Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/pagemap.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/gcd.h>
13#include <linux/lcm.h>
14#include <linux/jiffies.h>
15#include <linux/gfp.h>
16#include <linux/dma-mapping.h>
17
18#include "blk.h"
19#include "blk-wbt.h"
20
21void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
22{
23 q->rq_timeout = timeout;
24}
25EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
26
27/**
28 * blk_set_default_limits - reset limits to default values
29 * @lim: the queue_limits structure to reset
30 *
31 * Description:
32 * Returns a queue_limit struct to its default state.
33 */
34void blk_set_default_limits(struct queue_limits *lim)
35{
36 lim->max_segments = BLK_MAX_SEGMENTS;
37 lim->max_discard_segments = 1;
38 lim->max_integrity_segments = 0;
39 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
40 lim->virt_boundary_mask = 0;
41 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
42 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
43 lim->max_dev_sectors = 0;
44 lim->chunk_sectors = 0;
45 lim->max_write_zeroes_sectors = 0;
46 lim->max_zone_append_sectors = 0;
47 lim->max_discard_sectors = 0;
48 lim->max_hw_discard_sectors = 0;
49 lim->max_secure_erase_sectors = 0;
50 lim->discard_granularity = 0;
51 lim->discard_alignment = 0;
52 lim->discard_misaligned = 0;
53 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
54 lim->bounce = BLK_BOUNCE_NONE;
55 lim->alignment_offset = 0;
56 lim->io_opt = 0;
57 lim->misaligned = 0;
58 lim->zoned = BLK_ZONED_NONE;
59 lim->zone_write_granularity = 0;
60 lim->dma_alignment = 511;
61}
62
63/**
64 * blk_set_stacking_limits - set default limits for stacking devices
65 * @lim: the queue_limits structure to reset
66 *
67 * Description:
68 * Returns a queue_limit struct to its default state. Should be used
69 * by stacking drivers like DM that have no internal limits.
70 */
71void blk_set_stacking_limits(struct queue_limits *lim)
72{
73 blk_set_default_limits(lim);
74
75 /* Inherit limits from component devices */
76 lim->max_segments = USHRT_MAX;
77 lim->max_discard_segments = USHRT_MAX;
78 lim->max_hw_sectors = UINT_MAX;
79 lim->max_segment_size = UINT_MAX;
80 lim->max_sectors = UINT_MAX;
81 lim->max_dev_sectors = UINT_MAX;
82 lim->max_write_zeroes_sectors = UINT_MAX;
83 lim->max_zone_append_sectors = UINT_MAX;
84}
85EXPORT_SYMBOL(blk_set_stacking_limits);
86
87/**
88 * blk_queue_bounce_limit - set bounce buffer limit for queue
89 * @q: the request queue for the device
90 * @bounce: bounce limit to enforce
91 *
92 * Description:
93 * Force bouncing for ISA DMA ranges or highmem.
94 *
95 * DEPRECATED, don't use in new code.
96 **/
97void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
98{
99 q->limits.bounce = bounce;
100}
101EXPORT_SYMBOL(blk_queue_bounce_limit);
102
103/**
104 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
105 * @q: the request queue for the device
106 * @max_hw_sectors: max hardware sectors in the usual 512b unit
107 *
108 * Description:
109 * Enables a low level driver to set a hard upper limit,
110 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
111 * the device driver based upon the capabilities of the I/O
112 * controller.
113 *
114 * max_dev_sectors is a hard limit imposed by the storage device for
115 * READ/WRITE requests. It is set by the disk driver.
116 *
117 * max_sectors is a soft limit imposed by the block layer for
118 * filesystem type requests. This value can be overridden on a
119 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
120 * The soft limit can not exceed max_hw_sectors.
121 **/
122void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
123{
124 struct queue_limits *limits = &q->limits;
125 unsigned int max_sectors;
126
127 if ((max_hw_sectors << 9) < PAGE_SIZE) {
128 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
129 printk(KERN_INFO "%s: set to minimum %d\n",
130 __func__, max_hw_sectors);
131 }
132
133 max_hw_sectors = round_down(max_hw_sectors,
134 limits->logical_block_size >> SECTOR_SHIFT);
135 limits->max_hw_sectors = max_hw_sectors;
136
137 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
138 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
139 max_sectors = round_down(max_sectors,
140 limits->logical_block_size >> SECTOR_SHIFT);
141 limits->max_sectors = max_sectors;
142
143 if (!q->disk)
144 return;
145 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
146}
147EXPORT_SYMBOL(blk_queue_max_hw_sectors);
148
149/**
150 * blk_queue_chunk_sectors - set size of the chunk for this queue
151 * @q: the request queue for the device
152 * @chunk_sectors: chunk sectors in the usual 512b unit
153 *
154 * Description:
155 * If a driver doesn't want IOs to cross a given chunk size, it can set
156 * this limit and prevent merging across chunks. Note that the block layer
157 * must accept a page worth of data at any offset. So if the crossing of
158 * chunks is a hard limitation in the driver, it must still be prepared
159 * to split single page bios.
