Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/pagemap.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/gcd.h>
13#include <linux/lcm.h>
14#include <linux/jiffies.h>
15#include <linux/gfp.h>
16#include <linux/dma-mapping.h>
17
18#include "blk.h"
19#include "blk-wbt.h"
20
21void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
22{
23 q->rq_timeout = timeout;
24}
25EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
26
27/**
28 * blk_set_default_limits - reset limits to default values
29 * @lim: the queue_limits structure to reset
30 *
31 * Description:
32 * Returns a queue_limit struct to its default state.
33 */
34void blk_set_default_limits(struct queue_limits *lim)
35{
36 lim->max_segments = BLK_MAX_SEGMENTS;
37 lim->max_discard_segments = 1;
38 lim->max_integrity_segments = 0;
39 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
40 lim->virt_boundary_mask = 0;
41 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
42 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
43 lim->max_dev_sectors = 0;
44 lim->chunk_sectors = 0;
45 lim->max_write_zeroes_sectors = 0;
46 lim->max_zone_append_sectors = 0;
47 lim->max_discard_sectors = 0;
48 lim->max_hw_discard_sectors = 0;
49 lim->max_secure_erase_sectors = 0;
50 lim->discard_granularity = 0;
51 lim->discard_alignment = 0;
52 lim->discard_misaligned = 0;
53 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
54 lim->bounce = BLK_BOUNCE_NONE;
55 lim->alignment_offset = 0;
56 lim->io_opt = 0;
57 lim->misaligned = 0;
58 lim->zoned = BLK_ZONED_NONE;
59 lim->zone_write_granularity = 0;
60 lim->dma_alignment = 511;
61}
62
63/**
64 * blk_set_stacking_limits - set default limits for stacking devices
65 * @lim: the queue_limits structure to reset
66 *
67 * Description:
68 * Returns a queue_limit struct to its default state. Should be used
69 * by stacking drivers like DM that have no internal limits.
70 */
71void blk_set_stacking_limits(struct queue_limits *lim)
72{
73 blk_set_default_limits(lim);
74
75 /* Inherit limits from component devices */
76 lim->max_segments = USHRT_MAX;
77 lim->max_discard_segments = USHRT_MAX;
78 lim->max_hw_sectors = UINT_MAX;
79 lim->max_segment_size = UINT_MAX;
80 lim->max_sectors = UINT_MAX;
81 lim->max_dev_sectors = UINT_MAX;
82 lim->max_write_zeroes_sectors = UINT_MAX;
83 lim->max_zone_append_sectors = UINT_MAX;
84}
85EXPORT_SYMBOL(blk_set_stacking_limits);
86
87/**
88 * blk_queue_bounce_limit - set bounce buffer limit for queue
89 * @q: the request queue for the device
90 * @bounce: bounce limit to enforce
91 *
92 * Description:
93 * Force bouncing for ISA DMA ranges or highmem.
94 *
95 * DEPRECATED, don't use in new code.
96 **/
97void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
98{
99 q->limits.bounce = bounce;
100}
101EXPORT_SYMBOL(blk_queue_bounce_limit);
102
103/**
104 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
105 * @q: the request queue for the device
106 * @max_hw_sectors: max hardware sectors in the usual 512b unit
107 *
108 * Description:
109 * Enables a low level driver to set a hard upper limit,
110 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
111 * the device driver based upon the capabilities of the I/O
112 * controller.
113 *
114 * max_dev_sectors is a hard limit imposed by the storage device for
115 * READ/WRITE requests. It is set by the disk driver.
116 *
117 * max_sectors is a soft limit imposed by the block layer for
118 * filesystem type requests. This value can be overridden on a
119 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
120 * The soft limit can not exceed max_hw_sectors.
121 **/
122void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
123{
124 struct queue_limits *limits = &q->limits;
125 unsigned int max_sectors;
126
127 if ((max_hw_sectors << 9) < PAGE_SIZE) {
128 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
129 printk(KERN_INFO "%s: set to minimum %d\n",
130 __func__, max_hw_sectors);
131 }
132
133 max_hw_sectors = round_down(max_hw_sectors,
134 limits->logical_block_size >> SECTOR_SHIFT);
135 limits->max_hw_sectors = max_hw_sectors;
136
137 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
138 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
139 max_sectors = round_down(max_sectors,
140 limits->logical_block_size >> SECTOR_SHIFT);
141 limits->max_sectors = max_sectors;
142
143 if (!q->disk)
144 return;
145 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
146}
147EXPORT_SYMBOL(blk_queue_max_hw_sectors);
148
149/**
150 * blk_queue_chunk_sectors - set size of the chunk for this queue
151 * @q: the request queue for the device
152 * @chunk_sectors: chunk sectors in the usual 512b unit
153 *
154 * Description:
155 * If a driver doesn't want IOs to cross a given chunk size, it can set
156 * this limit and prevent merging across chunks. Note that the block layer
157 * must accept a page worth of data at any offset. So if the crossing of
158 * chunks is a hard limitation in the driver, it must still be prepared
159 * to split single page bios.
160 **/
161void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
162{
163 q->limits.chunk_sectors = chunk_sectors;
164}
165EXPORT_SYMBOL(blk_queue_chunk_sectors);
166
167/**
168 * blk_queue_max_discard_sectors - set max sectors for a single discard
169 * @q: the request queue for the device
170 * @max_discard_sectors: maximum number of sectors to discard
171 **/
172void blk_queue_max_discard_sectors(struct request_queue *q,
173 unsigned int max_discard_sectors)
174{
175 q->limits.max_hw_discard_sectors = max_discard_sectors;
176 q->limits.max_discard_sectors = max_discard_sectors;
177}
178EXPORT_SYMBOL(blk_queue_max_discard_sectors);
179
180/**
181 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
182 * @q: the request queue for the device
183 * @max_sectors: maximum number of sectors to secure_erase
184 **/
185void blk_queue_max_secure_erase_sectors(struct request_queue *q,
186 unsigned int max_sectors)
187{
188 q->limits.max_secure_erase_sectors = max_sectors;
189}
190EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
191
192/**
193 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
194 * write zeroes
195 * @q: the request queue for the device
196 * @max_write_zeroes_sectors: maximum number of sectors to write per command
197 **/
198void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
199 unsigned int max_write_zeroes_sectors)
200{
201 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
202}
203EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
204
205/**
206 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
207 * @q: the request queue for the device
208 * @max_zone_append_sectors: maximum number of sectors to write per command
209 **/
210void blk_queue_max_zone_append_sectors(struct request_queue *q,
211 unsigned int max_zone_append_sectors)
212{
213 unsigned int max_sectors;
214
215 if (WARN_ON(!blk_queue_is_zoned(q)))
216 return;
217
218 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
219 max_sectors = min(q->limits.chunk_sectors, max_sectors);
220
221 /*
222 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
223 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
224 * or the max_hw_sectors limit not set.
225 */
226 WARN_ON(!max_sectors);
227
228 q->limits.max_zone_append_sectors = max_sectors;
229}
230EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
231
232/**
233 * blk_queue_max_segments - set max hw segments for a request for this queue
234 * @q: the request queue for the device
235 * @max_segments: max number of segments
236 *
237 * Description:
238 * Enables a low level driver to set an upper limit on the number of
239 * hw data segments in a request.
240 **/
241void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
242{
243 if (!max_segments) {
244 max_segments = 1;
245 printk(KERN_INFO "%s: set to minimum %d\n",
246 __func__, max_segments);
247 }
248
249 q->limits.max_segments = max_segments;
250}
251EXPORT_SYMBOL(blk_queue_max_segments);
252
253/**
254 * blk_queue_max_discard_segments - set max segments for discard requests
255 * @q: the request queue for the device
256 * @max_segments: max number of segments
257 *
258 * Description:
259 * Enables a low level driver to set an upper limit on the number of
260 * segments in a discard request.
