ide: sanitize ide*_pm_* enums
[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, u8 stat, u8 error)
136 {
137         struct request_pm_state *pm = rq->data;
138
139         if (drive->media != ide_disk)
140                 return;
141
142         switch (pm->pm_step) {
143         case IDE_PM_FLUSH_CACHE:        /* Suspend step 1 (flush cache) */
144                 if (pm->pm_state == PM_EVENT_FREEZE)
145                         pm->pm_step = IDE_PM_COMPLETED;
146                 else
147                         pm->pm_step = IDE_PM_STANDBY;
148                 break;
149         case IDE_PM_STANDBY:            /* Suspend step 2 (standby) */
150                 pm->pm_step = IDE_PM_COMPLETED;
151                 break;
152         case IDE_PM_RESTORE_PIO:        /* Resume step 1 (restore PIO) */
153                 pm->pm_step = IDE_PM_IDLE;
154                 break;
155         case IDE_PM_IDLE:               /* Resume step 2 (idle)*/
156                 pm->pm_step = IDE_PM_RESTORE_DMA;
157                 break;
158         }
159 }
160
161 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
162 {
163         struct request_pm_state *pm = rq->data;
164         ide_task_t *args = rq->special;
165
166         memset(args, 0, sizeof(*args));
167
168         switch (pm->pm_step) {
169         case IDE_PM_FLUSH_CACHE:        /* Suspend step 1 (flush cache) */
170                 if (drive->media != ide_disk)
171                         break;
172                 /* Not supported? Switch to next step now. */
173                 if (ata_id_flush_enabled(drive->id) == 0 ||
174                     (drive->dev_flags & IDE_DFLAG_WCACHE) == 0) {
175                         ide_complete_power_step(drive, rq, 0, 0);
176                         return ide_stopped;
177                 }
178                 if (ata_id_flush_ext_enabled(drive->id))
179                         args->tf.command = ATA_CMD_FLUSH_EXT;
180                 else
181                         args->tf.command = ATA_CMD_FLUSH;
182                 goto out_do_tf;
183         case IDE_PM_STANDBY:            /* Suspend step 2 (standby) */
184                 args->tf.command = ATA_CMD_STANDBYNOW1;
185                 goto out_do_tf;
186         case IDE_PM_RESTORE_PIO:        /* Resume step 1 (restore PIO) */
187                 ide_set_max_pio(drive);
188                 /*
189                  * skip IDE_PM_IDLE for ATAPI devices
190                  */
191                 if (drive->media != ide_disk)
192                         pm->pm_step = IDE_PM_RESTORE_DMA;
193                 else
194                         ide_complete_power_step(drive, rq, 0, 0);
195                 return ide_stopped;
196         case IDE_PM_IDLE:               /* Resume step 2 (idle) */
197                 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
198                 goto out_do_tf;
199         case IDE_PM_RESTORE_DMA:        /* Resume step 3 (restore DMA) */
200                 /*
201                  * Right now, all we do is call ide_set_dma(drive),
202                  * we could be smarter and check for current xfer_speed
203                  * in struct drive etc...
204                  */
205                 if (drive->hwif->dma_ops == NULL)
206                         break;
207                 /*
208                  * TODO: respect IDE_DFLAG_USING_DMA
209                  */
210                 ide_set_dma(drive);
211                 break;
212         }
213
214         pm->pm_step = IDE_PM_COMPLETED;
215         return ide_stopped;
216
217 out_do_tf:
218         args->tf_flags   = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
219         args->data_phase = TASKFILE_NO_DATA;
220         return do_rw_taskfile(drive, args);
221 }
222
223 /**
224  *      ide_end_dequeued_request        -       complete an IDE I/O
225  *      @drive: IDE device for the I/O
226  *      @uptodate:
227  *      @nr_sectors: number of sectors completed
228  *
229  *      Complete an I/O that is no longer on the request queue. This
230  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
231  *      We must still finish the old request but we must not tamper with the
232  *      queue in the meantime.
233  *
234  *      NOTE: This path does not handle barrier, but barrier is not supported
235  *      on ide-cd anyway.
236  */
237
238 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
239                              int uptodate, int nr_sectors)
240 {
241         unsigned long flags;
242         int ret;
243
244         spin_lock_irqsave(&ide_lock, flags);
245         BUG_ON(!blk_rq_started(rq));
246         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
247         spin_unlock_irqrestore(&ide_lock, flags);
248
249         return ret;
250 }
251 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
252
253
254 /**
255  *      ide_complete_pm_request - end the current Power Management request
256  *      @drive: target drive
257  *      @rq: request
258  *
259  *      This function cleans up the current PM request and stops the queue
260  *      if necessary.
261  */
262 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
263 {
264         unsigned long flags;
265
266 #ifdef DEBUG_PM
267         printk("%s: completing PM request, %s\n", drive->name,
268                blk_pm_suspend_request(rq) ? "suspend" : "resume");
269 #endif
270         spin_lock_irqsave(&ide_lock, flags);
271         if (blk_pm_suspend_request(rq)) {
272                 blk_stop_queue(drive->queue);
273         } else {
274                 drive->dev_flags &= ~IDE_DFLAG_BLOCKED;
275                 blk_start_queue(drive->queue);
276         }
277         HWGROUP(drive)->rq = NULL;
278         if (__blk_end_request(rq, 0, 0))
279                 BUG();
280         spin_unlock_irqrestore(&ide_lock, flags);
281 }
282
283 /**
284  *      ide_end_drive_cmd       -       end an explicit drive command
285  *      @drive: command 
286  *      @stat: status bits
287  *      @err: error bits
288  *
289  *      Clean up after success/failure of an explicit drive command.
290  *      These get thrown onto the queue so they are synchronized with
291  *      real I/O operations on the drive.
292  *
293  *      In LBA48 mode we have to read the register set twice to get
294  *      all the extra information out.
