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 > max_hw_sectors ||
150 lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
152 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
154 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
156 lim->max_sectors = round_down(lim->max_sectors,
157 lim->logical_block_size >> SECTOR_SHIFT);
160 * Random default for the maximum number of segments. Driver should not
161 * rely on this and set their own.
163 if (!lim->max_segments)
164 lim->max_segments = BLK_MAX_SEGMENTS;
166 lim->max_discard_sectors =
167 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
169 if (!lim->max_discard_segments)
170 lim->max_discard_segments = 1;
172 if (lim->discard_granularity < lim->physical_block_size)
173 lim->discard_granularity = lim->physical_block_size;
176 * By default there is no limit on the segment boundary alignment,
177 * but if there is one it can't be smaller than the page size as
178 * that would break all the normal I/O patterns.
180 if (!lim->seg_boundary_mask)
181 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
182 if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
186 * Devices that require a virtual boundary do not support scatter/gather
187 * I/O natively, but instead require a descriptor list entry for each
188 * page (which might not be identical to the Linux PAGE_SIZE). Because
189 * of that they are not limited by our notion of "segment size".
191 if (lim->virt_boundary_mask) {
192 if (WARN_ON_ONCE(lim->max_segment_size &&
193 lim->max_segment_size != UINT_MAX))
195 lim->max_segment_size = UINT_MAX;
198 * The maximum segment size has an odd historic 64k default that
199 * drivers probably should override. Just like the I/O size we
200 * require drivers to at least handle a full page per segment.
202 if (!lim->max_segment_size)
203 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
204 if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
209 * We require drivers to at least do logical block aligned I/O, but
210 * historically could not check for that due to the separate calls
211 * to set the limits. Once the transition is finished the check
212 * below should be narrowed down to check the logical block size.
214 if (!lim->dma_alignment)
215 lim->dma_alignment = SECTOR_SIZE - 1;
216 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
219 if (lim->alignment_offset) {
220 lim->alignment_offset &= (lim->physical_block_size - 1);
224 return blk_validate_zoned_limits(lim);
228 * Set the default limits for a newly allocated queue. @lim contains the
229 * initial limits set by the driver, which could be no limit in which case
230 * all fields are cleared to zero.
232 int blk_set_default_limits(struct queue_limits *lim)
235 * Most defaults are set by capping the bounds in blk_validate_limits,
236 * but max_user_discard_sectors is special and needs an explicit
237 * initialization to the max value here.
239 lim->max_user_discard_sectors = UINT_MAX;
240 return blk_validate_limits(lim);
244 * queue_limits_commit_update - commit an atomic update of queue limits
245 * @q: queue to update
246 * @lim: limits to apply
248 * Apply the limits in @lim that were obtained from queue_limits_start_update()
249 * and updated by the caller to @q.
251 * Returns 0 if successful, else a negative error code.
253 int queue_limits_commit_update(struct request_queue *q,
254 struct queue_limits *lim)
255 __releases(q->limits_lock)
257 int error = blk_validate_limits(lim);
262 blk_apply_bdi_limits(q->disk->bdi, lim);
264 mutex_unlock(&q->limits_lock);
267 EXPORT_SYMBOL_GPL(queue_limits_commit_update);
270 * queue_limits_set - apply queue limits to queue
271 * @q: queue to update
272 * @lim: limits to apply
274 * Apply the limits in @lim that were freshly initialized to @q.
275 * To update existing limits use queue_limits_start_update() and
276 * queue_limits_commit_update() instead.
278 * Returns 0 if successful, else a negative error code.
280 int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
282 mutex_lock(&q->limits_lock);
283 return queue_limits_commit_update(q, lim);
285 EXPORT_SYMBOL_GPL(queue_limits_set);
288 * blk_queue_bounce_limit - set bounce buffer limit for queue
289 * @q: the request queue for the device
290 * @bounce: bounce limit to enforce
293 * Force bouncing for ISA DMA ranges or highmem.
295 * DEPRECATED, don't use in new code.
297 void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
299 q->limits.bounce = bounce;
301 EXPORT_SYMBOL(blk_queue_bounce_limit);
304 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
305 * @q: the request queue for the device
306 * @max_hw_sectors: max hardware sectors in the usual 512b unit
309 * Enables a low level driver to set a hard upper limit,
310 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
311 * the device driver based upon the capabilities of the I/O
314 * max_dev_sectors is a hard limit imposed by the storage device for
315 * READ/WRITE requests. It is set by the disk driver.
