| 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/pagemap.h> |
| 11 | #include <linux/backing-dev-defs.h> |
| 12 | #include <linux/gcd.h> |
| 13 | #include <linux/lcm.h> |
| 14 | #include <linux/jiffies.h> |
| 15 | #include <linux/gfp.h> |
| 16 | #include <linux/dma-mapping.h> |
| 17 | |
| 18 | #include "blk.h" |
| 19 | #include "blk-rq-qos.h" |
| 20 | #include "blk-wbt.h" |
| 21 | |
| 22 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
| 23 | { |
| 24 | q->rq_timeout = timeout; |
| 25 | } |
| 26 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); |
| 27 | |
| 28 | /** |
| 29 | * blk_set_default_limits - reset limits to default values |
| 30 | * @lim: the queue_limits structure to reset |
| 31 | * |
| 32 | * Description: |
| 33 | * Returns a queue_limit struct to its default state. |
| 34 | */ |
| 35 | void blk_set_default_limits(struct queue_limits *lim) |
| 36 | { |
| 37 | lim->max_segments = BLK_MAX_SEGMENTS; |
| 38 | lim->max_discard_segments = 1; |
| 39 | lim->max_integrity_segments = 0; |
| 40 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
| 41 | lim->virt_boundary_mask = 0; |
| 42 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; |
| 43 | lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; |
| 44 | lim->max_user_sectors = lim->max_dev_sectors = 0; |
| 45 | lim->chunk_sectors = 0; |
| 46 | lim->max_write_zeroes_sectors = 0; |
| 47 | lim->max_zone_append_sectors = 0; |
| 48 | lim->max_discard_sectors = 0; |
| 49 | lim->max_hw_discard_sectors = 0; |
| 50 | lim->max_secure_erase_sectors = 0; |
| 51 | lim->discard_granularity = 0; |
| 52 | lim->discard_alignment = 0; |
| 53 | lim->discard_misaligned = 0; |
| 54 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
| 55 | lim->bounce = BLK_BOUNCE_NONE; |
| 56 | lim->alignment_offset = 0; |
| 57 | lim->io_opt = 0; |
| 58 | lim->misaligned = 0; |
| 59 | lim->zoned = BLK_ZONED_NONE; |
| 60 | lim->zone_write_granularity = 0; |
| 61 | lim->dma_alignment = 511; |
| 62 | } |
| 63 | |
| 64 | /** |
| 65 | * blk_set_stacking_limits - set default limits for stacking devices |
| 66 | * @lim: the queue_limits structure to reset |
| 67 | * |
| 68 | * Description: |
| 69 | * Returns a queue_limit struct to its default state. Should be used |
| 70 | * by stacking drivers like DM that have no internal limits. |
| 71 | */ |
| 72 | void blk_set_stacking_limits(struct queue_limits *lim) |
| 73 | { |
| 74 | blk_set_default_limits(lim); |
| 75 | |
| 76 | /* Inherit limits from component devices */ |
| 77 | lim->max_segments = USHRT_MAX; |
| 78 | lim->max_discard_segments = USHRT_MAX; |
| 79 | lim->max_hw_sectors = UINT_MAX; |
| 80 | lim->max_segment_size = UINT_MAX; |
| 81 | lim->max_sectors = UINT_MAX; |
| 82 | lim->max_dev_sectors = UINT_MAX; |
| 83 | lim->max_write_zeroes_sectors = UINT_MAX; |
| 84 | lim->max_zone_append_sectors = UINT_MAX; |
| 85 | } |
| 86 | EXPORT_SYMBOL(blk_set_stacking_limits); |
| 87 | |
| 88 | /** |
| 89 | * blk_queue_bounce_limit - set bounce buffer limit for queue |
| 90 | * @q: the request queue for the device |
| 91 | * @bounce: bounce limit to enforce |
| 92 | * |
| 93 | * Description: |
| 94 | * Force bouncing for ISA DMA ranges or highmem. |
| 95 | * |
| 96 | * DEPRECATED, don't use in new code. |
| 97 | **/ |
| 98 | void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce) |
| 99 | { |
| 100 | q->limits.bounce = bounce; |
| 101 | } |
| 102 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
| 103 | |
| 104 | /** |
| 105 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue |
| 106 | * @q: the request queue for the device |
| 107 | * @max_hw_sectors: max hardware sectors in the usual 512b unit |
| 108 | * |
| 109 | * Description: |
| 110 | * Enables a low level driver to set a hard upper limit, |
| 111 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by |
| 112 | * the device driver based upon the capabilities of the I/O |
| 113 | * controller. |
| 114 | * |
| 115 | * max_dev_sectors is a hard limit imposed by the storage device for |
| 116 | * READ/WRITE requests. It is set by the disk driver. |
| 117 | * |
| 118 | * max_sectors is a soft limit imposed by the block layer for |
| 119 | * filesystem type requests. This value can be overridden on a |
| 120 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. |
| 121 | * The soft limit can not exceed max_hw_sectors. |
| 122 | **/ |
| 123 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) |
| 124 | { |
| 125 | struct queue_limits *limits = &q->limits; |
| 126 | unsigned int max_sectors; |
| 127 | |
| 128 | if ((max_hw_sectors << 9) < PAGE_SIZE) { |
| 129 | max_hw_sectors = 1 << (PAGE_SHIFT - 9); |
| 130 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 131 | __func__, max_hw_sectors); |
| 132 | } |
| 133 | |