261 **/
262void blk_queue_max_discard_segments(struct request_queue *q,
263 unsigned short max_segments)
264{
265 q->limits.max_discard_segments = max_segments;
266}
267EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
268
269/**
270 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
271 * @q: the request queue for the device
272 * @max_size: max size of segment in bytes
273 *
274 * Description:
275 * Enables a low level driver to set an upper limit on the size of a
276 * coalesced segment
277 **/
278void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
279{
280 if (max_size < PAGE_SIZE) {
281 max_size = PAGE_SIZE;
282 printk(KERN_INFO "%s: set to minimum %d\n",
283 __func__, max_size);
284 }
285
286 /* see blk_queue_virt_boundary() for the explanation */
287 WARN_ON_ONCE(q->limits.virt_boundary_mask);
288
289 q->limits.max_segment_size = max_size;
290}
291EXPORT_SYMBOL(blk_queue_max_segment_size);
292
293/**
294 * blk_queue_logical_block_size - set logical block size for the queue
295 * @q: the request queue for the device
296 * @size: the logical block size, in bytes
297 *
298 * Description:
299 * This should be set to the lowest possible block size that the
300 * storage device can address. The default of 512 covers most
301 * hardware.
302 **/
303void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
304{
305 struct queue_limits *limits = &q->limits;
306
307 limits->logical_block_size = size;
308
309 if (limits->physical_block_size < size)
310 limits->physical_block_size = size;
311
312 if (limits->io_min < limits->physical_block_size)
313 limits->io_min = limits->physical_block_size;
314
315 limits->max_hw_sectors =
316 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
317 limits->max_sectors =
318 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
319}
320EXPORT_SYMBOL(blk_queue_logical_block_size);
321
322/**
323 * blk_queue_physical_block_size - set physical block size for the queue
324 * @q: the request queue for the device
325 * @size: the physical block size, in bytes
326 *
327 * Description:
328 * This should be set to the lowest possible sector size that the
329 * hardware can operate on without reverting to read-modify-write
330 * operations.
331 */
332void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
333{
334 q->limits.physical_block_size = size;
335
336 if (q->limits.physical_block_size < q->limits.logical_block_size)
337 q->limits.physical_block_size = q->limits.logical_block_size;
338
339 if (q->limits.io_min < q->limits.physical_block_size)
340 q->limits.io_min = q->limits.physical_block_size;
341}
342EXPORT_SYMBOL(blk_queue_physical_block_size);
343
344/**
345 * blk_queue_zone_write_granularity - set zone write granularity for the queue
346 * @q: the request queue for the zoned device
347 * @size: the zone write granularity size, in bytes
348 *
349 * Description:
350 * This should be set to the lowest possible size allowing to write in
351 * sequential zones of a zoned block device.
352 */
353void blk_queue_zone_write_granularity(struct request_queue *q,
354 unsigned int size)
355{
356 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
357 return;
358
359 q->limits.zone_write_granularity = size;
360
361 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
362 q->limits.zone_write_granularity = q->limits.logical_block_size;
363}
364EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
365
366/**
367 * blk_queue_alignment_offset - set physical block alignment offset
368 * @q: the request queue for the device
369 * @offset: alignment offset in bytes
370 *
371 * Description:
372 * Some devices are naturally misaligned to compensate for things like
373 * the legacy DOS partition table 63-sector offset. Low-level drivers
374 * should call this function for devices whose first sector is not
375 * naturally aligned.
376 */
377void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
378{
379 q->limits.alignment_offset =
380 offset & (q->limits.physical_block_size - 1);
381 q->limits.misaligned = 0;
382}
383EXPORT_SYMBOL(blk_queue_alignment_offset);
384
385void disk_update_readahead(struct gendisk *disk)
386{
387 struct request_queue *q = disk->queue;
388
389 /*
390 * For read-ahead of large files to be effective, we need to read ahead
391 * at least twice the optimal I/O size.
392 */
393 disk->bdi->ra_pages =
394 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
395 disk->bdi->io_pages = queue_max_sectors(q) >> (PAGE_SHIFT - 9);
396}
397EXPORT_SYMBOL_GPL(disk_update_readahead);
398
399/**
400 * blk_limits_io_min - set minimum request size for a device
401 * @limits: the queue limits
402 * @min: smallest I/O size in bytes
403 *
404 * Description:
405 * Some devices have an internal block size bigger than the reported
406 * hardware sector size. This function can be used to signal the
407 * smallest I/O the device can perform without incurring a performance
408 * penalty.
409 */
410void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
411{
412 limits->io_min = min;
413
414 if (limits->io_min < limits->logical_block_size)
415 limits->io_min = limits->logical_block_size;
416
417 if (limits->io_min < limits->physical_block_size)
418 limits->io_min = limits->physical_block_size;
419}
420EXPORT_SYMBOL(blk_limits_io_min);
421
422/**
423 * blk_queue_io_min - set minimum request size for the queue
424 * @q: the request queue for the device
425 * @min: smallest I/O size in bytes
426 *
427 * Description:
428 * Storage devices may report a granularity or preferred minimum I/O
429 * size which is the smallest request the device can perform without
430 * incurring a performance penalty. For disk drives this is often the
431 * physical block size. For RAID arrays it is often the stripe chunk
432 * size. A properly aligned multiple of minimum_io_size is the
433 * preferred request size for workloads where a high number of I/O
434 * operations is desired.
435 */
436void blk_queue_io_min(struct request_queue *q, unsigned int min)
437{
438 blk_limits_io_min(&q->limits, min);
439}
440EXPORT_SYMBOL(blk_queue_io_min);
441
442/**
443 * blk_limits_io_opt - set optimal request size for a device
444 * @limits: the queue limits
445 * @opt: smallest I/O size in bytes
446 *
447 * Description:
448 * Storage devices may report an optimal I/O size, which is the
449 * device's preferred unit for sustained I/O. This is rarely reported
450 * for disk drives. For RAID arrays it is usually the stripe width or
451 * the internal track size. A properly aligned multiple of
452 * optimal_io_size is the preferred request size for workloads where
453 * sustained throughput is desired.
454 */
455void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
456{
457 limits->io_opt = opt;
458}
459EXPORT_SYMBOL(blk_limits_io_opt);
460
461/**
462 * blk_queue_io_opt - set optimal request size for the queue
463 * @q: the request queue for the device
464 * @opt: optimal request size in bytes
465 *
466 * Description:
467 * Storage devices may report an optimal I/O size, which is the
468 * device's preferred unit for sustained I/O. This is rarely reported
469 * for disk drives. For RAID arrays it is usually the stripe width or
470 * the internal track size. A properly aligned multiple of
471 * optimal_io_size is the preferred request size for workloads where
472 * sustained throughput is desired.
473 */
474void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
475{
476 blk_limits_io_opt(&q->limits, opt);
477 if (!q->disk)
478 return;
479 q->disk->bdi->ra_pages =
480 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
481}
482EXPORT_SYMBOL(blk_queue_io_opt);
483
484static int queue_limit_alignment_offset(const struct queue_limits *lim,
485 sector_t sector)
486{
487 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
488 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
489 << SECTOR_SHIFT;
490
491 return (granularity + lim->alignment_offset - alignment) % granularity;
492}
493
494static unsigned int queue_limit_discard_alignment(
495 const struct queue_limits *lim, sector_t sector)
496{
497 unsigned int alignment, granularity, offset;
498
499 if (!lim->max_discard_sectors)
500 return 0;
501
502 /* Why are these in bytes, not sectors? */
503 alignment = lim->discard_alignment >> SECTOR_SHIFT;
504 granularity = lim->discard_granularity >> SECTOR_SHIFT;
505 if (!granularity)
506 return 0;
507
508 /* Offset of the partition start in 'granularity' sectors */
509 offset = sector_div(sector, granularity);
510
511 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
512 offset = (granularity + alignment - offset) % granularity;
513
514 /* Turn it back into bytes, gaah */
515 return offset << SECTOR_SHIFT;
516}
517
518static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
519{
520 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
521 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
522 sectors = PAGE_SIZE >> SECTOR_SHIFT;
523 return sectors;
524}
525
526/**
527 * blk_stack_limits - adjust queue_limits for stacked devices
528 * @t: the stacking driver limits (top device)
529 * @b: the underlying queue limits (bottom, component device)
530 * @start: first data sector within component device
531 *
532 * Description:
533 * This function is used by stacking drivers like MD and DM to ensure
534 * that all component devices have compatible block sizes and
535 * alignments. The stacking driver must provide a queue_limits
536 * struct (top) and then iteratively call the stacking function for
537 * all component (bottom) devices. The stacking function will
538 * attempt to combine the values and ensure proper alignment.
