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 int __ide_end_request(ide_drive_t *drive, struct request *rq, int uptodate,
  59                       int nr_sectors)
  60 {
  61         int ret = 1;
  62
  63         BUG_ON(!(rq->flags & REQ_STARTED));
  64
  65         /*
  66          * if failfast is set on a request, override number of sectors and
  67          * complete the whole request right now
  68          */
  69         if (blk_noretry_request(rq) && end_io_error(uptodate))
  70                 nr_sectors = rq->hard_nr_sectors;
  71
  72         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
  73                 rq->errors = -EIO;
  74
  75         /*
  76          * decide whether to reenable DMA -- 3 is a random magic for now,
  77          * if we DMA timeout more than 3 times, just stay in PIO
  78          */
  79         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
  80                 drive->state = 0;
  81                 HWGROUP(drive)->hwif->ide_dma_on(drive);
  82         }
  83
  84         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
  85                 add_disk_randomness(rq->rq_disk);
  86                 blkdev_dequeue_request(rq);
  87                 HWGROUP(drive)->rq = NULL;
  88                 end_that_request_last(rq, uptodate);
  89                 ret = 0;
  90         }
  91
  92         return ret;
  93 }
  94 EXPORT_SYMBOL(__ide_end_request);
  95
  96 /**
  97  *      ide_end_request         -       complete an IDE I/O
  98  *      @drive: IDE device for the I/O
  99  *      @uptodate:
 100  *      @nr_sectors: number of sectors completed
 101  *
 102  *      This is our end_request wrapper function. We complete the I/O
 103  *      update random number input and dequeue the request, which if
 104  *      it was tagged may be out of order.
 105  */
 106
 107 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
 108 {
 109         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_sectors)
 121                 nr_sectors = rq->hard_cur_sectors;
 122
 123         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
 124
 125         spin_unlock_irqrestore(&ide_lock, flags);
 126         return ret;
 127 }
 128 EXPORT_SYMBOL(ide_end_request);
 129
 130 /*
 131  * Power Management state machine. This one is rather trivial for now,
 132  * we should probably add more, like switching back to PIO on suspend
 133  * to help some BIOSes, re-do the door locking on resume, etc...
 134  */
 135
 136 enum {
 137         ide_pm_flush_cache      = ide_pm_state_start_suspend,
 138         idedisk_pm_standby,
 139
 140         idedisk_pm_idle         = ide_pm_state_start_resume,
 141         ide_pm_restore_dma,
 142 };
 143
 144 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
 145 {
 146         if (drive->media != ide_disk)
 147                 return;
 148
 149         switch (rq->pm->pm_step) {
 150         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
 151                 if (rq->pm->pm_state == PM_EVENT_FREEZE)
 152                         rq->pm->pm_step = ide_pm_state_completed;
 153                 else
 154                         rq->pm->pm_step = idedisk_pm_standby;
 155                 break;
 156         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
 157                 rq->pm->pm_step = ide_pm_state_completed;
 158                 break;
 159         case idedisk_pm_idle:           /* Resume step 1 (idle) complete */
 160                 rq->pm->pm_step = ide_pm_restore_dma;
 161                 break;
 162         }
 163 }
 164
 165 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
 166 {
 167         ide_task_t *args = rq->special;
 168
 169         memset(args, 0, sizeof(*args));
 170
 171         if (drive->media != ide_disk) {
 172                 /* skip idedisk_pm_idle for ATAPI devices */
 173                 if (rq->pm->pm_step == idedisk_pm_idle)
 174                         rq->pm->pm_step = ide_pm_restore_dma;
 175         }
 176
 177         switch (rq->pm->pm_step) {
 178         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
 179                 if (drive->media != ide_disk)
 180                         break;
 181                 /* Not supported? Switch to next step now. */
 182                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
 183                         ide_complete_power_step(drive, rq, 0, 0);
 184                         return ide_stopped;
 185                 }
 186                 if (ide_id_has_flush_cache_ext(drive->id))
 187                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
 188                 else
 189                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
 190                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 191                 args->handler      = &task_no_data_intr;
 192                 return do_rw_taskfile(drive, args);
 193
 194         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
 195                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
 196                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 197                 args->handler      = &task_no_data_intr;
 198                 return do_rw_taskfile(drive, args);
 199
 200         case idedisk_pm_idle:           /* Resume step 1 (idle) */
 201                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
 202                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 203                 args->handler = task_no_data_intr;
 204                 return do_rw_taskfile(drive, args);
 205
 206         case ide_pm_restore_dma:        /* Resume step 2 (restore DMA) */
 207                 /*
 208                  * Right now, all we do is call hwif->ide_dma_check(drive),
 209                  * we could be smarter and check for current xfer_speed
 210                  * in struct drive etc...
 211                  */
 212                 if ((drive->id->capability & 1) == 0)
 213                         break;
 214                 if (drive->hwif->ide_dma_check == NULL)
 215                         break;
 216                 drive->hwif->ide_dma_check(drive);
 217                 break;
 218         }
 219         rq->pm->pm_step = ide_pm_state_completed;
 220         return ide_stopped;
 221 }
 222
 223 /**
 224  *      ide_complete_pm_request - end the current Power Management request
 225  *      @drive: target drive
 226  *      @rq: request
 227  *
 228  *      This function cleans up the current PM request and stops the queue
 229  *      if necessary.
 230  */
 231 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
 232 {
 233         unsigned long flags;
 234
 235 #ifdef DEBUG_PM
 236         printk("%s: completing PM request, %s\n", drive->name,
 237                blk_pm_suspend_request(rq) ? "suspend" : "resume");
 238 #endif
 239         spin_lock_irqsave(&ide_lock, flags);
 240         if (blk_pm_suspend_request(rq)) {
 241                 blk_stop_queue(drive->queue);
 242         } else {
 243                 drive->blocked = 0;
 244                 blk_start_queue(drive->queue);
 245         }
 246         blkdev_dequeue_request(rq);
 247         HWGROUP(drive)->rq = NULL;
 248         end_that_request_last(rq, 1);
 249         spin_unlock_irqrestore(&ide_lock, flags);
 250 }
 251
 252 /*
 253  * FIXME: probably move this somewhere else, name is bad too :)
 254  */
 255 u64 ide_get_error_location(ide_drive_t *drive, char *args)
 256 {
 257         u32 high, low;
 258         u8 hcyl, lcyl, sect;
 259         u64 sector;
 260
 261         high = 0;
 262         hcyl = args[5];
 263         lcyl = args[4];
 264         sect = args[3];
 265
 266         if (ide_id_has_flush_cache_ext(drive->id)) {
 267                 low = (hcyl << 16) | (lcyl << 8) | sect;
 268                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 269                 high = ide_read_24(drive);
 270         } else {
 271                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
 272                 if (cur & 0x40) {
 273                         high = cur & 0xf;
 274                         low = (hcyl << 16) | (lcyl << 8) | sect;
 275                 } else {
 276                         low = hcyl * drive->head * drive->sect;
 277                         low += lcyl * drive->sect;
 278                         low += sect - 1;
 279                 }
 280         }
 281
 282         sector = ((u64) high << 24) | low;
 283         return sector;
 284 }
 285 EXPORT_SYMBOL(ide_get_error_location);
 286
 287 /**
 288  *      ide_end_drive_cmd       -       end an explicit drive command
 289  *      @drive: command
 290  *      @stat: status bits
 291  *      @err: error bits
 292  *
 293  *      Clean up after success/failure of an explicit drive command.
