Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
32 #include <linux/mm.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/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->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
82                 drive->state = 0;
83                 ide_dma_on(drive);
84         }
85
86         if (!__blk_end_request(rq, error, nr_bytes)) {
87                 if (dequeue)
88                         HWGROUP(drive)->rq = NULL;
89                 ret = 0;
90         }
91
92         return ret;
93 }
94
95 /**
96  *      ide_end_request         -       complete an IDE I/O
97  *      @drive: IDE device for the I/O
98  *      @uptodate:
99  *      @nr_sectors: number of sectors completed
100  *
101  *      This is our end_request wrapper function. We complete the I/O
102  *      update random number input and dequeue the request, which if
103  *      it was tagged may be out of order.
104  */
105
106 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
107 {
108         unsigned int nr_bytes = nr_sectors << 9;
109         struct request *rq;
110         unsigned long flags;
111         int ret = 1;
112
113         /*
114          * room for locking improvements here, the calls below don't
115          * need the queue lock held at all
116          */
117         spin_lock_irqsave(&ide_lock, flags);
118         rq = HWGROUP(drive)->rq;
119
120         if (!nr_bytes) {
121                 if (blk_pc_request(rq))
122                         nr_bytes = rq->data_len;
123                 else
124                         nr_bytes = rq->hard_cur_sectors << 9;
125         }
126
127         ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
128
129         spin_unlock_irqrestore(&ide_lock, flags);
130         return ret;
131 }
132 EXPORT_SYMBOL(ide_end_request);
133
134 /*
135  * Power Management state machine. This one is rather trivial for now,
136  * we should probably add more, like switching back to PIO on suspend
137  * to help some BIOSes, re-do the door locking on resume, etc...
138  */
139
140 enum {
141         ide_pm_flush_cache      = ide_pm_state_start_suspend,
142         idedisk_pm_standby,
143
144         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
145         idedisk_pm_idle,
146         ide_pm_restore_dma,
147 };
148
149 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
150 {
151         struct request_pm_state *pm = rq->data;
152
153         if (drive->media != ide_disk)
154                 return;
155
156         switch (pm->pm_step) {
157         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
158                 if (pm->pm_state == PM_EVENT_FREEZE)
159                         pm->pm_step = ide_pm_state_completed;
160                 else
161                         pm->pm_step = idedisk_pm_standby;
162                 break;
163         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
164                 pm->pm_step = ide_pm_state_completed;
165                 break;
166         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
167                 pm->pm_step = idedisk_pm_idle;
168                 break;
169         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
170                 pm->pm_step = ide_pm_restore_dma;
171                 break;
172         }
173 }
174
175 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
176 {
177         struct request_pm_state *pm = rq->data;
178         ide_task_t *args = rq->special;
179
180         memset(args, 0, sizeof(*args));
181
182         switch (pm->pm_step) {
183         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
184                 if (drive->media != ide_disk)
185                         break;
186                 /* Not supported? Switch to next step now. */
187                 if (!drive->wcache || ata_id_flush_enabled(drive->id) == 0) {
188                         ide_complete_power_step(drive, rq, 0, 0);
189                         return ide_stopped;
190                 }
191                 if (ata_id_flush_ext_enabled(drive->id))
192                         args->tf.command = ATA_CMD_FLUSH_EXT;
193                 else
194                         args->tf.command = ATA_CMD_FLUSH;
195                 goto out_do_tf;
196
197         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
198                 args->tf.command = ATA_CMD_STANDBYNOW1;
199                 goto out_do_tf;
200
201         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
202                 ide_set_max_pio(drive);
203                 /*
204                  * skip idedisk_pm_idle for ATAPI devices
205                  */
206                 if (drive->media != ide_disk)
207                         pm->pm_step = ide_pm_restore_dma;
208                 else
209                         ide_complete_power_step(drive, rq, 0, 0);
210                 return ide_stopped;
211
212         case idedisk_pm_idle:           /* Resume step 2 (idle) */
213                 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
214                 goto out_do_tf;
215
216         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
217                 /*
218                  * Right now, all we do is call ide_set_dma(drive),
219                  * we could be smarter and check for current xfer_speed
220                  * in struct drive etc...
221                  */
222                 if (drive->hwif->dma_ops == NULL)
223                         break;
224                 /*
225                  * TODO: respect ->using_dma setting
226                  */
227                 ide_set_dma(drive);
228                 break;
229         }
230         pm->pm_step = ide_pm_state_completed;
231         return ide_stopped;
232
233 out_do_tf:
234         args->tf_flags   = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
235         args->data_phase = TASKFILE_NO_DATA;
236         return do_rw_taskfile(drive, args);
237 }
238
239 /**
240  *      ide_end_dequeued_request        -       complete an IDE I/O
241  *      @drive: IDE device for the I/O
242  *      @uptodate:
243  *      @nr_sectors: number of sectors completed
244  *
245  *      Complete an I/O that is no longer on the request queue. This
246  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
247  *      We must still finish the old request but we must not tamper with the
248  *      queue in the meantime.
249  *
250  *      NOTE: This path does not handle barrier, but barrier is not supported
251  *      on ide-cd anyway.
252  */
253
254 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
255                              int uptodate, int nr_sectors)
256 {
257         unsigned long flags;
258         int ret;
259
260         spin_lock_irqsave(&ide_lock, flags);
261         BUG_ON(!blk_rq_started(rq));
262         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
263         spin_unlock_irqrestore(&ide_lock, flags);
264
265         return ret;
266 }
267 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
268
269
270 /**
271  *      ide_complete_pm_request - end the current Power Management request
272  *      @drive: target drive
273  *      @rq: request
274  *
275  *      This function cleans up the current PM request and stops the queue
276  *      if necessary.
