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/lcm.h> 12 #include <linux/jiffies.h> 13 #include <linux/gfp.h> 14 15 #include "blk.h" 16 17 unsigned long blk_max_low_pfn; 18 EXPORT_SYMBOL(blk_max_low_pfn); 19 20 unsigned long blk_max_pfn; 21 22 /** 23 * blk_queue_prep_rq - set a prepare_request function for queue 24 * @q: queue 25 * @pfn: prepare_request function 26 * 27 * It's possible for a queue to register a prepare_request callback which 28 * is invoked before the request is handed to the request_fn. The goal of 29 * the function is to prepare a request for I/O, it can be used to build a 30 * cdb from the request data for instance. 31 * 32 */ 33 void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) 34 { 35 q->prep_rq_fn = pfn; 36 } 37 EXPORT_SYMBOL(blk_queue_prep_rq); 38 39 /** 40 * blk_queue_unprep_rq - set an unprepare_request function for queue 41 * @q: queue 42 * @ufn: unprepare_request function 43 * 44 * It's possible for a queue to register an unprepare_request callback 45 * which is invoked before the request is finally completed. The goal 46 * of the function is to deallocate any data that was allocated in the 47 * prepare_request callback. 48 * 49 */ 50 void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn) 51 { 52 q->unprep_rq_fn = ufn; 53 } 54 EXPORT_SYMBOL(blk_queue_unprep_rq); 55 56 /** 57 * blk_queue_merge_bvec - set a merge_bvec function for queue 58 * @q: queue 59 * @mbfn: merge_bvec_fn 60 * 61 * Usually queues have static limitations on the max sectors or segments that 62 * we can put in a request. Stacking drivers may have some settings that 63 * are dynamic, and thus we have to query the queue whether it is ok to 64 * add a new bio_vec to a bio at a given offset or not. If the block device 65 * has such limitations, it needs to register a merge_bvec_fn to control 66 * the size of bio's sent to it. Note that a block device *must* allow a 67 * single page to be added to an empty bio. The block device driver may want 68 * to use the bio_split() function to deal with these bio's. By default 69 * no merge_bvec_fn is defined for a queue, and only the fixed limits are 70 * honored. 71 */ 72 void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) 73 { 74 q->merge_bvec_fn = mbfn; 75 } 76 EXPORT_SYMBOL(blk_queue_merge_bvec); 77 78 void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) 79 { 80 q->softirq_done_fn = fn; 81 } 82 EXPORT_SYMBOL(blk_queue_softirq_done); 83 84 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) 85 { 86 q->rq_timeout = timeout; 87 } 88 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); 89 90 void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) 91 { 92 q->rq_timed_out_fn = fn; 93 } 94 EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); 95 96 void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) 97 { 98 q->lld_busy_fn = fn; 99 } 100 EXPORT_SYMBOL_GPL(blk_queue_lld_busy); 101 102 /** 103 * blk_set_default_limits - reset limits to default values 104 * @lim: the queue_limits structure to reset 105 * 106 * Description: 107 * Returns a queue_limit struct to its default state. 108 */ 109 void blk_set_default_limits(struct queue_limits *lim) 110 { 111 lim->max_segments = BLK_MAX_SEGMENTS; 112 lim->max_integrity_segments = 0; 113 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; 114 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; 115 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; 116 lim->chunk_sectors = 0; 117 lim->max_write_same_sectors = 0; 118 lim->max_discard_sectors = 0; 119 lim->discard_granularity = 0; 120 lim->discard_alignment = 0; 121 lim->discard_misaligned = 0; 122 lim->discard_zeroes_data = 0; 123 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; 124 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); 125 lim->alignment_offset = 0; 126 lim->io_opt = 0; 127 lim->misaligned = 0; 128 lim->cluster = 1; 129 } 130 EXPORT_SYMBOL(blk_set_default_limits); 131 132 /** 133 * blk_set_stacking_limits - set default limits for stacking devices 134 * @lim: the queue_limits structure to reset 135 * 136 * Description: 137 * Returns a queue_limit struct to its default state. Should be used 138 * by stacking drivers like DM that have no internal limits. 