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 #include "blk-wbt.h" 17 18 unsigned long blk_max_low_pfn; 19 EXPORT_SYMBOL(blk_max_low_pfn); 20 21 unsigned long blk_max_pfn; 22 23 /** 24 * blk_queue_prep_rq - set a prepare_request function for queue 25 * @q: queue 26 * @pfn: prepare_request function 27 * 28 * It's possible for a queue to register a prepare_request callback which 29 * is invoked before the request is handed to the request_fn. The goal of 30 * the function is to prepare a request for I/O, it can be used to build a 31 * cdb from the request data for instance. 32 * 33 */ 34 void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) 35 { 36 q->prep_rq_fn = pfn; 37 } 38 EXPORT_SYMBOL(blk_queue_prep_rq); 39 40 /** 41 * blk_queue_unprep_rq - set an unprepare_request function for queue 42 * @q: queue 43 * @ufn: unprepare_request function 44 * 45 * It's possible for a queue to register an unprepare_request callback 46 * which is invoked before the request is finally completed. The goal 47 * of the function is to deallocate any data that was allocated in the 48 * prepare_request callback. 49 * 50 */ 51 void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn) 52 { 53 q->unprep_rq_fn = ufn; 54 } 55 EXPORT_SYMBOL(blk_queue_unprep_rq); 56 57 void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) 58 { 59 q->softirq_done_fn = fn; 60 } 61 EXPORT_SYMBOL(blk_queue_softirq_done); 62 63 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) 64 { 65 q->rq_timeout = timeout; 66 } 67 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); 68 69 void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) 70 { 71 q->rq_timed_out_fn = fn; 72 } 73 EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); 74 75 void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) 76 { 77 q->lld_busy_fn = fn; 78 } 79 EXPORT_SYMBOL_GPL(blk_queue_lld_busy); 80 81 /** 82 * blk_set_default_limits - reset limits to default values 83 * @lim: the queue_limits structure to reset 84 * 85 * Description: 86 * Returns a queue_limit struct to its default state. 87 */ 88 void blk_set_default_limits(struct queue_limits *lim) 89 { 90 lim->max_segments = BLK_MAX_SEGMENTS; 91 lim->max_integrity_segments = 0; 92 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; 93 lim->virt_boundary_mask = 0; 94 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; 95 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; 96 lim->max_dev_sectors = 0; 97 lim->chunk_sectors = 0; 98 lim->max_write_same_sectors = 0; 99 lim->max_write_zeroes_sectors = 0; 100 lim->max_discard_sectors = 0; 101 lim->max_hw_discard_sectors = 0; 102 lim->discard_granularity = 0; 103 lim->discard_alignment = 0; 104 lim->discard_misaligned = 0; 105 lim->discard_zeroes_data = 0; 106 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; 107 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); 108 lim->alignment_offset = 0; 109 lim->io_opt = 0; 110 lim->misaligned = 0; 111 lim->cluster = 1; 112 lim->zoned = BLK_ZONED_NONE; 113 } 114 EXPORT_SYMBOL(blk_set_default_limits); 115 116 /** 117 * blk_set_stacking_limits - set default limits for stacking devices 118 * @lim: the queue_limits structure to reset 119 * 120 * Description: 121 * Returns a queue_limit struct to its default state. Should be used 122 * by stacking drivers like DM that have no internal limits. 123 */ 124 void blk_set_stacking_limits(struct queue_limits *lim) 125 { 126 blk_set_default_limits(lim); 127 128 /* Inherit limits from component devices */ 129 lim->discard_zeroes_data = 1; 130 lim->max_segments = USHRT_MAX; 131 lim->max_hw_sectors = UINT_MAX; 132 lim->max_segment_size = UINT_MAX; 133 lim->max_sectors = UINT_MAX; 134 lim->max_dev_sectors = UINT_MAX; 135 lim->max_write_same_sectors = UINT_MAX; 136 lim->max_write_zeroes_sectors = UINT_MAX; 137 } 138 EXPORT_SYMBOL(blk_set_stacking_limits); 139 140 /** 141 * blk_queue_make_request - define an alternate make_request function for a device 142 * @q: the request queue for the device to be affected 143 * @mfn: the alternate make_request function 144 * 145 * Description: 146 * The normal way for &struct bios to be passed to a device 147 * driver is for them to be collected into requests on a request 148 * queue, and then to allow the device driver to select requests 149 * off that queue when it is ready. This works well for many block 150 * devices. However some block devices (typically virtual devices 151 * such as md or lvm) do not benefit from the processing on the 152 * request queue, and are served best by having the requests passed 153 * directly to them. This can be achieved by providing a function 154 * to blk_queue_make_request(). 