| 134 | max_hw_sectors = round_down(max_hw_sectors, |
| 135 | limits->logical_block_size >> SECTOR_SHIFT); |
| 136 | limits->max_hw_sectors = max_hw_sectors; |
| 137 | |
| 138 | max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors); |
| 139 | |
| 140 | if (limits->max_user_sectors) |
| 141 | max_sectors = min(max_sectors, limits->max_user_sectors); |
| 142 | else |
| 143 | max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS); |
| 144 | |
| 145 | max_sectors = round_down(max_sectors, |
| 146 | limits->logical_block_size >> SECTOR_SHIFT); |
| 147 | limits->max_sectors = max_sectors; |
| 148 | |
| 149 | if (!q->disk) |
| 150 | return; |
| 151 | q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9); |
| 152 | } |
| 153 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
| 154 | |
| 155 | /** |
| 156 | * blk_queue_chunk_sectors - set size of the chunk for this queue |
| 157 | * @q: the request queue for the device |
| 158 | * @chunk_sectors: chunk sectors in the usual 512b unit |
| 159 | * |
| 160 | * Description: |
| 161 | * If a driver doesn't want IOs to cross a given chunk size, it can set |
| 162 | * this limit and prevent merging across chunks. Note that the block layer |
| 163 | * must accept a page worth of data at any offset. So if the crossing of |
| 164 | * chunks is a hard limitation in the driver, it must still be prepared |
| 165 | * to split single page bios. |
| 166 | **/ |
| 167 | void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) |
| 168 | { |
| 169 | q->limits.chunk_sectors = chunk_sectors; |
| 170 | } |
| 171 | EXPORT_SYMBOL(blk_queue_chunk_sectors); |
| 172 | |
| 173 | /** |
| 174 | * blk_queue_max_discard_sectors - set max sectors for a single discard |
| 175 | * @q: the request queue for the device |
| 176 | * @max_discard_sectors: maximum number of sectors to discard |
| 177 | **/ |
| 178 | void blk_queue_max_discard_sectors(struct request_queue *q, |
| 179 | unsigned int max_discard_sectors) |
| 180 | { |
| 181 | q->limits.max_hw_discard_sectors = max_discard_sectors; |
| 182 | q->limits.max_discard_sectors = max_discard_sectors; |
| 183 | } |
| 184 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); |
| 185 | |
| 186 | /** |
| 187 | * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase |
| 188 | * @q: the request queue for the device |
| 189 | * @max_sectors: maximum number of sectors to secure_erase |
| 190 | **/ |
| 191 | void blk_queue_max_secure_erase_sectors(struct request_queue *q, |
| 192 | unsigned int max_sectors) |
| 193 | { |
| 194 | q->limits.max_secure_erase_sectors = max_sectors; |
| 195 | } |
| 196 | EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors); |
| 197 | |
| 198 | /** |
| 199 | * blk_queue_max_write_zeroes_sectors - set max sectors for a single |
| 200 | * write zeroes |
| 201 | * @q: the request queue for the device |
| 202 | * @max_write_zeroes_sectors: maximum number of sectors to write per command |
| 203 | **/ |
| 204 | void blk_queue_max_write_zeroes_sectors(struct request_queue *q, |
| 205 | unsigned int max_write_zeroes_sectors) |
| 206 | { |
| 207 | q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors; |
| 208 | } |
| 209 | EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors); |
| 210 | |
| 211 | /** |
| 212 | * blk_queue_max_zone_append_sectors - set max sectors for a single zone append |
| 213 | * @q: the request queue for the device |
| 214 | * @max_zone_append_sectors: maximum number of sectors to write per command |
| 215 | **/ |
| 216 | void blk_queue_max_zone_append_sectors(struct request_queue *q, |
| 217 | unsigned int max_zone_append_sectors) |
| 218 | { |
| 219 | unsigned int max_sectors; |
| 220 | |
| 221 | if (WARN_ON(!blk_queue_is_zoned(q))) |
| 222 | return; |
| 223 | |
| 224 | max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors); |
| 225 | max_sectors = min(q->limits.chunk_sectors, max_sectors); |
| 226 | |
| 227 | /* |
| 228 | * Signal eventual driver bugs resulting in the max_zone_append sectors limit |
| 229 | * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set, |
| 230 | * or the max_hw_sectors limit not set. |
| 231 | */ |
| 232 | WARN_ON(!max_sectors); |
| 233 | |
| 234 | q->limits.max_zone_append_sectors = max_sectors; |
| 235 | } |
| 236 | EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors); |
| 237 | |
| 238 | /** |
| 239 | * blk_queue_max_segments - set max hw segments for a request for this queue |
| 240 | * @q: the request queue for the device |
| 241 | * @max_segments: max number of segments |
| 242 | * |
| 243 | * Description: |
| 244 | * Enables a low level driver to set an upper limit on the number of |
| 245 | * hw data segments in a request. |
| 246 | **/ |
| 247 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) |
