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 void ide_softirq_done(struct request *rq)
  59 {
  60         request_queue_t *q = rq->q;
  61
  62         add_disk_randomness(rq->rq_disk);
  63         end_that_request_chunk(rq, 1, rq->data_len);
  64
  65         spin_lock_irq(q->queue_lock);
  66         end_that_request_last(rq, 1);
  67         spin_unlock_irq(q->queue_lock);
  68 }
  69
  70 int __ide_end_request(ide_drive_t *drive, struct request *rq, int uptodate,
  71                       int nr_sectors)
  72 {
  73         unsigned int nbytes;
  74         int ret = 1;
  75
  76         BUG_ON(!(rq->flags & REQ_STARTED));
  77
  78         /*
  79          * if failfast is set on a request, override number of sectors and
  80          * complete the whole request right now
  81          */
  82         if (blk_noretry_request(rq) && end_io_error(uptodate))
  83                 nr_sectors = rq->hard_nr_sectors;
  84
  85         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
  86                 rq->errors = -EIO;
  87
  88         /*
  89          * decide whether to reenable DMA -- 3 is a random magic for now,
  90          * if we DMA timeout more than 3 times, just stay in PIO
  91          */
  92         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
  93                 drive->state = 0;
  94                 HWGROUP(drive)->hwif->ide_dma_on(drive);
  95         }
  96
  97         /*
  98          * For partial completions (or non fs/pc requests), use the regular
  99          * direct completion path. Same thing for requests that failed, to
 100          * preserve the ->errors value we use the normal completion path
 101          * for those
 102          */
 103         nbytes = nr_sectors << 9;
 104         if (!rq->errors && rq_all_done(rq, nbytes)) {
 105                 rq->data_len = nbytes;
 106                 blkdev_dequeue_request(rq);
 107                 HWGROUP(drive)->rq = NULL;
 108                 blk_complete_request(rq);
 109                 ret = 0;
 110         } else {
 111                 if (!end_that_request_first(rq, uptodate, nr_sectors)) {
 112                         add_disk_randomness(rq->rq_disk);
 113                         blkdev_dequeue_request(rq);
 114                         HWGROUP(drive)->rq = NULL;
 115                         end_that_request_last(rq, uptodate);
 116                         ret = 0;
 117                 }
 118         }
 119
 120         return ret;
 121 }
 122 EXPORT_SYMBOL(__ide_end_request);
 123
 124 /**
 125  *      ide_end_request         -       complete an IDE I/O
 126  *      @drive: IDE device for the I/O
 127  *      @uptodate:
 128  *      @nr_sectors: number of sectors completed
 129  *
 130  *      This is our end_request wrapper function. We complete the I/O
 131  *      update random number input and dequeue the request, which if
 132  *      it was tagged may be out of order.
 133  */
 134
 135 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
 136 {
 137         struct request *rq;
 138         unsigned long flags;
 139         int ret = 1;
 140
 141         /*
 142          * room for locking improvements here, the calls below don't
 143          * need the queue lock held at all
 144          */
 145         spin_lock_irqsave(&ide_lock, flags);
 146         rq = HWGROUP(drive)->rq;
 147
 148         if (!nr_sectors)
 149                 nr_sectors = rq->hard_cur_sectors;
 150
 151         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
 152
 153         spin_unlock_irqrestore(&ide_lock, flags);
 154         return ret;
 155 }
 156 EXPORT_SYMBOL(ide_end_request);
 157
 158 /*
 159  * Power Management state machine. This one is rather trivial for now,
 160  * we should probably add more, like switching back to PIO on suspend
 161  * to help some BIOSes, re-do the door locking on resume, etc...
 162  */
 163
 164 enum {
 165         ide_pm_flush_cache      = ide_pm_state_start_suspend,
 166         idedisk_pm_standby,
 167
 168         idedisk_pm_idle         = ide_pm_state_start_resume,
 169         ide_pm_restore_dma,
 170 };
 171
 172 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
 173 {
 174         if (drive->media != ide_disk)
 175                 return;
 176
 177         switch (rq->pm->pm_step) {
 178         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
 179                 if (rq->pm->pm_state == PM_EVENT_FREEZE)
 180                         rq->pm->pm_step = ide_pm_state_completed;
 181                 else
 182                         rq->pm->pm_step = idedisk_pm_standby;
 183                 break;
 184         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
 185                 rq->pm->pm_step = ide_pm_state_completed;
 186                 break;
 187         case idedisk_pm_idle:           /* Resume step 1 (idle) complete */
 188                 rq->pm->pm_step = ide_pm_restore_dma;
 189                 break;
 190         }
 191 }
 192
 193 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
 194 {
 195         ide_task_t *args = rq->special;
 196
 197         memset(args, 0, sizeof(*args));
 198
 199         if (drive->media != ide_disk) {
 200                 /* skip idedisk_pm_idle for ATAPI devices */
 201                 if (rq->pm->pm_step == idedisk_pm_idle)
 202                         rq->pm->pm_step = ide_pm_restore_dma;
 203         }
 204
 205         switch (rq->pm->pm_step) {
 206         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
 207                 if (drive->media != ide_disk)
 208                         break;
 209                 /* Not supported? Switch to next step now. */
 210                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
 211                         ide_complete_power_step(drive, rq, 0, 0);
 212                         return ide_stopped;
 213                 }
 214                 if (ide_id_has_flush_cache_ext(drive->id))
 215                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
 216                 else
 217                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
 218                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 219                 args->handler      = &task_no_data_intr;
 220                 return do_rw_taskfile(drive, args);
 221
 222         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
 223                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
 224                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 225                 args->handler      = &task_no_data_intr;
 226                 return do_rw_taskfile(drive, args);
 227
 228         case idedisk_pm_idle:           /* Resume step 1 (idle) */
 229                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
 230                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 231                 args->handler = task_no_data_intr;
 232                 return do_rw_taskfile(drive, args);
 233
 234         case ide_pm_restore_dma:        /* Resume step 2 (restore DMA) */
 235                 /*
 236                  * Right now, all we do is call hwif->ide_dma_check(drive),
 237                  * we could be smarter and check for current xfer_speed
 238                  * in struct drive etc...
 239                  */
 240                 if ((drive->id->capability & 1) == 0)
 241                         break;
 242                 if (drive->hwif->ide_dma_check == NULL)
 243                         break;
 244                 drive->hwif->ide_dma_check(drive);
 245                 break;
 246         }
 247         rq->pm->pm_step = ide_pm_state_completed;
 248         return ide_stopped;
 249 }
 250
 251 /**
 252  *      ide_complete_pm_request - end the current Power Management request
 253  *      @drive: target drive
 254  *      @rq: request
 255  *
 256  *      This function cleans up the current PM request and stops the queue
 257  *      if necessary.
