1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Functions related to setting various queue properties from drivers 4 */ 5 #include <linux/kernel.h> 6 #include <linux/module.h> 7 #include <linux/init.h> 8 #include <linux/bio.h> 9 #include <linux/blkdev.h> 10 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */ 11 #include <linux/gcd.h> 12 #include <linux/lcm.h> 13 #include <linux/jiffies.h> 14 #include <linux/gfp.h> 15 #include <linux/dma-mapping.h> 16 17 #include "blk.h" 18 #include "blk-wbt.h" 19 20 unsigned long blk_max_low_pfn; 21 EXPORT_SYMBOL(blk_max_low_pfn); 22 23 unsigned long blk_max_pfn; 24 25 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) 26 { 27 q->rq_timeout = timeout; 28 } 29 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); 30 31 /** 32 * blk_set_default_limits - reset limits to default values 33 * @lim: the queue_limits structure to reset 34 * 35 * Description: 36 * Returns a queue_limit struct to its default state. 37 */ 38 void blk_set_default_limits(struct queue_limits *lim) 39 { 40 lim->max_segments = BLK_MAX_SEGMENTS; 41 lim->max_discard_segments = 1; 42 lim->max_integrity_segments = 0; 43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; 44 lim->virt_boundary_mask = 0; 45 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; 46 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; 47 lim->max_dev_sectors = 0; 48 lim->chunk_sectors = 0; 49 lim->max_write_same_sectors = 0; 50 lim->max_write_zeroes_sectors = 0; 51 lim->max_zone_append_sectors = 0; 52 lim->max_discard_sectors = 0; 53 lim->max_hw_discard_sectors = 0; 54 lim->discard_granularity = 0; 55 lim->discard_alignment = 0; 56 lim->discard_misaligned = 0; 57 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; 58 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); 59 lim->alignment_offset = 0; 60 lim->io_opt = 0; 61 lim->misaligned = 0; 62 lim->zoned = BLK_ZONED_NONE; 63 lim->zone_write_granularity = 0; 64 } 65 EXPORT_SYMBOL(blk_set_default_limits); 66 67 /** 68 * blk_set_stacking_limits - set default limits for stacking devices 69 * @lim: the queue_limits structure to reset 70 * 71 * Description: 72 * Returns a queue_limit struct to its default state. Should be used 73 * by stacking drivers like DM that have no internal limits. 74 */ 75 void blk_set_stacking_limits(struct queue_limits *lim) 76 { 77 blk_set_default_limits(lim); 78 79 /* Inherit limits from component devices */ 80 lim->max_segments = USHRT_MAX; 81 lim->max_discard_segments = USHRT_MAX; 82 lim->max_hw_sectors = UINT_MAX; 83 lim->max_segment_size = UINT_MAX; 84 lim->max_sectors = UINT_MAX; 85 lim->max_dev_sectors = UINT_MAX; 86 lim->max_write_same_sectors = UINT_MAX; 87 lim->max_write_zeroes_sectors = UINT_MAX; 88 lim->max_zone_append_sectors = UINT_MAX; 89 } 90 EXPORT_SYMBOL(blk_set_stacking_limits); 91 92 /** 93 * blk_queue_bounce_limit - set bounce buffer limit for queue 94 * @q: the request queue for the device 95 * @max_addr: the maximum address the device can handle 96 * 97 * Description: 98 * Different hardware can have different requirements as to what pages 99 * it can do I/O directly to. A low level driver can call 100 * blk_queue_bounce_limit to have lower memory pages allocated as bounce 101 * buffers for doing I/O to pages residing above @max_addr. 102 **/ 103 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr) 104 { 105 unsigned long b_pfn = max_addr >> PAGE_SHIFT; 106 int dma = 0; 107 108 q->bounce_gfp = GFP_NOIO; 109 #if BITS_PER_LONG == 64 110 /* 111 * Assume anything <= 4GB can be handled by IOMMU. Actually 112 * some IOMMUs can handle everything, but I don't know of a 113 * way to test this here. 114 */ 115 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) 116 dma = 1; 117 q->limits.bounce_pfn = max(max_low_pfn, b_pfn); 118 #else 119 if (b_pfn < blk_max_low_pfn) 120 dma = 1; 121 q->limits.bounce_pfn = b_pfn; 122 #endif 