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