1 // SPDX-License-Identifier: GPL-2.0
3 * Functions related to setting various queue properties from drivers
5 #include <linux/kernel.h>
6 #include <linux/module.h>
7 #include <linux/init.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>
19 #include "blk-rq-qos.h"
22 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
24 q->rq_timeout = timeout;
26 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
29 * blk_set_stacking_limits - set default limits for stacking devices
30 * @lim: the queue_limits structure to reset
32 * Prepare queue limits for applying limits from underlying devices using
35 void blk_set_stacking_limits(struct queue_limits *lim)
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;
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;
56 EXPORT_SYMBOL(blk_set_stacking_limits);
58 static void blk_apply_bdi_limits(struct backing_dev_info *bdi,
59 struct queue_limits *lim)
62 * For read-ahead of large files to be effective, we need to read ahead
63 * at least twice the optimal I/O size.
65 bdi->ra_pages = max(lim->io_opt * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
66 bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT;
69 static int blk_validate_zoned_limits(struct queue_limits *lim)
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))
80 if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)))
83 if (lim->zone_write_granularity < lim->logical_block_size)
84 lim->zone_write_granularity = lim->logical_block_size;
86 if (lim->max_zone_append_sectors) {
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.
91 lim->max_zone_append_sectors =
92 min3(lim->max_hw_sectors,
93 lim->max_zone_append_sectors,
101 * Check that the limits in lim are valid, initialize defaults for unset
102 * values, and cap values based on others where needed.
104 static int blk_validate_limits(struct queue_limits *lim)
106 unsigned int max_hw_sectors;
109 * Unless otherwise specified, default to 512 byte logical blocks and a
110 * physical block size equal to the logical block size.
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;
118 * The minimum I/O size defaults to the physical block size unless
119 * explicitly overridden.
121 if (lim->io_min < lim->physical_block_size)
122 lim->io_min = lim->physical_block_size;
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
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.
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))
137 lim->max_hw_sectors = round_down(lim->max_hw_sectors,
138 lim->logical_block_size >> SECTOR_SHIFT);
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.
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)
151 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
153 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
155 lim->max_sectors = round_down(lim->max_sectors,
156 lim->logical_block_size >> SECTOR_SHIFT);
159 * Random default for the maximum number of segments. Driver should not
160 * rely on this and set their own.
162 if (!lim->max_segments)
163 lim->max_segments = BLK_MAX_SEGMENTS;
165 lim->max_discard_sectors =
166 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
168 if (!lim->max_discard_segments)
169 lim->max_discard_segments = 1;
171 if (lim->discard_granularity < lim->physical_block_size)
172 lim->discard_granularity = lim->physical_block_size;
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.
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))
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
191 if (lim->virt_boundary_mask) {
192 if (!lim->max_segment_size)
193 lim->max_segment_size = UINT_MAX;
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.
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))
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.
212 if (!lim->dma_alignment)
213 lim->dma_alignment = SECTOR_SIZE - 1;
214 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
217 if (lim->alignment_offset) {
218 lim->alignment_offset &= (lim->physical_block_size - 1);
222 return blk_validate_zoned_limits(lim);
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.
230 int blk_set_default_limits(struct queue_limits *lim)
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.
237 lim->max_user_discard_sectors = UINT_MAX;
238 return blk_validate_limits(lim);
242 * queue_limits_commit_update - commit an atomic update of queue limits
243 * @q: queue to update
244 * @lim: limits to apply
246 * Apply the limits in @lim that were obtained from queue_limits_start_update()
247 * and updated by the caller to @q.
249 * Returns 0 if successful, else a negative error code.
251 int queue_limits_commit_update(struct request_queue *q,
252 struct queue_limits *lim)
253 __releases(q->limits_lock)
255 int error = blk_validate_limits(lim);
260 blk_apply_bdi_limits(q->disk->bdi, lim);
262 mutex_unlock(&q->limits_lock);
265 EXPORT_SYMBOL_GPL(queue_limits_commit_update);
268 * queue_limits_set - apply queue limits to queue
269 * @q: queue to update
270 * @lim: limits to apply
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.
276 * Returns 0 if successful, else a negative error code.
278 int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
280 mutex_lock(&q->limits_lock);
281 return queue_limits_commit_update(q, lim);
283 EXPORT_SYMBOL_GPL(queue_limits_set);
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
291 * Force bouncing for ISA DMA ranges or highmem.
