1 /* 2 * Functions related to setting various queue properties from drivers 3 */ 4 #include <linux/kernel.h> 5 #include <linux/module.h> 6 #include <linux/init.h> 7 #include <linux/bio.h> 8 #include <linux/blkdev.h> 9 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ 10 #include <linux/gcd.h> 11 #include <linux/jiffies.h> 12 13 #include "blk.h" 14 15 unsigned long blk_max_low_pfn; 16 EXPORT_SYMBOL(blk_max_low_pfn); 17 18 unsigned long blk_max_pfn; 19 20 /** 21 * blk_queue_prep_rq - set a prepare_request function for queue 22 * @q: queue 23 * @pfn: prepare_request function 24 * 25 * It's possible for a queue to register a prepare_request callback which 26 * is invoked before the request is handed to the request_fn. The goal of 27 * the function is to prepare a request for I/O, it can be used to build a 28 * cdb from the request data for instance. 29 * 30 */ 31 void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) 32 { 33 q->prep_rq_fn = pfn; 34 } 35 EXPORT_SYMBOL(blk_queue_prep_rq); 36 37 /** 38 * blk_queue_merge_bvec - set a merge_bvec function for queue 39 * @q: queue 40 * @mbfn: merge_bvec_fn 41 * 42 * Usually queues have static limitations on the max sectors or segments that 43 * we can put in a request. Stacking drivers may have some settings that 44 * are dynamic, and thus we have to query the queue whether it is ok to 45 * add a new bio_vec to a bio at a given offset or not. If the block device 46 * has such limitations, it needs to register a merge_bvec_fn to control 47 * the size of bio's sent to it. Note that a block device *must* allow a 48 * single page to be added to an empty bio. The block device driver may want 49 * to use the bio_split() function to deal with these bio's. By default 50 * no merge_bvec_fn is defined for a queue, and only the fixed limits are 51 * honored. 52 */ 53 void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) 54 { 55 q->merge_bvec_fn = mbfn; 56 } 57 EXPORT_SYMBOL(blk_queue_merge_bvec); 58 59 void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) 60 { 61 q->softirq_done_fn = fn; 62 } 63 EXPORT_SYMBOL(blk_queue_softirq_done); 64 65 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) 66 { 67 q->rq_timeout = timeout; 68 } 69 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); 70 71 void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) 72 { 73 q->rq_timed_out_fn = fn; 74 } 75 EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); 76 77 void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) 78 { 79 q->lld_busy_fn = fn; 80 } 81 EXPORT_SYMBOL_GPL(blk_queue_lld_busy); 82 83 /** 84 * blk_set_default_limits - reset limits to default values 85 * @lim: the queue_limits structure to reset 86 * 87 * Description: 88 * Returns a queue_limit struct to its default state. Can be used by 89 * stacking drivers like DM that stage table swaps and reuse an 90 * existing device queue. 91 */ 92 void blk_set_default_limits(struct queue_limits *lim) 93 { 94 lim->max_phys_segments = MAX_PHYS_SEGMENTS; 95 lim->max_hw_segments = MAX_HW_SEGMENTS; 96 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; 97 lim->max_segment_size = MAX_SEGMENT_SIZE; 98 lim->max_sectors = BLK_DEF_MAX_SECTORS; 99 lim->max_hw_sectors = INT_MAX; 100 lim->max_discard_sectors = 0; 101 lim->discard_granularity = 0; 102 lim->discard_alignment = 0; 103 lim->discard_misaligned = 0; 104 lim->discard_zeroes_data = -1; 105 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; 106 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); 107 lim->alignment_offset = 0; 108 lim->io_opt = 0; 109 lim->misaligned = 0; 110 lim->no_cluster = 0; 111 } 112 EXPORT_SYMBOL(blk_set_default_limits); 113 114 /** 115 * blk_queue_make_request - define an alternate make_request function for a device 116 * @q: the request queue for the device to be affected 117 * @mfn: the alternate make_request function 118 * 119 * Description: 120 * The normal way for &struct bios to be passed to a device 121 * driver is for them to be collected into requests on a request 122 * queue, and then to allow the device driver to select requests 123 * off that queue when it is ready. This works well for many block 124 * devices. However some block devices (typically virtual devices 125 * such as md or lvm) do not benefit from the processing on the 126 * request queue, and are served best by having the requests passed 127 * directly to them. This can be achieved by providing a function 128 * to blk_queue_make_request(). 