155 * 156 * Caveat: 157 * The driver that does this *must* be able to deal appropriately 158 * with buffers in "highmemory". This can be accomplished by either calling 159 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling 160 * blk_queue_bounce() to create a buffer in normal memory. 161 **/ 162 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) 163 { 164 /* 165 * set defaults 166 */ 167 q->nr_requests = BLKDEV_MAX_RQ; 168 169 q->make_request_fn = mfn; 170 blk_queue_dma_alignment(q, 511); 171 blk_queue_congestion_threshold(q); 172 q->nr_batching = BLK_BATCH_REQ; 173 174 blk_set_default_limits(&q->limits); 175 176 /* 177 * by default assume old behaviour and bounce for any highmem page 178 */ 179 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); 180 } 181 EXPORT_SYMBOL(blk_queue_make_request); 182 183 /** 184 * blk_queue_bounce_limit - set bounce buffer limit for queue 185 * @q: the request queue for the device 186 * @max_addr: the maximum address the device can handle 187 * 188 * Description: 189 * Different hardware can have different requirements as to what pages 190 * it can do I/O directly to. A low level driver can call 191 * blk_queue_bounce_limit to have lower memory pages allocated as bounce 192 * buffers for doing I/O to pages residing above @max_addr. 193 **/ 194 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr) 195 { 196 unsigned long b_pfn = max_addr >> PAGE_SHIFT; 197 int dma = 0; 198 199 q->bounce_gfp = GFP_NOIO; 200 #if BITS_PER_LONG == 64 201 /* 202 * Assume anything <= 4GB can be handled by IOMMU. Actually 203 * some IOMMUs can handle everything, but I don't know of a 204 * way to test this here. 205 */ 206 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) 207 dma = 1; 208 q->limits.bounce_pfn = max(max_low_pfn, b_pfn); 209 #else 210 if (b_pfn < blk_max_low_pfn) 211 dma = 1; 212 q->limits.bounce_pfn = b_pfn; 213 #endif 214 if (dma) { 215 init_emergency_isa_pool(); 216 q->bounce_gfp = GFP_NOIO | GFP_DMA; 217 q->limits.bounce_pfn = b_pfn; 218 } 219 } 220 EXPORT_SYMBOL(blk_queue_bounce_limit); 221 222 /** 223 * blk_queue_max_hw_sectors - set max sectors for a request for this queue 224 * @q: the request queue for the device 225 * @max_hw_sectors: max hardware sectors in the usual 512b unit 226 * 227 * Description: 228 * Enables a low level driver to set a hard upper limit, 229 * max_hw_sectors, on the size of requests. max_hw_sectors is set by 230 * the device driver based upon the capabilities of the I/O 231 * controller. 232 * 233 * max_dev_sectors is a hard limit imposed by the storage device for 234 * READ/WRITE requests. It is set by the disk driver. 235 * 236 * max_sectors is a soft limit imposed by the block layer for 237 * filesystem type requests. This value can be overridden on a 238 * per-device basis in /sys/block/<device>/queue/max_sectors_kb. 239 * The soft limit can not exceed max_hw_sectors. 240 **/ 241 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) 242 { 243 struct queue_limits *limits = &q->limits; 244 unsigned int max_sectors; 245 246 if ((max_hw_sectors << 9) < PAGE_SIZE) { 247 max_hw_sectors = 1 << (PAGE_SHIFT - 9); 248 printk(KERN_INFO "%s: set to minimum %d\n", 249 __func__, max_hw_sectors); 250 } 251 252 limits->max_hw_sectors = max_hw_sectors; 253 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors); 254 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS); 255 limits->max_sectors = max_sectors; 256 } 257 EXPORT_SYMBOL(blk_queue_max_hw_sectors); 258 259 /** 260 * blk_queue_chunk_sectors - set size of the chunk for this queue 261 * @q: the request queue for the device 262 * @chunk_sectors: chunk sectors in the usual 512b unit 263 * 264 * Description: 265 * If a driver doesn't want IOs to cross a given chunk size, it can set 266 * this limit and prevent merging across chunks. Note that the chunk size 267 * must currently be a power-of-2 in sectors. Also note that the block 268 * layer must accept a page worth of data at any offset. So if the 269 * crossing of chunks is a hard limitation in the driver, it must still be 270 * prepared to split single page bios. 271 **/ 272 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) 273 { 274 BUG_ON(!is_power_of_2(chunk_sectors)); 275 q->limits.chunk_sectors = chunk_sectors; 276 } 277 EXPORT_SYMBOL(blk_queue_chunk_sectors); 278 279 /** 280 * blk_queue_max_discard_sectors - set max sectors for a single discard 281 * @q: the request queue for the device 282 * @max_discard_sectors: maximum number of sectors to discard 283 **/ 284 void blk_queue_max_discard_sectors(struct request_queue *q, 285 unsigned int max_discard_sectors) 286 { 287 q->limits.max_hw_discard_sectors = max_discard_sectors; 288 q->limits.max_discard_sectors = max_discard_sectors; 289 } 290 EXPORT_SYMBOL(blk_queue_max_discard_sectors); 291 292 /** 293 * blk_queue_max_write_same_sectors - set max sectors for a single write same 294 * @q: the request queue for the device 295 * @max_write_same_sectors: maximum number of sectors to write per command 296 **/ 297 void blk_queue_max_write_same_sectors(struct request_queue *q, 298 unsigned int max_write_same_sectors) 299 { 300 q->limits.max_write_same_sectors = max_write_same_sectors; 301 } 302 EXPORT_SYMBOL(blk_queue_max_write_same_sectors); 303 304 /** 305 * blk_queue_max_write_zeroes_sectors - set max sectors for a single 306 * write zeroes 307 * @q: the request queue for the device 308 * @max_write_zeroes_sectors: maximum number of sectors to write per command 309 **/ 310 void blk_queue_max_write_zeroes_sectors(struct request_queue *q, 311 unsigned int max_write_zeroes_sectors) 312 { 313 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors; 314 } 315 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors); 316 317 /** 318 * blk_queue_max_segments - set max hw segments for a request for this queue 319 * @q: the request queue for the device 320 * @max_segments: max number of segments 321 * 322 * Description: 323 * Enables a low level driver to set an upper limit on the number of 324 * hw data segments in a request. 325 **/ 326 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) 327 { 328 if (!max_segments) { 329 max_segments = 1; 330 printk(KERN_INFO "%s: set to minimum %d\n", 331 __func__, max_segments); 332 } 333 334 q->limits.max_segments = max_segments; 335 } 336 EXPORT_SYMBOL(blk_queue_max_segments); 337 338 /** 339 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg 340 * @q: the request queue for the device 341 * @max_size: max size of segment in bytes 342 * 343 * Description: 344 * Enables a low level driver to set an upper limit on the size of a 345 * coalesced segment 346 **/ 347 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) 348 { 349 if (max_size < PAGE_SIZE) { 350 max_size = PAGE_SIZE; 351 printk(KERN_INFO "%s: set to minimum %d\n", 352 __func__, max_size); 353 } 354 355 q->limits.max_segment_size = max_size; 356 } 357 EXPORT_SYMBOL(blk_queue_max_segment_size); 358 359 /** 360 * blk_queue_logical_block_size - set logical block size for the queue 361 * @q: the request queue for the device 362 * @size: the logical block size, in bytes 363 * 364 * Description: 365 * This should be set to the lowest possible block size that the 366 * storage device can address. The default of 512 covers most 367 * hardware. 368 **/ 369 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) 370 { 371 q->limits.logical_block_size = size; 372 373 if (q->limits.physical_block_size < size) 374 q->limits.physical_block_size = size; 375 376 if (q->limits.io_min < q->limits.physical_block_size) 377 q->limits.io_min = q->limits.physical_block_size; 378 } 379 EXPORT_SYMBOL(blk_queue_logical_block_size); 380 381 /** 382 * blk_queue_physical_block_size - set physical block size for the queue 383 * @q: the request queue for the device 384 * @size: the physical block size, in bytes 385 * 386 * Description: 387 * This should be set to the lowest possible sector size that the 388 * hardware can operate on without reverting to read-modify-write 389 * operations. 390 */ 391 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) 392 { 393 q->limits.physical_block_size = size; 394 395 if (q->limits.physical_block_size < q->limits.logical_block_size) 396 q->limits.physical_block_size = q->limits.logical_block_size; 397 398 if (q->limits.io_min < q->limits.physical_block_size) 399 q->limits.io_min = q->limits.physical_block_size; 400 } 401 EXPORT_SYMBOL(blk_queue_physical_block_size); 402 403 /** 404 * blk_queue_alignment_offset - set physical block alignment offset 405 * @q: the request queue for the device 406 * @offset: alignment offset in bytes 407 * 408 * Description: 409 * Some devices are naturally misaligned to compensate for things like 410 * the legacy DOS partition table 63-sector offset. Low-level drivers 411 * should call this function for devices whose first sector is not 412 * naturally aligned. 