317 * max_sectors is a soft limit imposed by the block layer for
318 * filesystem type requests. This value can be overridden on a
319 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
320 * The soft limit can not exceed max_hw_sectors.
322 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
324 struct queue_limits *limits = &q->limits;
325 unsigned int max_sectors;
327 if ((max_hw_sectors << 9) < PAGE_SIZE) {
328 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
329 pr_info("%s: set to minimum %u\n", __func__, max_hw_sectors);
332 max_hw_sectors = round_down(max_hw_sectors,
333 limits->logical_block_size >> SECTOR_SHIFT);
334 limits->max_hw_sectors = max_hw_sectors;
336 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
338 if (limits->max_user_sectors)
339 max_sectors = min(max_sectors, limits->max_user_sectors);
341 max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS_CAP);
343 max_sectors = round_down(max_sectors,
344 limits->logical_block_size >> SECTOR_SHIFT);
345 limits->max_sectors = max_sectors;
349 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
351 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
354 * blk_queue_chunk_sectors - set size of the chunk for this queue
355 * @q: the request queue for the device
356 * @chunk_sectors: chunk sectors in the usual 512b unit
359 * If a driver doesn't want IOs to cross a given chunk size, it can set
360 * this limit and prevent merging across chunks. Note that the block layer
361 * must accept a page worth of data at any offset. So if the crossing of
362 * chunks is a hard limitation in the driver, it must still be prepared
363 * to split single page bios.
365 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
367 q->limits.chunk_sectors = chunk_sectors;
369 EXPORT_SYMBOL(blk_queue_chunk_sectors);
372 * blk_queue_max_discard_sectors - set max sectors for a single discard
373 * @q: the request queue for the device
374 * @max_discard_sectors: maximum number of sectors to discard
376 void blk_queue_max_discard_sectors(struct request_queue *q,
377 unsigned int max_discard_sectors)
379 struct queue_limits *lim = &q->limits;
381 lim->max_hw_discard_sectors = max_discard_sectors;
382 lim->max_discard_sectors =
383 min(max_discard_sectors, lim->max_user_discard_sectors);
385 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
388 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
389 * @q: the request queue for the device
390 * @max_sectors: maximum number of sectors to secure_erase
392 void blk_queue_max_secure_erase_sectors(struct request_queue *q,
393 unsigned int max_sectors)
395 q->limits.max_secure_erase_sectors = max_sectors;
397 EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
400 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
402 * @q: the request queue for the device
403 * @max_write_zeroes_sectors: maximum number of sectors to write per command
405 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
406 unsigned int max_write_zeroes_sectors)
408 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
410 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
413 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
414 * @q: the request queue for the device
415 * @max_zone_append_sectors: maximum number of sectors to write per command
417 void blk_queue_max_zone_append_sectors(struct request_queue *q,
418 unsigned int max_zone_append_sectors)
420 unsigned int max_sectors;
422 if (WARN_ON(!blk_queue_is_zoned(q)))
425 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
426 max_sectors = min(q->limits.chunk_sectors, max_sectors);
429 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
430 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
431 * or the max_hw_sectors limit not set.
433 WARN_ON(!max_sectors);
435 q->limits.max_zone_append_sectors = max_sectors;
437 EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
440 * blk_queue_max_segments - set max hw segments for a request for this queue
441 * @q: the request queue for the device
442 * @max_segments: max number of segments
445 * Enables a low level driver to set an upper limit on the number of
446 * hw data segments in a request.
448 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
452 pr_info("%s: set to minimum %u\n", __func__, max_segments);
455 q->limits.max_segments = max_segments;
457 EXPORT_SYMBOL(blk_queue_max_segments);
460 * blk_queue_max_discard_segments - set max segments for discard requests
461 * @q: the request queue for the device
462 * @max_segments: max number of segments
465 * Enables a low level driver to set an upper limit on the number of
466 * segments in a discard request.
