Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/pagemap.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/gcd.h>
13#include <linux/lcm.h>
14#include <linux/jiffies.h>
15#include <linux/gfp.h>
16#include <linux/dma-mapping.h>
17
18#include "blk.h"
19#include "blk-wbt.h"
20
21void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
22{
23 q->rq_timeout = timeout;
24}
25EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
26
27/**
28 * blk_set_default_limits - reset limits to default values
29 * @lim: the queue_limits structure to reset
30 *
31 * Description:
32 * Returns a queue_limit struct to its default state.
33 */
34void blk_set_default_limits(struct queue_limits *lim)
35{
36 lim->max_segments = BLK_MAX_SEGMENTS;
37 lim->max_discard_segments = 1;
38 lim->max_integrity_segments = 0;
39 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
40 lim->virt_boundary_mask = 0;
41 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
42 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
43 lim->max_dev_sectors = 0;
44 lim->chunk_sectors = 0;
45 lim->max_write_zeroes_sectors = 0;
46 lim->max_zone_append_sectors = 0;
47 lim->max_discard_sectors = 0;
48 lim->max_hw_discard_sectors = 0;
49 lim->max_secure_erase_sectors = 0;
50 lim->discard_granularity = 0;
51 lim->discard_alignment = 0;
52 lim->discard_misaligned = 0;
53 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
54 lim->bounce = BLK_BOUNCE_NONE;
55 lim->alignment_offset = 0;
56 lim->io_opt = 0;
57 lim->misaligned = 0;
58 lim->zoned = BLK_ZONED_NONE;
59 lim->zone_write_granularity = 0;
60 lim->dma_alignment = 511;
61}
62
63/**
64 * blk_set_stacking_limits - set default limits for stacking devices
65 * @lim: the queue_limits structure to reset
66 *
67 * Description:
68 * Returns a queue_limit struct to its default state. Should be used
69 * by stacking drivers like DM that have no internal limits.
70 */
71void blk_set_stacking_limits(struct queue_limits *lim)
72{
73 blk_set_default_limits(lim);
74
75 /* Inherit limits from component devices */
76 lim->max_segments = USHRT_MAX;
77 lim->max_discard_segments = USHRT_MAX;
78 lim->max_hw_sectors = UINT_MAX;
79 lim->max_segment_size = UINT_MAX;
80 lim->max_sectors = UINT_MAX;
81 lim->max_dev_sectors = UINT_MAX;
82 lim->max_write_zeroes_sectors = UINT_MAX;
83 lim->max_zone_append_sectors = UINT_MAX;
84}
85EXPORT_SYMBOL(blk_set_stacking_limits);
86
87/**
88 * blk_queue_bounce_limit - set bounce buffer limit for queue
89 * @q: the request queue for the device
90 * @bounce: bounce limit to enforce
91 *
92 * Description:
93 * Force bouncing for ISA DMA ranges or highmem.
94 *
95 * DEPRECATED, don't use in new code.
96 **/
97void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
98{
99 q->limits.bounce = bounce;
100}
101EXPORT_SYMBOL(blk_queue_bounce_limit);
102
103/**
104 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
105 * @q: the request queue for the device
106 * @max_hw_sectors: max hardware sectors in the usual 512b unit
107 *
108 * Description:
109 * Enables a low level driver to set a hard upper limit,
110 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
111 * the device driver based upon the capabilities of the I/O
112 * controller.
113 *
114 * max_dev_sectors is a hard limit imposed by the storage device for
115 * READ/WRITE requests. It is set by the disk driver.
116 *
117 * max_sectors is a soft limit imposed by the block layer for
118 * filesystem type requests. This value can be overridden on a
119 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
120 * The soft limit can not exceed max_hw_sectors.
121 **/
122void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
123{
124 struct queue_limits *limits = &q->limits;
125 unsigned int max_sectors;
126
127 if ((max_hw_sectors << 9) < PAGE_SIZE) {
128 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
129 printk(KERN_INFO "%s: set to minimum %d\n",
130 __func__, max_hw_sectors);
131 }
132
133 max_hw_sectors = round_down(max_hw_sectors,
134 limits->logical_block_size >> SECTOR_SHIFT);
135 limits->max_hw_sectors = max_hw_sectors;
136
137 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
138 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
139 max_sectors = round_down(max_sectors,
140 limits->logical_block_size >> SECTOR_SHIFT);
141 limits->max_sectors = max_sectors;
142
143 if (!q->disk)
144 return;
145 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
146}
147EXPORT_SYMBOL(blk_queue_max_hw_sectors);
148
149/**
150 * blk_queue_chunk_sectors - set size of the chunk for this queue
151 * @q: the request queue for the device
152 * @chunk_sectors: chunk sectors in the usual 512b unit
153 *
154 * Description:
155 * If a driver doesn't want IOs to cross a given chunk size, it can set
156 * this limit and prevent merging across chunks. Note that the block layer
157 * must accept a page worth of data at any offset. So if the crossing of
158 * chunks is a hard limitation in the driver, it must still be prepared
159 * to split single page bios.
160 **/
161void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
162{
163 q->limits.chunk_sectors = chunk_sectors;
164}
165EXPORT_SYMBOL(blk_queue_chunk_sectors);
166
167/**
168 * blk_queue_max_discard_sectors - set max sectors for a single discard
169 * @q: the request queue for the device
170 * @max_discard_sectors: maximum number of sectors to discard
171 **/
172void blk_queue_max_discard_sectors(struct request_queue *q,
173 unsigned int max_discard_sectors)
174{
175 q->limits.max_hw_discard_sectors = max_discard_sectors;
176 q->limits.max_discard_sectors = max_discard_sectors;
177}
178EXPORT_SYMBOL(blk_queue_max_discard_sectors);
179
180/**
181 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
182 * @q: the request queue for the device
183 * @max_sectors: maximum number of sectors to secure_erase
184 **/
185void blk_queue_max_secure_erase_sectors(struct request_queue *q,
186 unsigned int max_sectors)
187{
188 q->limits.max_secure_erase_sectors = max_sectors;
189}
190EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
191
192/**
193 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
194 * write zeroes
195 * @q: the request queue for the device
196 * @max_write_zeroes_sectors: maximum number of sectors to write per command
197 **/
198void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
199 unsigned int max_write_zeroes_sectors)
200{
201 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
202}
203EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
204
205/**
206 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
207 * @q: the request queue for the device
208 * @max_zone_append_sectors: maximum number of sectors to write per command
209 **/
210void blk_queue_max_zone_append_sectors(struct request_queue *q,
211 unsigned int max_zone_append_sectors)
212{
213 unsigned int max_sectors;
214
215 if (WARN_ON(!blk_queue_is_zoned(q)))
216 return;
217
218 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
219 max_sectors = min(q->limits.chunk_sectors, max_sectors);
220
221 /*
222 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
223 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
224 * or the max_hw_sectors limit not set.
225 */
226 WARN_ON(!max_sectors);
227
228 q->limits.max_zone_append_sectors = max_sectors;
229}
230EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
231
232/**
233 * blk_queue_max_segments - set max hw segments for a request for this queue
234 * @q: the request queue for the device
235 * @max_segments: max number of segments
236 *
237 * Description:
238 * Enables a low level driver to set an upper limit on the number of
239 * hw data segments in a request.
240 **/
241void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
242{
243 if (!max_segments) {
244 max_segments = 1;
245 printk(KERN_INFO "%s: set to minimum %d\n",
246 __func__, max_segments);
247 }
248
249 q->limits.max_segments = max_segments;
250}
251EXPORT_SYMBOL(blk_queue_max_segments);
252
253/**
254 * blk_queue_max_discard_segments - set max segments for discard requests
255 * @q: the request queue for the device
256 * @max_segments: max number of segments
257 *
258 * Description:
259 * Enables a low level driver to set an upper limit on the number of
260 * segments in a discard request.