413 */ 414 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) 415 { 416 q->limits.alignment_offset = 417 offset & (q->limits.physical_block_size - 1); 418 q->limits.misaligned = 0; 419 } 420 EXPORT_SYMBOL(blk_queue_alignment_offset); 421 422 /** 423 * blk_limits_io_min - set minimum request size for a device 424 * @limits: the queue limits 425 * @min: smallest I/O size in bytes 426 * 427 * Description: 428 * Some devices have an internal block size bigger than the reported 429 * hardware sector size. This function can be used to signal the 430 * smallest I/O the device can perform without incurring a performance 431 * penalty. 432 */ 433 void blk_limits_io_min(struct queue_limits *limits, unsigned int min) 434 { 435 limits->io_min = min; 436 437 if (limits->io_min < limits->logical_block_size) 438 limits->io_min = limits->logical_block_size; 439 440 if (limits->io_min < limits->physical_block_size) 441 limits->io_min = limits->physical_block_size; 442 } 443 EXPORT_SYMBOL(blk_limits_io_min); 444 445 /** 446 * blk_queue_io_min - set minimum request size for the queue 447 * @q: the request queue for the device 448 * @min: smallest I/O size in bytes 449 * 450 * Description: 451 * Storage devices may report a granularity or preferred minimum I/O 452 * size which is the smallest request the device can perform without 453 * incurring a performance penalty. For disk drives this is often the 454 * physical block size. For RAID arrays it is often the stripe chunk 455 * size. A properly aligned multiple of minimum_io_size is the 456 * preferred request size for workloads where a high number of I/O 457 * operations is desired. 458 */ 459 void blk_queue_io_min(struct request_queue *q, unsigned int min) 460 { 461 blk_limits_io_min(&q->limits, min); 462 } 463 EXPORT_SYMBOL(blk_queue_io_min); 464 465 /** 466 * blk_limits_io_opt - set optimal request size for a device 467 * @limits: the queue limits 468 * @opt: smallest I/O size in bytes 469 * 470 * Description: 471 * Storage devices may report an optimal I/O size, which is the 472 * device's preferred unit for sustained I/O. This is rarely reported 473 * for disk drives. For RAID arrays it is usually the stripe width or 474 * the internal track size. A properly aligned multiple of 475 * optimal_io_size is the preferred request size for workloads where 476 * sustained throughput is desired. 477 */ 478 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) 479 { 480 limits->io_opt = opt; 481 } 482 EXPORT_SYMBOL(blk_limits_io_opt); 483 484 /** 485 * blk_queue_io_opt - set optimal request size for the queue 486 * @q: the request queue for the device 487 * @opt: optimal request size in bytes 488 * 489 * Description: 490 * Storage devices may report an optimal I/O size, which is the 491 * device's preferred unit for sustained I/O. This is rarely reported 492 * for disk drives. For RAID arrays it is usually the stripe width or 493 * the internal track size. A properly aligned multiple of 494 * optimal_io_size is the preferred request size for workloads where 495 * sustained throughput is desired. 496 */ 497 void blk_queue_io_opt(struct request_queue *q, unsigned int opt) 498 { 499 blk_limits_io_opt(&q->limits, opt); 500 } 501 EXPORT_SYMBOL(blk_queue_io_opt); 502 503 /** 504 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers 505 * @t: the stacking driver (top) 506 * @b: the underlying device (bottom) 507 **/ 508 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) 509 { 510 blk_stack_limits(&t->limits, &b->limits, 0); 511 } 512 EXPORT_SYMBOL(blk_queue_stack_limits); 513 514 /** 515 * blk_stack_limits - adjust queue_limits for stacked devices 516 * @t: the stacking driver limits (top device) 517 * @b: the underlying queue limits (bottom, component device) 518 * @start: first data sector within component device 519 * 520 * Description: 521 * This function is used by stacking drivers like MD and DM to ensure 522 * that all component devices have compatible block sizes and 523 * alignments. The stacking driver must provide a queue_limits 524 * struct (top) and then iteratively call the stacking function for 525 * all component (bottom) devices. The stacking function will 526 * attempt to combine the values and ensure proper alignment. 