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
  11#include "blk.h"
  12
  13unsigned long blk_max_low_pfn;
  14EXPORT_SYMBOL(blk_max_low_pfn);
  15
  16unsigned long blk_max_pfn;
  17
  18/**
  19 * blk_queue_prep_rq - set a prepare_request function for queue
  20 * @q:          queue
  21 * @pfn:        prepare_request function
  22 *
  23 * It's possible for a queue to register a prepare_request callback which
  24 * is invoked before the request is handed to the request_fn. The goal of
  25 * the function is to prepare a request for I/O, it can be used to build a
  26 * cdb from the request data for instance.
  27 *
  28 */
  29void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
  30{
  31        q->prep_rq_fn = pfn;
  32}
  33EXPORT_SYMBOL(blk_queue_prep_rq);
  34
  35/**
  36 * blk_queue_set_discard - set a discard_sectors function for queue
  37 * @q:          queue
  38 * @dfn:        prepare_discard function
  39 *
  40 * It's possible for a queue to register a discard callback which is used
  41 * to transform a discard request into the appropriate type for the
  42 * hardware. If none is registered, then discard requests are failed
  43 * with %EOPNOTSUPP.
  44 *
  45 */
  46void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
  47{
  48        q->prepare_discard_fn = dfn;
  49}
  50EXPORT_SYMBOL(blk_queue_set_discard);
  51
  52/**
  53 * blk_queue_merge_bvec - set a merge_bvec function for queue
  54 * @q:          queue
  55 * @mbfn:       merge_bvec_fn
  56 *
  57 * Usually queues have static limitations on the max sectors or segments that
  58 * we can put in a request. Stacking drivers may have some settings that
  59 * are dynamic, and thus we have to query the queue whether it is ok to
  60 * add a new bio_vec to a bio at a given offset or not. If the block device
  61 * has such limitations, it needs to register a merge_bvec_fn to control
  62 * the size of bio's sent to it. Note that a block device *must* allow a
  63 * single page to be added to an empty bio. The block device driver may want
  64 * to use the bio_split() function to deal with these bio's. By default
  65 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
  66 * honored.
  67 */
  68void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
  69{
  70        q->merge_bvec_fn = mbfn;
  71}
  72EXPORT_SYMBOL(blk_queue_merge_bvec);
  73
  74void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
  75{
  76        q->softirq_done_fn = fn;
  77}
  78EXPORT_SYMBOL(blk_queue_softirq_done);
  79
  80void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
  81{
  82        q->rq_timeout = timeout;
  83}
  84EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
  85
  86void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
  87{
  88        q->rq_timed_out_fn = fn;
  89}
  90EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
  91
  92void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
  93{
  94        q->lld_busy_fn = fn;
  95}
  96EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
  97
  98/**
  99 * blk_queue_make_request - define an alternate make_request function for a device
 100 * @q:  the request queue for the device to be affected
 101 * @mfn: the alternate make_request function
 102 *
 103 * Description:
 104 *    The normal way for &struct bios to be passed to a device
 105 *    driver is for them to be collected into requests on a request
 106 *    queue, and then to allow the device driver to select requests
 107 *    off that queue when it is ready.  This works well for many block
 108 *    devices. However some block devices (typically virtual devices
 109 *    such as md or lvm) do not benefit from the processing on the
 110 *    request queue, and are served best by having the requests passed
 111 *    directly to them.  This can be achieved by providing a function
 112 *    to blk_queue_make_request().
 113 *
 114 * Caveat:
 115 *    The driver that does this *must* be able to deal appropriately
 116 *    with buffers in "highmemory". This can be accomplished by either calling
 117 *    __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
 118 *    blk_queue_bounce() to create a buffer in normal memory.