123 if (dma) { 124 init_emergency_isa_pool(); 125 q->bounce_gfp = GFP_NOIO | GFP_DMA; 126 q->limits.bounce_pfn = b_pfn; 127 } 128 } 129 EXPORT_SYMBOL(blk_queue_bounce_limit); 130 131 /** 132 * blk_queue_max_hw_sectors - set max sectors for a request for this queue 133 * @q: the request queue for the device 134 * @max_hw_sectors: max hardware sectors in the usual 512b unit 135 * 136 * Description: 137 * Enables a low level driver to set a hard upper limit, 138 * max_hw_sectors, on the size of requests. max_hw_sectors is set by 139 * the device driver based upon the capabilities of the I/O 140 * controller. 141 * 142 * max_dev_sectors is a hard limit imposed by the storage device for 143 * READ/WRITE requests. It is set by the disk driver. 144 * 145 * max_sectors is a soft limit imposed by the block layer for 146 * filesystem type requests. This value can be overridden on a 147 * per-device basis in /sys/block/<device>/queue/max_sectors_kb. 148 * The soft limit can not exceed max_hw_sectors. 149 **/ 150 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) 151 { 152 struct queue_limits *limits = &q->limits; 153 unsigned int max_sectors; 154 155 if ((max_hw_sectors << 9) < PAGE_SIZE) { 156 max_hw_sectors = 1 << (PAGE_SHIFT - 9); 157 printk(KERN_INFO "%s: set to minimum %d\n", 158 __func__, max_hw_sectors); 159 } 160 161 max_hw_sectors = round_down(max_hw_sectors, 162 limits->logical_block_size >> SECTOR_SHIFT); 163 limits->max_hw_sectors = max_hw_sectors; 164 165 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors); 166 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS); 167 max_sectors = round_down(max_sectors, 168 limits->logical_block_size >> SECTOR_SHIFT); 169 limits->max_sectors = max_sectors; 170 171 q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9); 172 } 173 EXPORT_SYMBOL(blk_queue_max_hw_sectors); 174 175 /** 176 * blk_queue_chunk_sectors - set size of the chunk for this queue 177 * @q: the request queue for the device 178 * @chunk_sectors: chunk sectors in the usual 512b unit 179 * 180 * Description: 181 * If a driver doesn't want IOs to cross a given chunk size, it can set 182 * this limit and prevent merging across chunks. Note that the block layer 183 * must accept a page worth of data at any offset. So if the crossing of 184 * chunks is a hard limitation in the driver, it must still be prepared 185 * to split single page bios. 186 **/ 187 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) 188 { 189 q->limits.chunk_sectors = chunk_sectors; 190 } 191 EXPORT_SYMBOL(blk_queue_chunk_sectors); 192 193 /** 194 * blk_queue_max_discard_sectors - set max sectors for a single discard 195 * @q: the request queue for the device 196 * @max_discard_sectors: maximum number of sectors to discard 197 **/ 198 void blk_queue_max_discard_sectors(struct request_queue *q, 199 unsigned int max_discard_sectors) 200 { 201 q->limits.max_hw_discard_sectors = max_discard_sectors; 202 q->limits.max_discard_sectors = max_discard_sectors; 203 } 204 EXPORT_SYMBOL(blk_queue_max_discard_sectors); 205 206 /** 207 * blk_queue_max_write_same_sectors - set max sectors for a single write same 208 * @q: the request queue for the device 209 * @max_write_same_sectors: maximum number of sectors to write per command 210 **/ 211 void blk_queue_max_write_same_sectors(struct request_queue *q, 212 unsigned int max_write_same_sectors) 213 { 214 q->limits.max_write_same_sectors = max_write_same_sectors; 215 } 216 EXPORT_SYMBOL(blk_queue_max_write_same_sectors); 217 218 /** 219 * blk_queue_max_write_zeroes_sectors - set max sectors for a single 220 * write zeroes 221 * @q: the request queue for the device 222 * @max_write_zeroes_sectors: maximum number of sectors to write per command 223 **/ 224 void blk_queue_max_write_zeroes_sectors(struct request_queue *q, 225 unsigned int max_write_zeroes_sectors) 226 { 227 