Linux Audio

Check our new training course

Loading...
v3.5.6
  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.
108 */
109void blk_set_default_limits(struct queue_limits *lim)
110{
111	lim->max_segments = BLK_MAX_SEGMENTS;
 
112	lim->max_integrity_segments = 0;
113	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
 
114	lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
115	lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
 
 
 
 
116	lim->max_discard_sectors = 0;
 
117	lim->discard_granularity = 0;
118	lim->discard_alignment = 0;
119	lim->discard_misaligned = 0;
120	lim->discard_zeroes_data = 0;
121	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
122	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
123	lim->alignment_offset = 0;
124	lim->io_opt = 0;
125	lim->misaligned = 0;
126	lim->cluster = 1;
 
127}
128EXPORT_SYMBOL(blk_set_default_limits);
129
130/**
131 * blk_set_stacking_limits - set default limits for stacking devices
132 * @lim:  the queue_limits structure to reset
133 *
134 * Description:
135 *   Returns a queue_limit struct to its default state. Should be used
136 *   by stacking drivers like DM that have no internal limits.
137 */
138void blk_set_stacking_limits(struct queue_limits *lim)
139{
140	blk_set_default_limits(lim);
141
142	/* Inherit limits from component devices */
143	lim->discard_zeroes_data = 1;
144	lim->max_segments = USHRT_MAX;
 
145	lim->max_hw_sectors = UINT_MAX;
146
147	lim->max_sectors = BLK_DEF_MAX_SECTORS;
 
 
 
148}
149EXPORT_SYMBOL(blk_set_stacking_limits);
150
151/**
152 * blk_queue_make_request - define an alternate make_request function for a device
153 * @q:  the request queue for the device to be affected
154 * @mfn: the alternate make_request function
155 *
156 * Description:
157 *    The normal way for &struct bios to be passed to a device
158 *    driver is for them to be collected into requests on a request
159 *    queue, and then to allow the device driver to select requests
160 *    off that queue when it is ready.  This works well for many block
161 *    devices. However some block devices (typically virtual devices
162 *    such as md or lvm) do not benefit from the processing on the
163 *    request queue, and are served best by having the requests passed
164 *    directly to them.  This can be achieved by providing a function
165 *    to blk_queue_make_request().
166 *
167 * Caveat:
168 *    The driver that does this *must* be able to deal appropriately
169 *    with buffers in "highmemory". This can be accomplished by either calling
170 *    __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
171 *    blk_queue_bounce() to create a buffer in normal memory.
172 **/
173void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
174{
175	/*
176	 * set defaults
177	 */
178	q->nr_requests = BLKDEV_MAX_RQ;
179
180	q->make_request_fn = mfn;
181	blk_queue_dma_alignment(q, 511);
182	blk_queue_congestion_threshold(q);
183	q->nr_batching = BLK_BATCH_REQ;
184
185	blk_set_default_limits(&q->limits);
186
187	/*
188	 * by default assume old behaviour and bounce for any highmem page
189	 */
190	blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
191}
192EXPORT_SYMBOL(blk_queue_make_request);
193
194/**
195 * blk_queue_bounce_limit - set bounce buffer limit for queue
196 * @q: the request queue for the device
197 * @dma_mask: the maximum address the device can handle
198 *
199 * Description:
200 *    Different hardware can have different requirements as to what pages
201 *    it can do I/O directly to. A low level driver can call
202 *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
203 *    buffers for doing I/O to pages residing above @dma_mask.
204 **/
205void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
206{
207	unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
208	int dma = 0;
209
210	q->bounce_gfp = GFP_NOIO;
211#if BITS_PER_LONG == 64
212	/*
213	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
214	 * some IOMMUs can handle everything, but I don't know of a
215	 * way to test this here.
216	 */
217	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
218		dma = 1;
219	q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
220#else
221	if (b_pfn < blk_max_low_pfn)
222		dma = 1;
223	q->limits.bounce_pfn = b_pfn;
224#endif
225	if (dma) {
226		init_emergency_isa_pool();
227		q->bounce_gfp = GFP_NOIO | GFP_DMA;
228		q->limits.bounce_pfn = b_pfn;
229	}
230}
231EXPORT_SYMBOL(blk_queue_bounce_limit);
232
233/**
234 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
235 * @limits: the queue limits
236 * @max_hw_sectors:  max hardware sectors in the usual 512b unit
237 *
238 * Description:
239 *    Enables a low level driver to set a hard upper limit,
240 *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
241 *    the device driver based upon the combined capabilities of I/O
242 *    controller and storage device.
 
 
 
243 *
244 *    max_sectors is a soft limit imposed by the block layer for
245 *    filesystem type requests.  This value can be overridden on a
246 *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
247 *    The soft limit can not exceed max_hw_sectors.
248 **/
249void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
250{
251	if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
252		max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
 
 
 
253		printk(KERN_INFO "%s: set to minimum %d\n",
254		       __func__, max_hw_sectors);
255	}
256
257	limits->max_hw_sectors = max_hw_sectors;
258	limits->max_sectors = min_t(unsigned int, max_hw_sectors,
259				    BLK_DEF_MAX_SECTORS);
 
 
260}
261EXPORT_SYMBOL(blk_limits_max_hw_sectors);
262
263/**
264 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
265 * @q:  the request queue for the device
266 * @max_hw_sectors:  max hardware sectors in the usual 512b unit
267 *
268 * Description:
269 *    See description for blk_limits_max_hw_sectors().
 
 
 
 
 
270 **/
271void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
272{
273	blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
 
274}
275EXPORT_SYMBOL(blk_queue_max_hw_sectors);
276
277/**
278 * blk_queue_max_discard_sectors - set max sectors for a single discard
279 * @q:  the request queue for the device
280 * @max_discard_sectors: maximum number of sectors to discard
281 **/
282void blk_queue_max_discard_sectors(struct request_queue *q,
283		unsigned int max_discard_sectors)
284{
 