 119 **/
 120void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
 121{
 122        /*
 123         * set defaults
 124         */
 125        q->nr_requests = BLKDEV_MAX_RQ;
 126        blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
 127        blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
 128        blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK);
 129        blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
 130
 131        q->make_request_fn = mfn;
 132        q->backing_dev_info.ra_pages =
 133                        (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 134        q->backing_dev_info.state = 0;
 135        q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
 136        blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
 137        blk_queue_hardsect_size(q, 512);
 138        blk_queue_dma_alignment(q, 511);
 139        blk_queue_congestion_threshold(q);
 140        q->nr_batching = BLK_BATCH_REQ;
 141
 142        q->unplug_thresh = 4;           /* hmm */
 143        q->unplug_delay = (3 * HZ) / 1000;      /* 3 milliseconds */
 144        if (q->unplug_delay == 0)
 145                q->unplug_delay = 1;
 146
 147        q->unplug_timer.function = blk_unplug_timeout;
 148        q->unplug_timer.data = (unsigned long)q;
 149
 150        /*
 151         * by default assume old behaviour and bounce for any highmem page
 152         */
 153        blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
 154}
 155EXPORT_SYMBOL(blk_queue_make_request);
 156
 157/**
 158 * blk_queue_bounce_limit - set bounce buffer limit for queue
 159 * @q:  the request queue for the device
 160 * @dma_addr:   bus address limit
 161 *
 162 * Description:
 163 *    Different hardware can have different requirements as to what pages
 164 *    it can do I/O directly to. A low level driver can call
 165 *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
 166 *    buffers for doing I/O to pages residing above @dma_addr.
 167 **/
 168void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
 169{
 170        unsigned long b_pfn = dma_addr >> PAGE_SHIFT;
 171        int dma = 0;
 172
 173        q->bounce_gfp = GFP_NOIO;
 174#if BITS_PER_LONG == 64
 175        /* Assume anything <= 4GB can be handled by IOMMU.
 176           Actually some IOMMUs can handle everything, but I don't
 177           know of a way to test this here. */
 178        if (b_pfn < (min_t(u64, 0x100000000UL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
 179                dma = 1;
 180        q->bounce_pfn = max_low_pfn;
 181#else
 182        if (b_pfn < blk_max_low_pfn)
 183                dma = 1;
 184        q->bounce_pfn = b_pfn;
 185#endif
 186        if (dma) {
 187                init_emergency_isa_pool();
 188                q->bounce_gfp = GFP_NOIO | GFP_DMA;
 189                q->bounce_pfn = b_pfn;
 190        }
 191}
 192EXPORT_SYMBOL(blk_queue_bounce_limit);
 193
 194/**
 195 * blk_queue_max_sectors - set max sectors for a request for this queue
 196 * @q:  the request queue for the device
 197 * @max_sectors:  max sectors in the usual 512b unit
 198 *
 199 * Description:
 200 *    Enables a low level driver to set an upper limit on the size of
 201 *    received requests.
 202 **/
 203void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
 204{
 205        if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
 206                max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
 207                printk(KERN_INFO "%s: set to minimum %d\n",
 208                       __func__, max_sectors);
 209        }
 210
 211        if (BLK_DEF_MAX_SECTORS > max_sectors)
 212                q->max_hw_sectors = q->max_sectors = max_sectors;
 213        else {
 214                q->max_sectors = BLK_DEF_MAX_SECTORS;
 215                q->max_hw_sectors = max_sectors;
 216        }
 217}
 218EXPORT_SYMBOL(blk_queue_max_sectors);
 219
 220/**
 221 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
 222 * @q:  the request queue for the device
 223 * @max_segments:  max number of segments
 224 *
 225 * Description:
 226 *    Enables a low level driver to set an upper limit on the number of
 227 *    physical data segments in a request.  This would be the largest sized
 228 *    scatter list the driver could handle.
 229 **/
 230void blk_queue_max_phys_segments(struct request_queue *q,
 231                                 unsigned short max_segments)
 232{
 233        if (!max_segments) {
 234                max_segments = 1;
 235                printk(KERN_INFO "%s: set to minimum %d\n",
 236                       __func__, max_segments);
 237        }
 238
 239        q->max_phys_segments = max_segments;
 240}
 241EXPORT_SYMBOL(blk_queue_max_phys_segments);
 242
 243/**
 244 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
 245 * @q:  the request queue for the device
 246 * @max_segments:  max number of segments
 247 *
 248 * Description:
 249 *    Enables a low level driver to set an upper limit on the number of
 250 *    hw data segments in a request.  This would be the largest number of
 251 *    address/length pairs the host adapter can actually give at once
 252 *    to the device.