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors; 228 } 229 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors); 230 231 /** 232 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append 233 * @q: the request queue for the device 234 * @max_zone_append_sectors: maximum number of sectors to write per command 235 **/ 236 void blk_queue_max_zone_append_sectors(struct request_queue *q, 237 unsigned int max_zone_append_sectors) 238 { 239 unsigned int max_sectors; 240 241 if (WARN_ON(!blk_queue_is_zoned(q))) 242 return; 243 244 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors); 245 max_sectors = min(q->limits.chunk_sectors, max_sectors); 246 247 /* 248 * Signal eventual driver bugs resulting in the max_zone_append sectors limit 249 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set, 250 * or the max_hw_sectors limit not set. 251 */ 252 WARN_ON(!max_sectors); 253 254 q->limits.max_zone_append_sectors = max_sectors; 255 } 256 EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors); 257 258 /** 259 * blk_queue_max_segments - set max hw segments for a request for this queue 260 * @q: the request queue for the device 261 * @max_segments: max number of segments 262 * 263 * Description: 264 * Enables a low level driver to set an upper limit on the number of 265 * hw data segments in a request. 266 **/ 267 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) 268 { 269 if (!max_segments) { 270 max_segments = 1; 271 printk(KERN_INFO "%s: set to minimum %d\n", 272 __func__, max_segments); 273 } 274 275 q->limits.max_segments = max_segments; 276 } 277 EXPORT_SYMBOL(blk_queue_max_segments); 278 279 /** 280 * blk_queue_max_discard_segments - set max segments for discard requests 281 * @q: the request queue for the device 282 * @max_segments: max number of segments 283 * 284 * Description: 285 * Enables a low level driver to set an upper limit on the number of 286 * segments in a discard request. 287 **/ 288 void blk_queue_max_discard_segments(struct request_queue *q, 289 unsigned short max_segments) 290 { 291 q->limits.max_discard_segments = max_segments; 292 } 293 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments); 294 295 /** 296 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg 297 * @q: the request queue for the device 298 * @max_size: max size of segment in bytes 299 * 300 * Description: 301 * Enables a low level driver to set an upper limit on the size of a 302 * coalesced segment 303 **/ 304 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) 305 { 306 if (max_size < PAGE_SIZE) { 307 max_size = PAGE_SIZE; 308 printk(KERN_INFO "%s: set to minimum %d\n", 309 __func__, max_size); 310 } 311 312 /* see blk_queue_virt_boundary() for the explanation */ 313 WARN_ON_ONCE(q->limits.virt_boundary_mask); 314 315 q->limits.max_segment_size = max_size; 316 } 317 EXPORT_SYMBOL(blk_queue_max_segment_size); 318 319 /** 320 * blk_queue_logical_block_size - set logical block size for the queue 321 * @q: the request queue for the device 322 * @size: the logical block size, in bytes 323 * 324 * Description: 325 * This should be set to the lowest possible block size that the 326 * storage device can address. The default of 512 covers most 327 * hardware. 328 **/ 329 void blk_queue_logical_block_size(struct request_queue *q, unsigned int size) 330 { 331 struct queue_limits *limits = &q->limits; 332 333 limits->logical_block_size = size; 334 335 if (limits->physical_block_size < size) 336 limits->physical_block_size = size; 337 338 if (limits->io_min < limits->physical_block_size) 339 limits->io_min = limits->physical_block_size; 340 341 limits->max_hw_sectors = 342 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT); 343 limits->max_sectors = 344 round_down(limits->max_sectors, size >> SECTOR_SHIFT); 345 } 346 EXPORT_SYMBOL(blk_queue_logical_block_size); 347 348 /** 349 * blk_queue_physical_block_size - set physical block size for the queue 350 * @q: the request queue for the device 351 * @size: the physical block size, in bytes 352 * 353 * Description: 354 * This should be set to the lowest possible sector size that the 355 * hardware can operate on without reverting to read-modify-write 356 * operations. 