160 **/
161void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
162{
163 q->limits.chunk_sectors = chunk_sectors;
164}
165EXPORT_SYMBOL(blk_queue_chunk_sectors);
166
167/**
168 * blk_queue_max_discard_sectors - set max sectors for a single discard
169 * @q: the request queue for the device
170 * @max_discard_sectors: maximum number of sectors to discard
171 **/
172void blk_queue_max_discard_sectors(struct request_queue *q,
173 unsigned int max_discard_sectors)
174{
175 q->limits.max_hw_discard_sectors = max_discard_sectors;
176 q->limits.max_discard_sectors = max_discard_sectors;
177}
178EXPORT_SYMBOL(blk_queue_max_discard_sectors);
179
180/**
181 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
182 * @q: the request queue for the device
183 * @max_sectors: maximum number of sectors to secure_erase
184 **/
185void blk_queue_max_secure_erase_sectors(struct request_queue *q,
186 unsigned int max_sectors)
187{
188 q->limits.max_secure_erase_sectors = max_sectors;
189}
190EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
191
192/**
193 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
194 * write zeroes
195 * @q: the request queue for the device
196 * @max_write_zeroes_sectors: maximum number of sectors to write per command
197 **/
198void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
199 unsigned int max_write_zeroes_sectors)
200{
201 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
202}
203EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
204
205/**
206 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
207 * @q: the request queue for the device
208 * @max_zone_append_sectors: maximum number of sectors to write per command
209 **/
210void blk_queue_max_zone_append_sectors(struct request_queue *q,
211 unsigned int max_zone_append_sectors)
212{
213 unsigned int max_sectors;
214
215 if (WARN_ON(!blk_queue_is_zoned(q)))
216 return;
217
218 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
219 max_sectors = min(q->limits.chunk_sectors, max_sectors);
220
221 /*
222 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
223 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
224 * or the max_hw_sectors limit not set.
225 */
226 WARN_ON(!max_sectors);
227
228 q->limits.max_zone_append_sectors = max_sectors;
229}
230EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
231
232/**
233 * blk_queue_max_segments - set max hw segments for a request for this queue
234 * @q: the request queue for the device
235 * @max_segments: max number of segments
236 *
237 * Description:
238 * Enables a low level driver to set an upper limit on the number of
239 * hw data segments in a request.
240 **/
241void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
242{
243 if (!max_segments) {
244 max_segments = 1;
245 printk(KERN_INFO "%s: set to minimum %d\n",
246 __func__, max_segments);
247 }
248
249 q->limits.max_segments = max_segments;
250}
251EXPORT_SYMBOL(blk_queue_max_segments);
252
253/**
254 * blk_queue_max_discard_segments - set max segments for discard requests
255 * @q: the request queue for the device
256 * @max_segments: max number of segments
257 *
258 * Description:
259 * Enables a low level driver to set an upper limit on the number of
260 * segments in a discard request.
261 **/
262void blk_queue_max_discard_segments(struct request_queue *q,
263 unsigned short max_segments)
264{
265 q->limits.max_discard_segments = max_segments;
266}
267EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
268
269/**
270 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
271 * @q: the request queue for the device
272 * @max_size: max size of segment in bytes
273 *
274 * Description:
275 * Enables a low level driver to set an upper limit on the size of a
276 * coalesced segment
277 **/
278void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
279{
280 if (max_size < PAGE_SIZE) {
281 max_size = PAGE_SIZE;
282 printk(KERN_INFO "%s: set to minimum %d\n",
283 __func__, max_size);
284 }
285
286 /* see blk_queue_virt_boundary() for the explanation */
287 WARN_ON_ONCE(q->limits.virt_boundary_mask);
288
289 q->limits.max_segment_size = max_size;
290}
291EXPORT_SYMBOL(blk_queue_max_segment_size);
292
293/**
294 * blk_queue_logical_block_size - set logical block size for the queue
295 * @q: the request queue for the device
296 * @size: the logical block size, in bytes
297 *
298 * Description:
299 * This should be set to the lowest possible block size that the
300 * storage device can address. The default of 512 covers most
301 * hardware.
302 **/
303void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
304{
305 struct queue_limits *limits = &q->limits;
306
307 limits->logical_block_size = size;
308
309 if (limits->physical_block_size < size)
310 limits->physical_block_size = size;
311
312 if (limits->io_min < limits->physical_block_size)
313 limits->io_min = limits->physical_block_size;
314
315 limits->max_hw_sectors =
316 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
317 limits->max_sectors =
318 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
319}
320EXPORT_SYMBOL(blk_queue_logical_block_size);
321
322/**
323 * blk_queue_physical_block_size - set physical block size for the queue
324 * @q: the request queue for the device
325 * @size: the physical block size, in bytes
326 *
327 * Description:
328 * This should be set to the lowest possible sector size that the
329 * hardware can operate on without reverting to read-modify-write
330 * operations.
331 */
332void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
333{
334 q->limits.physical_block_size = size;
335
336 if (q->limits.physical_block_size < q->limits.logical_block_size)
337 q->limits.physical_block_size = q->limits.logical_block_size;
338
339 if (q->limits.io_min < q->limits.physical_block_size)
340 q->limits.io_min = q->limits.physical_block_size;
341}
342EXPORT_SYMBOL(blk_queue_physical_block_size);
343
344/**
345 * blk_queue_zone_write_granularity - set zone write granularity for the queue
346 * @q: the request queue for the zoned device
347 * @size: the zone write granularity size, in bytes
348 *
349 * Description:
350 * This should be set to the lowest possible size allowing to write in
351 * sequential zones of a zoned block device.
352 */
353void blk_queue_zone_write_granularity(struct request_queue *q,
354 unsigned int size)
355{
356 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
357 return;
358
359 q->limits.zone_write_granularity = size;
360
361 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
362 q->limits.zone_write_granularity = q->limits.logical_block_size;
363}
364EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
365
366/**
367 * blk_queue_alignment_offset - set physical block alignment offset
368 * @q: the request queue for the device
369 * @offset: alignment offset in bytes
370 *
371 * Description:
372 * Some devices are naturally misaligned to compensate for things like
373 * the legacy DOS partition table 63-sector offset. Low-level drivers
374 * should call this function for devices whose first sector is not
375 * naturally aligned.
376 */
377void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
378{
379 q->limits.alignment_offset =
380 offset & (q->limits.physical_block_size - 1);
381 q->limits.misaligned = 0;
382}
383EXPORT_SYMBOL(blk_queue_alignment_offset);
384
385void disk_update_readahead(struct gendisk *disk)
386{
387 struct request_queue *q = disk->queue;
388
389 /*
390 * For read-ahead of large files to be effective, we need to read ahead
391 * at least twice the optimal I/O size.
