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/config.h>
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/string.h>
  31 #include <linux/kernel.h>
  32 #include <linux/timer.h>
  33 #include <linux/mm.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/major.h>
  36 #include <linux/errno.h>
  37 #include <linux/genhd.h>
  38 #include <linux/blkpg.h>
  39 #include <linux/slab.h>
  40 #include <linux/init.h>
  41 #include <linux/pci.h>
  42 #include <linux/delay.h>
  43 #include <linux/ide.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
  52 #include <asm/byteorder.h>
  53 #include <asm/irq.h>
  54 #include <asm/uaccess.h>
  55 #include <asm/io.h>
  56 #include <asm/bitops.h>
  57
  58 static void ide_fill_flush_cmd(ide_drive_t *drive, struct request *rq)
  59 {
  60         char *buf = rq->cmd;
  61
  62         /*
  63          * reuse cdb space for ata command
  64          */
  65         memset(buf, 0, sizeof(rq->cmd));
  66
  67         rq->flags |= REQ_DRIVE_TASK | REQ_STARTED;
  68         rq->buffer = buf;
  69         rq->buffer[0] = WIN_FLUSH_CACHE;
  70
  71         if (ide_id_has_flush_cache_ext(drive->id) &&
  72             (drive->capacity64 >= (1UL << 28)))
  73                 rq->buffer[0] = WIN_FLUSH_CACHE_EXT;
  74 }
  75
  76 /*
  77  * preempt pending requests, and store this cache flush for immediate
  78  * execution
  79  */
  80 static struct request *ide_queue_flush_cmd(ide_drive_t *drive,
  81                                            struct request *rq, int post)
  82 {
  83         struct request *flush_rq = &HWGROUP(drive)->wrq;
  84
  85         /*
  86          * write cache disabled, clear the barrier bit and treat it like
  87          * an ordinary write
  88          */
  89         if (!drive->wcache) {
  90                 rq->flags |= REQ_BAR_PREFLUSH;
  91                 return rq;
  92         }
  93
  94         ide_init_drive_cmd(flush_rq);
  95         ide_fill_flush_cmd(drive, flush_rq);
  96
  97         flush_rq->special = rq;
  98         flush_rq->nr_sectors = rq->nr_sectors;
  99
 100         if (!post) {
 101                 drive->doing_barrier = 1;
 102                 flush_rq->flags |= REQ_BAR_PREFLUSH;
 103                 blkdev_dequeue_request(rq);
 104         } else
 105                 flush_rq->flags |= REQ_BAR_POSTFLUSH;
 106
 107         __elv_add_request(drive->queue, flush_rq, ELEVATOR_INSERT_FRONT, 0);
 108         HWGROUP(drive)->rq = NULL;
 109         return flush_rq;
 110 }
 111
 112 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
 113                              int uptodate, int nr_sectors)
 114 {
 115         int ret = 1;
 116
 117         BUG_ON(!(rq->flags & REQ_STARTED));
 118
 119         /*
 120          * if failfast is set on a request, override number of sectors and
 121          * complete the whole request right now
 122          */
 123         if (blk_noretry_request(rq) && end_io_error(uptodate))
 124                 nr_sectors = rq->hard_nr_sectors;
 125
 126         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
 127                 rq->errors = -EIO;
 128
 129         /*
 130          * decide whether to reenable DMA -- 3 is a random magic for now,
 131          * if we DMA timeout more than 3 times, just stay in PIO
 132          */
 133         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
 134                 drive->state = 0;
 135                 HWGROUP(drive)->hwif->ide_dma_on(drive);
 136         }
 137
 138         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
 139                 add_disk_randomness(rq->rq_disk);
 140
 141                 if (blk_rq_tagged(rq))
 142                         blk_queue_end_tag(drive->queue, rq);
 143
 144                 blkdev_dequeue_request(rq);
 145                 HWGROUP(drive)->rq = NULL;
 146                 end_that_request_last(rq);
 147                 ret = 0;
 148         }
 149         return ret;
 150 }
 151
 152 /**
 153  *      ide_end_request         -       complete an IDE I/O
 154  *      @drive: IDE device for the I/O
 155  *      @uptodate:
 156  *      @nr_sectors: number of sectors completed
 157  *
 158  *      This is our end_request wrapper function. We complete the I/O
 159  *      update random number input and dequeue the request, which if
 160  *      it was tagged may be out of order.
 161  */
 162
 163 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
 164 {
 165         struct request *rq;
 166         unsigned long flags;
 167         int ret = 1;
 168
 169         spin_lock_irqsave(&ide_lock, flags);
 170         rq = HWGROUP(drive)->rq;
 171
 172         if (!nr_sectors)
 173                 nr_sectors = rq->hard_cur_sectors;
 174
 175         if (!blk_barrier_rq(rq) || !drive->wcache)
 176                 ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
 177         else {
 178                 struct request *flush_rq = &HWGROUP(drive)->wrq;
 179
 180                 flush_rq->nr_sectors -= nr_sectors;
 181                 if (!flush_rq->nr_sectors) {
 182                         ide_queue_flush_cmd(drive, rq, 1);
 183                         ret = 0;
 184                 }
 185         }
 186
 187         spin_unlock_irqrestore(&ide_lock, flags);
 188         return ret;
 189 }
 190 EXPORT_SYMBOL(ide_end_request);
 191
 192 /**
 193  *      ide_complete_pm_request - end the current Power Management request
 194  *      @drive: target drive
 195  *      @rq: request
 196  *
 197  *      This function cleans up the current PM request and stops the queue
 198  *      if necessary.
 199  */
 200 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
 201 {
 202         unsigned long flags;
 203
 204 #ifdef DEBUG_PM
 205         printk("%s: completing PM request, %s\n", drive->name,
 206                blk_pm_suspend_request(rq) ? "suspend" : "resume");
 207 #endif
 208         spin_lock_irqsave(&ide_lock, flags);
 209         if (blk_pm_suspend_request(rq)) {
 210                 blk_stop_queue(drive->queue);
 211         } else {
 212                 drive->blocked = 0;
 213                 blk_start_queue(drive->queue);
 214         }
 215         blkdev_dequeue_request(rq);
 216         HWGROUP(drive)->rq = NULL;
 217         end_that_request_last(rq);
 218         spin_unlock_irqrestore(&ide_lock, flags);
 219 }
 220
 221 /*
 222  * FIXME: probably move this somewhere else, name is bad too :)
 223  */
 224 u64 ide_get_error_location(ide_drive_t *drive, char *args)
 225 {
 226         u32 high, low;
 227         u8 hcyl, lcyl, sect;
 228         u64 sector;
 229
 230         high = 0;
 231         hcyl = args[5];
 232         lcyl = args[4];
 233         sect = args[3];
 234
 235         if (ide_id_has_flush_cache_ext(drive->id)) {
 236                 low = (hcyl << 16) | (lcyl << 8) | sect;
 237                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 238                 high = ide_read_24(drive);
 239         } else {
 240                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
 241                 if (cur & 0x40) {
 242                         high = cur & 0xf;
 243                         low = (hcyl << 16) | (lcyl << 8) | sect;
 244                 } else {
 245                         low = hcyl * drive->head * drive->sect;
 246                         low += lcyl * drive->sect;
 247                         low += sect - 1;
 248                 }
 249         }
 250
 251         sector = ((u64) high << 24) | low;
 252         return sector;
 253 }
 254 EXPORT_SYMBOL(ide_get_error_location);
 255
 256 static void ide_complete_barrier(ide_drive_t *drive, struct request *rq,
 257                                  int error)
 258 {
 259         struct request *real_rq = rq->special;
 260         int good_sectors, bad_sectors;
 261         sector_t sector;
 262
 263         if (!error) {
 264                 if (blk_barrier_postflush(rq)) {
 265                         /*
 266                          * this completes the barrier write
 267                          */
 268                         __ide_end_request(drive, real_rq, 1, real_rq->hard_nr_sectors);
 269                         drive->doing_barrier = 0;
 270                 } else {
 271                         /*
 272                          * just indicate that we did the pre flush
 273                          */
 274                         real_rq->flags |= REQ_BAR_PREFLUSH;
 275                         elv_requeue_request(drive->queue, real_rq);
 276                 }
 277                 /*
 278                  * all is fine, return
 279                  */
 280                 return;
 281         }
 282
 283         /*
 284          * we need to end real_rq, but it's not on the queue currently.