357 */ 358 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) 359 { 360 q->limits.physical_block_size = size; 361 362 if (q->limits.physical_block_size < q->limits.logical_block_size) 363 q->limits.physical_block_size = q->limits.logical_block_size; 364 365 if (q->limits.io_min < q->limits.physical_block_size) 366 q->limits.io_min = q->limits.physical_block_size; 367 } 368 EXPORT_SYMBOL(blk_queue_physical_block_size); 369 370 /** 371 * blk_queue_zone_write_granularity - set zone write granularity for the queue 372 * @q: the request queue for the zoned device 373 * @size: the zone write granularity size, in bytes 374 * 375 * Description: 376 * This should be set to the lowest possible size allowing to write in 377 * sequential zones of a zoned block device. 378 */ 379 void blk_queue_zone_write_granularity(struct request_queue *q, 380 unsigned int size) 381 { 382 if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 383 return; 384 385 q->limits.zone_write_granularity = size; 386 387 if (q->limits.zone_write_granularity < q->limits.logical_block_size) 388 q->limits.zone_write_granularity = q->limits.logical_block_size; 389 } 390 EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity); 391 392 /** 393 * blk_queue_alignment_offset - set physical block alignment offset 394 * @q: the request queue for the device 395 * @offset: alignment offset in bytes 396 * 397 * Description: 398 * Some devices are naturally misaligned to compensate for things like 399 * the legacy DOS partition table 63-sector offset. Low-level drivers 400 * should call this function for devices whose first sector is not 401 * naturally aligned. 402 */ 403 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) 404 { 405 q->limits.alignment_offset = 406 offset & (q->limits.physical_block_size - 1); 407 q->limits.misaligned = 0; 408 } 409 EXPORT_SYMBOL(blk_queue_alignment_offset); 410 411 void blk_queue_update_readahead(struct request_queue *q) 412 { 413 /* 414 * For read-ahead of large files to be effective, we need to read ahead 415 * at least twice the optimal I/O size. 416 */ 417 q->backing_dev_info->ra_pages = 418 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES); 419 q->backing_dev_info->io_pages = 420 queue_max_sectors(q) >> (PAGE_SHIFT - 9); 421 } 422 EXPORT_SYMBOL_GPL(blk_queue_update_readahead); 423 424 /** 425 * blk_limits_io_min - set minimum request size for a device 426 * @limits: the queue limits 427 * @min: smallest I/O size in bytes 428 * 429 * Description: 430 * Some devices have an internal block size bigger than the reported 431 * hardware sector size. This function can be used to signal the 432 * smallest I/O the device can perform without incurring a performance 433 * penalty. 434 */ 435 void blk_limits_io_min(struct queue_limits *limits, unsigned int min) 436 { 437 limits->io_min = min; 438 439 if (limits->io_min < limits->logical_block_size) 440 limits->io_min = limits->logical_block_size; 441 442 if (limits->io_min < limits->physical_block_size) 443 limits->io_min = limits->physical_block_size; 444 } 445 EXPORT_SYMBOL(blk_limits_io_min); 446 447 /** 448 * blk_queue_io_min - set minimum request size for the queue 449 * @q: the request queue for the device 450 * @min: smallest I/O size in bytes 451 * 452 * Description: 453 * Storage devices may report a granularity or preferred minimum I/O 454 * size which is the smallest request the device can perform without 455 * incurring a performance penalty. For disk drives this is often the 456 * physical block size. For RAID arrays it is often the stripe chunk 457 * size. A properly aligned multiple of minimum_io_size is the 458 * preferred request size for workloads where a high number of I/O 459 * operations is desired. 