295  */
296  
297 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
298 {
299         unsigned long flags;
300         struct request *rq;
301
302         spin_lock_irqsave(&ide_lock, flags);
303         rq = HWGROUP(drive)->rq;
304         spin_unlock_irqrestore(&ide_lock, flags);
305
306         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
307                 ide_task_t *task = (ide_task_t *)rq->special;
308
309                 if (rq->errors == 0)
310                         rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
311
312                 if (task) {
313                         struct ide_taskfile *tf = &task->tf;
314
315                         tf->error = err;
316                         tf->status = stat;
317
318                         drive->hwif->tp_ops->tf_read(drive, task);
319
320                         if (task->tf_flags & IDE_TFLAG_DYN)
321                                 kfree(task);
322                 }
323         } else if (blk_pm_request(rq)) {
324                 struct request_pm_state *pm = rq->data;
325 #ifdef DEBUG_PM
326                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
327                         drive->name, rq->pm->pm_step, stat, err);
328 #endif
329                 ide_complete_power_step(drive, rq, stat, err);
330                 if (pm->pm_step == IDE_PM_COMPLETED)
331                         ide_complete_pm_request(drive, rq);
332                 return;
333         }
334
335         spin_lock_irqsave(&ide_lock, flags);
336         HWGROUP(drive)->rq = NULL;
337         rq->errors = err;
338         if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
339                                        blk_rq_bytes(rq))))
340                 BUG();
341         spin_unlock_irqrestore(&ide_lock, flags);
342 }
343
344 EXPORT_SYMBOL(ide_end_drive_cmd);
345
346 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
347 {
348         if (rq->rq_disk) {
349                 ide_driver_t *drv;
350
351                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
352                 drv->end_request(drive, 0, 0);
353         } else
354                 ide_end_request(drive, 0, 0);
355 }
356
357 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
358 {
359         ide_hwif_t *hwif = drive->hwif;
360
361         if ((stat & ATA_BUSY) ||
362             ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
363                 /* other bits are useless when BUSY */
364                 rq->errors |= ERROR_RESET;
365         } else if (stat & ATA_ERR) {
366                 /* err has different meaning on cdrom and tape */
367                 if (err == ATA_ABORTED) {
368                         if ((drive->dev_flags & IDE_DFLAG_LBA) &&
369                             /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
370                             hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
371                                 return ide_stopped;
372                 } else if ((err & BAD_CRC) == BAD_CRC) {
373                         /* UDMA crc error, just retry the operation */
374                         drive->crc_count++;
375                 } else if (err & (ATA_BBK | ATA_UNC)) {
376                         /* retries won't help these */
377                         rq->errors = ERROR_MAX;
378                 } else if (err & ATA_TRK0NF) {
379                         /* help it find track zero */
380                         rq->errors |= ERROR_RECAL;
381                 }
382         }
383
384         if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
385             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
386                 int nsect = drive->mult_count ? drive->mult_count : 1;
387
388                 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
389         }
390
391         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
392                 ide_kill_rq(drive, rq);
393                 return ide_stopped;
394         }
395
396         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
397                 rq->errors |= ERROR_RESET;
398
399         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
400                 ++rq->errors;
401                 return ide_do_reset(drive);
402         }
403
404         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
405                 drive->special.b.recalibrate = 1;
406
407         ++rq->errors;
408
409         return ide_stopped;
410 }
411
412 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
413 {
414         ide_hwif_t *hwif = drive->hwif;
415
416         if ((stat & ATA_BUSY) ||
417             ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
418                 /* other bits are useless when BUSY */
419                 rq->errors |= ERROR_RESET;
420         } else {
421                 /* add decoding error stuff */
422         }
423
424         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
425                 /* force an abort */
426                 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
427
428         if (rq->errors >= ERROR_MAX) {
429                 ide_kill_rq(drive, rq);
430         } else {
431                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
432                         ++rq->errors;
433                         return ide_do_reset(drive);
434                 }
435                 ++rq->errors;
436         }
437
438         return ide_stopped;
439 }
440
441 ide_startstop_t
442 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
443 {
444         if (drive->media == ide_disk)
445                 return ide_ata_error(drive, rq, stat, err);
446         return ide_atapi_error(drive, rq, stat, err);
447 }
448
449 EXPORT_SYMBOL_GPL(__ide_error);
450
451 /**
452  *      ide_error       -       handle an error on the IDE
453  *      @drive: drive the error occurred on
454  *      @msg: message to report
455  *      @stat: status bits
456  *
457  *      ide_error() takes action based on the error returned by the drive.
458  *      For normal I/O that may well include retries. We deal with
459  *      both new-style (taskfile) and old style command handling here.