261 **/
262void blk_queue_max_discard_segments(struct request_queue *q,
263 unsigned short max_segments)
264{
265 q->limits.max_discard_segments = max_segments;
266}
267EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
268
269/**
270 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
271 * @q: the request queue for the device
272 * @max_size: max size of segment in bytes
273 *
274 * Description:
275 * Enables a low level driver to set an upper limit on the size of a
276 * coalesced segment
277 **/
278void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
279{
280 if (max_size < PAGE_SIZE) {
281 max_size = PAGE_SIZE;
282 printk(KERN_INFO "%s: set to minimum %d\n",
283 __func__, max_size);
284 }
285
286 /* see blk_queue_virt_boundary() for the explanation */
287 WARN_ON_ONCE(q->limits.virt_boundary_mask);
288
289 q->limits.max_segment_size = max_size;
290}
291EXPORT_SYMBOL(blk_queue_max_segment_size);
292
293/**
294 * blk_queue_logical_block_size - set logical block size for the queue
295 * @q: the request queue for the device
296 * @size: the logical block size, in bytes
297 *
298 * Description:
299 * This should be set to the lowest possible block size that the
300 * storage device can address. The default of 512 covers most
301 * hardware.
302 **/
303void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
304{
305 struct queue_limits *limits = &q->limits;
306
307 limits->logical_block_size = size;
308
309 if (limits->physical_block_size < size)
310 limits->physical_block_size = size;
311
312 if (limits->io_min < limits->physical_block_size)
313 limits->io_min = limits->physical_block_size;
314
315 limits->max_hw_sectors =
316 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
317 limits->max_sectors =
318 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
319}
320EXPORT_SYMBOL(blk_queue_logical_block_size);
321
322/**
323 * blk_queue_physical_block_size - set physical block size for the queue
324 * @q: the request queue for the device
325 * @size: the physical block size, in bytes
326 *
327 * Description:
328 * This should be set to the lowest possible sector size that the
329 * hardware can operate on without reverting to read-modify-write
330 * operations.
331 */
332void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
333{
334 q->limits.physical_block_size = size;
335
336 if (q->limits.physical_block_size < q->limits.logical_block_size)
337 q->limits.physical_block_size = q->limits.logical_block_size;
338
339 if (q->limits.io_min < q->limits.physical_block_size)
340 q->limits.io_min = q->limits.physical_block_size;
341}
342EXPORT_SYMBOL(blk_queue_physical_block_size);
343
344/**
345 * blk_queue_zone_write_granularity - set zone write granularity for the queue
346 * @q: the request queue for the zoned device
347 * @size: the zone write granularity size, in bytes
348 *
349 * Description:
350 * This should be set to the lowest possible size allowing to write in
351 * sequential zones of a zoned block device.
352 */
353void blk_queue_zone_write_granularity(struct request_queue *q,
354 unsigned int size)
355{
356 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
357 return;
358
359 q->limits.zone_write_granularity = size;
360
361 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
362 q->limits.zone_write_granularity = q->limits.logical_block_size;
363}
364EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
365
366/**
367 * blk_queue_alignment_offset - set physical block alignment offset
368 * @q: the request queue for the device
369 * @offset: alignment offset in bytes
370 *
371 * Description:
372 * Some devices are naturally misaligned to compensate for things like
373 * the legacy DOS partition table 63-sector offset. Low-level drivers
374 * should call this function for devices whose first sector is not
375 * naturally aligned.
376 */
377void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
378{
379 q->limits.alignment_offset =
380 offset & (q->limits.physical_block_size - 1);
381 q->limits.misaligned = 0;
382}
383EXPORT_SYMBOL(blk_queue_alignment_offset);
384
385void disk_update_readahead(struct gendisk *disk)
386{
387 struct request_queue *q = disk->queue;
388
389 /*
390 * For read-ahead of large files to be effective, we need to read ahead
391 * at least twice the optimal I/O size.
392 */
393 disk->bdi->ra_pages =
394 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
395 disk->bdi->io_pages = queue_max_sectors(q) >> (PAGE_SHIFT - 9);
396}
397EXPORT_SYMBOL_GPL(disk_update_readahead);
398
399/**
400 * blk_limits_io_min - set minimum request size for a device
401 * @limits: the queue limits
402 * @min: smallest I/O size in bytes
403 *
404 * Description:
405 * Some devices have an internal block size bigger than the reported
406 * hardware sector size. This function can be used to signal the
407 * smallest I/O the device can perform without incurring a performance
408 * penalty.
409 */
410void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
411{
412 limits->io_min = min;
413
414 if (limits->io_min < limits->logical_block_size)
415 limits->io_min = limits->logical_block_size;
416
417 if (limits->io_min < limits->physical_block_size)
418 limits->io_min = limits->physical_block_size;
419}
420EXPORT_SYMBOL(blk_limits_io_min);
421
422/**
423 * blk_queue_io_min - set minimum request size for the queue
424 * @q: the request queue for the device
425 * @min: smallest I/O size in bytes
426 *
427 * Description:
428 * Storage devices may report a granularity or preferred minimum I/O
429 * size which is the smallest request the device can perform without
430 * incurring a performance penalty. For disk drives this is often the
431 * physical block size. For RAID arrays it is often the stripe chunk
432 * size. A properly aligned multiple of minimum_io_size is the
433 * preferred request size for workloads where a high number of I/O
434 * operations is desired.
435 */
436void blk_queue_io_min(struct request_queue *q, unsigned int min)
437{
438 blk_limits_io_min(&q->limits, min);
439}
440EXPORT_SYMBOL(blk_queue_io_min);
441
442/**
443 * blk_limits_io_opt - set optimal request size for a device
444 * @limits: the queue limits
445 * @opt: smallest I/O size in bytes
446 *
447 * Description:
448 * Storage devices may report an optimal I/O size, which is the
449 * device's preferred unit for sustained I/O. This is rarely reported
450 * for disk drives. For RAID arrays it is usually the stripe width or
451 * the internal track size. A properly aligned multiple of
452 * optimal_io_size is the preferred request size for workloads where
453 * sustained throughput is desired.
454 */
455void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
456{
457 limits->io_opt = opt;
458}
459EXPORT_SYMBOL(blk_limits_io_opt);
460
461/**
462 * blk_queue_io_opt - set optimal request size for the queue
463 * @q: the request queue for the device
464 * @opt: optimal request size in bytes
465 *
466 * Description:
467 * Storage devices may report an optimal I/O size, which is the
468 * device's preferred unit for sustained I/O. This is rarely reported
469 * for disk drives. For RAID arrays it is usually the stripe width or
470 * the internal track size. A properly aligned multiple of
471 * optimal_io_size is the preferred request size for workloads where
472 * sustained throughput is desired.