277  */
278 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
279 {
280         unsigned long flags;
281
282 #ifdef DEBUG_PM
283         printk("%s: completing PM request, %s\n", drive->name,
284                blk_pm_suspend_request(rq) ? "suspend" : "resume");
285 #endif
286         spin_lock_irqsave(&ide_lock, flags);
287         if (blk_pm_suspend_request(rq)) {
288                 blk_stop_queue(drive->queue);
289         } else {
290                 drive->blocked = 0;
291                 blk_start_queue(drive->queue);
292         }
293         HWGROUP(drive)->rq = NULL;
294         if (__blk_end_request(rq, 0, 0))
295                 BUG();
296         spin_unlock_irqrestore(&ide_lock, flags);
297 }
298
299 /**
300  *      ide_end_drive_cmd       -       end an explicit drive command
301  *      @drive: command 
302  *      @stat: status bits
303  *      @err: error bits
304  *
305  *      Clean up after success/failure of an explicit drive command.
306  *      These get thrown onto the queue so they are synchronized with
307  *      real I/O operations on the drive.
308  *
309  *      In LBA48 mode we have to read the register set twice to get
310  *      all the extra information out.
311  */
312  
313 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
314 {
315         unsigned long flags;
316         struct request *rq;
317
318         spin_lock_irqsave(&ide_lock, flags);
319         rq = HWGROUP(drive)->rq;
320         spin_unlock_irqrestore(&ide_lock, flags);
321
322         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
323                 ide_task_t *task = (ide_task_t *)rq->special;
324
325                 if (rq->errors == 0)
326                         rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
327
328                 if (task) {
329                         struct ide_taskfile *tf = &task->tf;
330
331                         tf->error = err;
332                         tf->status = stat;
333
334                         drive->hwif->tp_ops->tf_read(drive, task);
335
336                         if (task->tf_flags & IDE_TFLAG_DYN)
337                                 kfree(task);
338                 }
339         } else if (blk_pm_request(rq)) {
340                 struct request_pm_state *pm = rq->data;
341 #ifdef DEBUG_PM
342                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
343                         drive->name, rq->pm->pm_step, stat, err);
344 #endif
345                 ide_complete_power_step(drive, rq, stat, err);
346                 if (pm->pm_step == ide_pm_state_completed)
347                         ide_complete_pm_request(drive, rq);
348                 return;
349         }
350
351         spin_lock_irqsave(&ide_lock, flags);
352         HWGROUP(drive)->rq = NULL;
353         rq->errors = err;
354         if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
355                                        blk_rq_bytes(rq))))
356                 BUG();
357         spin_unlock_irqrestore(&ide_lock, flags);
358 }
359
360 EXPORT_SYMBOL(ide_end_drive_cmd);
361
362 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
363 {
364         if (rq->rq_disk) {
365                 ide_driver_t *drv;
366
367                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
368                 drv->end_request(drive, 0, 0);
369         } else
370                 ide_end_request(drive, 0, 0);
371 }
372
373 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
374 {
375         ide_hwif_t *hwif = drive->hwif;
376
377         if ((stat & ATA_BUSY) || ((stat & ATA_DF) && !drive->nowerr)) {
378                 /* other bits are useless when BUSY */
379                 rq->errors |= ERROR_RESET;
380         } else if (stat & ATA_ERR) {
381                 /* err has different meaning on cdrom and tape */
382                 if (err == ATA_ABORTED) {
383                         if (drive->select.b.lba &&
384                             /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
385                             hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
386                                 return ide_stopped;
387                 } else if ((err & BAD_CRC) == BAD_CRC) {
388                         /* UDMA crc error, just retry the operation */
389                         drive->crc_count++;
390                 } else if (err & (ATA_BBK | ATA_UNC)) {
391                         /* retries won't help these */
392                         rq->errors = ERROR_MAX;
393                 } else if (err & ATA_TRK0NF) {
394                         /* help it find track zero */
395                         rq->errors |= ERROR_RECAL;
396                 }
397         }
398
399         if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
400             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
401                 int nsect = drive->mult_count ? drive->mult_count : 1;
402
403                 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
404         }
405
406         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
407                 ide_kill_rq(drive, rq);
408                 return ide_stopped;
409         }
410
411         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
412                 rq->errors |= ERROR_RESET;
413
414         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
415                 ++rq->errors;
416                 return ide_do_reset(drive);
417         }
418
419         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
420                 drive->special.b.recalibrate = 1;
421
422         ++rq->errors;
423
424         return ide_stopped;
425 }
426
427 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
428 {
429         ide_hwif_t *hwif = drive->hwif;
430
431         if ((stat & ATA_BUSY) || ((stat & ATA_DF) && !drive->nowerr)) {
432                 /* other bits are useless when BUSY */
433                 rq->errors |= ERROR_RESET;
434         } else {
435                 /* add decoding error stuff */
436         }
437
438         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
439                 /* force an abort */
440                 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
441
442         if (rq->errors >= ERROR_MAX) {
443                 ide_kill_rq(drive, rq);
444         } else {
445                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
446                         ++rq->errors;
447                         return ide_do_reset(drive);
448                 }
449                 ++rq->errors;
450         }
451
452         return ide_stopped;
453 }
454
455 ide_startstop_t
456 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
457 {
458         if (drive->media == ide_disk)
459                 return ide_ata_error(drive, rq, stat, err);
460         return ide_atapi_error(drive, rq, stat, err);
461 }
462
463 EXPORT_SYMBOL_GPL(__ide_error);
464
465 /**
466  *      ide_error       -       handle an error on the IDE
467  *      @drive: drive the error occurred on
468  *      @msg: message to report
469  *      @stat: status bits
470  *
471  *      ide_error() takes action based on the error returned by the drive.
472  *      For normal I/O that may well include retries. We deal with
473  *      both new-style (taskfile) and old style command handling here.