 253 **/
 254void blk_queue_max_hw_segments(struct request_queue *q,
 255                               unsigned short max_segments)
 256{
 257        if (!max_segments) {
 258                max_segments = 1;
 259                printk(KERN_INFO "%s: set to minimum %d\n",
 260                       __func__, max_segments);
 261        }
 262
 263        q->max_hw_segments = max_segments;
 264}
 265EXPORT_SYMBOL(blk_queue_max_hw_segments);
 266
 267/**
 268 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
 269 * @q:  the request queue for the device
 270 * @max_size:  max size of segment in bytes
 271 *
 272 * Description:
 273 *    Enables a low level driver to set an upper limit on the size of a
 274 *    coalesced segment
 275 **/
 276void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
 277{
 278        if (max_size < PAGE_CACHE_SIZE) {
 279                max_size = PAGE_CACHE_SIZE;
 280                printk(KERN_INFO "%s: set to minimum %d\n",
 281                       __func__, max_size);
 282        }
 283
 284        q->max_segment_size = max_size;
 285}
 286EXPORT_SYMBOL(blk_queue_max_segment_size);
 287
 288/**
 289 * blk_queue_hardsect_size - set hardware sector size for the queue
 290 * @q:  the request queue for the device
 291 * @size:  the hardware sector size, in bytes
 292 *
 293 * Description:
 294 *   This should typically be set to the lowest possible sector size
 295 *   that the hardware can operate on (possible without reverting to
 296 *   even internal read-modify-write operations). Usually the default
 297 *   of 512 covers most hardware.
 298 **/
 299void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
 300{
 301        q->hardsect_size = size;
 302}
 303EXPORT_SYMBOL(blk_queue_hardsect_size);
 304
 305/*
 306 * Returns the minimum that is _not_ zero, unless both are zero.
 307 */
 308#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
 309
 310/**
 311 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
 312 * @t:  the stacking driver (top)
 313 * @b:  the underlying device (bottom)
 314 **/
 315void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
 316{
 317        /* zero is "infinity" */
 318        t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
 319        t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
 320        t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, b->seg_boundary_mask);
 321
 322        t->max_phys_segments = min(t->max_phys_segments, b->max_phys_segments);
 323        t->max_hw_segments = min(t->max_hw_segments, b->max_hw_segments);
 324        t->max_segment_size = min(t->max_segment_size, b->max_segment_size);
 325        t->hardsect_size = max(t->hardsect_size, b->hardsect_size);
 326        if (!t->queue_lock)
 327                WARN_ON_ONCE(1);
 328        else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
 329                unsigned long flags;
 330                spin_lock_irqsave(t->queue_lock, flags);
 331                queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
 332                spin_unlock_irqrestore(t->queue_lock, flags);
 333        }
 334}
 335EXPORT_SYMBOL(blk_queue_stack_limits);
 336
 337/**
 338 * blk_queue_dma_pad - set pad mask
 339 * @q:     the request queue for the device
 340 * @mask:  pad mask
 341 *
 342 * Set dma pad mask.
 343 *
 344 * Appending pad buffer to a request modifies the last entry of a
 345 * scatter list such that it includes the pad buffer.
 346 **/
 347void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
 348{
 349        q->dma_pad_mask = mask;
 350}
 351EXPORT_SYMBOL(blk_queue_dma_pad);
 352
 353/**
 354 * blk_queue_update_dma_pad - update pad mask
 355 * @q:     the request queue for the device
 356 * @mask:  pad mask
 357 *
 358 * Update dma pad mask.
 359 *
 360 * Appending pad buffer to a request modifies the last entry of a
 361 * scatter list such that it includes the pad buffer.
