ide-io.c revision ed67b92385a5afddc98d5ff0894b2854c4a54dac
1/* 2 * IDE I/O functions 3 * 4 * Basic PIO and command management functionality. 5 * 6 * This code was split off from ide.c. See ide.c for history and original 7 * copyrights. 8 * 9 * This program is free software; you can redistribute it and/or modify it 10 * under the terms of the GNU General Public License as published by the 11 * Free Software Foundation; either version 2, or (at your option) any 12 * later version. 13 * 14 * This program is distributed in the hope that it will be useful, but 15 * WITHOUT ANY WARRANTY; without even the implied warranty of 16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 17 * General Public License for more details. 18 * 19 * For the avoidance of doubt the "preferred form" of this code is one which 20 * is in an open non patent encumbered format. Where cryptographic key signing 21 * forms part of the process of creating an executable the information 22 * including keys needed to generate an equivalently functional executable 23 * are deemed to be part of the source code. 24 */ 25 26 27#include <linux/module.h> 28#include <linux/types.h> 29#include <linux/string.h> 30#include <linux/kernel.h> 31#include <linux/timer.h> 32#include <linux/mm.h> 33#include <linux/interrupt.h> 34#include <linux/major.h> 35#include <linux/errno.h> 36#include <linux/genhd.h> 37#include <linux/blkpg.h> 38#include <linux/slab.h> 39#include <linux/init.h> 40#include <linux/pci.h> 41#include <linux/delay.h> 42#include <linux/ide.h> 43#include <linux/completion.h> 44#include <linux/reboot.h> 45#include <linux/cdrom.h> 46#include <linux/seq_file.h> 47#include <linux/device.h> 48#include <linux/kmod.h> 49#include <linux/scatterlist.h> 50 51#include <asm/byteorder.h> 52#include <asm/irq.h> 53#include <asm/uaccess.h> 54#include <asm/io.h> 55#include <asm/bitops.h> 56 57static int __ide_end_request(ide_drive_t *drive, struct request *rq, 58 int uptodate, unsigned int nr_bytes) 59{ 60 int ret = 1; 61 62 /* 63 * if failfast is set on a request, override number of sectors and 64 * complete the whole request right now 65 */ 66 if (blk_noretry_request(rq) && end_io_error(uptodate)) 67 nr_bytes = rq->hard_nr_sectors << 9; 68 69 if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors) 70 rq->errors = -EIO; 71 72 /* 73 * decide whether to reenable DMA -- 3 is a random magic for now, 74 * if we DMA timeout more than 3 times, just stay in PIO 75 */ 76 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) { 77 drive->state = 0; 78 HWGROUP(drive)->hwif->ide_dma_on(drive); 79 } 80 81 if (!end_that_request_chunk(rq, uptodate, nr_bytes)) { 82 add_disk_randomness(rq->rq_disk); 83 if (!list_empty(&rq->queuelist)) 84 blkdev_dequeue_request(rq); 85 HWGROUP(drive)->rq = NULL; 86 end_that_request_last(rq, uptodate); 87 ret = 0; 88 } 89 90 return ret; 91} 92 93/** 94 * ide_end_request - complete an IDE I/O 95 * @drive: IDE device for the I/O 96 * @uptodate: 97 * @nr_sectors: number of sectors completed 98 * 99 * This is our end_request wrapper function. We complete the I/O 100 * update random number input and dequeue the request, which if 101 * it was tagged may be out of order. 102 */ 103 104int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors) 105{ 106 unsigned int nr_bytes = nr_sectors << 9; 107 struct request *rq; 108 unsigned long flags; 109 int ret = 1; 110 111 /* 112 * room for locking improvements here, the calls below don't 113 * need the queue lock held at all 114 */ 115 spin_lock_irqsave(&ide_lock, flags); 116 rq = HWGROUP(drive)->rq; 117 118 if (!nr_bytes) { 119 if (blk_pc_request(rq)) 120 nr_bytes = rq->data_len; 121 else 122 nr_bytes = rq->hard_cur_sectors << 9; 123 } 124 125 ret = __ide_end_request(drive, rq, uptodate, nr_bytes); 126 127 spin_unlock_irqrestore(&ide_lock, flags); 128 return ret; 129} 130EXPORT_SYMBOL(ide_end_request); 131 132/* 133 * Power Management state machine. This one is rather trivial for now, 134 * we should probably add more, like switching back to PIO on suspend 135 * to help some BIOSes, re-do the door locking on resume, etc... 136 */ 137 138enum { 139 ide_pm_flush_cache = ide_pm_state_start_suspend, 140 idedisk_pm_standby, 141 142 idedisk_pm_restore_pio = ide_pm_state_start_resume, 143 idedisk_pm_idle, 144 ide_pm_restore_dma, 145}; 146 147static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error) 148{ 149 struct request_pm_state *pm = rq->data; 150 151 if (drive->media != ide_disk) 152 return; 153 154 switch (pm->pm_step) { 155 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) complete */ 156 if (pm->pm_state == PM_EVENT_FREEZE) 157 pm->pm_step = ide_pm_state_completed; 158 else 159 pm->pm_step = idedisk_pm_standby; 160 break; 161 case idedisk_pm_standby: /* Suspend step 2 (standby) complete */ 162 pm->pm_step = ide_pm_state_completed; 163 break; 164 case idedisk_pm_restore_pio: /* Resume step 1 complete */ 165 pm->pm_step = idedisk_pm_idle; 166 break; 167 case idedisk_pm_idle: /* Resume step 2 (idle) complete */ 168 pm->pm_step = ide_pm_restore_dma; 169 break; 170 } 171} 172 173static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq) 174{ 175 struct request_pm_state *pm = rq->data; 176 ide_task_t *args = rq->special; 177 178 memset(args, 0, sizeof(*args)); 179 180 switch (pm->pm_step) { 181 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) */ 182 if (drive->media != ide_disk) 183 break; 184 /* Not supported? Switch to next step now. */ 185 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) { 186 ide_complete_power_step(drive, rq, 0, 0); 187 return ide_stopped; 188 } 189 if (ide_id_has_flush_cache_ext(drive->id)) 190 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT; 191 else 192 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE; 193 args->command_type = IDE_DRIVE_TASK_NO_DATA; 194 args->handler = &task_no_data_intr; 195 return do_rw_taskfile(drive, args); 196 197 case idedisk_pm_standby: /* Suspend step 2 (standby) */ 198 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1; 199 args->command_type = IDE_DRIVE_TASK_NO_DATA; 200 args->handler = &task_no_data_intr; 201 return do_rw_taskfile(drive, args); 202 203 case idedisk_pm_restore_pio: /* Resume step 1 (restore PIO) */ 204 ide_set_max_pio(drive); 205 /* 206 * skip idedisk_pm_idle for ATAPI devices 207 */ 208 if (drive->media != ide_disk) 209 pm->pm_step = ide_pm_restore_dma; 210 else 211 ide_complete_power_step(drive, rq, 0, 0); 212 return ide_stopped; 213 214 case idedisk_pm_idle: /* Resume step 2 (idle) */ 215 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE; 216 args->command_type = IDE_DRIVE_TASK_NO_DATA; 217 args->handler = task_no_data_intr; 218 return do_rw_taskfile(drive, args); 219 220 case ide_pm_restore_dma: /* Resume step 3 (restore DMA) */ 221 /* 222 * Right now, all we do is call ide_set_dma(drive), 223 * we could be smarter and check for current xfer_speed 224 * in struct drive etc... 225 */ 226 if (drive->hwif->ide_dma_on == NULL) 227 break; 228 drive->hwif->dma_off_quietly(drive); 229 /* 230 * TODO: respect ->using_dma setting 231 */ 232 ide_set_dma(drive); 233 break; 234 } 235 pm->pm_step = ide_pm_state_completed; 236 return ide_stopped; 237} 238 239/** 240 * ide_end_dequeued_request - complete an IDE I/O 241 * @drive: IDE device for the I/O 242 * @uptodate: 243 * @nr_sectors: number of sectors completed 244 * 245 * Complete an I/O that is no longer on the request queue. This 246 * typically occurs when we pull the request and issue a REQUEST_SENSE. 