392 */
393 disk->bdi->ra_pages =
394 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
395 disk->bdi->io_pages = queue_max_sectors(q) >> (PAGE_SHIFT - 9);
396}
397EXPORT_SYMBOL_GPL(disk_update_readahead);
398
399/**
400 * blk_limits_io_min - set minimum request size for a device
401 * @limits: the queue limits
402 * @min: smallest I/O size in bytes
403 *
404 * Description:
405 * Some devices have an internal block size bigger than the reported
406 * hardware sector size. This function can be used to signal the
407 * smallest I/O the device can perform without incurring a performance
408 * penalty.
409 */
410void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
411{
412 limits->io_min = min;
413
414 if (limits->io_min < limits->logical_block_size)
415 limits->io_min = limits->logical_block_size;
416
417 if (limits->io_min < limits->physical_block_size)
418 limits->io_min = limits->physical_block_size;
419}
420EXPORT_SYMBOL(blk_limits_io_min);
421
422/**
423 * blk_queue_io_min - set minimum request size for the queue
424 * @q: the request queue for the device
425 * @min: smallest I/O size in bytes
426 *
427 * Description:
428 * Storage devices may report a granularity or preferred minimum I/O
429 * size which is the smallest request the device can perform without
430 * incurring a performance penalty. For disk drives this is often the
431 * physical block size. For RAID arrays it is often the stripe chunk
432 * size. A properly aligned multiple of minimum_io_size is the
433 * preferred request size for workloads where a high number of I/O
434 * operations is desired.
435 */
436void blk_queue_io_min(struct request_queue *q, unsigned int min)
437{
438 blk_limits_io_min(&q->limits, min);
439}
440EXPORT_SYMBOL(blk_queue_io_min);
441
442/**
443 * blk_limits_io_opt - set optimal request size for a device
444 * @limits: the queue limits
445 * @opt: smallest I/O size in bytes
446 *
447 * Description:
448 * Storage devices may report an optimal I/O size, which is the
449 * device's preferred unit for sustained I/O. This is rarely reported
450 * for disk drives. For RAID arrays it is usually the stripe width or
451 * the internal track size. A properly aligned multiple of
452 * optimal_io_size is the preferred request size for workloads where
453 * sustained throughput is desired.
454 */
455void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
456{
457 limits->io_opt = opt;
458}
459EXPORT_SYMBOL(blk_limits_io_opt);
460
461/**
462 * blk_queue_io_opt - set optimal request size for the queue
463 * @q: the request queue for the device
464 * @opt: optimal request size in bytes
465 *
466 * Description:
467 * Storage devices may report an optimal I/O size, which is the
468 * device's preferred unit for sustained I/O. This is rarely reported
469 * for disk drives. For RAID arrays it is usually the stripe width or
470 * the internal track size. A properly aligned multiple of
471 * optimal_io_size is the preferred request size for workloads where
472 * sustained throughput is desired.
473 */
474void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
475{
476 blk_limits_io_opt(&q->limits, opt);
477 if (!q->disk)
478 return;
479 q->disk->bdi->ra_pages =
480 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
481}
482EXPORT_SYMBOL(blk_queue_io_opt);
483
484static int queue_limit_alignment_offset(const struct queue_limits *lim,
485 sector_t sector)
486{
487 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
488 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
489 << SECTOR_SHIFT;
490
491 return (granularity + lim->alignment_offset - alignment) % granularity;
492}
493
494static unsigned int queue_limit_discard_alignment(
495 const struct queue_limits *lim, sector_t sector)
496{
497 unsigned int alignment, granularity, offset;
498
499 if (!lim->max_discard_sectors)
500 return 0;
501
502 /* Why are these in bytes, not sectors? */
503 alignment = lim->discard_alignment >> SECTOR_SHIFT;
504 granularity = lim->discard_granularity >> SECTOR_SHIFT;
505 if (!granularity)
506 return 0;
507
508 /* Offset of the partition start in 'granularity' sectors */
509 offset = sector_div(sector, granularity);
510
511 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
512 offset = (granularity + alignment - offset) % granularity;
513
514 /* Turn it back into bytes, gaah */
515 return offset << SECTOR_SHIFT;
516}
517
518static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
519{
520 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
521 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
522 sectors = PAGE_SIZE >> SECTOR_SHIFT;
523 return sectors;
524}
525
526/**
527 * blk_stack_limits - adjust queue_limits for stacked devices
528 * @t: the stacking driver limits (top device)
529 * @b: the underlying queue limits (bottom, component device)
530 * @start: first data sector within component device
531 *
532 * Description:
533 * This function is used by stacking drivers like MD and DM to ensure
534 * that all component devices have compatible block sizes and
535 * alignments. The stacking driver must provide a queue_limits
536 * struct (top) and then iteratively call the stacking function for
537 * all component (bottom) devices. The stacking function will
538 * attempt to combine the values and ensure proper alignment.
539 *
540 * Returns 0 if the top and bottom queue_limits are compatible. The
541 * top device's block sizes and alignment offsets may be adjusted to
542 * ensure alignment with the bottom device. If no compatible sizes
543 * and alignments exist, -1 is returned and the resulting top
544 * queue_limits will have the misaligned flag set to indicate that
545 * the alignment_offset is undefined.
546 */
547int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
548 sector_t start)
549{
550 unsigned int top, bottom, alignment, ret = 0;
551
552 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
553 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
554 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
555 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
556 b->max_write_zeroes_sectors);
557 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
558 b->max_zone_append_sectors);
559 t->bounce = max(t->bounce, b->bounce);
560
561 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
562 b->seg_boundary_mask);
563 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
564 b->virt_boundary_mask);
565
566 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
567 t->max_discard_segments = min_not_zero(t->max_discard_segments,
568 b->max_discard_segments);
569 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
570 b->max_integrity_segments);
571
572 t->max_segment_size = min_not_zero(t->max_segment_size,
573 b->max_segment_size);
574
575 t->misaligned |= b->misaligned;
576
577 alignment = queue_limit_alignment_offset(b, start);
578
579 /* Bottom device has different alignment. Check that it is
580 * compatible with the current top alignment.