473 */
474void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
475{
476 blk_limits_io_opt(&q->limits, opt);
477 if (!q->disk)
478 return;
479 q->disk->bdi->ra_pages =
480 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
481}
482EXPORT_SYMBOL(blk_queue_io_opt);
483
484static int queue_limit_alignment_offset(const struct queue_limits *lim,
485 sector_t sector)
486{
487 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
488 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
489 << SECTOR_SHIFT;
490
491 return (granularity + lim->alignment_offset - alignment) % granularity;
492}
493
494static unsigned int queue_limit_discard_alignment(
495 const struct queue_limits *lim, sector_t sector)
496{
497 unsigned int alignment, granularity, offset;
498
499 if (!lim->max_discard_sectors)
500 return 0;
501
502 /* Why are these in bytes, not sectors? */
503 alignment = lim->discard_alignment >> SECTOR_SHIFT;
504 granularity = lim->discard_granularity >> SECTOR_SHIFT;
505 if (!granularity)
506 return 0;
507
508 /* Offset of the partition start in 'granularity' sectors */
509 offset = sector_div(sector, granularity);
510
511 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
512 offset = (granularity + alignment - offset) % granularity;
513
514 /* Turn it back into bytes, gaah */
515 return offset << SECTOR_SHIFT;
516}
517
518static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
519{
520 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
521 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
522 sectors = PAGE_SIZE >> SECTOR_SHIFT;
523 return sectors;
524}
525
526/**
527 * blk_stack_limits - adjust queue_limits for stacked devices
528 * @t: the stacking driver limits (top device)
529 * @b: the underlying queue limits (bottom, component device)
530 * @start: first data sector within component device
531 *
532 * Description:
533 * This function is used by stacking drivers like MD and DM to ensure
534 * that all component devices have compatible block sizes and
535 * alignments. The stacking driver must provide a queue_limits
536 * struct (top) and then iteratively call the stacking function for
537 * all component (bottom) devices. The stacking function will
538 * attempt to combine the values and ensure proper alignment.
539 *
540 * Returns 0 if the top and bottom queue_limits are compatible. The
541 * top device's block sizes and alignment offsets may be adjusted to
542 * ensure alignment with the bottom device. If no compatible sizes
543 * and alignments exist, -1 is returned and the resulting top
544 * queue_limits will have the misaligned flag set to indicate that
545 * the alignment_offset is undefined.
546 */
547int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
548 sector_t start)
549{
550 unsigned int top, bottom, alignment, ret = 0;
551
552 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
553 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
554 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
555 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
556 b->max_write_zeroes_sectors);
557 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
558 b->max_zone_append_sectors);
559 t->bounce = max(t->bounce, b->bounce);
560
561 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
562 b->seg_boundary_mask);
563 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
564 b->virt_boundary_mask);
565
566 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
567 t->max_discard_segments = min_not_zero(t->max_discard_segments,
568 b->max_discard_segments);
569 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
570 b->max_integrity_segments);
571
572 t->max_segment_size = min_not_zero(t->max_segment_size,
573 b->max_segment_size);
574
575 t->misaligned |= b->misaligned;
576
577 alignment = queue_limit_alignment_offset(b, start);
578
579 /* Bottom device has different alignment. Check that it is
580 * compatible with the current top alignment.
581 */
582 if (t->alignment_offset != alignment) {
583
584 top = max(t->physical_block_size, t->io_min)
585 + t->alignment_offset;
586 bottom = max(b->physical_block_size, b->io_min) + alignment;
587
588 /* Verify that top and bottom intervals line up */
589 if (max(top, bottom) % min(top, bottom)) {
590 t->misaligned = 1;
591 ret = -1;
592 }
593 }
594
595 t->logical_block_size = max(t->logical_block_size,
596 b->logical_block_size);
597
598 t->physical_block_size = max(t->physical_block_size,
599 b->physical_block_size);
600
601 t->io_min = max(t->io_min, b->io_min);
602 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
603 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
604
605 /* Set non-power-of-2 compatible chunk_sectors boundary */
606 if (b->chunk_sectors)
607 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
608
609 /* Physical block size a multiple of the logical block size? */
610 if (t->physical_block_size & (t->logical_block_size - 1)) {
611 t->physical_block_size = t->logical_block_size;
612 t->misaligned = 1;
613 ret = -1;
614 }
615
616 /* Minimum I/O a multiple of the physical block size? */
617 if (t->io_min & (t->physical_block_size - 1)) {
618 t->io_min = t->physical_block_size;
619 t->misaligned = 1;
620 ret = -1;
621 }
622
623 /* Optimal I/O a multiple of the physical block size? */
624 if (t->io_opt & (t->physical_block_size - 1)) {
625 t->io_opt = 0;
626 t->misaligned = 1;
627 ret = -1;
628 }
629
630 /* chunk_sectors a multiple of the physical block size? */
631 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
632 t->chunk_sectors = 0;
633 t->misaligned = 1;
634 ret = -1;
635 }
636
637 t->raid_partial_stripes_expensive =
638 max(t->raid_partial_stripes_expensive,
639 b->raid_partial_stripes_expensive);
640
641 /* Find lowest common alignment_offset */
642 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
643 % max(t->physical_block_size, t->io_min);
644
645 /* Verify that new alignment_offset is on a logical block boundary */
646 if (t->alignment_offset & (t->logical_block_size - 1)) {
647 t->misaligned = 1;
648 ret = -1;
649 }
650
651 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
652 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
653 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
654
655 /* Discard alignment and granularity */
656 if (b->discard_granularity) {
657 alignment = queue_limit_discard_alignment(b, start);
658
659 if (t->discard_granularity != 0 &&
660 t->discard_alignment != alignment) {
661 top = t->discard_granularity + t->discard_alignment;
662 bottom = b->discard_granularity + alignment;
663
664 /* Verify that top and bottom intervals line up */
665 if ((max(top, bottom) % min(top, bottom)) != 0)
666 t->discard_misaligned = 1;
667 }
668
669 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
670 b->max_discard_sectors);
671 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
672 b->max_hw_discard_sectors);
673 t->discard_granularity = max(t->discard_granularity,
674 b->discard_granularity);
675 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
676 t->discard_granularity;
677 }
678 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
679 b->max_secure_erase_sectors);
680 t->zone_write_granularity = max(t->zone_write_granularity,
681 b->zone_write_granularity);
682 t->zoned = max(t->zoned, b->zoned);
683 return ret;
684}
685EXPORT_SYMBOL(blk_stack_limits);
686
687/**
688 * disk_stack_limits - adjust queue limits for stacked drivers
689 * @disk: MD/DM gendisk (top)
690 * @bdev: the underlying block device (bottom)
691 * @offset: offset to beginning of data within component device
692 *
693 * Description:
694 * Merges the limits for a top level gendisk and a bottom level
695 * block_device.
696 */
697void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
698 sector_t offset)
699{
700 struct request_queue *t = disk->queue;
701
702 if (blk_stack_limits(&t->limits, &bdev_get_queue(bdev)->limits,
703 get_start_sect(bdev) + (offset >> 9)) < 0)
704 pr_notice("%s: Warning: Device %pg is misaligned\n",
705 disk->disk_name, bdev);
706
707 disk_update_readahead(disk);
708}
709EXPORT_SYMBOL(disk_stack_limits);
710
711/**
712 * blk_queue_update_dma_pad - update pad mask
713 * @q: the request queue for the device
714 * @mask: pad mask
715 *
716 * Update dma pad mask.
717 *
718 * Appending pad buffer to a request modifies the last entry of a
719 * scatter list such that it includes the pad buffer.
720 **/
721void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
722{
723 if (mask > q->dma_pad_mask)
724 q->dma_pad_mask = mask;
725}
726EXPORT_SYMBOL(blk_queue_update_dma_pad);
727
728/**
729 * blk_queue_segment_boundary - set boundary rules for segment merging
730 * @q: the request queue for the device
731 * @mask: the memory boundary mask
732 **/
733void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
734{
735 if (mask < PAGE_SIZE - 1) {
736 mask = PAGE_SIZE - 1;
737 printk(KERN_INFO "%s: set to minimum %lx\n",
738 __func__, mask);
739 }
740
741 q->limits.seg_boundary_mask = mask;
742}
743EXPORT_SYMBOL(blk_queue_segment_boundary);
744
745/**
746 * blk_queue_virt_boundary - set boundary rules for bio merging
747 * @q: the request queue for the device
748 * @mask: the memory boundary mask
749 **/
750void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
751{
752 q->limits.virt_boundary_mask = mask;
753
754 /*
755 * Devices that require a virtual boundary do not support scatter/gather
756 * I/O natively, but instead require a descriptor list entry for each
757 * page (which might not be idential to the Linux PAGE_SIZE). Because
758 * of that they are not limited by our notion of "segment size".