 258  */
 259 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
 260 {
 261         unsigned long flags;
 262
 263 #ifdef DEBUG_PM
 264         printk("%s: completing PM request, %s\n", drive->name,
 265                blk_pm_suspend_request(rq) ? "suspend" : "resume");
 266 #endif
 267         spin_lock_irqsave(&ide_lock, flags);
 268         if (blk_pm_suspend_request(rq)) {
 269                 blk_stop_queue(drive->queue);
 270         } else {
 271                 drive->blocked = 0;
 272                 blk_start_queue(drive->queue);
 273         }
 274         blkdev_dequeue_request(rq);
 275         HWGROUP(drive)->rq = NULL;
 276         end_that_request_last(rq, 1);
 277         spin_unlock_irqrestore(&ide_lock, flags);
 278 }
 279
 280 /*
 281  * FIXME: probably move this somewhere else, name is bad too :)
 282  */
 283 u64 ide_get_error_location(ide_drive_t *drive, char *args)
 284 {
 285         u32 high, low;
 286         u8 hcyl, lcyl, sect;
 287         u64 sector;
 288
 289         high = 0;
 290         hcyl = args[5];
 291         lcyl = args[4];
 292         sect = args[3];
 293
 294         if (ide_id_has_flush_cache_ext(drive->id)) {
 295                 low = (hcyl << 16) | (lcyl << 8) | sect;
 296                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 297                 high = ide_read_24(drive);
 298         } else {
 299                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
 300                 if (cur & 0x40) {
 301                         high = cur & 0xf;
 302                         low = (hcyl << 16) | (lcyl << 8) | sect;
 303                 } else {
 304                         low = hcyl * drive->head * drive->sect;
 305                         low += lcyl * drive->sect;
 306                         low += sect - 1;
 307                 }
 308         }
 309
 310         sector = ((u64) high << 24) | low;
 311         return sector;
 312 }
 313 EXPORT_SYMBOL(ide_get_error_location);
 314
 315 /**
 316  *      ide_end_drive_cmd       -       end an explicit drive command
 317  *      @drive: command
 318  *      @stat: status bits
 319  *      @err: error bits
 320  *
 321  *      Clean up after success/failure of an explicit drive command.
 322  *      These get thrown onto the queue so they are synchronized with
 323  *      real I/O operations on the drive.
 324  *
 325  *      In LBA48 mode we have to read the register set twice to get
 326  *      all the extra information out.
 327  */
 328
 329 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
 330 {
 331         ide_hwif_t *hwif = HWIF(drive);
 332         unsigned long flags;
 333         struct request *rq;
 334
 335         spin_lock_irqsave(&ide_lock, flags);
 336         rq = HWGROUP(drive)->rq;
 337         spin_unlock_irqrestore(&ide_lock, flags);
 338
 339         if (rq->flags & REQ_DRIVE_CMD) {
 340                 u8 *args = (u8 *) rq->buffer;
 341                 if (rq->errors == 0)
 342                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 343
 344                 if (args) {
 345                         args[0] = stat;
 346                         args[1] = err;
 347                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 348                 }
 349         } else if (rq->flags & REQ_DRIVE_TASK) {
 350                 u8 *args = (u8 *) rq->buffer;
 351                 if (rq->errors == 0)
 352                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 353
 354                 if (args) {
 355                         args[0] = stat;
 356                         args[1] = err;
 357                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 358                         args[3] = hwif->INB(IDE_SECTOR_REG);
 359                         args[4] = hwif->INB(IDE_LCYL_REG);
 360                         args[5] = hwif->INB(IDE_HCYL_REG);
 361                         args[6] = hwif->INB(IDE_SELECT_REG);
 362                 }
 363         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
 364                 ide_task_t *args = (ide_task_t *) rq->special;
 365                 if (rq->errors == 0)
 366                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 367
 368                 if (args) {
 369                         if (args->tf_in_flags.b.data) {
 370                                 u16 data                                = hwif->INW(IDE_DATA_REG);
 371                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
 372                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
 373                         }
 374                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
 375                         /* be sure we're looking at the low order bits */
 376                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
 377                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
 378                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
 379                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
 380                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
 381                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
 382                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
 383
 384                         if (drive->addressing == 1) {
 385                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 386                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
 387                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
 388                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
 389                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
 390                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
 391                         }
 392                 }
 393         } else if (blk_pm_request(rq)) {
 394 #ifdef DEBUG_PM
 395                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
 396                         drive->name, rq->pm->pm_step, stat, err);
 397 #endif
 398                 ide_complete_power_step(drive, rq, stat, err);
 399                 if (rq->pm->pm_step == ide_pm_state_completed)
 400                         ide_complete_pm_request(drive, rq);
 401                 return;
 402         }
 403
 404         spin_lock_irqsave(&ide_lock, flags);
 405         blkdev_dequeue_request(rq);
 406         HWGROUP(drive)->rq = NULL;
 407         rq->errors = err;
 408         end_that_request_last(rq, !rq->errors);
 409         spin_unlock_irqrestore(&ide_lock, flags);
 410 }
 411
 412 EXPORT_SYMBOL(ide_end_drive_cmd);
 413
 414 /**
 415  *      try_to_flush_leftover_data      -       flush junk
 416  *      @drive: drive to flush
 417  *
 418  *      try_to_flush_leftover_data() is invoked in response to a drive
 419  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
 420  *      resetting the drive, this routine tries to clear the condition
 421  *      by read a sector's worth of data from the drive.  Of course,
 422  *      this may not help if the drive is *waiting* for data from *us*.
 423  */
 424 static void try_to_flush_leftover_data (ide_drive_t *drive)
 425 {
 426         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
 427
 428         if (drive->media != ide_disk)
 429                 return;
 430         while (i > 0) {
 431                 u32 buffer[16];
 432                 u32 wcount = (i > 16) ? 16 : i;
 433
 434                 i -= wcount;
 435                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
 436         }
 437 }
 438
 439 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
 440 {
 441         if (rq->rq_disk) {
 442                 ide_driver_t *drv;
 443
 444                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 445                 drv->end_request(drive, 0, 0);
 446         } else
 447                 ide_end_request(drive, 0, 0);
 448 }
 449
 450 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 451 {
 452         ide_hwif_t *hwif = drive->hwif;
 453
 454         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 455                 /* other bits are useless when BUSY */
 456                 rq->errors |= ERROR_RESET;
 457         } else if (stat & ERR_STAT) {
 458                 /* err has different meaning on cdrom and tape */
 459                 if (err == ABRT_ERR) {
 460                         if (drive->select.b.lba &&
 461                             /* some newer drives don't support WIN_SPECIFY */
 462                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
 463                                 return ide_stopped;
 464                 } else if ((err & BAD_CRC) == BAD_CRC) {
 465                         /* UDMA crc error, just retry the operation */
 466                         drive->crc_count++;
 467                 } else if (err & (BBD_ERR | ECC_ERR)) {
 468                         /* retries won't help these */
 469                         rq->errors = ERROR_MAX;
 470                 } else if (err & TRK0_ERR) {
 471                         /* help it find track zero */
 472                         rq->errors |= ERROR_RECAL;
 473                 }
 474         }
 475
 476         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ)
 477                 try_to_flush_leftover_data(drive);
 478
 479         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 480                 /* force an abort */
 481                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 482
 483         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
 484                 ide_kill_rq(drive, rq);
 485         else {
 486                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 487                         ++rq->errors;
 488                         return ide_do_reset(drive);
 489                 }
 490                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
 491                         drive->special.b.recalibrate = 1;
 492                 ++rq->errors;
 493         }
 494         return ide_stopped;
 495 }
 496
 497 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 498 {
 499         ide_hwif_t *hwif = drive->hwif;
 500
 501         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 502                 /* other bits are useless when BUSY */
 503                 rq->errors |= ERROR_RESET;
 504         } else {
 505                 /* add decoding error stuff */
 506         }
 507
 508         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 509                 /* force an abort */
 510                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 511
 512         if (rq->errors >= ERROR_MAX) {
 513                 ide_kill_rq(drive, rq);
 514         } else {
 515                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 516                         ++rq->errors;
 517                         return ide_do_reset(drive);
 518                 }
 519                 ++rq->errors;
 520         }
 521
 522         return ide_stopped;
 523 }
 524
 525 ide_startstop_t
 526 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 527 {
 528         if (drive->media == ide_disk)
 529                 return ide_ata_error(drive, rq, stat, err);
 530         return ide_atapi_error(drive, rq, stat, err);
 531 }
 532
 533 EXPORT_SYMBOL_GPL(__ide_error);
 534
 535 /**
 536  *      ide_error       -       handle an error on the IDE
 537  *      @drive: drive the error occurred on
 538  *      @msg: message to report
 539  *      @stat: status bits
 540  *
 541  *      ide_error() takes action based on the error returned by the drive.