129 * 130 * Caveat: 131 * The driver that does this *must* be able to deal appropriately 132 * with buffers in "highmemory". This can be accomplished by either calling 133 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling 134 * blk_queue_bounce() to create a buffer in normal memory. 135 **/ 136 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) 137 { 138 /* 139 * set defaults 140 */ 141 q->nr_requests = BLKDEV_MAX_RQ; 142 143 q->make_request_fn = mfn; 144 blk_queue_dma_alignment(q, 511); 145 blk_queue_congestion_threshold(q); 146 q->nr_batching = BLK_BATCH_REQ; 147 148 q->unplug_thresh = 4; /* hmm */ 149 q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */ 150 if (q->unplug_delay == 0) 151 q->unplug_delay = 1; 152 153 q->unplug_timer.function = blk_unplug_timeout; 154 q->unplug_timer.data = (unsigned long)q; 155 156 blk_set_default_limits(&q->limits); 157 blk_queue_max_sectors(q, SAFE_MAX_SECTORS); 158 159 /* 160 * If the caller didn't supply a lock, fall back to our embedded 161 * per-queue locks 162 */ 163 if (!q->queue_lock) 164 q->queue_lock = &q->__queue_lock; 165 166 /* 167 * by default assume old behaviour and bounce for any highmem page 168 */ 169 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); 170 } 171 EXPORT_SYMBOL(blk_queue_make_request); 172 173 /** 174 * blk_queue_bounce_limit - set bounce buffer limit for queue 175 * @q: the request queue for the device 176 * @dma_mask: the maximum address the device can handle 177 * 178 * Description: 179 * Different hardware can have different requirements as to what pages 180 * it can do I/O directly to. A low level driver can call 181 * blk_queue_bounce_limit to have lower memory pages allocated as bounce 182 * buffers for doing I/O to pages residing above @dma_mask. 183 **/ 184 void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) 185 { 186 unsigned long b_pfn = dma_mask >> PAGE_SHIFT; 187 int dma = 0; 188 189 q->bounce_gfp = GFP_NOIO; 190 #if BITS_PER_LONG == 64 191 /* 192 * Assume anything <= 4GB can be handled by IOMMU. Actually 193 * some IOMMUs can handle everything, but I don't know of a 194 * way to test this here. 195 */ 196 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) 197 dma = 1; 198 q->limits.bounce_pfn = max_low_pfn; 199 #else 200 if (b_pfn < blk_max_low_pfn) 201 dma = 1; 202 q->limits.bounce_pfn = b_pfn; 203 #endif 204 if (dma) { 205 init_emergency_isa_pool(); 206 q->bounce_gfp = GFP_NOIO | GFP_DMA; 207 q->limits.bounce_pfn = b_pfn; 208 } 209 } 210 EXPORT_SYMBOL(blk_queue_bounce_limit); 211 212 /** 213 * blk_queue_max_sectors - set max sectors for a request for this queue 214 * @q: the request queue for the device 215 * @max_sectors: max sectors in the usual 512b unit 216 * 217 * Description: 218 * Enables a low level driver to set an upper limit on the size of 219 * received requests. 220 **/ 221 void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) 222 { 223 if ((max_sectors << 9) < PAGE_CACHE_SIZE) { 224 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); 225 printk(KERN_INFO "%s: set to minimum %d\n", 226 __func__, max_sectors); 227 } 228 229 if (BLK_DEF_MAX_SECTORS > max_sectors) 230 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors; 231 else { 232 q->limits.max_sectors = BLK_DEF_MAX_SECTORS; 233 q->limits.max_hw_sectors = max_sectors; 234 } 235 } 236 EXPORT_SYMBOL(blk_queue_max_sectors); 237 238 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors) 239 { 240 if (BLK_DEF_MAX_SECTORS > max_sectors) 241 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS; 