460 */ 461 void blk_queue_io_min(struct request_queue *q, unsigned int min) 462 { 463 blk_limits_io_min(&q->limits, min); 464 } 465 EXPORT_SYMBOL(blk_queue_io_min); 466 467 /** 468 * blk_limits_io_opt - set optimal request size for a device 469 * @limits: the queue limits 470 * @opt: smallest I/O size in bytes 471 * 472 * Description: 473 * Storage devices may report an optimal I/O size, which is the 474 * device's preferred unit for sustained I/O. This is rarely reported 475 * for disk drives. For RAID arrays it is usually the stripe width or 476 * the internal track size. A properly aligned multiple of 477 * optimal_io_size is the preferred request size for workloads where 478 * sustained throughput is desired. 479 */ 480 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) 481 { 482 limits->io_opt = opt; 483 } 484 EXPORT_SYMBOL(blk_limits_io_opt); 485 486 /** 487 * blk_queue_io_opt - set optimal request size for the queue 488 * @q: the request queue for the device 489 * @opt: optimal request size in bytes 490 * 491 * Description: 492 * Storage devices may report an optimal I/O size, which is the 493 * device's preferred unit for sustained I/O. This is rarely reported 494 * for disk drives. For RAID arrays it is usually the stripe width or 495 * the internal track size. A properly aligned multiple of 496 * optimal_io_size is the preferred request size for workloads where 497 * sustained throughput is desired. 498 */ 499 void blk_queue_io_opt(struct request_queue *q, unsigned int opt) 500 { 501 blk_limits_io_opt(&q->limits, opt); 502 q->backing_dev_info->ra_pages = 503 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES); 504 } 505 EXPORT_SYMBOL(blk_queue_io_opt); 506 507 /** 508 * blk_stack_limits - adjust queue_limits for stacked devices 509 * @t: the stacking driver limits (top device) 510 * @b: the underlying queue limits (bottom, component device) 511 * @start: first data sector within component device 512 * 513 * Description: 514 * This function is used by stacking drivers like MD and DM to ensure 515 * that all component devices have compatible block sizes and 516 * alignments. The stacking driver must provide a queue_limits 517 * struct (top) and then iteratively call the stacking function for 518 * all component (bottom) devices. The stacking function will 519 * attempt to combine the values and ensure proper alignment. 520 * 521 * Returns 0 if the top and bottom queue_limits are compatible. The 522 * top device's block sizes and alignment offsets may be adjusted to 523 * ensure alignment with the bottom device. If no compatible sizes 524 * and alignments exist, -1 is returned and the resulting top 525 * queue_limits will have the misaligned flag set to indicate that 526 * the alignment_offset is undefined. 527 */ 528 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, 529 sector_t start) 530 { 531 unsigned int top, bottom, alignment, 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->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); 536 t->max_write_same_sectors = min(t->max_write_same_sectors, 537 b->max_write_same_sectors); 538 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, 539 b->max_write_zeroes_sectors); 540 t->max_zone_append_sectors = min(t->max_zone_append_sectors, 541 b->max_zone_append_sectors); 542 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); 543 544 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, 545 b->seg_boundary_mask); 546 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, 547 b->virt_boundary_mask); 548 549 t->max_segments = min_not_zero(t->max_segments, b->max_segments); 550 t->max_discard_segments = min_not_zero(t->max_discard_segments, 551 b->max_discard_segments); 552 t->max_integrity_segments = min_not_zero(t->max_integrity_segments, 553 b->max_integrity_segments); 554 555 t->max_segment_size = min_not_zero(t->max_segment_size, 556 b->max_segment_size); 557 558 t->misaligned |= b->misaligned; 559 560 alignment = queue_limit_alignment_offset(b, start); 561 562 /* Bottom device has different alignment. Check that it is 563 * compatible with the current top alignment. 