsect   = drive->sect;
496         tf->lbal    = drive->sect;
497         tf->lbam    = drive->cyl;
498         tf->lbah    = drive->cyl >> 8;
499         tf->device  = (drive->head - 1) | drive->select;
500         tf->command = ATA_CMD_INIT_DEV_PARAMS;
501 }
502
503 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
504 {
505         tf->nsect   = drive->sect;
506         tf->command = ATA_CMD_RESTORE;
507 }
508
509 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
510 {
511         tf->nsect   = drive->mult_req;
512         tf->command = ATA_CMD_SET_MULTI;
513 }
514
515 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
516 {
517         special_t *s = &drive->special;
518         ide_task_t args;
519
520         memset(&args, 0, sizeof(ide_task_t));
521         args.data_phase = TASKFILE_NO_DATA;
522
523         if (s->b.set_geometry) {
524                 s->b.set_geometry = 0;
525                 ide_tf_set_specify_cmd(drive, &args.tf);
526         } else if (s->b.recalibrate) {
527                 s->b.recalibrate = 0;
528                 ide_tf_set_restore_cmd(drive, &args.tf);
529         } else if (s->b.set_multmode) {
530                 s->b.set_multmode = 0;
531                 ide_tf_set_setmult_cmd(drive, &args.tf);
532         } else if (s->all) {
533                 int special = s->all;
534                 s->all = 0;
535                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
536                 return ide_stopped;
537         }
538
539         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
540                         IDE_TFLAG_CUSTOM_HANDLER;
541
542         do_rw_taskfile(drive, &args);
543
544         return ide_started;
545 }
546
547 /**
548  *      do_special              -       issue some special commands
549  *      @drive: drive the command is for
550  *
551  *      do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
552  *      ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
553  *
554  *      It used to do much more, but has been scaled back.
555  */
556
557 static ide_startstop_t do_special (ide_drive_t *drive)
558 {
559         special_t *s = &drive->special;
560
561 #ifdef DEBUG
562         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
563 #endif
564         if (drive->media == ide_disk)
565                 return ide_disk_special(drive);
566
567         s->all = 0;
568         drive->mult_req = 0;
569         return ide_stopped;
570 }
571
572 void ide_map_sg(ide_drive_t *drive, struct request *rq)
573 {
574         ide_hwif_t *hwif = drive->hwif;
575         struct scatterlist *sg = hwif->sg_table;
576
577         if (hwif->sg_mapped)    /* needed by ide-scsi */
578                 return;
579
580         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
581                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
582         } else {
583                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
584                 hwif->sg_nents = 1;
585         }
586 }
587
588 EXPORT_SYMBOL_GPL(ide_map_sg);
589
590 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
591 {
592         ide_hwif_t *hwif = drive->hwif;
593
594         hwif->nsect = hwif->nleft = rq->nr_sectors;
595         hwif->cursg_ofs = 0;
596         hwif->cursg = NULL;
597 }
598
599 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
600
601 /**
602  *      execute_drive_command   -       issue special drive command
603  *      @drive: the drive to issue the command on
604  *      @rq: the request structure holding the command
605  *
606  *      execute_drive_cmd() issues a special drive command,  usually 
607  *      initiated by ioctl() from the external hdparm program. The
608  *      command can be a drive command, drive task or taskfile 
609  *      operation. Weirdly you can call it with NULL to wait for
610  *      all commands to finish. Don't do this as that is due to change
611  */
612
613 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
614                 struct request *rq)
615 {
616         ide_hwif_t *hwif = HWIF(drive);
617         ide_task_t *task = rq->special;
618
619         if (task) {
620                 hwif->data_phase = task->data_phase;
621
622                 switch (hwif->data_phase) {
623                 case TASKFILE_MULTI_OUT:
624                 case TASKFILE_OUT:
625                 case TASKFILE_MULTI_IN:
626                 case TASKFILE_IN:
627                         ide_init_sg_cmd(drive, rq);
628                         ide_map_sg(drive, rq);
629                 default:
630                         break;
631                 }
632
633                 return do_rw_taskfile(drive, task);
634         }
635
636         /*
637          * NULL is actually a valid way of waiting for
638          * all current requests to be flushed from the queue.