242 else 243 q->limits.max_hw_sectors = max_sectors; 244 } 245 EXPORT_SYMBOL(blk_queue_max_hw_sectors); 246 247 /** 248 * blk_queue_max_discard_sectors - set max sectors for a single discard 249 * @q: the request queue for the device 250 * @max_discard_sectors: maximum number of sectors to discard 251 **/ 252 void blk_queue_max_discard_sectors(struct request_queue *q, 253 unsigned int max_discard_sectors) 254 { 255 q->limits.max_discard_sectors = max_discard_sectors; 256 } 257 EXPORT_SYMBOL(blk_queue_max_discard_sectors); 258 259 /** 260 * blk_queue_max_phys_segments - set max phys segments for a request for this queue 261 * @q: the request queue for the device 262 * @max_segments: max number of segments 263 * 264 * Description: 265 * Enables a low level driver to set an upper limit on the number of 266 * physical data segments in a request. This would be the largest sized 267 * scatter list the driver could handle. 268 **/ 269 void blk_queue_max_phys_segments(struct request_queue *q, 270 unsigned short max_segments) 271 { 272 if (!max_segments) { 273 max_segments = 1; 274 printk(KERN_INFO "%s: set to minimum %d\n", 275 __func__, max_segments); 276 } 277 278 q->limits.max_phys_segments = max_segments; 279 } 280 EXPORT_SYMBOL(blk_queue_max_phys_segments); 281 282 /** 283 * blk_queue_max_hw_segments - set max hw segments for a request for this queue 284 * @q: the request queue for the device 285 * @max_segments: max number of segments 286 * 287 * Description: 288 * Enables a low level driver to set an upper limit on the number of 289 * hw data segments in a request. This would be the largest number of 290 * address/length pairs the host adapter can actually give at once 291 * to the device. 292 **/ 293 void blk_queue_max_hw_segments(struct request_queue *q, 294 unsigned short max_segments) 295 { 296 if (!max_segments) { 297 max_segments = 1; 298 printk(KERN_INFO "%s: set to minimum %d\n", 299 __func__, max_segments); 300 } 301 302 q->limits.max_hw_segments = max_segments; 303 } 304 EXPORT_SYMBOL(blk_queue_max_hw_segments); 305 306 /** 307 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg 308 * @q: the request queue for the device 309 * @max_size: max size of segment in bytes 310 * 311 * Description: 312 * Enables a low level driver to set an upper limit on the size of a 313 * coalesced segment 314 **/ 315 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) 316 { 317 if (max_size < PAGE_CACHE_SIZE) { 318 max_size = PAGE_CACHE_SIZE; 319 printk(KERN_INFO "%s: set to minimum %d\n", 320 __func__, max_size); 321 } 322 323 q->limits.max_segment_size = max_size; 324 } 325 EXPORT_SYMBOL(blk_queue_max_segment_size); 326 327 /** 328 * blk_queue_logical_block_size - set logical block size for the queue 329 * @q: the request queue for the device 330 * @size: the logical block size, in bytes 331 * 332 * Description: 333 * This should be set to the lowest possible block size that the 334 * storage device can address. The default of 512 covers most 335 * hardware. 336 **/ 337 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) 338 { 339 q->limits.logical_block_size = size; 340 341 if (q->limits.physical_block_size < size) 342 q->limits.physical_block_size = size; 343 344 if (q->limits.io_min < q->limits.physical_block_size) 345 q->limits.io_min = q->limits.physical_block_size; 346 } 347 EXPORT_SYMBOL(blk_queue_logical_block_size); 348 349 /** 350 * blk_queue_physical_block_size - set physical block size for the queue 351 * @q: the request queue for the device 352 * @size: the physical block size, in bytes 353 * 354 * Description: 355 * This should be set to the lowest possible sector size that the 356 * hardware can operate on without reverting to read-modify-write 357 * operations. 