247 * We must still finish the old request but we must not tamper with the 248 * queue in the meantime. 249 * 250 * NOTE: This path does not handle barrier, but barrier is not supported 251 * on ide-cd anyway. 252 */ 253 254int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq, 255 int uptodate, int nr_sectors) 256{ 257 unsigned long flags; 258 int ret = 1; 259 260 spin_lock_irqsave(&ide_lock, flags); 261 262 BUG_ON(!blk_rq_started(rq)); 263 264 /* 265 * if failfast is set on a request, override number of sectors and 266 * complete the whole request right now 267 */ 268 if (blk_noretry_request(rq) && end_io_error(uptodate)) 269 nr_sectors = rq->hard_nr_sectors; 270 271 if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors) 272 rq->errors = -EIO; 273 274 /* 275 * decide whether to reenable DMA -- 3 is a random magic for now, 276 * if we DMA timeout more than 3 times, just stay in PIO 277 */ 278 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) { 279 drive->state = 0; 280 HWGROUP(drive)->hwif->ide_dma_on(drive); 281 } 282 283 if (!end_that_request_first(rq, uptodate, nr_sectors)) { 284 add_disk_randomness(rq->rq_disk); 285 if (blk_rq_tagged(rq)) 286 blk_queue_end_tag(drive->queue, rq); 287 end_that_request_last(rq, uptodate); 288 ret = 0; 289 } 290 spin_unlock_irqrestore(&ide_lock, flags); 291 return ret; 292} 293EXPORT_SYMBOL_GPL(ide_end_dequeued_request); 294 295 296/** 297 * ide_complete_pm_request - end the current Power Management request 298 * @drive: target drive 299 * @rq: request 300 * 301 * This function cleans up the current PM request and stops the queue 302 * if necessary. 303 */ 304static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq) 305{ 306 unsigned long flags; 307 308#ifdef DEBUG_PM 309 printk("%s: completing PM request, %s\n", drive->name, 310 blk_pm_suspend_request(rq) ? "suspend" : "resume"); 311#endif 312 spin_lock_irqsave(&ide_lock, flags); 313 if (blk_pm_suspend_request(rq)) { 314 blk_stop_queue(drive->queue); 315 } else { 316 drive->blocked = 0; 317 blk_start_queue(drive->queue); 318 } 319 blkdev_dequeue_request(rq); 320 HWGROUP(drive)->rq = NULL; 321 end_that_request_last(rq, 1); 322 spin_unlock_irqrestore(&ide_lock, flags); 323} 324 325/** 326 * ide_end_drive_cmd - end an explicit drive command 327 * @drive: command 328 * @stat: status bits 329 * @err: error bits 330 * 331 * Clean up after success/failure of an explicit drive command. 332 * These get thrown onto the queue so they are synchronized with 333 * real I/O operations on the drive. 334 * 335 * In LBA48 mode we have to read the register set twice to get 336 * all the extra information out. 337 */ 338 339void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err) 340{ 341 ide_hwif_t *hwif = HWIF(drive); 342 unsigned long flags; 343 struct request *rq; 344 345 spin_lock_irqsave(&ide_lock, flags); 346 rq = HWGROUP(drive)->rq; 347 spin_unlock_irqrestore(&ide_lock, flags); 348 349 if (rq->cmd_type == REQ_TYPE_ATA_CMD) { 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 } 359 } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) { 360 u8 *args = (u8 *) rq->buffer; 361 if (rq->errors == 0) 362 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT); 363 364 if (args) { 365 args[0] = stat; 366 args[1] = err; 367 args[2] = hwif->INB(IDE_NSECTOR_REG); 368 args[3] = hwif->INB(IDE_SECTOR_REG); 369 args[4] = hwif->INB(IDE_LCYL_REG); 370 args[5] = hwif->INB(IDE_HCYL_REG); 371 args[6] = hwif->INB(IDE_SELECT_REG); 372 } 373 } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) { 374 ide_task_t *args = (ide_task_t *) rq->special; 375 if (rq->errors == 0) 376 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT); 377 378 if (args) { 379 if (args->tf_in_flags.b.data) { 380 u16 data = hwif->INW(IDE_DATA_REG); 381 args->tfRegister[IDE_DATA_OFFSET] = (data) & 0xFF; 382 args->hobRegister[IDE_DATA_OFFSET] = (data >> 8) & 0xFF; 383 } 384 args->tfRegister[IDE_ERROR_OFFSET] = err; 385 /* be sure we're looking at the low order bits */ 386 hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG); 387 args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG); 388 args->tfRegister[IDE_SECTOR_OFFSET] = hwif->INB(IDE_SECTOR_REG); 389 args->tfRegister[IDE_LCYL_OFFSET] = hwif->INB(IDE_LCYL_REG); 390 args->tfRegister[IDE_HCYL_OFFSET] = hwif->INB(IDE_HCYL_REG); 391 args->tfRegister[IDE_SELECT_OFFSET] = hwif->INB(IDE_SELECT_REG); 392 args->tfRegister[IDE_STATUS_OFFSET] = stat; 393 394 if (drive->addressing == 1) { 395 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG); 396 args->hobRegister[IDE_FEATURE_OFFSET] = hwif->INB(IDE_FEATURE_REG); 397 args->hobRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG); 398 args->hobRegister[IDE_SECTOR_OFFSET] = hwif->INB(IDE_SECTOR_REG); 399 args->hobRegister[IDE_LCYL_OFFSET] = hwif->INB(IDE_LCYL_REG); 400 args->hobRegister[IDE_HCYL_OFFSET] = hwif->INB(IDE_HCYL_REG); 401 } 402 } 403 } else if (blk_pm_request(rq)) { 404 struct request_pm_state *pm = rq->data; 405#ifdef DEBUG_PM 406 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n", 407 drive->name, rq->pm->pm_step, stat, err); 408#endif 409 ide_complete_power_step(drive, rq, stat, err); 410 if (pm->pm_step == ide_pm_state_completed) 411 ide_complete_pm_request(drive, rq); 412 return; 413 } 414 415 spin_lock_irqsave(&ide_lock, flags); 416 blkdev_dequeue_request(rq); 417 HWGROUP(drive)->rq = NULL; 418 rq->errors = err; 419 end_that_request_last(rq, !rq->errors); 420 spin_unlock_irqrestore(&ide_lock, flags); 421} 422 423EXPORT_SYMBOL(ide_end_drive_cmd); 424 425/** 426 * try_to_flush_leftover_data - flush junk 427 * @drive: drive to flush 428 * 429 * try_to_flush_leftover_data() is invoked in response to a drive 430 * unexpectedly having its DRQ_STAT bit set. As an alternative to 431 * resetting the drive, this routine tries to clear the condition 432 * by read a sector's worth of data from the drive. Of course, 433 * this may not help if the drive is *waiting* for data from *us*. 434 */ 435static void try_to_flush_leftover_data (ide_drive_t *drive) 436{ 437 int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS; 438 439 if (drive->media != ide_disk) 440 return; 441 while (i > 0) { 442 u32 buffer[16]; 443 u32 wcount = (i > 16) ? 