| 248 | { |
| 249 | if (!max_segments) { |
| 250 | max_segments = 1; |
| 251 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 252 | __func__, max_segments); |
| 253 | } |
| 254 | |
| 255 | q->limits.max_segments = max_segments; |
| 256 | } |
| 257 | EXPORT_SYMBOL(blk_queue_max_segments); |
| 258 | |
| 259 | /** |
| 260 | * blk_queue_max_discard_segments - set max segments for discard requests |
| 261 | * @q: the request queue for the device |
| 262 | * @max_segments: max number of segments |
| 263 | * |
| 264 | * Description: |
| 265 | * Enables a low level driver to set an upper limit on the number of |
| 266 | * segments in a discard request. |
| 267 | **/ |
| 268 | void blk_queue_max_discard_segments(struct request_queue *q, |
| 269 | unsigned short max_segments) |
| 270 | { |
| 271 | q->limits.max_discard_segments = max_segments; |
| 272 | } |
| 273 | EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments); |
| 274 | |
| 275 | /** |
| 276 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg |
| 277 | * @q: the request queue for the device |
| 278 | * @max_size: max size of segment in bytes |
| 279 | * |
| 280 | * Description: |
| 281 | * Enables a low level driver to set an upper limit on the size of a |
| 282 | * coalesced segment |
| 283 | **/ |
| 284 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) |
| 285 | { |
| 286 | if (max_size < PAGE_SIZE) { |
| 287 | max_size = PAGE_SIZE; |
| 288 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 289 | __func__, max_size); |
| 290 | } |
| 291 | |
| 292 | /* see blk_queue_virt_boundary() for the explanation */ |
| 293 | WARN_ON_ONCE(q->limits.virt_boundary_mask); |
| 294 | |
| 295 | q->limits.max_segment_size = max_size; |
| 296 | } |
| 297 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
| 298 | |
| 299 | /** |
| 300 | * blk_queue_logical_block_size - set logical block size for the queue |
| 301 | * @q: the request queue for the device |
| 302 | * @size: the logical block size, in bytes |
| 303 | * |
| 304 | * Description: |
| 305 | * This should be set to the lowest possible block size that the |
| 306 | * storage device can address. The default of 512 covers most |
| 307 | * hardware. |
| 308 | **/ |
| 309 | void blk_queue_logical_block_size(struct request_queue *q, unsigned int size) |
| 310 | { |
| 311 | struct queue_limits *limits = &q->limits; |
| 312 | |
| 313 | limits->logical_block_size = size; |
| 314 | |
| 315 | if (limits->physical_block_size < size) |
| 316 | limits->physical_block_size = size; |
| 317 | |
| 318 | if (limits->io_min < limits->physical_block_size) |
| 319 | limits->io_min = limits->physical_block_size; |
| 320 | |
| 321 | limits->max_hw_sectors = |
| 322 | round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT); |
| 323 | limits->max_sectors = |
| 324 | round_down(limits->max_sectors, size >> SECTOR_SHIFT); |
| 325 | } |
| 326 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
| 327 | |
| 328 | /** |
| 329 | * blk_queue_physical_block_size - set physical block size for the queue |
| 330 | * @q: the request queue for the device |
| 331 | * @size: the physical block size, in bytes |
| 332 | * |
| 333 | * Description: |
| 334 | * This should be set to the lowest possible sector size that the |
| 335 | * hardware can operate on without reverting to read-modify-write |
| 336 | * operations. |
| 337 | */ |
| 338 | void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) |
| 339 | { |
| 340 | q->limits.physical_block_size = size; |
| 341 | |
| 342 | if (q->limits.physical_block_size < q->limits.logical_block_size) |
| 343 | q->limits.physical_block_size = q->limits.logical_block_size; |
| 344 | |
| 345 | if (q->limits.io_min < q->limits.physical_block_size) |
| 346 | q->limits.io_min = q->limits.physical_block_size; |
| 347 | } |
| 348 | EXPORT_SYMBOL(blk_queue_physical_block_size); |
| 349 | |
| 350 | /** |
| 351 | * blk_queue_zone_write_granularity - set zone write granularity for the queue |
| 352 | * @q: the request queue for the zoned device |
| 353 | * @size: the zone write granularity size, in bytes |
| 354 | * |
| 355 | * Description: |
| 356 | * This should be set to the lowest possible size allowing to write in |
| 357 | * sequential zones of a zoned block device. |
| 358 | */ |
| 359 | void blk_queue_zone_write_granularity(struct request_queue *q, |
| 360 | unsigned int size) |
| 361 | { |
| 362 | if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) |
| 363 | return; |
| 364 | |
| 365 | q->limits.zone_write_granularity = size; |
| 366 | |
| 367 | if (q->limits.zone_write_granularity < q->limits.logical_block_size) |
| 368 | q->limits.zone_write_granularity = q->limits.logical_block_size; |
| 369 | } |
| 370 | EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity); |
