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 * Devices that require a virtual boundary do not support scatter/gather
186 * I/O natively, but instead require a descriptor list entry for each
187 * page (which might not be identical to the Linux PAGE_SIZE). Because
188 * of that they are not limited by our notion of "segment size".
190 if (lim->virt_boundary_mask) {
191 if (WARN_ON_ONCE(lim->max_segment_size &&
192 lim->max_segment_size != UINT_MAX))
194 lim->max_segment_size = UINT_MAX;
197 * The maximum segment size has an odd historic 64k default that
198 * drivers probably should override. Just like the I/O size we
199 * require drivers to at least handle a full page per segment.
201 if (!lim->max_segment_size)
202 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
203 if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
208 * We require drivers to at least do logical block aligned I/O, but
209 * historically could not check for that due to the separate calls
210 * to set the limits. Once the transition is finished the check
211 * below should be narrowed down to check the logical block size.
213 if (!lim->dma_alignment)
214 lim->dma_alignment = SECTOR_SIZE - 1;
215 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
218 if (lim->alignment_offset) {
219 lim->alignment_offset &= (lim->physical_block_size - 1);
223 return blk_validate_zoned_limits(lim);
227 * Set the default limits for a newly allocated queue. @lim contains the
228 * initial limits set by the driver, which could be no limit in which case
229 * all fields are cleared to zero.
231 int blk_set_default_limits(struct queue_limits *lim)
234 * Most defaults are set by capping the bounds in blk_validate_limits,
235 * but max_user_discard_sectors is special and needs an explicit
236 * initialization to the max value here.
238 lim->max_user_discard_sectors = UINT_MAX;
239 return blk_validate_limits(lim);
243 * queue_limits_commit_update - commit an atomic update of queue limits
244 * @q: queue to update
245 * @lim: limits to apply
247 * Apply the limits in @lim that were obtained from queue_limits_start_update()
248 * and updated by the caller to @q.
250 * Returns 0 if successful, else a negative error code.
252 int queue_limits_commit_update(struct request_queue *q,
253 struct queue_limits *lim)
254 __releases(q->limits_lock)
256 int error = blk_validate_limits(lim);
261 blk_apply_bdi_limits(q->disk->bdi, lim);
263 mutex_unlock(&q->limits_lock);
266 EXPORT_SYMBOL_GPL(queue_limits_commit_update);
269 * queue_limits_set - apply queue limits to queue
270 * @q: queue to update
271 * @lim: limits to apply
273 * Apply the limits in @lim that were freshly initialized to @q.
274 * To update existing limits use queue_limits_start_update() and
275 * queue_limits_commit_update() instead.
277 * Returns 0 if successful, else a negative error code.
279 int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
281 mutex_lock(&q->limits_lock);
282 return queue_limits_commit_update(q, lim);
284 EXPORT_SYMBOL_GPL(queue_limits_set);
287 * blk_queue_bounce_limit - set bounce buffer limit for queue
288 * @q: the request queue for the device
289 * @bounce: bounce limit to enforce
292 * Force bouncing for ISA DMA ranges or highmem.
294 * DEPRECATED, don't use in new code.
296 void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
298 q->limits.bounce = bounce;
300 EXPORT_SYMBOL(blk_queue_bounce_limit);
303 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
304 * @q: the request queue for the device
305 * @max_hw_sectors: max hardware sectors in the usual 512b unit
308 * Enables a low level driver to set a hard upper limit,
309 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
310 * the device driver based upon the capabilities of the I/O
313 * max_dev_sectors is a hard limit imposed by the storage device for
314 * READ/WRITE requests. It is set by the disk driver.
316 * max_sectors is a soft limit imposed by the block layer for
317 * filesystem type requests. This value can be overridden on a
318 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
319 * The soft limit can not exceed max_hw_sectors.