474  *      In the case of taskfile command handling there is work left to
475  *      do
476  */
477  
478 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
479 {
480         struct request *rq;
481         u8 err;
482
483         err = ide_dump_status(drive, msg, stat);
484
485         if ((rq = HWGROUP(drive)->rq) == NULL)
486                 return ide_stopped;
487
488         /* retry only "normal" I/O: */
489         if (!blk_fs_request(rq)) {
490                 rq->errors = 1;
491                 ide_end_drive_cmd(drive, stat, err);
492                 return ide_stopped;
493         }
494
495         if (rq->rq_disk) {
496                 ide_driver_t *drv;
497
498                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
499                 return drv->error(drive, rq, stat, err);
500         } else
501                 return __ide_error(drive, rq, stat, err);
502 }
503
504 EXPORT_SYMBOL_GPL(ide_error);
505
506 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
507 {
508         tf->nsect   = drive->sect;
509         tf->lbal    = drive->sect;
510         tf->lbam    = drive->cyl;
511         tf->lbah    = drive->cyl >> 8;
512         tf->device  = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
513         tf->command = ATA_CMD_INIT_DEV_PARAMS;
514 }
515
516 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
517 {
518         tf->nsect   = drive->sect;
519         tf->command = ATA_CMD_RESTORE;
520 }
521
522 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
523 {
524         tf->nsect   = drive->mult_req;
525         tf->command = ATA_CMD_SET_MULTI;
526 }
527
528 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
529 {
530         special_t *s = &drive->special;
531         ide_task_t args;
532
533         memset(&args, 0, sizeof(ide_task_t));
534         args.data_phase = TASKFILE_NO_DATA;
535
536         if (s->b.set_geometry) {
537                 s->b.set_geometry = 0;
538                 ide_tf_set_specify_cmd(drive, &args.tf);
539         } else if (s->b.recalibrate) {
540                 s->b.recalibrate = 0;
541                 ide_tf_set_restore_cmd(drive, &args.tf);
542         } else if (s->b.set_multmode) {
543                 s->b.set_multmode = 0;
544                 ide_tf_set_setmult_cmd(drive, &args.tf);
545         } else if (s->all) {
546                 int special = s->all;
547                 s->all = 0;
548                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
549                 return ide_stopped;
550         }
551
552         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
553                         IDE_TFLAG_CUSTOM_HANDLER;
554
555         do_rw_taskfile(drive, &args);
556
557         return ide_started;
558 }
559
560 /*
561  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
562  */
563 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
564 {
565         switch (req_pio) {
566         case 202:
567         case 201:
568         case 200:
569         case 102:
570         case 101:
571         case 100:
572                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
573         case 9:
574         case 8:
575                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
576         case 7:
577         case 6:
578                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
579         default:
580                 return 0;
581         }
582 }
583
584 /**
585  *      do_special              -       issue some special commands
586  *      @drive: drive the command is for
587  *
588  *      do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
589  *      ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
590  *
591  *      It used to do much more, but has been scaled back.
592  */
593
594 static ide_startstop_t do_special (ide_drive_t *drive)
595 {
596         special_t *s = &drive->special;
597
598 #ifdef DEBUG
599         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
600 #endif
601         if (s->b.set_tune) {
602                 ide_hwif_t *hwif = drive->hwif;
603                 const struct ide_port_ops *port_ops = hwif->port_ops;
604                 u8 req_pio = drive->tune_req;
605
606                 s->b.set_tune = 0;
607
608                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
609                         /*
610                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
611                          */
612                         if (req_pio == 8 || req_pio == 9) {
613                                 unsigned long flags;
614
615                                 spin_lock_irqsave(&ide_lock, flags);
616                                 port_ops->set_pio_mode(drive, req_pio);
617                                 spin_unlock_irqrestore(&ide_lock, flags);
618                         } else
619                                 port_ops->set_pio_mode(drive, req_pio);
620                 } else {
621                         int keep_dma = drive->using_dma;
622
623                         ide_set_pio(drive, req_pio);
624
625                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
626                                 if (keep_dma)
627                                         ide_dma_on(drive);
628                         }
629                 }
630
631                 return ide_stopped;
632         } else {
633                 if (drive->media == ide_disk)
634                         return ide_disk_special(drive);
635
636                 s->all = 0;
637                 drive->mult_req = 0;
638                 return ide_stopped;
639         }
640 }
641
642 void ide_map_sg(ide_drive_t *drive, struct request *rq)
643 {
644         ide_hwif_t *hwif = drive->hwif;
645         struct scatterlist *sg = hwif->sg_table;
646
647         if (hwif->sg_mapped)    /* needed by ide-scsi */
648                 return;
649
650         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
651                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
652         } else {
653                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
654                 hwif->sg_nents = 1;
655         }
656 }
657
658 EXPORT_SYMBOL_GPL(ide_map_sg);
659
660 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
661 {
662         ide_hwif_t *hwif = drive->hwif;
663
664         hwif->nsect = hwif->nleft = rq->nr_sectors;
665         hwif->cursg_ofs = 0;
666         hwif->cursg = NULL;
667 }
668
669 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
670
671 /**
672  *      execute_drive_command   -       issue special drive command
673  *      @drive: the drive to issue the command on
674  *      @rq: the request structure holding the command
675  *
676  *      execute_drive_cmd() issues a special drive command,  usually 
677  *      initiated by ioctl() from the external hdparm program. The
678  *      command can be a drive command, drive task or taskfile 
679  *      operation. Weirdly you can call it with NULL to wait for
680  *      all commands to finish. Don't do this as that is due to change
681  */
682
683 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
684                 struct request *rq)
685 {
686         ide_hwif_t *hwif = HWIF(drive);
687         ide_task_t *task = rq->special;
688
689         if (task) {
690                 hwif->data_phase = task->data_phase;
691
692                 switch (hwif->data_phase) {
693                 case TASKFILE_MULTI_OUT:
694                 case TASKFILE_OUT:
695                 case TASKFILE_MULTI_IN:
696                 case TASKFILE_IN:
697                         ide_init_sg_cmd(drive, rq);
698                         ide_map_sg(drive, rq);
699                 default:
700                         break;
701                 }
702
703                 return do_rw_taskfile(drive, task);
704         }
705
706         /*
707          * NULL is actually a valid way of waiting for
708          * all current requests to be flushed from the queue.