 294  *      These get thrown onto the queue so they are synchronized with
 295  *      real I/O operations on the drive.
 296  *
 297  *      In LBA48 mode we have to read the register set twice to get
 298  *      all the extra information out.
 299  */
 300
 301 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
 302 {
 303         ide_hwif_t *hwif = HWIF(drive);
 304         unsigned long flags;
 305         struct request *rq;
 306
 307         spin_lock_irqsave(&ide_lock, flags);
 308         rq = HWGROUP(drive)->rq;
 309         spin_unlock_irqrestore(&ide_lock, flags);
 310
 311         if (rq->flags & REQ_DRIVE_CMD) {
 312                 u8 *args = (u8 *) rq->buffer;
 313                 if (rq->errors == 0)
 314                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 315
 316                 if (args) {
 317                         args[0] = stat;
 318                         args[1] = err;
 319                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 320                 }
 321         } else if (rq->flags & REQ_DRIVE_TASK) {
 322                 u8 *args = (u8 *) rq->buffer;
 323                 if (rq->errors == 0)
 324                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 325
 326                 if (args) {
 327                         args[0] = stat;
 328                         args[1] = err;
 329                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 330                         args[3] = hwif->INB(IDE_SECTOR_REG);
 331                         args[4] = hwif->INB(IDE_LCYL_REG);
 332                         args[5] = hwif->INB(IDE_HCYL_REG);
 333                         args[6] = hwif->INB(IDE_SELECT_REG);
 334                 }
 335         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
 336                 ide_task_t *args = (ide_task_t *) rq->special;
 337                 if (rq->errors == 0)
 338                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 339
 340                 if (args) {
 341                         if (args->tf_in_flags.b.data) {
 342                                 u16 data                                = hwif->INW(IDE_DATA_REG);
 343                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
 344                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
 345                         }
 346                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
 347                         /* be sure we're looking at the low order bits */
 348                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
 349                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
 350                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
 351                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
 352                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
 353                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
 354                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
 355
 356                         if (drive->addressing == 1) {
 357                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 358                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
 359                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
 360                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
 361                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
 362                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
 363                         }
 364                 }
 365         } else if (blk_pm_request(rq)) {
 366 #ifdef DEBUG_PM
 367                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
 368                         drive->name, rq->pm->pm_step, stat, err);
 369 #endif
 370                 ide_complete_power_step(drive, rq, stat, err);
 371                 if (rq->pm->pm_step == ide_pm_state_completed)
 372                         ide_complete_pm_request(drive, rq);
 373                 return;
 374         }
 375
 376         spin_lock_irqsave(&ide_lock, flags);
 377         blkdev_dequeue_request(rq);
 378         HWGROUP(drive)->rq = NULL;
 379         rq->errors = err;
 380         end_that_request_last(rq, !rq->errors);
 381         spin_unlock_irqrestore(&ide_lock, flags);
 382 }
 383
 384 EXPORT_SYMBOL(ide_end_drive_cmd);
 385
 386 /**
 387  *      try_to_flush_leftover_data      -       flush junk
 388  *      @drive: drive to flush
 389  *
 390  *      try_to_flush_leftover_data() is invoked in response to a drive
 391  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
 392  *      resetting the drive, this routine tries to clear the condition
 393  *      by read a sector's worth of data from the drive.  Of course,
 394  *      this may not help if the drive is *waiting* for data from *us*.
 395  */
 396 static void try_to_flush_leftover_data (ide_drive_t *drive)
 397 {
 398         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
 399
 400         if (drive->media != ide_disk)
 401                 return;
 402         while (i > 0) {
 403                 u32 buffer[16];
 404                 u32 wcount = (i > 16) ? 16 : i;
 405
 406                 i -= wcount;
 407                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
 408         }
 409 }
 410
 411 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
 412 {
 413         if (rq->rq_disk) {
 414                 ide_driver_t *drv;
 415
 416                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 417                 drv->end_request(drive, 0, 0);
 418         } else
 419                 ide_end_request(drive, 0, 0);
 420 }
 421
 422 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 423 {
 424         ide_hwif_t *hwif = drive->hwif;
 425
 426         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 427                 /* other bits are useless when BUSY */
 428                 rq->errors |= ERROR_RESET;
 429         } else if (stat & ERR_STAT) {
 430                 /* err has different meaning on cdrom and tape */
 431                 if (err == ABRT_ERR) {
 432                         if (drive->select.b.lba &&
 433                             /* some newer drives don't support WIN_SPECIFY */
 434                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
 435                                 return ide_stopped;
 436                 } else if ((err & BAD_CRC) == BAD_CRC) {
 437                         /* UDMA crc error, just retry the operation */
 438                         drive->crc_count++;
 439                 } else if (err & (BBD_ERR | ECC_ERR)) {
 440                         /* retries won't help these */
 441                         rq->errors = ERROR_MAX;
 442                 } else if (err & TRK0_ERR) {
 443                         /* help it find track zero */
 444                         rq->errors |= ERROR_RECAL;
 445                 }
 446         }
 447
 448         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ)
 449                 try_to_flush_leftover_data(drive);
 450
 451         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 452                 /* force an abort */
 453                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 454
 455         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
 456                 ide_kill_rq(drive, rq);
 457         else {
 458                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 459                         ++rq->errors;
 460                         return ide_do_reset(drive);
 461                 }
 462                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
 463                         drive->special.b.recalibrate = 1;
 464                 ++rq->errors;
 465         }
 466         return ide_stopped;
 467 }
 468
 469 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 470 {
 471         ide_hwif_t *hwif = drive->hwif;
 472
 473         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 474                 /* other bits are useless when BUSY */
 475                 rq->errors |= ERROR_RESET;
 476         } else {
 477                 /* add decoding error stuff */
 478         }
 479
 480         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 481                 /* force an abort */
 482                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 483
 484         if (rq->errors >= ERROR_MAX) {
 485                 ide_kill_rq(drive, rq);
 486         } else {
 487                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 488                         ++rq->errors;
 489                         return ide_do_reset(drive);
 490                 }
 491                 ++rq->errors;
 492         }
 493
 494         return ide_stopped;
 495 }
 496
 497 ide_startstop_t
 498 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 499 {
 500         if (drive->media == ide_disk)
 501                 return ide_ata_error(drive, rq, stat, err);
 502         return ide_atapi_error(drive, rq, stat, err);
 503 }
 504
 505 EXPORT_SYMBOL_GPL(__ide_error);
 506
 507 /**
 508  *      ide_error       -       handle an error on the IDE
 509  *      @drive: drive the error occurred on
 510  *      @msg: message to report
 511  *      @stat: status bits
 512  *
 513  *      ide_error() takes action based on the error returned by the drive.