139 */ 140 void blk_set_stacking_limits(struct queue_limits *lim) 141 { 142 blk_set_default_limits(lim); 143 144 /* Inherit limits from component devices */ 145 lim->discard_zeroes_data = 1; 146 lim->max_segments = USHRT_MAX; 147 lim->max_hw_sectors = UINT_MAX; 148 lim->max_segment_size = UINT_MAX; 149 lim->max_sectors = UINT_MAX; 150 lim->max_write_same_sectors = UINT_MAX; 151 } 152 EXPORT_SYMBOL(blk_set_stacking_limits); 153 154 /** 155 * blk_queue_make_request - define an alternate make_request function for a device 156 * @q: the request queue for the device to be affected 157 * @mfn: the alternate make_request function 158 * 159 * Description: 160 * The normal way for &struct bios to be passed to a device 161 * driver is for them to be collected into requests on a request 162 * queue, and then to allow the device driver to select requests 163 * off that queue when it is ready. This works well for many block 164 * devices. However some block devices (typically virtual devices 165 * such as md or lvm) do not benefit from the processing on the 166 * request queue, and are served best by having the requests passed 167 * directly to them. This can be achieved by providing a function 168 * to blk_queue_make_request(). 169 * 170 * Caveat: 171 * The driver that does this *must* be able to deal appropriately 172 * with buffers in "highmemory". This can be accomplished by either calling 173 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling 174 * blk_queue_bounce() to create a buffer in normal memory. 175 **/ 176 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) 177 { 178 /* 179 * set defaults 180 */ 181 q->nr_requests = BLKDEV_MAX_RQ; 182 183 q->make_request_fn = mfn; 184 blk_queue_dma_alignment(q, 511); 185 blk_queue_congestion_threshold(q); 186 q->nr_batching = BLK_BATCH_REQ; 187 188 blk_set_default_limits(&q->limits); 189 190 /* 191 * by default assume old behaviour and bounce for any highmem page 192 */ 193 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); 194 } 195 EXPORT_SYMBOL(blk_queue_make_request); 196 197 /** 198 * blk_queue_bounce_limit - set bounce buffer limit for queue 199 * @q: the request queue for the device 200 * @max_addr: the maximum address the device can handle 201 * 202 * Description: 203 * Different hardware can have different requirements as to what pages 204 * it can do I/O directly to. A low level driver can call 205 * blk_queue_bounce_limit to have lower memory pages allocated as bounce 206 * buffers for doing I/O to pages residing above @max_addr. 207 **/ 208 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr) 209 { 210 unsigned long b_pfn = max_addr >> PAGE_SHIFT; 211 int dma = 0; 212 213 q->bounce_gfp = GFP_NOIO; 214 #if BITS_PER_LONG == 64 215 /* 216 * Assume anything <= 4GB can be handled by IOMMU. Actually 217 * some IOMMUs can handle everything, but I don't know of a 218 * way to test this here. 219 */ 220 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) 221 dma = 1; 222 q->limits.bounce_pfn = max(max_low_pfn, b_pfn); 223 #else 224 if (b_pfn < blk_max_low_pfn) 225 dma = 1; 226 q->limits.bounce_pfn = b_pfn; 227 #endif 228 if (dma) { 229 init_emergency_isa_pool(); 230 q->bounce_gfp = GFP_NOIO | GFP_DMA; 231 q->limits.bounce_pfn = b_pfn; 232 } 233 } 234 EXPORT_SYMBOL(blk_queue_bounce_limit); 235 236 /** 237 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request 238 * @limits: the queue limits 239 * @max_hw_sectors: max hardware sectors in the usual 512b unit 240 * 241 * Description: 242 * Enables a low level driver to set a hard upper limit, 243 * max_hw_sectors, on the size of requests. max_hw_sectors is set by 244 * the device driver based upon the combined capabilities of I/O 245 * controller and storage device. 246 * 247 * max_sectors is a soft limit imposed by the block layer for 248 * filesystem type requests. This value can be overridden on a 249 * per-device basis in /sys/block/<device>/queue/max_sectors_kb. 250 * The soft limit can not exceed max_hw_sectors. 