 362 **/
 363void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
 364{
 365        if (mask > q->dma_pad_mask)
 366                q->dma_pad_mask = mask;
 367}
 368EXPORT_SYMBOL(blk_queue_update_dma_pad);
 369
 370/**
 371 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
 372 * @q:  the request queue for the device
 373 * @dma_drain_needed: fn which returns non-zero if drain is necessary
 374 * @buf:        physically contiguous buffer
 375 * @size:       size of the buffer in bytes
 376 *
 377 * Some devices have excess DMA problems and can't simply discard (or
 378 * zero fill) the unwanted piece of the transfer.  They have to have a
 379 * real area of memory to transfer it into.  The use case for this is
 380 * ATAPI devices in DMA mode.  If the packet command causes a transfer
 381 * bigger than the transfer size some HBAs will lock up if there
 382 * aren't DMA elements to contain the excess transfer.  What this API
 383 * does is adjust the queue so that the buf is always appended
 384 * silently to the scatterlist.
 385 *
 386 * Note: This routine adjusts max_hw_segments to make room for
 387 * appending the drain buffer.  If you call
 388 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
 389 * calling this routine, you must set the limit to one fewer than your
 390 * device can support otherwise there won't be room for the drain
 391 * buffer.
 392 */
 393int blk_queue_dma_drain(struct request_queue *q,
 394                               dma_drain_needed_fn *dma_drain_needed,
 395                               void *buf, unsigned int size)
 396{
 397        if (q->max_hw_segments < 2 || q->max_phys_segments < 2)
 398                return -EINVAL;
 399        /* make room for appending the drain */
 400        --q->max_hw_segments;
 401        --q->max_phys_segments;
 402        q->dma_drain_needed = dma_drain_needed;
 403        q->dma_drain_buffer = buf;
 404        q->dma_drain_size = size;
 405
 406        return 0;
 407}
 408EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
 409
 410/**
 411 * blk_queue_segment_boundary - set boundary rules for segment merging
 412 * @q:  the request queue for the device
 413 * @mask:  the memory boundary mask
 414 **/
 415void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
 416{
 417        if (mask < PAGE_CACHE_SIZE - 1) {
 418                mask = PAGE_CACHE_SIZE - 1;
 419                printk(KERN_INFO "%s: set to minimum %lx\n",
 420                       __func__, mask);
 421        }
 422
 423        q->seg_boundary_mask = mask;
 424}
 425EXPORT_SYMBOL(blk_queue_segment_boundary);
 426
 427/**
 428 * blk_queue_dma_alignment - set dma length and memory alignment
 429 * @q:     the request queue for the device
 430 * @mask:  alignment mask
 431 *
 432 * description:
 433 *    set required memory and length alignment for direct dma transactions.
 434 *    this is used when buiding direct io requests for the queue.
 435 *
 436 **/
 437void blk_queue_dma_alignment(struct request_queue *q, int mask)
 438{
 439        q->dma_alignment = mask;
 440}
 441EXPORT_SYMBOL(blk_queue_dma_alignment);
 442
 443/**
 444 * blk_queue_update_dma_alignment - update dma length and memory alignment
 445 * @q:     the request queue for the device
 446 * @mask:  alignment mask
 447 *
 448 * description:
 449 *    update required memory and length alignment for direct dma transactions.
 450 *    If the requested alignment is larger than the current alignment, then
 451 *    the current queue alignment is updated to the new value, otherwise it
 452 *    is left alone.  The design of this is to allow multiple objects
 453 *    (driver, device, transport etc) to set their respective
 454 *    alignments without having them interfere.
 455 *
 456 **/
 457void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
 458{
 459        BUG_ON(mask > PAGE_SIZE);
 460
 461        if (mask > q->dma_alignment)
 462                q->dma_alignment = mask;
 463}
 464EXPORT_SYMBOL(blk_queue_update_dma_alignment);
 465
 466static int __init blk_settings_init(void)
 467{
 468        blk_max_low_pfn = max_low_pfn - 1;
 469        blk_max_pfn = max_pfn - 1;
 470        return 0;
 471}
 472subsys_initcall(blk_settings_init);
 473
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.