 514  *      For normal I/O that may well include retries. We deal with
 515  *      both new-style (taskfile) and old style command handling here.
 516  *      In the case of taskfile command handling there is work left to
 517  *      do
 518  */
 519
 520 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
 521 {
 522         struct request *rq;
 523         u8 err;
 524
 525         err = ide_dump_status(drive, msg, stat);
 526
 527         if ((rq = HWGROUP(drive)->rq) == NULL)
 528                 return ide_stopped;
 529
 530         /* retry only "normal" I/O: */
 531         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 532                 rq->errors = 1;
 533                 ide_end_drive_cmd(drive, stat, err);
 534                 return ide_stopped;
 535         }
 536
 537         if (rq->rq_disk) {
 538                 ide_driver_t *drv;
 539
 540                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 541                 return drv->error(drive, rq, stat, err);
 542         } else
 543                 return __ide_error(drive, rq, stat, err);
 544 }
 545
 546 EXPORT_SYMBOL_GPL(ide_error);
 547
 548 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
 549 {
 550         if (drive->media != ide_disk)
 551                 rq->errors |= ERROR_RESET;
 552
 553         ide_kill_rq(drive, rq);
 554
 555         return ide_stopped;
 556 }
 557
 558 EXPORT_SYMBOL_GPL(__ide_abort);
 559
 560 /**
 561  *      ide_abort       -       abort pending IDE operations
 562  *      @drive: drive the error occurred on
 563  *      @msg: message to report
 564  *
 565  *      ide_abort kills and cleans up when we are about to do a
 566  *      host initiated reset on active commands. Longer term we
 567  *      want handlers to have sensible abort handling themselves
 568  *
 569  *      This differs fundamentally from ide_error because in
 570  *      this case the command is doing just fine when we
 571  *      blow it away.
 572  */
 573
 574 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
 575 {
 576         struct request *rq;
 577
 578         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
 579                 return ide_stopped;
 580
 581         /* retry only "normal" I/O: */
 582         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 583                 rq->errors = 1;
 584                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
 585                 return ide_stopped;
 586         }
 587
 588         if (rq->rq_disk) {
 589                 ide_driver_t *drv;
 590
 591                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 592                 return drv->abort(drive, rq);
 593         } else
 594                 return __ide_abort(drive, rq);
 595 }
 596
 597 /**
 598  *      ide_cmd         -       issue a simple drive command
 599  *      @drive: drive the command is for
 600  *      @cmd: command byte
 601  *      @nsect: sector byte
 602  *      @handler: handler for the command completion
 603  *
 604  *      Issue a simple drive command with interrupts.
 605  *      The drive must be selected beforehand.
 606  */
 607
 608 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
 609                 ide_handler_t *handler)
 610 {
 611         ide_hwif_t *hwif = HWIF(drive);
 612         if (IDE_CONTROL_REG)
 613                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
 614         SELECT_MASK(drive,0);
 615         hwif->OUTB(nsect,IDE_NSECTOR_REG);
 616         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
 617 }
 618
 619 /**
 620  *      drive_cmd_intr          -       drive command completion interrupt
 621  *      @drive: drive the completion interrupt occurred on
 622  *
 623  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
 624  *      We do any necessary data reading and then wait for the drive to
 625  *      go non busy. At that point we may read the error data and complete
 626  *      the request
 627  */
 628
 629 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
 630 {
 631         struct request *rq = HWGROUP(drive)->rq;
 632         ide_hwif_t *hwif = HWIF(drive);
 633         u8 *args = (u8 *) rq->buffer;
 634         u8 stat = hwif->INB(IDE_STATUS_REG);
 635         int retries = 10;
 636
 637         local_irq_enable();
 638         if ((stat & DRQ_STAT) && args && args[3]) {
 639                 u8 io_32bit = drive->io_32bit;
 640                 drive->io_32bit = 0;
 641                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
 642                 drive->io_32bit = io_32bit;
 643                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
 644                         udelay(100);
 645         }
 646
 647         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
 648                 return ide_error(drive, "drive_cmd", stat);
 649                 /* calls ide_end_drive_cmd */
 650         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
 651         return ide_stopped;
 652 }
 653
 654 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
 655 {
 656         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
 657         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
 658         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
 659         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
 660         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
 661         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
 662
 663         task->handler = &set_geometry_intr;
 664 }
 665
 666 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
 667 {
 668         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
 669         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
 670
 671         task->handler = &recal_intr;
 672 }
 673
 674 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
 675 {
 676         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
 677         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
 678
 679         task->handler = &set_multmode_intr;
 680 }
 681
 682 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
 683 {
 684         special_t *s = &drive->special;
 685         ide_task_t args;
 686
 687         memset(&args, 0, sizeof(ide_task_t));
 688         args.command_type = IDE_DRIVE_TASK_NO_DATA;
 689
 690         if (s->b.set_geometry) {
 691                 s->b.set_geometry = 0;
 692                 ide_init_specify_cmd(drive, &args);
 693         } else if (s->b.recalibrate) {
 694                 s->b.recalibrate = 0;
 695                 ide_init_restore_cmd(drive, &args);
 696         } else if (s->b.set_multmode) {
 697                 s->b.set_multmode = 0;
 698                 if (drive->mult_req > drive->id->max_multsect)
 699                         drive->mult_req = drive->id->max_multsect;
 700                 ide_init_setmult_cmd(drive, &args);
 701         } else if (s->all) {
 702                 int special = s->all;
 703                 s->all = 0;
 704                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
 705                 return ide_stopped;
 706         }
 707
 708         do_rw_taskfile(drive, &args);
 709
 710         return ide_started;
 711 }
 712
 713 /**
 714  *      do_special              -       issue some special commands
 715  *      @drive: drive the command is for
 716  *
 717  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
 718  *      commands to a drive.  It used to do much more, but has been scaled
 719  *      back.