321 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
323 struct queue_limits *limits = &q->limits;
324 unsigned int max_sectors;
326 if ((max_hw_sectors << 9) < PAGE_SIZE) {
327 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
328 pr_info("%s: set to minimum %u\n", __func__, max_hw_sectors);
331 max_hw_sectors = round_down(max_hw_sectors,
332 limits->logical_block_size >> SECTOR_SHIFT);
333 limits->max_hw_sectors = max_hw_sectors;
335 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
337 if (limits->max_user_sectors)
338 max_sectors = min(max_sectors, limits->max_user_sectors);
340 max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS_CAP);
342 max_sectors = round_down(max_sectors,
343 limits->logical_block_size >> SECTOR_SHIFT);
344 limits->max_sectors = max_sectors;
348 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
350 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
353 * blk_queue_chunk_sectors - set size of the chunk for this queue
354 * @q: the request queue for the device
355 * @chunk_sectors: chunk sectors in the usual 512b unit
358 * If a driver doesn't want IOs to cross a given chunk size, it can set
359 * this limit and prevent merging across chunks. Note that the block layer
360 * must accept a page worth of data at any offset. So if the crossing of
361 * chunks is a hard limitation in the driver, it must still be prepared
362 * to split single page bios.
364 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
366 q->limits.chunk_sectors = chunk_sectors;
368 EXPORT_SYMBOL(blk_queue_chunk_sectors);
371 * blk_queue_max_discard_sectors - set max sectors for a single discard
372 * @q: the request queue for the device
373 * @max_discard_sectors: maximum number of sectors to discard
375 void blk_queue_max_discard_sectors(struct request_queue *q,
376 unsigned int max_discard_sectors)
378 struct queue_limits *lim = &q->limits;
380 lim->max_hw_discard_sectors = max_discard_sectors;
381 lim->max_discard_sectors =
382 min(max_discard_sectors, lim->max_user_discard_sectors);
384 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
387 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
388 * @q: the request queue for the device
389 * @max_sectors: maximum number of sectors to secure_erase
391 void blk_queue_max_secure_erase_sectors(struct request_queue *q,
392 unsigned int max_sectors)
394 q->limits.max_secure_erase_sectors = max_sectors;
396 EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
399 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
401 * @q: the request queue for the device
402 * @max_write_zeroes_sectors: maximum number of sectors to write per command
404 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
405 unsigned int max_write_zeroes_sectors)
407 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
409 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
412 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
413 * @q: the request queue for the device
414 * @max_zone_append_sectors: maximum number of sectors to write per command
416 void blk_queue_max_zone_append_sectors(struct request_queue *q,
417 unsigned int max_zone_append_sectors)
419 unsigned int max_sectors;
421 if (WARN_ON(!blk_queue_is_zoned(q)))
424 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
425 max_sectors = min(q->limits.chunk_sectors, max_sectors);
428 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
429 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
430 * or the max_hw_sectors limit not set.
432 WARN_ON(!max_sectors);
434 q->limits.max_zone_append_sectors = max_sectors;
436 EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
439 * blk_queue_max_segments - set max hw segments for a request for this queue
440 * @q: the request queue for the device
441 * @max_segments: max number of segments
444 * Enables a low level driver to set an upper limit on the number of
445 * hw data segments in a request.
447 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
451 pr_info("%s: set to minimum %u\n", __func__, max_segments);
454 q->limits.max_segments = max_segments;
456 EXPORT_SYMBOL(blk_queue_max_segments);
459 * blk_queue_max_discard_segments - set max segments for discard requests
460 * @q: the request queue for the device
461 * @max_segments: max number of segments
464 * Enables a low level driver to set an upper limit on the number of
465 * segments in a discard request.