759 */
760 if (mask)
761 q->limits.max_segment_size = UINT_MAX;
762}
763EXPORT_SYMBOL(blk_queue_virt_boundary);
764
765/**
766 * blk_queue_dma_alignment - set dma length and memory alignment
767 * @q: the request queue for the device
768 * @mask: alignment mask
769 *
770 * description:
771 * set required memory and length alignment for direct dma transactions.
772 * this is used when building direct io requests for the queue.
773 *
774 **/
775void blk_queue_dma_alignment(struct request_queue *q, int mask)
776{
777 q->limits.dma_alignment = mask;
778}
779EXPORT_SYMBOL(blk_queue_dma_alignment);
780
781/**
782 * blk_queue_update_dma_alignment - update dma length and memory alignment
783 * @q: the request queue for the device
784 * @mask: alignment mask
785 *
786 * description:
787 * update required memory and length alignment for direct dma transactions.
788 * If the requested alignment is larger than the current alignment, then
789 * the current queue alignment is updated to the new value, otherwise it
790 * is left alone. The design of this is to allow multiple objects
791 * (driver, device, transport etc) to set their respective
792 * alignments without having them interfere.
793 *
794 **/
795void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
796{
797 BUG_ON(mask > PAGE_SIZE);
798
799 if (mask > q->limits.dma_alignment)
800 q->limits.dma_alignment = mask;
801}
802EXPORT_SYMBOL(blk_queue_update_dma_alignment);
803
804/**
805 * blk_set_queue_depth - tell the block layer about the device queue depth
806 * @q: the request queue for the device
807 * @depth: queue depth
808 *
809 */
810void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
811{
812 q->queue_depth = depth;
813 rq_qos_queue_depth_changed(q);
814}
815EXPORT_SYMBOL(blk_set_queue_depth);
816
817/**
818 * blk_queue_write_cache - configure queue's write cache
819 * @q: the request queue for the device
820 * @wc: write back cache on or off
821 * @fua: device supports FUA writes, if true
822 *
823 * Tell the block layer about the write cache of @q.
824 */
825void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
826{
827 if (wc)
828 blk_queue_flag_set(QUEUE_FLAG_WC, q);
829 else
830 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
831 if (fua)
832 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
833 else
834 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
835
836 wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
837}
838EXPORT_SYMBOL_GPL(blk_queue_write_cache);
839
840/**
841 * blk_queue_required_elevator_features - Set a queue required elevator features
842 * @q: the request queue for the target device
843 * @features: Required elevator features OR'ed together
844 *
845 * Tell the block layer that for the device controlled through @q, only the
846 * only elevators that can be used are those that implement at least the set of
847 * features specified by @features.
848 */
849void blk_queue_required_elevator_features(struct request_queue *q,
850 unsigned int features)
851{
852 q->required_elevator_features = features;
853}
854EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
855
856/**
857 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
858 * @q: the request queue for the device
859 * @dev: the device pointer for dma
860 *
861 * Tell the block layer about merging the segments by dma map of @q.
862 */
863bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
864 struct device *dev)
865{
866 unsigned long boundary = dma_get_merge_boundary(dev);
867
868 if (!boundary)
869 return false;
870
871 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
872 blk_queue_virt_boundary(q, boundary);
873
874 return true;
875}
876EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
877
878static bool disk_has_partitions(struct gendisk *disk)
879{
880 unsigned long idx;
881 struct block_device *part;
882 bool ret = false;
883
884 rcu_read_lock();
885 xa_for_each(&disk->part_tbl, idx, part) {
886 if (bdev_is_partition(part)) {
887 ret = true;
888 break;
889 }
890 }
891 rcu_read_unlock();
892
893 return ret;
894}
895
896/**
897 * disk_set_zoned - configure the zoned model for a disk
898 * @disk: the gendisk of the queue to configure
899 * @model: the zoned model to set
900 *
901 * Set the zoned model of @disk to @model.
902 *
903 * When @model is BLK_ZONED_HM (host managed), this should be called only
904 * if zoned block device support is enabled (CONFIG_BLK_DEV_ZONED option).
905 * If @model specifies BLK_ZONED_HA (host aware), the effective model used
906 * depends on CONFIG_BLK_DEV_ZONED settings and on the existence of partitions
907 * on the disk.
908 */
909void disk_set_zoned(struct gendisk *disk, enum blk_zoned_model model)
910{
911 struct request_queue *q = disk->queue;
912
913 switch (model) {
914 case BLK_ZONED_HM:
915 /*
916 * Host managed devices are supported only if
917 * CONFIG_BLK_DEV_ZONED is enabled.
918 */
919 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
920 break;
921 case BLK_ZONED_HA:
922 /*
923 * Host aware devices can be treated either as regular block
924 * devices (similar to drive managed devices) or as zoned block
925 * devices to take advantage of the zone command set, similarly
926 * to host managed devices. We try the latter if there are no
927 * partitions and zoned block device support is enabled, else
928 * we do nothing special as far as the block layer is concerned.
929 */
930 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) ||
931 disk_has_partitions(disk))
932 model = BLK_ZONED_NONE;
933 break;
934 case BLK_ZONED_NONE:
935 default:
936 if (WARN_ON_ONCE(model != BLK_ZONED_NONE))
937 model = BLK_ZONED_NONE;
938 break;
939 }
940
941 q->limits.zoned = model;
942 if (model != BLK_ZONED_NONE) {
943 /*
944 * Set the zone write granularity to the device logical block
945 * size by default. The driver can change this value if needed.
946 */
947 blk_queue_zone_write_granularity(q,
948 queue_logical_block_size(q));
949 } else {
950 disk_clear_zone_settings(disk);
951 }
952}
953EXPORT_SYMBOL_GPL(disk_set_zoned);
954
955int bdev_alignment_offset(struct block_device *bdev)
956{
957 struct request_queue *q = bdev_get_queue(bdev);
958
959 if (q->limits.misaligned)
960 return -1;
961 if (bdev_is_partition(bdev))
962 return queue_limit_alignment_offset(&q->limits,
963 bdev->bd_start_sect);
964 return q->limits.alignment_offset;
965}
966EXPORT_SYMBOL_GPL(bdev_alignment_offset);
967
968unsigned int bdev_discard_alignment(struct block_device *bdev)
969{
970 struct request_queue *q = bdev_get_queue(bdev);
971
972 if (bdev_is_partition(bdev))
973 return queue_limit_discard_alignment(&q->limits,
974 bdev->bd_start_sect);
975 return q->limits.discard_alignment;
976}
977EXPORT_SYMBOL_GPL(bdev_discard_alignment);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blk-integrity.h>
10#include <linux/pagemap.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/gcd.h>
13#include <linux/lcm.h>
14#include <linux/jiffies.h>
15#include <linux/gfp.h>
16#include <linux/dma-mapping.h>
17
18#include "blk.h"
19#include "blk-rq-qos.h"
20#include "blk-wbt.h"
21
22void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
23{
24 q->rq_timeout = timeout;
25}
26EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
27
28/**
29 * blk_set_stacking_limits - set default limits for stacking devices
30 * @lim: the queue_limits structure to reset
31 *
32 * Prepare queue limits for applying limits from underlying devices using
33 * blk_stack_limits().