709          */
710 #ifdef DEBUG
711         printk("%s: DRIVE_CMD (null)\n", drive->name);
712 #endif
713         ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
714                           ide_read_error(drive));
715
716         return ide_stopped;
717 }
718
719 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
720                        int arg)
721 {
722         struct request_queue *q = drive->queue;
723         struct request *rq;
724         int ret = 0;
725
726         if (!(setting->flags & DS_SYNC))
727                 return setting->set(drive, arg);
728
729         rq = blk_get_request(q, READ, GFP_KERNEL);
730         if (!rq)
731                 return -ENOMEM;
732
733         rq->cmd_type = REQ_TYPE_SPECIAL;
734         rq->cmd_len = 5;
735         rq->cmd[0] = REQ_DEVSET_EXEC;
736         *(int *)&rq->cmd[1] = arg;
737         rq->special = setting->set;
738
739         if (blk_execute_rq(q, NULL, rq, 0))
740                 ret = rq->errors;
741         blk_put_request(rq);
742
743         return ret;
744 }
745 EXPORT_SYMBOL_GPL(ide_devset_execute);
746
747 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
748 {
749         switch (rq->cmd[0]) {
750         case REQ_DEVSET_EXEC:
751         {
752                 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
753
754                 err = setfunc(drive, *(int *)&rq->cmd[1]);
755                 if (err)
756                         rq->errors = err;
757                 else
758                         err = 1;
759                 ide_end_request(drive, err, 0);
760                 return ide_stopped;
761         }
762         case REQ_DRIVE_RESET:
763                 return ide_do_reset(drive);
764         default:
765                 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
766                 ide_end_request(drive, 0, 0);
767                 return ide_stopped;
768         }
769 }
770
771 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
772 {
773         struct request_pm_state *pm = rq->data;
774
775         if (blk_pm_suspend_request(rq) &&
776             pm->pm_step == ide_pm_state_start_suspend)
777                 /* Mark drive blocked when starting the suspend sequence. */
778                 drive->blocked = 1;
779         else if (blk_pm_resume_request(rq) &&
780                  pm->pm_step == ide_pm_state_start_resume) {
781                 /* 
782                  * The first thing we do on wakeup is to wait for BSY bit to
783                  * go away (with a looong timeout) as a drive on this hwif may
784                  * just be POSTing itself.
785                  * We do that before even selecting as the "other" device on
786                  * the bus may be broken enough to walk on our toes at this
787                  * point.
788                  */
789                 ide_hwif_t *hwif = drive->hwif;
790                 int rc;
791 #ifdef DEBUG_PM
792                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
793 #endif
794                 rc = ide_wait_not_busy(hwif, 35000);
795                 if (rc)
796                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
797                 SELECT_DRIVE(drive);
798                 hwif->tp_ops->set_irq(hwif, 1);
799                 rc = ide_wait_not_busy(hwif, 100000);
800                 if (rc)
801                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
802         }
803 }
804
805 /**
806  *      start_request   -       start of I/O and command issuing for IDE
807  *
808  *      start_request() initiates handling of a new I/O request. It
809  *      accepts commands and I/O (read/write) requests.
810  *
811  *      FIXME: this function needs a rename
812  */
813  
814 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
815 {
816         ide_startstop_t startstop;
817
818         BUG_ON(!blk_rq_started(rq));
819
820 #ifdef DEBUG
821         printk("%s: start_request: current=0x%08lx\n",
822                 HWIF(drive)->name, (unsigned long) rq);
823 #endif
824
825         /* bail early if we've exceeded max_failures */
826         if (drive->max_failures && (drive->failures > drive->max_failures)) {
827                 rq->cmd_flags |= REQ_FAILED;
828                 goto kill_rq;
829         }
830
831         if (blk_pm_request(rq))
832                 ide_check_pm_state(drive, rq);
833
834         SELECT_DRIVE(drive);
835         if (ide_wait_stat(&startstop, drive, drive->ready_stat,
836                           ATA_BUSY | ATA_DRQ, WAIT_READY)) {
837                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
838                 return startstop;
839         }
840         if (!drive->special.all) {
841                 ide_driver_t *drv;
842
843                 /*
844                  * We reset the drive so we need to issue a SETFEATURES.
845                  * Do it _after_ do_special() restored device parameters.
846                  */
847                 if (drive->current_speed == 0xff)
848                         ide_config_drive_speed(drive, drive->desired_speed);
849
850                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
851                         return execute_drive_cmd(drive, rq);
852                 else if (blk_pm_request(rq)) {
853                         struct request_pm_state *pm = rq->data;
854 #ifdef DEBUG_PM
855                         printk("%s: start_power_step(step: %d)\n",
856                                 drive->name, rq->pm->pm_step);
857 #endif
858                         startstop = ide_start_power_step(drive, rq);
859                         if (startstop == ide_stopped &&
860                             pm->pm_step == ide_pm_state_completed)
861                                 ide_complete_pm_request(drive, rq);
862                         return startstop;
863                 } else if (!rq->rq_disk && blk_special_request(rq))
864                         /*
865                          * TODO: Once all ULDs have been modified to
866                          * check for specific op codes rather than
867                          * blindly accepting any special request, the
868                          * check for ->rq_disk above may be replaced
869                          * by a more suitable mechanism or even
870                          * dropped entirely.
871                          */
872                         return ide_special_rq(drive, rq);
873
874                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
875
876                 return drv->do_request(drive, rq, rq->sector);
877         }
878         return do_special(drive);
879 kill_rq:
880         ide_kill_rq(drive, rq);
881         return ide_stopped;
882 }
883
884 /**
885  *      ide_stall_queue         -       pause an IDE device
886  *      @drive: drive to stall
887  *      @timeout: time to stall for (jiffies)
888  *
889  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
890  *      to the hwgroup by sleeping for timeout jiffies.
891  */
892  
893 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
894 {
895         if (timeout > WAIT_WORSTCASE)
896                 timeout = WAIT_WORSTCASE;
897         drive->sleep = timeout + jiffies;
898         drive->sleeping = 1;
899 }
900
901 EXPORT_SYMBOL(ide_stall_queue);
902
903 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
904
905 /**
906  *      choose_drive            -       select a drive to service
907  *      @hwgroup: hardware group to select on
908  *
909  *      choose_drive() selects the next drive which will be serviced.