285	q->limits.max_discard_sectors = max_discard_sectors;
286}
287EXPORT_SYMBOL(blk_queue_max_discard_sectors);
288
289/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
290 * blk_queue_max_segments - set max hw segments for a request for this queue
291 * @q:  the request queue for the device
292 * @max_segments:  max number of segments
293 *
294 * Description:
295 *    Enables a low level driver to set an upper limit on the number of
296 *    hw data segments in a request.
297 **/
298void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
299{
300	if (!max_segments) {
301		max_segments = 1;
302		printk(KERN_INFO "%s: set to minimum %d\n",
303		       __func__, max_segments);
304	}
305
306	q->limits.max_segments = max_segments;
307}
308EXPORT_SYMBOL(blk_queue_max_segments);
309
310/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
311 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
312 * @q:  the request queue for the device
313 * @max_size:  max size of segment in bytes
314 *
315 * Description:
316 *    Enables a low level driver to set an upper limit on the size of a
317 *    coalesced segment
318 **/
319void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
320{
321	if (max_size < PAGE_CACHE_SIZE) {
322		max_size = PAGE_CACHE_SIZE;
323		printk(KERN_INFO "%s: set to minimum %d\n",
324		       __func__, max_size);
325	}
326
327	q->limits.max_segment_size = max_size;
328}
329EXPORT_SYMBOL(blk_queue_max_segment_size);
330
331/**
332 * blk_queue_logical_block_size - set logical block size for the queue
333 * @q:  the request queue for the device
334 * @size:  the logical block size, in bytes
335 *
336 * Description:
337 *   This should be set to the lowest possible block size that the
338 *   storage device can address.  The default of 512 covers most
339 *   hardware.
340 **/
341void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
342{
343	q->limits.logical_block_size = size;
344
345	if (q->limits.physical_block_size < size)
346		q->limits.physical_block_size = size;
347
348	if (q->limits.io_min < q->limits.physical_block_size)
349		q->limits.io_min = q->limits.physical_block_size;
350}
351EXPORT_SYMBOL(blk_queue_logical_block_size);
352
353/**
354 * blk_queue_physical_block_size - set physical block size for the queue
355 * @q:  the request queue for the device
356 * @size:  the physical block size, in bytes
357 *
358 * Description:
359 *   This should be set to the lowest possible sector size that the
360 *   hardware can operate on without reverting to read-modify-write
361 *   operations.
362 */
363void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
364{
365	q->limits.physical_block_size = size;
366
367	if (q->limits.physical_block_size < q->limits.logical_block_size)
368		q->limits.physical_block_size = q->limits.logical_block_size;
369
370	if (q->limits.io_min < q->limits.physical_block_size)
371		q->limits.io_min = q->limits.physical_block_size;
372}
373EXPORT_SYMBOL(blk_queue_physical_block_size);
374
375/**
376 * blk_queue_alignment_offset - set physical block alignment offset
377 * @q:	the request queue for the device
378 * @offset: alignment offset in bytes
379 *
380 * Description:
381 *   Some devices are naturally misaligned to compensate for things like
382 *   the legacy DOS partition table 63-sector offset.  Low-level drivers
383 *   should call this function for devices whose first sector is not
384 *   naturally aligned.
385 */
386void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
387{
388	q->limits.alignment_offset =
389		offset & (q->limits.physical_block_size - 1);
390	q->limits.misaligned = 0;
391}
392EXPORT_SYMBOL(blk_queue_alignment_offset);
393
394/**
395 * blk_limits_io_min - set minimum request size for a device
396 * @limits: the queue limits
397 * @min:  smallest I/O size in bytes
398 *
399 * Description:
400 *   Some devices have an internal block size bigger than the reported
401 *   hardware sector size.  This function can be used to signal the
402 *   smallest I/O the device can perform without incurring a performance
403 *   penalty.
404 */
405void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
406{
407	limits->io_min = min;
408
409	if (limits->io_min < limits->logical_block_size)
410		limits->io_min = limits->logical_block_size;
411
412	if (limits->io_min < limits->physical_block_size)
413		limits->io_min = limits->physical_block_size;
414}
415EXPORT_SYMBOL(blk_limits_io_min);
416
417/**
418 * blk_queue_io_min - set minimum request size for the queue
419 * @q:	the request queue for the device
420 * @min:  smallest I/O size in bytes
421 *
422 * Description:
423 *   Storage devices may report a granularity or preferred minimum I/O
424 *   size which is the smallest request the device can perform without
425 *   incurring a performance penalty.  For disk drives this is often the
426 *   physical block size.  For RAID arrays it is often the stripe chunk
427 *   size.  A properly aligned multiple of minimum_io_size is the
428 *   preferred request size for workloads where a high number of I/O
429 *   operations is desired.
430 */
431void blk_queue_io_min(struct request_queue *q, unsigned int min)
432{
433	blk_limits_io_min(&q->limits, min);
434}
435EXPORT_SYMBOL(blk_queue_io_min);
436
437/**
438 * blk_limits_io_opt - set optimal request size for a device
439 * @limits: the queue limits
440 * @opt:  smallest I/O size in bytes
441 *
442 * Description:
443 *   Storage devices may report an optimal I/O size, which is the
444 *   device's preferred unit for sustained I/O.  This is rarely reported
445 *   for disk drives.  For RAID arrays it is usually the stripe width or
446 *   the internal track size.  A properly aligned multiple of
447 *   optimal_io_size is the preferred request size for workloads where
448 *   sustained throughput is desired.
449 */
450void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
451{
452	limits->io_opt = opt;
453}
454EXPORT_SYMBOL(blk_limits_io_opt);
455
456/**
457 * blk_queue_io_opt - set optimal request size for the queue
458 * @q:	the request queue for the device
459 * @opt:  optimal request size in bytes
460 *
461 * Description:
462 *   Storage devices may report an optimal I/O size, which is the
463 *   device's preferred unit for sustained I/O.  This is rarely reported
464 *   for disk drives.  For RAID arrays it is usually the stripe width or
465 *   the internal track size.  A properly aligned multiple of
466 *   optimal_io_size is the preferred request size for workloads where
467 *   sustained throughput is desired.
468 */
469void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
470{
471	blk_limits_io_opt(&q->limits, opt);
472}
473EXPORT_SYMBOL(blk_queue_io_opt);
474
475/**
476 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
477 * @t:	the stacking driver (top)
478 * @b:  the underlying device (bottom)
479 **/
480void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
481{
482	blk_stack_limits(&t->limits, &b->limits, 0);
483}
484EXPORT_SYMBOL(blk_queue_stack_limits);
485
486/**
487 * blk_stack_limits - adjust queue_limits for stacked devices
488 * @t:	the stacking driver limits (top device)
489 * @b:  the underlying queue limits (bottom, component device)
490 * @start:  first data sector within component device
491 *
492 * Description:
493 *    This function is used by stacking drivers like MD and DM to ensure
494 *    that all component devices have compatible block sizes and
495 *    alignments.  The stacking driver must provide a queue_limits
496 *    struct (top) and then iteratively call the stacking function for
497 *    all component (bottom) devices.  The stacking function will
498 *    attempt to combine the values and ensure proper alignment.
499 *
500 *    Returns 0 if the top and bottom queue_limits are compatible.  The
501 *    top device's block sizes and alignment offsets may be adjusted to
502 *    ensure alignment with the bottom device. If no compatible sizes
503 *    and alignments exist, -1 is returned and the resulting top
504 *    queue_limits will have the misaligned flag set to indicate that
505 *    the alignment_offset is undefined.
506 */
507int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
508		     sector_t start)
509{
510	unsigned int top, bottom, alignment, ret = 0;
511
512	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
513	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
 