358 */ 359 void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) 360 { 361 q->limits.physical_block_size = size; 362 363 if (q->limits.physical_block_size < q->limits.logical_block_size) 364 q->limits.physical_block_size = q->limits.logical_block_size; 365 366 if (q->limits.io_min < q->limits.physical_block_size) 367 q->limits.io_min = q->limits.physical_block_size; 368 } 369 EXPORT_SYMBOL(blk_queue_physical_block_size); 370 371 /** 372 * blk_queue_alignment_offset - set physical block alignment offset 373 * @q: the request queue for the device 374 * @offset: alignment offset in bytes 375 * 376 * Description: 377 * Some devices are naturally misaligned to compensate for things like 378 * the legacy DOS partition table 63-sector offset. Low-level drivers 379 * should call this function for devices whose first sector is not 380 * naturally aligned. 381 */ 382 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) 383 { 384 q->limits.alignment_offset = 385 offset & (q->limits.physical_block_size - 1); 386 q->limits.misaligned = 0; 387 } 388 EXPORT_SYMBOL(blk_queue_alignment_offset); 389 390 /** 391 * blk_limits_io_min - set minimum request size for a device 392 * @limits: the queue limits 393 * @min: smallest I/O size in bytes 394 * 395 * Description: 396 * Some devices have an internal block size bigger than the reported 397 * hardware sector size. This function can be used to signal the 398 * smallest I/O the device can perform without incurring a performance 399 * penalty. 400 */ 401 void blk_limits_io_min(struct queue_limits *limits, unsigned int min) 402 { 403 limits->io_min = min; 404 405 if (limits->io_min < limits->logical_block_size) 406 limits->io_min = limits->logical_block_size; 407 408 if (limits->io_min < limits->physical_block_size) 409 limits->io_min = limits->physical_block_size; 410 } 411 EXPORT_SYMBOL(blk_limits_io_min); 412 413 /** 414 * blk_queue_io_min - set minimum request size for the queue 415 * @q: the request queue for the device 416 * @min: smallest I/O size in bytes 417 * 418 * Description: 419 * Storage devices may report a granularity or preferred minimum I/O 420 * size which is the smallest request the device can perform without 421 * incurring a performance penalty. For disk drives this is often the 422 * physical block size. For RAID arrays it is often the stripe chunk 423 * size. A properly aligned multiple of minimum_io_size is the 424 * preferred request size for workloads where a high number of I/O 425 * operations is desired. 426 */ 427 void blk_queue_io_min(struct request_queue *q, unsigned int min) 428 { 429 blk_limits_io_min(&q->limits, min); 430 } 431 EXPORT_SYMBOL(blk_queue_io_min); 432 433 /** 434 * blk_limits_io_opt - set optimal request size for a device 435 * @limits: the queue limits 436 * @opt: smallest I/O size in bytes 437 * 438 * Description: 439 * Storage devices may report an optimal I/O size, which is the 440 * device's preferred unit for sustained I/O. This is rarely reported 441 * for disk drives. For RAID arrays it is usually the stripe width or 442 * the internal track size. A properly aligned multiple of 443 * optimal_io_size is the preferred request size for workloads where 444 * sustained throughput is desired. 445 */ 446 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) 447 { 448 limits->io_opt = opt; 449 } 450 EXPORT_SYMBOL(blk_limits_io_opt); 451 452 /** 453 * blk_queue_io_opt - set optimal request size for the queue 454 * @q: the request queue for the device 455 * @opt: optimal request size in bytes 456 * 457 * Description: 458 * Storage devices may report an optimal I/O size, which is the 459 * device's preferred unit for sustained I/O. This is rarely reported 460 * for disk drives. For RAID arrays it is usually the stripe width or 461 * the internal track size. A properly aligned multiple of 462 * optimal_io_size is the preferred request size for workloads where 463 * sustained throughput is desired. 