16 : i; 444 445 i -= wcount; 446 HWIF(drive)->ata_input_data(drive, buffer, wcount); 447 } 448} 449 450static void ide_kill_rq(ide_drive_t *drive, struct request *rq) 451{ 452 if (rq->rq_disk) { 453 ide_driver_t *drv; 454 455 drv = *(ide_driver_t **)rq->rq_disk->private_data; 456 drv->end_request(drive, 0, 0); 457 } else 458 ide_end_request(drive, 0, 0); 459} 460 461static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err) 462{ 463 ide_hwif_t *hwif = drive->hwif; 464 465 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) { 466 /* other bits are useless when BUSY */ 467 rq->errors |= ERROR_RESET; 468 } else if (stat & ERR_STAT) { 469 /* err has different meaning on cdrom and tape */ 470 if (err == ABRT_ERR) { 471 if (drive->select.b.lba && 472 /* some newer drives don't support WIN_SPECIFY */ 473 hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY) 474 return ide_stopped; 475 } else if ((err & BAD_CRC) == BAD_CRC) { 476 /* UDMA crc error, just retry the operation */ 477 drive->crc_count++; 478 } else if (err & (BBD_ERR | ECC_ERR)) { 479 /* retries won't help these */ 480 rq->errors = ERROR_MAX; 481 } else if (err & TRK0_ERR) { 482 /* help it find track zero */ 483 rq->errors |= ERROR_RECAL; 484 } 485 } 486 487 if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && 488 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) 489 try_to_flush_leftover_data(drive); 490 491 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) { 492 ide_kill_rq(drive, rq); 493 return ide_stopped; 494 } 495 496 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) 497 rq->errors |= ERROR_RESET; 498 499 if ((rq->errors & ERROR_RESET) == ERROR_RESET) { 500 ++rq->errors; 501 return ide_do_reset(drive); 502 } 503 504 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL) 505 drive->special.b.recalibrate = 1; 506 507 ++rq->errors; 508 509 return ide_stopped; 510} 511 512static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err) 513{ 514 ide_hwif_t *hwif = drive->hwif; 515 516 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) { 517 /* other bits are useless when BUSY */ 518 rq->errors |= ERROR_RESET; 519 } else { 520 /* add decoding error stuff */ 521 } 522 523 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) 524 /* force an abort */ 525 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG); 526 527 if (rq->errors >= ERROR_MAX) { 528 ide_kill_rq(drive, rq); 529 } else { 530 if ((rq->errors & ERROR_RESET) == ERROR_RESET) { 531 ++rq->errors; 532 return ide_do_reset(drive); 533 } 534 ++rq->errors; 535 } 536 537 return ide_stopped; 538} 539 540ide_startstop_t 541__ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err) 542{ 543 if (drive->media == ide_disk) 544 return ide_ata_error(drive, rq, stat, err); 545 return ide_atapi_error(drive, rq, stat, err); 546} 547 548EXPORT_SYMBOL_GPL(__ide_error); 549 550/** 551 * ide_error - handle an error on the IDE 552 * @drive: drive the error occurred on 553 * @msg: message to report 554 * @stat: status bits 555 * 556 * ide_error() takes action based on the error returned by the drive. 557 * For normal I/O that may well include retries. We deal with 558 * both new-style (taskfile) and old style command handling here. 559 * In the case of taskfile command handling there is work left to 560 * do 561 */ 562 563ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat) 564{ 565 struct request *rq; 566 u8 err; 567 568 err = ide_dump_status(drive, msg, stat); 569 570 if ((rq = HWGROUP(drive)->rq) == NULL) 571 return ide_stopped; 572 573 /* retry only "normal" I/O: */ 574 if (!blk_fs_request(rq)) { 575 rq->errors = 1; 576 ide_end_drive_cmd(drive, stat, err); 577 return ide_stopped; 578 } 579 580 if (rq->rq_disk) { 581 ide_driver_t *drv; 582 583 drv = *(ide_driver_t **)rq->rq_disk->private_data; 584 return drv->error(drive, rq, stat, err); 585 } else 586 return __ide_error(drive, rq, stat, err); 587} 588 589EXPORT_SYMBOL_GPL(ide_error); 590 591ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq) 592{ 593 if (drive->media != ide_disk) 594 rq->errors |= ERROR_RESET; 595 596 ide_kill_rq(drive, rq); 597 598 return ide_stopped; 599} 600 601EXPORT_SYMBOL_GPL(__ide_abort); 602 603/** 604 * ide_abort - abort pending IDE operations 605 * @drive: drive the error occurred on 606 * @msg: message to report 607 * 608 * ide_abort kills and cleans up when we are about to do a 609 * host initiated reset on active commands. Longer term we 610 * want handlers to have sensible abort handling themselves 611 * 612 * This differs fundamentally from ide_error because in 613 * this case the command is doing just fine when we 614 * blow it away. 615 */ 616 617ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg) 618{ 619 struct request *rq; 620 621 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL) 622 return ide_stopped; 623 624 /* retry only "normal" I/O: */ 625 if (!blk_fs_request(rq)) { 626 rq->errors = 1; 627 ide_end_drive_cmd(drive, BUSY_STAT, 0); 628 return ide_stopped; 629 } 630 631 if (rq->rq_disk) { 632 ide_driver_t *drv; 633 634 drv = *(ide_driver_t **)rq->rq_disk->private_data; 635 return drv->abort(drive, rq); 636 } else 637 return __ide_abort(drive, rq); 638} 639 640/** 641 * ide_cmd - issue a simple drive command 642 * @drive: drive the command is for 643 * @cmd: command byte 644 * @nsect: sector byte 645 * @handler: handler for the command completion 646 * 647 * Issue a simple drive command with interrupts. 648 * The drive must be selected beforehand. 649 */ 650 651static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, 652 ide_handler_t *handler) 653{ 654 ide_hwif_t *hwif = HWIF(drive); 655 if (IDE_CONTROL_REG) 656 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */ 657 SELECT_MASK(drive,0); 658 hwif->OUTB(nsect,IDE_NSECTOR_REG); 659 ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL); 660} 661 662/** 663 * drive_cmd_intr - drive command completion interrupt 664 * @drive: drive the completion interrupt occurred on 665 * 666 * drive_cmd_intr() is invoked on completion of a special DRIVE_CMD. 667 * We do any necessary data reading and then wait for the drive to 668 * go non busy. At that point we may read the error data and complete 669 * the request 670 */ 671 672static ide_startstop_t drive_cmd_intr (ide_drive_t *drive) 673{ 674 struct request *rq = HWGROUP(drive)->rq; 675 ide_hwif_t *hwif = HWIF(drive); 676 u8 *args = (u8 *) rq->buffer; 677 u8 stat = hwif->INB(IDE_STATUS_REG); 678 int retries = 10; 679 680 local_irq_enable_in_hardirq(); 681 if ((stat & DRQ_STAT) && args && args[3]) { 682 u8 io_32bit = drive->io_32bit; 683 drive->io_32bit = 0; 684 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS); 685 drive->io_32bit = io_32bit; 686 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--) 687 udelay(100); 688 } 689 690 if (!OK_STAT(stat, READY_STAT, BAD_STAT)) 691 return ide_error(drive, "drive_cmd", stat); 692 /* calls ide_end_drive_cmd */ 693 ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG)); 694 return ide_stopped; 695} 696 697static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task) 698{ 699 task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect; 700 task->tfRegister[IDE_SECTOR_OFFSET] = drive->sect; 701 task->tfRegister[IDE_LCYL_OFFSET] = drive->cyl; 702 task->tfRegister[IDE_HCYL_OFFSET] = drive->cyl>>8; 703 task->tfRegister[IDE_SELECT_OFFSET] = ((drive->head-1)|drive->select.all)&0xBF; 704 