 
 
 
 
514	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
515
516	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
517					    b->seg_boundary_mask);
 
 
518
519	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
 
 
520	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
521						 b->max_integrity_segments);
522
523	t->max_segment_size = min_not_zero(t->max_segment_size,
524					   b->max_segment_size);
525
526	t->misaligned |= b->misaligned;
527
528	alignment = queue_limit_alignment_offset(b, start);
529
530	/* Bottom device has different alignment.  Check that it is
531	 * compatible with the current top alignment.
532	 */
533	if (t->alignment_offset != alignment) {
534
535		top = max(t->physical_block_size, t->io_min)
536			+ t->alignment_offset;
537		bottom = max(b->physical_block_size, b->io_min) + alignment;
538
539		/* Verify that top and bottom intervals line up */
540		if (max(top, bottom) & (min(top, bottom) - 1)) {
541			t->misaligned = 1;
542			ret = -1;
543		}
544	}
545
546	t->logical_block_size = max(t->logical_block_size,
547				    b->logical_block_size);
548
549	t->physical_block_size = max(t->physical_block_size,
550				     b->physical_block_size);
551
552	t->io_min = max(t->io_min, b->io_min);
553	t->io_opt = lcm(t->io_opt, b->io_opt);
554
555	t->cluster &= b->cluster;
556	t->discard_zeroes_data &= b->discard_zeroes_data;
557
558	/* Physical block size a multiple of the logical block size? */
559	if (t->physical_block_size & (t->logical_block_size - 1)) {
560		t->physical_block_size = t->logical_block_size;
561		t->misaligned = 1;
562		ret = -1;
563	}
564
565	/* Minimum I/O a multiple of the physical block size? */
566	if (t->io_min & (t->physical_block_size - 1)) {
567		t->io_min = t->physical_block_size;
568		t->misaligned = 1;
569		ret = -1;
570	}
571
572	/* Optimal I/O a multiple of the physical block size? */
573	if (t->io_opt & (t->physical_block_size - 1)) {
574		t->io_opt = 0;
575		t->misaligned = 1;
576		ret = -1;
577	}
578
 
 
 
 
579	/* Find lowest common alignment_offset */
580	t->alignment_offset = lcm(t->alignment_offset, alignment)
581		& (max(t->physical_block_size, t->io_min) - 1);
582
583	/* Verify that new alignment_offset is on a logical block boundary */
584	if (t->alignment_offset & (t->logical_block_size - 1)) {
585		t->misaligned = 1;
586		ret = -1;
587	}
588
589	/* Discard alignment and granularity */
590	if (b->discard_granularity) {
591		alignment = queue_limit_discard_alignment(b, start);
592
593		if (t->discard_granularity != 0 &&
594		    t->discard_alignment != alignment) {
595			top = t->discard_granularity + t->discard_alignment;
596			bottom = b->discard_granularity + alignment;
597
598			/* Verify that top and bottom intervals line up */
599			if (max(top, bottom) & (min(top, bottom) - 1))
600				t->discard_misaligned = 1;
601		}
602
603		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
604						      b->max_discard_sectors);
 
 
605		t->discard_granularity = max(t->discard_granularity,
606					     b->discard_granularity);
607		t->discard_alignment = lcm(t->discard_alignment, alignment) &
608			(t->discard_granularity - 1);
609	}
610
 
 
 
 
611	return ret;
612}
613EXPORT_SYMBOL(blk_stack_limits);
614
615/**
616 * bdev_stack_limits - adjust queue limits for stacked drivers
617 * @t:	the stacking driver limits (top device)
618 * @bdev:  the component block_device (bottom)
619 * @start:  first data sector within component device
620 *
621 * Description:
622 *    Merges queue limits for a top device and a block_device.  Returns
623 *    0 if alignment didn't change.  Returns -1 if adding the bottom
624 *    device caused misalignment.
625 */
626int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
627		      sector_t start)
628{
629	struct request_queue *bq = bdev_get_queue(bdev);
630
631	start += get_start_sect(bdev);
632
633	return blk_stack_limits(t, &bq->limits, start);
634}
635EXPORT_SYMBOL(bdev_stack_limits);
636
637/**
638 * disk_stack_limits - adjust queue limits for stacked drivers
639 * @disk:  MD/DM gendisk (top)
640 * @bdev:  the underlying block device (bottom)
641 * @offset:  offset to beginning of data within component device
642 *
643 * Description:
644 *    Merges the limits for a top level gendisk and a bottom level
645 *    block_device.
646 */
647void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
648		       sector_t offset)
649{
650	struct request_queue *t = disk->queue;
651
652	if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
653		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
654
655		disk_name(disk, 0, top);
656		bdevname(bdev, bottom);
657
658		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
659		       top, bottom);
660	}
661}
662EXPORT_SYMBOL(disk_stack_limits);
663
664/**
665 * blk_queue_dma_pad - set pad mask
666 * @q:     the request queue for the device
667 * @mask:  pad mask
668 *
669 * Set 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_dma_pad(struct request_queue *q, unsigned int mask)
675{
676	q->dma_pad_mask = mask;
677}
678EXPORT_SYMBOL(blk_queue_dma_pad);
679
680/**
681 * blk_queue_update_dma_pad - update pad mask
682 * @q:     the request queue for the device
683 * @mask:  pad mask
684 *
685 * Update dma pad mask.
686 *
687 * Appending pad buffer to a request modifies the last entry of a
688 * scatter list such that it includes the pad buffer.
689 **/
690void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
691{
692	if (mask > q->dma_pad_mask)
693		q->dma_pad_mask = mask;
694}
695EXPORT_SYMBOL(blk_queue_update_dma_pad);
696
697/**
698 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
699 * @q:  the request queue for the device
700 * @dma_drain_needed: fn which returns non-zero if drain is necessary
701 * @buf:	physically contiguous buffer
702 * @size:	size of the buffer in bytes
703 *
704 * Some devices have excess DMA problems and can't simply discard (or
705 * zero fill) the unwanted piece of the transfer.  They have to have a
706 * real area of memory to transfer it into.  The use case for this is
707 * ATAPI devices in DMA mode.  If the packet command causes a transfer
708 * bigger than the transfer size some HBAs will lock up if there
709 * aren't DMA elements to contain the excess transfer.  What this API
710 * does is adjust the queue so that the buf is always appended
711 * silently to the scatterlist.
712 *
713 * Note: This routine adjusts max_hw_segments to make room for appending
714 * the drain buffer.  If you call blk_queue_max_segments() after calling
715 * this routine, you must set the limit to one fewer than your device
716 * can support otherwise there won't be room for the drain buffer.
717 */
718int blk_queue_dma_drain(struct request_queue *q,
719			       dma_drain_needed_fn *dma_drain_needed,
720			       void *buf, unsigned int size)
721{
722	if (queue_max_segments(q) < 2)
723		return -EINVAL;
724	/* make room for appending the drain */
725	blk_queue_max_segments(q, queue_max_segments(q) - 1);
726	q->dma_drain_needed = dma_drain_needed;
727	q->dma_drain_buffer = buf;
728	q->dma_drain_size = size;
729
730	return 0;
731}
732EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
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 **/
739void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
740{
741	if (mask < PAGE_CACHE_SIZE - 1) {
742		mask = PAGE_CACHE_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}
749EXPORT_SYMBOL(blk_queue_segment_boundary);
750
751/**
 