 285          * put it back on the queue, so we don't have to special case
 286          * anything else for completing it
 287          */
 288         if (!blk_barrier_postflush(rq))
 289                 elv_requeue_request(drive->queue, real_rq);
 290
 291         /*
 292          * drive aborted flush command, assume FLUSH_CACHE_* doesn't
 293          * work and disable barrier support
 294          */
 295         if (error & ABRT_ERR) {
 296                 printk(KERN_ERR "%s: barrier support doesn't work\n", drive->name);
 297                 __ide_end_request(drive, real_rq, -EOPNOTSUPP, real_rq->hard_nr_sectors);
 298                 blk_queue_ordered(drive->queue, 0);
 299                 blk_queue_issue_flush_fn(drive->queue, NULL);
 300         } else {
 301                 /*
 302                  * find out what part of the request failed
 303                  */
 304                 good_sectors = 0;
 305                 if (blk_barrier_postflush(rq)) {
 306                         sector = ide_get_error_location(drive, rq->buffer);
 307
 308                         if ((sector >= real_rq->hard_sector) &&
 309                             (sector < real_rq->hard_sector + real_rq->hard_nr_sectors))
 310                                 good_sectors = sector - real_rq->hard_sector;
 311                 } else
 312                         sector = real_rq->hard_sector;
 313
 314                 bad_sectors = real_rq->hard_nr_sectors - good_sectors;
 315                 if (good_sectors)
 316                         __ide_end_request(drive, real_rq, 1, good_sectors);
 317                 if (bad_sectors)
 318                         __ide_end_request(drive, real_rq, 0, bad_sectors);
 319
 320                 printk(KERN_ERR "%s: failed barrier write: "
 321                                 "sector=%Lx(good=%d/bad=%d)\n",
 322                                 drive->name, (unsigned long long)sector,
 323                                 good_sectors, bad_sectors);
 324         }
 325
 326         drive->doing_barrier = 0;
 327 }
 328
 329 /**
 330  *      ide_end_drive_cmd       -       end an explicit drive command
 331  *      @drive: command
 332  *      @stat: status bits
 333  *      @err: error bits
 334  *
 335  *      Clean up after success/failure of an explicit drive command.
 336  *      These get thrown onto the queue so they are synchronized with
 337  *      real I/O operations on the drive.
 338  *
 339  *      In LBA48 mode we have to read the register set twice to get
 340  *      all the extra information out.
 341  */
 342
 343 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
 344 {
 345         ide_hwif_t *hwif = HWIF(drive);
 346         unsigned long flags;
 347         struct request *rq;
 348
 349         spin_lock_irqsave(&ide_lock, flags);
 350         rq = HWGROUP(drive)->rq;
 351         spin_unlock_irqrestore(&ide_lock, flags);
 352
 353         if (rq->flags & REQ_DRIVE_CMD) {
 354                 u8 *args = (u8 *) rq->buffer;
 355                 if (rq->errors == 0)
 356                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 357
 358                 if (args) {
 359                         args[0] = stat;
 360                         args[1] = err;
 361                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 362                 }
 363         } else if (rq->flags & REQ_DRIVE_TASK) {
 364                 u8 *args = (u8 *) rq->buffer;
 365                 if (rq->errors == 0)
 366                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 367
 368                 if (args) {
 369                         args[0] = stat;
 370                         args[1] = err;
 371                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 372                         args[3] = hwif->INB(IDE_SECTOR_REG);
 373                         args[4] = hwif->INB(IDE_LCYL_REG);
 374                         args[5] = hwif->INB(IDE_HCYL_REG);
 375                         args[6] = hwif->INB(IDE_SELECT_REG);
 376                 }
 377         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
 378                 ide_task_t *args = (ide_task_t *) rq->special;
 379                 if (rq->errors == 0)
 380                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 381
 382                 if (args) {
 383                         if (args->tf_in_flags.b.data) {
 384                                 u16 data                                = hwif->INW(IDE_DATA_REG);
 385                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
 386                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
 387                         }
 388                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
 389                         /* be sure we're looking at the low order bits */
 390                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
 391                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
 392                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
 393                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
 394                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
 395                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
 396                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
 397
 398                         if (drive->addressing == 1) {
 399                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 400                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
 401                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
 402                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
 403                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
 404                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
 405                         }
 406                 }
 407         } else if (blk_pm_request(rq)) {
 408 #ifdef DEBUG_PM
 409                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
 410                         drive->name, rq->pm->pm_step, stat, err);
 411 #endif
 412                 DRIVER(drive)->complete_power_step(drive, rq, stat, err);
 413                 if (rq->pm->pm_step == ide_pm_state_completed)
 414                         ide_complete_pm_request(drive, rq);
 415                 return;
 416         }
 417
 418         spin_lock_irqsave(&ide_lock, flags);
 419         blkdev_dequeue_request(rq);
 420
 421         if (blk_barrier_preflush(rq) || blk_barrier_postflush(rq))
 422                 ide_complete_barrier(drive, rq, err);
 423
 424         HWGROUP(drive)->rq = NULL;
 425         end_that_request_last(rq);
 426         spin_unlock_irqrestore(&ide_lock, flags);
 427 }
 428
 429 EXPORT_SYMBOL(ide_end_drive_cmd);
 430
 431 /**
 432  *      try_to_flush_leftover_data      -       flush junk
 433  *      @drive: drive to flush
 434  *
 435  *      try_to_flush_leftover_data() is invoked in response to a drive
 436  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
 437  *      resetting the drive, this routine tries to clear the condition
 438  *      by read a sector's worth of data from the drive.  Of course,
 439  *      this may not help if the drive is *waiting* for data from *us*.