581 */
582 if (t->alignment_offset != alignment) {
583
584 top = max(t->physical_block_size, t->io_min)
585 + t->alignment_offset;
586 bottom = max(b->physical_block_size, b->io_min) + alignment;
587
588 /* Verify that top and bottom intervals line up */
589 if (max(top, bottom) % min(top, bottom)) {
590 t->misaligned = 1;
591 ret = -1;
592 }
593 }
594
595 t->logical_block_size = max(t->logical_block_size,
596 b->logical_block_size);
597
598 t->physical_block_size = max(t->physical_block_size,
599 b->physical_block_size);
600
601 t->io_min = max(t->io_min, b->io_min);
602 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
603 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
604
605 /* Set non-power-of-2 compatible chunk_sectors boundary */
606 if (b->chunk_sectors)
607 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
608
609 /* Physical block size a multiple of the logical block size? */
610 if (t->physical_block_size & (t->logical_block_size - 1)) {
611 t->physical_block_size = t->logical_block_size;
612 t->misaligned = 1;
613 ret = -1;
614 }
615
616 /* Minimum I/O a multiple of the physical block size? */
617 if (t->io_min & (t->physical_block_size - 1)) {
618 t->io_min = t->physical_block_size;
619 t->misaligned = 1;
620 ret = -1;
621 }
622
623 /* Optimal I/O a multiple of the physical block size? */
624 if (t->io_opt & (t->physical_block_size - 1)) {
625 t->io_opt = 0;
626 t->misaligned = 1;
627 ret = -1;
628 }
629
630 /* chunk_sectors a multiple of the physical block size? */
631 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
632 t->chunk_sectors = 0;
633 t->misaligned = 1;
634 ret = -1;
635 }
636
637 t->raid_partial_stripes_expensive =
638 max(t->raid_partial_stripes_expensive,
639 b->raid_partial_stripes_expensive);
640
641 /* Find lowest common alignment_offset */
642 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
643 % max(t->physical_block_size, t->io_min);
644
645 /* Verify that new alignment_offset is on a logical block boundary */
646 if (t->alignment_offset & (t->logical_block_size - 1)) {
647 t->misaligned = 1;
648 ret = -1;
649 }
650
651 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
652 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
653 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
654
655 /* Discard alignment and granularity */
656 if (b->discard_granularity) {
657 alignment = queue_limit_discard_alignment(b, start);
658
659 if (t->discard_granularity != 0 &&
660 t->discard_alignment != alignment) {
661 top = t->discard_granularity + t->discard_alignment;
662 bottom = b->discard_granularity + alignment;
663
664 /* Verify that top and bottom intervals line up */
665 if ((max(top, bottom) % min(top, bottom)) != 0)
666 t->discard_misaligned = 1;
667 }
668
669 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
670 b->max_discard_sectors);
671 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
672 b->max_hw_discard_sectors);
673 t->discard_granularity = max(t->discard_granularity,
674 b->discard_granularity);
675 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
676 t->discard_granularity;
677 }
678 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
679 b->max_secure_erase_sectors);
680 t->zone_write_granularity = max(t->zone_write_granularity,
681 b->zone_write_granularity);
682 t->zoned = max(t->zoned, b->zoned);
683 return ret;
684}
685EXPORT_SYMBOL(blk_stack_limits);
686
687/**
688 * disk_stack_limits - adjust queue limits for stacked drivers
689 * @disk: MD/DM gendisk (top)
690 * @bdev: the underlying block device (bottom)
691 * @offset: offset to beginning of data within component device
692 *
693 * Description:
694 * Merges the limits for a top level gendisk and a bottom level
695 * block_device.
696 */
697void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
698 sector_t offset)
699{
700 struct request_queue *t = disk->queue;
701
702 if (blk_stack_limits(&t->limits, &bdev_get_queue(bdev)->limits,
703 get_start_sect(bdev) + (offset >> 9)) < 0)
704 pr_notice("%s: Warning: Device %pg is misaligned\n",
705 disk->disk_name, bdev);
706
707 disk_update_readahead(disk);
708}
709EXPORT_SYMBOL(disk_stack_limits);
710
711/**
712 * blk_queue_update_dma_pad - update pad mask
713 * @q: the request queue for the device
714 * @mask: pad mask
715 *
716 * Update dma pad mask.
717 *
718 * Appending pad buffer to a request modifies the last entry of a
719 * scatter list such that it includes the pad buffer.
720 **/
721void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
722{
723 if (mask > q->dma_pad_mask)
724 q->dma_pad_mask = mask;
725}
726EXPORT_SYMBOL(blk_queue_update_dma_pad);
727
728/**
729 * blk_queue_segment_boundary - set boundary rules for segment merging
730 * @q: the request queue for the device
731 * @mask: the memory boundary mask
732 **/
733void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
734{
735 if (mask < PAGE_SIZE - 1) {
736 mask = PAGE_SIZE - 1;
737 printk(KERN_INFO "%s: set to minimum %lx\n",
738 __func__, mask);
739 }
740
741 q->limits.seg_boundary_mask = mask;
742}
743EXPORT_SYMBOL(blk_queue_segment_boundary);
744
745/**
746 * blk_queue_virt_boundary - set boundary rules for bio merging
747 * @q: the request queue for the device
748 * @mask: the memory boundary mask
749 **/
750void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
751{
752 q->limits.virt_boundary_mask = mask;
753
754 /*
755 * Devices that require a virtual boundary do not support scatter/gather
756 * I/O natively, but instead require a descriptor list entry for each
757 * page (which might not be idential to the Linux PAGE_SIZE). Because
758 * of that they are not limited by our notion of "segment size".