564 */ 565 if (t->alignment_offset != alignment) { 566 567 top = max(t->physical_block_size, t->io_min) 568 + t->alignment_offset; 569 bottom = max(b->physical_block_size, b->io_min) + alignment; 570 571 /* Verify that top and bottom intervals line up */ 572 if (max(top, bottom) % min(top, bottom)) { 573 t->misaligned = 1; 574 ret = -1; 575 } 576 } 577 578 t->logical_block_size = max(t->logical_block_size, 579 b->logical_block_size); 580 581 t->physical_block_size = max(t->physical_block_size, 582 b->physical_block_size); 583 584 t->io_min = max(t->io_min, b->io_min); 585 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt); 586 587 /* Set non-power-of-2 compatible chunk_sectors boundary */ 588 if (b->chunk_sectors) 589 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors); 590 591 /* Physical block size a multiple of the logical block size? */ 592 if (t->physical_block_size & (t->logical_block_size - 1)) { 593 t->physical_block_size = t->logical_block_size; 594 t->misaligned = 1; 595 ret = -1; 596 } 597 598 /* Minimum I/O a multiple of the physical block size? */ 599 if (t->io_min & (t->physical_block_size - 1)) { 600 t->io_min = t->physical_block_size; 601 t->misaligned = 1; 602 ret = -1; 603 } 604 605 /* Optimal I/O a multiple of the physical block size? */ 606 if (t->io_opt & (t->physical_block_size - 1)) { 607 t->io_opt = 0; 608 t->misaligned = 1; 609 ret = -1; 610 } 611 612 /* chunk_sectors a multiple of the physical block size? */ 613 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) { 614 t->chunk_sectors = 0; 615 t->misaligned = 1; 616 ret = -1; 617 } 618 619 t->raid_partial_stripes_expensive = 620 max(t->raid_partial_stripes_expensive, 621 b->raid_partial_stripes_expensive); 622 623 /* Find lowest common alignment_offset */ 624 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment) 625 % max(t->physical_block_size, t->io_min); 626 627 /* Verify that new alignment_offset is on a logical block boundary */ 628 if (t->alignment_offset & (t->logical_block_size - 1)) { 629 t->misaligned = 1; 630 ret = -1; 631 } 632 633 /* Discard alignment and granularity */ 634 if (b->discard_granularity) { 635 alignment = queue_limit_discard_alignment(b, start); 636 637 if (t->discard_granularity != 0 && 638 t->discard_alignment != alignment) { 639 top = t->discard_granularity + t->discard_alignment; 640 bottom = b->discard_granularity + alignment; 641 642 /* Verify that top and bottom intervals line up */ 643 if ((max(top, bottom) % min(top, bottom)) != 0) 644 t->discard_misaligned = 1; 645 } 646 647 t->max_discard_sectors = min_not_zero(t->max_discard_sectors, 648 b->max_discard_sectors); 649 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, 650 b->max_hw_discard_sectors); 651 t->discard_granularity = max(t->discard_granularity, 652 b->discard_granularity); 653 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) % 654 t->discard_granularity; 655 } 656 657 t->zone_write_granularity = max(t->zone_write_granularity, 658 b->zone_write_granularity); 659 t->zoned = max(t->zoned, b->zoned); 660 return ret; 661 } 662 EXPORT_SYMBOL(blk_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 a top level gendisk and a bottom level 672 * block_device. 