639          */
640 #ifdef DEBUG
641         printk("%s: DRIVE_CMD (null)\n", drive->name);
642 #endif
643         ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
644                           ide_read_error(drive));
645
646         return ide_stopped;
647 }
648
649 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
650                        int arg)
651 {
652         struct request_queue *q = drive->queue;
653         struct request *rq;
654         int ret = 0;
655
656         if (!(setting->flags & DS_SYNC))
657                 return setting->set(drive, arg);
658
659         rq = blk_get_request(q, READ, __GFP_WAIT);
660         rq->cmd_type = REQ_TYPE_SPECIAL;
661         rq->cmd_len = 5;
662         rq->cmd[0] = REQ_DEVSET_EXEC;
663         *(int *)&rq->cmd[1] = arg;
664         rq->special = setting->set;
665
666         if (blk_execute_rq(q, NULL, rq, 0))
667                 ret = rq->errors;
668         blk_put_request(rq);
669
670         return ret;
671 }
672 EXPORT_SYMBOL_GPL(ide_devset_execute);
673
674 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
675 {
676         u8 cmd = rq->cmd[0];
677
678         if (cmd == REQ_PARK_HEADS || cmd == REQ_UNPARK_HEADS) {
679                 ide_task_t task;
680                 struct ide_taskfile *tf = &task.tf;
681
682                 memset(&task, 0, sizeof(task));
683                 if (cmd == REQ_PARK_HEADS) {
684                         drive->sleep = *(unsigned long *)rq->special;
685                         drive->dev_flags |= IDE_DFLAG_SLEEPING;
686                         tf->command = ATA_CMD_IDLEIMMEDIATE;
687                         tf->feature = 0x44;
688                         tf->lbal = 0x4c;
689                         tf->lbam = 0x4e;
690                         tf->lbah = 0x55;
691                         task.tf_flags |= IDE_TFLAG_CUSTOM_HANDLER;
692                 } else          /* cmd == REQ_UNPARK_HEADS */
693                         tf->command = ATA_CMD_CHK_POWER;
694
695                 task.tf_flags |= IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
696                 task.rq = rq;
697                 drive->hwif->data_phase = task.data_phase = TASKFILE_NO_DATA;
698                 return do_rw_taskfile(drive, &task);
699         }
700
701         switch (cmd) {
702         case REQ_DEVSET_EXEC:
703         {
704                 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
705
706                 err = setfunc(drive, *(int *)&rq->cmd[1]);
707                 if (err)
708                         rq->errors = err;
709                 else
710                         err = 1;
711                 ide_end_request(drive, err, 0);
712                 return ide_stopped;
713         }
714         case REQ_DRIVE_RESET:
715                 return ide_do_reset(drive);
716         default:
717                 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
718                 ide_end_request(drive, 0, 0);
719                 return ide_stopped;
720         }
721 }
722
723 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
724 {
725         struct request_pm_state *pm = rq->data;
726
727         if (blk_pm_suspend_request(rq) &&
728             pm->pm_step == IDE_PM_START_SUSPEND)
729                 /* Mark drive blocked when starting the suspend sequence. */
730                 drive->dev_flags |= IDE_DFLAG_BLOCKED;
731         else if (blk_pm_resume_request(rq) &&
732                  pm->pm_step == IDE_PM_START_RESUME) {
733                 /* 
734                  * The first thing we do on wakeup is to wait for BSY bit to
735                  * go away (with a looong timeout) as a drive on this hwif may
736                  * just be POSTing itself.
737                  * We do that before even selecting as the "other" device on
738                  * the bus may be broken enough to walk on our toes at this
739                  * point.
740                  */
741                 ide_hwif_t *hwif = drive->hwif;
742                 int rc;
743 #ifdef DEBUG_PM
744                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
745 #endif
746                 rc = ide_wait_not_busy(hwif, 35000);
747                 if (rc)
748                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
749                 SELECT_DRIVE(drive);
750                 hwif->tp_ops->set_irq(hwif, 1);
751                 rc = ide_wait_not_busy(hwif, 100000);
752                 if (rc)
753                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
754         }
755 }
756
757 /**
758  *      start_request   -       start of I/O and command issuing for IDE
759  *
760  *      start_request() initiates handling of a new I/O request. It
761  *      accepts commands and I/O (read/write) requests.