527 * 528 * Returns 0 if the top and bottom queue_limits are compatible. The 529 * top device's block sizes and alignment offsets may be adjusted to 530 * ensure alignment with the bottom device. If no compatible sizes 531 * and alignments exist, -1 is returned and the resulting top 532 * queue_limits will have the misaligned flag set to indicate that 533 * the alignment_offset is undefined. 534 */ 535 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, 536 sector_t start) 537 { 538 unsigned int top, bottom, alignment, ret = 0; 539 540 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); 541 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); 542 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); 543 t->max_write_same_sectors = min(t->max_write_same_sectors, 544 b->max_write_same_sectors); 545 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, 546 b->max_write_zeroes_sectors); 547 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); 548 549 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, 550 b->seg_boundary_mask); 551 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, 552 b->virt_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_not_zero(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_not_zero(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->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, 645 b->max_hw_discard_sectors); 646 t->discard_granularity = max(t->discard_granularity, 647 b->discard_granularity); 648 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) % 649 t->discard_granularity; 650 } 651 652 if (b->chunk_sectors) 653 t->chunk_sectors = min_not_zero(t->chunk_sectors, 654 b->chunk_sectors); 655 656 return ret; 657 } 658 EXPORT_SYMBOL(blk_stack_limits); 659 660 /** 661 * bdev_stack_limits - adjust queue limits for stacked drivers 662 * @t: the stacking driver limits (top device) 663 * @bdev: the component block_device (bottom) 664 * @start: first data sector within component device 665 * 666 * Description: 667 * Merges queue limits for a top device and a block_device. Returns 668 * 0 if alignment didn't change. Returns -1 if adding the bottom 669 * device caused misalignment. 670 */ 671 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, 672 sector_t start) 673 { 674 struct request_queue *bq = bdev_get_queue(bdev); 675 676 start += get_start_sect(bdev); 677 678 return blk_stack_limits(t, &bq->limits, start); 679 } 680 EXPORT_SYMBOL(bdev_stack_limits); 681 682 /** 683 * disk_stack_limits - adjust queue limits for stacked drivers 684 * @disk: MD/DM gendisk (top) 685 * @bdev: the underlying block device (bottom) 686 * @offset: offset to beginning of data within component device 687 * 688 * Description: 689 * Merges the limits for a top level gendisk and a bottom level 690 * block_device. 691 */ 692 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, 693 sector_t offset) 694 { 695 struct request_queue *t = disk->queue; 696 697 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { 698 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; 699 700 disk_name(disk, 0, top); 701 bdevname(bdev, bottom); 702 703 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", 704 top, bottom); 705 } 706 } 707 EXPORT_SYMBOL(disk_stack_limits); 708 709 /** 710 * blk_queue_dma_pad - set pad mask 711 * @q: the request queue for the device 712 * @mask: pad mask 713 * 714 * Set dma pad mask. 715 * 716 * Appending pad buffer to a request modifies the last entry of a 717 * scatter list such that it includes the pad buffer. 718 **/ 719 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) 720 { 721 q->dma_pad_mask = mask; 722 } 723 EXPORT_SYMBOL(blk_queue_dma_pad); 724 725 /** 726 * blk_queue_update_dma_pad - update pad mask 727 * @q: the request queue for the device 728 * @mask: pad mask 729 * 730 * Update dma pad mask. 731 * 732 * Appending pad buffer to a request modifies the last entry of a 733 * scatter list such that it includes the pad buffer. 734 **/ 735 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) 736 { 737 if (mask > q->dma_pad_mask) 738 q->dma_pad_mask = mask; 739 } 740 EXPORT_SYMBOL(blk_queue_update_dma_pad); 741 742 /** 743 * blk_queue_dma_drain - Set up a drain buffer for excess dma. 744 * @q: the request queue for the device 745 * @dma_drain_needed: fn which returns non-zero if drain is necessary 746 * @buf: physically contiguous buffer 747 * @size: size of the buffer in bytes 748 * 749 * Some devices have excess DMA problems and can't simply discard (or 750 * zero fill) the unwanted piece of the transfer. They have to have a 751 * real area of memory to transfer it into. The use case for this is 752 * ATAPI devices in DMA mode. If the packet command causes a transfer 753 * bigger than the transfer size some HBAs will lock up if there 754 * aren't DMA elements to contain the excess transfer. What this API 755 * does is adjust the queue so that the buf is always appended 756 * silently to the scatterlist. 