 542  *      For normal I/O that may well include retries. We deal with
 543  *      both new-style (taskfile) and old style command handling here.
 544  *      In the case of taskfile command handling there is work left to
 545  *      do
 546  */
 547
 548 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
 549 {
 550         struct request *rq;
 551         u8 err;
 552
 553         err = ide_dump_status(drive, msg, stat);
 554
 555         if ((rq = HWGROUP(drive)->rq) == NULL)
 556                 return ide_stopped;
 557
 558         /* retry only "normal" I/O: */
 559         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 560                 rq->errors = 1;
 561                 ide_end_drive_cmd(drive, stat, err);
 562                 return ide_stopped;
 563         }
 564
 565         if (rq->rq_disk) {
 566                 ide_driver_t *drv;
 567
 568                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 569                 return drv->error(drive, rq, stat, err);
 570         } else
 571                 return __ide_error(drive, rq, stat, err);
 572 }
 573
 574 EXPORT_SYMBOL_GPL(ide_error);
 575
 576 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
 577 {
 578         if (drive->media != ide_disk)
 579                 rq->errors |= ERROR_RESET;
 580
 581         ide_kill_rq(drive, rq);
 582
 583         return ide_stopped;
 584 }
 585
 586 EXPORT_SYMBOL_GPL(__ide_abort);
 587
 588 /**
 589  *      ide_abort       -       abort pending IDE operations
 590  *      @drive: drive the error occurred on
 591  *      @msg: message to report
 592  *
 593  *      ide_abort kills and cleans up when we are about to do a
 594  *      host initiated reset on active commands. Longer term we
 595  *      want handlers to have sensible abort handling themselves
 596  *
 597  *      This differs fundamentally from ide_error because in
 598  *      this case the command is doing just fine when we
 599  *      blow it away.
 600  */
 601
 602 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
 603 {
 604         struct request *rq;
 605
 606         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
 607                 return ide_stopped;
 608
 609         /* retry only "normal" I/O: */
 610         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 611                 rq->errors = 1;
 612                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
 613                 return ide_stopped;
 614         }
 615
 616         if (rq->rq_disk) {
 617                 ide_driver_t *drv;
 618
 619                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 620                 return drv->abort(drive, rq);
 621         } else
 622                 return __ide_abort(drive, rq);
 623 }
 624
 625 /**
 626  *      ide_cmd         -       issue a simple drive command
 627  *      @drive: drive the command is for
 628  *      @cmd: command byte
 629  *      @nsect: sector byte
 630  *      @handler: handler for the command completion
 631  *
 632  *      Issue a simple drive command with interrupts.
 633  *      The drive must be selected beforehand.
 634  */
 635
 636 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
 637                 ide_handler_t *handler)
 638 {
 639         ide_hwif_t *hwif = HWIF(drive);
 640         if (IDE_CONTROL_REG)
 641                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
 642         SELECT_MASK(drive,0);
 643         hwif->OUTB(nsect,IDE_NSECTOR_REG);
 644         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
 645 }
 646
 647 /**
 648  *      drive_cmd_intr          -       drive command completion interrupt
 649  *      @drive: drive the completion interrupt occurred on
 650  *
 651  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
 652  *      We do any necessary data reading and then wait for the drive to
 653  *      go non busy. At that point we may read the error data and complete
 654  *      the request
 655  */
 656
 657 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
 658 {
 659         struct request *rq = HWGROUP(drive)->rq;
 660         ide_hwif_t *hwif = HWIF(drive);
 661         u8 *args = (u8 *) rq->buffer;
 662         u8 stat = hwif->INB(IDE_STATUS_REG);
 663         int retries = 10;
 664
 665         local_irq_enable();
 666         if ((stat & DRQ_STAT) && args && args[3]) {
 667                 u8 io_32bit = drive->io_32bit;
 668                 drive->io_32bit = 0;
 669                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
 670                 drive->io_32bit = io_32bit;
 671                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
 672                         udelay(100);
 673         }
 674
 675         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
 676                 return ide_error(drive, "drive_cmd", stat);
 677                 /* calls ide_end_drive_cmd */
 678         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
 679         return ide_stopped;
 680 }
 681
 682 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
 683 {
 684         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
 685         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
 686         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
 687         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
 688         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
 689         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
 690
 691         task->handler = &set_geometry_intr;
 692 }
 693
 694 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
 695 {
 696         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
 697         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
 698
 699         task->handler = &recal_intr;
 700 }
 701
 702 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
 703 {
 704         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
 705         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
 706
 707         task->handler = &set_multmode_intr;
 708 }
 709
 710 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
 711 {
 712         special_t *s = &drive->special;
 713         ide_task_t args;
 714
 715         memset(&args, 0, sizeof(ide_task_t));
 716         args.command_type = IDE_DRIVE_TASK_NO_DATA;
 717
 718         if (s->b.set_geometry) {
 719                 s->b.set_geometry = 0;
 720                 ide_init_specify_cmd(drive, &args);
 721         } else if (s->b.recalibrate) {
 722                 s->b.recalibrate = 0;
 723                 ide_init_restore_cmd(drive, &args);
 724         } else if (s->b.set_multmode) {
 725                 s->b.set_multmode = 0;
 726                 if (drive->mult_req > drive->id->max_multsect)
 727                         drive->mult_req = drive->id->max_multsect;
 728                 ide_init_setmult_cmd(drive, &args);
 729         } else if (s->all) {
 730                 int special = s->all;
 731                 s->all = 0;
 732                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
 733                 return ide_stopped;
 734         }
 735
 736         do_rw_taskfile(drive, &args);
 737
 738         return ide_started;
 739 }
 740
 741 /**
 742  *      do_special              -       issue some special commands
 743  *      @drive: drive the command is for
 744  *
 745  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
 746  *      commands to a drive.  It used to do much more, but has been scaled
 747  *      back.