762  *
763  *      FIXME: this function needs a rename
764  */
765  
766 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
767 {
768         ide_startstop_t startstop;
769
770         BUG_ON(!blk_rq_started(rq));
771
772 #ifdef DEBUG
773         printk("%s: start_request: current=0x%08lx\n",
774                 HWIF(drive)->name, (unsigned long) rq);
775 #endif
776
777         /* bail early if we've exceeded max_failures */
778         if (drive->max_failures && (drive->failures > drive->max_failures)) {
779                 rq->cmd_flags |= REQ_FAILED;
780                 goto kill_rq;
781         }
782
783         if (blk_pm_request(rq))
784                 ide_check_pm_state(drive, rq);
785
786         SELECT_DRIVE(drive);
787         if (ide_wait_stat(&startstop, drive, drive->ready_stat,
788                           ATA_BUSY | ATA_DRQ, WAIT_READY)) {
789                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
790                 return startstop;
791         }
792         if (!drive->special.all) {
793                 ide_driver_t *drv;
794
795                 /*
796                  * We reset the drive so we need to issue a SETFEATURES.
797                  * Do it _after_ do_special() restored device parameters.
798                  */
799                 if (drive->current_speed == 0xff)
800                         ide_config_drive_speed(drive, drive->desired_speed);
801
802                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
803                         return execute_drive_cmd(drive, rq);
804                 else if (blk_pm_request(rq)) {
805                         struct request_pm_state *pm = rq->data;
806 #ifdef DEBUG_PM
807                         printk("%s: start_power_step(step: %d)\n",
808                                 drive->name, pm->pm_step);
809 #endif
810                         startstop = ide_start_power_step(drive, rq);
811                         if (startstop == ide_stopped &&
812                             pm->pm_step == IDE_PM_COMPLETED)
813                                 ide_complete_pm_request(drive, rq);
814                         return startstop;
815                 } else if (!rq->rq_disk && blk_special_request(rq))
816                         /*
817                          * TODO: Once all ULDs have been modified to
818                          * check for specific op codes rather than
819                          * blindly accepting any special request, the
820                          * check for ->rq_disk above may be replaced
821                          * by a more suitable mechanism or even
822                          * dropped entirely.
823                          */
824                         return ide_special_rq(drive, rq);
825
826                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
827
828                 return drv->do_request(drive, rq, rq->sector);
829         }
830         return do_special(drive);
831 kill_rq:
832         ide_kill_rq(drive, rq);
833         return ide_stopped;
834 }
835
836 /**
837  *      ide_stall_queue         -       pause an IDE device
838  *      @drive: drive to stall
839  *      @timeout: time to stall for (jiffies)
840  *
841  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
842  *      to the hwgroup by sleeping for timeout jiffies.
843  */
844  
845 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
846 {
847         if (timeout > WAIT_WORSTCASE)
848                 timeout = WAIT_WORSTCASE;
849         drive->sleep = timeout + jiffies;
850         drive->dev_flags |= IDE_DFLAG_SLEEPING;
851 }
852
853 EXPORT_SYMBOL(ide_stall_queue);
854
855 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
856
857 /**
858  *      choose_drive            -       select a drive to service
859  *      @hwgroup: hardware group to select on
860  *
861  *      choose_drive() selects the next drive which will be serviced.
862  *      This is necessary because the IDE layer can't issue commands
863  *      to both drives on the same cable, unlike SCSI.
864  */
865  
866 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
867 {
868         ide_drive_t *drive, *best;
869
870 repeat: 
871         best = NULL;
872         drive = hwgroup->drive;
873
874         /*
875          * drive is doing pre-flush, ordered write, post-flush sequence. even
876          * though that is 3 requests, it must be seen as a single transaction.
877          * we must not preempt this drive until that is complete
878          */
879         if (blk_queue_flushing(drive->queue)) {
880                 /*
881                  * small race where queue could get replugged during
882                  * the 3-request flush cycle, just yank the plug since
883                  * we want it to finish asap
884                  */
885                 blk_remove_plug(drive->queue);
886                 return drive;
887         }
888
889         do {
890                 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
891                 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
892
893                 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
894                     !elv_queue_empty(drive->queue)) {
895                         if (best == NULL ||
896                             (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
897                             (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
898                                 if (!blk_queue_plugged(drive->queue))
899                                         best = drive;
900                         }
901                 }
902         } while ((drive = drive->next) != hwgroup->drive);
903
904         if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
905             (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
906             best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
907                 long t = (signed long)(WAKEUP(best) - jiffies);
908                 if (t >= WAIT_MIN_SLEEP) {
909                 /*
910                  * We *may* have some time to spare, but first let's see if
911                  * someone can potentially benefit from our nice mood today..