460  *      In the case of taskfile command handling there is work left to
461  *      do
462  */
463  
464 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
465 {
466         struct request *rq;
467         u8 err;
468
469         err = ide_dump_status(drive, msg, stat);
470
471         if ((rq = HWGROUP(drive)->rq) == NULL)
472                 return ide_stopped;
473
474         /* retry only "normal" I/O: */
475         if (!blk_fs_request(rq)) {
476                 rq->errors = 1;
477                 ide_end_drive_cmd(drive, stat, err);
478                 return ide_stopped;
479         }
480
481         if (rq->rq_disk) {
482                 ide_driver_t *drv;
483
484                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
485                 return drv->error(drive, rq, stat, err);
486         } else
487                 return __ide_error(drive, rq, stat, err);
488 }
489
490 EXPORT_SYMBOL_GPL(ide_error);
491
492 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
493 {
494         tf->nsect   = drive->sect;
495         tf->lbal    = drive->sect;
496         tf->lbam    = drive->cyl;
497         tf->lbah    = drive->cyl >> 8;
498         tf->device  = (drive->head - 1) | drive->select.all;
499         tf->command = ATA_CMD_INIT_DEV_PARAMS;
500 }
501
502 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
503 {
504         tf->nsect   = drive->sect;
505         tf->command = ATA_CMD_RESTORE;
506 }
507
508 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
509 {
510         tf->nsect   = drive->mult_req;
511         tf->command = ATA_CMD_SET_MULTI;
512 }
513
514 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
515 {
516         special_t *s = &drive->special;
517         ide_task_t args;
518
519         memset(&args, 0, sizeof(ide_task_t));
520         args.data_phase = TASKFILE_NO_DATA;
521
522         if (s->b.set_geometry) {
523                 s->b.set_geometry = 0;
524                 ide_tf_set_specify_cmd(drive, &args.tf);
525         } else if (s->b.recalibrate) {
526                 s->b.recalibrate = 0;
527                 ide_tf_set_restore_cmd(drive, &args.tf);
528         } else if (s->b.set_multmode) {
529                 s->b.set_multmode = 0;
530                 ide_tf_set_setmult_cmd(drive, &args.tf);
531         } else if (s->all) {
532                 int special = s->all;
533                 s->all = 0;
534                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
535                 return ide_stopped;
536         }
537
538         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
539                         IDE_TFLAG_CUSTOM_HANDLER;
540
541         do_rw_taskfile(drive, &args);
542
543         return ide_started;
544 }
545
546 /*
547  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
548  */
549 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
550 {
551         switch (req_pio) {
552         case 202:
553         case 201:
554         case 200:
555         case 102:
556         case 101:
557         case 100:
558                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
559         case 9:
560         case 8:
561                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
562         case 7:
563         case 6:
564                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
565         default:
566                 return 0;
567         }
568 }
569
570 /**
571  *      do_special              -       issue some special commands
572  *      @drive: drive the command is for
573  *
574  *      do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
575  *      ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
576  *
577  *      It used to do much more, but has been scaled back.
578  */
579
580 static ide_startstop_t do_special (ide_drive_t *drive)
581 {
582         special_t *s = &drive->special;
583
584 #ifdef DEBUG
585         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
586 #endif
587         if (s->b.set_tune) {
588                 ide_hwif_t *hwif = drive->hwif;
589                 const struct ide_port_ops *port_ops = hwif->port_ops;
590                 u8 req_pio = drive->tune_req;
591
592                 s->b.set_tune = 0;
593
594                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
595                         /*
596                          * take ide_lock for IDE_DFLAG_[NO_]UNMASK/[NO_]IO_32BIT
597                          */
598                         if (req_pio == 8 || req_pio == 9) {
599                                 unsigned long flags;
600
601                                 spin_lock_irqsave(&ide_lock, flags);
602                                 port_ops->set_pio_mode(drive, req_pio);
603                                 spin_unlock_irqrestore(&ide_lock, flags);
604                         } else
605                                 port_ops->set_pio_mode(drive, req_pio);
606                 } else {
607                         int keep_dma =
608                                 !!(drive->dev_flags & IDE_DFLAG_USING_DMA);
609
610                         ide_set_pio(drive, req_pio);
611
612                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
613                                 if (keep_dma)
614                                         ide_dma_on(drive);
615                         }
616                 }
617
618                 return ide_stopped;
619         } else {
620                 if (drive->media == ide_disk)
621                         return ide_disk_special(drive);
622
623                 s->all = 0;
624                 drive->mult_req = 0;
625                 return ide_stopped;
626         }
627 }
628
629 void ide_map_sg(ide_drive_t *drive, struct request *rq)
630 {
631         ide_hwif_t *hwif = drive->hwif;
632         struct scatterlist *sg = hwif->sg_table;
633
634         if (hwif->sg_mapped)    /* needed by ide-scsi */
635                 return;
636
637         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
638                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
639         } else {
640                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
641                 hwif->sg_nents = 1;
642         }
643 }
644
645 EXPORT_SYMBOL_GPL(ide_map_sg);
646
647 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
648 {
649         ide_hwif_t *hwif = drive->hwif;
650
651         hwif->nsect = hwif->nleft = rq->nr_sectors;
652         hwif->cursg_ofs = 0;
653         hwif->cursg = NULL;
654 }
655
656 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
657
658 /**
659  *      execute_drive_command   -       issue special drive command
660  *      @drive: the drive to issue the command on
661  *      @rq: the request structure holding the command
662  *
663  *      execute_drive_cmd() issues a special drive command,  usually 
664  *      initiated by ioctl() from the external hdparm program. The
665  *      command can be a drive command, drive task or taskfile 
666  *      operation. Weirdly you can call it with NULL to wait for
667  *      all commands to finish. Don't do this as that is due to change
668  */
669
670 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
671                 struct request *rq)
672 {
673         ide_hwif_t *hwif = HWIF(drive);
674         ide_task_t *task = rq->special;
675
676         if (task) {
677                 hwif->data_phase = task->data_phase;
678
679                 switch (hwif->data_phase) {
680                 case TASKFILE_MULTI_OUT:
681                 case TASKFILE_OUT:
682                 case TASKFILE_MULTI_IN:
683                 case TASKFILE_IN:
684                         ide_init_sg_cmd(drive, rq);
685                         ide_map_sg(drive, rq);
686                 default:
687                         break;
688                 }
689
690                 return do_rw_taskfile(drive, task);
691         }
692
693         /*
694          * NULL is actually a valid way of waiting for
695          * all current requests to be flushed from the queue.