 748  */
 749
 750 static ide_startstop_t do_special (ide_drive_t *drive)
 751 {
 752         special_t *s = &drive->special;
 753
 754 #ifdef DEBUG
 755         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
 756 #endif
 757         if (s->b.set_tune) {
 758                 s->b.set_tune = 0;
 759                 if (HWIF(drive)->tuneproc != NULL)
 760                         HWIF(drive)->tuneproc(drive, drive->tune_req);
 761                 return ide_stopped;
 762         } else {
 763                 if (drive->media == ide_disk)
 764                         return ide_disk_special(drive);
 765
 766                 s->all = 0;
 767                 drive->mult_req = 0;
 768                 return ide_stopped;
 769         }
 770 }
 771
 772 void ide_map_sg(ide_drive_t *drive, struct request *rq)
 773 {
 774         ide_hwif_t *hwif = drive->hwif;
 775         struct scatterlist *sg = hwif->sg_table;
 776
 777         if (hwif->sg_mapped)    /* needed by ide-scsi */
 778                 return;
 779
 780         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
 781                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
 782         } else {
 783                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
 784                 hwif->sg_nents = 1;
 785         }
 786 }
 787
 788 EXPORT_SYMBOL_GPL(ide_map_sg);
 789
 790 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
 791 {
 792         ide_hwif_t *hwif = drive->hwif;
 793
 794         hwif->nsect = hwif->nleft = rq->nr_sectors;
 795         hwif->cursg = hwif->cursg_ofs = 0;
 796 }
 797
 798 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
 799
 800 /**
 801  *      execute_drive_command   -       issue special drive command
 802  *      @drive: the drive to issue the command on
 803  *      @rq: the request structure holding the command
 804  *
 805  *      execute_drive_cmd() issues a special drive command,  usually
 806  *      initiated by ioctl() from the external hdparm program. The
 807  *      command can be a drive command, drive task or taskfile
 808  *      operation. Weirdly you can call it with NULL to wait for
 809  *      all commands to finish. Don't do this as that is due to change
 810  */
 811
 812 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
 813                 struct request *rq)
 814 {
 815         ide_hwif_t *hwif = HWIF(drive);
 816         if (rq->flags & REQ_DRIVE_TASKFILE) {
 817                 ide_task_t *args = rq->special;
 818
 819                 if (!args)
 820                         goto done;
 821
 822                 hwif->data_phase = args->data_phase;
 823
 824                 switch (hwif->data_phase) {
 825                 case TASKFILE_MULTI_OUT:
 826                 case TASKFILE_OUT:
 827                 case TASKFILE_MULTI_IN:
 828                 case TASKFILE_IN:
 829                         ide_init_sg_cmd(drive, rq);
 830                         ide_map_sg(drive, rq);
 831                 default:
 832                         break;
 833                 }
 834
 835                 if (args->tf_out_flags.all != 0)
 836                         return flagged_taskfile(drive, args);
 837                 return do_rw_taskfile(drive, args);
 838         } else if (rq->flags & REQ_DRIVE_TASK) {
 839                 u8 *args = rq->buffer;
 840                 u8 sel;
 841
 842                 if (!args)
 843                         goto done;
 844 #ifdef DEBUG
 845                 printk("%s: DRIVE_TASK_CMD ", drive->name);
 846                 printk("cmd=0x%02x ", args[0]);
 847                 printk("fr=0x%02x ", args[1]);
 848                 printk("ns=0x%02x ", args[2]);
 849                 printk("sc=0x%02x ", args[3]);
 850                 printk("lcyl=0x%02x ", args[4]);
 851                 printk("hcyl=0x%02x ", args[5]);
 852                 printk("sel=0x%02x\n", args[6]);
 853 #endif
 854                 hwif->OUTB(args[1], IDE_FEATURE_REG);
 855                 hwif->OUTB(args[3], IDE_SECTOR_REG);
 856                 hwif->OUTB(args[4], IDE_LCYL_REG);
 857                 hwif->OUTB(args[5], IDE_HCYL_REG);
 858                 sel = (args[6] & ~0x10);
 859                 if (drive->select.b.unit)
 860                         sel |= 0x10;
 861                 hwif->OUTB(sel, IDE_SELECT_REG);
 862                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
 863                 return ide_started;
 864         } else if (rq->flags & REQ_DRIVE_CMD) {
 865                 u8 *args = rq->buffer;
 866
 867                 if (!args)
 868                         goto done;
 869 #ifdef DEBUG
 870                 printk("%s: DRIVE_CMD ", drive->name);
 871                 printk("cmd=0x%02x ", args[0]);
 872                 printk("sc=0x%02x ", args[1]);
 873                 printk("fr=0x%02x ", args[2]);
 874                 printk("xx=0x%02x\n", args[3]);
 875 #endif
 876                 if (args[0] == WIN_SMART) {
 877                         hwif->OUTB(0x4f, IDE_LCYL_REG);
 878                         hwif->OUTB(0xc2, IDE_HCYL_REG);
 879                         hwif->OUTB(args[2],IDE_FEATURE_REG);
 880                         hwif->OUTB(args[1],IDE_SECTOR_REG);
 881                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
 882                         return ide_started;
 883                 }
 884                 hwif->OUTB(args[2],IDE_FEATURE_REG);
 885                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
 886                 return ide_started;
 887         }
 888
 889 done:
 890         /*
 891          * NULL is actually a valid way of waiting for
 892          * all current requests to be flushed from the queue.
 893          */
 894 #ifdef DEBUG
 895         printk("%s: DRIVE_CMD (null)\n", drive->name);
 896 #endif
 897         ide_end_drive_cmd(drive,
 898                         hwif->INB(IDE_STATUS_REG),
 899                         hwif->INB(IDE_ERROR_REG));
 900         return ide_stopped;
 901 }
 902
 903 /**
 904  *      start_request   -       start of I/O and command issuing for IDE
 905  *
 906  *      start_request() initiates handling of a new I/O request. It
 907  *      accepts commands and I/O (read/write) requests. It also does
 908  *      the final remapping for weird stuff like EZDrive. Once
 909  *      device mapper can work sector level the EZDrive stuff can go away
 910  *
 911  *      FIXME: this function needs a rename
 912  */
 913
 914 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
 915 {
 916         ide_startstop_t startstop;
 917         sector_t block;
 918
 919         BUG_ON(!(rq->flags & REQ_STARTED));
 920
 921 #ifdef DEBUG
 922         printk("%s: start_request: current=0x%08lx\n",
 923                 HWIF(drive)->name, (unsigned long) rq);
 924 #endif
 925
 926         /* bail early if we've exceeded max_failures */
 927         if (drive->max_failures && (drive->failures > drive->max_failures)) {
 928                 goto kill_rq;
 929         }
 930
 931         block    = rq->sector;
 932         if (blk_fs_request(rq) &&
 933             (drive->media == ide_disk || drive->media == ide_floppy)) {
 934                 block += drive->sect0;
 935         }
 936         /* Yecch - this will shift the entire interval,
 937            possibly killing some innocent following sector */
 938         if (block == 0 && drive->remap_0_to_1 == 1)
 939                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
 940
 941         if (blk_pm_suspend_request(rq) &&
 942             rq->pm->pm_step == ide_pm_state_start_suspend)
 943                 /* Mark drive blocked when starting the suspend sequence. */
 944                 drive->blocked = 1;
 945         else if (blk_pm_resume_request(rq) &&
 946                  rq->pm->pm_step == ide_pm_state_start_resume) {
 947                 /*
 948                  * The first thing we do on wakeup is to wait for BSY bit to
 949                  * go away (with a looong timeout) as a drive on this hwif may
 950                  * just be POSTing itself.