912                  */
913                         drive = best->next;
914                         do {
915                                 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
916                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
917                                  && time_before(WAKEUP(drive), jiffies + t))
918                                 {
919                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
920                                         goto repeat;
921                                 }
922                         } while ((drive = drive->next) != best);
923                 }
924         }
925         return best;
926 }
927
928 /*
929  * Issue a new request to a drive from hwgroup
930  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
931  *
932  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
933  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
934  * may have both interfaces in a single hwgroup to "serialize" access.
935  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
936  * together into one hwgroup for serialized access.
937  *
938  * Note also that several hwgroups can end up sharing a single IRQ,
939  * possibly along with many other devices.  This is especially common in
940  * PCI-based systems with off-board IDE controller cards.
941  *
942  * The IDE driver uses the single global ide_lock spinlock to protect
943  * access to the request queues, and to protect the hwgroup->busy flag.
944  *
945  * The first thread into the driver for a particular hwgroup sets the
946  * hwgroup->busy flag to indicate that this hwgroup is now active,
947  * and then initiates processing of the top request from the request queue.
948  *
949  * Other threads attempting entry notice the busy setting, and will simply
950  * queue their new requests and exit immediately.  Note that hwgroup->busy
951  * remains set even when the driver is merely awaiting the next interrupt.
952  * Thus, the meaning is "this hwgroup is busy processing a request".
953  *
954  * When processing of a request completes, the completing thread or IRQ-handler
955  * will start the next request from the queue.  If no more work remains,
956  * the driver will clear the hwgroup->busy flag and exit.
957  *
958  * The ide_lock (spinlock) is used to protect all access to the
959  * hwgroup->busy flag, but is otherwise not needed for most processing in
960  * the driver.  This makes the driver much more friendlier to shared IRQs
961  * than previous designs, while remaining 100% (?) SMP safe and capable.
962  */
963 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
964 {
965         ide_drive_t     *drive;
966         ide_hwif_t      *hwif;
967         struct request  *rq;
968         ide_startstop_t startstop;
969         int             loops = 0;
970
971         /* caller must own ide_lock */
972         BUG_ON(!irqs_disabled());
973
974         while (!hwgroup->busy) {
975                 hwgroup->busy = 1;
976                 /* for atari only */
977                 ide_get_lock(ide_intr, hwgroup);
978                 drive = choose_drive(hwgroup);
979                 if (drive == NULL) {
980                         int sleeping = 0;
981                         unsigned long sleep = 0; /* shut up, gcc */
982                         hwgroup->rq = NULL;
983                         drive = hwgroup->drive;
984                         do {
985                                 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
986                                     (sleeping == 0 ||
987                                      time_before(drive->sleep, sleep))) {
988                                         sleeping = 1;
989                                         sleep = drive->sleep;
990                                 }
991                         } while ((drive = drive->next) != hwgroup->drive);
992                         if (sleeping) {
993                 /*
994                  * Take a short snooze, and then wake up this hwgroup again.
995                  * This gives other hwgroups on the same a chance to
996                  * play fairly with us, just in case there are big differences
997                  * in relative throughputs.. don't want to hog the cpu too much.
998                  */
999                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1000                                         sleep = jiffies + WAIT_MIN_SLEEP;
1001 #if 1
1002                                 if (timer_pending(&hwgroup->timer))
1003                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1004 #endif
1005                                 /* so that ide_timer_expiry knows what to do */
1006                                 hwgroup->sleeping = 1;
1007                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1008                                 mod_timer(&hwgroup->timer, sleep);
1009                                 /* we purposely leave hwgroup->busy==1
1010                                  * while sleeping */
1011                         } else {
1012                                 /* Ugly, but how can we sleep for the lock
1013                                  * otherwise? perhaps from tq_disk?
1014                                  */
1015
1016                                 /* for atari only */
1017                                 ide_release_lock();
1018                                 hwgroup->busy = 0;
1019                         }
1020
1021                         /* no more work for this hwgroup (for now) */
1022                         return;
1023                 }
1024         again:
1025                 hwif = HWIF(drive);
1026                 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1027                         /*
1028                          * set nIEN for previous hwif, drives in the
1029                          * quirk_list may not like intr setups/cleanups
1030                          */
1031                         if (drive->quirk_list != 1)
1032                                 hwif->tp_ops->set_irq(hwif, 0);
1033                 }
1034                 hwgroup->hwif = hwif;
1035                 hwgroup->drive = drive;
1036                 drive->dev_flags &= ~(IDE_DFLAG_SLEEPING | IDE_DFLAG_PARKED);
1037                 drive->service_start = jiffies;
1038
1039                 if (blk_queue_plugged(drive->queue)) {
1040                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1041                         break;
1042                 }
1043
1044                 /*
1045                  * we know that the queue isn't empty, but this can happen
1046                  * if the q->prep_rq_fn() decides to kill a request
1047                  */
1048                 rq = elv_next_request(drive->queue);
1049                 if (!rq) {
1050                         hwgroup->busy = 0;
1051                         break;
1052                 }
1053
1054                 /*
1055                  * Sanity: don't accept a request that isn't a PM request
1056                  * if we are currently power managed. This is very important as
1057                  * blk_stop_queue() doesn't prevent the elv_next_request()
1058                  * above to return us whatever is in the queue. Since we call
1059                  * ide_do_request() ourselves, we end up taking requests while
1060                  * the queue is blocked...