 440  */
 441 static void try_to_flush_leftover_data (ide_drive_t *drive)
 442 {
 443         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
 444
 445         if (drive->media != ide_disk)
 446                 return;
 447         while (i > 0) {
 448                 u32 buffer[16];
 449                 u32 wcount = (i > 16) ? 16 : i;
 450
 451                 i -= wcount;
 452                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
 453         }
 454 }
 455
 456 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 457 {
 458         ide_hwif_t *hwif = drive->hwif;
 459
 460         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 461                 /* other bits are useless when BUSY */
 462                 rq->errors |= ERROR_RESET;
 463         } else if (stat & ERR_STAT) {
 464                 /* err has different meaning on cdrom and tape */
 465                 if (err == ABRT_ERR) {
 466                         if (drive->select.b.lba &&
 467                             /* some newer drives don't support WIN_SPECIFY */
 468                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
 469                                 return ide_stopped;
 470                 } else if ((err & BAD_CRC) == BAD_CRC) {
 471                         /* UDMA crc error, just retry the operation */
 472                         drive->crc_count++;
 473                 } else if (err & (BBD_ERR | ECC_ERR)) {
 474                         /* retries won't help these */
 475                         rq->errors = ERROR_MAX;
 476                 } else if (err & TRK0_ERR) {
 477                         /* help it find track zero */
 478                         rq->errors |= ERROR_RECAL;
 479                 }
 480         }
 481
 482         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ)
 483                 try_to_flush_leftover_data(drive);
 484
 485         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 486                 /* force an abort */
 487                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 488
 489         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
 490                 drive->driver->end_request(drive, 0, 0);
 491         else {
 492                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 493                         ++rq->errors;
 494                         return ide_do_reset(drive);
 495                 }
 496                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
 497                         drive->special.b.recalibrate = 1;
 498                 ++rq->errors;
 499         }
 500         return ide_stopped;
 501 }
 502
 503 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 504 {
 505         ide_hwif_t *hwif = drive->hwif;
 506
 507         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 508                 /* other bits are useless when BUSY */
 509                 rq->errors |= ERROR_RESET;
 510         } else {
 511                 /* add decoding error stuff */
 512         }
 513
 514         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 515                 /* force an abort */
 516                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 517
 518         if (rq->errors >= ERROR_MAX) {
 519                 drive->driver->end_request(drive, 0, 0);
 520         } else {
 521                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 522                         ++rq->errors;
 523                         return ide_do_reset(drive);
 524                 }
 525                 ++rq->errors;
 526         }
 527
 528         return ide_stopped;
 529 }
 530
 531 ide_startstop_t
 532 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 533 {
 534         if (drive->media == ide_disk)
 535                 return ide_ata_error(drive, rq, stat, err);
 536         return ide_atapi_error(drive, rq, stat, err);
 537 }
 538
 539 /**
 540  *      ide_error       -       handle an error on the IDE
 541  *      @drive: drive the error occurred on
 542  *      @msg: message to report
 543  *      @stat: status bits
 544  *
 545  *      ide_error() takes action based on the error returned by the drive.
 546  *      For normal I/O that may well include retries. We deal with
 547  *      both new-style (taskfile) and old style command handling here.
 548  *      In the case of taskfile command handling there is work left to
 549  *      do
 550  */
 551
 552 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
 553 {
 554         struct request *rq;
 555         u8 err;
 556
 557         err = ide_dump_status(drive, msg, stat);
 558
 559         if ((rq = HWGROUP(drive)->rq) == NULL)
 560                 return ide_stopped;
 561
 562         /* retry only "normal" I/O: */
 563         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 564                 rq->errors = 1;
 565                 ide_end_drive_cmd(drive, stat, err);
 566                 return ide_stopped;
 567         }
 568
 569         return drive->driver->error(drive, rq, stat, err);
 570 }
 571
 572 EXPORT_SYMBOL_GPL(ide_error);
 573
 574 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
 575 {
 576         if (drive->media != ide_disk)
 577                 rq->errors |= ERROR_RESET;
 578
 579         DRIVER(drive)->end_request(drive, 0, 0);
 580         return ide_stopped;
 581 }
 582
 583 /**
 584  *      ide_abort       -       abort pending IDE operatins
 585  *      @drive: drive the error occurred on
 586  *      @msg: message to report
 587  *
 588  *      ide_abort kills and cleans up when we are about to do a
 589  *      host initiated reset on active commands. Longer term we
 590  *      want handlers to have sensible abort handling themselves
 591  *
 592  *      This differs fundamentally from ide_error because in
 593  *      this case the command is doing just fine when we
 594  *      blow it away.
 595  */
 596
 597 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
 598 {
 599         struct request *rq;
 600
 601         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
 602                 return ide_stopped;
 603
 604         /* retry only "normal" I/O: */
 605         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 606                 rq->errors = 1;
 607                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
 608                 return ide_stopped;
 609         }
 610
 611         return drive->driver->abort(drive, rq);
 612 }
 613
 614 /**
 615  *      ide_cmd         -       issue a simple drive command
 616  *      @drive: drive the command is for
 617  *      @cmd: command byte
 618  *      @nsect: sector byte
 619  *      @handler: handler for the command completion
 620  *
 621  *      Issue a simple drive command with interrupts.
 622  *      The drive must be selected beforehand.
 623  */
 624
 625 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
 626                 ide_handler_t *handler)
 627 {
 628         ide_hwif_t *hwif = HWIF(drive);
 629         if (IDE_CONTROL_REG)
 630                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
 631         SELECT_MASK(drive,0);
 632         hwif->OUTB(nsect,IDE_NSECTOR_REG);
 633         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
 634 }
 635
 636 /**
 637  *      drive_cmd_intr          -       drive command completion interrupt
 638  *      @drive: drive the completion interrupt occurred on
 639  *
 640  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
 641  *      We do any necessary daya reading and then wait for the drive to
 642  *      go non busy. At that point we may read the error data and complete
 643  *      the request
 644  */
 645
 646 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
 647 {
 648         struct request *rq = HWGROUP(drive)->rq;
 649         ide_hwif_t *hwif = HWIF(drive);
 650         u8 *args = (u8 *) rq->buffer;
 651         u8 stat = hwif->INB(IDE_STATUS_REG);
 652         int retries = 10;
 653
 654         local_irq_enable();
 655         if ((stat & DRQ_STAT) && args && args[3]) {
 656                 u8 io_32bit = drive->io_32bit;
 657                 drive->io_32bit = 0;
 658                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
 659                 drive->io_32bit = io_32bit;
 660                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
 661                         udelay(100);
 662         }
 663
 664         if (!OK_STAT(stat, READY_STAT, BAD_STAT) && DRIVER(drive) != NULL)
 665                 return ide_error(drive, "drive_cmd", stat);
 666                 /* calls ide_end_drive_cmd */
 667         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
 668         return ide_stopped;
 669 }
 670
 671 /**
 672  *      do_special              -       issue some special commands
 673  *      @drive: drive the command is for
 674  *
 675  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
 676  *      commands to a drive.  It used to do much more, but has been scaled
 677  *      back.