 951                  * We do that before even selecting as the "other" device on
 952                  * the bus may be broken enough to walk on our toes at this
 953                  * point.
 954                  */
 955                 int rc;
 956 #ifdef DEBUG_PM
 957                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
 958 #endif
 959                 rc = ide_wait_not_busy(HWIF(drive), 35000);
 960                 if (rc)
 961                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
 962                 SELECT_DRIVE(drive);
 963                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
 964                 rc = ide_wait_not_busy(HWIF(drive), 10000);
 965                 if (rc)
 966                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
 967         }
 968
 969         SELECT_DRIVE(drive);
 970         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
 971                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
 972                 return startstop;
 973         }
 974         if (!drive->special.all) {
 975                 ide_driver_t *drv;
 976
 977                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
 978                         return execute_drive_cmd(drive, rq);
 979                 else if (rq->flags & REQ_DRIVE_TASKFILE)
 980                         return execute_drive_cmd(drive, rq);
 981                 else if (blk_pm_request(rq)) {
 982 #ifdef DEBUG_PM
 983                         printk("%s: start_power_step(step: %d)\n",
 984                                 drive->name, rq->pm->pm_step);
 985 #endif
 986                         startstop = ide_start_power_step(drive, rq);
 987                         if (startstop == ide_stopped &&
 988                             rq->pm->pm_step == ide_pm_state_completed)
 989                                 ide_complete_pm_request(drive, rq);
 990                         return startstop;
 991                 }
 992
 993                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 994                 return drv->do_request(drive, rq, block);
 995         }
 996         return do_special(drive);
 997 kill_rq:
 998         ide_kill_rq(drive, rq);
 999         return ide_stopped;
1000 }
1001
1002 /**
1003  *      ide_stall_queue         -       pause an IDE device
1004  *      @drive: drive to stall
1005  *      @timeout: time to stall for (jiffies)
1006  *
1007  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1008  *      to the hwgroup by sleeping for timeout jiffies.
1009  */
1010
1011 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1012 {
1013         if (timeout > WAIT_WORSTCASE)
1014                 timeout = WAIT_WORSTCASE;
1015         drive->sleep = timeout + jiffies;
1016         drive->sleeping = 1;
1017 }
1018
1019 EXPORT_SYMBOL(ide_stall_queue);
1020
1021 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1022
1023 /**
1024  *      choose_drive            -       select a drive to service
1025  *      @hwgroup: hardware group to select on
1026  *
1027  *      choose_drive() selects the next drive which will be serviced.
1028  *      This is necessary because the IDE layer can't issue commands
1029  *      to both drives on the same cable, unlike SCSI.
1030  */
1031
1032 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1033 {
1034         ide_drive_t *drive, *best;
1035
1036 repeat:
1037         best = NULL;
1038         drive = hwgroup->drive;
1039
1040         /*
1041          * drive is doing pre-flush, ordered write, post-flush sequence. even
1042          * though that is 3 requests, it must be seen as a single transaction.
1043          * we must not preempt this drive until that is complete
1044          */
1045         if (blk_queue_flushing(drive->queue)) {
1046                 /*
1047                  * small race where queue could get replugged during
1048                  * the 3-request flush cycle, just yank the plug since
1049                  * we want it to finish asap
1050                  */
1051                 blk_remove_plug(drive->queue);
1052                 return drive;
1053         }
1054
1055         do {
1056                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1057                     && !elv_queue_empty(drive->queue)) {
1058                         if (!best
1059                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1060                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1061                         {
1062                                 if (!blk_queue_plugged(drive->queue))
1063                                         best = drive;
1064                         }
1065                 }
1066         } while ((drive = drive->next) != hwgroup->drive);
1067         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1068                 long t = (signed long)(WAKEUP(best) - jiffies);
1069                 if (t >= WAIT_MIN_SLEEP) {
1070                 /*
1071                  * We *may* have some time to spare, but first let's see if
1072                  * someone can potentially benefit from our nice mood today..
1073                  */
1074                         drive = best->next;
1075                         do {
1076                                 if (!drive->sleeping
1077                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1078                                  && time_before(WAKEUP(drive), jiffies + t))
1079                                 {
1080                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1081                                         goto repeat;
1082                                 }
1083                         } while ((drive = drive->next) != best);
1084                 }
1085         }
1086         return best;
1087 }
1088
1089 /*
1090  * Issue a new request to a drive from hwgroup
1091  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1092  *
1093  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1094  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1095  * may have both interfaces in a single hwgroup to "serialize" access.
1096  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1097  * together into one hwgroup for serialized access.
1098  *
1099  * Note also that several hwgroups can end up sharing a single IRQ,
1100  * possibly along with many other devices.  This is especially common in
1101  * PCI-based systems with off-board IDE controller cards.
1102  *
1103  * The IDE driver uses the single global ide_lock spinlock to protect
1104  * access to the request queues, and to protect the hwgroup->busy flag.
1105  *
1106  * The first thread into the driver for a particular hwgroup sets the
1107  * hwgroup->busy flag to indicate that this hwgroup is now active,
1108  * and then initiates processing of the top request from the request queue.
1109  *
1110  * Other threads attempting entry notice the busy setting, and will simply
1111  * queue their new requests and exit immediately.  Note that hwgroup->busy
1112  * remains set even when the driver is merely awaiting the next interrupt.
1113  * Thus, the meaning is "this hwgroup is busy processing a request".
1114  *
1115  * When processing of a request completes, the completing thread or IRQ-handler
1116  * will start the next request from the queue.  If no more work remains,
1117  * the driver will clear the hwgroup->busy flag and exit.
1118  *
1119  * The ide_lock (spinlock) is used to protect all access to the
1120  * hwgroup->busy flag, but is otherwise not needed for most processing in
1121  * the driver.  This makes the driver much more friendlier to shared IRQs
1122  * than previous designs, while remaining 100% (?) SMP safe and capable.