757 * 758 * Note: This routine adjusts max_hw_segments to make room for appending 759 * the drain buffer. If you call blk_queue_max_segments() after calling 760 * this routine, you must set the limit to one fewer than your device 761 * can support otherwise there won't be room for the drain buffer. 762 */ 763 int blk_queue_dma_drain(struct request_queue *q, 764 dma_drain_needed_fn *dma_drain_needed, 765 void *buf, unsigned int size) 766 { 767 if (queue_max_segments(q) < 2) 768 return -EINVAL; 769 /* make room for appending the drain */ 770 blk_queue_max_segments(q, queue_max_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_SIZE - 1) { 787 mask = PAGE_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_virt_boundary - set boundary rules for bio merging 798 * @q: the request queue for the device 799 * @mask: the memory boundary mask 800 **/ 801 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask) 802 { 803 q->limits.virt_boundary_mask = mask; 804 } 805 EXPORT_SYMBOL(blk_queue_virt_boundary); 806 807 /** 808 * blk_queue_dma_alignment - set dma length and memory alignment 809 * @q: the request queue for the device 810 * @mask: alignment mask 811 * 812 * description: 813 * set required memory and length alignment for direct dma transactions. 814 * this is used when building direct io requests for the queue. 815 * 816 **/ 817 void blk_queue_dma_alignment(struct request_queue *q, int mask) 818 { 819 q->dma_alignment = mask; 820 } 821 EXPORT_SYMBOL(blk_queue_dma_alignment); 822 823 /** 824 * blk_queue_update_dma_alignment - update dma length and memory alignment 825 * @q: the request queue for the device 826 * @mask: alignment mask 827 * 828 * description: 829 * update required memory and length alignment for direct dma transactions. 830 * If the requested alignment is larger than the current alignment, then 831 * the current queue alignment is updated to the new value, otherwise it 832 * is left alone. The design of this is to allow multiple objects 833 * (driver, device, transport etc) to set their respective 834 * alignments without having them interfere. 835 * 836 **/ 837 void blk_queue_update_dma_alignment(struct request_queue *q, int mask) 838 { 839 BUG_ON(mask > PAGE_SIZE); 840 841 if (mask > q->dma_alignment) 842 q->dma_alignment = mask; 843 } 844 EXPORT_SYMBOL(blk_queue_update_dma_alignment); 845 846 void blk_queue_flush_queueable(struct request_queue *q, bool queueable) 847 { 848 spin_lock_irq(q->queue_lock); 849 if (queueable) 850 clear_bit(QUEUE_FLAG_FLUSH_NQ, &q->queue_flags); 851 else 852 set_bit(QUEUE_FLAG_FLUSH_NQ, &q->queue_flags); 853 spin_unlock_irq(q->queue_lock); 854 } 855 EXPORT_SYMBOL_GPL(blk_queue_flush_queueable); 856 857 /** 858 * blk_set_queue_depth - tell the block layer about the device queue depth 859 * @q: the request queue for the device 860 * @depth: queue depth 861 * 862 */ 863 void blk_set_queue_depth(struct request_queue *q, unsigned int depth) 864 { 865 q->queue_depth = depth; 866 wbt_set_queue_depth(q->rq_wb, depth); 867 } 868 EXPORT_SYMBOL(blk_set_queue_depth); 869 870 /** 871 * blk_queue_write_cache - configure queue's write cache 872 * @q: the request queue for the device 873 * @wc: write back cache on or off 874 * @fua: device supports FUA writes, if true 875 * 876 * Tell the block layer about the write cache of @q. 877 */ 878 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua) 879 { 880 spin_lock_irq(q->queue_lock); 881 if (wc) 882 queue_flag_set(QUEUE_FLAG_WC, q); 883 else 884 queue_flag_clear(QUEUE_FLAG_WC, q); 885 if (fua) 886 queue_flag_set(QUEUE_FLAG_FUA, q); 887 else 888 queue_flag_clear(QUEUE_FLAG_FUA, q); 889 spin_unlock_irq(q->queue_lock); 890 891 wbt_set_write_cache(q->rq_wb, test_bit(QUEUE_FLAG_WC, &q->queue_flags)); 892 } 893 EXPORT_SYMBOL_GPL(blk_queue_write_cache); 894 895 static int __init blk_settings_init(void) 896 { 897 blk_max_low_pfn = max_low_pfn - 1; 898 blk_max_pfn = max_pfn - 1; 899 return 0; 900 } 901 subsys_initcall(blk_settings_init); 902