696          */
697 #ifdef DEBUG
698         printk("%s: DRIVE_CMD (null)\n", drive->name);
699 #endif
700         ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
701                           ide_read_error(drive));
702
703         return ide_stopped;
704 }
705
706 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
707                        int arg)
708 {
709         struct request_queue *q = drive->queue;
710         struct request *rq;
711         int ret = 0;
712
713         if (!(setting->flags & DS_SYNC))
714                 return setting->set(drive, arg);
715
716         rq = blk_get_request(q, READ, GFP_KERNEL);
717         if (!rq)
718                 return -ENOMEM;
719
720         rq->cmd_type = REQ_TYPE_SPECIAL;
721         rq->cmd_len = 5;
722         rq->cmd[0] = REQ_DEVSET_EXEC;
723         *(int *)&rq->cmd[1] = arg;
724         rq->special = setting->set;
725
726         if (blk_execute_rq(q, NULL, rq, 0))
727                 ret = rq->errors;
728         blk_put_request(rq);
729
730         return ret;
731 }
732 EXPORT_SYMBOL_GPL(ide_devset_execute);
733
734 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
735 {
736         switch (rq->cmd[0]) {
737         case REQ_DEVSET_EXEC:
738         {
739                 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
740
741                 err = setfunc(drive, *(int *)&rq->cmd[1]);
742                 if (err)
743                         rq->errors = err;
744                 else
745                         err = 1;
746                 ide_end_request(drive, err, 0);
747                 return ide_stopped;
748         }
749         case REQ_DRIVE_RESET:
750                 return ide_do_reset(drive);
751         default:
752                 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
753                 ide_end_request(drive, 0, 0);
754                 return ide_stopped;
755         }
756 }
757
758 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
759 {
760         struct request_pm_state *pm = rq->data;
761
762         if (blk_pm_suspend_request(rq) &&
763             pm->pm_step == IDE_PM_START_SUSPEND)
764                 /* Mark drive blocked when starting the suspend sequence. */
765                 drive->dev_flags |= IDE_DFLAG_BLOCKED;
766         else if (blk_pm_resume_request(rq) &&
767                  pm->pm_step == IDE_PM_START_RESUME) {
768                 /* 
769                  * The first thing we do on wakeup is to wait for BSY bit to
770                  * go away (with a looong timeout) as a drive on this hwif may
771                  * just be POSTing itself.
772                  * We do that before even selecting as the "other" device on
773                  * the bus may be broken enough to walk on our toes at this
774                  * point.
775                  */
776                 ide_hwif_t *hwif = drive->hwif;
777                 int rc;
778 #ifdef DEBUG_PM
779                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
780 #endif
781                 rc = ide_wait_not_busy(hwif, 35000);
782                 if (rc)
783                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
784                 SELECT_DRIVE(drive);
785                 hwif->tp_ops->set_irq(hwif, 1);
786                 rc = ide_wait_not_busy(hwif, 100000);
787                 if (rc)
788                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
789         }
790 }
791
792 /**
793  *      start_request   -       start of I/O and command issuing for IDE
794  *
795  *      start_request() initiates handling of a new I/O request. It
796  *      accepts commands and I/O (read/write) requests.
797  *
798  *      FIXME: this function needs a rename
799  */
800  
801 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
802 {
803         ide_startstop_t startstop;
804
805         BUG_ON(!blk_rq_started(rq));
806
807 #ifdef DEBUG
808         printk("%s: start_request: current=0x%08lx\n",
809                 HWIF(drive)->name, (unsigned long) rq);
810 #endif
811
812         /* bail early if we've exceeded max_failures */
813         if (drive->max_failures && (drive->failures > drive->max_failures)) {
814                 rq->cmd_flags |= REQ_FAILED;
815                 goto kill_rq;
816         }
817
818         if (blk_pm_request(rq))
819                 ide_check_pm_state(drive, rq);
820
821         SELECT_DRIVE(drive);
822         if (ide_wait_stat(&startstop, drive, drive->ready_stat,
823                           ATA_BUSY | ATA_DRQ, WAIT_READY)) {
824                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
825                 return startstop;
826         }
827         if (!drive->special.all) {
828                 ide_driver_t *drv;
829
830                 /*
831                  * We reset the drive so we need to issue a SETFEATURES.
832                  * Do it _after_ do_special() restored device parameters.
833                  */
834                 if (drive->current_speed == 0xff)
835                         ide_config_drive_speed(drive, drive->desired_speed);
836
837                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
838                         return execute_drive_cmd(drive, rq);
839                 else if (blk_pm_request(rq)) {
840                         struct request_pm_state *pm = rq->data;
841 #ifdef DEBUG_PM
842                         printk("%s: start_power_step(step: %d)\n",
843                                 drive->name, rq->pm->pm_step);
844 #endif
845                         startstop = ide_start_power_step(drive, rq);
846                         if (startstop == ide_stopped &&
847                             pm->pm_step == IDE_PM_COMPLETED)
848                                 ide_complete_pm_request(drive, rq);
849                         return startstop;
850                 } else if (!rq->rq_disk && blk_special_request(rq))
851                         /*
852                          * TODO: Once all ULDs have been modified to
853                          * check for specific op codes rather than
854                          * blindly accepting any special request, the
855                          * check for ->rq_disk above may be replaced
856                          * by a more suitable mechanism or even
857                          * dropped entirely.
858                          */
859                         return ide_special_rq(drive, rq);
860
861                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
862
863                 return drv->do_request(drive, rq, rq->sector);
864         }
865         return do_special(drive);
866 kill_rq:
867         ide_kill_rq(drive, rq);
868         return ide_stopped;
869 }
870
871 /**
872  *      ide_stall_queue         -       pause an IDE device
873  *      @drive: drive to stall
874  *      @timeout: time to stall for (jiffies)
875  *
876  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
877  *      to the hwgroup by sleeping for timeout jiffies.
878  */
879  
880 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
881 {
882         if (timeout > WAIT_WORSTCASE)
883                 timeout = WAIT_WORSTCASE;
884         drive->sleep = timeout + jiffies;
885         drive->dev_flags |= IDE_DFLAG_SLEEPING;
886 }
887
888 EXPORT_SYMBOL(ide_stall_queue);
889
890 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
891
892 /**
893  *      choose_drive            -       select a drive to service
894  *      @hwgroup: hardware group to select on
895  *
896  *      choose_drive() selects the next drive which will be serviced.
897  *      This is necessary because the IDE layer can't issue commands
898  *      to both drives on the same cable, unlike SCSI.
899  */
900  
901 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
902 {
903         ide_drive_t *drive, *best;
904
905 repeat: 
906         best = NULL;
907         drive = hwgroup->drive;
908
909         /*
910          * drive is doing pre-flush, ordered write, post-flush sequence. even
911          * though that is 3 requests, it must be seen as a single transaction.
