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