539 *
540 * Returns 0 if the top and bottom queue_limits are compatible. The
541 * top device's block sizes and alignment offsets may be adjusted to
542 * ensure alignment with the bottom device. If no compatible sizes
543 * and alignments exist, -1 is returned and the resulting top
544 * queue_limits will have the misaligned flag set to indicate that
545 * the alignment_offset is undefined.
546 */
547int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
548 sector_t start)
549{
550 unsigned int top, bottom, alignment, ret = 0;
551
552 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
553 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
554 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
555 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
556 b->max_write_zeroes_sectors);
557 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
558 b->max_zone_append_sectors);
559 t->bounce = max(t->bounce, b->bounce);
560
561 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
562 b->seg_boundary_mask);
563 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
564 b->virt_boundary_mask);
565
566 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
567 t->max_discard_segments = min_not_zero(t->max_discard_segments,
568 b->max_discard_segments);
569 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
570 b->max_integrity_segments);
571
572 t->max_segment_size = min_not_zero(t->max_segment_size,
573 b->max_segment_size);
574
575 t->misaligned |= b->misaligned;
576
577 alignment = queue_limit_alignment_offset(b, start);
578
579 /* Bottom device has different alignment. Check that it is
580 * compatible with the current top alignment.
581 */
582 if (t->alignment_offset != alignment) {
583
584 top = max(t->physical_block_size, t->io_min)
585 + t->alignment_offset;
586 bottom = max(b->physical_block_size, b->io_min) + alignment;
587
588 /* Verify that top and bottom intervals line up */
589 if (max(top, bottom) % min(top, bottom)) {
590 t->misaligned = 1;
591 ret = -1;
592 }
593 }
594
595 t->logical_block_size = max(t->logical_block_size,
596 b->logical_block_size);
597
598 t->physical_block_size = max(t->physical_block_size,
599 b->physical_block_size);
600
601 t->io_min = max(t->io_min, b->io_min);
602 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
603 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
604
605 /* Set non-power-of-2 compatible chunk_sectors boundary */
606 if (b->chunk_sectors)
607 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
608
609 /* Physical block size a multiple of the logical block size? */
610 if (t->physical_block_size & (t->logical_block_size - 1)) {
611 t->physical_block_size = t->logical_block_size;
612 t->misaligned = 1;
613 ret = -1;
614 }
615
616 /* Minimum I/O a multiple of the physical block size? */
617 if (t->io_min & (t->physical_block_size - 1)) {
618 t->io_min = t->physical_block_size;
619 t->misaligned = 1;
620 ret = -1;
621 }
622
623 /* Optimal I/O a multiple of the physical block size? */
624 if (t->io_opt & (t->physical_block_size - 1)) {
625 t->io_opt = 0;
626 t->misaligned = 1;
627 ret = -1;
628 }
629
630 /* chunk_sectors a multiple of the physical block size? */
631 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
632 t->chunk_sectors = 0;
633 t->misaligned = 1;
634 ret = -1;
635 }
636
637 t->raid_partial_stripes_expensive =
638 max(t->raid_partial_stripes_expensive,
639 b->raid_partial_stripes_expensive);
640
641 /* Find lowest common alignment_offset */
642 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
643 % max(t->physical_block_size, t->io_min);
644
645 /* Verify that new alignment_offset is on a logical block boundary */
646 if (t->alignment_offset & (t->logical_block_size - 1)) {
647 t->misaligned = 1;
648 ret = -1;
649 }
650
651 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
652 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
653 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
654
655 /* Discard alignment and granularity */
656 if (b->discard_granularity) {
657 alignment = queue_limit_discard_alignment(b, start);
658
659 if (t->discard_granularity != 0 &&
660 t->discard_alignment != alignment) {
661 top = t->discard_granularity + t->discard_alignment;
662 bottom = b->discard_granularity + alignment;
663
664 /* Verify that top and bottom intervals line up */
665 if ((max(top, bottom) % min(top, bottom)) != 0)
666 t->discard_misaligned = 1;
667 }
668
669 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
670 b->max_discard_sectors);
671 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
672 b->max_hw_discard_sectors);
673 t->discard_granularity = max(t->discard_granularity,
674 b->discard_granularity);
675 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
676 t->discard_granularity;
677 }
678 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
679 b->max_secure_erase_sectors);
680 t->zone_write_granularity = max(t->zone_write_granularity,
681 b->zone_write_granularity);
682 t->zoned = max(t->zoned, b->zoned);
683 return ret;
684}
685EXPORT_SYMBOL(blk_stack_limits);
686
687/**
688 * disk_stack_limits - adjust queue limits for stacked drivers
689 * @disk: MD/DM gendisk (top)
690 * @bdev: the underlying block device (bottom)
691 * @offset: offset to beginning of data within component device
692 *
693 * Description:
694 * Merges the limits for a top level gendisk and a bottom level
695 * block_device.
696 */
697void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
698 sector_t offset)
699{
700 struct request_queue *t = disk->queue;
701
702 if (blk_stack_limits(&t->limits, &bdev_get_queue(bdev)->limits,
703 get_start_sect(bdev) + (offset >> 9)) < 0)
704 pr_notice("%s: Warning: Device %pg is misaligned\n",
705 disk->disk_name, bdev);
706
707 disk_update_readahead(disk);
708}
709EXPORT_SYMBOL(disk_stack_limits);
710
711/**
712 * blk_queue_update_dma_pad - update pad mask
713 * @q: the request queue for the device
714 * @mask: pad mask
715 *
716 * Update dma pad mask.
717 *
718 * Appending pad buffer to a request modifies the last entry of a
719 * scatter list such that it includes the pad buffer.
720 **/
721void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
722{
723 if (mask > q->dma_pad_mask)
724 q->dma_pad_mask = mask;
725}
726EXPORT_SYMBOL(blk_queue_update_dma_pad);
727
728/**
729 * blk_queue_segment_boundary - set boundary rules for segment merging
730 * @q: the request queue for the device
731 * @mask: the memory boundary mask
732 **/
733void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
734{
735 if (mask < PAGE_SIZE - 1) {
736 mask = PAGE_SIZE - 1;
737 printk(KERN_INFO "%s: set to minimum %lx\n",
738 __func__, mask);
739 }
740
741 q->limits.seg_boundary_mask = mask;
742}
743EXPORT_SYMBOL(blk_queue_segment_boundary);
744
745/**
746 * blk_queue_virt_boundary - set boundary rules for bio merging
747 * @q: the request queue for the device
748 * @mask: the memory boundary mask
749 **/
750void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
751{
752 q->limits.virt_boundary_mask = mask;
753
754 /*
755 * Devices that require a virtual boundary do not support scatter/gather
756 * I/O natively, but instead require a descriptor list entry for each
757 * page (which might not be idential to the Linux PAGE_SIZE). Because
758 * of that they are not limited by our notion of "segment size".
759 */
760 if (mask)
761 q->limits.max_segment_size = UINT_MAX;
762}
763EXPORT_SYMBOL(blk_queue_virt_boundary);
764
765/**
766 * blk_queue_dma_alignment - set dma length and memory alignment
767 * @q: the request queue for the device
768 * @mask: alignment mask
769 *
770 * description:
771 * set required memory and length alignment for direct dma transactions.
772 * this is used when building direct io requests for the queue.
773 *
774 **/
775void blk_queue_dma_alignment(struct request_queue *q, int mask)
776{
777 q->limits.dma_alignment = mask;
778}
779EXPORT_SYMBOL(blk_queue_dma_alignment);
780
781/**
782 * blk_queue_update_dma_alignment - update dma length and memory alignment
783 * @q: the request queue for the device
784 * @mask: alignment mask
785 *
786 * description:
787 * update required memory and length alignment for direct dma transactions.
788 * If the requested alignment is larger than the current alignment, then
789 * the current queue alignment is updated to the new value, otherwise it
790 * is left alone. The design of this is to allow multiple objects
791 * (driver, device, transport etc) to set their respective
792 * alignments without having them interfere.