task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY; 705 706 task->handler = &set_geometry_intr; 707} 708 709static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task) 710{ 711 task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect; 712 task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE; 713 714 task->handler = &recal_intr; 715} 716 717static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task) 718{ 719 task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req; 720 task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT; 721 722 task->handler = &set_multmode_intr; 723} 724 725static ide_startstop_t ide_disk_special(ide_drive_t *drive) 726{ 727 special_t *s = &drive->special; 728 ide_task_t args; 729 730 memset(&args, 0, sizeof(ide_task_t)); 731 args.command_type = IDE_DRIVE_TASK_NO_DATA; 732 733 if (s->b.set_geometry) { 734 s->b.set_geometry = 0; 735 ide_init_specify_cmd(drive, &args); 736 } else if (s->b.recalibrate) { 737 s->b.recalibrate = 0; 738 ide_init_restore_cmd(drive, &args); 739 } else if (s->b.set_multmode) { 740 s->b.set_multmode = 0; 741 if (drive->mult_req > drive->id->max_multsect) 742 drive->mult_req = drive->id->max_multsect; 743 ide_init_setmult_cmd(drive, &args); 744 } else if (s->all) { 745 int special = s->all; 746 s->all = 0; 747 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special); 748 return ide_stopped; 749 } 750 751 do_rw_taskfile(drive, &args); 752 753 return ide_started; 754} 755 756/* 757 * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away 758 */ 759static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio) 760{ 761 switch (req_pio) { 762 case 202: 763 case 201: 764 case 200: 765 case 102: 766 case 101: 767 case 100: 768 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0; 769 case 9: 770 case 8: 771 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0; 772 case 7: 773 case 6: 774 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0; 775 default: 776 return 0; 777 } 778} 779 780/** 781 * do_special - issue some special commands 782 * @drive: drive the command is for 783 * 784 * do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT 785 * commands to a drive. It used to do much more, but has been scaled 786 * back. 787 */ 788 789static ide_startstop_t do_special (ide_drive_t *drive) 790{ 791 special_t *s = &drive->special; 792 793#ifdef DEBUG 794 printk("%s: do_special: 0x%02x\n", drive->name, s->all); 795#endif 796 if (s->b.set_tune) { 797 ide_hwif_t *hwif = drive->hwif; 798 u8 req_pio = drive->tune_req; 799 800 s->b.set_tune = 0; 801 802 if (set_pio_mode_abuse(drive->hwif, req_pio)) { 803 if (hwif->set_pio_mode) 804 hwif->set_pio_mode(drive, req_pio); 805 } else { 806 int keep_dma = drive->using_dma; 807 808 ide_set_pio(drive, req_pio); 809 810 if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) { 811 if (keep_dma) 812 hwif->ide_dma_on(drive); 813 } 814 } 815 816 return ide_stopped; 817 } else { 818 if (drive->media == ide_disk) 819 return ide_disk_special(drive); 820 821 s->all = 0; 822 drive->mult_req = 0; 823 return ide_stopped; 824 } 825} 826 827void ide_map_sg(ide_drive_t *drive, struct request *rq) 828{ 829 ide_hwif_t *hwif = drive->hwif; 830 struct scatterlist *sg = hwif->sg_table; 831 832 if (hwif->sg_mapped) /* needed by ide-scsi */ 833 return; 834 835 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) { 836 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg); 837 } else { 838 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE); 839 hwif->sg_nents = 1; 840 } 841} 842 843EXPORT_SYMBOL_GPL(ide_map_sg); 844 845void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq) 846{ 847 ide_hwif_t *hwif = drive->hwif; 848 849 hwif->nsect = hwif->nleft = rq->nr_sectors; 850 hwif->cursg_ofs = 0; 851 hwif->cursg = NULL; 852} 853 854EXPORT_SYMBOL_GPL(ide_init_sg_cmd); 855 856/** 857 * execute_drive_command - issue special drive command 858 * @drive: the drive to issue the command on 859 * @rq: the request structure holding the command 860 * 861 * execute_drive_cmd() issues a special drive command, usually 862 * initiated by ioctl() from the external hdparm program. The 863 * command can be a drive command, drive task or taskfile 864 * operation. Weirdly you can call it with NULL to wait for 865 * all commands to finish. Don't do this as that is due to change 866 */ 867 868static ide_startstop_t execute_drive_cmd (ide_drive_t *drive, 869 struct request *rq) 870{ 871 ide_hwif_t *hwif = HWIF(drive); 872 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) { 873 ide_task_t *args = rq->special; 874 875 if (!args) 876 goto done; 877 878 hwif->data_phase = args->data_phase; 879 880 switch (hwif->data_phase) { 881 case TASKFILE_MULTI_OUT: 882 case TASKFILE_OUT: 883 case TASKFILE_MULTI_IN: 884 case TASKFILE_IN: 885 ide_init_sg_cmd(drive, rq); 886 ide_map_sg(drive, rq); 887 default: 888 break; 889 } 890 891 if (args->tf_out_flags.all != 0) 892 return flagged_taskfile(drive, args); 893 return do_rw_taskfile(drive, args); 894 } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) { 895 u8 *args = rq->buffer; 896 u8 sel; 897 898 if (!args) 899 goto done; 900#ifdef DEBUG 901 printk("%s: DRIVE_TASK_CMD ", drive->name); 902 printk("cmd=0x%02x ", args[0]); 903 printk("fr=0x%02x ", args[1]); 904 printk("ns=0x%02x ", args[2]); 905 printk("sc=0x%02x ", args[3]); 906 printk("lcyl=0x%02x ", args[4]); 907 printk("hcyl=0x%02x ", args[5]); 908 printk("sel=0x%02x\n", args[6]); 909#endif 910 hwif->OUTB(args[1], IDE_FEATURE_REG); 911 hwif->OUTB(args[3], IDE_SECTOR_REG); 912 hwif->OUTB(args[4], IDE_LCYL_REG); 913 hwif->OUTB(args[5], IDE_HCYL_REG); 914 sel = (args[6] & ~0x10); 915 if (drive->select.b.unit) 916 sel |= 0x10; 917 hwif->OUTB(sel, IDE_SELECT_REG); 918 ide_cmd(drive, args[0], args[2], &drive_cmd_intr); 919 return ide_started; 920 } else if (rq->cmd_type == REQ_TYPE_ATA_CMD) { 921 u8 *args = rq->buffer; 922 923 if (!args) 924 goto done; 925#ifdef DEBUG 926 printk("%s: DRIVE_CMD ", drive->name); 927 printk("cmd=0x%02x ", args[0]); 928 printk("sc=0x%02x ", args[1]); 929 printk("fr=0x%02x ", args[2]); 930 printk("xx=0x%02x\n", args[3]); 931#endif 932 if (args[0] == WIN_SMART) { 933 hwif->OUTB(0x4f, IDE_LCYL_REG); 934 hwif->OUTB(0xc2, IDE_HCYL_REG); 935 hwif->OUTB(args[2],IDE_FEATURE_REG); 936 hwif->OUTB(args[1],IDE_SECTOR_REG); 937 ide_cmd(drive, args[0], args[3], &drive_cmd_intr); 938 return ide_started; 939 } 940 hwif->OUTB(args[2],IDE_FEATURE_REG); 941 ide_cmd(drive, args[0], args[1], &drive_cmd_intr); 942 return ide_started; 943 } 944 945done: 946 /* 947 * NULL is actually a valid way of waiting for 948 * all current requests to be flushed from the queue. 949 */ 950#ifdef DEBUG 951 printk("%s: DRIVE_CMD (null)\n", drive->name); 952#endif 953 ide_end_drive_cmd(drive, 954 hwif->INB(IDE_STATUS_REG), 955 hwif->INB(IDE_ERROR_REG)); 956 return ide_stopped; 957} 958 959static void ide_check_pm_state(ide_drive_t *drive, struct request *rq) 960{ 961 struct request_pm_state *pm = rq->data; 962 963 if (blk_pm_suspend_request(rq) && 964 pm->pm_step == ide_pm_state_start_suspend) 965 /* Mark drive blocked when starting the suspend sequence. */ 966 drive->blocked = 1; 967 else if (blk_pm_resume_request(rq) && 968 pm->pm_step == ide_pm_state_start_resume) { 969 /* 970 * The first thing we do on wakeup is to wait for BSY bit to 971 * go away (with a looong timeout) as a drive on this hwif may 972 * just be POSTing itself. 