34 */
35void blk_set_stacking_limits(struct queue_limits *lim)
36{
37 memset(lim, 0, sizeof(*lim));
38 lim->logical_block_size = SECTOR_SIZE;
39 lim->physical_block_size = SECTOR_SIZE;
40 lim->io_min = SECTOR_SIZE;
41 lim->discard_granularity = SECTOR_SIZE;
42 lim->dma_alignment = SECTOR_SIZE - 1;
43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
44
45 /* Inherit limits from component devices */
46 lim->max_segments = USHRT_MAX;
47 lim->max_discard_segments = USHRT_MAX;
48 lim->max_hw_sectors = UINT_MAX;
49 lim->max_segment_size = UINT_MAX;
50 lim->max_sectors = UINT_MAX;
51 lim->max_dev_sectors = UINT_MAX;
52 lim->max_write_zeroes_sectors = UINT_MAX;
53 lim->max_hw_zone_append_sectors = UINT_MAX;
54 lim->max_user_discard_sectors = UINT_MAX;
55}
56EXPORT_SYMBOL(blk_set_stacking_limits);
57
58void blk_apply_bdi_limits(struct backing_dev_info *bdi,
59 struct queue_limits *lim)
60{
61 /*
62 * For read-ahead of large files to be effective, we need to read ahead
63 * at least twice the optimal I/O size.
64 */
65 bdi->ra_pages = max(lim->io_opt * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
66 bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT;
67}
68
69static int blk_validate_zoned_limits(struct queue_limits *lim)
70{
71 if (!(lim->features & BLK_FEAT_ZONED)) {
72 if (WARN_ON_ONCE(lim->max_open_zones) ||
73 WARN_ON_ONCE(lim->max_active_zones) ||
74 WARN_ON_ONCE(lim->zone_write_granularity) ||
75 WARN_ON_ONCE(lim->max_zone_append_sectors))
76 return -EINVAL;
77 return 0;
78 }
79
80 if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)))
81 return -EINVAL;
82
83 /*
84 * Given that active zones include open zones, the maximum number of
85 * open zones cannot be larger than the maximum number of active zones.
86 */
87 if (lim->max_active_zones &&
88 lim->max_open_zones > lim->max_active_zones)
89 return -EINVAL;
90
91 if (lim->zone_write_granularity < lim->logical_block_size)
92 lim->zone_write_granularity = lim->logical_block_size;
93
94 /*
95 * The Zone Append size is limited by the maximum I/O size and the zone
96 * size given that it can't span zones.
97 *
98 * If no max_hw_zone_append_sectors limit is provided, the block layer
99 * will emulated it, else we're also bound by the hardware limit.
100 */
101 lim->max_zone_append_sectors =
102 min_not_zero(lim->max_hw_zone_append_sectors,
103 min(lim->chunk_sectors, lim->max_hw_sectors));
104 return 0;
105}
106
107static int blk_validate_integrity_limits(struct queue_limits *lim)
108{
109 struct blk_integrity *bi = &lim->integrity;
110
111 if (!bi->tuple_size) {
112 if (bi->csum_type != BLK_INTEGRITY_CSUM_NONE ||
113 bi->tag_size || ((bi->flags & BLK_INTEGRITY_REF_TAG))) {
114 pr_warn("invalid PI settings.\n");
115 return -EINVAL;
116 }
117 return 0;
118 }
119
120 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) {
121 pr_warn("integrity support disabled.\n");
122 return -EINVAL;
123 }
124
125 if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE &&
126 (bi->flags & BLK_INTEGRITY_REF_TAG)) {
127 pr_warn("ref tag not support without checksum.\n");
128 return -EINVAL;
129 }
130
131 if (!bi->interval_exp)
132 bi->interval_exp = ilog2(lim->logical_block_size);
133
134 return 0;
135}
136
137/*
138 * Returns max guaranteed bytes which we can fit in a bio.
139 *
140 * We request that an atomic_write is ITER_UBUF iov_iter (so a single vector),
141 * so we assume that we can fit in at least PAGE_SIZE in a segment, apart from
142 * the first and last segments.
143 */
144static unsigned int blk_queue_max_guaranteed_bio(struct queue_limits *lim)
145{
146 unsigned int max_segments = min(BIO_MAX_VECS, lim->max_segments);
147 unsigned int length;
148
149 length = min(max_segments, 2) * lim->logical_block_size;
150 if (max_segments > 2)
151 length += (max_segments - 2) * PAGE_SIZE;
152
153 return length;
154}
155
156static void blk_atomic_writes_update_limits(struct queue_limits *lim)
157{
158 unsigned int unit_limit = min(lim->max_hw_sectors << SECTOR_SHIFT,
159 blk_queue_max_guaranteed_bio(lim));
160
161 unit_limit = rounddown_pow_of_two(unit_limit);
162
163 lim->atomic_write_max_sectors =
164 min(lim->atomic_write_hw_max >> SECTOR_SHIFT,
165 lim->max_hw_sectors);
166 lim->atomic_write_unit_min =
167 min(lim->atomic_write_hw_unit_min, unit_limit);
168 lim->atomic_write_unit_max =
169 min(lim->atomic_write_hw_unit_max, unit_limit);
170 lim->atomic_write_boundary_sectors =
171 lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
172}
173
174static void blk_validate_atomic_write_limits(struct queue_limits *lim)
175{
176 unsigned int boundary_sectors;
177
178 if (!lim->atomic_write_hw_max)
179 goto unsupported;
180
181 if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_min)))
182 goto unsupported;
183
184 if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_max)))
185 goto unsupported;
186
187 if (WARN_ON_ONCE(lim->atomic_write_hw_unit_min >
188 lim->atomic_write_hw_unit_max))
189 goto unsupported;
190
191 if (WARN_ON_ONCE(lim->atomic_write_hw_unit_max >
192 lim->atomic_write_hw_max))
193 goto unsupported;
194
195 boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
196
197 if (boundary_sectors) {
198 if (WARN_ON_ONCE(lim->atomic_write_hw_max >
199 lim->atomic_write_hw_boundary))
200 goto unsupported;
201 /*
202 * A feature of boundary support is that it disallows bios to
203 * be merged which would result in a merged request which
204 * crosses either a chunk sector or atomic write HW boundary,
205 * even though chunk sectors may be just set for performance.
206 * For simplicity, disallow atomic writes for a chunk sector
207 * which is non-zero and smaller than atomic write HW boundary.
208 * Furthermore, chunk sectors must be a multiple of atomic
209 * write HW boundary. Otherwise boundary support becomes
210 * complicated.
211 * Devices which do not conform to these rules can be dealt
212 * with if and when they show up.
213 */
214 if (WARN_ON_ONCE(lim->chunk_sectors % boundary_sectors))
215 goto unsupported;
216
217 /*
218 * The boundary size just needs to be a multiple of unit_max
219 * (and not necessarily a power-of-2), so this following check
220 * could be relaxed in future.
221 * Furthermore, if needed, unit_max could even be reduced so
222 * that it is compliant with a !power-of-2 boundary.
223 */
224 if (!is_power_of_2(boundary_sectors))
225 goto unsupported;
226 }
227
228 blk_atomic_writes_update_limits(lim);
229 return;
230
231unsupported:
232 lim->atomic_write_max_sectors = 0;
233 lim->atomic_write_boundary_sectors = 0;
234 lim->atomic_write_unit_min = 0;
235 lim->atomic_write_unit_max = 0;
236}
237
238/*
239 * Check that the limits in lim are valid, initialize defaults for unset
240 * values, and cap values based on others where needed.
241 */
242int blk_validate_limits(struct queue_limits *lim)
243{
244 unsigned int max_hw_sectors;
245 unsigned int logical_block_sectors;
246 int err;
247
248 /*
249 * Unless otherwise specified, default to 512 byte logical blocks and a
250 * physical block size equal to the logical block size.