793 *
794 **/
795void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
796{
797 BUG_ON(mask > PAGE_SIZE);
798
799 if (mask > q->limits.dma_alignment)
800 q->limits.dma_alignment = mask;
801}
802EXPORT_SYMBOL(blk_queue_update_dma_alignment);
803
804/**
805 * blk_set_queue_depth - tell the block layer about the device queue depth
806 * @q: the request queue for the device
807 * @depth: queue depth
808 *
809 */
810void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
811{
812 q->queue_depth = depth;
813 rq_qos_queue_depth_changed(q);
814}
815EXPORT_SYMBOL(blk_set_queue_depth);
816
817/**
818 * blk_queue_write_cache - configure queue's write cache
819 * @q: the request queue for the device
820 * @wc: write back cache on or off
821 * @fua: device supports FUA writes, if true
822 *
823 * Tell the block layer about the write cache of @q.
824 */
825void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
826{
827 if (wc)
828 blk_queue_flag_set(QUEUE_FLAG_WC, q);
829 else
830 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
831 if (fua)
832 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
833 else
834 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
835
836 wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
837}
838EXPORT_SYMBOL_GPL(blk_queue_write_cache);
839
840/**
841 * blk_queue_required_elevator_features - Set a queue required elevator features
842 * @q: the request queue for the target device
843 * @features: Required elevator features OR'ed together
844 *
845 * Tell the block layer that for the device controlled through @q, only the
846 * only elevators that can be used are those that implement at least the set of
847 * features specified by @features.
848 */
849void blk_queue_required_elevator_features(struct request_queue *q,
850 unsigned int features)
851{
852 q->required_elevator_features = features;
853}
854EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
855
856/**
857 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
858 * @q: the request queue for the device
859 * @dev: the device pointer for dma
860 *
861 * Tell the block layer about merging the segments by dma map of @q.
862 */
863bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
864 struct device *dev)
865{
866 unsigned long boundary = dma_get_merge_boundary(dev);
867
868 if (!boundary)
869 return false;
870
871 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
872 blk_queue_virt_boundary(q, boundary);
873
874 return true;
875}
876EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
877
878static bool disk_has_partitions(struct gendisk *disk)
879{
880 unsigned long idx;
881 struct block_device *part;
882 bool ret = false;
883
884 rcu_read_lock();
885 xa_for_each(&disk->part_tbl, idx, part) {
886 if (bdev_is_partition(part)) {
887 ret = true;
888 break;
889 }
890 }
891 rcu_read_unlock();
892
893 return ret;
894}
895
896/**
897 * disk_set_zoned - configure the zoned model for a disk
898 * @disk: the gendisk of the queue to configure
899 * @model: the zoned model to set
900 *
901 * Set the zoned model of @disk to @model.
902 *
903 * When @model is BLK_ZONED_HM (host managed), this should be called only
904 * if zoned block device support is enabled (CONFIG_BLK_DEV_ZONED option).
905 * If @model specifies BLK_ZONED_HA (host aware), the effective model used
906 * depends on CONFIG_BLK_DEV_ZONED settings and on the existence of partitions
907 * on the disk.
908 */
909void disk_set_zoned(struct gendisk *disk, enum blk_zoned_model model)
910{
911 struct request_queue *q = disk->queue;
912
913 switch (model) {
914 case BLK_ZONED_HM:
915 /*
916 * Host managed devices are supported only if
917 * CONFIG_BLK_DEV_ZONED is enabled.
918 */
919 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
920 break;
921 case BLK_ZONED_HA:
922 /*
923 * Host aware devices can be treated either as regular block
924 * devices (similar to drive managed devices) or as zoned block
925 * devices to take advantage of the zone command set, similarly
926 * to host managed devices. We try the latter if there are no
927 * partitions and zoned block device support is enabled, else
928 * we do nothing special as far as the block layer is concerned.
929 */
930 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) ||
931 disk_has_partitions(disk))
932 model = BLK_ZONED_NONE;
933 break;
934 case BLK_ZONED_NONE:
935 default:
936 if (WARN_ON_ONCE(model != BLK_ZONED_NONE))
937 model = BLK_ZONED_NONE;
938 break;
939 }
940
941 q->limits.zoned = model;
942 if (model != BLK_ZONED_NONE) {
943 /*
944 * Set the zone write granularity to the device logical block
945 * size by default. The driver can change this value if needed.
946 */
947 blk_queue_zone_write_granularity(q,
948 queue_logical_block_size(q));
949 } else {
950 disk_clear_zone_settings(disk);
951 }
952}
953EXPORT_SYMBOL_GPL(disk_set_zoned);
954
955int bdev_alignment_offset(struct block_device *bdev)
956{
957 struct request_queue *q = bdev_get_queue(bdev);
958
959 if (q->limits.misaligned)
960 return -1;
961 if (bdev_is_partition(bdev))
962 return queue_limit_alignment_offset(&q->limits,
963 bdev->bd_start_sect);
964 return q->limits.alignment_offset;
965}
966EXPORT_SYMBOL_GPL(bdev_alignment_offset);
967
968unsigned int bdev_discard_alignment(struct block_device *bdev)
969{
970 struct request_queue *q = bdev_get_queue(bdev);
971
972 if (bdev_is_partition(bdev))
973 return queue_limit_discard_alignment(&q->limits,
974 bdev->bd_start_sect);
975 return q->limits.discard_alignment;
976}
977EXPORT_SYMBOL_GPL(bdev_discard_alignment);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/pagemap.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/gcd.h>
13#include <linux/lcm.h>
14#include <linux/jiffies.h>
15#include <linux/gfp.h>
16#include <linux/dma-mapping.h>
17
18#include "blk.h"
19#include "blk-rq-qos.h"
20#include "blk-wbt.h"
21
22void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
23{
24 q->rq_timeout = timeout;
25}
26EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
27
28/**
29 * blk_set_stacking_limits - set default limits for stacking devices
30 * @lim: the queue_limits structure to reset
31 *
32 * Prepare queue limits for applying limits from underlying devices using
33 * blk_stack_limits().
34 */
35void blk_set_stacking_limits(struct queue_limits *lim)
36{
37 memset(lim, 0, sizeof(*lim));
38 lim->logical_block_size = SECTOR_SIZE;
39 lim->physical_block_size = SECTOR_SIZE;
40 lim->io_min = SECTOR_SIZE;
41 lim->discard_granularity = SECTOR_SIZE;
42 lim->dma_alignment = SECTOR_SIZE - 1;
43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
44
45 /* Inherit limits from component devices */
46 lim->max_segments = USHRT_MAX;
47 lim->max_discard_segments = USHRT_MAX;
48 lim->max_hw_sectors = UINT_MAX;
49 lim->max_segment_size = UINT_MAX;
50 lim->max_sectors = UINT_MAX;
51 lim->max_dev_sectors = UINT_MAX;
52 lim->max_write_zeroes_sectors = UINT_MAX;
53 lim->max_zone_append_sectors = UINT_MAX;
54 lim->max_user_discard_sectors = UINT_MAX;
55}
56EXPORT_SYMBOL(blk_set_stacking_limits);
57
58static void blk_apply_bdi_limits(struct backing_dev_info *bdi,
59 struct queue_limits *lim)
60{
61 /*
62 * For read-ahead of large files to be effective, we need to read ahead
63 * at least twice the optimal I/O size.
64 */
65 bdi->ra_pages = max(lim->io_opt * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
66 bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT;
67}
68
69static int blk_validate_zoned_limits(struct queue_limits *lim)
70{
71 if (!lim->zoned) {
72 if (WARN_ON_ONCE(lim->max_open_zones) ||
73 WARN_ON_ONCE(lim->max_active_zones) ||
74 WARN_ON_ONCE(lim->zone_write_granularity) ||
75 WARN_ON_ONCE(lim->max_zone_append_sectors))
76 return -EINVAL;
77 return 0;
78 }
79
80 if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)))
81 return -EINVAL;
82
83 if (lim->zone_write_granularity < lim->logical_block_size)
84 lim->zone_write_granularity = lim->logical_block_size;
85
86 if (lim->max_zone_append_sectors) {
87 /*
88 * The Zone Append size is limited by the maximum I/O size
89 * and the zone size given that it can't span zones.