 
 
 
 
 
 
 
 
 
 
752 * blk_queue_dma_alignment - set dma length and memory alignment
753 * @q:     the request queue for the device
754 * @mask:  alignment mask
755 *
756 * description:
757 *    set required memory and length alignment for direct dma transactions.
758 *    this is used when building direct io requests for the queue.
759 *
760 **/
761void blk_queue_dma_alignment(struct request_queue *q, int mask)
762{
763	q->dma_alignment = mask;
764}
765EXPORT_SYMBOL(blk_queue_dma_alignment);
766
767/**
768 * blk_queue_update_dma_alignment - update dma length and memory alignment
769 * @q:     the request queue for the device
770 * @mask:  alignment mask
771 *
772 * description:
773 *    update required memory and length alignment for direct dma transactions.
774 *    If the requested alignment is larger than the current alignment, then
775 *    the current queue alignment is updated to the new value, otherwise it
776 *    is left alone.  The design of this is to allow multiple objects
777 *    (driver, device, transport etc) to set their respective
778 *    alignments without having them interfere.
779 *
780 **/
781void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
782{
783	BUG_ON(mask > PAGE_SIZE);
784
785	if (mask > q->dma_alignment)
786		q->dma_alignment = mask;
787}
788EXPORT_SYMBOL(blk_queue_update_dma_alignment);
789
 
 
 
 
 
 
 
 
 
790/**
791 * blk_queue_flush - configure queue's cache flush capability
792 * @q:		the request queue for the device
793 * @flush:	0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
794 *
795 * Tell block layer cache flush capability of @q.  If it supports
796 * flushing, REQ_FLUSH should be set.  If it supports bypassing
797 * write cache for individual writes, REQ_FUA should be set.
798 */
799void blk_queue_flush(struct request_queue *q, unsigned int flush)
800{
801	WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
802
803	if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
804		flush &= ~REQ_FUA;
805
806	q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
807}
808EXPORT_SYMBOL_GPL(blk_queue_flush);
809
810void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
 
 
 
 
 
 
 
 
811{
812	q->flush_not_queueable = !queueable;
 
 
 
 
 
 
 
 
 
 
 
813}
814EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
815
816static int __init blk_settings_init(void)
817{
818	blk_max_low_pfn = max_low_pfn - 1;
819	blk_max_pfn = max_pfn - 1;
820	return 0;
821}
822subsys_initcall(blk_settings_init);
v4.17
  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#include "blk-wbt.h"
 17
 18unsigned long blk_max_low_pfn;
 19EXPORT_SYMBOL(blk_max_low_pfn);
 20
 21unsigned long blk_max_pfn;
 22
 23/**
 24 * blk_queue_prep_rq - set a prepare_request function for queue
 25 * @q:		queue
 26 * @pfn:	prepare_request function
 27 *
 28 * It's possible for a queue to register a prepare_request callback which
 29 * is invoked before the request is handed to the request_fn. The goal of
 30 * the function is to prepare a request for I/O, it can be used to build a
 31 * cdb from the request data for instance.
 32 *
 33 */
 34void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
 35{
 36	q->prep_rq_fn = pfn;
 37}
 38EXPORT_SYMBOL(blk_queue_prep_rq);
 39
 40/**
 41 * blk_queue_unprep_rq - set an unprepare_request function for queue
 42 * @q:		queue
 43 * @ufn:	unprepare_request function
 44 *
 45 * It's possible for a queue to register an unprepare_request callback
 46 * which is invoked before the request is finally completed. The goal
 47 * of the function is to deallocate any data that was allocated in the
 48 * prepare_request callback.
 49 *
 50 */
 51void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
 52{
 53	q->unprep_rq_fn = ufn;
 54}
 55EXPORT_SYMBOL(blk_queue_unprep_rq);
 56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 57void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
 58{
 59	q->softirq_done_fn = fn;
 60}
 61EXPORT_SYMBOL(blk_queue_softirq_done);
 62
 63void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
 64{
 65	q->rq_timeout = timeout;
 66}
 67EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
 68
 69void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
 70{
 71	WARN_ON_ONCE(q->mq_ops);
 72	q->rq_timed_out_fn = fn;
 73}
 74EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
 75
 76void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
 77{
 78	q->lld_busy_fn = fn;
 79}
 80EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
 81
 82/**
 83 * blk_set_default_limits - reset limits to default values
 84 * @lim:  the queue_limits structure to reset
 85 *
 86 * Description:
 87 *   Returns a queue_limit struct to its default state.
 88 */
 89void blk_set_default_limits(struct queue_limits *lim)
 90{
 91	lim->max_segments = BLK_MAX_SEGMENTS;
 92	lim->max_discard_segments = 1;
 93	lim->max_integrity_segments = 0;
 94	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
 95	lim->virt_boundary_mask = 0;
 96	lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
 97	lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
 98	lim->max_dev_sectors = 0;
 99	lim->chunk_sectors = 0;
100	lim->max_write_same_sectors = 0;
101	lim->max_write_zeroes_sectors = 0;
102	lim->max_discard_sectors = 0;
103	lim->max_hw_discard_sectors = 0;
104	lim->discard_granularity = 0;
105	lim->discard_alignment = 0;
106	lim->discard_misaligned = 0;
 