 720  */
 721
 722 static ide_startstop_t do_special (ide_drive_t *drive)
 723 {
 724         special_t *s = &drive->special;
 725
 726 #ifdef DEBUG
 727         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
 728 #endif
 729         if (s->b.set_tune) {
 730                 s->b.set_tune = 0;
 731                 if (HWIF(drive)->tuneproc != NULL)
 732                         HWIF(drive)->tuneproc(drive, drive->tune_req);
 733                 return ide_stopped;
 734         } else {
 735                 if (drive->media == ide_disk)
 736                         return ide_disk_special(drive);
 737
 738                 s->all = 0;
 739                 drive->mult_req = 0;
 740                 return ide_stopped;
 741         }
 742 }
 743
 744 void ide_map_sg(ide_drive_t *drive, struct request *rq)
 745 {
 746         ide_hwif_t *hwif = drive->hwif;
 747         struct scatterlist *sg = hwif->sg_table;
 748
 749         if (hwif->sg_mapped)    /* needed by ide-scsi */
 750                 return;
 751
 752         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
 753                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
 754         } else {
 755                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
 756                 hwif->sg_nents = 1;
 757         }
 758 }
 759
 760 EXPORT_SYMBOL_GPL(ide_map_sg);
 761
 762 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
 763 {
 764         ide_hwif_t *hwif = drive->hwif;
 765
 766         hwif->nsect = hwif->nleft = rq->nr_sectors;
 767         hwif->cursg = hwif->cursg_ofs = 0;
 768 }
 769
 770 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
 771
 772 /**
 773  *      execute_drive_command   -       issue special drive command
 774  *      @drive: the drive to issue the command on
 775  *      @rq: the request structure holding the command
 776  *
 777  *      execute_drive_cmd() issues a special drive command,  usually
 778  *      initiated by ioctl() from the external hdparm program. The
 779  *      command can be a drive command, drive task or taskfile
 780  *      operation. Weirdly you can call it with NULL to wait for
 781  *      all commands to finish. Don't do this as that is due to change
 782  */
 783
 784 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
 785                 struct request *rq)
 786 {
 787         ide_hwif_t *hwif = HWIF(drive);
 788         if (rq->flags & REQ_DRIVE_TASKFILE) {
 789                 ide_task_t *args = rq->special;
 790
 791                 if (!args)
 792                         goto done;
 793
 794                 hwif->data_phase = args->data_phase;
 795
 796                 switch (hwif->data_phase) {
 797                 case TASKFILE_MULTI_OUT:
 798                 case TASKFILE_OUT:
 799                 case TASKFILE_MULTI_IN:
 800                 case TASKFILE_IN:
 801                         ide_init_sg_cmd(drive, rq);
 802                         ide_map_sg(drive, rq);
 803                 default:
 804                         break;
 805                 }
 806
 807                 if (args->tf_out_flags.all != 0)
 808                         return flagged_taskfile(drive, args);
 809                 return do_rw_taskfile(drive, args);
 810         } else if (rq->flags & REQ_DRIVE_TASK) {
 811                 u8 *args = rq->buffer;
 812                 u8 sel;
 813
 814                 if (!args)
 815                         goto done;
 816 #ifdef DEBUG
 817                 printk("%s: DRIVE_TASK_CMD ", drive->name);
 818                 printk("cmd=0x%02x ", args[0]);
 819                 printk("fr=0x%02x ", args[1]);
 820                 printk("ns=0x%02x ", args[2]);
 821                 printk("sc=0x%02x ", args[3]);
 822                 printk("lcyl=0x%02x ", args[4]);
 823                 printk("hcyl=0x%02x ", args[5]);
 824                 printk("sel=0x%02x\n", args[6]);
 825 #endif
 826                 hwif->OUTB(args[1], IDE_FEATURE_REG);
 827                 hwif->OUTB(args[3], IDE_SECTOR_REG);
 828                 hwif->OUTB(args[4], IDE_LCYL_REG);
 829                 hwif->OUTB(args[5], IDE_HCYL_REG);
 830                 sel = (args[6] & ~0x10);
 831                 if (drive->select.b.unit)
 832                         sel |= 0x10;
 833                 hwif->OUTB(sel, IDE_SELECT_REG);
 834                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
 835                 return ide_started;
 836         } else if (rq->flags & REQ_DRIVE_CMD) {
 837                 u8 *args = rq->buffer;
 838
 839                 if (!args)
 840                         goto done;
 841 #ifdef DEBUG
 842                 printk("%s: DRIVE_CMD ", drive->name);
 843                 printk("cmd=0x%02x ", args[0]);
 844                 printk("sc=0x%02x ", args[1]);
 845                 printk("fr=0x%02x ", args[2]);
 846                 printk("xx=0x%02x\n", args[3]);
 847 #endif
 848                 if (args[0] == WIN_SMART) {
 849                         hwif->OUTB(0x4f, IDE_LCYL_REG);
 850                         hwif->OUTB(0xc2, IDE_HCYL_REG);
 851                         hwif->OUTB(args[2],IDE_FEATURE_REG);
 852                         hwif->OUTB(args[1],IDE_SECTOR_REG);
 853                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
 854                         return ide_started;
 855                 }
 856                 hwif->OUTB(args[2],IDE_FEATURE_REG);
 857                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
 858                 return ide_started;
 859         }
 860
 861 done:
 862         /*
 863          * NULL is actually a valid way of waiting for
 864          * all current requests to be flushed from the queue.
 865          */
 866 #ifdef DEBUG
 867         printk("%s: DRIVE_CMD (null)\n", drive->name);
 868 #endif
 869         ide_end_drive_cmd(drive,
 870                         hwif->INB(IDE_STATUS_REG),
 871                         hwif->INB(IDE_ERROR_REG));
 872         return ide_stopped;
 873 }
 874
 875 /**
 876  *      start_request   -       start of I/O and command issuing for IDE
 877  *
 878  *      start_request() initiates handling of a new I/O request. It
 879  *      accepts commands and I/O (read/write) requests. It also does
 880  *      the final remapping for weird stuff like EZDrive. Once
 881  *      device mapper can work sector level the EZDrive stuff can go away
 882  *
 883  *      FIXME: this function needs a rename
 884  */
 885
 886 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
 887 {
 888         ide_startstop_t startstop;
 889         sector_t block;
 890
 891         BUG_ON(!(rq->flags & REQ_STARTED));
 892
 893 #ifdef DEBUG
 894         printk("%s: start_request: current=0x%08lx\n",
 895                 HWIF(drive)->name, (unsigned long) rq);
 896 #endif
 897
 898         /* bail early if we've exceeded max_failures */
 899         if (drive->max_failures && (drive->failures > drive->max_failures)) {
 900                 goto kill_rq;
 901         }
 902
 903         block    = rq->sector;
 904         if (blk_fs_request(rq) &&
 905             (drive->media == ide_disk || drive->media == ide_floppy)) {
 906                 block += drive->sect0;
 907         }
 908         /* Yecch - this will shift the entire interval,
 909            possibly killing some innocent following sector */
 910         if (block == 0 && drive->remap_0_to_1 == 1)
 911                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
 912
 913         if (blk_pm_suspend_request(rq) &&
 914             rq->pm->pm_step == ide_pm_state_start_suspend)
 915                 /* Mark drive blocked when starting the suspend sequence. */
 916                 drive->blocked = 1;
 917         else if (blk_pm_resume_request(rq) &&
 918                  rq->pm->pm_step == ide_pm_state_start_resume) {
 919                 /*
 920                  * The first thing we do on wakeup is to wait for BSY bit to
 921                  * go away (with a looong timeout) as a drive on this hwif may
 922                  * just be POSTing itself.