464 */ 465 void blk_queue_io_opt(struct request_queue *q, unsigned int opt) 466 { 467 blk_limits_io_opt(&q->limits, opt); 468 } 469 EXPORT_SYMBOL(blk_queue_io_opt); 470 471 /* 472 * Returns the minimum that is _not_ zero, unless both are zero. 473 */ 474 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) 475 476 /** 477 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers 478 * @t: the stacking driver (top) 479 * @b: the underlying device (bottom) 480 **/ 481 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) 482 { 483 blk_stack_limits(&t->limits, &b->limits, 0); 484 485 if (!t->queue_lock) 486 WARN_ON_ONCE(1); 487 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { 488 unsigned long flags; 489 spin_lock_irqsave(t->queue_lock, flags); 490 queue_flag_clear(QUEUE_FLAG_CLUSTER, t); 491 spin_unlock_irqrestore(t->queue_lock, flags); 492 } 493 } 494 EXPORT_SYMBOL(blk_queue_stack_limits); 495 496 static unsigned int lcm(unsigned int a, unsigned int b) 497 { 498 if (a && b) 499 return (a * b) / gcd(a, b); 500 else if (b) 501 return b; 502 503 return a; 504 } 505 506 /** 507 * blk_stack_limits - adjust queue_limits for stacked devices 508 * @t: the stacking driver limits (top device) 509 * @b: the underlying queue limits (bottom, component device) 510 * @offset: offset to beginning of data within component device 511 * 512 * Description: 513 * This function is used by stacking drivers like MD and DM to ensure 514 * that all component devices have compatible block sizes and 515 * alignments. The stacking driver must provide a queue_limits 516 * struct (top) and then iteratively call the stacking function for 517 * all component (bottom) devices. The stacking function will 518 * attempt to combine the values and ensure proper alignment. 519 * 520 * Returns 0 if the top and bottom queue_limits are compatible. The 521 * top device's block sizes and alignment offsets may be adjusted to 522 * ensure alignment with the bottom device. If no compatible sizes 523 * and alignments exist, -1 is returned and the resulting top 524 * queue_limits will have the misaligned flag set to indicate that 525 * the alignment_offset is undefined. 526 */ 527 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, 528 sector_t offset) 529 { 530 sector_t alignment; 531 unsigned int top, bottom, ret = 0; 532 533 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); 534 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); 535 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); 536 537 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, 538 b->seg_boundary_mask); 539 540 t->max_phys_segments = min_not_zero(t->max_phys_segments, 541 b->max_phys_segments); 542 543 t->max_hw_segments = min_not_zero(t->max_hw_segments, 544 b->max_hw_segments); 545 546 t->max_segment_size = min_not_zero(t->max_segment_size, 547 b->max_segment_size); 548 549 t->misaligned |= b->misaligned; 550 551 alignment = queue_limit_alignment_offset(b, offset); 552 553 /* Bottom device has different alignment. Check that it is 554 * compatible with the current top alignment. 555 */ 556 if (t->alignment_offset != alignment) { 557 558 top = max(t->physical_block_size, t->io_min) 559 + t->alignment_offset; 560 bottom = max(b->physical_block_size, b->io_min) + alignment; 561 562 /* Verify that top and bottom intervals line up */ 563 if (max(top, bottom) & (min(top, bottom) - 1)) { 564 t->misaligned = 1; 565 ret = -1; 566 } 567 } 568 569 t->logical_block_size = max(t->logical_block_size, 570 b->logical_block_size); 571 572 t->physical_block_size = max(t->physical_block_size, 573 b->physical_block_size); 574 575 t->io_min = max(t->io_min, b->io_min); 576 t->io_opt = lcm(t->io_opt, b->io_opt); 577 578 t->no_cluster |= b->no_cluster; 579 t->discard_zeroes_data &= b->discard_zeroes_data; 580 581 /* Physical block size a multiple of the logical block size? */ 582 if (t->physical_block_size & (t->logical_block_size - 1)) { 583 t->physical_block_size = t->logical_block_size; 584 t->misaligned = 1; 585 ret = -1; 586 } 587 588 /* Minimum I/O a multiple of the physical block size? */ 589 if (t->io_min & (t->physical_block_size - 1)) { 590 t->io_min = t->physical_block_size; 591 t->misaligned = 1; 592 ret = -1; 593 } 594 595 /* Optimal I/O a multiple of the physical block size? */ 596 if (t->io_opt & (t->physical_block_size - 1)) { 597 t->io_opt = 0; 598 t->misaligned = 1; 599 ret = -1; 600 } 601 602 /* Find lowest common alignment_offset */ 603 t->alignment_offset = lcm(t->alignment_offset, alignment) 604 & (max(t->physical_block_size, t->io_min) - 1); 605 606 /* Verify that new alignment_offset is on a logical block boundary */ 607 if (t->alignment_offset & (t->logical_block_size - 1)) { 608 t->misaligned = 1; 609 ret = -1; 610 } 611 612 /* Discard alignment and granularity */ 613 if (b->discard_granularity) { 614 unsigned int granularity = b->discard_granularity; 615 offset &= granularity - 1; 616 617 alignment = (granularity + b->discard_alignment - offset) 618 & (granularity - 1); 619 620 if (t->discard_granularity != 0 && 621 t->discard_alignment != alignment) { 622 top = t->discard_granularity + t->discard_alignment; 623 bottom = b->discard_granularity + alignment; 624 625 /* Verify that top and bottom intervals line up */ 626 if (max(top, bottom) & (min(top, bottom) - 1)) 627 t->discard_misaligned = 1; 628 } 629 630 t->max_discard_sectors = min_not_zero(t->max_discard_sectors, 631 b->max_discard_sectors); 632 t->discard_granularity = max(t->discard_granularity, 633 b->discard_granularity); 634 t->discard_alignment = lcm(t->discard_alignment, alignment) & 635 (t->discard_granularity - 1); 636 } 637 638 return ret; 639 } 640 EXPORT_SYMBOL(blk_stack_limits); 641 642 /** 643 * bdev_stack_limits - adjust queue limits for stacked drivers 644 * @t: the stacking driver limits (top device) 645 * @bdev: the component block_device (bottom) 646 * @start: first data sector within component device 647 * 648 * Description: 649 * Merges queue limits for a top device and a block_device. Returns 650 * 0 if alignment didn't change. Returns -1 if adding the bottom 651 * device caused misalignment. 652 */ 653 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, 654 sector_t start) 655 { 656 struct request_queue *bq = bdev_get_queue(bdev); 657 658 start += get_start_sect(bdev); 659 660 return blk_stack_limits(t, &bq->limits, start << 9); 661 } 662 EXPORT_SYMBOL(bdev_stack_limits); 663 664 /** 665 * disk_stack_limits - adjust queue limits for stacked drivers 666 * @disk: MD/DM gendisk (top) 667 * @bdev: the underlying block device (bottom) 668 * @offset: offset to beginning of data within component device 669 * 670 * Description: 671 * Merges the limits for two queues. Returns 0 if alignment 672 * didn't change. Returns -1 if adding the bottom device caused 673 * misalignment. 674 */ 675 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, 676 sector_t offset) 677 { 678 struct request_queue *t = disk->queue; 679 struct request_queue *b = bdev_get_queue(bdev); 680 681 offset += get_start_sect(bdev) << 9; 682 683 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) { 684 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; 685 686 disk_name(disk, 0, top); 687 bdevname(bdev, bottom); 688 689 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", 690 top, bottom); 691 } 692 693 if (!t->queue_lock) 694 WARN_ON_ONCE(1); 695 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { 696 unsigned long flags; 697 698 spin_lock_irqsave(t->queue_lock, flags); 699 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) 700 queue_flag_clear(QUEUE_FLAG_CLUSTER, t); 701 spin_unlock_irqrestore(t->queue_lock, flags); 702 } 703 } 704 EXPORT_SYMBOL(disk_stack_limits); 705 706 /** 707 * blk_queue_dma_pad - set pad mask 708 * @q: the request queue for the device 709 * @mask: pad mask 710 * 711 * Set dma pad mask. 712 * 713 * Appending pad buffer to a request modifies the last entry of a 714 * scatter list such that it includes the pad buffer. 715 **/ 716 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) 717 { 718 q->dma_pad_mask = mask; 719 } 720 EXPORT_SYMBOL(blk_queue_dma_pad); 721 722 /** 723 * blk_queue_update_dma_pad - update pad mask 724 * @q: the request queue for the device 725 * @mask: pad mask 726 * 727 * Update dma pad mask. 728 * 729 * Appending pad buffer to a request modifies the last entry of a 730 * scatter list such that it includes the pad buffer. 