block/blk-settings.c

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