293 * DEPRECATED, don't use in new code.
295 void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
297 q->limits.bounce = bounce;
299 EXPORT_SYMBOL(blk_queue_bounce_limit);
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
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
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.
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.
320 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
322 struct queue_limits *limits = &q->limits;
323 unsigned int max_sectors;
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);
330 max_hw_sectors = round_down(max_hw_sectors,
331 limits->logical_block_size >> SECTOR_SHIFT);
332 limits->max_hw_sectors = max_hw_sectors;
334 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
336 if (limits->max_user_sectors)
337 max_sectors = min(max_sectors, limits->max_user_sectors);
339 max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS_CAP);
341 max_sectors = round_down(max_sectors,
342 limits->logical_block_size >> SECTOR_SHIFT);
343 limits->max_sectors = max_sectors;
347 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
349 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
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
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.
363 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
365 q->limits.chunk_sectors = chunk_sectors;
367 EXPORT_SYMBOL(blk_queue_chunk_sectors);
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
374 void blk_queue_max_discard_sectors(struct request_queue *q,
375 unsigned int max_discard_sectors)
377 struct queue_limits *lim = &q->limits;
379 lim->max_hw_discard_sectors = max_discard_sectors;
380 lim->max_discard_sectors =
381 min(max_discard_sectors, lim->max_user_discard_sectors);
383 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
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
390 void blk_queue_max_secure_erase_sectors(struct request_queue *q,
391 unsigned int max_sectors)
393 q->limits.max_secure_erase_sectors = max_sectors;
395 EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
398 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
400 * @q: the request queue for the device
401 * @max_write_zeroes_sectors: maximum number of sectors to write per command
403 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
404 unsigned int max_write_zeroes_sectors)
406 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
408 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
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
415 void blk_queue_max_zone_append_sectors(struct request_queue *q,
416 unsigned int max_zone_append_sectors)
418 unsigned int max_sectors;
420 if (WARN_ON(!blk_queue_is_zoned(q)))
423 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
424 max_sectors = min(q->limits.chunk_sectors, max_sectors);
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.
431 WARN_ON(!max_sectors);
433 q->limits.max_zone_append_sectors = max_sectors;
435 EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
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
443 * Enables a low level driver to set an upper limit on the number of
444 * hw data segments in a request.
446 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
450 pr_info("%s: set to minimum %u\n", __func__, max_segments);
453 q->limits.max_segments = max_segments;
455 EXPORT_SYMBOL(blk_queue_max_segments);
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
463 * Enables a low level driver to set an upper limit on the number of
464 * segments in a discard request.
466 void blk_queue_max_discard_segments(struct request_queue *q,
467 unsigned short max_segments)
469 q->limits.max_discard_segments = max_segments;
471 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
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
479 * Enables a low level driver to set an upper limit on the size of a
482 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
484 if (max_size < PAGE_SIZE) {
485 max_size = PAGE_SIZE;
486 pr_info("%s: set to minimum %u\n", __func__, max_size);
489 /* see blk_queue_virt_boundary() for the explanation */
490 WARN_ON_ONCE(q->limits.virt_boundary_mask);
492 q->limits.max_segment_size = max_size;
494 EXPORT_SYMBOL(blk_queue_max_segment_size);
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
502 * This should be set to the lowest possible block size that the
503 * storage device can address. The default of 512 covers most
506 void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
508 struct queue_limits *limits = &q->limits;
510 limits->logical_block_size = size;
512 if (limits->discard_granularity < limits->logical_block_size)
513 limits->discard_granularity = limits->logical_block_size;
515 if (limits->physical_block_size < size)
516 limits->physical_block_size = size;
518 if (limits->io_min < limits->physical_block_size)
519 limits->io_min = limits->physical_block_size;
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);
526 EXPORT_SYMBOL(blk_queue_logical_block_size);
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
534 * This should be set to the lowest possible sector size that the
535 * hardware can operate on without reverting to read-modify-write
538 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
540 q->limits.physical_block_size = size;
542 if (q->limits.physical_block_size < q->limits.logical_block_size)
543 q->limits.physical_block_size = q->limits.logical_block_size;
545 if (q->limits.discard_granularity < q->limits.physical_block_size)
546 q->limits.discard_granularity = q->limits.physical_block_size;
548 if (q->limits.io_min < q->limits.physical_block_size)
549 q->limits.io_min = q->limits.physical_block_size;
551 EXPORT_SYMBOL(blk_queue_physical_block_size);
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
559 * This should be set to the lowest possible size allowing to write in
560 * sequential zones of a zoned block device.