910  *      This is necessary because the IDE layer can't issue commands
911  *      to both drives on the same cable, unlike SCSI.
912  */
913  
914 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
915 {
916         ide_drive_t *drive, *best;
917
918 repeat: 
919         best = NULL;
920         drive = hwgroup->drive;
921
922         /*
923          * drive is doing pre-flush, ordered write, post-flush sequence. even
924          * though that is 3 requests, it must be seen as a single transaction.
925          * we must not preempt this drive until that is complete
926          */
927         if (blk_queue_flushing(drive->queue)) {
928                 /*
929                  * small race where queue could get replugged during
930                  * the 3-request flush cycle, just yank the plug since
931                  * we want it to finish asap
932                  */
933                 blk_remove_plug(drive->queue);
934                 return drive;
935         }
936
937         do {
938                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
939                     && !elv_queue_empty(drive->queue)) {
940                         if (!best
941                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
942                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
943                         {
944                                 if (!blk_queue_plugged(drive->queue))
945                                         best = drive;
946                         }
947                 }
948         } while ((drive = drive->next) != hwgroup->drive);
949         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
950                 long t = (signed long)(WAKEUP(best) - jiffies);
951                 if (t >= WAIT_MIN_SLEEP) {
952                 /*
953                  * We *may* have some time to spare, but first let's see if
954                  * someone can potentially benefit from our nice mood today..
955                  */
956                         drive = best->next;
957                         do {
958                                 if (!drive->sleeping
959                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
960                                  && time_before(WAKEUP(drive), jiffies + t))
961                                 {
962                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
963                                         goto repeat;
964                                 }
965                         } while ((drive = drive->next) != best);
966                 }
967         }
968         return best;
969 }
970
971 /*
972  * Issue a new request to a drive from hwgroup
973  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
974  *
975  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
976  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
977  * may have both interfaces in a single hwgroup to "serialize" access.
978  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
979  * together into one hwgroup for serialized access.
980  *
981  * Note also that several hwgroups can end up sharing a single IRQ,
982  * possibly along with many other devices.  This is especially common in
983  * PCI-based systems with off-board IDE controller cards.
984  *
985  * The IDE driver uses the single global ide_lock spinlock to protect
986  * access to the request queues, and to protect the hwgroup->busy flag.
987  *
988  * The first thread into the driver for a particular hwgroup sets the
989  * hwgroup->busy flag to indicate that this hwgroup is now active,
990  * and then initiates processing of the top request from the request queue.
991  *
992  * Other threads attempting entry notice the busy setting, and will simply
993  * queue their new requests and exit immediately.  Note that hwgroup->busy
994  * remains set even when the driver is merely awaiting the next interrupt.
995  * Thus, the meaning is "this hwgroup is busy processing a request".
996  *
997  * When processing of a request completes, the completing thread or IRQ-handler
998  * will start the next request from the queue.  If no more work remains,
999  * the driver will clear the hwgroup->busy flag and exit.
1000  *
1001  * The ide_lock (spinlock) is used to protect all access to the
1002  * hwgroup->busy flag, but is otherwise not needed for most processing in
1003  * the driver.  This makes the driver much more friendlier to shared IRQs
1004  * than previous designs, while remaining 100% (?) SMP safe and capable.
1005  */
1006 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1007 {
1008         ide_drive_t     *drive;
1009         ide_hwif_t      *hwif;
1010         struct request  *rq;
1011         ide_startstop_t startstop;
1012         int             loops = 0;
1013
1014         /* for atari only: POSSIBLY BROKEN HERE(?) */
1015         ide_get_lock(ide_intr, hwgroup);
1016
1017         /* caller must own ide_lock */
1018         BUG_ON(!irqs_disabled());
1019
1020         while (!hwgroup->busy) {
1021                 hwgroup->busy = 1;
1022                 drive = choose_drive(hwgroup);
1023                 if (drive == NULL) {
1024                         int sleeping = 0;
1025                         unsigned long sleep = 0; /* shut up, gcc */
1026                         hwgroup->rq = NULL;
1027                         drive = hwgroup->drive;
1028                         do {
1029                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1030                                         sleeping = 1;
1031                                         sleep = drive->sleep;
1032                                 }
1033                         } while ((drive = drive->next) != hwgroup->drive);
1034                         if (sleeping) {
1035                 /*
1036                  * Take a short snooze, and then wake up this hwgroup again.
1037                  * This gives other hwgroups on the same a chance to
1038                  * play fairly with us, just in case there are big differences
1039                  * in relative throughputs.. don't want to hog the cpu too much.
1040                  */
1041                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1042                                         sleep = jiffies + WAIT_MIN_SLEEP;
1043 #if 1
1044                                 if (timer_pending(&hwgroup->timer))
1045                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1046 #endif
1047                                 /* so that ide_timer_expiry knows what to do */
1048                                 hwgroup->sleeping = 1;
1049                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1050                                 mod_timer(&hwgroup->timer, sleep);
1051                                 /* we purposely leave hwgroup->busy==1
1052                                  * while sleeping */
1053                         } else {
1054                                 /* Ugly, but how can we sleep for the lock
1055                                  * otherwise? perhaps from tq_disk?