973 * We do that before even selecting as the "other" device on 974 * the bus may be broken enough to walk on our toes at this 975 * point. 976 */ 977 int rc; 978#ifdef DEBUG_PM 979 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name); 980#endif 981 rc = ide_wait_not_busy(HWIF(drive), 35000); 982 if (rc) 983 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name); 984 SELECT_DRIVE(drive); 985 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]); 986 rc = ide_wait_not_busy(HWIF(drive), 100000); 987 if (rc) 988 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name); 989 } 990} 991 992/** 993 * start_request - start of I/O and command issuing for IDE 994 * 995 * start_request() initiates handling of a new I/O request. It 996 * accepts commands and I/O (read/write) requests. It also does 997 * the final remapping for weird stuff like EZDrive. Once 998 * device mapper can work sector level the EZDrive stuff can go away 999 * 1000 * FIXME: this function needs a rename 1001 */ 1002 1003static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq) 1004{ 1005 ide_startstop_t startstop; 1006 sector_t block; 1007 1008 BUG_ON(!blk_rq_started(rq)); 1009 1010#ifdef DEBUG 1011 printk("%s: start_request: current=0x%08lx\n", 1012 HWIF(drive)->name, (unsigned long) rq); 1013#endif 1014 1015 /* bail early if we've exceeded max_failures */ 1016 if (drive->max_failures && (drive->failures > drive->max_failures)) { 1017 goto kill_rq; 1018 } 1019 1020 block = rq->sector; 1021 if (blk_fs_request(rq) && 1022 (drive->media == ide_disk || drive->media == ide_floppy)) { 1023 block += drive->sect0; 1024 } 1025 /* Yecch - this will shift the entire interval, 1026 possibly killing some innocent following sector */ 1027 if (block == 0 && drive->remap_0_to_1 == 1) 1028 block = 1; /* redirect MBR access to EZ-Drive partn table */ 1029 1030 if (blk_pm_request(rq)) 1031 ide_check_pm_state(drive, rq); 1032 1033 SELECT_DRIVE(drive); 1034 if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) { 1035 printk(KERN_ERR "%s: drive not ready for command\n", drive->name); 1036 return startstop; 1037 } 1038 if (!drive->special.all) { 1039 ide_driver_t *drv; 1040 1041 /* 1042 * We reset the drive so we need to issue a SETFEATURES. 1043 * Do it _after_ do_special() restored device parameters. 1044 */ 1045 if (drive->current_speed == 0xff) 1046 ide_config_drive_speed(drive, drive->desired_speed); 1047 1048 if (rq->cmd_type == REQ_TYPE_ATA_CMD || 1049 rq->cmd_type == REQ_TYPE_ATA_TASK || 1050 rq->cmd_type == REQ_TYPE_ATA_TASKFILE) 1051 return execute_drive_cmd(drive, rq); 1052 else if (blk_pm_request(rq)) { 1053 struct request_pm_state *pm = rq->data; 1054#ifdef DEBUG_PM 1055 printk("%s: start_power_step(step: %d)\n", 1056 drive->name, rq->pm->pm_step); 1057#endif 1058 startstop = ide_start_power_step(drive, rq); 1059 if (startstop == ide_stopped && 1060 pm->pm_step == ide_pm_state_completed) 1061 ide_complete_pm_request(drive, rq); 1062 return startstop; 1063 } 1064 1065 drv = *(ide_driver_t **)rq->rq_disk->private_data; 1066 return drv->do_request(drive, rq, block); 1067 } 1068 return do_special(drive); 1069kill_rq: 1070 ide_kill_rq(drive, rq); 1071 return ide_stopped; 1072} 1073 1074/** 1075 * ide_stall_queue - pause an IDE device 1076 * @drive: drive to stall 1077 * @timeout: time to stall for (jiffies) 1078 * 1079 * ide_stall_queue() can be used by a drive to give excess bandwidth back 1080 * to the hwgroup by sleeping for timeout jiffies. 1081 */ 1082 1083void ide_stall_queue (ide_drive_t *drive, unsigned long timeout) 1084{ 1085 if (timeout > WAIT_WORSTCASE) 1086 timeout = WAIT_WORSTCASE; 1087 drive->sleep = timeout + jiffies; 1088 drive->sleeping = 1; 1089} 1090 1091EXPORT_SYMBOL(ide_stall_queue); 1092 1093#define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time) 1094 1095/** 1096 * choose_drive - select a drive to service 1097 * @hwgroup: hardware group to select on 1098 * 1099 * choose_drive() selects the next drive which will be serviced. 1100 * This is necessary because the IDE layer can't issue commands 1101 * to both drives on the same cable, unlike SCSI. 1102 */ 1103 1104static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup) 1105{ 1106 ide_drive_t *drive, *best; 1107 1108repeat: 1109 best = NULL; 1110 drive = hwgroup->drive; 1111 1112 /* 1113 * drive is doing pre-flush, ordered write, post-flush sequence. even 1114 * though that is 3 requests, it must be seen as a single transaction. 1115 * we must not preempt this drive until that is complete 1116 */ 1117 if (blk_queue_flushing(drive->queue)) { 1118 /* 1119 * small race where queue could get replugged during 1120 * the 3-request flush cycle, just yank the plug since 1121 * we want it to finish asap 1122 */ 1123 blk_remove_plug(drive->queue); 1124 return drive; 1125 } 1126 1127 do { 1128 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep)) 1129 && !elv_queue_empty(drive->queue)) { 1130 if (!best 1131 || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep))) 1132 || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best)))) 1133 { 1134 if (!blk_queue_plugged(drive->queue)) 1135 best = drive; 1136 } 1137 } 1138 } while ((drive = drive->next) != hwgroup->drive); 1139 if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) { 1140 long t = (signed long)(WAKEUP(best) - jiffies); 1141 if (t >= WAIT_MIN_SLEEP) { 1142 /* 1143 * We *may* have some time to spare, but first let's see if 1144 * someone can potentially benefit from our nice mood today.. 1145 */ 1146 drive = best->next; 1147 do { 1148 if (!drive->sleeping 1149 && time_before(jiffies - best->service_time, WAKEUP(drive)) 1150 && time_before(WAKEUP(drive), jiffies + t)) 1151 { 1152 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP)); 1153 goto repeat; 1154 } 1155 } while ((drive = drive->next) != best); 1156 } 1157 } 1158 return best; 1159} 1160 1161/* 1162 * Issue a new request to a drive from hwgroup 1163 * Caller must have already done spin_lock_irqsave(&ide_lock, ..); 1164 * 1165 * A hwgroup is a serialized group of IDE interfaces. Usually there is 1166 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640) 1167 * may have both interfaces in a single hwgroup to "serialize" access. 1168 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped 1169 * together into one hwgroup for serialized access. 1170 * 1171 * Note also that several hwgroups can end up sharing a single IRQ, 1172 * possibly along with many other devices. This is especially common in 1173 * PCI-based systems with off-board IDE controller cards. 