251 **/ 252 void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors) 253 { 254 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) { 255 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9); 256 printk(KERN_INFO "%s: set to minimum %d\n", 257 __func__, max_hw_sectors); 258 } 259 260 limits->max_sectors = limits->max_hw_sectors = max_hw_sectors; 261 } 262 EXPORT_SYMBOL(blk_limits_max_hw_sectors); 263 264 /** 265 * blk_queue_max_hw_sectors - set max sectors for a request for this queue 266 * @q: the request queue for the device 267 * @max_hw_sectors: max hardware sectors in the usual 512b unit 268 * 269 * Description: 270 * See description for blk_limits_max_hw_sectors(). 271 **/ 272 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) 273 { 274 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors); 275 } 276 EXPORT_SYMBOL(blk_queue_max_hw_sectors); 277 278 /** 279 * blk_queue_chunk_sectors - set size of the chunk for this queue 280 * @q: the request queue for the device 281 * @chunk_sectors: chunk sectors in the usual 512b unit 282 * 283 * Description: 284 * If a driver doesn't want IOs to cross a given chunk size, it can set 285 * this limit and prevent merging across chunks. Note that the chunk size 286 * must currently be a power-of-2 in sectors. Also note that the block 287 * layer must accept a page worth of data at any offset. So if the 288 * crossing of chunks is a hard limitation in the driver, it must still be 289 * prepared to split single page bios. 290 **/ 291 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) 292 { 293 BUG_ON(!is_power_of_2(chunk_sectors)); 294 q->limits.chunk_sectors = chunk_sectors; 295 } 296 EXPORT_SYMBOL(blk_queue_chunk_sectors); 297 298 /** 299 * blk_queue_max_discard_sectors - set max sectors for a single discard 300 * @q: the request queue for the device 301 * @max_discard_sectors: maximum number of sectors to discard 302 **/ 303 void blk_queue_max_discard_sectors(struct request_queue *q, 304 unsigned int max_discard_sectors) 305 { 306 q->limits.max_discard_sectors = max_discard_sectors; 307 } 308 EXPORT_SYMBOL(blk_queue_max_discard_sectors); 309 310 /** 311 * blk_queue_max_write_same_sectors - set max sectors for a single write same 312 * @q: the request queue for the device 313 * @max_write_same_sectors: maximum number of sectors to write per command 314 **/ 315 void blk_queue_max_write_same_sectors(struct request_queue *q, 316 unsigned int max_write_same_sectors) 317 { 318 q->limits.max_write_same_sectors = max_write_same_sectors; 319 } 320 EXPORT_SYMBOL(blk_queue_max_write_same_sectors); 321 322 /** 323 * blk_queue_max_segments - set max hw segments for a request for this queue 324 * @q: the request queue for the device 325 * @max_segments: max number of segments 326 * 327 * Description: 328 * Enables a low level driver to set an upper limit on the number of 329 * hw data segments in a request. 330 **/ 331 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) 332 { 333 if (!max_segments) { 334 max_segments = 1; 335 printk(KERN_INFO "%s: set to minimum %d\n", 336 __func__, max_segments); 337 } 338 339 q->limits.max_segments = max_segments; 340 } 341 EXPORT_SYMBOL(blk_queue_max_segments); 342 343 /** 344 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg 345 * @q: the request queue for the device 346 * @max_size: max size of segment in bytes 347 * 348 * Description: 349 * Enables a low level driver to set an upper limit on the size of a 350 * coalesced segment 351 **/ 352 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) 353 { 354 if (max_size < PAGE_CACHE_SIZE) { 355 max_size = PAGE_CACHE_SIZE; 356 printk(KERN_INFO "%s: set to minimum %d\n", 357 __func__, max_size); 358 } 359 360 q->limits.max_segment_size = max_size; 361 } 362 EXPORT_SYMBOL(blk_queue_max_segment_size); 363 364 /** 365 * blk_queue_logical_block_size - set logical block size for the queue 366 * @q: the request queue for the device 367 * @size: the logical block size, in bytes 368 * 369 * Description: 370 * This should be set to the lowest possible block size that the 371 * storage device can address. The default of 512 covers most 372 * hardware. 