| 371 | |
| 372 | /** |
| 373 | * blk_queue_alignment_offset - set physical block alignment offset |
| 374 | * @q: the request queue for the device |
| 375 | * @offset: alignment offset in bytes |
| 376 | * |
| 377 | * Description: |
| 378 | * Some devices are naturally misaligned to compensate for things like |
| 379 | * the legacy DOS partition table 63-sector offset. Low-level drivers |
| 380 | * should call this function for devices whose first sector is not |
| 381 | * naturally aligned. |
| 382 | */ |
| 383 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) |
| 384 | { |
| 385 | q->limits.alignment_offset = |
| 386 | offset & (q->limits.physical_block_size - 1); |
| 387 | q->limits.misaligned = 0; |
| 388 | } |
| 389 | EXPORT_SYMBOL(blk_queue_alignment_offset); |
| 390 | |
| 391 | void disk_update_readahead(struct gendisk *disk) |
| 392 | { |
| 393 | struct request_queue *q = disk->queue; |
| 394 | |
| 395 | /* |
| 396 | * For read-ahead of large files to be effective, we need to read ahead |
| 397 | * at least twice the optimal I/O size. |
| 398 | */ |
| 399 | disk->bdi->ra_pages = |
| 400 | max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES); |
| 401 | disk->bdi->io_pages = queue_max_sectors(q) >> (PAGE_SHIFT - 9); |
| 402 | } |
| 403 | EXPORT_SYMBOL_GPL(disk_update_readahead); |
| 404 | |
| 405 | /** |
| 406 | * blk_limits_io_min - set minimum request size for a device |
| 407 | * @limits: the queue limits |
| 408 | * @min: smallest I/O size in bytes |
| 409 | * |
| 410 | * Description: |
| 411 | * Some devices have an internal block size bigger than the reported |
| 412 | * hardware sector size. This function can be used to signal the |
| 413 | * smallest I/O the device can perform without incurring a performance |
| 414 | * penalty. |
| 415 | */ |
| 416 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) |
| 417 | { |
| 418 | limits->io_min = min; |
| 419 | |
| 420 | if (limits->io_min < limits->logical_block_size) |
| 421 | limits->io_min = limits->logical_block_size; |
| 422 | |
| 423 | if (limits->io_min < limits->physical_block_size) |
| 424 | limits->io_min = limits->physical_block_size; |
| 425 | } |
| 426 | EXPORT_SYMBOL(blk_limits_io_min); |
| 427 | |
| 428 | /** |
| 429 | * blk_queue_io_min - set minimum request size for the queue |
| 430 | * @q: the request queue for the device |
| 431 | * @min: smallest I/O size in bytes |
| 432 | * |
| 433 | * Description: |
| 434 | * Storage devices may report a granularity or preferred minimum I/O |
| 435 | * size which is the smallest request the device can perform without |
| 436 | * incurring a performance penalty. For disk drives this is often the |
| 437 | * physical block size. For RAID arrays it is often the stripe chunk |
| 438 | * size. A properly aligned multiple of minimum_io_size is the |
| 439 | * preferred request size for workloads where a high number of I/O |
| 440 | * operations is desired. |
| 441 | */ |
| 442 | void blk_queue_io_min(struct request_queue *q, unsigned int min) |
| 443 | { |
| 444 | blk_limits_io_min(&q->limits, min); |
| 445 | } |
| 446 | EXPORT_SYMBOL(blk_queue_io_min); |
| 447 | |
| 448 | /** |
| 449 | * blk_limits_io_opt - set optimal request size for a device |
| 450 | * @limits: the queue limits |
| 451 | * @opt: smallest I/O size in bytes |
| 452 | * |
| 453 | * Description: |
| 454 | * Storage devices may report an optimal I/O size, which is the |
| 455 | * device's preferred unit for sustained I/O. This is rarely reported |
| 456 | * for disk drives. For RAID arrays it is usually the stripe width or |
| 457 | * the internal track size. A properly aligned multiple of |
| 458 | * optimal_io_size is the preferred request size for workloads where |
| 459 | * sustained throughput is desired. |
| 460 | */ |
| 461 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) |
| 462 | { |
| 463 | limits->io_opt = opt; |
| 464 | } |
| 465 | EXPORT_SYMBOL(blk_limits_io_opt); |
| 466 | |
| 467 | /** |
| 468 | * blk_queue_io_opt - set optimal request size for the queue |
| 469 | * @q: the request queue for the device |
| 470 | * @opt: optimal request 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_queue_io_opt(struct request_queue *q, unsigned int opt) |
| 481 | { |
| 482 | blk_limits_io_opt(&q->limits, opt); |
| 483 | if (!q->disk) |
| 484 | return; |
| 485 | q->disk->bdi->ra_pages = |
| 486 | max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES); |
| 487 | } |
| 488 | EXPORT_SYMBOL(blk_queue_io_opt); |
| 489 | |
| 490 | static int queue_limit_alignment_offset(const struct queue_limits *lim, |
| 491 | sector_t sector) |
| 492 | { |
| 493 | unsigned int granularity = max(lim->physical_block_size, lim->io_min); |