1174 * 1175 * The IDE driver uses the single global ide_lock spinlock to protect 1176 * access to the request queues, and to protect the hwgroup->busy flag. 1177 * 1178 * The first thread into the driver for a particular hwgroup sets the 1179 * hwgroup->busy flag to indicate that this hwgroup is now active, 1180 * and then initiates processing of the top request from the request queue. 1181 * 1182 * Other threads attempting entry notice the busy setting, and will simply 1183 * queue their new requests and exit immediately. Note that hwgroup->busy 1184 * remains set even when the driver is merely awaiting the next interrupt. 1185 * Thus, the meaning is "this hwgroup is busy processing a request". 1186 * 1187 * When processing of a request completes, the completing thread or IRQ-handler 1188 * will start the next request from the queue. If no more work remains, 1189 * the driver will clear the hwgroup->busy flag and exit. 1190 * 1191 * The ide_lock (spinlock) is used to protect all access to the 1192 * hwgroup->busy flag, but is otherwise not needed for most processing in 1193 * the driver. This makes the driver much more friendlier to shared IRQs 1194 * than previous designs, while remaining 100% (?) SMP safe and capable. 1195 */ 1196static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq) 1197{ 1198 ide_drive_t *drive; 1199 ide_hwif_t *hwif; 1200 struct request *rq; 1201 ide_startstop_t startstop; 1202 int loops = 0; 1203 1204 /* for atari only: POSSIBLY BROKEN HERE(?) */ 1205 ide_get_lock(ide_intr, hwgroup); 1206 1207 /* caller must own ide_lock */ 1208 BUG_ON(!irqs_disabled()); 1209 1210 while (!hwgroup->busy) { 1211 hwgroup->busy = 1; 1212 drive = choose_drive(hwgroup); 1213 if (drive == NULL) { 1214 int sleeping = 0; 1215 unsigned long sleep = 0; /* shut up, gcc */ 1216 hwgroup->rq = NULL; 1217 drive = hwgroup->drive; 1218 do { 1219 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) { 1220 sleeping = 1; 1221 sleep = drive->sleep; 1222 } 1223 } while ((drive = drive->next) != hwgroup->drive); 1224 if (sleeping) { 1225 /* 1226 * Take a short snooze, and then wake up this hwgroup again. 1227 * This gives other hwgroups on the same a chance to 1228 * play fairly with us, just in case there are big differences 1229 * in relative throughputs.. don't want to hog the cpu too much. 1230 */ 1231 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP)) 1232 sleep = jiffies + WAIT_MIN_SLEEP; 1233#if 1 1234 if (timer_pending(&hwgroup->timer)) 1235 printk(KERN_CRIT "ide_set_handler: timer already active\n"); 1236#endif 1237 /* so that ide_timer_expiry knows what to do */ 1238 hwgroup->sleeping = 1; 1239 hwgroup->req_gen_timer = hwgroup->req_gen; 1240 mod_timer(&hwgroup->timer, sleep); 1241 /* we purposely leave hwgroup->busy==1 1242 * while sleeping */ 1243 } else { 1244 /* Ugly, but how can we sleep for the lock 1245 * otherwise? perhaps from tq_disk? 1246 */ 1247 1248 /* for atari only */ 1249 ide_release_lock(); 1250 hwgroup->busy = 0; 1251 } 1252 1253 /* no more work for this hwgroup (for now) */ 1254 return; 1255 } 1256 again: 1257 hwif = HWIF(drive); 1258 if (hwgroup->hwif->sharing_irq && 1259 hwif != hwgroup->hwif && 1260 hwif->io_ports[IDE_CONTROL_OFFSET]) { 1261 /* set nIEN for previous hwif */ 1262 SELECT_INTERRUPT(drive); 1263 } 1264 hwgroup->hwif = hwif; 1265 hwgroup->drive = drive; 1266 drive->sleeping = 0; 1267 drive->service_start = jiffies; 1268 1269 if (blk_queue_plugged(drive->queue)) { 1270 printk(KERN_ERR "ide: huh? queue was plugged!\n"); 1271 break; 1272 } 1273 1274 /* 1275 * we know that the queue isn't empty, but this can happen 1276 * if the q->prep_rq_fn() decides to kill a request 1277 */ 1278 rq = elv_next_request(drive->queue); 1279 if (!rq) { 1280 hwgroup->busy = 0; 1281 break; 1282 } 1283 1284 /* 1285 * Sanity: don't accept a request that isn't a PM request 1286 * if we are currently power managed. This is very important as 1287 * blk_stop_queue() doesn't prevent the elv_next_request() 1288 * above to return us whatever is in the queue. Since we call 1289 * ide_do_request() ourselves, we end up taking requests while 1290 * the queue is blocked... 1291 * 1292 * We let requests forced at head of queue with ide-preempt 1293 * though. I hope that doesn't happen too much, hopefully not 1294 * unless the subdriver triggers such a thing in its own PM 1295 * state machine. 1296 * 1297 * We count how many times we loop here to make sure we service 1298 * all drives in the hwgroup without looping for ever 1299 */ 1300 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) { 1301 drive = drive->next ? drive->next : hwgroup->drive; 1302 if (loops++ < 4 && !blk_queue_plugged(drive->queue)) 1303 goto again; 1304 /* We clear busy, there should be no pending ATA command at this point. */ 1305 hwgroup->busy = 0; 1306 break; 1307 } 1308 1309 hwgroup->rq = rq; 1310 1311 /* 1312 * Some systems have trouble with IDE IRQs arriving while 1313 * the driver is still setting things up. So, here we disable 1314 * the IRQ used by this interface while the request is being started. 1315 * This may look bad at first, but pretty much the same thing 1316 * happens anyway when any interrupt comes in, IDE or otherwise 1317 * -- the kernel masks the IRQ while it is being handled. 1318 */ 1319 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq) 1320 disable_irq_nosync(hwif->irq); 1321 spin_unlock(&ide_lock); 1322 local_irq_enable_in_hardirq(); 1323 /* allow other IRQs while we start this request */ 1324 startstop = start_request(drive, rq); 1325 spin_lock_irq(&ide_lock); 1326 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq) 1327 enable_irq(hwif->irq); 1328 if (startstop == ide_stopped) 1329 hwgroup->busy = 0; 1330 } 1331} 1332 1333/* 1334 * Passes the stuff to ide_do_request 1335 */ 1336void do_ide_request(struct request_queue *q) 1337{ 1338 ide_drive_t *drive = q->queuedata; 1339 1340 ide_do_request(HWGROUP(drive), IDE_NO_IRQ); 1341} 1342 1343/* 1344 * un-busy the hwgroup etc, and clear any pending DMA status. we want to 1345 * retry the current request in pio mode instead of risking tossing it 1346 * all away 1347 */ 1348static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error) 1349{ 1350 ide_hwif_t *hwif = HWIF(drive); 1351 struct request *rq; 1352 ide_startstop_t ret = ide_stopped; 1353 1354 /* 1355 * end current dma transaction 1356 */ 1357 1358 if (error < 0) { 1359 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name); 1360 (void)HWIF(drive)->ide_dma_end(drive); 1361 ret = ide_error(drive, "dma timeout error", 1362 hwif->INB(IDE_STATUS_REG)); 1363 } else { 1364 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name); 1365 hwif->dma_timeout(drive); 1366 } 1367 1368 /* 1369 * disable dma for now, but remember that we did so because of 1370 * a timeout -- we'll reenable after we finish this next request 1371 * (or rather the first chunk of it) in pio. 1372 */ 1373 drive->retry_pio++; 1374 drive->state = DMA_PIO_RETRY; 1375 hwif->dma_off_quietly(drive); 1376 1377 /* 1378 * un-busy drive etc (hwgroup->busy is cleared on return) and 1379 * make sure request is sane 1380 */ 1381 rq = HWGROUP(drive)->rq; 1382 1383 if (!rq) 1384 goto out; 