 678  */
 679
 680 static ide_startstop_t do_special (ide_drive_t *drive)
 681 {
 682         special_t *s = &drive->special;
 683
 684 #ifdef DEBUG
 685         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
 686 #endif
 687         if (s->b.set_tune) {
 688                 s->b.set_tune = 0;
 689                 if (HWIF(drive)->tuneproc != NULL)
 690                         HWIF(drive)->tuneproc(drive, drive->tune_req);
 691                 return ide_stopped;
 692         }
 693         else
 694                 return DRIVER(drive)->special(drive);
 695 }
 696
 697 void ide_map_sg(ide_drive_t *drive, struct request *rq)
 698 {
 699         ide_hwif_t *hwif = drive->hwif;
 700         struct scatterlist *sg = hwif->sg_table;
 701
 702         if (hwif->sg_mapped)    /* needed by ide-scsi */
 703                 return;
 704
 705         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
 706                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
 707         } else {
 708                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
 709                 hwif->sg_nents = 1;
 710         }
 711 }
 712
 713 EXPORT_SYMBOL_GPL(ide_map_sg);
 714
 715 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
 716 {
 717         ide_hwif_t *hwif = drive->hwif;
 718
 719         hwif->nsect = hwif->nleft = rq->nr_sectors;
 720         hwif->cursg = hwif->cursg_ofs = 0;
 721 }
 722
 723 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
 724
 725 /**
 726  *      execute_drive_command   -       issue special drive command
 727  *      @drive: the drive to issue th command on
 728  *      @rq: the request structure holding the command
 729  *
 730  *      execute_drive_cmd() issues a special drive command,  usually
 731  *      initiated by ioctl() from the external hdparm program. The
 732  *      command can be a drive command, drive task or taskfile
 733  *      operation. Weirdly you can call it with NULL to wait for
 734  *      all commands to finish. Don't do this as that is due to change
 735  */
 736
 737 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
 738                 struct request *rq)
 739 {
 740         ide_hwif_t *hwif = HWIF(drive);
 741         if (rq->flags & REQ_DRIVE_TASKFILE) {
 742                 ide_task_t *args = rq->special;
 743
 744                 if (!args)
 745                         goto done;
 746
 747                 hwif->data_phase = args->data_phase;
 748
 749                 switch (hwif->data_phase) {
 750                 case TASKFILE_MULTI_OUT:
 751                 case TASKFILE_OUT:
 752                 case TASKFILE_MULTI_IN:
 753                 case TASKFILE_IN:
 754                         ide_init_sg_cmd(drive, rq);
 755                         ide_map_sg(drive, rq);
 756                 default:
 757                         break;
 758                 }
 759
 760                 if (args->tf_out_flags.all != 0)
 761                         return flagged_taskfile(drive, args);
 762                 return do_rw_taskfile(drive, args);
 763         } else if (rq->flags & REQ_DRIVE_TASK) {
 764                 u8 *args = rq->buffer;
 765                 u8 sel;
 766
 767                 if (!args)
 768                         goto done;
 769 #ifdef DEBUG
 770                 printk("%s: DRIVE_TASK_CMD ", drive->name);
 771                 printk("cmd=0x%02x ", args[0]);
 772                 printk("fr=0x%02x ", args[1]);
 773                 printk("ns=0x%02x ", args[2]);
 774                 printk("sc=0x%02x ", args[3]);
 775                 printk("lcyl=0x%02x ", args[4]);
 776                 printk("hcyl=0x%02x ", args[5]);
 777                 printk("sel=0x%02x\n", args[6]);
 778 #endif
 779                 hwif->OUTB(args[1], IDE_FEATURE_REG);
 780                 hwif->OUTB(args[3], IDE_SECTOR_REG);
 781                 hwif->OUTB(args[4], IDE_LCYL_REG);
 782                 hwif->OUTB(args[5], IDE_HCYL_REG);
 783                 sel = (args[6] & ~0x10);
 784                 if (drive->select.b.unit)
 785                         sel |= 0x10;
 786                 hwif->OUTB(sel, IDE_SELECT_REG);
 787                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
 788                 return ide_started;
 789         } else if (rq->flags & REQ_DRIVE_CMD) {
 790                 u8 *args = rq->buffer;
 791
 792                 if (!args)
 793                         goto done;
 794 #ifdef DEBUG
 795                 printk("%s: DRIVE_CMD ", drive->name);
 796                 printk("cmd=0x%02x ", args[0]);
 797                 printk("sc=0x%02x ", args[1]);
 798                 printk("fr=0x%02x ", args[2]);
 799                 printk("xx=0x%02x\n", args[3]);
 800 #endif
 801                 if (args[0] == WIN_SMART) {
 802                         hwif->OUTB(0x4f, IDE_LCYL_REG);
 803                         hwif->OUTB(0xc2, IDE_HCYL_REG);
 804                         hwif->OUTB(args[2],IDE_FEATURE_REG);
 805                         hwif->OUTB(args[1],IDE_SECTOR_REG);
 806                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
 807                         return ide_started;
 808                 }
 809                 hwif->OUTB(args[2],IDE_FEATURE_REG);
 810                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
 811                 return ide_started;
 812         }
 813
 814 done:
 815         /*
 816          * NULL is actually a valid way of waiting for
 817          * all current requests to be flushed from the queue.
 818          */
 819 #ifdef DEBUG
 820         printk("%s: DRIVE_CMD (null)\n", drive->name);
 821 #endif
 822         ide_end_drive_cmd(drive,
 823                         hwif->INB(IDE_STATUS_REG),
 824                         hwif->INB(IDE_ERROR_REG));
 825         return ide_stopped;
 826 }
 827
 828 /**
 829  *      start_request   -       start of I/O and command issuing for IDE
 830  *
 831  *      start_request() initiates handling of a new I/O request. It
 832  *      accepts commands and I/O (read/write) requests. It also does
 833  *      the final remapping for weird stuff like EZDrive. Once
 834  *      device mapper can work sector level the EZDrive stuff can go away
 835  *
 836  *      FIXME: this function needs a rename
 837  */
 838
 839 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
 840 {
 841         ide_startstop_t startstop;
 842         sector_t block;
 843
 844         BUG_ON(!(rq->flags & REQ_STARTED));
 845
 846 #ifdef DEBUG
 847         printk("%s: start_request: current=0x%08lx\n",
 848                 HWIF(drive)->name, (unsigned long) rq);
 849 #endif
 850
 851         /* bail early if we've exceeded max_failures */
 852         if (drive->max_failures && (drive->failures > drive->max_failures)) {
 853                 goto kill_rq;
 854         }
 855
 856         block    = rq->sector;
 857         if (blk_fs_request(rq) &&
 858             (drive->media == ide_disk || drive->media == ide_floppy)) {
 859                 block += drive->sect0;
 860         }
 861         /* Yecch - this will shift the entire interval,
 862            possibly killing some innocent following sector */
 863         if (block == 0 && drive->remap_0_to_1 == 1)
 864                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
 865
 866         if (blk_pm_suspend_request(rq) &&
 867             rq->pm->pm_step == ide_pm_state_start_suspend)
 868                 /* Mark drive blocked when starting the suspend sequence. */
 869                 drive->blocked = 1;
 870         else if (blk_pm_resume_request(rq) &&
 871                  rq->pm->pm_step == ide_pm_state_start_resume) {
 872                 /*
 873                  * The first thing we do on wakeup is to wait for BSY bit to
 874                  * go away (with a looong timeout) as a drive on this hwif may
 875                  * just be POSTing itself.