912          * we must not preempt this drive until that is complete
913          */
914         if (blk_queue_flushing(drive->queue)) {
915                 /*
916                  * small race where queue could get replugged during
917                  * the 3-request flush cycle, just yank the plug since
918                  * we want it to finish asap
919                  */
920                 blk_remove_plug(drive->queue);
921                 return drive;
922         }
923
924         do {
925                 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
926                 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
927
928                 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
929                     !elv_queue_empty(drive->queue)) {
930                         if (best == NULL ||
931                             (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
932                             (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
933                                 if (!blk_queue_plugged(drive->queue))
934                                         best = drive;
935                         }
936                 }
937         } while ((drive = drive->next) != hwgroup->drive);
938
939         if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
940             (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
941             best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
942                 long t = (signed long)(WAKEUP(best) - jiffies);
943                 if (t >= WAIT_MIN_SLEEP) {
944                 /*
945                  * We *may* have some time to spare, but first let's see if
946                  * someone can potentially benefit from our nice mood today..
947                  */
948                         drive = best->next;
949                         do {
950                                 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
951                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
952                                  && time_before(WAKEUP(drive), jiffies + t))
953                                 {
954                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
955                                         goto repeat;
956                                 }
957                         } while ((drive = drive->next) != best);
958                 }
959         }
960         return best;
961 }
962
963 /*
964  * Issue a new request to a drive from hwgroup
965  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
966  *
967  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
968  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
969  * may have both interfaces in a single hwgroup to "serialize" access.
970  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
971  * together into one hwgroup for serialized access.
972  *
973  * Note also that several hwgroups can end up sharing a single IRQ,
974  * possibly along with many other devices.  This is especially common in
975  * PCI-based systems with off-board IDE controller cards.
976  *
977  * The IDE driver uses the single global ide_lock spinlock to protect
978  * access to the request queues, and to protect the hwgroup->busy flag.
979  *
980  * The first thread into the driver for a particular hwgroup sets the
981  * hwgroup->busy flag to indicate that this hwgroup is now active,
982  * and then initiates processing of the top request from the request queue.
983  *
984  * Other threads attempting entry notice the busy setting, and will simply
985  * queue their new requests and exit immediately.  Note that hwgroup->busy
986  * remains set even when the driver is merely awaiting the next interrupt.
987  * Thus, the meaning is "this hwgroup is busy processing a request".
988  *
989  * When processing of a request completes, the completing thread or IRQ-handler
990  * will start the next request from the queue.  If no more work remains,
991  * the driver will clear the hwgroup->busy flag and exit.
992  *
993  * The ide_lock (spinlock) is used to protect all access to the
994  * hwgroup->busy flag, but is otherwise not needed for most processing in
995  * the driver.  This makes the driver much more friendlier to shared IRQs
996  * than previous designs, while remaining 100% (?) SMP safe and capable.
997  */
998 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
999 {
1000         ide_drive_t     *drive;
1001         ide_hwif_t      *hwif;
1002         struct request  *rq;
1003         ide_startstop_t startstop;
1004         int             loops = 0;
1005
1006         /* for atari only: POSSIBLY BROKEN HERE(?) */
1007         ide_get_lock(ide_intr, hwgroup);
1008
1009         /* caller must own ide_lock */
1010         BUG_ON(!irqs_disabled());
1011
1012         while (!hwgroup->busy) {
1013                 hwgroup->busy = 1;
1014                 drive = choose_drive(hwgroup);
1015                 if (drive == NULL) {
1016                         int sleeping = 0;
1017                         unsigned long sleep = 0; /* shut up, gcc */
1018                         hwgroup->rq = NULL;
1019                         drive = hwgroup->drive;
1020                         do {
1021                                 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
1022                                     (sleeping == 0 ||
1023                                      time_before(drive->sleep, sleep))) {
1024                                         sleeping = 1;
1025                                         sleep = drive->sleep;
1026                                 }
1027                         } while ((drive = drive->next) != hwgroup->drive);
1028                         if (sleeping) {
1029                 /*
1030                  * Take a short snooze, and then wake up this hwgroup again.
1031                  * This gives other hwgroups on the same a chance to
1032                  * play fairly with us, just in case there are big differences
1033                  * in relative throughputs.. don't want to hog the cpu too much.
1034                  */
1035                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1036                                         sleep = jiffies + WAIT_MIN_SLEEP;
1037 #if 1
1038                                 if (timer_pending(&hwgroup->timer))
1039                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1040 #endif
1041                                 /* so that ide_timer_expiry knows what to do */
1042                                 hwgroup->sleeping = 1;
1043                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1044                                 mod_timer(&hwgroup->timer, sleep);
1045                                 /* we purposely leave hwgroup->busy==1
1046                                  * while sleeping */
1047                         } else {
1048                                 /* Ugly, but how can we sleep for the lock
1049                                  * otherwise? perhaps from tq_disk?
1050                                  */
1051
1052                                 /* for atari only */
1053                                 ide_release_lock();
1054                                 hwgroup->busy = 0;
1055                         }
1056
1057                         /* no more work for this hwgroup (for now) */
1058                         return;
1059                 }
1060         again:
1061                 hwif = HWIF(drive);
1062                 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1063                         /*
1064                          * set nIEN for previous hwif, drives in the
1065                          * quirk_list may not like intr setups/cleanups
1066                          */
1067                         if (drive->quirk_list != 1)
1068                                 hwif->tp_ops->set_irq(hwif, 0);
1069                 }
1070                 hwgroup->hwif = hwif;
1071                 hwgroup->drive = drive;
1072                 drive->dev_flags &= ~IDE_DFLAG_SLEEPING;
1073                 drive->service_start = jiffies;
1074
1075                 if (blk_queue_plugged(drive->queue)) {
1076                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1077                         break;
1078                 }
1079
1080                 /*
1081                  * we know that the queue isn't empty, but this can happen
1082                  * if the q->prep_rq_fn() decides to kill a request
1083                  */
1084                 rq = elv_next_request(drive->queue);
1085                 if (!rq) {
1086                         hwgroup->busy = 0;
1087                         break;
1088                 }
1089
1090                 /*
1091                  * Sanity: don't accept a request that isn't a PM request
1092                  * if we are currently power managed. This is very important as
1093                  * blk_stop_queue() doesn't prevent the elv_next_request()
1094                  * above to return us whatever is in the queue. Since we call
1095                  * ide_do_request() ourselves, we end up taking requests while
1096                  * the queue is blocked...