1385 1386 HWGROUP(drive)->rq = NULL; 1387 1388 rq->errors = 0; 1389 1390 if (!rq->bio) 1391 goto out; 1392 1393 rq->sector = rq->bio->bi_sector; 1394 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9; 1395 rq->hard_cur_sectors = rq->current_nr_sectors; 1396 rq->buffer = bio_data(rq->bio); 1397out: 1398 return ret; 1399} 1400 1401/** 1402 * ide_timer_expiry - handle lack of an IDE interrupt 1403 * @data: timer callback magic (hwgroup) 1404 * 1405 * An IDE command has timed out before the expected drive return 1406 * occurred. At this point we attempt to clean up the current 1407 * mess. If the current handler includes an expiry handler then 1408 * we invoke the expiry handler, and providing it is happy the 1409 * work is done. If that fails we apply generic recovery rules 1410 * invoking the handler and checking the drive DMA status. We 1411 * have an excessively incestuous relationship with the DMA 1412 * logic that wants cleaning up. 1413 */ 1414 1415void ide_timer_expiry (unsigned long data) 1416{ 1417 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data; 1418 ide_handler_t *handler; 1419 ide_expiry_t *expiry; 1420 unsigned long flags; 1421 unsigned long wait = -1; 1422 1423 spin_lock_irqsave(&ide_lock, flags); 1424 1425 if (((handler = hwgroup->handler) == NULL) || 1426 (hwgroup->req_gen != hwgroup->req_gen_timer)) { 1427 /* 1428 * Either a marginal timeout occurred 1429 * (got the interrupt just as timer expired), 1430 * or we were "sleeping" to give other devices a chance. 1431 * Either way, we don't really want to complain about anything. 1432 */ 1433 if (hwgroup->sleeping) { 1434 hwgroup->sleeping = 0; 1435 hwgroup->busy = 0; 1436 } 1437 } else { 1438 ide_drive_t *drive = hwgroup->drive; 1439 if (!drive) { 1440 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n"); 1441 hwgroup->handler = NULL; 1442 } else { 1443 ide_hwif_t *hwif; 1444 ide_startstop_t startstop = ide_stopped; 1445 if (!hwgroup->busy) { 1446 hwgroup->busy = 1; /* paranoia */ 1447 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name); 1448 } 1449 if ((expiry = hwgroup->expiry) != NULL) { 1450 /* continue */ 1451 if ((wait = expiry(drive)) > 0) { 1452 /* reset timer */ 1453 hwgroup->timer.expires = jiffies + wait; 1454 hwgroup->req_gen_timer = hwgroup->req_gen; 1455 add_timer(&hwgroup->timer); 1456 spin_unlock_irqrestore(&ide_lock, flags); 1457 return; 1458 } 1459 } 1460 hwgroup->handler = NULL; 1461 /* 1462 * We need to simulate a real interrupt when invoking 1463 * the handler() function, which means we need to 1464 * globally mask the specific IRQ: 1465 */ 1466 spin_unlock(&ide_lock); 1467 hwif = HWIF(drive); 1468#if DISABLE_IRQ_NOSYNC 1469 disable_irq_nosync(hwif->irq); 1470#else 1471 /* disable_irq_nosync ?? */ 1472 disable_irq(hwif->irq); 1473#endif /* DISABLE_IRQ_NOSYNC */ 1474 /* local CPU only, 1475 * as if we were handling an interrupt */ 1476 local_irq_disable(); 1477 if (hwgroup->polling) { 1478 startstop = handler(drive); 1479 } else if (drive_is_ready(drive)) { 1480 if (drive->waiting_for_dma) 1481 hwgroup->hwif->dma_lost_irq(drive); 1482 (void)ide_ack_intr(hwif); 1483 printk(KERN_WARNING "%s: lost interrupt\n", drive->name); 1484 startstop = handler(drive); 1485 } else { 1486 if (drive->waiting_for_dma) { 1487 startstop = ide_dma_timeout_retry(drive, wait); 1488 } else 1489 startstop = 1490 ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG)); 1491 } 1492 drive->service_time = jiffies - drive->service_start; 1493 spin_lock_irq(&ide_lock); 1494 enable_irq(hwif->irq); 1495 if (startstop == ide_stopped) 1496 hwgroup->busy = 0; 1497 } 1498 } 1499 ide_do_request(hwgroup, IDE_NO_IRQ); 1500 spin_unlock_irqrestore(&ide_lock, flags); 1501} 1502 1503/** 1504 * unexpected_intr - handle an unexpected IDE interrupt 1505 * @irq: interrupt line 1506 * @hwgroup: hwgroup being processed 1507 * 1508 * There's nothing really useful we can do with an unexpected interrupt, 1509 * other than reading the status register (to clear it), and logging it. 1510 * There should be no way that an irq can happen before we're ready for it, 1511 * so we needn't worry much about losing an "important" interrupt here. 1512 * 1513 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever 1514 * the drive enters "idle", "standby", or "sleep" mode, so if the status 1515 * looks "good", we just ignore the interrupt completely. 1516 * 1517 * This routine assumes __cli() is in effect when called. 1518 * 1519 * If an unexpected interrupt happens on irq15 while we are handling irq14 1520 * and if the two interfaces are "serialized" (CMD640), then it looks like 1521 * we could screw up by interfering with a new request being set up for 1522 * irq15. 1523 * 1524 * In reality, this is a non-issue. The new command is not sent unless 1525 * the drive is ready to accept one, in which case we know the drive is 1526 * not trying to interrupt us. And ide_set_handler() is always invoked 1527 * before completing the issuance of any new drive command, so we will not 1528 * be accidentally invoked as a result of any valid command completion 1529 * interrupt. 1530 * 1531 * Note that we must walk the entire hwgroup here. We know which hwif 1532 * is doing the current command, but we don't know which hwif burped 1533 * mysteriously. 1534 */ 1535 1536static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup) 1537{ 1538 u8 stat; 1539 ide_hwif_t *hwif = hwgroup->hwif; 1540 1541 /* 1542 * handle the unexpected interrupt 1543 */ 1544 do { 1545 if (hwif->irq == irq) { 1546 stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]); 1547 if (!OK_STAT(stat, READY_STAT, BAD_STAT)) { 1548 /* Try to not flood the console with msgs */ 1549 static unsigned long last_msgtime, count; 1550 ++count; 1551 if (time_after(jiffies, last_msgtime + HZ)) { 1552 last_msgtime = jiffies; 1553 printk(KERN_ERR "%s%s: unexpected interrupt, " 1554 "status=0x%02x, count=%ld\n", 1555 hwif->name, 1556 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count); 1557 } 1558 } 1559 } 1560 } while ((hwif = hwif->next) != hwgroup->hwif); 1561} 1562 1563/** 1564 * ide_intr - default IDE interrupt handler 1565 * @irq: interrupt number 1566 * @dev_id: hwif group 1567 * @regs: unused weirdness from the kernel irq layer 1568 * 1569 * This is the default IRQ handler for the IDE layer. You should 1570 * not need to override it. If you do be aware it is subtle in 1571 * places 1572 * 1573 * hwgroup->hwif is the interface in the group currently performing 1574 * a command. hwgroup->drive is the drive and hwgroup->handler is 1575 * the IRQ handler to call. As we issue a command the handlers 1576 * step through multiple states, reassigning the handler to the 1577 * next step in the process. Unlike a smart SCSI controller IDE 1578 * expects the main processor to sequence the various transfer 1579 * stages. We also manage a poll timer to catch up with most 1580 * timeout situations. There are still a few where the handlers 1581 * don't ever decide to give up. 1582 * 1583 * The handler eventually returns ide_stopped to indicate the 1584 * request completed. At this point we issue the next request 1585 * on the hwgroup and the process begins again. 