 923                  * We do that before even selecting as the "other" device on
 924                  * the bus may be broken enough to walk on our toes at this
 925                  * point.
 926                  */
 927                 int rc;
 928 #ifdef DEBUG_PM
 929                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
 930 #endif
 931                 rc = ide_wait_not_busy(HWIF(drive), 35000);
 932                 if (rc)
 933                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
 934                 SELECT_DRIVE(drive);
 935                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
 936                 rc = ide_wait_not_busy(HWIF(drive), 10000);
 937                 if (rc)
 938                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
 939         }
 940
 941         SELECT_DRIVE(drive);
 942         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
 943                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
 944                 return startstop;
 945         }
 946         if (!drive->special.all) {
 947                 ide_driver_t *drv;
 948
 949                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
 950                         return execute_drive_cmd(drive, rq);
 951                 else if (rq->flags & REQ_DRIVE_TASKFILE)
 952                         return execute_drive_cmd(drive, rq);
 953                 else if (blk_pm_request(rq)) {
 954 #ifdef DEBUG_PM
 955                         printk("%s: start_power_step(step: %d)\n",
 956                                 drive->name, rq->pm->pm_step);
 957 #endif
 958                         startstop = ide_start_power_step(drive, rq);
 959                         if (startstop == ide_stopped &&
 960                             rq->pm->pm_step == ide_pm_state_completed)
 961                                 ide_complete_pm_request(drive, rq);
 962                         return startstop;
 963                 }
 964
 965                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 966                 return drv->do_request(drive, rq, block);
 967         }
 968         return do_special(drive);
 969 kill_rq:
 970         ide_kill_rq(drive, rq);
 971         return ide_stopped;
 972 }
 973
 974 /**
 975  *      ide_stall_queue         -       pause an IDE device
 976  *      @drive: drive to stall
 977  *      @timeout: time to stall for (jiffies)
 978  *
 979  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
 980  *      to the hwgroup by sleeping for timeout jiffies.
 981  */
 982
 983 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
 984 {
 985         if (timeout > WAIT_WORSTCASE)
 986                 timeout = WAIT_WORSTCASE;
 987         drive->sleep = timeout + jiffies;
 988         drive->sleeping = 1;
 989 }
 990
 991 EXPORT_SYMBOL(ide_stall_queue);
 992
 993 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
 994
 995 /**
 996  *      choose_drive            -       select a drive to service
 997  *      @hwgroup: hardware group to select on
 998  *
 999  *      choose_drive() selects the next drive which will be serviced.
1000  *      This is necessary because the IDE layer can't issue commands
1001  *      to both drives on the same cable, unlike SCSI.
1002  */
1003
1004 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1005 {
1006         ide_drive_t *drive, *best;
1007
1008 repeat:
1009         best = NULL;
1010         drive = hwgroup->drive;
1011
1012         /*
1013          * drive is doing pre-flush, ordered write, post-flush sequence. even
1014          * though that is 3 requests, it must be seen as a single transaction.
1015          * we must not preempt this drive until that is complete
1016          */
1017         if (blk_queue_flushing(drive->queue)) {
1018                 /*
1019                  * small race where queue could get replugged during
1020                  * the 3-request flush cycle, just yank the plug since
1021                  * we want it to finish asap
1022                  */
1023                 blk_remove_plug(drive->queue);
1024                 return drive;
1025         }
1026
1027         do {
1028                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1029                     && !elv_queue_empty(drive->queue)) {
1030                         if (!best
1031                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1032                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1033                         {
1034                                 if (!blk_queue_plugged(drive->queue))
1035                                         best = drive;
1036                         }
1037                 }
1038         } while ((drive = drive->next) != hwgroup->drive);
1039         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1040                 long t = (signed long)(WAKEUP(best) - jiffies);
1041                 if (t >= WAIT_MIN_SLEEP) {
1042                 /*
1043                  * We *may* have some time to spare, but first let's see if
1044                  * someone can potentially benefit from our nice mood today..
1045                  */
1046                         drive = best->next;
1047                         do {
1048                                 if (!drive->sleeping
1049                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1050                                  && time_before(WAKEUP(drive), jiffies + t))
1051                                 {
1052                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1053                                         goto repeat;
1054                                 }
1055                         } while ((drive = drive->next) != best);
1056                 }
1057         }
1058         return best;
1059 }
1060
1061 /*
1062  * Issue a new request to a drive from hwgroup
1063  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1064  *
1065  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1066  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1067  * may have both interfaces in a single hwgroup to "serialize" access.
1068  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1069  * together into one hwgroup for serialized access.
1070  *
1071  * Note also that several hwgroups can end up sharing a single IRQ,
1072  * possibly along with many other devices.  This is especially common in
1073  * PCI-based systems with off-board IDE controller cards.
1074  *
1075  * The IDE driver uses the single global ide_lock spinlock to protect
1076  * access to the request queues, and to protect the hwgroup->busy flag.
1077  *
1078  * The first thread into the driver for a particular hwgroup sets the
1079  * hwgroup->busy flag to indicate that this hwgroup is now active,
1080  * and then initiates processing of the top request from the request queue.
1081  *
1082  * Other threads attempting entry notice the busy setting, and will simply
1083  * queue their new requests and exit immediately.  Note that hwgroup->busy
1084  * remains set even when the driver is merely awaiting the next interrupt.
1085  * Thus, the meaning is "this hwgroup is busy processing a request".
1086  *
1087  * When processing of a request completes, the completing thread or IRQ-handler
1088  * will start the next request from the queue.  If no more work remains,
1089  * the driver will clear the hwgroup->busy flag and exit.
1090  *
1091  * The ide_lock (spinlock) is used to protect all access to the
1092  * hwgroup->busy flag, but is otherwise not needed for most processing in
1093  * the driver.  This makes the driver much more friendlier to shared IRQs
1094  * than previous designs, while remaining 100% (?) SMP safe and capable.