90 */
91 lim->max_zone_append_sectors =
92 min3(lim->max_hw_sectors,
93 lim->max_zone_append_sectors,
94 lim->chunk_sectors);
95 }
96
97 return 0;
98}
99
100/*
101 * Check that the limits in lim are valid, initialize defaults for unset
102 * values, and cap values based on others where needed.
103 */
104static int blk_validate_limits(struct queue_limits *lim)
105{
106 unsigned int max_hw_sectors;
107
108 /*
109 * Unless otherwise specified, default to 512 byte logical blocks and a
110 * physical block size equal to the logical block size.
111 */
112 if (!lim->logical_block_size)
113 lim->logical_block_size = SECTOR_SIZE;
114 if (lim->physical_block_size < lim->logical_block_size)
115 lim->physical_block_size = lim->logical_block_size;
116
117 /*
118 * The minimum I/O size defaults to the physical block size unless
119 * explicitly overridden.
120 */
121 if (lim->io_min < lim->physical_block_size)
122 lim->io_min = lim->physical_block_size;
123
124 /*
125 * max_hw_sectors has a somewhat weird default for historical reason,
126 * but driver really should set their own instead of relying on this
127 * value.
128 *
129 * The block layer relies on the fact that every driver can
130 * handle at lest a page worth of data per I/O, and needs the value
131 * aligned to the logical block size.
132 */
133 if (!lim->max_hw_sectors)
134 lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
135 if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
136 return -EINVAL;
137 lim->max_hw_sectors = round_down(lim->max_hw_sectors,
138 lim->logical_block_size >> SECTOR_SHIFT);
139
140 /*
141 * The actual max_sectors value is a complex beast and also takes the
142 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
143 * value into account. The ->max_sectors value is always calculated
144 * from these, so directly setting it won't have any effect.
145 */
146 max_hw_sectors = min_not_zero(lim->max_hw_sectors,
147 lim->max_dev_sectors);
148 if (lim->max_user_sectors) {
149 if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
150 return -EINVAL;
151 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
152 } else {
153 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
154 }
155 lim->max_sectors = round_down(lim->max_sectors,
156 lim->logical_block_size >> SECTOR_SHIFT);
157
158 /*
159 * Random default for the maximum number of segments. Driver should not
160 * rely on this and set their own.
161 */
162 if (!lim->max_segments)
163 lim->max_segments = BLK_MAX_SEGMENTS;
164
165 lim->max_discard_sectors =
166 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
167
168 if (!lim->max_discard_segments)
169 lim->max_discard_segments = 1;
170
171 if (lim->discard_granularity < lim->physical_block_size)
172 lim->discard_granularity = lim->physical_block_size;
173
174 /*
175 * By default there is no limit on the segment boundary alignment,
176 * but if there is one it can't be smaller than the page size as
177 * that would break all the normal I/O patterns.
178 */
179 if (!lim->seg_boundary_mask)
180 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
181 if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
182 return -EINVAL;
183
184 /*
185 * Stacking device may have both virtual boundary and max segment
186 * size limit, so allow this setting now, and long-term the two
187 * might need to move out of stacking limits since we have immutable
188 * bvec and lower layer bio splitting is supposed to handle the two
189 * correctly.
190 */
191 if (lim->virt_boundary_mask) {
192 if (!lim->max_segment_size)
193 lim->max_segment_size = UINT_MAX;
194 } else {
195 /*
196 * The maximum segment size has an odd historic 64k default that
197 * drivers probably should override. Just like the I/O size we
198 * require drivers to at least handle a full page per segment.
199 */
200 if (!lim->max_segment_size)
201 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
202 if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
203 return -EINVAL;
204 }
205
206 /*
207 * We require drivers to at least do logical block aligned I/O, but
208 * historically could not check for that due to the separate calls
209 * to set the limits. Once the transition is finished the check
210 * below should be narrowed down to check the logical block size.
211 */
212 if (!lim->dma_alignment)
213 lim->dma_alignment = SECTOR_SIZE - 1;
214 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
215 return -EINVAL;
216
217 if (lim->alignment_offset) {
218 lim->alignment_offset &= (lim->physical_block_size - 1);
219 lim->misaligned = 0;
220 }
221
222 return blk_validate_zoned_limits(lim);
223}
224
225/*
226 * Set the default limits for a newly allocated queue. @lim contains the
227 * initial limits set by the driver, which could be no limit in which case
228 * all fields are cleared to zero.
229 */
230int blk_set_default_limits(struct queue_limits *lim)
231{
232 /*
233 * Most defaults are set by capping the bounds in blk_validate_limits,
234 * but max_user_discard_sectors is special and needs an explicit
235 * initialization to the max value here.
236 */
237 lim->max_user_discard_sectors = UINT_MAX;
238 return blk_validate_limits(lim);
239}
240
241/**
242 * queue_limits_commit_update - commit an atomic update of queue limits
243 * @q: queue to update
244 * @lim: limits to apply
245 *
246 * Apply the limits in @lim that were obtained from queue_limits_start_update()
247 * and updated by the caller to @q.
248 *
249 * Returns 0 if successful, else a negative error code.
250 */
251int queue_limits_commit_update(struct request_queue *q,
252 struct queue_limits *lim)
253 __releases(q->limits_lock)
254{
255 int error = blk_validate_limits(lim);
256
257 if (!error) {
258 q->limits = *lim;
259 if (q->disk)
260 blk_apply_bdi_limits(q->disk->bdi, lim);
261 }
262 mutex_unlock(&q->limits_lock);
263 return error;
264}
265EXPORT_SYMBOL_GPL(queue_limits_commit_update);
266
267/**
268 * queue_limits_set - apply queue limits to queue
269 * @q: queue to update
270 * @lim: limits to apply
271 *
272 * Apply the limits in @lim that were freshly initialized to @q.
273 * To update existing limits use queue_limits_start_update() and
274 * queue_limits_commit_update() instead.
275 *
276 * Returns 0 if successful, else a negative error code.
277 */
278int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
279{
280 mutex_lock(&q->limits_lock);
281 return queue_limits_commit_update(q, lim);
282}
283EXPORT_SYMBOL_GPL(queue_limits_set);
284
285/**
286 * blk_queue_bounce_limit - set bounce buffer limit for queue
287 * @q: the request queue for the device
288 * @bounce: bounce limit to enforce
289 *
290 * Description:
291 * Force bouncing for ISA DMA ranges or highmem.
292 *
293 * DEPRECATED, don't use in new code.
294 **/
295void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
296{
297 q->limits.bounce = bounce;
298}
299EXPORT_SYMBOL(blk_queue_bounce_limit);
300
301/**
302 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
303 * @q: the request queue for the device
304 * @max_hw_sectors: max hardware sectors in the usual 512b unit
305 *
306 * Description:
307 * Enables a low level driver to set a hard upper limit,
308 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
309 * the device driver based upon the capabilities of the I/O
310 * controller.
311 *
312 * max_dev_sectors is a hard limit imposed by the storage device for
313 * READ/WRITE requests. It is set by the disk driver.
314 *
315 * max_sectors is a soft limit imposed by the block layer for
316 * filesystem type requests. This value can be overridden on a
317 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
318 * The soft limit can not exceed max_hw_sectors.