1061                  * 
1062                  * We let requests forced at head of queue with ide-preempt
1063                  * though. I hope that doesn't happen too much, hopefully not
1064                  * unless the subdriver triggers such a thing in its own PM
1065                  * state machine.
1066                  *
1067                  * We count how many times we loop here to make sure we service
1068                  * all drives in the hwgroup without looping for ever
1069                  */
1070                 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
1071                     blk_pm_request(rq) == 0 &&
1072                     (rq->cmd_flags & REQ_PREEMPT) == 0) {
1073                         drive = drive->next ? drive->next : hwgroup->drive;
1074                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1075                                 goto again;
1076                         /* We clear busy, there should be no pending ATA command at this point. */
1077                         hwgroup->busy = 0;
1078                         break;
1079                 }
1080
1081                 hwgroup->rq = rq;
1082
1083                 /*
1084                  * Some systems have trouble with IDE IRQs arriving while
1085                  * the driver is still setting things up.  So, here we disable
1086                  * the IRQ used by this interface while the request is being started.
1087                  * This may look bad at first, but pretty much the same thing
1088                  * happens anyway when any interrupt comes in, IDE or otherwise
1089                  *  -- the kernel masks the IRQ while it is being handled.
1090                  */
1091                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1092                         disable_irq_nosync(hwif->irq);
1093                 spin_unlock(&ide_lock);
1094                 local_irq_enable_in_hardirq();
1095                         /* allow other IRQs while we start this request */
1096                 startstop = start_request(drive, rq);
1097                 spin_lock_irq(&ide_lock);
1098                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1099                         enable_irq(hwif->irq);
1100                 if (startstop == ide_stopped)
1101                         hwgroup->busy = 0;
1102         }
1103 }
1104
1105 /*
1106  * Passes the stuff to ide_do_request
1107  */
1108 void do_ide_request(struct request_queue *q)
1109 {
1110         ide_drive_t *drive = q->queuedata;
1111
1112         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1113 }
1114
1115 /*
1116  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1117  * retry the current request in pio mode instead of risking tossing it
1118  * all away
1119  */
1120 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1121 {
1122         ide_hwif_t *hwif = HWIF(drive);
1123         struct request *rq;
1124         ide_startstop_t ret = ide_stopped;
1125
1126         /*
1127          * end current dma transaction
1128          */
1129
1130         if (error < 0) {
1131                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1132                 (void)hwif->dma_ops->dma_end(drive);
1133                 ret = ide_error(drive, "dma timeout error",
1134                                 hwif->tp_ops->read_status(hwif));
1135         } else {
1136                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1137                 hwif->dma_ops->dma_timeout(drive);
1138         }
1139
1140         /*
1141          * disable dma for now, but remember that we did so because of
1142          * a timeout -- we'll reenable after we finish this next request
1143          * (or rather the first chunk of it) in pio.
1144          */
1145         drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
1146         drive->retry_pio++;
1147         ide_dma_off_quietly(drive);
1148
1149         /*
1150          * un-busy drive etc (hwgroup->busy is cleared on return) and
1151          * make sure request is sane
1152          */
1153         rq = HWGROUP(drive)->rq;
1154
1155         if (!rq)
1156                 goto out;
1157
1158         HWGROUP(drive)->rq = NULL;
1159
1160         rq->errors = 0;
1161
1162         if (!rq->bio)
1163                 goto out;
1164
1165         rq->sector = rq->bio->bi_sector;
1166         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1167         rq->hard_cur_sectors = rq->current_nr_sectors;
1168         rq->buffer = bio_data(rq->bio);
1169 out:
1170         return ret;
1171 }
1172
1173 /**
1174  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1175  *      @data: timer callback magic (hwgroup)
1176  *
1177  *      An IDE command has timed out before the expected drive return
1178  *      occurred. At this point we attempt to clean up the current
1179  *      mess. If the current handler includes an expiry handler then
1180  *      we invoke the expiry handler, and providing it is happy the
1181  *      work is done. If that fails we apply generic recovery rules
1182  *      invoking the handler and checking the drive DMA status. We
1183  *      have an excessively incestuous relationship with the DMA
1184  *      logic that wants cleaning up.