| 494 | unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT) |
| 495 | << SECTOR_SHIFT; |
| 496 | |
| 497 | return (granularity + lim->alignment_offset - alignment) % granularity; |
| 498 | } |
| 499 | |
| 500 | static unsigned int queue_limit_discard_alignment( |
| 501 | const struct queue_limits *lim, sector_t sector) |
| 502 | { |
| 503 | unsigned int alignment, granularity, offset; |
| 504 | |
| 505 | if (!lim->max_discard_sectors) |
| 506 | return 0; |
| 507 | |
| 508 | /* Why are these in bytes, not sectors? */ |
| 509 | alignment = lim->discard_alignment >> SECTOR_SHIFT; |
| 510 | granularity = lim->discard_granularity >> SECTOR_SHIFT; |
| 511 | if (!granularity) |
| 512 | return 0; |
| 513 | |
| 514 | /* Offset of the partition start in 'granularity' sectors */ |
| 515 | offset = sector_div(sector, granularity); |
| 516 | |
| 517 | /* And why do we do this modulus *again* in blkdev_issue_discard()? */ |
| 518 | offset = (granularity + alignment - offset) % granularity; |
| 519 | |
| 520 | /* Turn it back into bytes, gaah */ |
| 521 | return offset << SECTOR_SHIFT; |
| 522 | } |
| 523 | |
| 524 | static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs) |
| 525 | { |
| 526 | sectors = round_down(sectors, lbs >> SECTOR_SHIFT); |
| 527 | if (sectors < PAGE_SIZE >> SECTOR_SHIFT) |
| 528 | sectors = PAGE_SIZE >> SECTOR_SHIFT; |
| 529 | return sectors; |
| 530 | } |
| 531 | |
| 532 | /** |
| 533 | * blk_stack_limits - adjust queue_limits for stacked devices |
| 534 | * @t: the stacking driver limits (top device) |
| 535 | * @b: the underlying queue limits (bottom, component device) |
| 536 | * @start: first data sector within component device |
| 537 | * |
| 538 | * Description: |
| 539 | * This function is used by stacking drivers like MD and DM to ensure |
| 540 | * that all component devices have compatible block sizes and |
| 541 | * alignments. The stacking driver must provide a queue_limits |
| 542 | * struct (top) and then iteratively call the stacking function for |
| 543 | * all component (bottom) devices. The stacking function will |
| 544 | * attempt to combine the values and ensure proper alignment. |
| 545 | * |
| 546 | * Returns 0 if the top and bottom queue_limits are compatible. The |
| 547 | * top device's block sizes and alignment offsets may be adjusted to |
| 548 | * ensure alignment with the bottom device. If no compatible sizes |
| 549 | * and alignments exist, -1 is returned and the resulting top |
| 550 | * queue_limits will have the misaligned flag set to indicate that |
| 551 | * the alignment_offset is undefined. |
| 552 | */ |
| 553 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, |
| 554 | sector_t start) |
| 555 | { |
| 556 | unsigned int top, bottom, alignment, ret = 0; |
| 557 | |
| 558 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
| 559 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); |
| 560 | t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); |
| 561 | t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, |
| 562 | b->max_write_zeroes_sectors); |
| 563 | t->max_zone_append_sectors = min(t->max_zone_append_sectors, |
| 564 | b->max_zone_append_sectors); |
| 565 | t->bounce = max(t->bounce, b->bounce); |
| 566 | |
| 567 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, |
| 568 | b->seg_boundary_mask); |
| 569 | t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, |
| 570 | b->virt_boundary_mask); |
| 571 | |
| 572 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
| 573 | t->max_discard_segments = min_not_zero(t->max_discard_segments, |
| 574 | b->max_discard_segments); |
| 575 | t->max_integrity_segments = min_not_zero(t->max_integrity_segments, |
| 576 | b->max_integrity_segments); |
| 577 | |
| 578 | t->max_segment_size = min_not_zero(t->max_segment_size, |
| 579 | b->max_segment_size); |
| 580 | |
| 581 | t->misaligned |= b->misaligned; |
| 582 | |
| 583 | alignment = queue_limit_alignment_offset(b, start); |
| 584 | |
| 585 | /* Bottom device has different alignment. Check that it is |
| 586 | * compatible with the current top alignment. |
| 587 | */ |
| 588 | if (t->alignment_offset != alignment) { |
| 589 | |
| 590 | top = max(t->physical_block_size, t->io_min) |
| 591 | + t->alignment_offset; |
| 592 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
| 593 | |
| 594 | /* Verify that top and bottom intervals line up */ |
| 595 | if (max(top, bottom) % min(top, bottom)) { |
| 596 | t->misaligned = 1; |
| 597 | ret = -1; |
| 598 | } |
| 599 | } |
| 600 | |
| 601 | t->logical_block_size = max(t->logical_block_size, |
| 602 | b->logical_block_size); |
| 603 | |
| 604 | t->physical_block_size = max(t->physical_block_size, |
| 605 | b->physical_block_size); |
| 606 | |
| 607 | t->io_min = max(t->io_min, b->io_min); |