319 **/
320void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
321{
322 struct queue_limits *limits = &q->limits;
323 unsigned int max_sectors;
324
325 if ((max_hw_sectors << 9) < PAGE_SIZE) {
326 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
327 pr_info("%s: set to minimum %u\n", __func__, max_hw_sectors);
328 }
329
330 max_hw_sectors = round_down(max_hw_sectors,
331 limits->logical_block_size >> SECTOR_SHIFT);
332 limits->max_hw_sectors = max_hw_sectors;
333
334 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
335
336 if (limits->max_user_sectors)
337 max_sectors = min(max_sectors, limits->max_user_sectors);
338 else
339 max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS_CAP);
340
341 max_sectors = round_down(max_sectors,
342 limits->logical_block_size >> SECTOR_SHIFT);
343 limits->max_sectors = max_sectors;
344
345 if (!q->disk)
346 return;
347 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
348}
349EXPORT_SYMBOL(blk_queue_max_hw_sectors);
350
351/**
352 * blk_queue_chunk_sectors - set size of the chunk for this queue
353 * @q: the request queue for the device
354 * @chunk_sectors: chunk sectors in the usual 512b unit
355 *
356 * Description:
357 * If a driver doesn't want IOs to cross a given chunk size, it can set
358 * this limit and prevent merging across chunks. Note that the block layer
359 * must accept a page worth of data at any offset. So if the crossing of
360 * chunks is a hard limitation in the driver, it must still be prepared
361 * to split single page bios.
362 **/
363void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
364{
365 q->limits.chunk_sectors = chunk_sectors;
366}
367EXPORT_SYMBOL(blk_queue_chunk_sectors);
368
369/**
370 * blk_queue_max_discard_sectors - set max sectors for a single discard
371 * @q: the request queue for the device
372 * @max_discard_sectors: maximum number of sectors to discard
373 **/
374void blk_queue_max_discard_sectors(struct request_queue *q,
375 unsigned int max_discard_sectors)
376{
377 struct queue_limits *lim = &q->limits;
378
379 lim->max_hw_discard_sectors = max_discard_sectors;
380 lim->max_discard_sectors =
381 min(max_discard_sectors, lim->max_user_discard_sectors);
382}
383EXPORT_SYMBOL(blk_queue_max_discard_sectors);
384
385/**
386 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
387 * @q: the request queue for the device
388 * @max_sectors: maximum number of sectors to secure_erase
389 **/
390void blk_queue_max_secure_erase_sectors(struct request_queue *q,
391 unsigned int max_sectors)
392{
393 q->limits.max_secure_erase_sectors = max_sectors;
394}
395EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
396
397/**
398 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
399 * write zeroes
400 * @q: the request queue for the device
401 * @max_write_zeroes_sectors: maximum number of sectors to write per command
402 **/
403void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
404 unsigned int max_write_zeroes_sectors)
405{
406 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
407}
408EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
409
410/**
411 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
412 * @q: the request queue for the device
413 * @max_zone_append_sectors: maximum number of sectors to write per command
414 **/
415void blk_queue_max_zone_append_sectors(struct request_queue *q,
416 unsigned int max_zone_append_sectors)
417{
418 unsigned int max_sectors;
419
420 if (WARN_ON(!blk_queue_is_zoned(q)))
421 return;
422
423 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
424 max_sectors = min(q->limits.chunk_sectors, max_sectors);
425
426 /*
427 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
428 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
429 * or the max_hw_sectors limit not set.
430 */
431 WARN_ON(!max_sectors);
432
433 q->limits.max_zone_append_sectors = max_sectors;
434}
435EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
436
437/**
438 * blk_queue_max_segments - set max hw segments for a request for this queue
439 * @q: the request queue for the device
440 * @max_segments: max number of segments
441 *
442 * Description:
443 * Enables a low level driver to set an upper limit on the number of
444 * hw data segments in a request.
445 **/
446void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
447{
448 if (!max_segments) {
449 max_segments = 1;
450 pr_info("%s: set to minimum %u\n", __func__, max_segments);
451 }
452
453 q->limits.max_segments = max_segments;
454}
455EXPORT_SYMBOL(blk_queue_max_segments);
456
457/**
458 * blk_queue_max_discard_segments - set max segments for discard requests
459 * @q: the request queue for the device
460 * @max_segments: max number of segments
461 *
462 * Description:
463 * Enables a low level driver to set an upper limit on the number of
464 * segments in a discard request.
465 **/
466void blk_queue_max_discard_segments(struct request_queue *q,
467 unsigned short max_segments)
468{
469 q->limits.max_discard_segments = max_segments;
470}
471EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
472
473/**
474 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
475 * @q: the request queue for the device
476 * @max_size: max size of segment in bytes
477 *
478 * Description:
479 * Enables a low level driver to set an upper limit on the size of a
480 * coalesced segment
481 **/
482void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
483{
484 if (max_size < PAGE_SIZE) {
485 max_size = PAGE_SIZE;
486 pr_info("%s: set to minimum %u\n", __func__, max_size);
487 }
488
489 /* see blk_queue_virt_boundary() for the explanation */
490 WARN_ON_ONCE(q->limits.virt_boundary_mask);
491
492 q->limits.max_segment_size = max_size;
493}
494EXPORT_SYMBOL(blk_queue_max_segment_size);
495
496/**
497 * blk_queue_logical_block_size - set logical block size for the queue
498 * @q: the request queue for the device
499 * @size: the logical block size, in bytes
500 *
501 * Description:
502 * This should be set to the lowest possible block size that the
503 * storage device can address. The default of 512 covers most
504 * hardware.
505 **/
506void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
507{
508 struct queue_limits *limits = &q->limits;
509
510 limits->logical_block_size = size;
511
512 if (limits->discard_granularity < limits->logical_block_size)
513 limits->discard_granularity = limits->logical_block_size;
514
515 if (limits->physical_block_size < size)
516 limits->physical_block_size = size;
517
518 if (limits->io_min < limits->physical_block_size)
519 limits->io_min = limits->physical_block_size;
520
521 limits->max_hw_sectors =
522 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
523 limits->max_sectors =
524 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
525}
526EXPORT_SYMBOL(blk_queue_logical_block_size);
527
528/**
529 * blk_queue_physical_block_size - set physical block size for the queue
530 * @q: the request queue for the device
531 * @size: the physical block size, in bytes
532 *
533 * Description:
534 * This should be set to the lowest possible sector size that the
535 * hardware can operate on without reverting to read-modify-write
536 * operations.
537 */
538void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
539{
540 q->limits.physical_block_size = size;
541
542 if (q->limits.physical_block_size < q->limits.logical_block_size)
543 q->limits.physical_block_size = q->limits.logical_block_size;
544
545 if (q->limits.discard_granularity < q->limits.physical_block_size)
546 q->limits.discard_granularity = q->limits.physical_block_size;
547
548 if (q->limits.io_min < q->limits.physical_block_size)
549 q->limits.io_min = q->limits.physical_block_size;
550}
551EXPORT_SYMBOL(blk_queue_physical_block_size);
552
553/**
554 * blk_queue_zone_write_granularity - set zone write granularity for the queue
555 * @q: the request queue for the zoned device
556 * @size: the zone write granularity size, in bytes
557 *
558 * Description:
559 * This should be set to the lowest possible size allowing to write in
560 * sequential zones of a zoned block device.
561 */
562void blk_queue_zone_write_granularity(struct request_queue *q,
563 unsigned int size)
564{
565 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
566 return;
567
568 q->limits.zone_write_granularity = size;
569
570 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
571 q->limits.zone_write_granularity = q->limits.logical_block_size;
572}
573EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
574
575/**
576 * blk_queue_alignment_offset - set physical block alignment offset
577 * @q: the request queue for the device
578 * @offset: alignment offset in bytes
579 *
580 * Description:
581 * Some devices are naturally misaligned to compensate for things like
582 * the legacy DOS partition table 63-sector offset. Low-level drivers
583 * should call this function for devices whose first sector is not
584 * naturally aligned.
585 */
586void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
587{
588 q->limits.alignment_offset =
589 offset & (q->limits.physical_block_size - 1);
590 q->limits.misaligned = 0;
591}
592EXPORT_SYMBOL(blk_queue_alignment_offset);
593
594void disk_update_readahead(struct gendisk *disk)
595{
596 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
597}
598EXPORT_SYMBOL_GPL(disk_update_readahead);
599
600/**
601 * blk_limits_io_min - set minimum request size for a device
602 * @limits: the queue limits
603 * @min: smallest I/O size in bytes
604 *
605 * Description:
606 * Some devices have an internal block size bigger than the reported
607 * hardware sector size. This function can be used to signal the
608 * smallest I/O the device can perform without incurring a performance
609 * penalty.