1185  */
1186  
1187 void ide_timer_expiry (unsigned long data)
1188 {
1189         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1190         ide_handler_t   *handler;
1191         ide_expiry_t    *expiry;
1192         unsigned long   flags;
1193         unsigned long   wait = -1;
1194
1195         spin_lock_irqsave(&ide_lock, flags);
1196
1197         if (((handler = hwgroup->handler) == NULL) ||
1198             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1199                 /*
1200                  * Either a marginal timeout occurred
1201                  * (got the interrupt just as timer expired),
1202                  * or we were "sleeping" to give other devices a chance.
1203                  * Either way, we don't really want to complain about anything.
1204                  */
1205                 if (hwgroup->sleeping) {
1206                         hwgroup->sleeping = 0;
1207                         hwgroup->busy = 0;
1208                 }
1209         } else {
1210                 ide_drive_t *drive = hwgroup->drive;
1211                 if (!drive) {
1212                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1213                         hwgroup->handler = NULL;
1214                 } else {
1215                         ide_hwif_t *hwif;
1216                         ide_startstop_t startstop = ide_stopped;
1217                         if (!hwgroup->busy) {
1218                                 hwgroup->busy = 1;      /* paranoia */
1219                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1220                         }
1221                         if ((expiry = hwgroup->expiry) != NULL) {
1222                                 /* continue */
1223                                 if ((wait = expiry(drive)) > 0) {
1224                                         /* reset timer */
1225                                         hwgroup->timer.expires  = jiffies + wait;
1226                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1227                                         add_timer(&hwgroup->timer);
1228                                         spin_unlock_irqrestore(&ide_lock, flags);
1229                                         return;
1230                                 }
1231                         }
1232                         hwgroup->handler = NULL;
1233                         /*
1234                          * We need to simulate a real interrupt when invoking
1235                          * the handler() function, which means we need to
1236                          * globally mask the specific IRQ:
1237                          */
1238                         spin_unlock(&ide_lock);
1239                         hwif  = HWIF(drive);
1240                         /* disable_irq_nosync ?? */
1241                         disable_irq(hwif->irq);
1242                         /* local CPU only,
1243                          * as if we were handling an interrupt */
1244                         local_irq_disable();
1245                         if (hwgroup->polling) {
1246                                 startstop = handler(drive);
1247                         } else if (drive_is_ready(drive)) {
1248                                 if (drive->waiting_for_dma)
1249                                         hwif->dma_ops->dma_lost_irq(drive);
1250                                 (void)ide_ack_intr(hwif);
1251                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1252                                 startstop = handler(drive);
1253                         } else {
1254                                 if (drive->waiting_for_dma) {
1255                                         startstop = ide_dma_timeout_retry(drive, wait);
1256                                 } else
1257                                         startstop =
1258                                         ide_error(drive, "irq timeout",
1259                                                   hwif->tp_ops->read_status(hwif));
1260                         }
1261                         drive->service_time = jiffies - drive->service_start;
1262                         spin_lock_irq(&ide_lock);
1263                         enable_irq(hwif->irq);
1264                         if (startstop == ide_stopped)
1265                                 hwgroup->busy = 0;
1266                 }
1267         }
1268         ide_do_request(hwgroup, IDE_NO_IRQ);
1269         spin_unlock_irqrestore(&ide_lock, flags);
1270 }
1271
1272 /**
1273  *      unexpected_intr         -       handle an unexpected IDE interrupt
1274  *      @irq: interrupt line
1275  *      @hwgroup: hwgroup being processed
1276  *
1277  *      There's nothing really useful we can do with an unexpected interrupt,
1278  *      other than reading the status register (to clear it), and logging it.
1279  *      There should be no way that an irq can happen before we're ready for it,
1280  *      so we needn't worry much about losing an "important" interrupt here.
1281  *
1282  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1283  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1284  *      looks "good", we just ignore the interrupt completely.
1285  *
1286  *      This routine assumes __cli() is in effect when called.
1287  *
1288  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1289  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1290  *      we could screw up by interfering with a new request being set up for 
1291  *      irq15.