468 void blk_queue_max_discard_segments(struct request_queue *q,
469 unsigned short max_segments)
471 q->limits.max_discard_segments = max_segments;
473 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
476 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
477 * @q: the request queue for the device
478 * @max_size: max size of segment in bytes
481 * Enables a low level driver to set an upper limit on the size of a
484 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
486 if (max_size < PAGE_SIZE) {
487 max_size = PAGE_SIZE;
488 pr_info("%s: set to minimum %u\n", __func__, max_size);
491 /* see blk_queue_virt_boundary() for the explanation */
492 WARN_ON_ONCE(q->limits.virt_boundary_mask);
494 q->limits.max_segment_size = max_size;
496 EXPORT_SYMBOL(blk_queue_max_segment_size);
499 * blk_queue_logical_block_size - set logical block size for the queue
500 * @q: the request queue for the device
501 * @size: the logical block size, in bytes
504 * This should be set to the lowest possible block size that the
505 * storage device can address. The default of 512 covers most
508 void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
510 struct queue_limits *limits = &q->limits;
512 limits->logical_block_size = size;
514 if (limits->discard_granularity < limits->logical_block_size)
515 limits->discard_granularity = limits->logical_block_size;
517 if (limits->physical_block_size < size)
518 limits->physical_block_size = size;
520 if (limits->io_min < limits->physical_block_size)
521 limits->io_min = limits->physical_block_size;
523 limits->max_hw_sectors =
524 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
525 limits->max_sectors =
526 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
528 EXPORT_SYMBOL(blk_queue_logical_block_size);
531 * blk_queue_physical_block_size - set physical block size for the queue
532 * @q: the request queue for the device
533 * @size: the physical block size, in bytes
536 * This should be set to the lowest possible sector size that the
537 * hardware can operate on without reverting to read-modify-write
540 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
542 q->limits.physical_block_size = size;
544 if (q->limits.physical_block_size < q->limits.logical_block_size)
545 q->limits.physical_block_size = q->limits.logical_block_size;
547 if (q->limits.discard_granularity < q->limits.physical_block_size)
548 q->limits.discard_granularity = q->limits.physical_block_size;
550 if (q->limits.io_min < q->limits.physical_block_size)
551 q->limits.io_min = q->limits.physical_block_size;
553 EXPORT_SYMBOL(blk_queue_physical_block_size);
556 * blk_queue_zone_write_granularity - set zone write granularity for the queue
557 * @q: the request queue for the zoned device
558 * @size: the zone write granularity size, in bytes
561 * This should be set to the lowest possible size allowing to write in
562 * sequential zones of a zoned block device.
564 void blk_queue_zone_write_granularity(struct request_queue *q,
567 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
570 q->limits.zone_write_granularity = size;
572 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
573 q->limits.zone_write_granularity = q->limits.logical_block_size;
575 EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
578 * blk_queue_alignment_offset - set physical block alignment offset
579 * @q: the request queue for the device
580 * @offset: alignment offset in bytes
583 * Some devices are naturally misaligned to compensate for things like
584 * the legacy DOS partition table 63-sector offset. Low-level drivers
585 * should call this function for devices whose first sector is not
588 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
590 q->limits.alignment_offset =
591 offset & (q->limits.physical_block_size - 1);
592 q->limits.misaligned = 0;
594 EXPORT_SYMBOL(blk_queue_alignment_offset);
596 void disk_update_readahead(struct gendisk *disk)
598 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
600 EXPORT_SYMBOL_GPL(disk_update_readahead);
603 * blk_limits_io_min - set minimum request size for a device
604 * @limits: the queue limits
605 * @min: smallest I/O size in bytes
608 * Some devices have an internal block size bigger than the reported
609 * hardware sector size. This function can be used to signal the
610 * smallest I/O the device can perform without incurring a performance
613 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
615 limits->io_min = min;
617 if (limits->io_min < limits->logical_block_size)
618 limits->io_min = limits->logical_block_size;
620 if (limits->io_min < limits->physical_block_size)
621 limits->io_min = limits->physical_block_size;
623 EXPORT_SYMBOL(blk_limits_io_min);
626 * blk_queue_io_min - set minimum request size for the queue
627 * @q: the request queue for the device
628 * @min: smallest I/O size in bytes
631 * Storage devices may report a granularity or preferred minimum I/O
632 * size which is the smallest request the device can perform without
633 * incurring a performance penalty. For disk drives this is often the
634 * physical block size. For RAID arrays it is often the stripe chunk
635 * size. A properly aligned multiple of minimum_io_size is the
636 * preferred request size for workloads where a high number of I/O
637 * operations is desired.