251 */
252 if (!lim->logical_block_size)
253 lim->logical_block_size = SECTOR_SIZE;
254 else if (blk_validate_block_size(lim->logical_block_size)) {
255 pr_warn("Invalid logical block size (%d)\n", lim->logical_block_size);
256 return -EINVAL;
257 }
258 if (lim->physical_block_size < lim->logical_block_size)
259 lim->physical_block_size = lim->logical_block_size;
260
261 /*
262 * The minimum I/O size defaults to the physical block size unless
263 * explicitly overridden.
264 */
265 if (lim->io_min < lim->physical_block_size)
266 lim->io_min = lim->physical_block_size;
267
268 /*
269 * The optimal I/O size may not be aligned to physical block size
270 * (because it may be limited by dma engines which have no clue about
271 * block size of the disks attached to them), so we round it down here.
272 */
273 lim->io_opt = round_down(lim->io_opt, lim->physical_block_size);
274
275 /*
276 * max_hw_sectors has a somewhat weird default for historical reason,
277 * but driver really should set their own instead of relying on this
278 * value.
279 *
280 * The block layer relies on the fact that every driver can
281 * handle at lest a page worth of data per I/O, and needs the value
282 * aligned to the logical block size.
283 */
284 if (!lim->max_hw_sectors)
285 lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
286 if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
287 return -EINVAL;
288 logical_block_sectors = lim->logical_block_size >> SECTOR_SHIFT;
289 if (WARN_ON_ONCE(logical_block_sectors > lim->max_hw_sectors))
290 return -EINVAL;
291 lim->max_hw_sectors = round_down(lim->max_hw_sectors,
292 logical_block_sectors);
293
294 /*
295 * The actual max_sectors value is a complex beast and also takes the
296 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
297 * value into account. The ->max_sectors value is always calculated
298 * from these, so directly setting it won't have any effect.
299 */
300 max_hw_sectors = min_not_zero(lim->max_hw_sectors,
301 lim->max_dev_sectors);
302 if (lim->max_user_sectors) {
303 if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
304 return -EINVAL;
305 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
306 } else if (lim->io_opt > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
307 lim->max_sectors =
308 min(max_hw_sectors, lim->io_opt >> SECTOR_SHIFT);
309 } else if (lim->io_min > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
310 lim->max_sectors =
311 min(max_hw_sectors, lim->io_min >> SECTOR_SHIFT);
312 } else {
313 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
314 }
315 lim->max_sectors = round_down(lim->max_sectors,
316 logical_block_sectors);
317
318 /*
319 * Random default for the maximum number of segments. Driver should not
320 * rely on this and set their own.
321 */
322 if (!lim->max_segments)
323 lim->max_segments = BLK_MAX_SEGMENTS;
324
325 lim->max_discard_sectors =
326 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
327
328 if (!lim->max_discard_segments)
329 lim->max_discard_segments = 1;
330
331 if (lim->discard_granularity < lim->physical_block_size)
332 lim->discard_granularity = lim->physical_block_size;
333
334 /*
335 * By default there is no limit on the segment boundary alignment,
336 * but if there is one it can't be smaller than the page size as
337 * that would break all the normal I/O patterns.
338 */
339 if (!lim->seg_boundary_mask)
340 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
341 if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
342 return -EINVAL;
343
344 /*
345 * Stacking device may have both virtual boundary and max segment
346 * size limit, so allow this setting now, and long-term the two
347 * might need to move out of stacking limits since we have immutable
348 * bvec and lower layer bio splitting is supposed to handle the two
349 * correctly.
350 */
351 if (lim->virt_boundary_mask) {
352 if (!lim->max_segment_size)
353 lim->max_segment_size = UINT_MAX;
354 } else {
355 /*
356 * The maximum segment size has an odd historic 64k default that
357 * drivers probably should override. Just like the I/O size we
358 * require drivers to at least handle a full page per segment.
359 */
360 if (!lim->max_segment_size)
361 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
362 if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
363 return -EINVAL;
364 }
365
366 /*
367 * We require drivers to at least do logical block aligned I/O, but
368 * historically could not check for that due to the separate calls
369 * to set the limits. Once the transition is finished the check
370 * below should be narrowed down to check the logical block size.
371 */
372 if (!lim->dma_alignment)
373 lim->dma_alignment = SECTOR_SIZE - 1;
374 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
375 return -EINVAL;
376
377 if (lim->alignment_offset) {
378 lim->alignment_offset &= (lim->physical_block_size - 1);
379 lim->flags &= ~BLK_FLAG_MISALIGNED;
380 }
381
382 if (!(lim->features & BLK_FEAT_WRITE_CACHE))
383 lim->features &= ~BLK_FEAT_FUA;
384
385 blk_validate_atomic_write_limits(lim);
386
387 err = blk_validate_integrity_limits(lim);
388 if (err)
389 return err;
390 return blk_validate_zoned_limits(lim);
391}
392EXPORT_SYMBOL_GPL(blk_validate_limits);
393
394/*
395 * Set the default limits for a newly allocated queue. @lim contains the
396 * initial limits set by the driver, which could be no limit in which case
397 * all fields are cleared to zero.
398 */
399int blk_set_default_limits(struct queue_limits *lim)
400{
401 /*
402 * Most defaults are set by capping the bounds in blk_validate_limits,
403 * but max_user_discard_sectors is special and needs an explicit
404 * initialization to the max value here.
405 */
406 lim->max_user_discard_sectors = UINT_MAX;
407 return blk_validate_limits(lim);
408}
409
410/**
411 * queue_limits_commit_update - commit an atomic update of queue limits
412 * @q: queue to update
413 * @lim: limits to apply
414 *
415 * Apply the limits in @lim that were obtained from queue_limits_start_update()
416 * and updated by the caller to @q.
417 *
418 * Returns 0 if successful, else a negative error code.
419 */
420int queue_limits_commit_update(struct request_queue *q,
421 struct queue_limits *lim)
422{
423 int error;
424
425 error = blk_validate_limits(lim);
426 if (error)
427 goto out_unlock;
428
429#ifdef CONFIG_BLK_INLINE_ENCRYPTION
430 if (q->crypto_profile && lim->integrity.tag_size) {
431 pr_warn("blk-integrity: Integrity and hardware inline encryption are not supported together.\n");
432 error = -EINVAL;
433 goto out_unlock;
434 }
435#endif
436
437 q->limits = *lim;
438 if (q->disk)
439 blk_apply_bdi_limits(q->disk->bdi, lim);
440out_unlock:
441 mutex_unlock(&q->limits_lock);
442 return error;
443}
444EXPORT_SYMBOL_GPL(queue_limits_commit_update);
445
446/**
447 * queue_limits_commit_update_frozen - commit an atomic update of queue limits
448 * @q: queue to update
449 * @lim: limits to apply
450 *
451 * Apply the limits in @lim that were obtained from queue_limits_start_update()
452 * and updated with the new values by the caller to @q. Freezes the queue
453 * before the update and unfreezes it after.
454 *
455 * Returns 0 if successful, else a negative error code.
456 */
457int queue_limits_commit_update_frozen(struct request_queue *q,
458 struct queue_limits *lim)
459{
460 int ret;
461
462 blk_mq_freeze_queue(q);
463 ret = queue_limits_commit_update(q, lim);
464 blk_mq_unfreeze_queue(q);
465
466 return ret;
467}
468EXPORT_SYMBOL_GPL(queue_limits_commit_update_frozen);
469
470/**
471 * queue_limits_set - apply queue limits to queue
472 * @q: queue to update
473 * @lim: limits to apply
474 *
475 * Apply the limits in @lim that were freshly initialized to @q.
476 * To update existing limits use queue_limits_start_update() and
477 * queue_limits_commit_update() instead.
478 *
479 * Returns 0 if successful, else a negative error code.