673 */ 674 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, 675 sector_t offset) 676 { 677 struct request_queue *t = disk->queue; 678 679 if (blk_stack_limits(&t->limits, &bdev_get_queue(bdev)->limits, 680 get_start_sect(bdev) + (offset >> 9)) < 0) { 681 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; 682 683 disk_name(disk, 0, top); 684 bdevname(bdev, bottom); 685 686 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", 687 top, bottom); 688 } 689 690 blk_queue_update_readahead(disk->queue); 691 } 692 EXPORT_SYMBOL(disk_stack_limits); 693 694 /** 695 * blk_queue_update_dma_pad - update pad mask 696 * @q: the request queue for the device 697 * @mask: pad mask 698 * 699 * Update dma pad mask. 700 * 701 * Appending pad buffer to a request modifies the last entry of a 702 * scatter list such that it includes the pad buffer. 703 **/ 704 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) 705 { 706 if (mask > q->dma_pad_mask) 707 q->dma_pad_mask = mask; 708 } 709 EXPORT_SYMBOL(blk_queue_update_dma_pad); 710 711 /** 712 * blk_queue_segment_boundary - set boundary rules for segment merging 713 * @q: the request queue for the device 714 * @mask: the memory boundary mask 715 **/ 716 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) 717 { 718 if (mask < PAGE_SIZE - 1) { 719 mask = PAGE_SIZE - 1; 720 printk(KERN_INFO "%s: set to minimum %lx\n", 721 __func__, mask); 722 } 723 724 q->limits.seg_boundary_mask = mask; 725 } 726 EXPORT_SYMBOL(blk_queue_segment_boundary); 727 728 /** 729 * blk_queue_virt_boundary - set boundary rules for bio merging 730 * @q: the request queue for the device 731 * @mask: the memory boundary mask 732 **/ 733 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask) 734 { 735 q->limits.virt_boundary_mask = mask; 736 737 /* 738 * Devices that require a virtual boundary do not support scatter/gather 739 * I/O natively, but instead require a descriptor list entry for each 740 * page (which might not be idential to the Linux PAGE_SIZE). Because 741 * of that they are not limited by our notion of "segment size". 742 */ 743 if (mask) 744 q->limits.max_segment_size = UINT_MAX; 745 } 746 EXPORT_SYMBOL(blk_queue_virt_boundary); 747 748 /** 749 * blk_queue_dma_alignment - set dma length and memory alignment 750 * @q: the request queue for the device 751 * @mask: alignment mask 752 * 753 * description: 754 * set required memory and length alignment for direct dma transactions. 755 * this is used when building direct io requests for the queue. 756 * 757 **/ 758 void blk_queue_dma_alignment(struct request_queue *q, int mask) 759 { 760 q->dma_alignment = mask; 761 } 762 EXPORT_SYMBOL(blk_queue_dma_alignment); 763 764 /** 765 * blk_queue_update_dma_alignment - update dma length and memory alignment 766 * @q: the request queue for the device 767 * @mask: alignment mask 768 * 769 * description: 770 * update required memory and length alignment for direct dma transactions. 771 * If the requested alignment is larger than the current alignment, then 772 * the current queue alignment is updated to the new value, otherwise it 773 * is left alone. The design of this is to allow multiple objects 774 * (driver, device, transport etc) to set their respective 775 * alignments without having them interfere. 776 * 777 **/ 778 void blk_queue_update_dma_alignment(struct request_queue *q, int mask) 779 { 780 BUG_ON(mask > PAGE_SIZE); 781 782 if (mask > q->dma_alignment) 783 q->dma_alignment = mask; 784 } 785 EXPORT_SYMBOL(blk_queue_update_dma_alignment); 786 787 /** 788 * blk_set_queue_depth - tell the block layer about the device queue depth 789 * @q: the request queue for the device 790 * @depth: queue depth 791 * 792 */ 793 void blk_set_queue_depth(struct request_queue *q, unsigned int depth) 794 { 795 q->queue_depth = depth; 796 rq_qos_queue_depth_changed(q); 797 } 798 EXPORT_SYMBOL(blk_set_queue_depth); 799 800 /** 801 * blk_queue_write_cache - configure queue's write cache 802 * @q: the request queue for the device 803 * @wc: write back cache on or off 804 * @fua: device supports FUA writes, if true 805 * 806 * Tell the block layer about the write cache of @q. 807 */ 808 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua) 