759 */
760 if (mask)
761 q->limits.max_segment_size = UINT_MAX;
762}
763EXPORT_SYMBOL(blk_queue_virt_boundary);
764
765/**
766 * blk_queue_dma_alignment - set dma length and memory alignment
767 * @q: the request queue for the device
768 * @mask: alignment mask
769 *
770 * description:
771 * set required memory and length alignment for direct dma transactions.
772 * this is used when building direct io requests for the queue.
773 *
774 **/
775void blk_queue_dma_alignment(struct request_queue *q, int mask)
776{
777 q->limits.dma_alignment = mask;
778}
779EXPORT_SYMBOL(blk_queue_dma_alignment);
780
781/**
782 * blk_queue_update_dma_alignment - update dma length and memory alignment
783 * @q: the request queue for the device
784 * @mask: alignment mask
785 *
786 * description:
787 * update required memory and length alignment for direct dma transactions.
788 * If the requested alignment is larger than the current alignment, then
789 * the current queue alignment is updated to the new value, otherwise it
790 * is left alone. The design of this is to allow multiple objects
791 * (driver, device, transport etc) to set their respective
792 * alignments without having them interfere.
793 *
794 **/
795void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
796{
797 BUG_ON(mask > PAGE_SIZE);
798
799 if (mask > q->limits.dma_alignment)
800 q->limits.dma_alignment = mask;
801}
802EXPORT_SYMBOL(blk_queue_update_dma_alignment);
803
804/**
805 * blk_set_queue_depth - tell the block layer about the device queue depth
806 * @q: the request queue for the device
807 * @depth: queue depth
808 *
809 */
810void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
811{
812 q->queue_depth = depth;
813 rq_qos_queue_depth_changed(q);
814}
815EXPORT_SYMBOL(blk_set_queue_depth);
816
817/**
818 * blk_queue_write_cache - configure queue's write cache
819 * @q: the request queue for the device
820 * @wc: write back cache on or off
821 * @fua: device supports FUA writes, if true
822 *
823 * Tell the block layer about the write cache of @q.
824 */
825void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
826{
827 if (wc)
828 blk_queue_flag_set(QUEUE_FLAG_WC, q);
829 else
830 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
831 if (fua)
832 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
833 else
834 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
835
836 wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
837}
838EXPORT_SYMBOL_GPL(blk_queue_write_cache);
839
840/**
841 * blk_queue_required_elevator_features - Set a queue required elevator features
842 * @q: the request queue for the target device
843 * @features: Required elevator features OR'ed together
844 *
845 * Tell the block layer that for the device controlled through @q, only the
846 * only elevators that can be used are those that implement at least the set of
847 * features specified by @features.
848 */
849void blk_queue_required_elevator_features(struct request_queue *q,
850 unsigned int features)
851{
852 q->required_elevator_features = features;
853}
854EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
855
856/**
857 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
858 * @q: the request queue for the device
859 * @dev: the device pointer for dma
860 *
861 * Tell the block layer about merging the segments by dma map of @q.
862 */
863bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
864 struct device *dev)
865{
866 unsigned long boundary = dma_get_merge_boundary(dev);
867
868 if (!boundary)
869 return false;
870
871 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
872 blk_queue_virt_boundary(q, boundary);
873
874 return true;
875}
876EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
877
878static bool disk_has_partitions(struct gendisk *disk)
879{
880 unsigned long idx;
881 struct block_device *part;
882 bool ret = false;
883
884 rcu_read_lock();
885 xa_for_each(&disk->part_tbl, idx, part) {
886 if (bdev_is_partition(part)) {
887 ret = true;
888 break;
889 }
890 }
891 rcu_read_unlock();
892
893 return ret;
894}
895
896/**
897 * disk_set_zoned - configure the zoned model for a disk
898 * @disk: the gendisk of the queue to configure
899 * @model: the zoned model to set
900 *
901 * Set the zoned model of @disk to @model.
902 *
903 * When @model is BLK_ZONED_HM (host managed), this should be called only
904 * if zoned block device support is enabled (CONFIG_BLK_DEV_ZONED option).
905 * If @model specifies BLK_ZONED_HA (host aware), the effective model used
906 * depends on CONFIG_BLK_DEV_ZONED settings and on the existence of partitions
907 * on the disk.
908 */
909void disk_set_zoned(struct gendisk *disk, enum blk_zoned_model model)
910{
911 struct request_queue *q = disk->queue;
912
913 switch (model) {
914 case BLK_ZONED_HM:
915 /*
916 * Host managed devices are supported only if
917 * CONFIG_BLK_DEV_ZONED is enabled.
918 */
919 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
920 break;
921 case BLK_ZONED_HA:
922 /*
923 * Host aware devices can be treated either as regular block
924 * devices (similar to drive managed devices) or as zoned block
925 * devices to take advantage of the zone command set, similarly
926 * to host managed devices. We try the latter if there are no
927 * partitions and zoned block device support is enabled, else
928 * we do nothing special as far as the block layer is concerned.
929 */
930 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) ||
931 disk_has_partitions(disk))
932 model = BLK_ZONED_NONE;
933 break;
934 case BLK_ZONED_NONE:
935 default:
936 if (WARN_ON_ONCE(model != BLK_ZONED_NONE))
937 model = BLK_ZONED_NONE;
938 break;
939 }
940
941 q->limits.zoned = model;
942 if (model != BLK_ZONED_NONE) {
943 /*
944 * Set the zone write granularity to the device logical block
945 * size by default. The driver can change this value if needed.