 876                  * We do that before even selecting as the "other" device on
 877                  * the bus may be broken enough to walk on our toes at this
 878                  * point.
 879                  */
 880                 int rc;
 881 #ifdef DEBUG_PM
 882                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
 883 #endif
 884                 rc = ide_wait_not_busy(HWIF(drive), 35000);
 885                 if (rc)
 886                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
 887                 SELECT_DRIVE(drive);
 888                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
 889                 rc = ide_wait_not_busy(HWIF(drive), 10000);
 890                 if (rc)
 891                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
 892         }
 893
 894         SELECT_DRIVE(drive);
 895         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
 896                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
 897                 return startstop;
 898         }
 899         if (!drive->special.all) {
 900                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
 901                         return execute_drive_cmd(drive, rq);
 902                 else if (rq->flags & REQ_DRIVE_TASKFILE)
 903                         return execute_drive_cmd(drive, rq);
 904                 else if (blk_pm_request(rq)) {
 905 #ifdef DEBUG_PM
 906                         printk("%s: start_power_step(step: %d)\n",
 907                                 drive->name, rq->pm->pm_step);
 908 #endif
 909                         startstop = DRIVER(drive)->start_power_step(drive, rq);
 910                         if (startstop == ide_stopped &&
 911                             rq->pm->pm_step == ide_pm_state_completed)
 912                                 ide_complete_pm_request(drive, rq);
 913                         return startstop;
 914                 }
 915                 return (DRIVER(drive)->do_request(drive, rq, block));
 916         }
 917         return do_special(drive);
 918 kill_rq:
 919         DRIVER(drive)->end_request(drive, 0, 0);
 920         return ide_stopped;
 921 }
 922
 923 /**
 924  *      ide_stall_queue         -       pause an IDE device
 925  *      @drive: drive to stall
 926  *      @timeout: time to stall for (jiffies)
 927  *
 928  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
 929  *      to the hwgroup by sleeping for timeout jiffies.
 930  */
 931
 932 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
 933 {
 934         if (timeout > WAIT_WORSTCASE)
 935                 timeout = WAIT_WORSTCASE;
 936         drive->sleep = timeout + jiffies;
 937         drive->sleeping = 1;
 938 }
 939
 940 EXPORT_SYMBOL(ide_stall_queue);
 941
 942 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
 943
 944 /**
 945  *      choose_drive            -       select a drive to service
 946  *      @hwgroup: hardware group to select on
 947  *
 948  *      choose_drive() selects the next drive which will be serviced.
 949  *      This is necessary because the IDE layer can't issue commands
 950  *      to both drives on the same cable, unlike SCSI.
 951  */
 952
 953 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
 954 {
 955         ide_drive_t *drive, *best;
 956
 957 repeat:
 958         best = NULL;
 959         drive = hwgroup->drive;
 960
 961         /*
 962          * drive is doing pre-flush, ordered write, post-flush sequence. even
 963          * though that is 3 requests, it must be seen as a single transaction.
 964          * we must not preempt this drive until that is complete
 965          */
 966         if (drive->doing_barrier) {
 967                 /*
 968                  * small race where queue could get replugged during
 969                  * the 3-request flush cycle, just yank the plug since
 970                  * we want it to finish asap
 971                  */
 972                 blk_remove_plug(drive->queue);
 973                 return drive;
 974         }
 975
 976         do {
 977                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
 978                     && !elv_queue_empty(drive->queue)) {
 979                         if (!best
 980                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
 981                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
 982                         {
 983                                 if (!blk_queue_plugged(drive->queue))
 984                                         best = drive;
 985                         }
 986                 }
 987         } while ((drive = drive->next) != hwgroup->drive);
 988         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
 989                 long t = (signed long)(WAKEUP(best) - jiffies);
 990                 if (t >= WAIT_MIN_SLEEP) {
 991                 /*
 992                  * We *may* have some time to spare, but first let's see if
 993                  * someone can potentially benefit from our nice mood today..
 994                  */
 995                         drive = best->next;
 996                         do {
 997                                 if (!drive->sleeping
 998                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
 999                                  && time_before(WAKEUP(drive), jiffies + t))
1000                                 {
1001                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1002                                         goto repeat;
1003                                 }
1004                         } while ((drive = drive->next) != best);
1005                 }
1006         }
1007         return best;
1008 }
1009
1010 /*
1011  * Issue a new request to a drive from hwgroup
1012  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1013  *
1014  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1015  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1016  * may have both interfaces in a single hwgroup to "serialize" access.
1017  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1018  * together into one hwgroup for serialized access.
1019  *
1020  * Note also that several hwgroups can end up sharing a single IRQ,
1021  * possibly along with many other devices.  This is especially common in
1022  * PCI-based systems with off-board IDE controller cards.
1023  *
1024  * The IDE driver uses the single global ide_lock spinlock to protect
1025  * access to the request queues, and to protect the hwgroup->busy flag.
1026  *
1027  * The first thread into the driver for a particular hwgroup sets the
1028  * hwgroup->busy flag to indicate that this hwgroup is now active,
1029  * and then initiates processing of the top request from the request queue.
1030  *
1031  * Other threads attempting entry notice the busy setting, and will simply
1032  * queue their new requests and exit immediately.  Note that hwgroup->busy
1033  * remains set even when the driver is merely awaiting the next interrupt.
1034  * Thus, the meaning is "this hwgroup is busy processing a request".
1035  *
1036  * When processing of a request completes, the completing thread or IRQ-handler
1037  * will start the next request from the queue.  If no more work remains,
1038  * the driver will clear the hwgroup->busy flag and exit.
1039  *
1040  * The ide_lock (spinlock) is used to protect all access to the
1041  * hwgroup->busy flag, but is otherwise not needed for most processing in
1042  * the driver.  This makes the driver much more friendlier to shared IRQs
1043  * than previous designs, while remaining 100% (?) SMP safe and capable.