480 */
481int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
482{
483 mutex_lock(&q->limits_lock);
484 return queue_limits_commit_update(q, lim);
485}
486EXPORT_SYMBOL_GPL(queue_limits_set);
487
488static int queue_limit_alignment_offset(const struct queue_limits *lim,
489 sector_t sector)
490{
491 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
492 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
493 << SECTOR_SHIFT;
494
495 return (granularity + lim->alignment_offset - alignment) % granularity;
496}
497
498static unsigned int queue_limit_discard_alignment(
499 const struct queue_limits *lim, sector_t sector)
500{
501 unsigned int alignment, granularity, offset;
502
503 if (!lim->max_discard_sectors)
504 return 0;
505
506 /* Why are these in bytes, not sectors? */
507 alignment = lim->discard_alignment >> SECTOR_SHIFT;
508 granularity = lim->discard_granularity >> SECTOR_SHIFT;
509
510 /* Offset of the partition start in 'granularity' sectors */
511 offset = sector_div(sector, granularity);
512
513 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
514 offset = (granularity + alignment - offset) % granularity;
515
516 /* Turn it back into bytes, gaah */
517 return offset << SECTOR_SHIFT;
518}
519
520static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
521{
522 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
523 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
524 sectors = PAGE_SIZE >> SECTOR_SHIFT;
525 return sectors;
526}
527
528/* Check if second and later bottom devices are compliant */
529static bool blk_stack_atomic_writes_tail(struct queue_limits *t,
530 struct queue_limits *b)
531{
532 /* We're not going to support different boundary sizes.. yet */
533 if (t->atomic_write_hw_boundary != b->atomic_write_hw_boundary)
534 return false;
535
536 /* Can't support this */
537 if (t->atomic_write_hw_unit_min > b->atomic_write_hw_unit_max)
538 return false;
539
540 /* Or this */
541 if (t->atomic_write_hw_unit_max < b->atomic_write_hw_unit_min)
542 return false;
543
544 t->atomic_write_hw_max = min(t->atomic_write_hw_max,
545 b->atomic_write_hw_max);
546 t->atomic_write_hw_unit_min = max(t->atomic_write_hw_unit_min,
547 b->atomic_write_hw_unit_min);
548 t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max,
549 b->atomic_write_hw_unit_max);
550 return true;
551}
552
553/* Check for valid boundary of first bottom device */
554static bool blk_stack_atomic_writes_boundary_head(struct queue_limits *t,
555 struct queue_limits *b)
556{
557 /*
558 * Ensure atomic write boundary is aligned with chunk sectors. Stacked
559 * devices store chunk sectors in t->io_min.
560 */
561 if (b->atomic_write_hw_boundary > t->io_min &&
562 b->atomic_write_hw_boundary % t->io_min)
563 return false;
564 if (t->io_min > b->atomic_write_hw_boundary &&
565 t->io_min % b->atomic_write_hw_boundary)
566 return false;
567
568 t->atomic_write_hw_boundary = b->atomic_write_hw_boundary;
569 return true;
570}
571
572
573/* Check stacking of first bottom device */
574static bool blk_stack_atomic_writes_head(struct queue_limits *t,
575 struct queue_limits *b)
576{
577 if (b->atomic_write_hw_boundary &&
578 !blk_stack_atomic_writes_boundary_head(t, b))
579 return false;
580
581 if (t->io_min <= SECTOR_SIZE) {
582 /* No chunk sectors, so use bottom device values directly */
583 t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
584 t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
585 t->atomic_write_hw_max = b->atomic_write_hw_max;
586 return true;
587 }
588
589 /*
590 * Find values for limits which work for chunk size.
591 * b->atomic_write_hw_unit_{min, max} may not be aligned with chunk
592 * size (t->io_min), as chunk size is not restricted to a power-of-2.
593 * So we need to find highest power-of-2 which works for the chunk
594 * size.
595 * As an example scenario, we could have b->unit_max = 16K and
596 * t->io_min = 24K. For this case, reduce t->unit_max to a value
597 * aligned with both limits, i.e. 8K in this example.
598 */
599 t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
600 while (t->io_min % t->atomic_write_hw_unit_max)
601 t->atomic_write_hw_unit_max /= 2;
602
603 t->atomic_write_hw_unit_min = min(b->atomic_write_hw_unit_min,
604 t->atomic_write_hw_unit_max);
605 t->atomic_write_hw_max = min(b->atomic_write_hw_max, t->io_min);
606
607 return true;
608}
609
610static void blk_stack_atomic_writes_limits(struct queue_limits *t,
611 struct queue_limits *b, sector_t start)
612{
613 if (!(t->features & BLK_FEAT_ATOMIC_WRITES_STACKED))
614 goto unsupported;
615
616 if (!b->atomic_write_unit_min)
617 goto unsupported;
618
619 if (!blk_atomic_write_start_sect_aligned(start, b))
620 goto unsupported;
621
622 /*
623 * If atomic_write_hw_max is set, we have already stacked 1x bottom
624 * device, so check for compliance.
625 */
626 if (t->atomic_write_hw_max) {
627 if (!blk_stack_atomic_writes_tail(t, b))
628 goto unsupported;
629 return;
630 }
631
632 if (!blk_stack_atomic_writes_head(t, b))
633 goto unsupported;
634 return;
635
636unsupported:
637 t->atomic_write_hw_max = 0;
638 t->atomic_write_hw_unit_max = 0;
639 t->atomic_write_hw_unit_min = 0;
640 t->atomic_write_hw_boundary = 0;
641 t->features &= ~BLK_FEAT_ATOMIC_WRITES_STACKED;
642}
643
644/**
645 * blk_stack_limits - adjust queue_limits for stacked devices
646 * @t: the stacking driver limits (top device)
647 * @b: the underlying queue limits (bottom, component device)
648 * @start: first data sector within component device
649 *
650 * Description:
651 * This function is used by stacking drivers like MD and DM to ensure
652 * that all component devices have compatible block sizes and
653 * alignments. The stacking driver must provide a queue_limits
654 * struct (top) and then iteratively call the stacking function for
655 * all component (bottom) devices. The stacking function will
656 * attempt to combine the values and ensure proper alignment.
657 *
658 * Returns 0 if the top and bottom queue_limits are compatible. The
659 * top device's block sizes and alignment offsets may be adjusted to
660 * ensure alignment with the bottom device. If no compatible sizes
661 * and alignments exist, -1 is returned and the resulting top
662 * queue_limits will have the misaligned flag set to indicate that
663 * the alignment_offset is undefined.
664 */
665int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
666 sector_t start)
667{
668 unsigned int top, bottom, alignment, ret = 0;
669
670 t->features |= (b->features & BLK_FEAT_INHERIT_MASK);
671
672 /*
673 * Some feaures need to be supported both by the stacking driver and all
674 * underlying devices. The stacking driver sets these flags before
675 * stacking the limits, and this will clear the flags if any of the
676 * underlying devices does not support it.
677 */
678 if (!(b->features & BLK_FEAT_NOWAIT))
679 t->features &= ~BLK_FEAT_NOWAIT;
680 if (!(b->features & BLK_FEAT_POLL))
681 t->features &= ~BLK_FEAT_POLL;
682
683 t->flags |= (b->flags & BLK_FLAG_MISALIGNED);
684
685 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
686 t->max_user_sectors = min_not_zero(t->max_user_sectors,
687 b->max_user_sectors);
688 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
689 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
690 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
691 b->max_write_zeroes_sectors);
692 t->max_hw_zone_append_sectors = min(t->max_hw_zone_append_sectors,
693 b->max_hw_zone_append_sectors);
694
695 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
696 b->seg_boundary_mask);
697 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
698 b->virt_boundary_mask);
699
700 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
701 t->max_discard_segments = min_not_zero(t->max_discard_segments,
702 b->max_discard_segments);
703 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
704 b->max_integrity_segments);
705
706 t->max_segment_size = min_not_zero(t->max_segment_size,
707 b->max_segment_size);
708
709 alignment = queue_limit_alignment_offset(b, start);
710
711 /* Bottom device has different alignment. Check that it is
712 * compatible with the current top alignment.