562 void blk_queue_zone_write_granularity(struct request_queue *q,
565 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
568 q->limits.zone_write_granularity = size;
570 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
571 q->limits.zone_write_granularity = q->limits.logical_block_size;
573 EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
576 * blk_queue_alignment_offset - set physical block alignment offset
577 * @q: the request queue for the device
578 * @offset: alignment offset in bytes
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
586 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
588 q->limits.alignment_offset =
589 offset & (q->limits.physical_block_size - 1);
590 q->limits.misaligned = 0;
592 EXPORT_SYMBOL(blk_queue_alignment_offset);
594 void disk_update_readahead(struct gendisk *disk)
596 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
598 EXPORT_SYMBOL_GPL(disk_update_readahead);
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
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
611 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
613 limits->io_min = min;
615 if (limits->io_min < limits->logical_block_size)
616 limits->io_min = limits->logical_block_size;
618 if (limits->io_min < limits->physical_block_size)
619 limits->io_min = limits->physical_block_size;
621 EXPORT_SYMBOL(blk_limits_io_min);
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
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.
637 void blk_queue_io_min(struct request_queue *q, unsigned int min)
639 blk_limits_io_min(&q->limits, min);
641 EXPORT_SYMBOL(blk_queue_io_min);
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
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.
656 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
658 limits->io_opt = opt;
660 EXPORT_SYMBOL(blk_limits_io_opt);
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
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.
675 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
677 blk_limits_io_opt(&q->limits, opt);
680 q->disk->bdi->ra_pages =
681 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
683 EXPORT_SYMBOL(blk_queue_io_opt);
685 static int queue_limit_alignment_offset(const struct queue_limits *lim,
688 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
689 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
692 return (granularity + lim->alignment_offset - alignment) % granularity;
695 static unsigned int queue_limit_discard_alignment(
696 const struct queue_limits *lim, sector_t sector)
698 unsigned int alignment, granularity, offset;
700 if (!lim->max_discard_sectors)
703 /* Why are these in bytes, not sectors? */
704 alignment = lim->discard_alignment >> SECTOR_SHIFT;
705 granularity = lim->discard_granularity >> SECTOR_SHIFT;
709 /* Offset of the partition start in 'granularity' sectors */
710 offset = sector_div(sector, granularity);
712 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
713 offset = (granularity + alignment - offset) % granularity;
715 /* Turn it back into bytes, gaah */
716 return offset << SECTOR_SHIFT;
719 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
721 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
722 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
723 sectors = PAGE_SIZE >> SECTOR_SHIFT;
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
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.
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.
748 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
751 unsigned int top, bottom, alignment, ret = 0;
753 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
754 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
755 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
756 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
757 b->max_write_zeroes_sectors);
758 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
759 b->max_zone_append_sectors);
760 t->bounce = max(t->bounce, b->bounce);
762 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
763 b->seg_boundary_mask);
764 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
765 b->virt_boundary_mask);
767 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
768 t->max_discard_segments = min_not_zero(t->max_discard_segments,
769 b->max_discard_segments);
770 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
771 b->max_integrity_segments);
773 t->max_segment_size = min_not_zero(t->max_segment_size,
774 b->max_segment_size);
776 t->misaligned |= b->misaligned;
778 alignment = queue_limit_alignment_offset(b, start);
780 /* Bottom device has different alignment. Check that it is
781 * compatible with the current top alignment.