1123  */
1124 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1125 {
1126         ide_drive_t     *drive;
1127         ide_hwif_t      *hwif;
1128         struct request  *rq;
1129         ide_startstop_t startstop;
1130         int             loops = 0;
1131
1132         /* for atari only: POSSIBLY BROKEN HERE(?) */
1133         ide_get_lock(ide_intr, hwgroup);
1134
1135         /* caller must own ide_lock */
1136         BUG_ON(!irqs_disabled());
1137
1138         while (!hwgroup->busy) {
1139                 hwgroup->busy = 1;
1140                 drive = choose_drive(hwgroup);
1141                 if (drive == NULL) {
1142                         int sleeping = 0;
1143                         unsigned long sleep = 0; /* shut up, gcc */
1144                         hwgroup->rq = NULL;
1145                         drive = hwgroup->drive;
1146                         do {
1147                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1148                                         sleeping = 1;
1149                                         sleep = drive->sleep;
1150                                 }
1151                         } while ((drive = drive->next) != hwgroup->drive);
1152                         if (sleeping) {
1153                 /*
1154                  * Take a short snooze, and then wake up this hwgroup again.
1155                  * This gives other hwgroups on the same a chance to
1156                  * play fairly with us, just in case there are big differences
1157                  * in relative throughputs.. don't want to hog the cpu too much.
1158                  */
1159                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1160                                         sleep = jiffies + WAIT_MIN_SLEEP;
1161 #if 1
1162                                 if (timer_pending(&hwgroup->timer))
1163                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1164 #endif
1165                                 /* so that ide_timer_expiry knows what to do */
1166                                 hwgroup->sleeping = 1;
1167                                 mod_timer(&hwgroup->timer, sleep);
1168                                 /* we purposely leave hwgroup->busy==1
1169                                  * while sleeping */
1170                         } else {
1171                                 /* Ugly, but how can we sleep for the lock
1172                                  * otherwise? perhaps from tq_disk?
1173                                  */
1174
1175                                 /* for atari only */
1176                                 ide_release_lock();
1177                                 hwgroup->busy = 0;
1178                         }
1179
1180                         /* no more work for this hwgroup (for now) */
1181                         return;
1182                 }
1183         again:
1184                 hwif = HWIF(drive);
1185                 if (hwgroup->hwif->sharing_irq &&
1186                     hwif != hwgroup->hwif &&
1187                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1188                         /* set nIEN for previous hwif */
1189                         SELECT_INTERRUPT(drive);
1190                 }
1191                 hwgroup->hwif = hwif;
1192                 hwgroup->drive = drive;
1193                 drive->sleeping = 0;
1194                 drive->service_start = jiffies;
1195
1196                 if (blk_queue_plugged(drive->queue)) {
1197                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1198                         break;
1199                 }
1200
1201                 /*
1202                  * we know that the queue isn't empty, but this can happen
1203                  * if the q->prep_rq_fn() decides to kill a request
1204                  */
1205                 rq = elv_next_request(drive->queue);
1206                 if (!rq) {
1207                         hwgroup->busy = 0;
1208                         break;
1209                 }
1210
1211                 /*
1212                  * Sanity: don't accept a request that isn't a PM request
1213                  * if we are currently power managed. This is very important as
1214                  * blk_stop_queue() doesn't prevent the elv_next_request()
1215                  * above to return us whatever is in the queue. Since we call
1216                  * ide_do_request() ourselves, we end up taking requests while
1217                  * the queue is blocked...
1218                  *
1219                  * We let requests forced at head of queue with ide-preempt
1220                  * though. I hope that doesn't happen too much, hopefully not
1221                  * unless the subdriver triggers such a thing in its own PM
1222                  * state machine.
1223                  *
1224                  * We count how many times we loop here to make sure we service
1225                  * all drives in the hwgroup without looping for ever
1226                  */
1227                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1228                         drive = drive->next ? drive->next : hwgroup->drive;
1229                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1230                                 goto again;
1231                         /* We clear busy, there should be no pending ATA command at this point. */
1232                         hwgroup->busy = 0;
1233                         break;
1234                 }
1235
1236                 hwgroup->rq = rq;
1237
1238                 /*
1239                  * Some systems have trouble with IDE IRQs arriving while
1240                  * the driver is still setting things up.  So, here we disable
1241                  * the IRQ used by this interface while the request is being started.
1242                  * This may look bad at first, but pretty much the same thing
1243                  * happens anyway when any interrupt comes in, IDE or otherwise
1244                  *  -- the kernel masks the IRQ while it is being handled.
1245                  */
1246                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1247                         disable_irq_nosync(hwif->irq);
1248                 spin_unlock(&ide_lock);
1249                 local_irq_enable();
1250                         /* allow other IRQs while we start this request */
1251                 startstop = start_request(drive, rq);
1252                 spin_lock_irq(&ide_lock);
1253                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1254                         enable_irq(hwif->irq);
1255                 if (startstop == ide_stopped)
1256                         hwgroup->busy = 0;
1257         }
1258 }
1259
1260 /*
1261  * Passes the stuff to ide_do_request
1262  */
1263 void do_ide_request(request_queue_t *q)
1264 {
1265         ide_drive_t *drive = q->queuedata;
1266
1267         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1268 }
1269
1270 /*
1271  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1272  * retry the current request in pio mode instead of risking tossing it
1273  * all away
1274  */
1275 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1276 {
1277         ide_hwif_t *hwif = HWIF(drive);
1278         struct request *rq;
1279         ide_startstop_t ret = ide_stopped;
1280
1281         /*
1282          * end current dma transaction
1283          */
1284
1285         if (error < 0) {
1286                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1287                 (void)HWIF(drive)->ide_dma_end(drive);
1288                 ret = ide_error(drive, "dma timeout error",
1289                                                 hwif->INB(IDE_STATUS_REG));
1290         } else {
1291                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1292                 (void) hwif->ide_dma_timeout(drive);
1293         }
1294
1295         /*
1296          * disable dma for now, but remember that we did so because of
1297          * a timeout -- we'll reenable after we finish this next request
1298          * (or rather the first chunk of it) in pio.
1299          */
1300         drive->retry_pio++;
1301         drive->state = DMA_PIO_RETRY;
1302         (void) hwif->ide_dma_off_quietly(drive);
1303
1304         /*
1305          * un-busy drive etc (hwgroup->busy is cleared on return) and
1306          * make sure request is sane
1307          */
1308         rq = HWGROUP(drive)->rq;
1309         HWGROUP(drive)->rq = NULL;
1310
1311         rq->errors = 0;
1312
1313         if (!rq->bio)
1314                 goto out;
1315
1316         rq->sector = rq->bio->bi_sector;
1317         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1318         rq->hard_cur_sectors = rq->current_nr_sectors;
1319         rq->buffer = bio_data(rq->bio);
1320 out:
1321         return ret;
1322 }
1323
1324 /**
1325  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1326  *      @data: timer callback magic (hwgroup)
1327  *
1328  *      An IDE command has timed out before the expected drive return
1329  *      occurred. At this point we attempt to clean up the current
1330  *      mess. If the current handler includes an expiry handler then
1331  *      we invoke the expiry handler, and providing it is happy the
1332  *      work is done. If that fails we apply generic recovery rules
1333  *      invoking the handler and checking the drive DMA status. We
1334  *      have an excessively incestuous relationship with the DMA
1335  *      logic that wants cleaning up.