107	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
108	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
109	lim->alignment_offset = 0;
110	lim->io_opt = 0;
111	lim->misaligned = 0;
112	lim->cluster = 1;
113	lim->zoned = BLK_ZONED_NONE;
114}
115EXPORT_SYMBOL(blk_set_default_limits);
116
117/**
118 * blk_set_stacking_limits - set default limits for stacking devices
119 * @lim:  the queue_limits structure to reset
120 *
121 * Description:
122 *   Returns a queue_limit struct to its default state. Should be used
123 *   by stacking drivers like DM that have no internal limits.
124 */
125void blk_set_stacking_limits(struct queue_limits *lim)
126{
127	blk_set_default_limits(lim);
128
129	/* Inherit limits from component devices */
 
130	lim->max_segments = USHRT_MAX;
131	lim->max_discard_segments = 1;
132	lim->max_hw_sectors = UINT_MAX;
133	lim->max_segment_size = UINT_MAX;
134	lim->max_sectors = UINT_MAX;
135	lim->max_dev_sectors = UINT_MAX;
136	lim->max_write_same_sectors = UINT_MAX;
137	lim->max_write_zeroes_sectors = UINT_MAX;
138}
139EXPORT_SYMBOL(blk_set_stacking_limits);
140
141/**
142 * blk_queue_make_request - define an alternate make_request function for a device
143 * @q:  the request queue for the device to be affected
144 * @mfn: the alternate make_request function
145 *
146 * Description:
147 *    The normal way for &struct bios to be passed to a device
148 *    driver is for them to be collected into requests on a request
149 *    queue, and then to allow the device driver to select requests
150 *    off that queue when it is ready.  This works well for many block
151 *    devices. However some block devices (typically virtual devices
152 *    such as md or lvm) do not benefit from the processing on the
153 *    request queue, and are served best by having the requests passed
154 *    directly to them.  This can be achieved by providing a function
155 *    to blk_queue_make_request().
156 *
157 * Caveat:
158 *    The driver that does this *must* be able to deal appropriately
159 *    with buffers in "highmemory". This can be accomplished by either calling
160 *    kmap_atomic() to get a temporary kernel mapping, or by calling
161 *    blk_queue_bounce() to create a buffer in normal memory.
162 **/
163void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
164{
165	/*
166	 * set defaults
167	 */
168	q->nr_requests = BLKDEV_MAX_RQ;
169
170	q->make_request_fn = mfn;
171	blk_queue_dma_alignment(q, 511);
172	blk_queue_congestion_threshold(q);
173	q->nr_batching = BLK_BATCH_REQ;
174
175	blk_set_default_limits(&q->limits);
 
 
 
 
 