1056                                  */
1057
1058                                 /* for atari only */
1059                                 ide_release_lock();
1060                                 hwgroup->busy = 0;
1061                         }
1062
1063                         /* no more work for this hwgroup (for now) */
1064                         return;
1065                 }
1066         again:
1067                 hwif = HWIF(drive);
1068                 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1069                         /*
1070                          * set nIEN for previous hwif, drives in the
1071                          * quirk_list may not like intr setups/cleanups
1072                          */
1073                         if (drive->quirk_list != 1)
1074                                 hwif->tp_ops->set_irq(hwif, 0);
1075                 }
1076                 hwgroup->hwif = hwif;
1077                 hwgroup->drive = drive;
1078                 drive->sleeping = 0;
1079                 drive->service_start = jiffies;
1080
1081                 if (blk_queue_plugged(drive->queue)) {
1082                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1083                         break;
1084                 }
1085
1086                 /*
1087                  * we know that the queue isn't empty, but this can happen
1088                  * if the q->prep_rq_fn() decides to kill a request
1089                  */
1090                 rq = elv_next_request(drive->queue);
1091                 if (!rq) {
1092                         hwgroup->busy = 0;
1093                         break;
1094                 }
1095
1096                 /*
1097                  * Sanity: don't accept a request that isn't a PM request
1098                  * if we are currently power managed. This is very important as
1099                  * blk_stop_queue() doesn't prevent the elv_next_request()
1100                  * above to return us whatever is in the queue. Since we call
1101                  * ide_do_request() ourselves, we end up taking requests while
1102                  * the queue is blocked...
1103                  * 
1104                  * We let requests forced at head of queue with ide-preempt
1105                  * though. I hope that doesn't happen too much, hopefully not
1106                  * unless the subdriver triggers such a thing in its own PM
1107                  * state machine.
1108                  *
1109                  * We count how many times we loop here to make sure we service
1110                  * all drives in the hwgroup without looping for ever
1111                  */
1112                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1113                         drive = drive->next ? drive->next : hwgroup->drive;
1114                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1115                                 goto again;
1116                         /* We clear busy, there should be no pending ATA command at this point. */
1117                         hwgroup->busy = 0;
1118                         break;
1119                 }
1120
1121                 hwgroup->rq = rq;
1122
1123                 /*
1124                  * Some systems have trouble with IDE IRQs arriving while
1125                  * the driver is still setting things up.  So, here we disable
1126                  * the IRQ used by this interface while the request is being started.
1127                  * This may look bad at first, but pretty much the same thing
1128                  * happens anyway when any interrupt comes in, IDE or otherwise
1129                  *  -- the kernel masks the IRQ while it is being handled.
1130                  */
1131                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1132                         disable_irq_nosync(hwif->irq);
1133                 spin_unlock(&ide_lock);
1134                 local_irq_enable_in_hardirq();
1135                         /* allow other IRQs while we start this request */
1136                 startstop = start_request(drive, rq);
1137                 spin_lock_irq(&ide_lock);
1138                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1139                         enable_irq(hwif->irq);
1140                 if (startstop == ide_stopped)
1141                         hwgroup->busy = 0;
1142         }
1143 }
1144
1145 /*
1146  * Passes the stuff to ide_do_request
1147  */
1148 void do_ide_request(struct request_queue *q)
1149 {
1150         ide_drive_t *drive = q->queuedata;
1151
1152         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1153 }
1154
1155 /*
1156  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1157  * retry the current request in pio mode instead of risking tossing it
1158  * all away
1159  */
1160 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1161 {
1162         ide_hwif_t *hwif = HWIF(drive);
1163         struct request *rq;
1164         ide_startstop_t ret = ide_stopped;
1165
1166         /*
1167          * end current dma transaction
1168          */
1169
1170         if (error < 0) {
1171                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1172                 (void)hwif->dma_ops->dma_end(drive);
1173                 ret = ide_error(drive, "dma timeout error",
1174                                 hwif->tp_ops->read_status(hwif));
1175         } else {
1176                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1177                 hwif->dma_ops->dma_timeout(drive);
1178         }
1179
1180         /*
1181          * disable dma for now, but remember that we did so because of
1182          * a timeout -- we'll reenable after we finish this next request
1183          * (or rather the first chunk of it) in pio.
1184          */
1185         drive->retry_pio++;
1186         drive->state = DMA_PIO_RETRY;
1187         ide_dma_off_quietly(drive);
1188
1189         /*
1190          * un-busy drive etc (hwgroup->busy is cleared on return) and
1191          * make sure request is sane
1192          */
1193         rq = HWGROUP(drive)->rq;
1194
1195         if (!rq)
1196                 goto out;
1197
1198         HWGROUP(drive)->rq = NULL;
1199
1200         rq->errors = 0;
1201
1202         if (!rq->bio)
1203                 goto out;
1204
1205         rq->sector = rq->bio->bi_sector;
1206         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1207         rq->hard_cur_sectors = rq->current_nr_sectors;
1208         rq->buffer = bio_data(rq->bio);
1209 out:
1210         return ret;
1211 }
1212
1213 /**
1214  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1215  *      @data: timer callback magic (hwgroup)
1216  *
1217  *      An IDE command has timed out before the expected drive return
1218  *      occurred. At this point we attempt to clean up the current
1219  *      mess. If the current handler includes an expiry handler then
1220  *      we invoke the expiry handler, and providing it is happy the
1221  *      work is done. If that fails we apply generic recovery rules
1222  *      invoking the handler and checking the drive DMA status. We
1223  *      have an excessively incestuous relationship with the DMA
1224  *      logic that wants cleaning up.
1225  */
1226  
1227 void ide_timer_expiry (unsigned long data)
1228 {
1229         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1230         ide_handler_t   *handler;
1231         ide_expiry_t    *expiry;
1232         unsigned long   flags;
1233         unsigned long   wait = -1;
1234
1235         spin_lock_irqsave(&ide_lock, flags);
1236
1237         if (((handler = hwgroup->handler) == NULL) ||
1238             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1239                 /*
1240                  * Either a marginal timeout occurred
1241                  * (got the interrupt just as timer expired),
1242                  * or we were "sleeping" to give other devices a chance.
1243                  * Either way, we don't really want to complain about anything.