1097                  * 
1098                  * We let requests forced at head of queue with ide-preempt
1099                  * though. I hope that doesn't happen too much, hopefully not
1100                  * unless the subdriver triggers such a thing in its own PM
1101                  * state machine.
1102                  *
1103                  * We count how many times we loop here to make sure we service
1104                  * all drives in the hwgroup without looping for ever
1105                  */
1106                 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
1107                     blk_pm_request(rq) == 0 &&
1108                     (rq->cmd_flags & REQ_PREEMPT) == 0) {
1109                         drive = drive->next ? drive->next : hwgroup->drive;
1110                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1111                                 goto again;
1112                         /* We clear busy, there should be no pending ATA command at this point. */
1113                         hwgroup->busy = 0;
1114                         break;
1115                 }
1116
1117                 hwgroup->rq = rq;
1118
1119                 /*
1120                  * Some systems have trouble with IDE IRQs arriving while
1121                  * the driver is still setting things up.  So, here we disable
1122                  * the IRQ used by this interface while the request is being started.
1123                  * This may look bad at first, but pretty much the same thing
1124                  * happens anyway when any interrupt comes in, IDE or otherwise
1125                  *  -- the kernel masks the IRQ while it is being handled.
1126                  */
1127                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1128                         disable_irq_nosync(hwif->irq);
1129                 spin_unlock(&ide_lock);
1130                 local_irq_enable_in_hardirq();
1131                         /* allow other IRQs while we start this request */
1132                 startstop = start_request(drive, rq);
1133                 spin_lock_irq(&ide_lock);
1134                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1135                         enable_irq(hwif->irq);
1136                 if (startstop == ide_stopped)
1137                         hwgroup->busy = 0;
1138         }
1139 }
1140
1141 /*
1142  * Passes the stuff to ide_do_request
1143  */
1144 void do_ide_request(struct request_queue *q)
1145 {
1146         ide_drive_t *drive = q->queuedata;
1147
1148         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1149 }
1150
1151 /*
1152  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1153  * retry the current request in pio mode instead of risking tossing it
1154  * all away
1155  */
1156 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1157 {
1158         ide_hwif_t *hwif = HWIF(drive);
1159         struct request *rq;
1160         ide_startstop_t ret = ide_stopped;
1161
1162         /*
1163          * end current dma transaction
1164          */
1165
1166         if (error < 0) {
1167                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1168                 (void)hwif->dma_ops->dma_end(drive);
1169                 ret = ide_error(drive, "dma timeout error",
1170                                 hwif->tp_ops->read_status(hwif));
1171         } else {
1172                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1173                 hwif->dma_ops->dma_timeout(drive);
1174         }
1175
1176         /*
1177          * disable dma for now, but remember that we did so because of
1178          * a timeout -- we'll reenable after we finish this next request
1179          * (or rather the first chunk of it) in pio.
1180          */
1181         drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
1182         drive->retry_pio++;
1183         ide_dma_off_quietly(drive);
1184
1185         /*
1186          * un-busy drive etc (hwgroup->busy is cleared on return) and
1187          * make sure request is sane
1188          */
1189         rq = HWGROUP(drive)->rq;
1190
1191         if (!rq)
1192                 goto out;
1193
1194         HWGROUP(drive)->rq = NULL;
1195
1196         rq->errors = 0;
1197
1198         if (!rq->bio)
1199                 goto out;
1200
1201         rq->sector = rq->bio->bi_sector;
1202         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1203         rq->hard_cur_sectors = rq->current_nr_sectors;
1204         rq->buffer = bio_data(rq->bio);
1205 out:
1206         return ret;
1207 }
1208
1209 /**
1210  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1211  *      @data: timer callback magic (hwgroup)
1212  *
1213  *      An IDE command has timed out before the expected drive return
1214  *      occurred. At this point we attempt to clean up the current
1215  *      mess. If the current handler includes an expiry handler then
1216  *      we invoke the expiry handler, and providing it is happy the
1217  *      work is done. If that fails we apply generic recovery rules
1218  *      invoking the handler and checking the drive DMA status. We
1219  *      have an excessively incestuous relationship with the DMA
1220  *      logic that wants cleaning up.
1221  */
1222  
1223 void ide_timer_expiry (unsigned long data)
1224 {
1225         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1226         ide_handler_t   *handler;
1227         ide_expiry_t    *expiry;
1228         unsigned long   flags;
1229         unsigned long   wait = -1;
1230
1231         spin_lock_irqsave(&ide_lock, flags);
1232
1233         if (((handler = hwgroup->handler) == NULL) ||
1234             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1235                 /*
1236                  * Either a marginal timeout occurred
1237                  * (got the interrupt just as timer expired),
1238                  * or we were "sleeping" to give other devices a chance.
1239                  * Either way, we don't really want to complain about anything.