176}
177EXPORT_SYMBOL(blk_queue_make_request);
178
179/**
180 * blk_queue_bounce_limit - set bounce buffer limit for queue
181 * @q: the request queue for the device
182 * @max_addr: the maximum address the device can handle
183 *
184 * Description:
185 *    Different hardware can have different requirements as to what pages
186 *    it can do I/O directly to. A low level driver can call
187 *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
188 *    buffers for doing I/O to pages residing above @max_addr.
189 **/
190void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
191{
192	unsigned long b_pfn = max_addr >> PAGE_SHIFT;
193	int dma = 0;
194
195	q->bounce_gfp = GFP_NOIO;
196#if BITS_PER_LONG == 64
197	/*
198	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
199	 * some IOMMUs can handle everything, but I don't know of a
200	 * way to test this here.
201	 */
202	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
203		dma = 1;
204	q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
205#else
206	if (b_pfn < blk_max_low_pfn)
207		dma = 1;
208	q->limits.bounce_pfn = b_pfn;
209#endif
210	if (dma) {
211		init_emergency_isa_pool();
212		q->bounce_gfp = GFP_NOIO | GFP_DMA;
213		q->limits.bounce_pfn = b_pfn;
214	}
215}
216EXPORT_SYMBOL(blk_queue_bounce_limit);
217
218/**
219 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
220 * @q:  the request queue for the device
221 * @max_hw_sectors:  max hardware sectors in the usual 512b unit
222 *
223 * Description:
224 *    Enables a low level driver to set a hard upper limit,
225 *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
226 *    the device driver based upon the capabilities of the I/O
227 *    controller.
228 *
229 *    max_dev_sectors is a hard limit imposed by the storage device for
230 *    READ/WRITE requests. It is set by the disk driver.
231 *
232 *    max_sectors is a soft limit imposed by the block layer for
233 *    filesystem type requests.  This value can be overridden on a
234 *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
235 *    The soft limit can not exceed max_hw_sectors.
236 **/
237void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
238{
239	struct queue_limits *limits = &q->limits;
240	unsigned int max_sectors;
241
242	if ((max_hw_sectors << 9) < PAGE_SIZE) {
243		max_hw_sectors = 1 << (PAGE_SHIFT - 9);
244		printk(KERN_INFO "%s: set to minimum %d\n",
245		       __func__, max_hw_sectors);
246	}
247
248	limits->max_hw_sectors = max_hw_sectors;
249	max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
250	max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
251	limits->max_sectors = max_sectors;
252	q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
253}
254EXPORT_SYMBOL(blk_queue_max_hw_sectors);
255
256/**
257 * blk_queue_chunk_sectors - set size of the chunk for this queue
258 * @q:  the request queue for the device
259 * @chunk_sectors:  chunk sectors in the usual 512b unit
260 *
261 * Description:
262 *    If a driver doesn't want IOs to cross a given chunk size, it can set
263 *    this limit and prevent merging across chunks. Note that the chunk size
264 *    must currently be a power-of-2 in sectors. Also note that the block
265 *    layer must accept a page worth of data at any offset. So if the
266 *    crossing of chunks is a hard limitation in the driver, it must still be
267 *    prepared to split single page bios.
268 **/
269void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
270{
271	BUG_ON(!is_power_of_2(chunk_sectors));
272	q->limits.chunk_sectors = chunk_sectors;
273}
274EXPORT_SYMBOL(blk_queue_chunk_sectors);
275
276/**
277 * blk_queue_max_discard_sectors - set max sectors for a single discard
278 * @q:  the request queue for the device
279 * @max_discard_sectors: maximum number of sectors to discard
280 **/
281void blk_queue_max_discard_sectors(struct request_queue *q,
282		unsigned int max_discard_sectors)
283{
284	q->limits.max_hw_discard_sectors = max_discard_sectors;
285	q->limits.max_discard_sectors = max_discard_sectors;
286}
287EXPORT_SYMBOL(blk_queue_max_discard_sectors);
288
289/**
290 * blk_queue_max_write_same_sectors - set max sectors for a single write same
291 * @q:  the request queue for the device
292 * @max_write_same_sectors: maximum number of sectors to write per command
293 **/
294void blk_queue_max_write_same_sectors(struct request_queue *q,
295				      unsigned int max_write_same_sectors)
296{
297	q->limits.max_write_same_sectors = max_write_same_sectors;
298}
299EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
300
301/**
302 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
303 *                                      write zeroes
304 * @q:  the request queue for the device
305 * @max_write_zeroes_sectors: maximum number of sectors to write per command
306 **/
307void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
308		unsigned int max_write_zeroes_sectors)
309{
310	q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
311}
312EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
313
314/**
315 * blk_queue_max_segments - set max hw segments for a request for this queue
316 * @q:  the request queue for the device
317 * @max_segments:  max number of segments
318 *
319 * Description:
320 *    Enables a low level driver to set an upper limit on the number of
321 *    hw data segments in a request.
322 **/
323void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
324{
325	if (!max_segments) {
326		max_segments = 1;
327		printk(KERN_INFO "%s: set to minimum %d\n",
328		       __func__, max_segments);
329	}
330
331	q->limits.max_segments = max_segments;
332}
333EXPORT_SYMBOL(blk_queue_max_segments);
334
335/**
336 * blk_queue_max_discard_segments - set max segments for discard requests
337 * @q:  the request queue for the device
338 * @max_segments:  max number of segments
339 *
340 * Description:
341 *    Enables a low level driver to set an upper limit on the number of
342 *    segments in a discard request.
343 **/
344void blk_queue_max_discard_segments(struct request_queue *q,
345		unsigned short max_segments)
346{
347	q->limits.max_discard_segments = max_segments;
348}
349EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
350
351/**
352 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
353 * @q:  the request queue for the device
354 * @max_size:  max size of segment in bytes
355 *
356 * Description:
357 *    Enables a low level driver to set an upper limit on the size of a
358 *    coalesced segment
359 **/
360void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
361{
362	if (max_size < PAGE_SIZE) {
363		max_size = PAGE_SIZE;
364		printk(KERN_INFO "%s: set to minimum %d\n",
365		       __func__, max_size);
366	}
367
368	q->limits.max_segment_size = max_size;
369}
370EXPORT_SYMBOL(blk_queue_max_segment_size);
371
372/**
373 * blk_queue_logical_block_size - set logical block size for the queue
374 * @q:  the request queue for the device
375 * @size:  the logical block size, in bytes
376 *
377 * Description:
378 *   This should be set to the lowest possible block size that the
379 *   storage device can address.  The default of 512 covers most
380 *   hardware.
381 **/
382void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
383{
384	q->limits.logical_block_size = size;
385
386	if (q->limits.physical_block_size < size)
387		q->limits.physical_block_size = size;
388
389	if (q->limits.io_min < q->limits.physical_block_size)
390		q->limits.io_min = q->limits.physical_block_size;
391}
392EXPORT_SYMBOL(blk_queue_logical_block_size);
393
394/**
395 * blk_queue_physical_block_size - set physical block size for the queue
396 * @q:  the request queue for the device
397 * @size:  the physical block size, in bytes
398 *
399 * Description:
400 *   This should be set to the lowest possible sector size that the
401 *   hardware can operate on without reverting to read-modify-write
402 *   operations.
403 */
404void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
405{
406	q->limits.physical_block_size = size;
407
408	if (q->limits.physical_block_size < q->limits.logical_block_size)
409		q->limits.physical_block_size = q->limits.logical_block_size;
410
411	if (q->limits.io_min < q->limits.physical_block_size)
412		q->limits.io_min = q->limits.physical_block_size;
413}
414EXPORT_SYMBOL(blk_queue_physical_block_size);
415
416/**
417 * blk_queue_alignment_offset - set physical block alignment offset
418 * @q:	the request queue for the device
419 * @offset: alignment offset in bytes
420 *
421 * Description:
422 *   Some devices are naturally misaligned to compensate for things like
423 *   the legacy DOS partition table 63-sector offset.  Low-level drivers
424 *   should call this function for devices whose first sector is not
425 *   naturally aligned.
426 */
427void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
428{
429	q->limits.alignment_offset =
430		offset & (q->limits.physical_block_size - 1);
431	q->limits.misaligned = 0;
432}
433EXPORT_SYMBOL(blk_queue_alignment_offset);
434
435/**
436 * blk_limits_io_min - set minimum request size for a device
437 * @limits: the queue limits
438 * @min:  smallest I/O size in bytes
439 *
440 * Description:
441 *   Some devices have an internal block size bigger than the reported
442 *   hardware sector size.  This function can be used to signal the
443 *   smallest I/O the device can perform without incurring a performance
444 *   penalty.
445 */
446void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
447{
448	limits->io_min = min;
449
450	if (limits->io_min < limits->logical_block_size)
451		limits->io_min = limits->logical_block_size;
452
453	if (limits->io_min < limits->physical_block_size)
454		limits->io_min = limits->physical_block_size;
455}
456EXPORT_SYMBOL(blk_limits_io_min);
457
458/**
459 * blk_queue_io_min - set minimum request size for the queue
460 * @q:	the request queue for the device
461 * @min:  smallest I/O size in bytes
462 *
463 * Description:
464 *   Storage devices may report a granularity or preferred minimum I/O
465 *   size which is the smallest request the device can perform without
466 *   incurring a performance penalty.  For disk drives this is often the
467 *   physical block size.  For RAID arrays it is often the stripe chunk
468 *   size.  A properly aligned multiple of minimum_io_size is the
469 *   preferred request size for workloads where a high number of I/O
470 *   operations is desired.
471 */
472void blk_queue_io_min(struct request_queue *q, unsigned int min)
473{
474	blk_limits_io_min(&q->limits, min);
475}
476EXPORT_SYMBOL(blk_queue_io_min);
477
478/**
479 * blk_limits_io_opt - set optimal request size for a device
480 * @limits: the queue limits
481 * @opt:  smallest I/O size in bytes
482 *
483 * Description:
484 *   Storage devices may report an optimal I/O size, which is the
485 *   device's preferred unit for sustained I/O.  This is rarely reported
486 *   for disk drives.  For RAID arrays it is usually the stripe width or
487 *   the internal track size.  A properly aligned multiple of
488 *   optimal_io_size is the preferred request size for workloads where
489 *   sustained throughput is desired.
490 */
491void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
492{
493	limits->io_opt = opt;
494}
495EXPORT_SYMBOL(blk_limits_io_opt);
496
497/**
498 * blk_queue_io_opt - set optimal request size for the queue
499 * @q:	the request queue for the device
500 * @opt:  optimal request size in bytes
501 *
502 * Description:
503 *   Storage devices may report an optimal I/O size, which is the
504 *   device's preferred unit for sustained I/O.  This is rarely reported
505 *   for disk drives.  For RAID arrays it is usually the stripe width or
506 *   the internal track size.  A properly aligned multiple of
507 *   optimal_io_size is the preferred request size for workloads where
508 *   sustained throughput is desired.
509 */
510void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
511{
512	blk_limits_io_opt(&q->limits, opt);
513}
514EXPORT_SYMBOL(blk_queue_io_opt);
515
516/**
517 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
518 * @t:	the stacking driver (top)
519 * @b:  the underlying device (bottom)
520 **/
521void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
522{
523	blk_stack_limits(&t->limits, &b->limits, 0);
524}
525EXPORT_SYMBOL(blk_queue_stack_limits);
526
527/**
528 * blk_stack_limits - adjust queue_limits for stacked devices
529 * @t:	the stacking driver limits (top device)
530 * @b:  the underlying queue limits (bottom, component device)
531 * @start:  first data sector within component device
532 *
533 * Description:
534 *    This function is used by stacking drivers like MD and DM to ensure
535 *    that all component devices have compatible block sizes and
536 *    alignments.  The stacking driver must provide a queue_limits
537 *    struct (top) and then iteratively call the stacking function for
538 *    all component (bottom) devices.  The stacking function will
539 *    attempt to combine the values and ensure proper alignment.
540 *
541 *    Returns 0 if the top and bottom queue_limits are compatible.  The
542 *    top device's block sizes and alignment offsets may be adjusted to
543 *    ensure alignment with the bottom device. If no compatible sizes
544 *    and alignments exist, -1 is returned and the resulting top
545 *    queue_limits will have the misaligned flag set to indicate that
546 *    the alignment_offset is undefined.
547 */
548int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
549		     sector_t start)
550{
551	unsigned int top, bottom, alignment, ret = 0;
552
553	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
554	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
555	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
556	t->max_write_same_sectors = min(t->max_write_same_sectors,
557					b->max_write_same_sectors);
558	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
559					b->max_write_zeroes_sectors);
560	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
561
562	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
563					    b->seg_boundary_mask);
564	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
565					    b->virt_boundary_mask);
566
567	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
568	t->max_discard_segments = min_not_zero(t->max_discard_segments,
569					       b->max_discard_segments);
570	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
571						 b->max_integrity_segments);
572
573	t->max_segment_size = min_not_zero(t->max_segment_size,
574					   b->max_segment_size);
575
576	t->misaligned |= b->misaligned;
577
578	alignment = queue_limit_alignment_offset(b, start);
579
580	/* Bottom device has different alignment.  Check that it is
581	 * compatible with the current top alignment.
582	 */
583	if (t->alignment_offset != alignment) {
584
585		top = max(t->physical_block_size, t->io_min)
586			+ t->alignment_offset;
587		bottom = max(b->physical_block_size, b->io_min) + alignment;
588
589		/* Verify that top and bottom intervals line up */
590		if (max(top, bottom) % min(top, bottom)) {
591			t->misaligned = 1;
592			ret = -1;
593		}
594	}
595
596	t->logical_block_size = max(t->logical_block_size,
597				    b->logical_block_size);
598
599	t->physical_block_size = max(t->physical_block_size,
600				     b->physical_block_size);
601
602	t->io_min = max(t->io_min, b->io_min);
603	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
604
605	t->cluster &= b->cluster;
 