1244                  */
1245                 if (hwgroup->sleeping) {
1246                         hwgroup->sleeping = 0;
1247                         hwgroup->busy = 0;
1248                 }
1249         } else {
1250                 ide_drive_t *drive = hwgroup->drive;
1251                 if (!drive) {
1252                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1253                         hwgroup->handler = NULL;
1254                 } else {
1255                         ide_hwif_t *hwif;
1256                         ide_startstop_t startstop = ide_stopped;
1257                         if (!hwgroup->busy) {
1258                                 hwgroup->busy = 1;      /* paranoia */
1259                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1260                         }
1261                         if ((expiry = hwgroup->expiry) != NULL) {
1262                                 /* continue */
1263                                 if ((wait = expiry(drive)) > 0) {
1264                                         /* reset timer */
1265                                         hwgroup->timer.expires  = jiffies + wait;
1266                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1267                                         add_timer(&hwgroup->timer);
1268                                         spin_unlock_irqrestore(&ide_lock, flags);
1269                                         return;
1270                                 }
1271                         }
1272                         hwgroup->handler = NULL;
1273                         /*
1274                          * We need to simulate a real interrupt when invoking
1275                          * the handler() function, which means we need to
1276                          * globally mask the specific IRQ:
1277                          */
1278                         spin_unlock(&ide_lock);
1279                         hwif  = HWIF(drive);
1280                         /* disable_irq_nosync ?? */
1281                         disable_irq(hwif->irq);
1282                         /* local CPU only,
1283                          * as if we were handling an interrupt */
1284                         local_irq_disable();
1285                         if (hwgroup->polling) {
1286                                 startstop = handler(drive);
1287                         } else if (drive_is_ready(drive)) {
1288                                 if (drive->waiting_for_dma)
1289                                         hwif->dma_ops->dma_lost_irq(drive);
1290                                 (void)ide_ack_intr(hwif);
1291                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1292                                 startstop = handler(drive);
1293                         } else {
1294                                 if (drive->waiting_for_dma) {
1295                                         startstop = ide_dma_timeout_retry(drive, wait);
1296                                 } else
1297                                         startstop =
1298                                         ide_error(drive, "irq timeout",
1299                                                   hwif->tp_ops->read_status(hwif));
1300                         }
1301                         drive->service_time = jiffies - drive->service_start;
1302                         spin_lock_irq(&ide_lock);
1303                         enable_irq(hwif->irq);
1304                         if (startstop == ide_stopped)
1305                                 hwgroup->busy = 0;
1306                 }
1307         }
1308         ide_do_request(hwgroup, IDE_NO_IRQ);
1309         spin_unlock_irqrestore(&ide_lock, flags);
1310 }
1311
1312 /**
1313  *      unexpected_intr         -       handle an unexpected IDE interrupt
1314  *      @irq: interrupt line
1315  *      @hwgroup: hwgroup being processed
1316  *
1317  *      There's nothing really useful we can do with an unexpected interrupt,
1318  *      other than reading the status register (to clear it), and logging it.
1319  *      There should be no way that an irq can happen before we're ready for it,
1320  *      so we needn't worry much about losing an "important" interrupt here.
1321  *
1322  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1323  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1324  *      looks "good", we just ignore the interrupt completely.
1325  *
1326  *      This routine assumes __cli() is in effect when called.
1327  *
1328  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1329  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1330  *      we could screw up by interfering with a new request being set up for 
1331  *      irq15.
1332  *
1333  *      In reality, this is a non-issue.  The new command is not sent unless 
1334  *      the drive is ready to accept one, in which case we know the drive is
1335  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1336  *      before completing the issuance of any new drive command, so we will not
1337  *      be accidentally invoked as a result of any valid command completion
1338  *      interrupt.
1339  *
1340  *      Note that we must walk the entire hwgroup here. We know which hwif
1341  *      is doing the current command, but we don't know which hwif burped
1342  *      mysteriously.
1343  */
1344  
1345 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1346 {
1347         u8 stat;
1348         ide_hwif_t *hwif = hwgroup->hwif;
1349
1350         /*
1351          * handle the unexpected interrupt
1352          */
1353         do {
1354                 if (hwif->irq == irq) {
1355                         stat = hwif->tp_ops->read_status(hwif);
1356
1357                         if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1358                                 /* Try to not flood the console with msgs */
1359                                 static unsigned long last_msgtime, count;
1360                                 ++count;
1361                                 if (time_after(jiffies, last_msgtime + HZ)) {
1362                                         last_msgtime = jiffies;
1363                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1364                                                 "status=0x%02x, count=%ld\n",
1365                                                 hwif->name,
1366                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1367                                 }
1368                         }
1369                 }
1370         } while ((hwif = hwif->next) != hwgroup->hwif);
1371 }
1372
1373 /**
1374  *      ide_intr        -       default IDE interrupt handler
1375  *      @irq: interrupt number
1376  *      @dev_id: hwif group
1377  *      @regs: unused weirdness from the kernel irq layer
1378  *
1379  *      This is the default IRQ handler for the IDE layer. You should
1380  *      not need to override it. If you do be aware it is subtle in
1381  *      places
1382  *
1383  *      hwgroup->hwif is the interface in the group currently performing
1384  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1385  *      the IRQ handler to call. As we issue a command the handlers
1386  *      step through multiple states, reassigning the handler to the
1387  *      next step in the process. Unlike a smart SCSI controller IDE
1388  *      expects the main processor to sequence the various transfer
1389  *      stages. We also manage a poll timer to catch up with most
1390  *      timeout situations. There are still a few where the handlers
1391  *      don't ever decide to give up.
1392  *
1393  *      The handler eventually returns ide_stopped to indicate the
1394  *      request completed. At this point we issue the next request
1395  *      on the hwgroup and the process begins again.
1396  */
1397  
1398 irqreturn_t ide_intr (int irq, void *dev_id)
1399 {
1400         unsigned long flags;
1401         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1402         ide_hwif_t *hwif;
1403         ide_drive_t *drive;
1404         ide_handler_t *handler;
1405         ide_startstop_t startstop;
1406
1407         spin_lock_irqsave(&ide_lock, flags);
1408         hwif = hwgroup->hwif;
1409
1410         if (!ide_ack_intr(hwif)) {
1411                 spin_unlock_irqrestore(&ide_lock, flags);
1412                 return IRQ_NONE;
1413         }
1414
1415         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1416                 /*
1417                  * Not expecting an interrupt from this drive.
1418                  * That means this could be:
1419                  *      (1) an interrupt from another PCI device
1420                  *      sharing the same PCI INT# as us.