639 void blk_queue_io_min(struct request_queue *q, unsigned int min)
641 blk_limits_io_min(&q->limits, min);
643 EXPORT_SYMBOL(blk_queue_io_min);
646 * blk_limits_io_opt - set optimal request size for a device
647 * @limits: the queue limits
648 * @opt: smallest I/O size in bytes
651 * Storage devices may report an optimal I/O size, which is the
652 * device's preferred unit for sustained I/O. This is rarely reported
653 * for disk drives. For RAID arrays it is usually the stripe width or
654 * the internal track size. A properly aligned multiple of
655 * optimal_io_size is the preferred request size for workloads where
656 * sustained throughput is desired.
658 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
660 limits->io_opt = opt;
662 EXPORT_SYMBOL(blk_limits_io_opt);
665 * blk_queue_io_opt - set optimal request size for the queue
666 * @q: the request queue for the device
667 * @opt: optimal request size in bytes
670 * Storage devices may report an optimal I/O size, which is the
671 * device's preferred unit for sustained I/O. This is rarely reported
672 * for disk drives. For RAID arrays it is usually the stripe width or
673 * the internal track size. A properly aligned multiple of
674 * optimal_io_size is the preferred request size for workloads where
675 * sustained throughput is desired.
677 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
679 blk_limits_io_opt(&q->limits, opt);
682 q->disk->bdi->ra_pages =
683 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
685 EXPORT_SYMBOL(blk_queue_io_opt);
687 static int queue_limit_alignment_offset(const struct queue_limits *lim,
690 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
691 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
694 return (granularity + lim->alignment_offset - alignment) % granularity;
697 static unsigned int queue_limit_discard_alignment(
698 const struct queue_limits *lim, sector_t sector)
700 unsigned int alignment, granularity, offset;
702 if (!lim->max_discard_sectors)
705 /* Why are these in bytes, not sectors? */
706 alignment = lim->discard_alignment >> SECTOR_SHIFT;
707 granularity = lim->discard_granularity >> SECTOR_SHIFT;
711 /* Offset of the partition start in 'granularity' sectors */
712 offset = sector_div(sector, granularity);
714 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
715 offset = (granularity + alignment - offset) % granularity;
717 /* Turn it back into bytes, gaah */
718 return offset << SECTOR_SHIFT;
721 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
723 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
724 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
725 sectors = PAGE_SIZE >> SECTOR_SHIFT;
730 * blk_stack_limits - adjust queue_limits for stacked devices
731 * @t: the stacking driver limits (top device)
732 * @b: the underlying queue limits (bottom, component device)
733 * @start: first data sector within component device
736 * This function is used by stacking drivers like MD and DM to ensure
737 * that all component devices have compatible block sizes and
738 * alignments. The stacking driver must provide a queue_limits
739 * struct (top) and then iteratively call the stacking function for
740 * all component (bottom) devices. The stacking function will
741 * attempt to combine the values and ensure proper alignment.
743 * Returns 0 if the top and bottom queue_limits are compatible. The
744 * top device's block sizes and alignment offsets may be adjusted to
745 * ensure alignment with the bottom device. If no compatible sizes
746 * and alignments exist, -1 is returned and the resulting top
747 * queue_limits will have the misaligned flag set to indicate that
748 * the alignment_offset is undefined.
750 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
753 unsigned int top, bottom, alignment, ret = 0;
755 t->max_sectors = min_not_zero(t->max_sectors, b->max_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);
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);
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);
775 t->max_segment_size = min_not_zero(t->max_segment_size,
776 b->max_segment_size);
778 t->misaligned |= b->misaligned;
780 alignment = queue_limit_alignment_offset(b, start);
782 /* Bottom device has different alignment. Check that it is
783 * compatible with the current top alignment.