1586 */ 1587 1588irqreturn_t ide_intr (int irq, void *dev_id) 1589{ 1590 unsigned long flags; 1591 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id; 1592 ide_hwif_t *hwif; 1593 ide_drive_t *drive; 1594 ide_handler_t *handler; 1595 ide_startstop_t startstop; 1596 1597 spin_lock_irqsave(&ide_lock, flags); 1598 hwif = hwgroup->hwif; 1599 1600 if (!ide_ack_intr(hwif)) { 1601 spin_unlock_irqrestore(&ide_lock, flags); 1602 return IRQ_NONE; 1603 } 1604 1605 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) { 1606 /* 1607 * Not expecting an interrupt from this drive. 1608 * That means this could be: 1609 * (1) an interrupt from another PCI device 1610 * sharing the same PCI INT# as us. 1611 * or (2) a drive just entered sleep or standby mode, 1612 * and is interrupting to let us know. 1613 * or (3) a spurious interrupt of unknown origin. 1614 * 1615 * For PCI, we cannot tell the difference, 1616 * so in that case we just ignore it and hope it goes away. 1617 * 1618 * FIXME: unexpected_intr should be hwif-> then we can 1619 * remove all the ifdef PCI crap 1620 */ 1621#ifdef CONFIG_BLK_DEV_IDEPCI 1622 if (hwif->pci_dev && !hwif->pci_dev->vendor) 1623#endif /* CONFIG_BLK_DEV_IDEPCI */ 1624 { 1625 /* 1626 * Probably not a shared PCI interrupt, 1627 * so we can safely try to do something about it: 1628 */ 1629 unexpected_intr(irq, hwgroup); 1630#ifdef CONFIG_BLK_DEV_IDEPCI 1631 } else { 1632 /* 1633 * Whack the status register, just in case 1634 * we have a leftover pending IRQ. 1635 */ 1636 (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]); 1637#endif /* CONFIG_BLK_DEV_IDEPCI */ 1638 } 1639 spin_unlock_irqrestore(&ide_lock, flags); 1640 return IRQ_NONE; 1641 } 1642 drive = hwgroup->drive; 1643 if (!drive) { 1644 /* 1645 * This should NEVER happen, and there isn't much 1646 * we could do about it here. 1647 * 1648 * [Note - this can occur if the drive is hot unplugged] 1649 */ 1650 spin_unlock_irqrestore(&ide_lock, flags); 1651 return IRQ_HANDLED; 1652 } 1653 if (!drive_is_ready(drive)) { 1654 /* 1655 * This happens regularly when we share a PCI IRQ with 1656 * another device. Unfortunately, it can also happen 1657 * with some buggy drives that trigger the IRQ before 1658 * their status register is up to date. Hopefully we have 1659 * enough advance overhead that the latter isn't a problem. 1660 */ 1661 spin_unlock_irqrestore(&ide_lock, flags); 1662 return IRQ_NONE; 1663 } 1664 if (!hwgroup->busy) { 1665 hwgroup->busy = 1; /* paranoia */ 1666 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name); 1667 } 1668 hwgroup->handler = NULL; 1669 hwgroup->req_gen++; 1670 del_timer(&hwgroup->timer); 1671 spin_unlock(&ide_lock); 1672 1673 /* Some controllers might set DMA INTR no matter DMA or PIO; 1674 * bmdma status might need to be cleared even for 1675 * PIO interrupts to prevent spurious/lost irq. 1676 */ 1677 if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma)) 1678 /* ide_dma_end() needs bmdma status for error checking. 1679 * So, skip clearing bmdma status here and leave it 1680 * to ide_dma_end() if this is dma interrupt. 1681 */ 1682 hwif->ide_dma_clear_irq(drive); 1683 1684 if (drive->unmask) 1685 local_irq_enable_in_hardirq(); 1686 /* service this interrupt, may set handler for next interrupt */ 1687 startstop = handler(drive); 1688 spin_lock_irq(&ide_lock); 1689 1690 /* 1691 * Note that handler() may have set things up for another 1692 * interrupt to occur soon, but it cannot happen until 1693 * we exit from this routine, because it will be the 1694 * same irq as is currently being serviced here, and Linux 1695 * won't allow another of the same (on any CPU) until we return. 1696 */ 1697 drive->service_time = jiffies - drive->service_start; 1698 if (startstop == ide_stopped) { 1699 if (hwgroup->handler == NULL) { /* paranoia */ 1700 hwgroup->busy = 0; 1701 ide_do_request(hwgroup, hwif->irq); 1702 } else { 1703 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler " 1704 "on exit\n", drive->name); 1705 } 1706 } 1707 spin_unlock_irqrestore(&ide_lock, flags); 1708 return IRQ_HANDLED; 1709} 1710 1711/** 1712 * ide_init_drive_cmd - initialize a drive command request 1713 * @rq: request object 1714 * 1715 * Initialize a request before we fill it in and send it down to 1716 * ide_do_drive_cmd. Commands must be set up by this function. Right 1717 * now it doesn't do a lot, but if that changes abusers will have a 1718 * nasty surprise. 1719 */ 1720 1721void ide_init_drive_cmd (struct request *rq) 1722{ 1723 memset(rq, 0, sizeof(*rq)); 1724 rq->cmd_type = REQ_TYPE_ATA_CMD; 1725 rq->ref_count = 1; 1726} 1727 1728EXPORT_SYMBOL(ide_init_drive_cmd); 1729 1730/** 1731 * ide_do_drive_cmd - issue IDE special command 1732 * @drive: device to issue command 1733 * @rq: request to issue 1734 * @action: action for processing 1735 * 1736 * This function issues a special IDE device request 1737 * onto the request queue. 1738 * 1739 * If action is ide_wait, then the rq is queued at the end of the 1740 * request queue, and the function sleeps until it has been processed. 1741 * This is for use when invoked from an ioctl handler. 1742 * 1743 * If action is ide_preempt, then the rq is queued at the head of 1744 * the request queue, displacing the currently-being-processed 1745 * request and this function returns immediately without waiting 1746 * for the new rq to be completed. This is VERY DANGEROUS, and is 1747 * intended for careful use by the ATAPI tape/cdrom driver code. 1748 * 1749 * If action is ide_end, then the rq is queued at the end of the 1750 * request queue, and the function returns immediately without waiting 1751 * for the new rq to be completed. This is again intended for careful 1752 * use by the ATAPI tape/cdrom driver code. 1753 */ 1754 1755int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action) 1756{ 1757 unsigned long flags; 1758 ide_hwgroup_t *hwgroup = HWGROUP(drive); 1759 DECLARE_COMPLETION_ONSTACK(wait); 1760 int where = ELEVATOR_INSERT_BACK, err; 1761 int must_wait = (action == ide_wait || action == ide_head_wait); 1762 1763 rq->errors = 0; 1764 1765 /* 1766 * we need to hold an extra reference to request for safe inspection 1767 * after completion 1768 */ 1769 if (must_wait) { 1770 rq->ref_count++; 1771 rq->end_io_data = &wait; 1772 rq->end_io = blk_end_sync_rq; 1773 } 1774 1775 spin_lock_irqsave(&ide_lock, flags); 1776 if (action == ide_preempt) 1777 hwgroup->rq = NULL; 1778 if (action == ide_preempt || action == ide_head_wait) { 1779 where = ELEVATOR_INSERT_FRONT; 1780 rq->cmd_flags |= REQ_PREEMPT; 1781 } 1782 __elv_add_request(drive->queue, rq, where, 0); 1783 ide_do_request(hwgroup, IDE_NO_IRQ); 1784 spin_unlock_irqrestore(&ide_lock, flags); 1785 1786 err = 0; 1787 if (must_wait) { 1788 wait_for_completion(&wait); 1789 if (rq->errors) 1790 err = -EIO; 1791 1792 blk_put_request(rq); 1793 } 1794 1795 return err; 1796} 1797 1798EXPORT_SYMBOL(ide_do_drive_cmd); 1799