1095  */
1096 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1097 {
1098         ide_drive_t     *drive;
1099         ide_hwif_t      *hwif;
1100         struct request  *rq;
1101         ide_startstop_t startstop;
1102         int             loops = 0;
1103
1104         /* for atari only: POSSIBLY BROKEN HERE(?) */
1105         ide_get_lock(ide_intr, hwgroup);
1106
1107         /* caller must own ide_lock */
1108         BUG_ON(!irqs_disabled());
1109
1110         while (!hwgroup->busy) {
1111                 hwgroup->busy = 1;
1112                 drive = choose_drive(hwgroup);
1113                 if (drive == NULL) {
1114                         int sleeping = 0;
1115                         unsigned long sleep = 0; /* shut up, gcc */
1116                         hwgroup->rq = NULL;
1117                         drive = hwgroup->drive;
1118                         do {
1119                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1120                                         sleeping = 1;
1121                                         sleep = drive->sleep;
1122                                 }
1123                         } while ((drive = drive->next) != hwgroup->drive);
1124                         if (sleeping) {
1125                 /*
1126                  * Take a short snooze, and then wake up this hwgroup again.
1127                  * This gives other hwgroups on the same a chance to
1128                  * play fairly with us, just in case there are big differences
1129                  * in relative throughputs.. don't want to hog the cpu too much.
1130                  */
1131                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1132                                         sleep = jiffies + WAIT_MIN_SLEEP;
1133 #if 1
1134                                 if (timer_pending(&hwgroup->timer))
1135                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1136 #endif
1137                                 /* so that ide_timer_expiry knows what to do */
1138                                 hwgroup->sleeping = 1;
1139                                 mod_timer(&hwgroup->timer, sleep);
1140                                 /* we purposely leave hwgroup->busy==1
1141                                  * while sleeping */
1142                         } else {
1143                                 /* Ugly, but how can we sleep for the lock
1144                                  * otherwise? perhaps from tq_disk?
1145                                  */
1146
1147                                 /* for atari only */
1148                                 ide_release_lock();
1149                                 hwgroup->busy = 0;
1150                         }
1151
1152                         /* no more work for this hwgroup (for now) */
1153                         return;
1154                 }
1155         again:
1156                 hwif = HWIF(drive);
1157                 if (hwgroup->hwif->sharing_irq &&
1158                     hwif != hwgroup->hwif &&
1159                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1160                         /* set nIEN for previous hwif */
1161                         SELECT_INTERRUPT(drive);
1162                 }
1163                 hwgroup->hwif = hwif;
1164                 hwgroup->drive = drive;
1165                 drive->sleeping = 0;
1166                 drive->service_start = jiffies;
1167
1168                 if (blk_queue_plugged(drive->queue)) {
1169                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1170                         break;
1171                 }
1172
1173                 /*
1174                  * we know that the queue isn't empty, but this can happen
1175                  * if the q->prep_rq_fn() decides to kill a request
1176                  */
1177                 rq = elv_next_request(drive->queue);
1178                 if (!rq) {
1179                         hwgroup->busy = 0;
1180                         break;
1181                 }
1182
1183                 /*
1184                  * Sanity: don't accept a request that isn't a PM request
1185                  * if we are currently power managed. This is very important as
1186                  * blk_stop_queue() doesn't prevent the elv_next_request()
1187                  * above to return us whatever is in the queue. Since we call
1188                  * ide_do_request() ourselves, we end up taking requests while
1189                  * the queue is blocked...
1190                  *
1191                  * We let requests forced at head of queue with ide-preempt
1192                  * though. I hope that doesn't happen too much, hopefully not
1193                  * unless the subdriver triggers such a thing in its own PM
1194                  * state machine.
1195                  *
1196                  * We count how many times we loop here to make sure we service
1197                  * all drives in the hwgroup without looping for ever
1198                  */
1199                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1200                         drive = drive->next ? drive->next : hwgroup->drive;
1201                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1202                                 goto again;
1203                         /* We clear busy, there should be no pending ATA command at this point. */
1204                         hwgroup->busy = 0;
1205                         break;
1206                 }
1207
1208                 hwgroup->rq = rq;
1209
1210                 /*
1211                  * Some systems have trouble with IDE IRQs arriving while
1212                  * the driver is still setting things up.  So, here we disable
1213                  * the IRQ used by this interface while the request is being started.
1214                  * This may look bad at first, but pretty much the same thing
1215                  * happens anyway when any interrupt comes in, IDE or otherwise
1216                  *  -- the kernel masks the IRQ while it is being handled.
1217                  */
1218                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1219                         disable_irq_nosync(hwif->irq);
1220                 spin_unlock(&ide_lock);
1221                 local_irq_enable();
1222                         /* allow other IRQs while we start this request */
1223                 startstop = start_request(drive, rq);
1224                 spin_lock_irq(&ide_lock);
1225                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1226                         enable_irq(hwif->irq);
1227                 if (startstop == ide_stopped)
1228                         hwgroup->busy = 0;
1229         }
1230 }
1231
1232 /*
1233  * Passes the stuff to ide_do_request
1234  */
1235 void do_ide_request(request_queue_t *q)
1236 {
1237         ide_drive_t *drive = q->queuedata;
1238
1239         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1240 }
1241
1242 /*
1243  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1244  * retry the current request in pio mode instead of risking tossing it
1245  * all away
1246  */
1247 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1248 {
1249         ide_hwif_t *hwif = HWIF(drive);
1250         struct request *rq;
1251         ide_startstop_t ret = ide_stopped;
1252
1253         /*
1254          * end current dma transaction
1255          */
1256
1257         if (error < 0) {
1258                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1259                 (void)HWIF(drive)->ide_dma_end(drive);
1260                 ret = ide_error(drive, "dma timeout error",
1261                                                 hwif->INB(IDE_STATUS_REG));
1262         } else {
1263                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1264                 (void) hwif->ide_dma_timeout(drive);
1265         }
1266
1267         /*
1268          * disable dma for now, but remember that we did so because of
1269          * a timeout -- we'll reenable after we finish this next request
1270          * (or rather the first chunk of it) in pio.
1271          */
1272         drive->retry_pio++;
1273         drive->state = DMA_PIO_RETRY;
1274         (void) hwif->ide_dma_off_quietly(drive);
1275
1276         /*
1277          * un-busy drive etc (hwgroup->busy is cleared on return) and
1278          * make sure request is sane
1279          */
1280         rq = HWGROUP(drive)->rq;
1281         HWGROUP(drive)->rq = NULL;
1282
1283         rq->errors = 0;
1284
1285         if (!rq->bio)
1286                 goto out;
1287
1288         rq->sector = rq->bio->bi_sector;
1289         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1290         rq->hard_cur_sectors = rq->current_nr_sectors;
1291         rq->buffer = bio_data(rq->bio);
1292 out:
1293         return ret;
1294 }
1295
1296 /**
1297  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1298  *      @data: timer callback magic (hwgroup)
1299  *
1300  *      An IDE command has timed out before the expected drive return
1301  *      occurred. At this point we attempt to clean up the current
1302  *      mess. If the current handler includes an expiry handler then
1303  *      we invoke the expiry handler, and providing it is happy the
1304  *      work is done. If that fails we apply generic recovery rules
1305  *      invoking the handler and checking the drive DMA status. We
1306  *      have an excessively incestuous relationship with the DMA
1307  *      logic that wants cleaning up.