1292  *
1293  *      In reality, this is a non-issue.  The new command is not sent unless 
1294  *      the drive is ready to accept one, in which case we know the drive is
1295  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1296  *      before completing the issuance of any new drive command, so we will not
1297  *      be accidentally invoked as a result of any valid command completion
1298  *      interrupt.
1299  *
1300  *      Note that we must walk the entire hwgroup here. We know which hwif
1301  *      is doing the current command, but we don't know which hwif burped
1302  *      mysteriously.
1303  */
1304  
1305 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1306 {
1307         u8 stat;
1308         ide_hwif_t *hwif = hwgroup->hwif;
1309
1310         /*
1311          * handle the unexpected interrupt
1312          */
1313         do {
1314                 if (hwif->irq == irq) {
1315                         stat = hwif->tp_ops->read_status(hwif);
1316
1317                         if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1318                                 /* Try to not flood the console with msgs */
1319                                 static unsigned long last_msgtime, count;
1320                                 ++count;
1321                                 if (time_after(jiffies, last_msgtime + HZ)) {
1322                                         last_msgtime = jiffies;
1323                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1324                                                 "status=0x%02x, count=%ld\n",
1325                                                 hwif->name,
1326                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1327                                 }
1328                         }
1329                 }
1330         } while ((hwif = hwif->next) != hwgroup->hwif);
1331 }
1332
1333 /**
1334  *      ide_intr        -       default IDE interrupt handler
1335  *      @irq: interrupt number
1336  *      @dev_id: hwif group
1337  *      @regs: unused weirdness from the kernel irq layer
1338  *
1339  *      This is the default IRQ handler for the IDE layer. You should
1340  *      not need to override it. If you do be aware it is subtle in
1341  *      places
1342  *
1343  *      hwgroup->hwif is the interface in the group currently performing
1344  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1345  *      the IRQ handler to call. As we issue a command the handlers
1346  *      step through multiple states, reassigning the handler to the
1347  *      next step in the process. Unlike a smart SCSI controller IDE
1348  *      expects the main processor to sequence the various transfer
1349  *      stages. We also manage a poll timer to catch up with most
1350  *      timeout situations. There are still a few where the handlers
1351  *      don't ever decide to give up.
1352  *
1353  *      The handler eventually returns ide_stopped to indicate the
1354  *      request completed. At this point we issue the next request
1355  *      on the hwgroup and the process begins again.
1356  */
1357  
1358 irqreturn_t ide_intr (int irq, void *dev_id)
1359 {
1360         unsigned long flags;
1361         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1362         ide_hwif_t *hwif;
1363         ide_drive_t *drive;
1364         ide_handler_t *handler;
1365         ide_startstop_t startstop;
1366
1367         spin_lock_irqsave(&ide_lock, flags);
1368         hwif = hwgroup->hwif;
1369
1370         if (!ide_ack_intr(hwif)) {
1371                 spin_unlock_irqrestore(&ide_lock, flags);
1372                 return IRQ_NONE;
1373         }
1374
1375         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1376                 /*
1377                  * Not expecting an interrupt from this drive.
1378                  * That means this could be:
1379                  *      (1) an interrupt from another PCI device
1380                  *      sharing the same PCI INT# as us.
1381                  * or   (2) a drive just entered sleep or standby mode,
1382                  *      and is interrupting to let us know.
1383                  * or   (3) a spurious interrupt of unknown origin.
1384                  *
1385                  * For PCI, we cannot tell the difference,
1386                  * so in that case we just ignore it and hope it goes away.
1387                  *
1388                  * FIXME: unexpected_intr should be hwif-> then we can
1389                  * remove all the ifdef PCI crap
1390                  */
1391 #ifdef CONFIG_BLK_DEV_IDEPCI
1392                 if (hwif->chipset != ide_pci)
1393 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1394                 {
1395                         /*
1396                          * Probably not a shared PCI interrupt,
1397                          * so we can safely try to do something about it:
1398                          */
1399                         unexpected_intr(irq, hwgroup);
1400 #ifdef CONFIG_BLK_DEV_IDEPCI
1401                 } else {
1402                         /*
1403                          * Whack the status register, just in case
1404                          * we have a leftover pending IRQ.
1405                          */
1406                         (void)hwif->tp_ops->read_status(hwif);
1407 #endif /* CONFIG_BLK_DEV_IDEPCI */
1408                 }
1409                 spin_unlock_irqrestore(&ide_lock, flags);
1410                 return IRQ_NONE;
1411         }
1412         drive = hwgroup->drive;
1413         if (!drive) {
1414                 /*
1415                  * This should NEVER happen, and there isn't much
1416                  * we could do about it here.