713 */
714 if (t->alignment_offset != alignment) {
715
716 top = max(t->physical_block_size, t->io_min)
717 + t->alignment_offset;
718 bottom = max(b->physical_block_size, b->io_min) + alignment;
719
720 /* Verify that top and bottom intervals line up */
721 if (max(top, bottom) % min(top, bottom)) {
722 t->flags |= BLK_FLAG_MISALIGNED;
723 ret = -1;
724 }
725 }
726
727 t->logical_block_size = max(t->logical_block_size,
728 b->logical_block_size);
729
730 t->physical_block_size = max(t->physical_block_size,
731 b->physical_block_size);
732
733 t->io_min = max(t->io_min, b->io_min);
734 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
735 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
736
737 /* Set non-power-of-2 compatible chunk_sectors boundary */
738 if (b->chunk_sectors)
739 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
740
741 /* Physical block size a multiple of the logical block size? */
742 if (t->physical_block_size & (t->logical_block_size - 1)) {
743 t->physical_block_size = t->logical_block_size;
744 t->flags |= BLK_FLAG_MISALIGNED;
745 ret = -1;
746 }
747
748 /* Minimum I/O a multiple of the physical block size? */
749 if (t->io_min & (t->physical_block_size - 1)) {
750 t->io_min = t->physical_block_size;
751 t->flags |= BLK_FLAG_MISALIGNED;
752 ret = -1;
753 }
754
755 /* Optimal I/O a multiple of the physical block size? */
756 if (t->io_opt & (t->physical_block_size - 1)) {
757 t->io_opt = 0;
758 t->flags |= BLK_FLAG_MISALIGNED;
759 ret = -1;
760 }
761
762 /* chunk_sectors a multiple of the physical block size? */
763 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
764 t->chunk_sectors = 0;
765 t->flags |= BLK_FLAG_MISALIGNED;
766 ret = -1;
767 }
768
769 /* Find lowest common alignment_offset */
770 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
771 % max(t->physical_block_size, t->io_min);
772
773 /* Verify that new alignment_offset is on a logical block boundary */
774 if (t->alignment_offset & (t->logical_block_size - 1)) {
775 t->flags |= BLK_FLAG_MISALIGNED;
776 ret = -1;
777 }
778
779 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
780 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
781 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
782
783 /* Discard alignment and granularity */
784 if (b->discard_granularity) {
785 alignment = queue_limit_discard_alignment(b, start);
786
787 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
788 b->max_discard_sectors);
789 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
790 b->max_hw_discard_sectors);
791 t->discard_granularity = max(t->discard_granularity,
792 b->discard_granularity);
793 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
794 t->discard_granularity;
795 }
796 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
797 b->max_secure_erase_sectors);
798 t->zone_write_granularity = max(t->zone_write_granularity,
799 b->zone_write_granularity);
800 if (!(t->features & BLK_FEAT_ZONED)) {
801 t->zone_write_granularity = 0;
802 t->max_zone_append_sectors = 0;
803 }
804 blk_stack_atomic_writes_limits(t, b, start);
805
806 return ret;
807}
808EXPORT_SYMBOL(blk_stack_limits);
809
810/**
811 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
812 * @t: the stacking driver limits (top device)
813 * @bdev: the underlying block device (bottom)
814 * @offset: offset to beginning of data within component device
815 * @pfx: prefix to use for warnings logged
816 *
817 * Description:
818 * This function is used by stacking drivers like MD and DM to ensure
819 * that all component devices have compatible block sizes and
820 * alignments. The stacking driver must provide a queue_limits
821 * struct (top) and then iteratively call the stacking function for
822 * all component (bottom) devices. The stacking function will
823 * attempt to combine the values and ensure proper alignment.
824 */
825void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
826 sector_t offset, const char *pfx)
827{
828 if (blk_stack_limits(t, bdev_limits(bdev),
829 get_start_sect(bdev) + offset))
830 pr_notice("%s: Warning: Device %pg is misaligned\n",
831 pfx, bdev);
832}
833EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
834
835/**
836 * queue_limits_stack_integrity - stack integrity profile
837 * @t: target queue limits
838 * @b: base queue limits
839 *
840 * Check if the integrity profile in the @b can be stacked into the
841 * target @t. Stacking is possible if either:
842 *
843 * a) does not have any integrity information stacked into it yet
844 * b) the integrity profile in @b is identical to the one in @t
845 *
846 * If @b can be stacked into @t, return %true. Else return %false and clear the
847 * integrity information in @t.
848 */
849bool queue_limits_stack_integrity(struct queue_limits *t,
850 struct queue_limits *b)
851{
852 struct blk_integrity *ti = &t->integrity;
853 struct blk_integrity *bi = &b->integrity;
854
855 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
856 return true;
857
858 if (!ti->tuple_size) {
859 /* inherit the settings from the first underlying device */
860 if (!(ti->flags & BLK_INTEGRITY_STACKED)) {
861 ti->flags = BLK_INTEGRITY_DEVICE_CAPABLE |
862 (bi->flags & BLK_INTEGRITY_REF_TAG);
863 ti->csum_type = bi->csum_type;
864 ti->tuple_size = bi->tuple_size;
865 ti->pi_offset = bi->pi_offset;
866 ti->interval_exp = bi->interval_exp;
867 ti->tag_size = bi->tag_size;
868 goto done;
869 }
870 if (!bi->tuple_size)
871 goto done;
872 }
873
874 if (ti->tuple_size != bi->tuple_size)
875 goto incompatible;
876 if (ti->interval_exp != bi->interval_exp)
877 goto incompatible;
878 if (ti->tag_size != bi->tag_size)
879 goto incompatible;
880 if (ti->csum_type != bi->csum_type)
881 goto incompatible;
882 if ((ti->flags & BLK_INTEGRITY_REF_TAG) !=
883 (bi->flags & BLK_INTEGRITY_REF_TAG))
884 goto incompatible;
885
886done:
887 ti->flags |= BLK_INTEGRITY_STACKED;
888 return true;
889
890incompatible:
891 memset(ti, 0, sizeof(*ti));
892 return false;
893}
894EXPORT_SYMBOL_GPL(queue_limits_stack_integrity);
895
896/**
897 * blk_set_queue_depth - tell the block layer about the device queue depth
898 * @q: the request queue for the device
899 * @depth: queue depth
900 *
901 */
902void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
903{
904 q->queue_depth = depth;
905 rq_qos_queue_depth_changed(q);
906}
907EXPORT_SYMBOL(blk_set_queue_depth);
908
909int bdev_alignment_offset(struct block_device *bdev)
910{
911 struct request_queue *q = bdev_get_queue(bdev);
912
913 if (q->limits.flags & BLK_FLAG_MISALIGNED)
914 return -1;
915 if (bdev_is_partition(bdev))
916 return queue_limit_alignment_offset(&q->limits,
917 bdev->bd_start_sect);
918 return q->limits.alignment_offset;
919}
920EXPORT_SYMBOL_GPL(bdev_alignment_offset);
921
922unsigned int bdev_discard_alignment(struct block_device *bdev)
923{
924 struct request_queue *q = bdev_get_queue(bdev);
925
926 if (bdev_is_partition(bdev))
927 return queue_limit_discard_alignment(&q->limits,
928 bdev->bd_start_sect);
929 return q->limits.discard_alignment;
930}
931EXPORT_SYMBOL_GPL(bdev_discard_alignment);