946 */
947 blk_queue_zone_write_granularity(q,
948 queue_logical_block_size(q));
949 } else {
950 disk_clear_zone_settings(disk);
951 }
952}
953EXPORT_SYMBOL_GPL(disk_set_zoned);
954
955int bdev_alignment_offset(struct block_device *bdev)
956{
957 struct request_queue *q = bdev_get_queue(bdev);
958
959 if (q->limits.misaligned)
960 return -1;
961 if (bdev_is_partition(bdev))
962 return queue_limit_alignment_offset(&q->limits,
963 bdev->bd_start_sect);
964 return q->limits.alignment_offset;
965}
966EXPORT_SYMBOL_GPL(bdev_alignment_offset);
967
968unsigned int bdev_discard_alignment(struct block_device *bdev)
969{
970 struct request_queue *q = bdev_get_queue(bdev);
971
972 if (bdev_is_partition(bdev))
973 return queue_limit_discard_alignment(&q->limits,
974 bdev->bd_start_sect);
975 return q->limits.discard_alignment;
976}
977EXPORT_SYMBOL_GPL(bdev_discard_alignment);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/memblock.h> /* for max_pfn/max_low_pfn */
11#include <linux/gcd.h>
12#include <linux/lcm.h>
13#include <linux/jiffies.h>
14#include <linux/gfp.h>
15#include <linux/dma-mapping.h>
16
17#include "blk.h"
18#include "blk-wbt.h"
19
20unsigned long blk_max_low_pfn;
21EXPORT_SYMBOL(blk_max_low_pfn);
22
23unsigned long blk_max_pfn;
24
25void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
26{
27 q->rq_timeout = timeout;
28}
29EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
30
31/**
32 * blk_set_default_limits - reset limits to default values
33 * @lim: the queue_limits structure to reset
34 *
35 * Description:
36 * Returns a queue_limit struct to its default state.
37 */
38void blk_set_default_limits(struct queue_limits *lim)
39{
40 lim->max_segments = BLK_MAX_SEGMENTS;
41 lim->max_discard_segments = 1;
42 lim->max_integrity_segments = 0;
43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
44 lim->virt_boundary_mask = 0;
45 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
46 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
47 lim->max_dev_sectors = 0;
48 lim->chunk_sectors = 0;
49 lim->max_write_same_sectors = 0;
50 lim->max_write_zeroes_sectors = 0;
51 lim->max_discard_sectors = 0;
52 lim->max_hw_discard_sectors = 0;
53 lim->discard_granularity = 0;
54 lim->discard_alignment = 0;
55 lim->discard_misaligned = 0;
56 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
57 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
58 lim->alignment_offset = 0;
59 lim->io_opt = 0;
60 lim->misaligned = 0;
61 lim->zoned = BLK_ZONED_NONE;
62}
63EXPORT_SYMBOL(blk_set_default_limits);
64
65/**
66 * blk_set_stacking_limits - set default limits for stacking devices
67 * @lim: the queue_limits structure to reset
68 *
69 * Description:
70 * Returns a queue_limit struct to its default state. Should be used
71 * by stacking drivers like DM that have no internal limits.
72 */
73void blk_set_stacking_limits(struct queue_limits *lim)
74{
75 blk_set_default_limits(lim);
76
77 /* Inherit limits from component devices */
78 lim->max_segments = USHRT_MAX;
79 lim->max_discard_segments = USHRT_MAX;
80 lim->max_hw_sectors = UINT_MAX;
81 lim->max_segment_size = UINT_MAX;
82 lim->max_sectors = UINT_MAX;
83 lim->max_dev_sectors = UINT_MAX;
84 lim->max_write_same_sectors = UINT_MAX;
85 lim->max_write_zeroes_sectors = UINT_MAX;
86}
87EXPORT_SYMBOL(blk_set_stacking_limits);
88
89/**
90 * blk_queue_make_request - define an alternate make_request function for a device
91 * @q: the request queue for the device to be affected
92 * @mfn: the alternate make_request function
93 *
94 * Description:
95 * The normal way for &struct bios to be passed to a device
96 * driver is for them to be collected into requests on a request
97 * queue, and then to allow the device driver to select requests
98 * off that queue when it is ready. This works well for many block
99 * devices. However some block devices (typically virtual devices
100 * such as md or lvm) do not benefit from the processing on the
101 * request queue, and are served best by having the requests passed
102 * directly to them. This can be achieved by providing a function
103 * to blk_queue_make_request().
104 *
105 * Caveat:
106 * The driver that does this *must* be able to deal appropriately
107 * with buffers in "highmemory". This can be accomplished by either calling
108 * kmap_atomic() to get a temporary kernel mapping, or by calling
109 * blk_queue_bounce() to create a buffer in normal memory.
110 **/
111void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
112{
113 /*
114 * set defaults
115 */
116 q->nr_requests = BLKDEV_MAX_RQ;
117
118 q->make_request_fn = mfn;
119 blk_queue_dma_alignment(q, 511);
120
121 blk_set_default_limits(&q->limits);
122}
123EXPORT_SYMBOL(blk_queue_make_request);
124
125/**
126 * blk_queue_bounce_limit - set bounce buffer limit for queue
127 * @q: the request queue for the device
128 * @max_addr: the maximum address the device can handle
129 *
130 * Description:
131 * Different hardware can have different requirements as to what pages
132 * it can do I/O directly to. A low level driver can call
133 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
134 * buffers for doing I/O to pages residing above @max_addr.
135 **/
136void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
137{
138 unsigned long b_pfn = max_addr >> PAGE_SHIFT;
139 int dma = 0;
140
141 q->bounce_gfp = GFP_NOIO;
142#if BITS_PER_LONG == 64
143 /*
144 * Assume anything <= 4GB can be handled by IOMMU. Actually
145 * some IOMMUs can handle everything, but I don't know of a
146 * way to test this here.