809 { 810 if (wc) 811 blk_queue_flag_set(QUEUE_FLAG_WC, q); 812 else 813 blk_queue_flag_clear(QUEUE_FLAG_WC, q); 814 if (fua) 815 blk_queue_flag_set(QUEUE_FLAG_FUA, q); 816 else 817 blk_queue_flag_clear(QUEUE_FLAG_FUA, q); 818 819 wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags)); 820 } 821 EXPORT_SYMBOL_GPL(blk_queue_write_cache); 822 823 /** 824 * blk_queue_required_elevator_features - Set a queue required elevator features 825 * @q: the request queue for the target device 826 * @features: Required elevator features OR'ed together 827 * 828 * Tell the block layer that for the device controlled through @q, only the 829 * only elevators that can be used are those that implement at least the set of 830 * features specified by @features. 831 */ 832 void blk_queue_required_elevator_features(struct request_queue *q, 833 unsigned int features) 834 { 835 q->required_elevator_features = features; 836 } 837 EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features); 838 839 /** 840 * blk_queue_can_use_dma_map_merging - configure queue for merging segments. 841 * @q: the request queue for the device 842 * @dev: the device pointer for dma 843 * 844 * Tell the block layer about merging the segments by dma map of @q. 845 */ 846 bool blk_queue_can_use_dma_map_merging(struct request_queue *q, 847 struct device *dev) 848 { 849 unsigned long boundary = dma_get_merge_boundary(dev); 850 851 if (!boundary) 852 return false; 853 854 /* No need to update max_segment_size. see blk_queue_virt_boundary() */ 855 blk_queue_virt_boundary(q, boundary); 856 857 return true; 858 } 859 EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging); 860 861 /** 862 * blk_queue_set_zoned - configure a disk queue zoned model. 863 * @disk: the gendisk of the queue to configure 864 * @model: the zoned model to set 865 * 866 * Set the zoned model of the request queue of @disk according to @model. 867 * When @model is BLK_ZONED_HM (host managed), this should be called only 868 * if zoned block device support is enabled (CONFIG_BLK_DEV_ZONED option). 869 * If @model specifies BLK_ZONED_HA (host aware), the effective model used 870 * depends on CONFIG_BLK_DEV_ZONED settings and on the existence of partitions 871 * on the disk. 872 */ 873 void blk_queue_set_zoned(struct gendisk *disk, enum blk_zoned_model model) 874 { 875 struct request_queue *q = disk->queue; 876 877 switch (model) { 878 case BLK_ZONED_HM: 879 /* 880 * Host managed devices are supported only if 881 * CONFIG_BLK_DEV_ZONED is enabled. 882 */ 883 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)); 884 break; 885 case BLK_ZONED_HA: 886 /* 887 * Host aware devices can be treated either as regular block 888 * devices (similar to drive managed devices) or as zoned block 889 * devices to take advantage of the zone command set, similarly 890 * to host managed devices. We try the latter if there are no 891 * partitions and zoned block device support is enabled, else 892 * we do nothing special as far as the block layer is concerned. 893 */ 894 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) || 895 !xa_empty(&disk->part_tbl)) 896 model = BLK_ZONED_NONE; 897 break; 898 case BLK_ZONED_NONE: 899 default: 900 if (WARN_ON_ONCE(model != BLK_ZONED_NONE)) 901 model = BLK_ZONED_NONE; 902 break; 903 } 904 905 q->limits.zoned = model; 906 if (model != BLK_ZONED_NONE) { 907 /* 908 * Set the zone write granularity to the device logical block 909 * size by default. The driver can change this value if needed. 910 */ 911 blk_queue_zone_write_granularity(q, 912 queue_logical_block_size(q)); 913 } else { 914 blk_queue_clear_zone_settings(q); 915 } 916 } 917 EXPORT_SYMBOL_GPL(blk_queue_set_zoned); 918 919 static int __init blk_settings_init(void) 920 { 921 blk_max_low_pfn = max_low_pfn - 1; 922 blk_max_pfn = max_pfn - 1; 923 return 0; 924 } 925 subsys_initcall(blk_settings_init); 926