| 608 | t->io_opt = lcm_not_zero(t->io_opt, b->io_opt); |
| 609 | t->dma_alignment = max(t->dma_alignment, b->dma_alignment); |
| 610 | |
| 611 | /* Set non-power-of-2 compatible chunk_sectors boundary */ |
| 612 | if (b->chunk_sectors) |
| 613 | t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors); |
| 614 | |
| 615 | /* Physical block size a multiple of the logical block size? */ |
| 616 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
| 617 | t->physical_block_size = t->logical_block_size; |
| 618 | t->misaligned = 1; |
| 619 | ret = -1; |
| 620 | } |
| 621 | |
| 622 | /* Minimum I/O a multiple of the physical block size? */ |
| 623 | if (t->io_min & (t->physical_block_size - 1)) { |
| 624 | t->io_min = t->physical_block_size; |
| 625 | t->misaligned = 1; |
| 626 | ret = -1; |
| 627 | } |
| 628 | |
| 629 | /* Optimal I/O a multiple of the physical block size? */ |
| 630 | if (t->io_opt & (t->physical_block_size - 1)) { |
| 631 | t->io_opt = 0; |
| 632 | t->misaligned = 1; |
| 633 | ret = -1; |
| 634 | } |
| 635 | |
| 636 | /* chunk_sectors a multiple of the physical block size? */ |
| 637 | if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) { |
| 638 | t->chunk_sectors = 0; |
| 639 | t->misaligned = 1; |
| 640 | ret = -1; |
| 641 | } |
| 642 | |
| 643 | t->raid_partial_stripes_expensive = |
| 644 | max(t->raid_partial_stripes_expensive, |
| 645 | b->raid_partial_stripes_expensive); |
| 646 | |
| 647 | /* Find lowest common alignment_offset */ |
| 648 | t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment) |
| 649 | % max(t->physical_block_size, t->io_min); |
| 650 | |
| 651 | /* Verify that new alignment_offset is on a logical block boundary */ |
| 652 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
| 653 | t->misaligned = 1; |
| 654 | ret = -1; |
| 655 | } |
| 656 | |
| 657 | t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size); |
| 658 | t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size); |
| 659 | t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size); |
| 660 | |
| 661 | /* Discard alignment and granularity */ |
| 662 | if (b->discard_granularity) { |
| 663 | alignment = queue_limit_discard_alignment(b, start); |
| 664 | |
| 665 | if (t->discard_granularity != 0 && |
| 666 | t->discard_alignment != alignment) { |
| 667 | top = t->discard_granularity + t->discard_alignment; |
| 668 | bottom = b->discard_granularity + alignment; |
| 669 | |
| 670 | /* Verify that top and bottom intervals line up */ |
| 671 | if ((max(top, bottom) % min(top, bottom)) != 0) |
| 672 | t->discard_misaligned = 1; |
| 673 | } |
| 674 | |
| 675 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
| 676 | b->max_discard_sectors); |
| 677 | t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, |
| 678 | b->max_hw_discard_sectors); |
| 679 | t->discard_granularity = max(t->discard_granularity, |
| 680 | b->discard_granularity); |
| 681 | t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) % |
| 682 | t->discard_granularity; |
| 683 | } |
| 684 | t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors, |
| 685 | b->max_secure_erase_sectors); |
| 686 | t->zone_write_granularity = max(t->zone_write_granularity, |
| 687 | b->zone_write_granularity); |
| 688 | t->zoned = max(t->zoned, b->zoned); |
| 689 | return ret; |
| 690 | } |
| 691 | EXPORT_SYMBOL(blk_stack_limits); |
| 692 | |
| 693 | /** |
| 694 | * disk_stack_limits - adjust queue limits for stacked drivers |
| 695 | * @disk: MD/DM gendisk (top) |
| 696 | * @bdev: the underlying block device (bottom) |
| 697 | * @offset: offset to beginning of data within component device |
| 698 | * |
| 699 | * Description: |
| 700 | * Merges the limits for a top level gendisk and a bottom level |
| 701 | * block_device. |
| 702 | */ |
| 703 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, |
| 704 | sector_t offset) |
| 705 | { |
| 706 | struct request_queue *t = disk->queue; |
| 707 | |
| 708 | if (blk_stack_limits(&t->limits, &bdev_get_queue(bdev)->limits, |
| 709 | get_start_sect(bdev) + (offset >> 9)) < 0) |
| 710 | pr_notice("%s: Warning: Device %pg is misaligned\n", |
| 711 | disk->disk_name, bdev); |
| 712 | |
| 713 | disk_update_readahead(disk); |
| 714 | } |
| 715 | EXPORT_SYMBOL(disk_stack_limits); |
| 716 | |
| 717 | /** |
| 718 | * blk_queue_update_dma_pad - update pad mask |
| 719 | * @q: the request queue for the device |
| 720 | * @mask: pad mask |
| 721 | * |
| 722 | * Update dma pad mask. |
| 723 | * |
| 724 | * Appending pad buffer to a request modifies the last entry of a |
| 725 | * scatter list such that it includes the pad buffer. |
| 726 | **/ |
| 727 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) |
| 728 | { |
| 729 | if (mask > q->dma_pad_mask) |