467 void blk_queue_max_discard_segments(struct request_queue *q,
468 unsigned short max_segments)
470 q->limits.max_discard_segments = max_segments;
472 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
475 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
476 * @q: the request queue for the device
477 * @max_size: max size of segment in bytes
480 * Enables a low level driver to set an upper limit on the size of a
483 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
485 if (max_size < PAGE_SIZE) {
486 max_size = PAGE_SIZE;
487 pr_info("%s: set to minimum %u\n", __func__, max_size);
490 /* see blk_queue_virt_boundary() for the explanation */
491 WARN_ON_ONCE(q->limits.virt_boundary_mask);
493 q->limits.max_segment_size = max_size;
495 EXPORT_SYMBOL(blk_queue_max_segment_size);
498 * blk_queue_logical_block_size - set logical block size for the queue
499 * @q: the request queue for the device
500 * @size: the logical block size, in bytes
503 * This should be set to the lowest possible block size that the
504 * storage device can address. The default of 512 covers most
507 void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
509 struct queue_limits *limits = &q->limits;
511 limits->logical_block_size = size;
513 if (limits->discard_granularity < limits->logical_block_size)
514 limits->discard_granularity = limits->logical_block_size;
516 if (limits->physical_block_size < size)
517 limits->physical_block_size = size;
519 if (limits->io_min < limits->physical_block_size)
520 limits->io_min = limits->physical_block_size;
522 limits->max_hw_sectors =
523 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
524 limits->max_sectors =
525 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
527 EXPORT_SYMBOL(blk_queue_logical_block_size);
530 * blk_queue_physical_block_size - set physical block size for the queue
531 * @q: the request queue for the device
532 * @size: the physical block size, in bytes
535 * This should be set to the lowest possible sector size that the
536 * hardware can operate on without reverting to read-modify-write
539 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
541 q->limits.physical_block_size = size;
543 if (q->limits.physical_block_size < q->limits.logical_block_size)
544 q->limits.physical_block_size = q->limits.logical_block_size;
546 if (q->limits.discard_granularity < q->limits.physical_block_size)
547 q->limits.discard_granularity = q->limits.physical_block_size;
549 if (q->limits.io_min < q->limits.physical_block_size)
550 q->limits.io_min = q->limits.physical_block_size;
552 EXPORT_SYMBOL(blk_queue_physical_block_size);
555 * blk_queue_zone_write_granularity - set zone write granularity for the queue
556 * @q: the request queue for the zoned device
557 * @size: the zone write granularity size, in bytes
560 * This should be set to the lowest possible size allowing to write in
561 * sequential zones of a zoned block device.
563 void blk_queue_zone_write_granularity(struct request_queue *q,
566 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
569 q->limits.zone_write_granularity = size;
571 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
572 q->limits.zone_write_granularity = q->limits.logical_block_size;
574 EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
577 * blk_queue_alignment_offset - set physical block alignment offset
578 * @q: the request queue for the device
579 * @offset: alignment offset in bytes
582 * Some devices are naturally misaligned to compensate for things like
583 * the legacy DOS partition table 63-sector offset. Low-level drivers
584 * should call this function for devices whose first sector is not
587 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
589 q->limits.alignment_offset =
590 offset & (q->limits.physical_block_size - 1);
591 q->limits.misaligned = 0;
593 EXPORT_SYMBOL(blk_queue_alignment_offset);
595 void disk_update_readahead(struct gendisk *disk)
597 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
599 EXPORT_SYMBOL_GPL(disk_update_readahead);
602 * blk_limits_io_min - set minimum request size for a device
603 * @limits: the queue limits
604 * @min: smallest I/O size in bytes
607 * Some devices have an internal block size bigger than the reported
608 * hardware sector size. This function can be used to signal the
609 * smallest I/O the device can perform without incurring a performance
612 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
614 limits->io_min = min;
616 if (limits->io_min < limits->logical_block_size)
617 limits->io_min = limits->logical_block_size;
619 if (limits->io_min < limits->physical_block_size)
620 limits->io_min = limits->physical_block_size;
622 EXPORT_SYMBOL(blk_limits_io_min);
625 * blk_queue_io_min - set minimum request size for the queue
626 * @q: the request queue for the device
627 * @min: smallest I/O size in bytes
630 * Storage devices may report a granularity or preferred minimum I/O
631 * size which is the smallest request the device can perform without
632 * incurring a performance penalty. For disk drives this is often the
633 * physical block size. For RAID arrays it is often the stripe chunk
634 * size. A properly aligned multiple of minimum_io_size is the
635 * preferred request size for workloads where a high number of I/O
636 * operations is desired.
638 void blk_queue_io_min(struct request_queue *q, unsigned int min)
640 blk_limits_io_min(&q->limits, min);
642 EXPORT_SYMBOL(blk_queue_io_min);
645 * blk_limits_io_opt - set optimal request size for a device
646 * @limits: the queue limits
647 * @opt: smallest I/O size in bytes
650 * Storage devices may report an optimal I/O size, which is the
651 * device's preferred unit for sustained I/O. This is rarely reported
652 * for disk drives. For RAID arrays it is usually the stripe width or
653 * the internal track size. A properly aligned multiple of
654 * optimal_io_size is the preferred request size for workloads where
655 * sustained throughput is desired.