610 */
611void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
612{
613 limits->io_min = min;
614
615 if (limits->io_min < limits->logical_block_size)
616 limits->io_min = limits->logical_block_size;
617
618 if (limits->io_min < limits->physical_block_size)
619 limits->io_min = limits->physical_block_size;
620}
621EXPORT_SYMBOL(blk_limits_io_min);
622
623/**
624 * blk_queue_io_min - set minimum request size for the queue
625 * @q: the request queue for the device
626 * @min: smallest I/O size in bytes
627 *
628 * Description:
629 * Storage devices may report a granularity or preferred minimum I/O
630 * size which is the smallest request the device can perform without
631 * incurring a performance penalty. For disk drives this is often the
632 * physical block size. For RAID arrays it is often the stripe chunk
633 * size. A properly aligned multiple of minimum_io_size is the
634 * preferred request size for workloads where a high number of I/O
635 * operations is desired.
636 */
637void blk_queue_io_min(struct request_queue *q, unsigned int min)
638{
639 blk_limits_io_min(&q->limits, min);
640}
641EXPORT_SYMBOL(blk_queue_io_min);
642
643/**
644 * blk_limits_io_opt - set optimal request size for a device
645 * @limits: the queue limits
646 * @opt: smallest I/O size in bytes
647 *
648 * Description:
649 * Storage devices may report an optimal I/O size, which is the
650 * device's preferred unit for sustained I/O. This is rarely reported
651 * for disk drives. For RAID arrays it is usually the stripe width or
652 * the internal track size. A properly aligned multiple of
653 * optimal_io_size is the preferred request size for workloads where
654 * sustained throughput is desired.
655 */
656void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
657{
658 limits->io_opt = opt;
659}
660EXPORT_SYMBOL(blk_limits_io_opt);
661
662/**
663 * blk_queue_io_opt - set optimal request size for the queue
664 * @q: the request queue for the device
665 * @opt: optimal request size in bytes
666 *
667 * Description:
668 * Storage devices may report an optimal I/O size, which is the
669 * device's preferred unit for sustained I/O. This is rarely reported
670 * for disk drives. For RAID arrays it is usually the stripe width or
671 * the internal track size. A properly aligned multiple of
672 * optimal_io_size is the preferred request size for workloads where
673 * sustained throughput is desired.
674 */
675void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
676{
677 blk_limits_io_opt(&q->limits, opt);
678 if (!q->disk)
679 return;
680 q->disk->bdi->ra_pages =
681 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
682}
683EXPORT_SYMBOL(blk_queue_io_opt);
684
685static int queue_limit_alignment_offset(const struct queue_limits *lim,
686 sector_t sector)
687{
688 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
689 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
690 << SECTOR_SHIFT;
691
692 return (granularity + lim->alignment_offset - alignment) % granularity;
693}
694
695static unsigned int queue_limit_discard_alignment(
696 const struct queue_limits *lim, sector_t sector)
697{
698 unsigned int alignment, granularity, offset;
699
700 if (!lim->max_discard_sectors)
701 return 0;
702
703 /* Why are these in bytes, not sectors? */
704 alignment = lim->discard_alignment >> SECTOR_SHIFT;
705 granularity = lim->discard_granularity >> SECTOR_SHIFT;
706 if (!granularity)
707 return 0;
708
709 /* Offset of the partition start in 'granularity' sectors */
710 offset = sector_div(sector, granularity);
711
712 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
713 offset = (granularity + alignment - offset) % granularity;
714
715 /* Turn it back into bytes, gaah */
716 return offset << SECTOR_SHIFT;
717}
718
719static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
720{
721 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
722 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
723 sectors = PAGE_SIZE >> SECTOR_SHIFT;
724 return sectors;
725}
726
727/**
728 * blk_stack_limits - adjust queue_limits for stacked devices
729 * @t: the stacking driver limits (top device)
730 * @b: the underlying queue limits (bottom, component device)
731 * @start: first data sector within component device
732 *
733 * Description:
734 * This function is used by stacking drivers like MD and DM to ensure
735 * that all component devices have compatible block sizes and
736 * alignments. The stacking driver must provide a queue_limits
737 * struct (top) and then iteratively call the stacking function for
738 * all component (bottom) devices. The stacking function will
739 * attempt to combine the values and ensure proper alignment.
740 *
741 * Returns 0 if the top and bottom queue_limits are compatible. The
742 * top device's block sizes and alignment offsets may be adjusted to
743 * ensure alignment with the bottom device. If no compatible sizes
744 * and alignments exist, -1 is returned and the resulting top
745 * queue_limits will have the misaligned flag set to indicate that
746 * the alignment_offset is undefined.
747 */
748int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
749 sector_t start)
750{
751 unsigned int top, bottom, alignment, ret = 0;
752
753 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
754 t->max_user_sectors = min_not_zero(t->max_user_sectors,
755 b->max_user_sectors);
756 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
757 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
758 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
759 b->max_write_zeroes_sectors);
760 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
761 b->max_zone_append_sectors);
762 t->bounce = max(t->bounce, b->bounce);
763
764 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
765 b->seg_boundary_mask);
766 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
767 b->virt_boundary_mask);
768
769 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
770 t->max_discard_segments = min_not_zero(t->max_discard_segments,
771 b->max_discard_segments);
772 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
773 b->max_integrity_segments);
774
775 t->max_segment_size = min_not_zero(t->max_segment_size,
776 b->max_segment_size);
777
778 t->misaligned |= b->misaligned;
779
780 alignment = queue_limit_alignment_offset(b, start);
781
782 /* Bottom device has different alignment. Check that it is
783 * compatible with the current top alignment.
784 */
785 if (t->alignment_offset != alignment) {
786
787 top = max(t->physical_block_size, t->io_min)
788 + t->alignment_offset;
789 bottom = max(b->physical_block_size, b->io_min) + alignment;
790
791 /* Verify that top and bottom intervals line up */
792 if (max(top, bottom) % min(top, bottom)) {
793 t->misaligned = 1;
794 ret = -1;
795 }
796 }
797
798 t->logical_block_size = max(t->logical_block_size,
799 b->logical_block_size);
800
801 t->physical_block_size = max(t->physical_block_size,
802 b->physical_block_size);
803
804 t->io_min = max(t->io_min, b->io_min);
805 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
806 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
807
808 /* Set non-power-of-2 compatible chunk_sectors boundary */
809 if (b->chunk_sectors)
810 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
811
812 /* Physical block size a multiple of the logical block size? */
813 if (t->physical_block_size & (t->logical_block_size - 1)) {
814 t->physical_block_size = t->logical_block_size;
815 t->misaligned = 1;
816 ret = -1;
817 }
818
819 /* Minimum I/O a multiple of the physical block size? */
820 if (t->io_min & (t->physical_block_size - 1)) {
821 t->io_min = t->physical_block_size;
822 t->misaligned = 1;
823 ret = -1;
824 }
825
826 /* Optimal I/O a multiple of the physical block size? */
827 if (t->io_opt & (t->physical_block_size - 1)) {
828 t->io_opt = 0;
829 t->misaligned = 1;
830 ret = -1;
831 }
832
833 /* chunk_sectors a multiple of the physical block size? */
834 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
835 t->chunk_sectors = 0;
836 t->misaligned = 1;
837 ret = -1;
838 }
839
840 t->raid_partial_stripes_expensive =
841 max(t->raid_partial_stripes_expensive,
842 b->raid_partial_stripes_expensive);
843
844 /* Find lowest common alignment_offset */
845 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
846 % max(t->physical_block_size, t->io_min);
847
848 /* Verify that new alignment_offset is on a logical block boundary */
849 if (t->alignment_offset & (t->logical_block_size - 1)) {
850 t->misaligned = 1;
851 ret = -1;
852 }
853
854 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
855 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
856 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
857
858 /* Discard alignment and granularity */
859 if (b->discard_granularity) {
860 alignment = queue_limit_discard_alignment(b, start);
861
862 if (t->discard_granularity != 0 &&
863 t->discard_alignment != alignment) {
864 top = t->discard_granularity + t->discard_alignment;
865 bottom = b->discard_granularity + alignment;
866
867 /* Verify that top and bottom intervals line up */
868 if ((max(top, bottom) % min(top, bottom)) != 0)
869 t->discard_misaligned = 1;
870 }
871
872 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
873 b->max_discard_sectors);
874 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
875 b->max_hw_discard_sectors);
876 t->discard_granularity = max(t->discard_granularity,
877 b->discard_granularity);
878 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
879 t->discard_granularity;
880 }
881 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
882 b->max_secure_erase_sectors);
883 t->zone_write_granularity = max(t->zone_write_granularity,
884 b->zone_write_granularity);
885 t->zoned = max(t->zoned, b->zoned);
886 if (!t->zoned) {
887 t->zone_write_granularity = 0;
888 t->max_zone_append_sectors = 0;
889 }
890 return ret;
891}
892EXPORT_SYMBOL(blk_stack_limits);
893
894/**
895 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
896 * @t: the stacking driver limits (top device)
897 * @bdev: the underlying block device (bottom)
898 * @offset: offset to beginning of data within component device
899 * @pfx: prefix to use for warnings logged
900 *
901 * Description:
902 * This function is used by stacking drivers like MD and DM to ensure
903 * that all component devices have compatible block sizes and
904 * alignments. The stacking driver must provide a queue_limits
905 * struct (top) and then iteratively call the stacking function for
906 * all component (bottom) devices. The stacking function will
907 * attempt to combine the values and ensure proper alignment.