785 if (t->alignment_offset != alignment) {
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;
791 /* Verify that top and bottom intervals line up */
792 if (max(top, bottom) % min(top, bottom)) {
798 t->logical_block_size = max(t->logical_block_size,
799 b->logical_block_size);
801 t->physical_block_size = max(t->physical_block_size,
802 b->physical_block_size);
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);
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);
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;
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;
826 /* Optimal I/O a multiple of the physical block size? */
827 if (t->io_opt & (t->physical_block_size - 1)) {
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;
840 t->raid_partial_stripes_expensive =
841 max(t->raid_partial_stripes_expensive,
842 b->raid_partial_stripes_expensive);
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);
848 /* Verify that new alignment_offset is on a logical block boundary */
849 if (t->alignment_offset & (t->logical_block_size - 1)) {
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);
858 /* Discard alignment and granularity */
859 if (b->discard_granularity) {
860 alignment = queue_limit_discard_alignment(b, start);
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;
867 /* Verify that top and bottom intervals line up */
868 if ((max(top, bottom) % min(top, bottom)) != 0)
869 t->discard_misaligned = 1;
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;
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);
887 t->zone_write_granularity = 0;
888 t->max_zone_append_sectors = 0;
892 EXPORT_SYMBOL(blk_stack_limits);
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
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.
909 void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
910 sector_t offset, const char *pfx)
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",
917 EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
920 * blk_queue_update_dma_pad - update pad mask
921 * @q: the request queue for the device
924 * Update dma pad mask.
926 * Appending pad buffer to a request modifies the last entry of a
927 * scatter list such that it includes the pad buffer.
929 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
931 if (mask > q->dma_pad_mask)
932 q->dma_pad_mask = mask;
934 EXPORT_SYMBOL(blk_queue_update_dma_pad);
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
941 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
943 if (mask < PAGE_SIZE - 1) {
944 mask = PAGE_SIZE - 1;
945 pr_info("%s: set to minimum %lx\n", __func__, mask);
948 q->limits.seg_boundary_mask = mask;
950 EXPORT_SYMBOL(blk_queue_segment_boundary);
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
957 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
959 q->limits.virt_boundary_mask = mask;
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".
968 q->limits.max_segment_size = UINT_MAX;
970 EXPORT_SYMBOL(blk_queue_virt_boundary);
973 * blk_queue_dma_alignment - set dma length and memory alignment
974 * @q: the request queue for the device
975 * @mask: alignment mask
978 * set required memory and length alignment for direct dma transactions.
979 * this is used when building direct io requests for the queue.
982 void blk_queue_dma_alignment(struct request_queue *q, int mask)
984 q->limits.dma_alignment = mask;
986 EXPORT_SYMBOL(blk_queue_dma_alignment);
989 * blk_queue_update_dma_alignment - update dma length and memory alignment
990 * @q: the request queue for the device
991 * @mask: alignment mask
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.
1002 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
1004 BUG_ON(mask > PAGE_SIZE);
1006 if (mask > q->limits.dma_alignment)
1007 q->limits.dma_alignment = mask;
1009 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
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
1017 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
1019 q->queue_depth = depth;
1020 rq_qos_queue_depth_changed(q);
1022 EXPORT_SYMBOL(blk_set_queue_depth);
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
1030 * Tell the block layer about the write cache of @q.
1032 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
1035 blk_queue_flag_set(QUEUE_FLAG_HW_WC, q);
1036 blk_queue_flag_set(QUEUE_FLAG_WC, q);
1038 blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q);
1039 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
1042 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
1044 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
1046 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
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
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.
1057 void blk_queue_required_elevator_features(struct request_queue *q,
1058 unsigned int features)
1060 q->required_elevator_features = features;
1062 EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
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
1069 * Tell the block layer about merging the segments by dma map of @q.
1071 bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
1074 unsigned long boundary = dma_get_merge_boundary(dev);
1079 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
1080 blk_queue_virt_boundary(q, boundary);
1084 EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
1087 * disk_set_zoned - inidicate a zoned device
1088 * @disk: gendisk to configure
1090 void disk_set_zoned(struct gendisk *disk)
1092 struct request_queue *q = disk->queue;
1094 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
1097 * Set the zone write granularity to the device logical block
1098 * size by default. The driver can change this value if needed.
1100 q->limits.zoned = true;
1101 blk_queue_zone_write_granularity(q, queue_logical_block_size(q));
1103 EXPORT_SYMBOL_GPL(disk_set_zoned);
1105 int bdev_alignment_offset(struct block_device *bdev)
1107 struct request_queue *q = bdev_get_queue(bdev);
1109 if (q->limits.misaligned)
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;
1116 EXPORT_SYMBOL_GPL(bdev_alignment_offset);
1118 unsigned int bdev_discard_alignment(struct block_device *bdev)
1120 struct request_queue *q = bdev_get_queue(bdev);
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;
1127 EXPORT_SYMBOL_GPL(bdev_discard_alignment);