657 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
659 limits->io_opt = opt;
661 EXPORT_SYMBOL(blk_limits_io_opt);
664 * blk_queue_io_opt - set optimal request size for the queue
665 * @q: the request queue for the device
666 * @opt: optimal request size in bytes
669 * Storage devices may report an optimal I/O size, which is the
670 * device's preferred unit for sustained I/O. This is rarely reported
671 * for disk drives. For RAID arrays it is usually the stripe width or
672 * the internal track size. A properly aligned multiple of
673 * optimal_io_size is the preferred request size for workloads where
674 * sustained throughput is desired.
676 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
678 blk_limits_io_opt(&q->limits, opt);
681 q->disk->bdi->ra_pages =
682 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
684 EXPORT_SYMBOL(blk_queue_io_opt);
686 static int queue_limit_alignment_offset(const struct queue_limits *lim,
689 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
690 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
693 return (granularity + lim->alignment_offset - alignment) % granularity;
696 static unsigned int queue_limit_discard_alignment(
697 const struct queue_limits *lim, sector_t sector)
699 unsigned int alignment, granularity, offset;
701 if (!lim->max_discard_sectors)
704 /* Why are these in bytes, not sectors? */
705 alignment = lim->discard_alignment >> SECTOR_SHIFT;
706 granularity = lim->discard_granularity >> SECTOR_SHIFT;
710 /* Offset of the partition start in 'granularity' sectors */
711 offset = sector_div(sector, granularity);
713 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
714 offset = (granularity + alignment - offset) % granularity;
716 /* Turn it back into bytes, gaah */
717 return offset << SECTOR_SHIFT;
720 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
722 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
723 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
724 sectors = PAGE_SIZE >> SECTOR_SHIFT;
729 * blk_stack_limits - adjust queue_limits for stacked devices
730 * @t: the stacking driver limits (top device)
731 * @b: the underlying queue limits (bottom, component device)
732 * @start: first data sector within component device
735 * This function is used by stacking drivers like MD and DM to ensure
736 * that all component devices have compatible block sizes and
737 * alignments. The stacking driver must provide a queue_limits
738 * struct (top) and then iteratively call the stacking function for
739 * all component (bottom) devices. The stacking function will
740 * attempt to combine the values and ensure proper alignment.
742 * Returns 0 if the top and bottom queue_limits are compatible. The
743 * top device's block sizes and alignment offsets may be adjusted to
744 * ensure alignment with the bottom device. If no compatible sizes
745 * and alignments exist, -1 is returned and the resulting top
746 * queue_limits will have the misaligned flag set to indicate that
747 * the alignment_offset is undefined.
749 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
752 unsigned int top, bottom, alignment, ret = 0;
754 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
755 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
756 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
757 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
758 b->max_write_zeroes_sectors);
759 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
760 b->max_zone_append_sectors);
761 t->bounce = max(t->bounce, b->bounce);
763 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
764 b->seg_boundary_mask);
765 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
766 b->virt_boundary_mask);
768 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
769 t->max_discard_segments = min_not_zero(t->max_discard_segments,
770 b->max_discard_segments);
771 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
772 b->max_integrity_segments);
774 t->max_segment_size = min_not_zero(t->max_segment_size,
775 b->max_segment_size);
777 t->misaligned |= b->misaligned;
779 alignment = queue_limit_alignment_offset(b, start);
781 /* Bottom device has different alignment. Check that it is
782 * compatible with the current top alignment.