373 **/ 374 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) 375 { 376 q->limits.logical_block_size = size; 377 378 if (q->limits.physical_block_size < size) 379 q->limits.physical_block_size = size; 380 381 if (q->limits.io_min < q->limits.physical_block_size) 382 q->limits.io_min = q->limits.physical_block_size; 383 } 384 EXPORT_SYMBOL(blk_queue_logical_block_size); 385 386 /** 387 * blk_queue_physical_block_size - set physical block size for the queue 388 * @q: the request queue for the device 389 * @size: the physical block size, in bytes 390 * 391 * Description: 392 * This should be set to the lowest possible sector size that the 393 * hardware can operate on without reverting to read-modify-write 394 * operations. 395 */ 396 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) 397 { 398 q->limits.physical_block_size = size; 399 400 if (q->limits.physical_block_size < q->limits.logical_block_size) 401 q->limits.physical_block_size = q->limits.logical_block_size; 402 403 if (q->limits.io_min < q->limits.physical_block_size) 404 q->limits.io_min = q->limits.physical_block_size; 405 } 406 EXPORT_SYMBOL(blk_queue_physical_block_size); 407 408 /** 409 * blk_queue_alignment_offset - set physical block alignment offset 410 * @q: the request queue for the device 411 * @offset: alignment offset in bytes 412 * 413 * Description: 414 * Some devices are naturally misaligned to compensate for things like 415 * the legacy DOS partition table 63-sector offset. Low-level drivers 416 * should call this function for devices whose first sector is not 417 * naturally aligned. 418 */ 419 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) 420 { 421 q->limits.alignment_offset = 422 offset & (q->limits.physical_block_size - 1); 423 q->limits.misaligned = 0; 424 } 425 EXPORT_SYMBOL(blk_queue_alignment_offset); 426 427 /** 428 * blk_limits_io_min - set minimum request size for a device 429 * @limits: the queue limits 430 * @min: smallest I/O size in bytes 431 * 432 * Description: 433 * Some devices have an internal block size bigger than the reported 434 * hardware sector size. This function can be used to signal the 435 * smallest I/O the device can perform without incurring a performance 436 * penalty. 437 */ 438 void blk_limits_io_min(struct queue_limits *limits, unsigned int min) 439 { 440 limits->io_min = min; 441 442 if (limits->io_min < limits->logical_block_size) 443 limits->io_min = limits->logical_block_size; 444 445 if (limits->io_min < limits->physical_block_size) 446 limits->io_min = limits->physical_block_size; 447 } 448 EXPORT_SYMBOL(blk_limits_io_min); 449 450 /** 451 * blk_queue_io_min - set minimum request size for the queue 452 * @q: the request queue for the device 453 * @min: smallest I/O size in bytes 454 * 455 * Description: 456 * Storage devices may report a granularity or preferred minimum I/O 457 * size which is the smallest request the device can perform without 458 * incurring a performance penalty. For disk drives this is often the 459 * physical block size. For RAID arrays it is often the stripe chunk 460 * size. A properly aligned multiple of minimum_io_size is the 461 * preferred request size for workloads where a high number of I/O 462 * operations is desired. 463 */ 464 void blk_queue_io_min(struct request_queue *q, unsigned int min) 465 { 466 blk_limits_io_min(&q->limits, min); 467 } 468 EXPORT_SYMBOL(blk_queue_io_min); 469 470 /** 471 * blk_limits_io_opt - set optimal request size for a device 472 * @limits: the queue limits 473 * @opt: smallest I/O size in bytes 474 * 475 * Description: 476 * Storage devices may report an optimal I/O size, which is the 477 * device's preferred unit for sustained I/O. This is rarely reported 478 * for disk drives. For RAID arrays it is usually the stripe width or 479 * the internal track size. A properly aligned multiple of 480 * optimal_io_size is the preferred request size for workloads where 481 * sustained throughput is desired. 