1336  */
1337
1338 void ide_timer_expiry (unsigned long data)
1339 {
1340         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1341         ide_handler_t   *handler;
1342         ide_expiry_t    *expiry;
1343         unsigned long   flags;
1344         unsigned long   wait = -1;
1345
1346         spin_lock_irqsave(&ide_lock, flags);
1347
1348         if ((handler = hwgroup->handler) == NULL) {
1349                 /*
1350                  * Either a marginal timeout occurred
1351                  * (got the interrupt just as timer expired),
1352                  * or we were "sleeping" to give other devices a chance.
1353                  * Either way, we don't really want to complain about anything.
1354                  */
1355                 if (hwgroup->sleeping) {
1356                         hwgroup->sleeping = 0;
1357                         hwgroup->busy = 0;
1358                 }
1359         } else {
1360                 ide_drive_t *drive = hwgroup->drive;
1361                 if (!drive) {
1362                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1363                         hwgroup->handler = NULL;
1364                 } else {
1365                         ide_hwif_t *hwif;
1366                         ide_startstop_t startstop = ide_stopped;
1367                         if (!hwgroup->busy) {
1368                                 hwgroup->busy = 1;      /* paranoia */
1369                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1370                         }
1371                         if ((expiry = hwgroup->expiry) != NULL) {
1372                                 /* continue */
1373                                 if ((wait = expiry(drive)) > 0) {
1374                                         /* reset timer */
1375                                         hwgroup->timer.expires  = jiffies + wait;
1376                                         add_timer(&hwgroup->timer);
1377                                         spin_unlock_irqrestore(&ide_lock, flags);
1378                                         return;
1379                                 }
1380                         }
1381                         hwgroup->handler = NULL;
1382                         /*
1383                          * We need to simulate a real interrupt when invoking
1384                          * the handler() function, which means we need to
1385                          * globally mask the specific IRQ:
1386                          */
1387                         spin_unlock(&ide_lock);
1388                         hwif  = HWIF(drive);
1389 #if DISABLE_IRQ_NOSYNC
1390                         disable_irq_nosync(hwif->irq);
1391 #else
1392                         /* disable_irq_nosync ?? */
1393                         disable_irq(hwif->irq);
1394 #endif /* DISABLE_IRQ_NOSYNC */
1395                         /* local CPU only,
1396                          * as if we were handling an interrupt */
1397                         local_irq_disable();
1398                         if (hwgroup->polling) {
1399                                 startstop = handler(drive);
1400                         } else if (drive_is_ready(drive)) {
1401                                 if (drive->waiting_for_dma)
1402                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1403                                 (void)ide_ack_intr(hwif);
1404                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1405                                 startstop = handler(drive);
1406                         } else {
1407                                 if (drive->waiting_for_dma) {
1408                                         startstop = ide_dma_timeout_retry(drive, wait);
1409                                 } else
1410                                         startstop =
1411                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1412                         }
1413                         drive->service_time = jiffies - drive->service_start;
1414                         spin_lock_irq(&ide_lock);
1415                         enable_irq(hwif->irq);
1416                         if (startstop == ide_stopped)
1417                                 hwgroup->busy = 0;
1418                 }
1419         }
1420         ide_do_request(hwgroup, IDE_NO_IRQ);
1421         spin_unlock_irqrestore(&ide_lock, flags);
1422 }
1423
1424 /**
1425  *      unexpected_intr         -       handle an unexpected IDE interrupt
1426  *      @irq: interrupt line
1427  *      @hwgroup: hwgroup being processed
1428  *
1429  *      There's nothing really useful we can do with an unexpected interrupt,
1430  *      other than reading the status register (to clear it), and logging it.
1431  *      There should be no way that an irq can happen before we're ready for it,
1432  *      so we needn't worry much about losing an "important" interrupt here.
1433  *
1434  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1435  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1436  *      looks "good", we just ignore the interrupt completely.
1437  *
1438  *      This routine assumes __cli() is in effect when called.
1439  *
1440  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1441  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1442  *      we could screw up by interfering with a new request being set up for
1443  *      irq15.
1444  *
1445  *      In reality, this is a non-issue.  The new command is not sent unless
1446  *      the drive is ready to accept one, in which case we know the drive is
1447  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1448  *      before completing the issuance of any new drive command, so we will not
1449  *      be accidentally invoked as a result of any valid command completion
1450  *      interrupt.
1451  *
1452  *      Note that we must walk the entire hwgroup here. We know which hwif
1453  *      is doing the current command, but we don't know which hwif burped
1454  *      mysteriously.
1455  */
1456
1457 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1458 {
1459         u8 stat;
1460         ide_hwif_t *hwif = hwgroup->hwif;
1461
1462         /*
1463          * handle the unexpected interrupt
1464          */
1465         do {
1466                 if (hwif->irq == irq) {
1467                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1468                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1469                                 /* Try to not flood the console with msgs */
1470                                 static unsigned long last_msgtime, count;
1471                                 ++count;
1472                                 if (time_after(jiffies, last_msgtime + HZ)) {
1473                                         last_msgtime = jiffies;
1474                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1475                                                 "status=0x%02x, count=%ld\n",
1476                                                 hwif->name,
1477                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1478                                 }
1479                         }
1480                 }
1481         } while ((hwif = hwif->next) != hwgroup->hwif);
1482 }
1483
1484 /**
1485  *      ide_intr        -       default IDE interrupt handler
1486  *      @irq: interrupt number
1487  *      @dev_id: hwif group
1488  *      @regs: unused weirdness from the kernel irq layer
1489  *
1490  *      This is the default IRQ handler for the IDE layer. You should
1491  *      not need to override it. If you do be aware it is subtle in
1492  *      places
1493  *
1494  *      hwgroup->hwif is the interface in the group currently performing
1495  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1496  *      the IRQ handler to call. As we issue a command the handlers
1497  *      step through multiple states, reassigning the handler to the
1498  *      next step in the process. Unlike a smart SCSI controller IDE
1499  *      expects the main processor to sequence the various transfer
1500  *      stages. We also manage a poll timer to catch up with most
1501  *      timeout situations. There are still a few where the handlers
1502  *      don't ever decide to give up.
1503  *
1504  *      The handler eventually returns ide_stopped to indicate the
1505  *      request completed. At this point we issue the next request
1506  *      on the hwgroup and the process begins again.