1044  */
1045 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1046 {
1047         ide_drive_t     *drive;
1048         ide_hwif_t      *hwif;
1049         struct request  *rq;
1050         ide_startstop_t startstop;
1051
1052         /* for atari only: POSSIBLY BROKEN HERE(?) */
1053         ide_get_lock(ide_intr, hwgroup);
1054
1055         /* caller must own ide_lock */
1056         BUG_ON(!irqs_disabled());
1057
1058         while (!hwgroup->busy) {
1059                 hwgroup->busy = 1;
1060                 drive = choose_drive(hwgroup);
1061                 if (drive == NULL) {
1062                         int sleeping = 0;
1063                         unsigned long sleep = 0; /* shut up, gcc */
1064                         hwgroup->rq = NULL;
1065                         drive = hwgroup->drive;
1066                         do {
1067                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1068                                         sleeping = 1;
1069                                         sleep = drive->sleep;
1070                                 }
1071                         } while ((drive = drive->next) != hwgroup->drive);
1072                         if (sleeping) {
1073                 /*
1074                  * Take a short snooze, and then wake up this hwgroup again.
1075                  * This gives other hwgroups on the same a chance to
1076                  * play fairly with us, just in case there are big differences
1077                  * in relative throughputs.. don't want to hog the cpu too much.
1078                  */
1079                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1080                                         sleep = jiffies + WAIT_MIN_SLEEP;
1081 #if 1
1082                                 if (timer_pending(&hwgroup->timer))
1083                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1084 #endif
1085                                 /* so that ide_timer_expiry knows what to do */
1086                                 hwgroup->sleeping = 1;
1087                                 mod_timer(&hwgroup->timer, sleep);
1088                                 /* we purposely leave hwgroup->busy==1
1089                                  * while sleeping */
1090                         } else {
1091                                 /* Ugly, but how can we sleep for the lock
1092                                  * otherwise? perhaps from tq_disk?
1093                                  */
1094
1095                                 /* for atari only */
1096                                 ide_release_lock();
1097                                 hwgroup->busy = 0;
1098                         }
1099
1100                         /* no more work for this hwgroup (for now) */
1101                         return;
1102                 }
1103                 hwif = HWIF(drive);
1104                 if (hwgroup->hwif->sharing_irq &&
1105                     hwif != hwgroup->hwif &&
1106                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1107                         /* set nIEN for previous hwif */
1108                         SELECT_INTERRUPT(drive);
1109                 }
1110                 hwgroup->hwif = hwif;
1111                 hwgroup->drive = drive;
1112                 drive->sleeping = 0;
1113                 drive->service_start = jiffies;
1114
1115                 if (blk_queue_plugged(drive->queue)) {
1116                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1117                         break;
1118                 }
1119
1120                 /*
1121                  * we know that the queue isn't empty, but this can happen
1122                  * if the q->prep_rq_fn() decides to kill a request
1123                  */
1124                 rq = elv_next_request(drive->queue);
1125                 if (!rq) {
1126                         hwgroup->busy = 0;
1127                         break;
1128                 }
1129
1130                 /*
1131                  * if rq is a barrier write, issue pre cache flush if not
1132                  * already done
1133                  */
1134                 if (blk_barrier_rq(rq) && !blk_barrier_preflush(rq))
1135                         rq = ide_queue_flush_cmd(drive, rq, 0);
1136
1137                 /*
1138                  * Sanity: don't accept a request that isn't a PM request
1139                  * if we are currently power managed. This is very important as
1140                  * blk_stop_queue() doesn't prevent the elv_next_request()
1141                  * above to return us whatever is in the queue. Since we call
1142                  * ide_do_request() ourselves, we end up taking requests while
1143                  * the queue is blocked...
1144                  *
1145                  * We let requests forced at head of queue with ide-preempt
1146                  * though. I hope that doesn't happen too much, hopefully not
1147                  * unless the subdriver triggers such a thing in its own PM
1148                  * state machine.
1149                  */
1150                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1151                         /* We clear busy, there should be no pending ATA command at this point. */
1152                         hwgroup->busy = 0;
1153                         break;
1154                 }
1155
1156                 hwgroup->rq = rq;
1157
1158                 /*
1159                  * Some systems have trouble with IDE IRQs arriving while
1160                  * the driver is still setting things up.  So, here we disable
1161                  * the IRQ used by this interface while the request is being started.
1162                  * This may look bad at first, but pretty much the same thing
1163                  * happens anyway when any interrupt comes in, IDE or otherwise
1164                  *  -- the kernel masks the IRQ while it is being handled.
1165                  */
1166                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1167                         disable_irq_nosync(hwif->irq);
1168                 spin_unlock(&ide_lock);
1169                 local_irq_enable();
1170                         /* allow other IRQs while we start this request */
1171                 startstop = start_request(drive, rq);
1172                 spin_lock_irq(&ide_lock);
1173                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1174                         enable_irq(hwif->irq);
1175                 if (startstop == ide_stopped)
1176                         hwgroup->busy = 0;
1177         }
1178 }
1179
1180 /*
1181  * Passes the stuff to ide_do_request
1182  */
1183 void do_ide_request(request_queue_t *q)
1184 {
1185         ide_drive_t *drive = q->queuedata;
1186
1187         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1188 }
1189
1190 /*
1191  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1192  * retry the current request in pio mode instead of risking tossing it
1193  * all away
1194  */
1195 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1196 {
1197         ide_hwif_t *hwif = HWIF(drive);
1198         struct request *rq;
1199         ide_startstop_t ret = ide_stopped;
1200
1201         /*
1202          * end current dma transaction
1203          */
1204
1205         if (error < 0) {
1206                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1207                 (void)HWIF(drive)->ide_dma_end(drive);
1208                 ret = ide_error(drive, "dma timeout error",
1209                                                 hwif->INB(IDE_STATUS_REG));
1210         } else {
1211                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1212                 (void) hwif->ide_dma_timeout(drive);
1213         }
1214
1215         /*
1216          * disable dma for now, but remember that we did so because of
1217          * a timeout -- we'll reenable after we finish this next request
1218          * (or rather the first chunk of it) in pio.
1219          */
1220         drive->retry_pio++;
1221         drive->state = DMA_PIO_RETRY;
1222         (void) hwif->ide_dma_off_quietly(drive);
1223
1224         /*
1225          * un-busy drive etc (hwgroup->busy is cleared on return) and
1226          * make sure request is sane
1227          */
1228         rq = HWGROUP(drive)->rq;
1229         HWGROUP(drive)->rq = NULL;
1230
1231         rq->errors = 0;
1232
1233         if (!rq->bio)
1234                 goto out;
1235
1236         rq->sector = rq->bio->bi_sector;
1237         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1238         rq->hard_cur_sectors = rq->current_nr_sectors;
1239         rq->buffer = bio_data(rq->bio);
1240 out:
1241         return ret;
1242 }
1243
1244 /**
1245  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1246  *      @data: timer callback magic (hwgroup)
1247  *
1248  *      An IDE command has timed out before the expected drive return
1249  *      occurred. At this point we attempt to clean up the current
1250  *      mess. If the current handler includes an expiry handler then
1251  *      we invoke the expiry handler, and providing it is happy the
1252  *      work is done. If that fails we apply generic recovery rules
1253  *      invoking the handler and checking the drive DMA status. We
1254  *      have an excessively incestuous relationship with the DMA
1255  *      logic that wants cleaning up.