731 **/ 732 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) 733 { 734 if (mask > q->dma_pad_mask) 735 q->dma_pad_mask = mask; 736 } 737 EXPORT_SYMBOL(blk_queue_update_dma_pad); 738 739 /** 740 * blk_queue_dma_drain - Set up a drain buffer for excess dma. 741 * @q: the request queue for the device 742 * @dma_drain_needed: fn which returns non-zero if drain is necessary 743 * @buf: physically contiguous buffer 744 * @size: size of the buffer in bytes 745 * 746 * Some devices have excess DMA problems and can't simply discard (or 747 * zero fill) the unwanted piece of the transfer. They have to have a 748 * real area of memory to transfer it into. The use case for this is 749 * ATAPI devices in DMA mode. If the packet command causes a transfer 750 * bigger than the transfer size some HBAs will lock up if there 751 * aren't DMA elements to contain the excess transfer. What this API 752 * does is adjust the queue so that the buf is always appended 753 * silently to the scatterlist. 754 * 755 * Note: This routine adjusts max_hw_segments to make room for 756 * appending the drain buffer. If you call 757 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after 758 * calling this routine, you must set the limit to one fewer than your 759 * device can support otherwise there won't be room for the drain 760 * buffer. 761 */ 762 int blk_queue_dma_drain(struct request_queue *q, 763 dma_drain_needed_fn *dma_drain_needed, 764 void *buf, unsigned int size) 765 { 766 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2) 767 return -EINVAL; 768 /* make room for appending the drain */ 769 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1); 770 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1); 771 q->dma_drain_needed = dma_drain_needed; 772 q->dma_drain_buffer = buf; 773 q->dma_drain_size = size; 774 775 return 0; 776 } 777 EXPORT_SYMBOL_GPL(blk_queue_dma_drain); 778 779 /** 780 * blk_queue_segment_boundary - set boundary rules for segment merging 781 * @q: the request queue for the device 782 * @mask: the memory boundary mask 783 **/ 784 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) 785 { 786 if (mask < PAGE_CACHE_SIZE - 1) { 787 mask = PAGE_CACHE_SIZE - 1; 788 printk(KERN_INFO "%s: set to minimum %lx\n", 789 __func__, mask); 790 } 791 792 q->limits.seg_boundary_mask = mask; 793 } 794 EXPORT_SYMBOL(blk_queue_segment_boundary); 795 796 /** 797 * blk_queue_dma_alignment - set dma length and memory alignment 798 * @q: the request queue for the device 799 * @mask: alignment mask 800 * 801 * description: 802 * set required memory and length alignment for direct dma transactions. 803 * this is used when building direct io requests for the queue. 804 * 805 **/ 806 void blk_queue_dma_alignment(struct request_queue *q, int mask) 807 { 808 q->dma_alignment = mask; 809 } 810 EXPORT_SYMBOL(blk_queue_dma_alignment); 811 812 /** 813 * blk_queue_update_dma_alignment - update dma length and memory alignment 814 * @q: the request queue for the device 815 * @mask: alignment mask 816 * 817 * description: 818 * update required memory and length alignment for direct dma transactions. 819 * If the requested alignment is larger than the current alignment, then 820 * the current queue alignment is updated to the new value, otherwise it 821 * is left alone. The design of this is to allow multiple objects 822 * (driver, device, transport etc) to set their respective 823 * alignments without having them interfere. 824 * 825 **/ 826 void blk_queue_update_dma_alignment(struct request_queue *q, int mask) 827 { 828 BUG_ON(mask > PAGE_SIZE); 829 830 if (mask > q->dma_alignment) 831 q->dma_alignment = mask; 832 } 833 EXPORT_SYMBOL(blk_queue_update_dma_alignment); 834 835 static int __init blk_settings_init(void) 836 { 837 blk_max_low_pfn = max_low_pfn - 1; 838 blk_max_pfn = max_pfn - 1; 839 return 0; 840 } 841 subsys_initcall(blk_settings_init); 842