1308  */
1309
1310 void ide_timer_expiry (unsigned long data)
1311 {
1312         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1313         ide_handler_t   *handler;
1314         ide_expiry_t    *expiry;
1315         unsigned long   flags;
1316         unsigned long   wait = -1;
1317
1318         spin_lock_irqsave(&ide_lock, flags);
1319
1320         if ((handler = hwgroup->handler) == NULL) {
1321                 /*
1322                  * Either a marginal timeout occurred
1323                  * (got the interrupt just as timer expired),
1324                  * or we were "sleeping" to give other devices a chance.
1325                  * Either way, we don't really want to complain about anything.
1326                  */
1327                 if (hwgroup->sleeping) {
1328                         hwgroup->sleeping = 0;
1329                         hwgroup->busy = 0;
1330                 }
1331         } else {
1332                 ide_drive_t *drive = hwgroup->drive;
1333                 if (!drive) {
1334                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1335                         hwgroup->handler = NULL;
1336                 } else {
1337                         ide_hwif_t *hwif;
1338                         ide_startstop_t startstop = ide_stopped;
1339                         if (!hwgroup->busy) {
1340                                 hwgroup->busy = 1;      /* paranoia */
1341                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1342                         }
1343                         if ((expiry = hwgroup->expiry) != NULL) {
1344                                 /* continue */
1345                                 if ((wait = expiry(drive)) > 0) {
1346                                         /* reset timer */
1347                                         hwgroup->timer.expires  = jiffies + wait;
1348                                         add_timer(&hwgroup->timer);
1349                                         spin_unlock_irqrestore(&ide_lock, flags);
1350                                         return;
1351                                 }
1352                         }
1353                         hwgroup->handler = NULL;
1354                         /*
1355                          * We need to simulate a real interrupt when invoking
1356                          * the handler() function, which means we need to
1357                          * globally mask the specific IRQ:
1358                          */
1359                         spin_unlock(&ide_lock);
1360                         hwif  = HWIF(drive);
1361 #if DISABLE_IRQ_NOSYNC
1362                         disable_irq_nosync(hwif->irq);
1363 #else
1364                         /* disable_irq_nosync ?? */
1365                         disable_irq(hwif->irq);
1366 #endif /* DISABLE_IRQ_NOSYNC */
1367                         /* local CPU only,
1368                          * as if we were handling an interrupt */
1369                         local_irq_disable();
1370                         if (hwgroup->polling) {
1371                                 startstop = handler(drive);
1372                         } else if (drive_is_ready(drive)) {
1373                                 if (drive->waiting_for_dma)
1374                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1375                                 (void)ide_ack_intr(hwif);
1376                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1377                                 startstop = handler(drive);
1378                         } else {
1379                                 if (drive->waiting_for_dma) {
1380                                         startstop = ide_dma_timeout_retry(drive, wait);
1381                                 } else
1382                                         startstop =
1383                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1384                         }
1385                         drive->service_time = jiffies - drive->service_start;
1386                         spin_lock_irq(&ide_lock);
1387                         enable_irq(hwif->irq);
1388                         if (startstop == ide_stopped)
1389                                 hwgroup->busy = 0;
1390                 }
1391         }
1392         ide_do_request(hwgroup, IDE_NO_IRQ);
1393         spin_unlock_irqrestore(&ide_lock, flags);
1394 }
1395
1396 /**
1397  *      unexpected_intr         -       handle an unexpected IDE interrupt
1398  *      @irq: interrupt line
1399  *      @hwgroup: hwgroup being processed
1400  *
1401  *      There's nothing really useful we can do with an unexpected interrupt,
1402  *      other than reading the status register (to clear it), and logging it.
1403  *      There should be no way that an irq can happen before we're ready for it,
1404  *      so we needn't worry much about losing an "important" interrupt here.
1405  *
1406  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1407  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1408  *      looks "good", we just ignore the interrupt completely.
1409  *
1410  *      This routine assumes __cli() is in effect when called.
1411  *
1412  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1413  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1414  *      we could screw up by interfering with a new request being set up for
1415  *      irq15.
1416  *
1417  *      In reality, this is a non-issue.  The new command is not sent unless
1418  *      the drive is ready to accept one, in which case we know the drive is
1419  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1420  *      before completing the issuance of any new drive command, so we will not
1421  *      be accidentally invoked as a result of any valid command completion
1422  *      interrupt.
1423  *
1424  *      Note that we must walk the entire hwgroup here. We know which hwif
1425  *      is doing the current command, but we don't know which hwif burped
1426  *      mysteriously.
1427  */
1428
1429 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1430 {
1431         u8 stat;
1432         ide_hwif_t *hwif = hwgroup->hwif;
1433
1434         /*
1435          * handle the unexpected interrupt
1436          */
1437         do {
1438                 if (hwif->irq == irq) {
1439                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1440                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1441                                 /* Try to not flood the console with msgs */
1442                                 static unsigned long last_msgtime, count;
1443                                 ++count;
1444                                 if (time_after(jiffies, last_msgtime + HZ)) {
1445                                         last_msgtime = jiffies;
1446                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1447                                                 "status=0x%02x, count=%ld\n",
1448                                                 hwif->name,
1449                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1450                                 }
1451                         }
1452                 }
1453         } while ((hwif = hwif->next) != hwgroup->hwif);
1454 }
1455
1456 /**
1457  *      ide_intr        -       default IDE interrupt handler
1458  *      @irq: interrupt number
1459  *      @dev_id: hwif group
1460  *      @regs: unused weirdness from the kernel irq layer
1461  *
1462  *      This is the default IRQ handler for the IDE layer. You should
1463  *      not need to override it. If you do be aware it is subtle in
1464  *      places
1465  *
1466  *      hwgroup->hwif is the interface in the group currently performing
1467  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1468  *      the IRQ handler to call. As we issue a command the handlers
1469  *      step through multiple states, reassigning the handler to the
1470  *      next step in the process. Unlike a smart SCSI controller IDE
1471  *      expects the main processor to sequence the various transfer
1472  *      stages. We also manage a poll timer to catch up with most
1473  *      timeout situations. There are still a few where the handlers
1474  *      don't ever decide to give up.
1475  *
1476  *      The handler eventually returns ide_stopped to indicate the
1477  *      request completed. At this point we issue the next request
1478  *      on the hwgroup and the process begins again.
1479  */
1480
1481 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1482 {
1483         unsigned long flags;
1484         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1485         ide_hwif_t *hwif;
1486         ide_drive_t *drive;
1487         ide_handler_t *handler;
1488         ide_startstop_t startstop;
1489
1490         spin_lock_irqsave(&ide_lock, flags);
1491         hwif = hwgroup->hwif;
1492
1493         if (!ide_ack_intr(hwif)) {
1494                 spin_unlock_irqrestore(&ide_lock, flags);
1495                 return IRQ_NONE;
1496         }
1497
1498         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1499                 /*
1500                  * Not expecting an interrupt from this drive.