1256  */
1257
1258 void ide_timer_expiry (unsigned long data)
1259 {
1260         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1261         ide_handler_t   *handler;
1262         ide_expiry_t    *expiry;
1263         unsigned long   flags;
1264         unsigned long   wait = -1;
1265
1266         spin_lock_irqsave(&ide_lock, flags);
1267
1268         if ((handler = hwgroup->handler) == NULL) {
1269                 /*
1270                  * Either a marginal timeout occurred
1271                  * (got the interrupt just as timer expired),
1272                  * or we were "sleeping" to give other devices a chance.
1273                  * Either way, we don't really want to complain about anything.
1274                  */
1275                 if (hwgroup->sleeping) {
1276                         hwgroup->sleeping = 0;
1277                         hwgroup->busy = 0;
1278                 }
1279         } else {
1280                 ide_drive_t *drive = hwgroup->drive;
1281                 if (!drive) {
1282                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1283                         hwgroup->handler = NULL;
1284                 } else {
1285                         ide_hwif_t *hwif;
1286                         ide_startstop_t startstop = ide_stopped;
1287                         if (!hwgroup->busy) {
1288                                 hwgroup->busy = 1;      /* paranoia */
1289                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1290                         }
1291                         if ((expiry = hwgroup->expiry) != NULL) {
1292                                 /* continue */
1293                                 if ((wait = expiry(drive)) > 0) {
1294                                         /* reset timer */
1295                                         hwgroup->timer.expires  = jiffies + wait;
1296                                         add_timer(&hwgroup->timer);
1297                                         spin_unlock_irqrestore(&ide_lock, flags);
1298                                         return;
1299                                 }
1300                         }
1301                         hwgroup->handler = NULL;
1302                         /*
1303                          * We need to simulate a real interrupt when invoking
1304                          * the handler() function, which means we need to
1305                          * globally mask the specific IRQ:
1306                          */
1307                         spin_unlock(&ide_lock);
1308                         hwif  = HWIF(drive);
1309 #if DISABLE_IRQ_NOSYNC
1310                         disable_irq_nosync(hwif->irq);
1311 #else
1312                         /* disable_irq_nosync ?? */
1313                         disable_irq(hwif->irq);
1314 #endif /* DISABLE_IRQ_NOSYNC */
1315                         /* local CPU only,
1316                          * as if we were handling an interrupt */
1317                         local_irq_disable();
1318                         if (hwgroup->polling) {
1319                                 startstop = handler(drive);
1320                         } else if (drive_is_ready(drive)) {
1321                                 if (drive->waiting_for_dma)
1322                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1323                                 (void)ide_ack_intr(hwif);
1324                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1325                                 startstop = handler(drive);
1326                         } else {
1327                                 if (drive->waiting_for_dma) {
1328                                         startstop = ide_dma_timeout_retry(drive, wait);
1329                                 } else
1330                                         startstop =
1331                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1332                         }
1333                         drive->service_time = jiffies - drive->service_start;
1334                         spin_lock_irq(&ide_lock);
1335                         enable_irq(hwif->irq);
1336                         if (startstop == ide_stopped)
1337                                 hwgroup->busy = 0;
1338                 }
1339         }
1340         ide_do_request(hwgroup, IDE_NO_IRQ);
1341         spin_unlock_irqrestore(&ide_lock, flags);
1342 }
1343
1344 /**
1345  *      unexpected_intr         -       handle an unexpected IDE interrupt
1346  *      @irq: interrupt line
1347  *      @hwgroup: hwgroup being processed
1348  *
1349  *      There's nothing really useful we can do with an unexpected interrupt,
1350  *      other than reading the status register (to clear it), and logging it.
1351  *      There should be no way that an irq can happen before we're ready for it,
1352  *      so we needn't worry much about losing an "important" interrupt here.
1353  *
1354  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1355  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1356  *      looks "good", we just ignore the interrupt completely.
1357  *
1358  *      This routine assumes __cli() is in effect when called.
1359  *
1360  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1361  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1362  *      we could screw up by interfering with a new request being set up for
1363  *      irq15.
1364  *
1365  *      In reality, this is a non-issue.  The new command is not sent unless
1366  *      the drive is ready to accept one, in which case we know the drive is
1367  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1368  *      before completing the issuance of any new drive command, so we will not
1369  *      be accidentally invoked as a result of any valid command completion
1370  *      interrupt.
1371  *
1372  *      Note that we must walk the entire hwgroup here. We know which hwif
1373  *      is doing the current command, but we don't know which hwif burped
1374  *      mysteriously.
1375  */
1376
1377 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1378 {
1379         u8 stat;
1380         ide_hwif_t *hwif = hwgroup->hwif;
1381
1382         /*
1383          * handle the unexpected interrupt
1384          */
1385         do {
1386                 if (hwif->irq == irq) {
1387                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1388                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1389                                 /* Try to not flood the console with msgs */
1390                                 static unsigned long last_msgtime, count;
1391                                 ++count;
1392                                 if (time_after(jiffies, last_msgtime + HZ)) {
1393                                         last_msgtime = jiffies;
1394                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1395                                                 "status=0x%02x, count=%ld\n",
1396                                                 hwif->name,
1397                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1398                                 }
1399                         }
1400                 }
1401         } while ((hwif = hwif->next) != hwgroup->hwif);
1402 }
1403
1404 /**
1405  *      ide_intr        -       default IDE interrupt handler
1406  *      @irq: interrupt number
1407  *      @dev_id: hwif group
1408  *      @regs: unused weirdness from the kernel irq layer
1409  *
1410  *      This is the default IRQ handler for the IDE layer. You should
1411  *      not need to override it. If you do be aware it is subtle in
1412  *      places
1413  *
1414  *      hwgroup->hwif is the interface in the group currently performing
1415  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1416  *      the IRQ handler to call. As we issue a command the handlers
1417  *      step through multiple states, reassigning the handler to the
1418  *      next step in the process. Unlike a smart SCSI controller IDE
1419  *      expects the main processor to sequence the various transfer
1420  *      stages. We also manage a poll timer to catch up with most
1421  *      timeout situations. There are still a few where the handlers
1422  *      don't ever decide to give up.
1423  *
1424  *      The handler eventually returns ide_stopped to indicate the
1425  *      request completed. At this point we issue the next request
1426  *      on the hwgroup and the process begins again.
1427  */
1428
1429 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1430 {
1431         unsigned long flags;
1432         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1433         ide_hwif_t *hwif;
1434         ide_drive_t *drive;
1435         ide_handler_t *handler;
1436         ide_startstop_t startstop;
1437
1438         spin_lock_irqsave(&ide_lock, flags);
1439         hwif = hwgroup->hwif;
1440
1441         if (!ide_ack_intr(hwif)) {
1442                 spin_unlock_irqrestore(&ide_lock, flags);
1443                 return IRQ_NONE;
1444         }
1445
1446         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1447                 /*
1448                  * Not expecting an interrupt from this drive.