1507  */
1508
1509 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1510 {
1511         unsigned long flags;
1512         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1513         ide_hwif_t *hwif;
1514         ide_drive_t *drive;
1515         ide_handler_t *handler;
1516         ide_startstop_t startstop;
1517
1518         spin_lock_irqsave(&ide_lock, flags);
1519         hwif = hwgroup->hwif;
1520
1521         if (!ide_ack_intr(hwif)) {
1522                 spin_unlock_irqrestore(&ide_lock, flags);
1523                 return IRQ_NONE;
1524         }
1525
1526         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1527                 /*
1528                  * Not expecting an interrupt from this drive.
1529                  * That means this could be:
1530                  *      (1) an interrupt from another PCI device
1531                  *      sharing the same PCI INT# as us.
1532                  * or   (2) a drive just entered sleep or standby mode,
1533                  *      and is interrupting to let us know.
1534                  * or   (3) a spurious interrupt of unknown origin.
1535                  *
1536                  * For PCI, we cannot tell the difference,
1537                  * so in that case we just ignore it and hope it goes away.
1538                  *
1539                  * FIXME: unexpected_intr should be hwif-> then we can
1540                  * remove all the ifdef PCI crap
1541                  */
1542 #ifdef CONFIG_BLK_DEV_IDEPCI
1543                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1544 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1545                 {
1546                         /*
1547                          * Probably not a shared PCI interrupt,
1548                          * so we can safely try to do something about it:
1549                          */
1550                         unexpected_intr(irq, hwgroup);
1551 #ifdef CONFIG_BLK_DEV_IDEPCI
1552                 } else {
1553                         /*
1554                          * Whack the status register, just in case
1555                          * we have a leftover pending IRQ.
1556                          */
1557                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1558 #endif /* CONFIG_BLK_DEV_IDEPCI */
1559                 }
1560                 spin_unlock_irqrestore(&ide_lock, flags);
1561                 return IRQ_NONE;
1562         }
1563         drive = hwgroup->drive;
1564         if (!drive) {
1565                 /*
1566                  * This should NEVER happen, and there isn't much
1567                  * we could do about it here.
1568                  *
1569                  * [Note - this can occur if the drive is hot unplugged]
1570                  */
1571                 spin_unlock_irqrestore(&ide_lock, flags);
1572                 return IRQ_HANDLED;
1573         }
1574         if (!drive_is_ready(drive)) {
1575                 /*
1576                  * This happens regularly when we share a PCI IRQ with
1577                  * another device.  Unfortunately, it can also happen
1578                  * with some buggy drives that trigger the IRQ before
1579                  * their status register is up to date.  Hopefully we have
1580                  * enough advance overhead that the latter isn't a problem.
1581                  */
1582                 spin_unlock_irqrestore(&ide_lock, flags);
1583                 return IRQ_NONE;
1584         }
1585         if (!hwgroup->busy) {
1586                 hwgroup->busy = 1;      /* paranoia */
1587                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1588         }
1589         hwgroup->handler = NULL;
1590         del_timer(&hwgroup->timer);
1591         spin_unlock(&ide_lock);
1592
1593         if (drive->unmask)
1594                 local_irq_enable();
1595         /* service this interrupt, may set handler for next interrupt */
1596         startstop = handler(drive);
1597         spin_lock_irq(&ide_lock);
1598
1599         /*
1600          * Note that handler() may have set things up for another
1601          * interrupt to occur soon, but it cannot happen until
1602          * we exit from this routine, because it will be the
1603          * same irq as is currently being serviced here, and Linux
1604          * won't allow another of the same (on any CPU) until we return.
1605          */
1606         drive->service_time = jiffies - drive->service_start;
1607         if (startstop == ide_stopped) {
1608                 if (hwgroup->handler == NULL) { /* paranoia */
1609                         hwgroup->busy = 0;
1610                         ide_do_request(hwgroup, hwif->irq);
1611                 } else {
1612                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1613                                 "on exit\n", drive->name);
1614                 }
1615         }
1616         spin_unlock_irqrestore(&ide_lock, flags);
1617         return IRQ_HANDLED;
1618 }
1619
1620 /**
1621  *      ide_init_drive_cmd      -       initialize a drive command request
1622  *      @rq: request object
1623  *
1624  *      Initialize a request before we fill it in and send it down to
1625  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1626  *      now it doesn't do a lot, but if that changes abusers will have a
1627  *      nasty suprise.
1628  */
1629
1630 void ide_init_drive_cmd (struct request *rq)
1631 {
1632         memset(rq, 0, sizeof(*rq));
1633         rq->flags = REQ_DRIVE_CMD;
1634         rq->ref_count = 1;
1635 }
1636
1637 EXPORT_SYMBOL(ide_init_drive_cmd);
1638
1639 /**
1640  *      ide_do_drive_cmd        -       issue IDE special command
1641  *      @drive: device to issue command
1642  *      @rq: request to issue
1643  *      @action: action for processing
1644  *
1645  *      This function issues a special IDE device request
1646  *      onto the request queue.
1647  *
1648  *      If action is ide_wait, then the rq is queued at the end of the
1649  *      request queue, and the function sleeps until it has been processed.
1650  *      This is for use when invoked from an ioctl handler.
1651  *
1652  *      If action is ide_preempt, then the rq is queued at the head of
1653  *      the request queue, displacing the currently-being-processed
1654  *      request and this function returns immediately without waiting
1655  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1656  *      intended for careful use by the ATAPI tape/cdrom driver code.
1657  *
1658  *      If action is ide_end, then the rq is queued at the end of the
1659  *      request queue, and the function returns immediately without waiting
1660  *      for the new rq to be completed. This is again intended for careful
1661  *      use by the ATAPI tape/cdrom driver code.
1662  */
1663
1664 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1665 {
1666         unsigned long flags;
1667         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1668         DECLARE_COMPLETION(wait);
1669         int where = ELEVATOR_INSERT_BACK, err;
1670         int must_wait = (action == ide_wait || action == ide_head_wait);
1671
1672         rq->errors = 0;
1673         rq->rq_status = RQ_ACTIVE;
1674
1675         /*
1676          * we need to hold an extra reference to request for safe inspection
1677          * after completion
1678          */
1679         if (must_wait) {
1680                 rq->ref_count++;
1681                 rq->waiting = &wait;
1682                 rq->end_io = blk_end_sync_rq;
1683         }
1684
1685         spin_lock_irqsave(&ide_lock, flags);
1686         if (action == ide_preempt)
1687                 hwgroup->rq = NULL;
1688         if (action == ide_preempt || action == ide_head_wait) {
1689                 where = ELEVATOR_INSERT_FRONT;
1690                 rq->flags |= REQ_PREEMPT;
1691         }
1692         __elv_add_request(drive->queue, rq, where, 0);
1693         ide_do_request(hwgroup, IDE_NO_IRQ);
1694         spin_unlock_irqrestore(&ide_lock, flags);
1695
1696         err = 0;
1697         if (must_wait) {
1698                 wait_for_completion(&wait);
1699                 rq->waiting = NULL;
1700                 if (rq->errors)
1701                         err = -EIO;
1702
1703                 blk_put_request(rq);
1704         }
1705
1706         return err;
1707 }
1708
1709 EXPORT_SYMBOL(ide_do_drive_cmd);