606
607	/* Physical block size a multiple of the logical block size? */
608	if (t->physical_block_size & (t->logical_block_size - 1)) {
609		t->physical_block_size = t->logical_block_size;
610		t->misaligned = 1;
611		ret = -1;
612	}
613
614	/* Minimum I/O a multiple of the physical block size? */
615	if (t->io_min & (t->physical_block_size - 1)) {
616		t->io_min = t->physical_block_size;
617		t->misaligned = 1;
618		ret = -1;
619	}
620
621	/* Optimal I/O a multiple of the physical block size? */
622	if (t->io_opt & (t->physical_block_size - 1)) {
623		t->io_opt = 0;
624		t->misaligned = 1;
625		ret = -1;
626	}
627
628	t->raid_partial_stripes_expensive =
629		max(t->raid_partial_stripes_expensive,
630		    b->raid_partial_stripes_expensive);
631
632	/* Find lowest common alignment_offset */
633	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
634		% max(t->physical_block_size, t->io_min);
635
636	/* Verify that new alignment_offset is on a logical block boundary */
637	if (t->alignment_offset & (t->logical_block_size - 1)) {
638		t->misaligned = 1;
639		ret = -1;
640	}
641
642	/* Discard alignment and granularity */
643	if (b->discard_granularity) {
644		alignment = queue_limit_discard_alignment(b, start);
645
646		if (t->discard_granularity != 0 &&
647		    t->discard_alignment != alignment) {
648			top = t->discard_granularity + t->discard_alignment;
649			bottom = b->discard_granularity + alignment;
650
651			/* Verify that top and bottom intervals line up */
652			if ((max(top, bottom) % min(top, bottom)) != 0)
653				t->discard_misaligned = 1;
654		}
655
656		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
657						      b->max_discard_sectors);
658		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
659							 b->max_hw_discard_sectors);
660		t->discard_granularity = max(t->discard_granularity,
661					     b->discard_granularity);
662		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
663			t->discard_granularity;
664	}
665
666	if (b->chunk_sectors)
667		t->chunk_sectors = min_not_zero(t->chunk_sectors,
668						b->chunk_sectors);
669
670	return ret;
671}
672EXPORT_SYMBOL(blk_stack_limits);
673
674/**
675 * bdev_stack_limits - adjust queue limits for stacked drivers
676 * @t:	the stacking driver limits (top device)
677 * @bdev:  the component block_device (bottom)
678 * @start:  first data sector within component device
679 *
680 * Description:
681 *    Merges queue limits for a top device and a block_device.  Returns
682 *    0 if alignment didn't change.  Returns -1 if adding the bottom
683 *    device caused misalignment.
684 */
685int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
686		      sector_t start)
687{
688	struct request_queue *bq = bdev_get_queue(bdev);
689
690	start += get_start_sect(bdev);
691
692	return blk_stack_limits(t, &bq->limits, start);
693}
694EXPORT_SYMBOL(bdev_stack_limits);
695
696/**
697 * disk_stack_limits - adjust queue limits for stacked drivers
698 * @disk:  MD/DM gendisk (top)
699 * @bdev:  the underlying block device (bottom)
700 * @offset:  offset to beginning of data within component device
701 *
702 * Description:
703 *    Merges the limits for a top level gendisk and a bottom level
704 *    block_device.
705 */
706void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
707		       sector_t offset)
708{
709	struct request_queue *t = disk->queue;
710
711	if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
712		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
713
714		disk_name(disk, 0, top);
715		bdevname(bdev, bottom);
716
717		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
718		       top, bottom);
719	}
720}
721EXPORT_SYMBOL(disk_stack_limits);
722
723/**
724 * blk_queue_dma_pad - set pad mask
725 * @q:     the request queue for the device
726 * @mask:  pad mask
727 *
728 * Set dma pad mask.
729 *
730 * Appending pad buffer to a request modifies the last entry of a
731 * scatter list such that it includes the pad buffer.
732 **/
733void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
734{
735	q->dma_pad_mask = mask;
736}
737EXPORT_SYMBOL(blk_queue_dma_pad);
738
739/**
740 * blk_queue_update_dma_pad - update pad mask
741 * @q:     the request queue for the device
742 * @mask:  pad mask
743 *
744 * Update dma pad mask.
745 *
746 * Appending pad buffer to a request modifies the last entry of a
747 * scatter list such that it includes the pad buffer.
748 **/
749void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
750{
751	if (mask > q->dma_pad_mask)
752		q->dma_pad_mask = mask;
753}
754EXPORT_SYMBOL(blk_queue_update_dma_pad);
755
756/**
757 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
758 * @q:  the request queue for the device
759 * @dma_drain_needed: fn which returns non-zero if drain is necessary
760 * @buf:	physically contiguous buffer
761 * @size:	size of the buffer in bytes
762 *
763 * Some devices have excess DMA problems and can't simply discard (or
764 * zero fill) the unwanted piece of the transfer.  They have to have a
765 * real area of memory to transfer it into.  The use case for this is
766 * ATAPI devices in DMA mode.  If the packet command causes a transfer
767 * bigger than the transfer size some HBAs will lock up if there
768 * aren't DMA elements to contain the excess transfer.  What this API
769 * does is adjust the queue so that the buf is always appended
770 * silently to the scatterlist.
771 *
772 * Note: This routine adjusts max_hw_segments to make room for appending
773 * the drain buffer.  If you call blk_queue_max_segments() after calling
774 * this routine, you must set the limit to one fewer than your device
775 * can support otherwise there won't be room for the drain buffer.
776 */
777int blk_queue_dma_drain(struct request_queue *q,
778			       dma_drain_needed_fn *dma_drain_needed,
779			       void *buf, unsigned int size)
780{
781	if (queue_max_segments(q) < 2)
782		return -EINVAL;
783	/* make room for appending the drain */
784	blk_queue_max_segments(q, queue_max_segments(q) - 1);
785	q->dma_drain_needed = dma_drain_needed;
786	q->dma_drain_buffer = buf;
787	q->dma_drain_size = size;
788
789	return 0;
790}
791EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
792
793/**
794 * blk_queue_segment_boundary - set boundary rules for segment merging
795 * @q:  the request queue for the device
796 * @mask:  the memory boundary mask
797 **/
798void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
799{
800	if (mask < PAGE_SIZE - 1) {
801		mask = PAGE_SIZE - 1;
802		printk(KERN_INFO "%s: set to minimum %lx\n",
803		       __func__, mask);
804	}
805
806	q->limits.seg_boundary_mask = mask;
807}
808EXPORT_SYMBOL(blk_queue_segment_boundary);
809
810/**
811 * blk_queue_virt_boundary - set boundary rules for bio merging
812 * @q:  the request queue for the device
813 * @mask:  the memory boundary mask
814 **/
815void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
816{
817	q->limits.virt_boundary_mask = mask;
818}
819EXPORT_SYMBOL(blk_queue_virt_boundary);
820
821/**
822 * blk_queue_dma_alignment - set dma length and memory alignment
823 * @q:     the request queue for the device
824 * @mask:  alignment mask
825 *
826 * description:
827 *    set required memory and length alignment for direct dma transactions.
828 *    this is used when building direct io requests for the queue.
829 *
830 **/
831void blk_queue_dma_alignment(struct request_queue *q, int mask)
832{
833	q->dma_alignment = mask;
834}
835EXPORT_SYMBOL(blk_queue_dma_alignment);
836
837/**
838 * blk_queue_update_dma_alignment - update dma length and memory alignment
839 * @q:     the request queue for the device
840 * @mask:  alignment mask
841 *
842 * description:
843 *    update required memory and length alignment for direct dma transactions.
844 *    If the requested alignment is larger than the current alignment, then
845 *    the current queue alignment is updated to the new value, otherwise it
846 *    is left alone.  The design of this is to allow multiple objects
847 *    (driver, device, transport etc) to set their respective
848 *    alignments without having them interfere.
849 *
850 **/
851void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
852{
853	BUG_ON(mask > PAGE_SIZE);
854
855	if (mask > q->dma_alignment)
856		q->dma_alignment = mask;
857}
858EXPORT_SYMBOL(blk_queue_update_dma_alignment);
859
860void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
861{
862	if (queueable)
863		blk_queue_flag_clear(QUEUE_FLAG_FLUSH_NQ, q);
864	else
865		blk_queue_flag_set(QUEUE_FLAG_FLUSH_NQ, q);
866}
867EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
868
869/**
870 * blk_set_queue_depth - tell the block layer about the device queue depth
871 * @q:		the request queue for the device
872 * @depth:		queue depth
873 *
 