1449                  * That means this could be:
1450                  *      (1) an interrupt from another PCI device
1451                  *      sharing the same PCI INT# as us.
1452                  * or   (2) a drive just entered sleep or standby mode,
1453                  *      and is interrupting to let us know.
1454                  * or   (3) a spurious interrupt of unknown origin.
1455                  *
1456                  * For PCI, we cannot tell the difference,
1457                  * so in that case we just ignore it and hope it goes away.
1458                  *
1459                  * FIXME: unexpected_intr should be hwif-> then we can
1460                  * remove all the ifdef PCI crap
1461                  */
1462 #ifdef CONFIG_BLK_DEV_IDEPCI
1463                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1464 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1465                 {
1466                         /*
1467                          * Probably not a shared PCI interrupt,
1468                          * so we can safely try to do something about it:
1469                          */
1470                         unexpected_intr(irq, hwgroup);
1471 #ifdef CONFIG_BLK_DEV_IDEPCI
1472                 } else {
1473                         /*
1474                          * Whack the status register, just in case
1475                          * we have a leftover pending IRQ.
1476                          */
1477                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1478 #endif /* CONFIG_BLK_DEV_IDEPCI */
1479                 }
1480                 spin_unlock_irqrestore(&ide_lock, flags);
1481                 return IRQ_NONE;
1482         }
1483         drive = hwgroup->drive;
1484         if (!drive) {
1485                 /*
1486                  * This should NEVER happen, and there isn't much
1487                  * we could do about it here.
1488                  *
1489                  * [Note - this can occur if the drive is hot unplugged]
1490                  */
1491                 spin_unlock_irqrestore(&ide_lock, flags);
1492                 return IRQ_HANDLED;
1493         }
1494         if (!drive_is_ready(drive)) {
1495                 /*
1496                  * This happens regularly when we share a PCI IRQ with
1497                  * another device.  Unfortunately, it can also happen
1498                  * with some buggy drives that trigger the IRQ before
1499                  * their status register is up to date.  Hopefully we have
1500                  * enough advance overhead that the latter isn't a problem.
1501                  */
1502                 spin_unlock_irqrestore(&ide_lock, flags);
1503                 return IRQ_NONE;
1504         }
1505         if (!hwgroup->busy) {
1506                 hwgroup->busy = 1;      /* paranoia */
1507                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1508         }
1509         hwgroup->handler = NULL;
1510         del_timer(&hwgroup->timer);
1511         spin_unlock(&ide_lock);
1512
1513         if (drive->unmask)
1514                 local_irq_enable();
1515         /* service this interrupt, may set handler for next interrupt */
1516         startstop = handler(drive);
1517         spin_lock_irq(&ide_lock);
1518
1519         /*
1520          * Note that handler() may have set things up for another
1521          * interrupt to occur soon, but it cannot happen until
1522          * we exit from this routine, because it will be the
1523          * same irq as is currently being serviced here, and Linux
1524          * won't allow another of the same (on any CPU) until we return.
1525          */
1526         drive->service_time = jiffies - drive->service_start;
1527         if (startstop == ide_stopped) {
1528                 if (hwgroup->handler == NULL) { /* paranoia */
1529                         hwgroup->busy = 0;
1530                         ide_do_request(hwgroup, hwif->irq);
1531                 } else {
1532                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1533                                 "on exit\n", drive->name);
1534                 }
1535         }
1536         spin_unlock_irqrestore(&ide_lock, flags);
1537         return IRQ_HANDLED;
1538 }
1539
1540 /**
1541  *      ide_init_drive_cmd      -       initialize a drive command request
1542  *      @rq: request object
1543  *
1544  *      Initialize a request before we fill it in and send it down to
1545  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1546  *      now it doesn't do a lot, but if that changes abusers will have a
1547  *      nasty suprise.
1548  */
1549
1550 void ide_init_drive_cmd (struct request *rq)
1551 {
1552         memset(rq, 0, sizeof(*rq));
1553         rq->flags = REQ_DRIVE_CMD;
1554         rq->ref_count = 1;
1555 }
1556
1557 EXPORT_SYMBOL(ide_init_drive_cmd);
1558
1559 /**
1560  *      ide_do_drive_cmd        -       issue IDE special command
1561  *      @drive: device to issue command
1562  *      @rq: request to issue
1563  *      @action: action for processing
1564  *
1565  *      This function issues a special IDE device request
1566  *      onto the request queue.
1567  *
1568  *      If action is ide_wait, then the rq is queued at the end of the
1569  *      request queue, and the function sleeps until it has been processed.
1570  *      This is for use when invoked from an ioctl handler.
1571  *
1572  *      If action is ide_preempt, then the rq is queued at the head of
1573  *      the request queue, displacing the currently-being-processed
1574  *      request and this function returns immediately without waiting
1575  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1576  *      intended for careful use by the ATAPI tape/cdrom driver code.
1577  *
1578  *      If action is ide_next, then the rq is queued immediately after
1579  *      the currently-being-processed-request (if any), and the function
1580  *      returns without waiting for the new rq to be completed.  As above,
1581  *      This is VERY DANGEROUS, and is intended for careful use by the
1582  *      ATAPI tape/cdrom driver code.
1583  *
1584  *      If action is ide_end, then the rq is queued at the end of the
1585  *      request queue, and the function returns immediately without waiting
1586  *      for the new rq to be completed. This is again intended for careful
1587  *      use by the ATAPI tape/cdrom driver code.
1588  */
1589
1590 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1591 {
1592         unsigned long flags;
1593         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1594         DECLARE_COMPLETION(wait);
1595         int where = ELEVATOR_INSERT_BACK, err;
1596         int must_wait = (action == ide_wait || action == ide_head_wait);
1597
1598         rq->errors = 0;
1599         rq->rq_status = RQ_ACTIVE;
1600
1601         rq->rq_disk = drive->disk;
1602
1603         /*
1604          * we need to hold an extra reference to request for safe inspection
1605          * after completion
1606          */
1607         if (must_wait) {
1608                 rq->ref_count++;
1609                 rq->waiting = &wait;
1610         }
1611
1612         spin_lock_irqsave(&ide_lock, flags);
1613         if (action == ide_preempt)
1614                 hwgroup->rq = NULL;
1615         if (action == ide_preempt || action == ide_head_wait) {
1616                 where = ELEVATOR_INSERT_FRONT;
1617                 rq->flags |= REQ_PREEMPT;
1618         }
1619         __elv_add_request(drive->queue, rq, where, 0);
1620         ide_do_request(hwgroup, IDE_NO_IRQ);
1621         spin_unlock_irqrestore(&ide_lock, flags);
1622
1623         err = 0;
1624         if (must_wait) {
1625                 wait_for_completion(&wait);
1626                 rq->waiting = NULL;
1627                 if (rq->errors)
1628                         err = -EIO;
1629
1630                 blk_put_request(rq);
1631         }
1632
1633         return err;
1634 }
1635
1636 EXPORT_SYMBOL(ide_do_drive_cmd);