147 */
148 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
149 dma = 1;
150 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
151#else
152 if (b_pfn < blk_max_low_pfn)
153 dma = 1;
154 q->limits.bounce_pfn = b_pfn;
155#endif
156 if (dma) {
157 init_emergency_isa_pool();
158 q->bounce_gfp = GFP_NOIO | GFP_DMA;
159 q->limits.bounce_pfn = b_pfn;
160 }
161}
162EXPORT_SYMBOL(blk_queue_bounce_limit);
163
164/**
165 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
166 * @q: the request queue for the device
167 * @max_hw_sectors: max hardware sectors in the usual 512b unit
168 *
169 * Description:
170 * Enables a low level driver to set a hard upper limit,
171 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
172 * the device driver based upon the capabilities of the I/O
173 * controller.
174 *
175 * max_dev_sectors is a hard limit imposed by the storage device for
176 * READ/WRITE requests. It is set by the disk driver.
177 *
178 * max_sectors is a soft limit imposed by the block layer for
179 * filesystem type requests. This value can be overridden on a
180 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
181 * The soft limit can not exceed max_hw_sectors.
182 **/
183void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
184{
185 struct queue_limits *limits = &q->limits;
186 unsigned int max_sectors;
187
188 if ((max_hw_sectors << 9) < PAGE_SIZE) {
189 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
190 printk(KERN_INFO "%s: set to minimum %d\n",
191 __func__, max_hw_sectors);
192 }
193
194 limits->max_hw_sectors = max_hw_sectors;
195 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
196 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
197 limits->max_sectors = max_sectors;
198 q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
199}
200EXPORT_SYMBOL(blk_queue_max_hw_sectors);
201
202/**
203 * blk_queue_chunk_sectors - set size of the chunk for this queue
204 * @q: the request queue for the device
205 * @chunk_sectors: chunk sectors in the usual 512b unit
206 *
207 * Description:
208 * If a driver doesn't want IOs to cross a given chunk size, it can set
209 * this limit and prevent merging across chunks. Note that the chunk size
210 * must currently be a power-of-2 in sectors. Also note that the block
211 * layer must accept a page worth of data at any offset. So if the
212 * crossing of chunks is a hard limitation in the driver, it must still be
213 * prepared to split single page bios.
214 **/
215void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
216{
217 BUG_ON(!is_power_of_2(chunk_sectors));
218 q->limits.chunk_sectors = chunk_sectors;
219}
220EXPORT_SYMBOL(blk_queue_chunk_sectors);
221
222/**
223 * blk_queue_max_discard_sectors - set max sectors for a single discard
224 * @q: the request queue for the device
225 * @max_discard_sectors: maximum number of sectors to discard
226 **/
227void blk_queue_max_discard_sectors(struct request_queue *q,
228 unsigned int max_discard_sectors)
229{
230 q->limits.max_hw_discard_sectors = max_discard_sectors;
231 q->limits.max_discard_sectors = max_discard_sectors;
232}
233EXPORT_SYMBOL(blk_queue_max_discard_sectors);
234
235/**
236 * blk_queue_max_write_same_sectors - set max sectors for a single write same
237 * @q: the request queue for the device
238 * @max_write_same_sectors: maximum number of sectors to write per command
239 **/
240void blk_queue_max_write_same_sectors(struct request_queue *q,
241 unsigned int max_write_same_sectors)
242{
243 q->limits.max_write_same_sectors = max_write_same_sectors;
244}
245EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
246
247/**
248 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
249 * write zeroes
250 * @q: the request queue for the device
251 * @max_write_zeroes_sectors: maximum number of sectors to write per command
252 **/
253void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
254 unsigned int max_write_zeroes_sectors)
255{
256 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
257}
258EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
259
260/**
261 * blk_queue_max_segments - set max hw segments for a request for this queue
262 * @q: the request queue for the device
263 * @max_segments: max number of segments
264 *
265 * Description:
266 * Enables a low level driver to set an upper limit on the number of
267 * hw data segments in a request.
268 **/
269void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
270{
271 if (!max_segments) {
272 max_segments = 1;
273 printk(KERN_INFO "%s: set to minimum %d\n",
274 __func__, max_segments);
275 }
276
277 q->limits.max_segments = max_segments;
278}
279EXPORT_SYMBOL(blk_queue_max_segments);
280
281/**
282 * blk_queue_max_discard_segments - set max segments for discard requests
283 * @q: the request queue for the device
284 * @max_segments: max number of segments
285 *
286 * Description:
287 * Enables a low level driver to set an upper limit on the number of
288 * segments in a discard request.
289 **/
290void blk_queue_max_discard_segments(struct request_queue *q,
291 unsigned short max_segments)
292{
293 q->limits.max_discard_segments = max_segments;
294}
295EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
296
297/**
298 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
299 * @q: the request queue for the device
300 * @max_size: max size of segment in bytes
301 *
302 * Description:
303 * Enables a low level driver to set an upper limit on the size of a
304 * coalesced segment
305 **/
306void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
307{
308 if (max_size < PAGE_SIZE) {
309 max_size = PAGE_SIZE;
310 printk(KERN_INFO "%s: set to minimum %d\n",
311 __func__, max_size);
312 }
313
314 /* see blk_queue_virt_boundary() for the explanation */
315 WARN_ON_ONCE(q->limits.virt_boundary_mask);
316
317 q->limits.max_segment_size = max_size;
318}
319EXPORT_SYMBOL(blk_queue_max_segment_size);
320
321/**
322 * blk_queue_logical_block_size - set logical block size for the queue
323 * @q: the request queue for the device
324 * @size: the logical block size, in bytes
325 *
326 * Description:
327 * This should be set to the lowest possible block size that the
328 * storage device can address. The default of 512 covers most
329 * hardware.