1240                  */
1241                 if (hwgroup->sleeping) {
1242                         hwgroup->sleeping = 0;
1243                         hwgroup->busy = 0;
1244                 }
1245         } else {
1246                 ide_drive_t *drive = hwgroup->drive;
1247                 if (!drive) {
1248                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1249                         hwgroup->handler = NULL;
1250                 } else {
1251                         ide_hwif_t *hwif;
1252                         ide_startstop_t startstop = ide_stopped;
1253                         if (!hwgroup->busy) {
1254                                 hwgroup->busy = 1;      /* paranoia */
1255                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1256                         }
1257                         if ((expiry = hwgroup->expiry) != NULL) {
1258                                 /* continue */
1259                                 if ((wait = expiry(drive)) > 0) {
1260                                         /* reset timer */
1261                                         hwgroup->timer.expires  = jiffies + wait;
1262                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1263                                         add_timer(&hwgroup->timer);
1264                                         spin_unlock_irqrestore(&ide_lock, flags);
1265                                         return;
1266                                 }
1267                         }
1268                         hwgroup->handler = NULL;
1269                         /*
1270                          * We need to simulate a real interrupt when invoking
1271                          * the handler() function, which means we need to
1272                          * globally mask the specific IRQ:
1273                          */
1274                         spin_unlock(&ide_lock);
1275                         hwif  = HWIF(drive);
1276                         /* disable_irq_nosync ?? */
1277                         disable_irq(hwif->irq);
1278                         /* local CPU only,
1279                          * as if we were handling an interrupt */
1280                         local_irq_disable();
1281                         if (hwgroup->polling) {
1282                                 startstop = handler(drive);
1283                         } else if (drive_is_ready(drive)) {
1284                                 if (drive->waiting_for_dma)
1285                                         hwif->dma_ops->dma_lost_irq(drive);
1286                                 (void)ide_ack_intr(hwif);
1287                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1288                                 startstop = handler(drive);
1289                         } else {
1290                                 if (drive->waiting_for_dma) {
1291                                         startstop = ide_dma_timeout_retry(drive, wait);
1292                                 } else
1293                                         startstop =
1294                                         ide_error(drive, "irq timeout",
1295                                                   hwif->tp_ops->read_status(hwif));
1296                         }
1297                         drive->service_time = jiffies - drive->service_start;
1298                         spin_lock_irq(&ide_lock);
1299                         enable_irq(hwif->irq);
1300                         if (startstop == ide_stopped)
1301                                 hwgroup->busy = 0;
1302                 }
1303         }
1304         ide_do_request(hwgroup, IDE_NO_IRQ);
1305         spin_unlock_irqrestore(&ide_lock, flags);
1306 }
1307
1308 /**
1309  *      unexpected_intr         -       handle an unexpected IDE interrupt
1310  *      @irq: interrupt line
1311  *      @hwgroup: hwgroup being processed
1312  *
1313  *      There's nothing really useful we can do with an unexpected interrupt,
1314  *      other than reading the status register (to clear it), and logging it.
1315  *      There should be no way that an irq can happen before we're ready for it,
1316  *      so we needn't worry much about losing an "important" interrupt here.
1317  *
1318  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1319  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1320  *      looks "good", we just ignore the interrupt completely.
1321  *
1322  *      This routine assumes __cli() is in effect when called.
1323  *
1324  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1325  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1326  *      we could screw up by interfering with a new request being set up for 
1327  *      irq15.
1328  *
1329  *      In reality, this is a non-issue.  The new command is not sent unless 
1330  *      the drive is ready to accept one, in which case we know the drive is
1331  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1332  *      before completing the issuance of any new drive command, so we will not
1333  *      be accidentally invoked as a result of any valid command completion
1334  *      interrupt.
1335  *
1336  *      Note that we must walk the entire hwgroup here. We know which hwif
1337  *      is doing the current command, but we don't know which hwif burped
1338  *      mysteriously.
1339  */
1340  
1341 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1342 {
1343         u8 stat;
1344         ide_hwif_t *hwif = hwgroup->hwif;
1345
1346         /*
1347          * handle the unexpected interrupt
1348          */
1349         do {
1350                 if (hwif->irq == irq) {
1351                         stat = hwif->tp_ops->read_status(hwif);
1352
1353                         if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1354                                 /* Try to not flood the console with msgs */
1355                                 static unsigned long last_msgtime, count;
1356                                 ++count;
1357                                 if (time_after(jiffies, last_msgtime + HZ)) {
1358                                         last_msgtime = jiffies;
1359                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1360                                                 "status=0x%02x, count=%ld\n",
1361                                                 hwif->name,
1362                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1363                                 }
1364                         }
1365                 }
1366         } while ((hwif = hwif->next) != hwgroup->hwif);
1367 }
1368
1369 /**
1370  *      ide_intr        -       default IDE interrupt handler
1371  *      @irq: interrupt number
1372  *      @dev_id: hwif group
1373  *      @regs: unused weirdness from the kernel irq layer
1374  *
1375  *      This is the default IRQ handler for the IDE layer. You should
1376  *      not need to override it. If you do be aware it is subtle in
1377  *      places
1378  *
1379  *      hwgroup->hwif is the interface in the group currently performing
1380  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1381  *      the IRQ handler to call. As we issue a command the handlers
1382  *      step through multiple states, reassigning the handler to the
1383  *      next step in the process. Unlike a smart SCSI controller IDE
1384  *      expects the main processor to sequence the various transfer
1385  *      stages. We also manage a poll timer to catch up with most
1386  *      timeout situations. There are still a few where the handlers
1387  *      don't ever decide to give up.
1388  *
1389  *      The handler eventually returns ide_stopped to indicate the
1390  *      request completed. At this point we issue the next request
1391  *      on the hwgroup and the process begins again.
1392  */
1393  
1394 irqreturn_t ide_intr (int irq, void *dev_id)
1395 {
1396         unsigned long flags;
1397         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1398         ide_hwif_t *hwif;
1399         ide_drive_t *drive;
1400         ide_handler_t *handler;
1401         ide_startstop_t startstop;
1402
1403         spin_lock_irqsave(&ide_lock, flags);
1404         hwif = hwgroup->hwif;
1405
1406         if (!ide_ack_intr(hwif)) {
1407                 spin_unlock_irqrestore(&ide_lock, flags);
1408                 return IRQ_NONE;
1409         }
1410
1411         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1412                 /*
1413                  * Not expecting an interrupt from this drive.
1414                  * That means this could be:
1415                  *      (1) an interrupt from another PCI device
1416                  *      sharing the same PCI INT# as us.