1417                  *
1418                  * [Note - this can occur if the drive is hot unplugged]
1419                  */
1420                 spin_unlock_irqrestore(&ide_lock, flags);
1421                 return IRQ_HANDLED;
1422         }
1423         if (!drive_is_ready(drive)) {
1424                 /*
1425                  * This happens regularly when we share a PCI IRQ with
1426                  * another device.  Unfortunately, it can also happen
1427                  * with some buggy drives that trigger the IRQ before
1428                  * their status register is up to date.  Hopefully we have
1429                  * enough advance overhead that the latter isn't a problem.
1430                  */
1431                 spin_unlock_irqrestore(&ide_lock, flags);
1432                 return IRQ_NONE;
1433         }
1434         if (!hwgroup->busy) {
1435                 hwgroup->busy = 1;      /* paranoia */
1436                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1437         }
1438         hwgroup->handler = NULL;
1439         hwgroup->req_gen++;
1440         del_timer(&hwgroup->timer);
1441         spin_unlock(&ide_lock);
1442
1443         if (hwif->port_ops && hwif->port_ops->clear_irq)
1444                 hwif->port_ops->clear_irq(drive);
1445
1446         if (drive->dev_flags & IDE_DFLAG_UNMASK)
1447                 local_irq_enable_in_hardirq();
1448
1449         /* service this interrupt, may set handler for next interrupt */
1450         startstop = handler(drive);
1451
1452         spin_lock_irq(&ide_lock);
1453         /*
1454          * Note that handler() may have set things up for another
1455          * interrupt to occur soon, but it cannot happen until
1456          * we exit from this routine, because it will be the
1457          * same irq as is currently being serviced here, and Linux
1458          * won't allow another of the same (on any CPU) until we return.
1459          */
1460         drive->service_time = jiffies - drive->service_start;
1461         if (startstop == ide_stopped) {
1462                 if (hwgroup->handler == NULL) { /* paranoia */
1463                         hwgroup->busy = 0;
1464                         ide_do_request(hwgroup, hwif->irq);
1465                 } else {
1466                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1467                                 "on exit\n", drive->name);
1468                 }
1469         }
1470         spin_unlock_irqrestore(&ide_lock, flags);
1471         return IRQ_HANDLED;
1472 }
1473
1474 /**
1475  *      ide_do_drive_cmd        -       issue IDE special command
1476  *      @drive: device to issue command
1477  *      @rq: request to issue
1478  *
1479  *      This function issues a special IDE device request
1480  *      onto the request queue.
1481  *
1482  *      the rq is queued at the head of the request queue, displacing
1483  *      the currently-being-processed request and this function
1484  *      returns immediately without waiting for the new rq to be
1485  *      completed.  This is VERY DANGEROUS, and is intended for
1486  *      careful use by the ATAPI tape/cdrom driver code.
1487  */
1488
1489 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1490 {
1491         unsigned long flags;
1492         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1493
1494         spin_lock_irqsave(&ide_lock, flags);
1495         hwgroup->rq = NULL;
1496         __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 0);
1497         blk_start_queueing(drive->queue);
1498         spin_unlock_irqrestore(&ide_lock, flags);
1499 }
1500
1501 EXPORT_SYMBOL(ide_do_drive_cmd);
1502
1503 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1504 {
1505         ide_hwif_t *hwif = drive->hwif;
1506         ide_task_t task;
1507
1508         memset(&task, 0, sizeof(task));
1509         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1510                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1511         task.tf.feature = dma;          /* Use PIO/DMA */
1512         task.tf.lbam    = bcount & 0xff;
1513         task.tf.lbah    = (bcount >> 8) & 0xff;
1514
1515         ide_tf_dump(drive->name, &task.tf);
1516         hwif->tp_ops->set_irq(hwif, 1);
1517         SELECT_MASK(drive, 0);
1518         hwif->tp_ops->tf_load(drive, &task);
1519 }
1520
1521 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1522
1523 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1524 {
1525         ide_hwif_t *hwif = drive->hwif;
1526         u8 buf[4] = { 0 };
1527
1528         while (len > 0) {
1529                 if (write)
1530                         hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1531                 else
1532                         hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1533                 len -= 4;
1534         }
1535 }
1536 EXPORT_SYMBOL_GPL(ide_pad_transfer);