1421                  * or   (2) a drive just entered sleep or standby mode,
1422                  *      and is interrupting to let us know.
1423                  * or   (3) a spurious interrupt of unknown origin.
1424                  *
1425                  * For PCI, we cannot tell the difference,
1426                  * so in that case we just ignore it and hope it goes away.
1427                  *
1428                  * FIXME: unexpected_intr should be hwif-> then we can
1429                  * remove all the ifdef PCI crap
1430                  */
1431 #ifdef CONFIG_BLK_DEV_IDEPCI
1432                 if (hwif->chipset != ide_pci)
1433 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1434                 {
1435                         /*
1436                          * Probably not a shared PCI interrupt,
1437                          * so we can safely try to do something about it:
1438                          */
1439                         unexpected_intr(irq, hwgroup);
1440 #ifdef CONFIG_BLK_DEV_IDEPCI
1441                 } else {
1442                         /*
1443                          * Whack the status register, just in case
1444                          * we have a leftover pending IRQ.
1445                          */
1446                         (void)hwif->tp_ops->read_status(hwif);
1447 #endif /* CONFIG_BLK_DEV_IDEPCI */
1448                 }
1449                 spin_unlock_irqrestore(&ide_lock, flags);
1450                 return IRQ_NONE;
1451         }
1452         drive = hwgroup->drive;
1453         if (!drive) {
1454                 /*
1455                  * This should NEVER happen, and there isn't much
1456                  * we could do about it here.
1457                  *
1458                  * [Note - this can occur if the drive is hot unplugged]
1459                  */
1460                 spin_unlock_irqrestore(&ide_lock, flags);
1461                 return IRQ_HANDLED;
1462         }
1463         if (!drive_is_ready(drive)) {
1464                 /*
1465                  * This happens regularly when we share a PCI IRQ with
1466                  * another device.  Unfortunately, it can also happen
1467                  * with some buggy drives that trigger the IRQ before
1468                  * their status register is up to date.  Hopefully we have
1469                  * enough advance overhead that the latter isn't a problem.
1470                  */
1471                 spin_unlock_irqrestore(&ide_lock, flags);
1472                 return IRQ_NONE;
1473         }
1474         if (!hwgroup->busy) {
1475                 hwgroup->busy = 1;      /* paranoia */
1476                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1477         }
1478         hwgroup->handler = NULL;
1479         hwgroup->req_gen++;
1480         del_timer(&hwgroup->timer);
1481         spin_unlock(&ide_lock);
1482
1483         /* Some controllers might set DMA INTR no matter DMA or PIO;
1484          * bmdma status might need to be cleared even for
1485          * PIO interrupts to prevent spurious/lost irq.
1486          */
1487         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1488                 /* ide_dma_end() needs bmdma status for error checking.
1489                  * So, skip clearing bmdma status here and leave it
1490                  * to ide_dma_end() if this is dma interrupt.
1491                  */
1492                 hwif->ide_dma_clear_irq(drive);
1493
1494         if (drive->unmask)
1495                 local_irq_enable_in_hardirq();
1496         /* service this interrupt, may set handler for next interrupt */
1497         startstop = handler(drive);
1498         spin_lock_irq(&ide_lock);
1499
1500         /*
1501          * Note that handler() may have set things up for another
1502          * interrupt to occur soon, but it cannot happen until
1503          * we exit from this routine, because it will be the
1504          * same irq as is currently being serviced here, and Linux
1505          * won't allow another of the same (on any CPU) until we return.
1506          */
1507         drive->service_time = jiffies - drive->service_start;
1508         if (startstop == ide_stopped) {
1509                 if (hwgroup->handler == NULL) { /* paranoia */
1510                         hwgroup->busy = 0;
1511                         ide_do_request(hwgroup, hwif->irq);
1512                 } else {
1513                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1514                                 "on exit\n", drive->name);
1515                 }
1516         }
1517         spin_unlock_irqrestore(&ide_lock, flags);
1518         return IRQ_HANDLED;
1519 }
1520
1521 /**
1522  *      ide_do_drive_cmd        -       issue IDE special command
1523  *      @drive: device to issue command
1524  *      @rq: request to issue
1525  *
1526  *      This function issues a special IDE device request
1527  *      onto the request queue.
1528  *
1529  *      the rq is queued at the head of the request queue, displacing
1530  *      the currently-being-processed request and this function
1531  *      returns immediately without waiting for the new rq to be
1532  *      completed.  This is VERY DANGEROUS, and is intended for
1533  *      careful use by the ATAPI tape/cdrom driver code.
1534  */
1535
1536 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1537 {
1538         unsigned long flags;
1539         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1540
1541         spin_lock_irqsave(&ide_lock, flags);
1542         hwgroup->rq = NULL;
1543         __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1544         __generic_unplug_device(drive->queue);
1545         spin_unlock_irqrestore(&ide_lock, flags);
1546 }
1547
1548 EXPORT_SYMBOL(ide_do_drive_cmd);
1549
1550 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1551 {
1552         ide_hwif_t *hwif = drive->hwif;
1553         ide_task_t task;
1554
1555         memset(&task, 0, sizeof(task));
1556         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1557                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1558         task.tf.feature = dma;          /* Use PIO/DMA */
1559         task.tf.lbam    = bcount & 0xff;
1560         task.tf.lbah    = (bcount >> 8) & 0xff;
1561
1562         ide_tf_dump(drive->name, &task.tf);
1563         hwif->tp_ops->set_irq(hwif, 1);
1564         SELECT_MASK(drive, 0);
1565         hwif->tp_ops->tf_load(drive, &task);
1566 }
1567
1568 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1569
1570 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1571 {
1572         ide_hwif_t *hwif = drive->hwif;
1573         u8 buf[4] = { 0 };
1574
1575         while (len > 0) {
1576                 if (write)
1577                         hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1578                 else
1579                         hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1580                 len -= 4;
1581         }
1582 }
1583 EXPORT_SYMBOL_GPL(ide_pad_transfer);