784 if (t->alignment_offset != alignment) {
786 top = max(t->physical_block_size, t->io_min)
787 + t->alignment_offset;
788 bottom = max(b->physical_block_size, b->io_min) + alignment;
790 /* Verify that top and bottom intervals line up */
791 if (max(top, bottom) % min(top, bottom)) {
797 t->logical_block_size = max(t->logical_block_size,
798 b->logical_block_size);
800 t->physical_block_size = max(t->physical_block_size,
801 b->physical_block_size);
803 t->io_min = max(t->io_min, b->io_min);
804 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
805 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
807 /* Set non-power-of-2 compatible chunk_sectors boundary */
808 if (b->chunk_sectors)
809 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
811 /* Physical block size a multiple of the logical block size? */
812 if (t->physical_block_size & (t->logical_block_size - 1)) {
813 t->physical_block_size = t->logical_block_size;
818 /* Minimum I/O a multiple of the physical block size? */
819 if (t->io_min & (t->physical_block_size - 1)) {
820 t->io_min = t->physical_block_size;
825 /* Optimal I/O a multiple of the physical block size? */
826 if (t->io_opt & (t->physical_block_size - 1)) {
832 /* chunk_sectors a multiple of the physical block size? */
833 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
834 t->chunk_sectors = 0;
839 t->raid_partial_stripes_expensive =
840 max(t->raid_partial_stripes_expensive,
841 b->raid_partial_stripes_expensive);
843 /* Find lowest common alignment_offset */
844 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
845 % max(t->physical_block_size, t->io_min);
847 /* Verify that new alignment_offset is on a logical block boundary */
848 if (t->alignment_offset & (t->logical_block_size - 1)) {
853 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
854 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
855 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
857 /* Discard alignment and granularity */
858 if (b->discard_granularity) {
859 alignment = queue_limit_discard_alignment(b, start);
861 if (t->discard_granularity != 0 &&
862 t->discard_alignment != alignment) {
863 top = t->discard_granularity + t->discard_alignment;
864 bottom = b->discard_granularity + alignment;
866 /* Verify that top and bottom intervals line up */
867 if ((max(top, bottom) % min(top, bottom)) != 0)
868 t->discard_misaligned = 1;
871 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
872 b->max_discard_sectors);
873 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
874 b->max_hw_discard_sectors);
875 t->discard_granularity = max(t->discard_granularity,
876 b->discard_granularity);
877 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
878 t->discard_granularity;
880 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
881 b->max_secure_erase_sectors);
882 t->zone_write_granularity = max(t->zone_write_granularity,
883 b->zone_write_granularity);
884 t->zoned = max(t->zoned, b->zoned);
886 t->zone_write_granularity = 0;
887 t->max_zone_append_sectors = 0;
891 EXPORT_SYMBOL(blk_stack_limits);
894 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
895 * @t: the stacking driver limits (top device)
896 * @bdev: the underlying block device (bottom)
897 * @offset: offset to beginning of data within component device
898 * @pfx: prefix to use for warnings logged
901 * This function is used by stacking drivers like MD and DM to ensure
902 * that all component devices have compatible block sizes and
903 * alignments. The stacking driver must provide a queue_limits
904 * struct (top) and then iteratively call the stacking function for
905 * all component (bottom) devices. The stacking function will
906 * attempt to combine the values and ensure proper alignment.
908 void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
909 sector_t offset, const char *pfx)
911 if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits,
912 get_start_sect(bdev) + offset))
913 pr_notice("%s: Warning: Device %pg is misaligned\n",
916 EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
919 * blk_queue_update_dma_pad - update pad mask
920 * @q: the request queue for the device
923 * Update dma pad mask.
925 * Appending pad buffer to a request modifies the last entry of a
926 * scatter list such that it includes the pad buffer.
928 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
930 if (mask > q->dma_pad_mask)
931 q->dma_pad_mask = mask;
933 EXPORT_SYMBOL(blk_queue_update_dma_pad);
936 * blk_queue_segment_boundary - set boundary rules for segment merging
937 * @q: the request queue for the device
938 * @mask: the memory boundary mask
940 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
942 if (mask < PAGE_SIZE - 1) {
943 mask = PAGE_SIZE - 1;
944 pr_info("%s: set to minimum %lx\n", __func__, mask);
947 q->limits.seg_boundary_mask = mask;
949 EXPORT_SYMBOL(blk_queue_segment_boundary);
952 * blk_queue_virt_boundary - set boundary rules for bio merging
953 * @q: the request queue for the device
954 * @mask: the memory boundary mask
956 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
958 q->limits.virt_boundary_mask = mask;
961 * Devices that require a virtual boundary do not support scatter/gather
962 * I/O natively, but instead require a descriptor list entry for each
963 * page (which might not be idential to the Linux PAGE_SIZE). Because
964 * of that they are not limited by our notion of "segment size".