482 */ 483 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) 484 { 485 limits->io_opt = opt; 486 } 487 EXPORT_SYMBOL(blk_limits_io_opt); 488 489 /** 490 * blk_queue_io_opt - set optimal request size for the queue 491 * @q: the request queue for the device 492 * @opt: optimal request size in bytes 493 * 494 * Description: 495 * Storage devices may report an optimal I/O size, which is the 496 * device's preferred unit for sustained I/O. This is rarely reported 497 * for disk drives. For RAID arrays it is usually the stripe width or 498 * the internal track size. A properly aligned multiple of 499 * optimal_io_size is the preferred request size for workloads where 500 * sustained throughput is desired. 501 */ 502 void blk_queue_io_opt(struct request_queue *q, unsigned int opt) 503 { 504 blk_limits_io_opt(&q->limits, opt); 505 } 506 EXPORT_SYMBOL(blk_queue_io_opt); 507 508 /** 509 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers 510 * @t: the stacking driver (top) 511 * @b: the underlying device (bottom) 512 **/ 513 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) 514 { 515 blk_stack_limits(&t->limits, &b->limits, 0); 516 } 517 EXPORT_SYMBOL(blk_queue_stack_limits); 518 519 /** 520 * blk_stack_limits - adjust queue_limits for stacked devices 521 * @t: the stacking driver limits (top device) 522 * @b: the underlying queue limits (bottom, component device) 523 * @start: first data sector within component device 524 * 525 * Description: 526 * This function is used by stacking drivers like MD and DM to ensure 527 * that all component devices have compatible block sizes and 528 * alignments. The stacking driver must provide a queue_limits 529 * struct (top) and then iteratively call the stacking function for 530 * all component (bottom) devices. The stacking function will 531 * attempt to combine the values and ensure proper alignment. 532 * 533 * Returns 0 if the top and bottom queue_limits are compatible. The 534 * top device's block sizes and alignment offsets may be adjusted to 535 * ensure alignment with the bottom device. If no compatible sizes 536 * and alignments exist, -1 is returned and the resulting top 537 * queue_limits will have the misaligned flag set to indicate that 538 * the alignment_offset is undefined. 539 */ 540 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, 541 sector_t start) 542 { 543 unsigned int top, bottom, alignment, ret = 0; 544 545 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); 546 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); 547 t->max_write_same_sectors = min(t->max_write_same_sectors, 548 b->max_write_same_sectors); 549 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); 550 551 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, 552 b->seg_boundary_mask); 553 554 t->max_segments = min_not_zero(t->max_segments, b->max_segments); 555 t->max_integrity_segments = min_not_zero(t->max_integrity_segments, 556 b->max_integrity_segments); 557 558 t->max_segment_size = min_not_zero(t->max_segment_size, 559 b->max_segment_size); 560 561 t->misaligned |= b->misaligned; 562 563 alignment = queue_limit_alignment_offset(b, start); 564 565 /* Bottom device has different alignment. Check that it is 566 * compatible with the current top alignment. 567 */ 568 if (t->alignment_offset != alignment) { 569 570 top = max(t->physical_block_size, t->io_min) 571 + t->alignment_offset; 572 bottom = max(b->physical_block_size, b->io_min) + alignment; 573 574 /* Verify that top and bottom intervals line up */ 575 if (max(top, bottom) % min(top, bottom)) { 576 t->misaligned = 1; 577 ret = -1; 578 } 579 } 580 581 t->logical_block_size = max(t->logical_block_size, 582 b->logical_block_size); 583 584 t->physical_block_size = max(t->physical_block_size, 585 b->physical_block_size); 586 587 t->io_min = max(t->io_min, b->io_min); 588 t->io_opt = lcm(t->io_opt, b->io_opt); 589 590 t->cluster &= b->cluster; 591 t->discard_zeroes_data &= b->discard_zeroes_data; 592 593 /* Physical block size a multiple of the logical block size? */ 594 if (t->physical_block_size & (t->logical_block_size - 1)) { 595 t->physical_block_size = t->logical_block_size; 596 t->misaligned = 1; 597 ret = -1; 598 } 599 600 /* Minimum I/O a multiple of the physical block size? */ 601 if (t->io_min & (t->physical_block_size - 1)) { 602 t->io_min = t->physical_block_size; 603 t->misaligned = 1; 604 ret = -1; 605 } 606 607 /* Optimal I/O a multiple of the physical