908 */
909void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
910 sector_t offset, const char *pfx)
911{
912 if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits,
913 get_start_sect(bdev) + offset))
914 pr_notice("%s: Warning: Device %pg is misaligned\n",
915 pfx, bdev);
916}
917EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
918
919/**
920 * blk_queue_update_dma_pad - update pad mask
921 * @q: the request queue for the device
922 * @mask: pad mask
923 *
924 * Update dma pad mask.
925 *
926 * Appending pad buffer to a request modifies the last entry of a
927 * scatter list such that it includes the pad buffer.
928 **/
929void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
930{
931 if (mask > q->dma_pad_mask)
932 q->dma_pad_mask = mask;
933}
934EXPORT_SYMBOL(blk_queue_update_dma_pad);
935
936/**
937 * blk_queue_segment_boundary - set boundary rules for segment merging
938 * @q: the request queue for the device
939 * @mask: the memory boundary mask
940 **/
941void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
942{
943 if (mask < PAGE_SIZE - 1) {
944 mask = PAGE_SIZE - 1;
945 pr_info("%s: set to minimum %lx\n", __func__, mask);
946 }
947
948 q->limits.seg_boundary_mask = mask;
949}
950EXPORT_SYMBOL(blk_queue_segment_boundary);
951
952/**
953 * blk_queue_virt_boundary - set boundary rules for bio merging
954 * @q: the request queue for the device
955 * @mask: the memory boundary mask
956 **/
957void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
958{
959 q->limits.virt_boundary_mask = mask;
960
961 /*
962 * Devices that require a virtual boundary do not support scatter/gather
963 * I/O natively, but instead require a descriptor list entry for each
964 * page (which might not be idential to the Linux PAGE_SIZE). Because
965 * of that they are not limited by our notion of "segment size".
966 */
967 if (mask)
968 q->limits.max_segment_size = UINT_MAX;
969}
970EXPORT_SYMBOL(blk_queue_virt_boundary);
971
972/**
973 * blk_queue_dma_alignment - set dma length and memory alignment
974 * @q: the request queue for the device
975 * @mask: alignment mask
976 *
977 * description:
978 * set required memory and length alignment for direct dma transactions.
979 * this is used when building direct io requests for the queue.
980 *
981 **/
982void blk_queue_dma_alignment(struct request_queue *q, int mask)
983{
984 q->limits.dma_alignment = mask;
985}
986EXPORT_SYMBOL(blk_queue_dma_alignment);
987
988/**
989 * blk_queue_update_dma_alignment - update dma length and memory alignment
990 * @q: the request queue for the device
991 * @mask: alignment mask
992 *
993 * description:
994 * update required memory and length alignment for direct dma transactions.
995 * If the requested alignment is larger than the current alignment, then
996 * the current queue alignment is updated to the new value, otherwise it
997 * is left alone. The design of this is to allow multiple objects
998 * (driver, device, transport etc) to set their respective
999 * alignments without having them interfere.
1000 *
1001 **/
1002void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
1003{
1004 BUG_ON(mask > PAGE_SIZE);
1005
1006 if (mask > q->limits.dma_alignment)
1007 q->limits.dma_alignment = mask;
1008}
1009EXPORT_SYMBOL(blk_queue_update_dma_alignment);
1010
1011/**
1012 * blk_set_queue_depth - tell the block layer about the device queue depth
1013 * @q: the request queue for the device
1014 * @depth: queue depth
1015 *
1016 */
1017void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
1018{
1019 q->queue_depth = depth;
1020 rq_qos_queue_depth_changed(q);
1021}
1022EXPORT_SYMBOL(blk_set_queue_depth);
1023
1024/**
1025 * blk_queue_write_cache - configure queue's write cache
1026 * @q: the request queue for the device
1027 * @wc: write back cache on or off
1028 * @fua: device supports FUA writes, if true
1029 *
1030 * Tell the block layer about the write cache of @q.
1031 */
1032void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
1033{
1034 if (wc) {
1035 blk_queue_flag_set(QUEUE_FLAG_HW_WC, q);
1036 blk_queue_flag_set(QUEUE_FLAG_WC, q);
1037 } else {
1038 blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q);
1039 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
1040 }
1041 if (fua)
1042 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
1043 else
1044 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
1045}
1046EXPORT_SYMBOL_GPL(blk_queue_write_cache);
1047
1048/**
1049 * blk_queue_required_elevator_features - Set a queue required elevator features
1050 * @q: the request queue for the target device
1051 * @features: Required elevator features OR'ed together
1052 *
1053 * Tell the block layer that for the device controlled through @q, only the
1054 * only elevators that can be used are those that implement at least the set of
1055 * features specified by @features.
1056 */
1057void blk_queue_required_elevator_features(struct request_queue *q,
1058 unsigned int features)
1059{
1060 q->required_elevator_features = features;
1061}
1062EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
1063
1064/**
1065 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
1066 * @q: the request queue for the device
1067 * @dev: the device pointer for dma
1068 *
1069 * Tell the block layer about merging the segments by dma map of @q.
1070 */
1071bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
1072 struct device *dev)
1073{
1074 unsigned long boundary = dma_get_merge_boundary(dev);
1075
1076 if (!boundary)
1077 return false;
1078
1079 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
1080 blk_queue_virt_boundary(q, boundary);
1081
1082 return true;
1083}
1084EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
1085
1086/**
1087 * disk_set_zoned - inidicate a zoned device
1088 * @disk: gendisk to configure
1089 */
1090void disk_set_zoned(struct gendisk *disk)
1091{
1092 struct request_queue *q = disk->queue;
1093
1094 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
1095
1096 /*
1097 * Set the zone write granularity to the device logical block
1098 * size by default. The driver can change this value if needed.
1099 */
1100 q->limits.zoned = true;
1101 blk_queue_zone_write_granularity(q, queue_logical_block_size(q));
1102}
1103EXPORT_SYMBOL_GPL(disk_set_zoned);
1104
1105int bdev_alignment_offset(struct block_device *bdev)
1106{
1107 struct request_queue *q = bdev_get_queue(bdev);
1108
1109 if (q->limits.misaligned)
1110 return -1;
1111 if (bdev_is_partition(bdev))
1112 return queue_limit_alignment_offset(&q->limits,
1113 bdev->bd_start_sect);
1114 return q->limits.alignment_offset;
1115}
1116EXPORT_SYMBOL_GPL(bdev_alignment_offset);
1117
1118unsigned int bdev_discard_alignment(struct block_device *bdev)
1119{
1120 struct request_queue *q = bdev_get_queue(bdev);
1121
1122 if (bdev_is_partition(bdev))
1123 return queue_limit_discard_alignment(&q->limits,
1124 bdev->bd_start_sect);
1125 return q->limits.discard_alignment;
1126}
1127EXPORT_SYMBOL_GPL(bdev_discard_alignment);