1501                  * That means this could be:
1502                  *      (1) an interrupt from another PCI device
1503                  *      sharing the same PCI INT# as us.
1504                  * or   (2) a drive just entered sleep or standby mode,
1505                  *      and is interrupting to let us know.
1506                  * or   (3) a spurious interrupt of unknown origin.
1507                  *
1508                  * For PCI, we cannot tell the difference,
1509                  * so in that case we just ignore it and hope it goes away.
1510                  *
1511                  * FIXME: unexpected_intr should be hwif-> then we can
1512                  * remove all the ifdef PCI crap
1513                  */
1514 #ifdef CONFIG_BLK_DEV_IDEPCI
1515                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1516 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1517                 {
1518                         /*
1519                          * Probably not a shared PCI interrupt,
1520                          * so we can safely try to do something about it:
1521                          */
1522                         unexpected_intr(irq, hwgroup);
1523 #ifdef CONFIG_BLK_DEV_IDEPCI
1524                 } else {
1525                         /*
1526                          * Whack the status register, just in case
1527                          * we have a leftover pending IRQ.
1528                          */
1529                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1530 #endif /* CONFIG_BLK_DEV_IDEPCI */
1531                 }
1532                 spin_unlock_irqrestore(&ide_lock, flags);
1533                 return IRQ_NONE;
1534         }
1535         drive = hwgroup->drive;
1536         if (!drive) {
1537                 /*
1538                  * This should NEVER happen, and there isn't much
1539                  * we could do about it here.
1540                  *
1541                  * [Note - this can occur if the drive is hot unplugged]
1542                  */
1543                 spin_unlock_irqrestore(&ide_lock, flags);
1544                 return IRQ_HANDLED;
1545         }
1546         if (!drive_is_ready(drive)) {
1547                 /*
1548                  * This happens regularly when we share a PCI IRQ with
1549                  * another device.  Unfortunately, it can also happen
1550                  * with some buggy drives that trigger the IRQ before
1551                  * their status register is up to date.  Hopefully we have
1552                  * enough advance overhead that the latter isn't a problem.
1553                  */
1554                 spin_unlock_irqrestore(&ide_lock, flags);
1555                 return IRQ_NONE;
1556         }
1557         if (!hwgroup->busy) {
1558                 hwgroup->busy = 1;      /* paranoia */
1559                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1560         }
1561         hwgroup->handler = NULL;
1562         del_timer(&hwgroup->timer);
1563         spin_unlock(&ide_lock);
1564
1565         if (drive->unmask)
1566                 local_irq_enable();
1567         /* service this interrupt, may set handler for next interrupt */
1568         startstop = handler(drive);
1569         spin_lock_irq(&ide_lock);
1570
1571         /*
1572          * Note that handler() may have set things up for another
1573          * interrupt to occur soon, but it cannot happen until
1574          * we exit from this routine, because it will be the
1575          * same irq as is currently being serviced here, and Linux
1576          * won't allow another of the same (on any CPU) until we return.
1577          */
1578         drive->service_time = jiffies - drive->service_start;
1579         if (startstop == ide_stopped) {
1580                 if (hwgroup->handler == NULL) { /* paranoia */
1581                         hwgroup->busy = 0;
1582                         ide_do_request(hwgroup, hwif->irq);
1583                 } else {
1584                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1585                                 "on exit\n", drive->name);
1586                 }
1587         }
1588         spin_unlock_irqrestore(&ide_lock, flags);
1589         return IRQ_HANDLED;
1590 }
1591
1592 /**
1593  *      ide_init_drive_cmd      -       initialize a drive command request
1594  *      @rq: request object
1595  *
1596  *      Initialize a request before we fill it in and send it down to
1597  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1598  *      now it doesn't do a lot, but if that changes abusers will have a
1599  *      nasty suprise.
1600  */
1601
1602 void ide_init_drive_cmd (struct request *rq)
1603 {
1604         memset(rq, 0, sizeof(*rq));
1605         rq->flags = REQ_DRIVE_CMD;
1606         rq->ref_count = 1;
1607 }
1608
1609 EXPORT_SYMBOL(ide_init_drive_cmd);
1610
1611 /**
1612  *      ide_do_drive_cmd        -       issue IDE special command
1613  *      @drive: device to issue command
1614  *      @rq: request to issue
1615  *      @action: action for processing
1616  *
1617  *      This function issues a special IDE device request
1618  *      onto the request queue.
1619  *
1620  *      If action is ide_wait, then the rq is queued at the end of the
1621  *      request queue, and the function sleeps until it has been processed.
1622  *      This is for use when invoked from an ioctl handler.
1623  *
1624  *      If action is ide_preempt, then the rq is queued at the head of
1625  *      the request queue, displacing the currently-being-processed
1626  *      request and this function returns immediately without waiting
1627  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1628  *      intended for careful use by the ATAPI tape/cdrom driver code.
1629  *
1630  *      If action is ide_end, then the rq is queued at the end of the
1631  *      request queue, and the function returns immediately without waiting
1632  *      for the new rq to be completed. This is again intended for careful
1633  *      use by the ATAPI tape/cdrom driver code.
1634  */
1635
1636 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1637 {
1638         unsigned long flags;
1639         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1640         DECLARE_COMPLETION(wait);
1641         int where = ELEVATOR_INSERT_BACK, err;
1642         int must_wait = (action == ide_wait || action == ide_head_wait);
1643
1644         rq->errors = 0;
1645         rq->rq_status = RQ_ACTIVE;
1646
1647         /*
1648          * we need to hold an extra reference to request for safe inspection
1649          * after completion
1650          */
1651         if (must_wait) {
1652                 rq->ref_count++;
1653                 rq->waiting = &wait;
1654                 rq->end_io = blk_end_sync_rq;
1655         }
1656
1657         spin_lock_irqsave(&ide_lock, flags);
1658         if (action == ide_preempt)
1659                 hwgroup->rq = NULL;
1660         if (action == ide_preempt || action == ide_head_wait) {
1661                 where = ELEVATOR_INSERT_FRONT;
1662                 rq->flags |= REQ_PREEMPT;
1663         }
1664         __elv_add_request(drive->queue, rq, where, 0);
1665         ide_do_request(hwgroup, IDE_NO_IRQ);
1666         spin_unlock_irqrestore(&ide_lock, flags);
1667
1668         err = 0;
1669         if (must_wait) {
1670                 wait_for_completion(&wait);
1671                 rq->waiting = NULL;
1672                 if (rq->errors)
1673                         err = -EIO;
1674
1675                 blk_put_request(rq);
1676         }
1677
1678         return err;
1679 }
1680
1681 EXPORT_SYMBOL(ide_do_drive_cmd);