block size? */ 608 if (t->io_opt & (t->physical_block_size - 1)) { 609 t->io_opt = 0; 610 t->misaligned = 1; 611 ret = -1; 612 } 613 614 t->raid_partial_stripes_expensive = 615 max(t->raid_partial_stripes_expensive, 616 b->raid_partial_stripes_expensive); 617 618 /* Find lowest common alignment_offset */ 619 t->alignment_offset = lcm(t->alignment_offset, alignment) 620 % max(t->physical_block_size, t->io_min); 621 622 /* Verify that new alignment_offset is on a logical block boundary */ 623 if (t->alignment_offset & (t->logical_block_size - 1)) { 624 t->misaligned = 1; 625 ret = -1; 626 } 627 628 /* Discard alignment and granularity */ 629 if (b->discard_granularity) { 630 alignment = queue_limit_discard_alignment(b, start); 631 632 if (t->discard_granularity != 0 && 633 t->discard_alignment != alignment) { 634 top = t->discard_granularity + t->discard_alignment; 635 bottom = b->discard_granularity + alignment; 636 637 /* Verify that top and bottom intervals line up */ 638 if ((max(top, bottom) % min(top, bottom)) != 0) 639 t->discard_misaligned = 1; 640 } 641 642 t->max_discard_sectors = min_not_zero(t->max_discard_sectors, 643 b->max_discard_sectors); 644 t->discard_granularity = max(t->discard_granularity, 645 b->discard_granularity); 646 t->discard_alignment = lcm(t->discard_alignment, alignment) % 647 t->discard_granularity; 648 } 649 650 return ret; 651 } 652 EXPORT_SYMBOL(blk_stack_limits); 653 654 /** 655 * bdev_stack_limits - adjust queue limits for stacked drivers 656 * @t: the stacking driver limits (top device) 657 * @bdev: the component block_device (bottom) 658 * @start: first data sector within component device 659 * 660 * Description: 661 * Merges queue limits for a top device and a block_device. Returns 662 * 0 if alignment didn't change. Returns -1 if adding the bottom 663 * device caused misalignment. 664 */ 665 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, 666 sector_t start) 667 { 668 struct request_queue *bq = bdev_get_queue(bdev); 669 670 start += get_start_sect(bdev); 671 672 return blk_stack_limits(t, &bq->limits, start); 673 } 674 EXPORT_SYMBOL(bdev_stack_limits); 675 676 /** 677 * disk_stack_limits - adjust queue limits for stacked drivers 678 * @disk: MD/DM gendisk (top) 679 * @bdev: the underlying block device (bottom) 680 * @offset: offset to beginning of data within component device 681 * 682 * Description: 683 * Merges the limits for a top level gendisk and a bottom level 684 * block_device. 685 */ 686 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, 687 sector_t offset) 688 { 689 struct request_queue *t = disk->queue; 690 691 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { 692 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; 693 694 disk_name(disk, 0, top); 695 bdevname(bdev, bottom); 696 697 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", 698 top, bottom); 699 } 700 } 701 EXPORT_SYMBOL(disk_stack_limits); 702 703 /** 704 * blk_queue_dma_pad - set pad mask 705 * @q: the request queue for the device 706 * @mask: pad mask 707 * 708 * Set dma pad mask. 709 * 710 * Appending pad buffer to a request modifies the last entry of a 711 * scatter list such that it includes the pad buffer. 712 **/ 713 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) 714 { 715 q->dma_pad_mask = mask; 716 } 717 EXPORT_SYMBOL(blk_queue_dma_pad); 718 719 /** 720 * blk_queue_update_dma_pad - update pad mask 721 * @q: the request queue for the device 722 * @mask: pad mask 723 * 724 * Update dma pad mask. 725 * 726 * Appending pad buffer to a request modifies the last entry of a 727 * scatter list such that it includes the pad buffer. 728 **/ 729 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) 730 { 731 if (mask > q->dma_pad_mask) 732 q->dma_pad_mask = mask; 733 } 734 EXPORT_SYMBOL(blk_queue_update_dma_pad); 735 736 /** 737 * blk_queue_dma_drain - Set up a drain buffer for excess dma. 738 * @q: the request queue for the device 739 * @dma_drain_needed: fn which returns non-zero if drain is necessary 740 * @buf: physically contiguous buffer 741 * @size: size of the buffer in bytes 742 * 743 * Some devices have excess DMA problems and can't simply discard (or 744 * zero fill) the unwanted piece of the transfer. They have to have a 745 * real area of memory to transfer it into. The use case for this is 746 * ATAPI devices in DMA mode. If the packet command causes a transfer 747 * bigger than the transfer size some HBAs will lock up if there 748 * aren't DMA elements to contain the excess transfer. What this API 749 * does is adjust the queue so that the buf is always appended 750 * silently to the scatterlist. 