1417                  * or   (2) a drive just entered sleep or standby mode,
1418                  *      and is interrupting to let us know.
1419                  * or   (3) a spurious interrupt of unknown origin.
1420                  *
1421                  * For PCI, we cannot tell the difference,
1422                  * so in that case we just ignore it and hope it goes away.
1423                  *
1424                  * FIXME: unexpected_intr should be hwif-> then we can
1425                  * remove all the ifdef PCI crap
1426                  */
1427 #ifdef CONFIG_BLK_DEV_IDEPCI
1428                 if (hwif->chipset != ide_pci)
1429 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1430                 {
1431                         /*
1432                          * Probably not a shared PCI interrupt,
1433                          * so we can safely try to do something about it:
1434                          */
1435                         unexpected_intr(irq, hwgroup);
1436 #ifdef CONFIG_BLK_DEV_IDEPCI
1437                 } else {
1438                         /*
1439                          * Whack the status register, just in case
1440                          * we have a leftover pending IRQ.
1441                          */
1442                         (void)hwif->tp_ops->read_status(hwif);
1443 #endif /* CONFIG_BLK_DEV_IDEPCI */
1444                 }
1445                 spin_unlock_irqrestore(&ide_lock, flags);
1446                 return IRQ_NONE;
1447         }
1448         drive = hwgroup->drive;
1449         if (!drive) {
1450                 /*
1451                  * This should NEVER happen, and there isn't much
1452                  * we could do about it here.
1453                  *
1454                  * [Note - this can occur if the drive is hot unplugged]
1455                  */
1456                 spin_unlock_irqrestore(&ide_lock, flags);
1457                 return IRQ_HANDLED;
1458         }
1459         if (!drive_is_ready(drive)) {
1460                 /*
1461                  * This happens regularly when we share a PCI IRQ with
1462                  * another device.  Unfortunately, it can also happen
1463                  * with some buggy drives that trigger the IRQ before
1464                  * their status register is up to date.  Hopefully we have
1465                  * enough advance overhead that the latter isn't a problem.
1466                  */
1467                 spin_unlock_irqrestore(&ide_lock, flags);
1468                 return IRQ_NONE;
1469         }
1470         if (!hwgroup->busy) {
1471                 hwgroup->busy = 1;      /* paranoia */
1472                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1473         }
1474         hwgroup->handler = NULL;
1475         hwgroup->req_gen++;
1476         del_timer(&hwgroup->timer);
1477         spin_unlock(&ide_lock);
1478
1479         /* Some controllers might set DMA INTR no matter DMA or PIO;
1480          * bmdma status might need to be cleared even for
1481          * PIO interrupts to prevent spurious/lost irq.
1482          */
1483         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1484                 /* ide_dma_end() needs bmdma status for error checking.
1485                  * So, skip clearing bmdma status here and leave it
1486                  * to ide_dma_end() if this is dma interrupt.
1487                  */
1488                 hwif->ide_dma_clear_irq(drive);
1489
1490         if (drive->dev_flags & IDE_DFLAG_UNMASK)
1491                 local_irq_enable_in_hardirq();
1492         /* service this interrupt, may set handler for next interrupt */
1493         startstop = handler(drive);
1494         spin_lock_irq(&ide_lock);
1495
1496         /*
1497          * Note that handler() may have set things up for another
1498          * interrupt to occur soon, but it cannot happen until
1499          * we exit from this routine, because it will be the
1500          * same irq as is currently being serviced here, and Linux
1501          * won't allow another of the same (on any CPU) until we return.
1502          */
1503         drive->service_time = jiffies - drive->service_start;
1504         if (startstop == ide_stopped) {
1505                 if (hwgroup->handler == NULL) { /* paranoia */
1506                         hwgroup->busy = 0;
1507                         ide_do_request(hwgroup, hwif->irq);
1508                 } else {
1509                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1510                                 "on exit\n", drive->name);
1511                 }
1512         }
1513         spin_unlock_irqrestore(&ide_lock, flags);
1514         return IRQ_HANDLED;
1515 }
1516
1517 /**
1518  *      ide_do_drive_cmd        -       issue IDE special command
1519  *      @drive: device to issue command
1520  *      @rq: request to issue
1521  *
1522  *      This function issues a special IDE device request
1523  *      onto the request queue.
1524  *
1525  *      the rq is queued at the head of the request queue, displacing
1526  *      the currently-being-processed request and this function
1527  *      returns immediately without waiting for the new rq to be
1528  *      completed.  This is VERY DANGEROUS, and is intended for
1529  *      careful use by the ATAPI tape/cdrom driver code.
1530  */
1531
1532 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1533 {
1534         unsigned long flags;
1535         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1536
1537         spin_lock_irqsave(&ide_lock, flags);
1538         hwgroup->rq = NULL;
1539         __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1540         __generic_unplug_device(drive->queue);
1541         spin_unlock_irqrestore(&ide_lock, flags);
1542 }
1543
1544 EXPORT_SYMBOL(ide_do_drive_cmd);
1545
1546 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1547 {
1548         ide_hwif_t *hwif = drive->hwif;
1549         ide_task_t task;
1550
1551         memset(&task, 0, sizeof(task));
1552         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1553                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1554         task.tf.feature = dma;          /* Use PIO/DMA */
1555         task.tf.lbam    = bcount & 0xff;
1556         task.tf.lbah    = (bcount >> 8) & 0xff;
1557
1558         ide_tf_dump(drive->name, &task.tf);
1559         hwif->tp_ops->set_irq(hwif, 1);
1560         SELECT_MASK(drive, 0);
1561         hwif->tp_ops->tf_load(drive, &task);
1562 }
1563
1564 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1565
1566 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1567 {
1568         ide_hwif_t *hwif = drive->hwif;
1569         u8 buf[4] = { 0 };
1570
1571         while (len > 0) {
1572                 if (write)
1573                         hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1574                 else
1575                         hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1576                 len -= 4;
1577         }
1578 }
1579 EXPORT_SYMBOL_GPL(ide_pad_transfer);