783 if (t->alignment_offset != alignment) {
785 top = max(t->physical_block_size, t->io_min)
786 + t->alignment_offset;
787 bottom = max(b->physical_block_size, b->io_min) + alignment;
789 /* Verify that top and bottom intervals line up */
790 if (max(top, bottom) % min(top, bottom)) {
796 t->logical_block_size = max(t->logical_block_size,
797 b->logical_block_size);
799 t->physical_block_size = max(t->physical_block_size,
800 b->physical_block_size);
802 t->io_min = max(t->io_min, b->io_min);
803 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
804 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
806 /* Set non-power-of-2 compatible chunk_sectors boundary */
807 if (b->chunk_sectors)
808 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
810 /* Physical block size a multiple of the logical block size? */
811 if (t->physical_block_size & (t->logical_block_size - 1)) {
812 t->physical_block_size = t->logical_block_size;
817 /* Minimum I/O a multiple of the physical block size? */
818 if (t->io_min & (t->physical_block_size - 1)) {
819 t->io_min = t->physical_block_size;
824 /* Optimal I/O a multiple of the physical block size? */
825 if (t->io_opt & (t->physical_block_size - 1)) {
831 /* chunk_sectors a multiple of the physical block size? */
832 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
833 t->chunk_sectors = 0;
838 t->raid_partial_stripes_expensive =
839 max(t->raid_partial_stripes_expensive,
840 b->raid_partial_stripes_expensive);
842 /* Find lowest common alignment_offset */
843 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
844 % max(t->physical_block_size, t->io_min);
846 /* Verify that new alignment_offset is on a logical block boundary */
847 if (t->alignment_offset & (t->logical_block_size - 1)) {
852 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
853 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
854 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
856 /* Discard alignment and granularity */
857 if (b->discard_granularity) {
858 alignment = queue_limit_discard_alignment(b, start);
860 if (t->discard_granularity != 0 &&
861 t->discard_alignment != alignment) {
862 top = t->discard_granularity + t->discard_alignment;
863 bottom = b->discard_granularity + alignment;
865 /* Verify that top and bottom intervals line up */
866 if ((max(top, bottom) % min(top, bottom)) != 0)
867 t->discard_misaligned = 1;
870 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
871 b->max_discard_sectors);
872 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
873 b->max_hw_discard_sectors);
874 t->discard_granularity = max(t->discard_granularity,
875 b->discard_granularity);
876 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
877 t->discard_granularity;
879 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
880 b->max_secure_erase_sectors);
881 t->zone_write_granularity = max(t->zone_write_granularity,
882 b->zone_write_granularity);
883 t->zoned = max(t->zoned, b->zoned);
885 t->zone_write_granularity = 0;
886 t->max_zone_append_sectors = 0;
890 EXPORT_SYMBOL(blk_stack_limits);
893 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
894 * @t: the stacking driver limits (top device)
895 * @bdev: the underlying block device (bottom)
896 * @offset: offset to beginning of data within component device
897 * @pfx: prefix to use for warnings logged
900 * This function is used by stacking drivers like MD and DM to ensure
901 * that all component devices have compatible block sizes and
902 * alignments. The stacking driver must provide a queue_limits
903 * struct (top) and then iteratively call the stacking function for
904 * all component (bottom) devices. The stacking function will
905 * attempt to combine the values and ensure proper alignment.
907 void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
908 sector_t offset, const char *pfx)
910 if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits,
911 get_start_sect(bdev) + offset))
912 pr_notice("%s: Warning: Device %pg is misaligned\n",
915 EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
918 * blk_queue_update_dma_pad - update pad mask
919 * @q: the request queue for the device
922 * Update dma pad mask.
924 * Appending pad buffer to a request modifies the last entry of a
925 * scatter list such that it includes the pad buffer.
927 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
929 if (mask > q->dma_pad_mask)
930 q->dma_pad_mask = mask;
932 EXPORT_SYMBOL(blk_queue_update_dma_pad);
935 * blk_queue_segment_boundary - set boundary rules for segment merging
936 * @q: the request queue for the device
937 * @mask: the memory boundary mask
939 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
941 if (mask < PAGE_SIZE - 1) {
942 mask = PAGE_SIZE - 1;
943 pr_info("%s: set to minimum %lx\n", __func__, mask);
946 q->limits.seg_boundary_mask = mask;
948 EXPORT_SYMBOL(blk_queue_segment_boundary);
951 * blk_queue_virt_boundary - set boundary rules for bio merging
952 * @q: the request queue for the device
953 * @mask: the memory boundary mask
955 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
957 q->limits.virt_boundary_mask = mask;
960 * Devices that require a virtual boundary do not support scatter/gather
961 * I/O natively, but instead require a descriptor list entry for each
962 * page (which might not be idential to the Linux PAGE_SIZE). Because
963 * of that they are not limited by our notion of "segment size".