| 730 | q->dma_pad_mask = mask; |
| 731 | } |
| 732 | EXPORT_SYMBOL(blk_queue_update_dma_pad); |
| 733 | |
| 734 | /** |
| 735 | * blk_queue_segment_boundary - set boundary rules for segment merging |
| 736 | * @q: the request queue for the device |
| 737 | * @mask: the memory boundary mask |
| 738 | **/ |
| 739 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) |
| 740 | { |
| 741 | if (mask < PAGE_SIZE - 1) { |
| 742 | mask = PAGE_SIZE - 1; |
| 743 | printk(KERN_INFO "%s: set to minimum %lx\n", |
| 744 | __func__, mask); |
| 745 | } |
| 746 | |
| 747 | q->limits.seg_boundary_mask = mask; |
| 748 | } |
| 749 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
| 750 | |
| 751 | /** |
| 752 | * blk_queue_virt_boundary - set boundary rules for bio merging |
| 753 | * @q: the request queue for the device |
| 754 | * @mask: the memory boundary mask |
| 755 | **/ |
| 756 | void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask) |
| 757 | { |
| 758 | q->limits.virt_boundary_mask = mask; |
| 759 | |
| 760 | /* |
| 761 | * Devices that require a virtual boundary do not support scatter/gather |
| 762 | * I/O natively, but instead require a descriptor list entry for each |
| 763 | * page (which might not be idential to the Linux PAGE_SIZE). Because |
| 764 | * of that they are not limited by our notion of "segment size". |
| 765 | */ |
| 766 | if (mask) |
| 767 | q->limits.max_segment_size = UINT_MAX; |
| 768 | } |
| 769 | EXPORT_SYMBOL(blk_queue_virt_boundary); |
| 770 | |
| 771 | /** |
| 772 | * blk_queue_dma_alignment - set dma length and memory alignment |
| 773 | * @q: the request queue for the device |
| 774 | * @mask: alignment mask |
| 775 | * |
| 776 | * description: |
| 777 | * set required memory and length alignment for direct dma transactions. |
| 778 | * this is used when building direct io requests for the queue. |
| 779 | * |
| 780 | **/ |
| 781 | void blk_queue_dma_alignment(struct request_queue *q, int mask) |
| 782 | { |
| 783 | q->limits.dma_alignment = mask; |
| 784 | } |
| 785 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
| 786 | |
| 787 | /** |
| 788 | * blk_queue_update_dma_alignment - update dma length and memory alignment |
| 789 | * @q: the request queue for the device |
| 790 | * @mask: alignment mask |
| 791 | * |
| 792 | * description: |
| 793 | * update required memory and length alignment for direct dma transactions. |
| 794 | * If the requested alignment is larger than the current alignment, then |
| 795 | * the current queue alignment is updated to the new value, otherwise it |
| 796 | * is left alone. The design of this is to allow multiple objects |
| 797 | * (driver, device, transport etc) to set their respective |
| 798 | * alignments without having them interfere. |
| 799 | * |
| 800 | **/ |
| 801 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) |
| 802 | { |
| 803 | BUG_ON(mask > PAGE_SIZE); |
| 804 | |
| 805 | if (mask > q->limits.dma_alignment) |
| 806 | q->limits.dma_alignment = mask; |
| 807 | } |
| 808 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
| 809 | |
| 810 | /** |
| 811 | * blk_set_queue_depth - tell the block layer about the device queue depth |
| 812 | * @q: the request queue for the device |
| 813 | * @depth: queue depth |
| 814 | * |
| 815 | */ |
| 816 | void blk_set_queue_depth(struct request_queue *q, unsigned int depth) |
| 817 | { |
| 818 | q->queue_depth = depth; |
| 819 | rq_qos_queue_depth_changed(q); |
| 820 | } |
| 821 | EXPORT_SYMBOL(blk_set_queue_depth); |
| 822 | |
| 823 | /** |
| 824 | * blk_queue_write_cache - configure queue's write cache |
| 825 | * @q: the request queue for the device |
| 826 | * @wc: write back cache on or off |
| 827 | * @fua: device supports FUA writes, if true |
| 828 | * |
| 829 | * Tell the block layer about the write cache of @q. |
| 830 | */ |
| 831 | void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua) |
| 832 | { |
| 833 | if (wc) |
| 834 | blk_queue_flag_set(QUEUE_FLAG_WC, q); |
| 835 | else |
| 836 | blk_queue_flag_clear(QUEUE_FLAG_WC, q); |
| 837 | if (fua) |
| 838 | blk_queue_flag_set(QUEUE_FLAG_FUA, q); |
| 839 | else |
| 840 | blk_queue_flag_clear(QUEUE_FLAG_FUA, q); |
| 841 | |
| 842 | wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags)); |
| 843 | } |
| 844 | EXPORT_SYMBOL_GPL(blk_queue_write_cache); |
| 845 | |
| 846 | /** |
| 847 | * blk_queue_required_elevator_features - Set a queue required elevator features |
| 848 | * @q: the request queue for the target device |
| 849 | * @features: Required elevator features OR'ed together |
| 850 | * |
| 851 | * Tell the block layer that for the device controlled through @q, only the |
| 852 | * only elevators that can be used are those that implement at least the set of |