751 * 752 * Note: This routine adjusts max_hw_segments to make room for appending 753 * the drain buffer. If you call blk_queue_max_segments() after calling 754 * this routine, you must set the limit to one fewer than your device 755 * can support otherwise there won't be room for the drain buffer. 756 */ 757 int blk_queue_dma_drain(struct request_queue *q, 758 dma_drain_needed_fn *dma_drain_needed, 759 void *buf, unsigned int size) 760 { 761 if (queue_max_segments(q) < 2) 762 return -EINVAL; 763 /* make room for appending the drain */ 764 blk_queue_max_segments(q, queue_max_segments(q) - 1); 765 q->dma_drain_needed = dma_drain_needed; 766 q->dma_drain_buffer = buf; 767 q->dma_drain_size = size; 768 769 return 0; 770 } 771 EXPORT_SYMBOL_GPL(blk_queue_dma_drain); 772 773 /** 774 * blk_queue_segment_boundary - set boundary rules for segment merging 775 * @q: the request queue for the device 776 * @mask: the memory boundary mask 777 **/ 778 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) 779 { 780 if (mask < PAGE_CACHE_SIZE - 1) { 781 mask = PAGE_CACHE_SIZE - 1; 782 printk(KERN_INFO "%s: set to minimum %lx\n", 783 __func__, mask); 784 } 785 786 q->limits.seg_boundary_mask = mask; 787 } 788 EXPORT_SYMBOL(blk_queue_segment_boundary); 789 790 /** 791 * blk_queue_dma_alignment - set dma length and memory alignment 792 * @q: the request queue for the device 793 * @mask: alignment mask 794 * 795 * description: 796 * set required memory and length alignment for direct dma transactions. 797 * this is used when building direct io requests for the queue. 798 * 799 **/ 800 void blk_queue_dma_alignment(struct request_queue *q, int mask) 801 { 802 q->dma_alignment = mask; 803 } 804 EXPORT_SYMBOL(blk_queue_dma_alignment); 805 806 /** 807 * blk_queue_update_dma_alignment - update dma length and memory alignment 808 * @q: the request queue for the device 809 * @mask: alignment mask 810 * 811 * description: 812 * update required memory and length alignment for direct dma transactions. 813 * If the requested alignment is larger than the current alignment, then 814 * the current queue alignment is updated to the new value, otherwise it 815 * is left alone. The design of this is to allow multiple objects 816 * (driver, device, transport etc) to set their respective 817 * alignments without having them interfere. 818 * 819 **/ 820 void blk_queue_update_dma_alignment(struct request_queue *q, int mask) 821 { 822 BUG_ON(mask > PAGE_SIZE); 823 824 if (mask > q->dma_alignment) 825 q->dma_alignment = mask; 826 } 827 EXPORT_SYMBOL(blk_queue_update_dma_alignment); 828 829 /** 830 * blk_queue_flush - configure queue's cache flush capability 831 * @q: the request queue for the device 832 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA 833 * 834 * Tell block layer cache flush capability of @q. If it supports 835 * flushing, REQ_FLUSH should be set. If it supports bypassing 836 * write cache for individual writes, REQ_FUA should be set. 837 */ 838 void blk_queue_flush(struct request_queue *q, unsigned int flush) 839 { 840 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA)); 841 842 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA))) 843 flush &= ~REQ_FUA; 844 845 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA); 846 } 847 EXPORT_SYMBOL_GPL(blk_queue_flush); 848 849 void blk_queue_flush_queueable(struct request_queue *q, bool queueable) 850 { 851 q->flush_not_queueable = !queueable; 852 } 853 EXPORT_SYMBOL_GPL(blk_queue_flush_queueable); 854 855 static int __init blk_settings_init(void) 856 { 857 blk_max_low_pfn = max_low_pfn - 1; 858 blk_max_pfn = max_pfn - 1; 859 return 0; 860 } 861 subsys_initcall(blk_settings_init); 862