967 q->limits.max_segment_size = UINT_MAX;
969 EXPORT_SYMBOL(blk_queue_virt_boundary);
972 * blk_queue_dma_alignment - set dma length and memory alignment
973 * @q: the request queue for the device
974 * @mask: alignment mask
977 * set required memory and length alignment for direct dma transactions.
978 * this is used when building direct io requests for the queue.
981 void blk_queue_dma_alignment(struct request_queue *q, int mask)
983 q->limits.dma_alignment = mask;
985 EXPORT_SYMBOL(blk_queue_dma_alignment);
988 * blk_queue_update_dma_alignment - update dma length and memory alignment
989 * @q: the request queue for the device
990 * @mask: alignment mask
993 * update required memory and length alignment for direct dma transactions.
994 * If the requested alignment is larger than the current alignment, then
995 * the current queue alignment is updated to the new value, otherwise it
996 * is left alone. The design of this is to allow multiple objects
997 * (driver, device, transport etc) to set their respective
998 * alignments without having them interfere.
1001 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
1003 BUG_ON(mask > PAGE_SIZE);
1005 if (mask > q->limits.dma_alignment)
1006 q->limits.dma_alignment = mask;
1008 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
1011 * blk_set_queue_depth - tell the block layer about the device queue depth
1012 * @q: the request queue for the device
1013 * @depth: queue depth
1016 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
1018 q->queue_depth = depth;
1019 rq_qos_queue_depth_changed(q);
1021 EXPORT_SYMBOL(blk_set_queue_depth);
1024 * blk_queue_write_cache - configure queue's write cache
1025 * @q: the request queue for the device
1026 * @wc: write back cache on or off
1027 * @fua: device supports FUA writes, if true
1029 * Tell the block layer about the write cache of @q.
1031 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
1034 blk_queue_flag_set(QUEUE_FLAG_HW_WC, q);
1035 blk_queue_flag_set(QUEUE_FLAG_WC, q);
1037 blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q);
1038 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
1041 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
1043 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
1045 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
1048 * blk_queue_required_elevator_features - Set a queue required elevator features
1049 * @q: the request queue for the target device
1050 * @features: Required elevator features OR'ed together
1052 * Tell the block layer that for the device controlled through @q, only the
1053 * only elevators that can be used are those that implement at least the set of
1054 * features specified by @features.
1056 void blk_queue_required_elevator_features(struct request_queue *q,
1057 unsigned int features)
1059 q->required_elevator_features = features;
1061 EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
1064 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
1065 * @q: the request queue for the device
1066 * @dev: the device pointer for dma
1068 * Tell the block layer about merging the segments by dma map of @q.
1070 bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
1073 unsigned long boundary = dma_get_merge_boundary(dev);
1078 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
1079 blk_queue_virt_boundary(q, boundary);
1083 EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
1086 * disk_set_zoned - inidicate a zoned device
1087 * @disk: gendisk to configure
1089 void disk_set_zoned(struct gendisk *disk)
1091 struct request_queue *q = disk->queue;
1093 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
1096 * Set the zone write granularity to the device logical block
1097 * size by default. The driver can change this value if needed.
1099 q->limits.zoned = true;
1100 blk_queue_zone_write_granularity(q, queue_logical_block_size(q));
1102 EXPORT_SYMBOL_GPL(disk_set_zoned);
1104 int bdev_alignment_offset(struct block_device *bdev)
1106 struct request_queue *q = bdev_get_queue(bdev);
1108 if (q->limits.misaligned)
1110 if (bdev_is_partition(bdev))
1111 return queue_limit_alignment_offset(&q->limits,
1112 bdev->bd_start_sect);
1113 return q->limits.alignment_offset;
1115 EXPORT_SYMBOL_GPL(bdev_alignment_offset);
1117 unsigned int bdev_discard_alignment(struct block_device *bdev)
1119 struct request_queue *q = bdev_get_queue(bdev);
1121 if (bdev_is_partition(bdev))
1122 return queue_limit_discard_alignment(&q->limits,
1123 bdev->bd_start_sect);
1124 return q->limits.discard_alignment;
1126 EXPORT_SYMBOL_GPL(bdev_discard_alignment);