| 853 | * features specified by @features. |
| 854 | */ |
| 855 | void blk_queue_required_elevator_features(struct request_queue *q, |
| 856 | unsigned int features) |
| 857 | { |
| 858 | q->required_elevator_features = features; |
| 859 | } |
| 860 | EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features); |
| 861 | |
| 862 | /** |
| 863 | * blk_queue_can_use_dma_map_merging - configure queue for merging segments. |
| 864 | * @q: the request queue for the device |
| 865 | * @dev: the device pointer for dma |
| 866 | * |
| 867 | * Tell the block layer about merging the segments by dma map of @q. |
| 868 | */ |
| 869 | bool blk_queue_can_use_dma_map_merging(struct request_queue *q, |
| 870 | struct device *dev) |
| 871 | { |
| 872 | unsigned long boundary = dma_get_merge_boundary(dev); |
| 873 | |
| 874 | if (!boundary) |
| 875 | return false; |
| 876 | |
| 877 | /* No need to update max_segment_size. see blk_queue_virt_boundary() */ |
| 878 | blk_queue_virt_boundary(q, boundary); |
| 879 | |
| 880 | return true; |
| 881 | } |
| 882 | EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging); |
| 883 | |
| 884 | static bool disk_has_partitions(struct gendisk *disk) |
| 885 | { |
| 886 | unsigned long idx; |
| 887 | struct block_device *part; |
| 888 | bool ret = false; |
| 889 | |
| 890 | rcu_read_lock(); |
| 891 | xa_for_each(&disk->part_tbl, idx, part) { |
| 892 | if (bdev_is_partition(part)) { |
| 893 | ret = true; |
| 894 | break; |
| 895 | } |
| 896 | } |
| 897 | rcu_read_unlock(); |
| 898 | |
| 899 | return ret; |
| 900 | } |
| 901 | |
| 902 | /** |
| 903 | * disk_set_zoned - configure the zoned model for a disk |
| 904 | * @disk: the gendisk of the queue to configure |
| 905 | * @model: the zoned model to set |
| 906 | * |
| 907 | * Set the zoned model of @disk to @model. |
| 908 | * |
| 909 | * When @model is BLK_ZONED_HM (host managed), this should be called only |
| 910 | * if zoned block device support is enabled (CONFIG_BLK_DEV_ZONED option). |
| 911 | * If @model specifies BLK_ZONED_HA (host aware), the effective model used |
| 912 | * depends on CONFIG_BLK_DEV_ZONED settings and on the existence of partitions |
| 913 | * on the disk. |
| 914 | */ |
| 915 | void disk_set_zoned(struct gendisk *disk, enum blk_zoned_model model) |
| 916 | { |
| 917 | struct request_queue *q = disk->queue; |
| 918 | |
| 919 | switch (model) { |
| 920 | case BLK_ZONED_HM: |
| 921 | /* |
| 922 | * Host managed devices are supported only if |
| 923 | * CONFIG_BLK_DEV_ZONED is enabled. |
| 924 | */ |
| 925 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)); |
| 926 | break; |
| 927 | case BLK_ZONED_HA: |
| 928 | /* |
| 929 | * Host aware devices can be treated either as regular block |
| 930 | * devices (similar to drive managed devices) or as zoned block |
| 931 | * devices to take advantage of the zone command set, similarly |
| 932 | * to host managed devices. We try the latter if there are no |
| 933 | * partitions and zoned block device support is enabled, else |
| 934 | * we do nothing special as far as the block layer is concerned. |
| 935 | */ |
| 936 | if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) || |
| 937 | disk_has_partitions(disk)) |
| 938 | model = BLK_ZONED_NONE; |
| 939 | break; |
| 940 | case BLK_ZONED_NONE: |
| 941 | default: |
| 942 | if (WARN_ON_ONCE(model != BLK_ZONED_NONE)) |
| 943 | model = BLK_ZONED_NONE; |
| 944 | break; |
| 945 | } |
| 946 | |
| 947 | q->limits.zoned = model; |
| 948 | if (model != BLK_ZONED_NONE) { |
| 949 | /* |
| 950 | * Set the zone write granularity to the device logical block |
| 951 | * size by default. The driver can change this value if needed. |
| 952 | */ |
| 953 | blk_queue_zone_write_granularity(q, |
| 954 | queue_logical_block_size(q)); |
| 955 | } else { |
| 956 | disk_clear_zone_settings(disk); |
| 957 | } |
| 958 | } |
| 959 | EXPORT_SYMBOL_GPL(disk_set_zoned); |
| 960 | |
| 961 | int bdev_alignment_offset(struct block_device *bdev) |
| 962 | { |
| 963 | struct request_queue *q = bdev_get_queue(bdev); |
| 964 | |
| 965 | if (q->limits.misaligned) |
| 966 | return -1; |
| 967 | if (bdev_is_partition(bdev)) |
| 968 | return queue_limit_alignment_offset(&q->limits, |
| 969 | bdev->bd_start_sect); |
| 970 | return q->limits.alignment_offset; |
| 971 | } |
| 972 | EXPORT_SYMBOL_GPL(bdev_alignment_offset); |
| 973 | |
| 974 | unsigned int bdev_discard_alignment(struct block_device *bdev) |
| 975 | { |
| 976 | struct request_queue *q = bdev_get_queue(bdev); |
| 977 | |
| 978 | if (bdev_is_partition(bdev)) |
| 979 | return queue_limit_discard_alignment(&q->limits, |
| 980 | bdev->bd_start_sect); |
| 981 | return q->limits.discard_alignment; |
| 982 | } |
| 983 | EXPORT_SYMBOL_GPL(bdev_discard_alignment); |