966 q->limits.max_segment_size = UINT_MAX;
968 EXPORT_SYMBOL(blk_queue_virt_boundary);
971 * blk_queue_dma_alignment - set dma length and memory alignment
972 * @q: the request queue for the device
973 * @mask: alignment mask
976 * set required memory and length alignment for direct dma transactions.
977 * this is used when building direct io requests for the queue.
980 void blk_queue_dma_alignment(struct request_queue *q, int mask)
982 q->limits.dma_alignment = mask;
984 EXPORT_SYMBOL(blk_queue_dma_alignment);
987 * blk_queue_update_dma_alignment - update dma length and memory alignment
988 * @q: the request queue for the device
989 * @mask: alignment mask
992 * update required memory and length alignment for direct dma transactions.
993 * If the requested alignment is larger than the current alignment, then
994 * the current queue alignment is updated to the new value, otherwise it
995 * is left alone. The design of this is to allow multiple objects
996 * (driver, device, transport etc) to set their respective
997 * alignments without having them interfere.
1000 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
1002 BUG_ON(mask > PAGE_SIZE);
1004 if (mask > q->limits.dma_alignment)
1005 q->limits.dma_alignment = mask;
1007 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
1010 * blk_set_queue_depth - tell the block layer about the device queue depth
1011 * @q: the request queue for the device
1012 * @depth: queue depth
1015 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
1017 q->queue_depth = depth;
1018 rq_qos_queue_depth_changed(q);
1020 EXPORT_SYMBOL(blk_set_queue_depth);
1023 * blk_queue_write_cache - configure queue's write cache
1024 * @q: the request queue for the device
1025 * @wc: write back cache on or off
1026 * @fua: device supports FUA writes, if true
1028 * Tell the block layer about the write cache of @q.
1030 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
1033 blk_queue_flag_set(QUEUE_FLAG_HW_WC, q);
1034 blk_queue_flag_set(QUEUE_FLAG_WC, q);
1036 blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q);
1037 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
1040 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
1042 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
1044 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
1047 * blk_queue_required_elevator_features - Set a queue required elevator features
1048 * @q: the request queue for the target device
1049 * @features: Required elevator features OR'ed together
1051 * Tell the block layer that for the device controlled through @q, only the
1052 * only elevators that can be used are those that implement at least the set of
1053 * features specified by @features.
1055 void blk_queue_required_elevator_features(struct request_queue *q,
1056 unsigned int features)
1058 q->required_elevator_features = features;
1060 EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
1063 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
1064 * @q: the request queue for the device
1065 * @dev: the device pointer for dma
1067 * Tell the block layer about merging the segments by dma map of @q.
1069 bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
1072 unsigned long boundary = dma_get_merge_boundary(dev);
1077 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
1078 blk_queue_virt_boundary(q, boundary);
1082 EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
1085 * disk_set_zoned - inidicate a zoned device
1086 * @disk: gendisk to configure
1088 void disk_set_zoned(struct gendisk *disk)
1090 struct request_queue *q = disk->queue;
1092 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
1095 * Set the zone write granularity to the device logical block
1096 * size by default. The driver can change this value if needed.
1098 q->limits.zoned = true;
1099 blk_queue_zone_write_granularity(q, queue_logical_block_size(q));
1101 EXPORT_SYMBOL_GPL(disk_set_zoned);
1103 int bdev_alignment_offset(struct block_device *bdev)
1105 struct request_queue *q = bdev_get_queue(bdev);
1107 if (q->limits.misaligned)
1109 if (bdev_is_partition(bdev))
1110 return queue_limit_alignment_offset(&q->limits,
1111 bdev->bd_start_sect);
1112 return q->limits.alignment_offset;
1114 EXPORT_SYMBOL_GPL(bdev_alignment_offset);
1116 unsigned int bdev_discard_alignment(struct block_device *bdev)
1118 struct request_queue *q = bdev_get_queue(bdev);
1120 if (bdev_is_partition(bdev))
1121 return queue_limit_discard_alignment(&q->limits,
1122 bdev->bd_start_sect);
1123 return q->limits.discard_alignment;
1125 EXPORT_SYMBOL_GPL(bdev_discard_alignment);