 
 
874 */
875void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
876{
877	q->queue_depth = depth;
878	wbt_set_queue_depth(q->rq_wb, depth);
 
 
 
 
879}
880EXPORT_SYMBOL(blk_set_queue_depth);
881
882/**
883 * blk_queue_write_cache - configure queue's write cache
884 * @q:		the request queue for the device
885 * @wc:		write back cache on or off
886 * @fua:	device supports FUA writes, if true
887 *
888 * Tell the block layer about the write cache of @q.
889 */
890void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
891{
892	spin_lock_irq(q->queue_lock);
893	if (wc)
894		queue_flag_set(QUEUE_FLAG_WC, q);
895	else
896		queue_flag_clear(QUEUE_FLAG_WC, q);
897	if (fua)
898		queue_flag_set(QUEUE_FLAG_FUA, q);
899	else
900		queue_flag_clear(QUEUE_FLAG_FUA, q);
901	spin_unlock_irq(q->queue_lock);
902
903	wbt_set_write_cache(q->rq_wb, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
904}
905EXPORT_SYMBOL_GPL(blk_queue_write_cache);
906
907static int __init blk_settings_init(void)
908{
909	blk_max_low_pfn = max_low_pfn - 1;
910	blk_max_pfn = max_pfn - 1;
911	return 0;
912}
913subsys_initcall(blk_settings_init);