330 **/
331void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
332{
333 q->limits.logical_block_size = size;
334
335 if (q->limits.physical_block_size < size)
336 q->limits.physical_block_size = size;
337
338 if (q->limits.io_min < q->limits.physical_block_size)
339 q->limits.io_min = q->limits.physical_block_size;
340}
341EXPORT_SYMBOL(blk_queue_logical_block_size);
342
343/**
344 * blk_queue_physical_block_size - set physical block size for the queue
345 * @q: the request queue for the device
346 * @size: the physical block size, in bytes
347 *
348 * Description:
349 * This should be set to the lowest possible sector size that the
350 * hardware can operate on without reverting to read-modify-write
351 * operations.
352 */
353void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
354{
355 q->limits.physical_block_size = size;
356
357 if (q->limits.physical_block_size < q->limits.logical_block_size)
358 q->limits.physical_block_size = q->limits.logical_block_size;
359
360 if (q->limits.io_min < q->limits.physical_block_size)
361 q->limits.io_min = q->limits.physical_block_size;
362}
363EXPORT_SYMBOL(blk_queue_physical_block_size);
364
365/**
366 * blk_queue_alignment_offset - set physical block alignment offset
367 * @q: the request queue for the device
368 * @offset: alignment offset in bytes
369 *
370 * Description:
371 * Some devices are naturally misaligned to compensate for things like
372 * the legacy DOS partition table 63-sector offset. Low-level drivers
373 * should call this function for devices whose first sector is not
374 * naturally aligned.
375 */
376void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
377{
378 q->limits.alignment_offset =
379 offset & (q->limits.physical_block_size - 1);
380 q->limits.misaligned = 0;
381}
382EXPORT_SYMBOL(blk_queue_alignment_offset);
383
384/**
385 * blk_limits_io_min - set minimum request size for a device
386 * @limits: the queue limits
387 * @min: smallest I/O size in bytes
388 *
389 * Description:
390 * Some devices have an internal block size bigger than the reported
391 * hardware sector size. This function can be used to signal the
392 * smallest I/O the device can perform without incurring a performance
393 * penalty.
394 */
395void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
396{
397 limits->io_min = min;
398
399 if (limits->io_min < limits->logical_block_size)
400 limits->io_min = limits->logical_block_size;
401
402 if (limits->io_min < limits->physical_block_size)
403 limits->io_min = limits->physical_block_size;
404}
405EXPORT_SYMBOL(blk_limits_io_min);
406
407/**
408 * blk_queue_io_min - set minimum request size for the queue
409 * @q: the request queue for the device
410 * @min: smallest I/O size in bytes
411 *
412 * Description:
413 * Storage devices may report a granularity or preferred minimum I/O
414 * size which is the smallest request the device can perform without
415 * incurring a performance penalty. For disk drives this is often the
416 * physical block size. For RAID arrays it is often the stripe chunk
417 * size. A properly aligned multiple of minimum_io_size is the
418 * preferred request size for workloads where a high number of I/O
419 * operations is desired.
420 */
421void blk_queue_io_min(struct request_queue *q, unsigned int min)
422{
423 blk_limits_io_min(&q->limits, min);
424}
425EXPORT_SYMBOL(blk_queue_io_min);
426
427/**
428 * blk_limits_io_opt - set optimal request size for a device
429 * @limits: the queue limits
430 * @opt: smallest I/O size in bytes
431 *
432 * Description:
433 * Storage devices may report an optimal I/O size, which is the
434 * device's preferred unit for sustained I/O. This is rarely reported
435 * for disk drives. For RAID arrays it is usually the stripe width or
436 * the internal track size. A properly aligned multiple of
437 * optimal_io_size is the preferred request size for workloads where
438 * sustained throughput is desired.
439 */
440void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
441{
442 limits->io_opt = opt;
443}
444EXPORT_SYMBOL(blk_limits_io_opt);
445
446/**
447 * blk_queue_io_opt - set optimal request size for the queue
448 * @q: the request queue for the device
449 * @opt: optimal request size in bytes
450 *
451 * Description:
452 * Storage devices may report an optimal I/O size, which is the
453 * device's preferred unit for sustained I/O. This is rarely reported
454 * for disk drives. For RAID arrays it is usually the stripe width or
455 * the internal track size. A properly aligned multiple of
456 * optimal_io_size is the preferred request size for workloads where
457 * sustained throughput is desired.
458 */
459void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
460{
461 blk_limits_io_opt(&q->limits, opt);
462}
463EXPORT_SYMBOL(blk_queue_io_opt);
464
465/**
466 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
467 * @t: the stacking driver (top)
468 * @b: the underlying device (bottom)
469 **/
470void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
471{
472 blk_stack_limits(&t->limits, &b->limits, 0);
473}
474EXPORT_SYMBOL(blk_queue_stack_limits);
475
476/**
477 * blk_stack_limits - adjust queue_limits for stacked devices
478 * @t: the stacking driver limits (top device)
479 * @b: the underlying queue limits (bottom, component device)
480 * @start: first data sector within component device
481 *
482 * Description:
483 * This function is used by stacking drivers like MD and DM to ensure
484 * that all component devices have compatible block sizes and
485 * alignments. The stacking driver must provide a queue_limits
486 * struct (top) and then iteratively call the stacking function for
487 * all component (bottom) devices. The stacking function will
488 * attempt to combine the values and ensure proper alignment.
489 *
490 * Returns 0 if the top and bottom queue_limits are compatible. The
491 * top device's block sizes and alignment offsets may be adjusted to
492 * ensure alignment with the bottom device. If no compatible sizes
493 * and alignments exist, -1 is returned and the resulting top
494 * queue_limits will have the misaligned flag set to indicate that
495 * the alignment_offset is undefined.
496 */
497int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
498 sector_t start)
499{
500 unsigned int top, bottom, alignment, ret = 0;
501
502 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
503 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
504 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
505 t->max_write_same_sectors = min(t->max_write_same_sectors,
506 b->max_write_same_sectors);
507 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
508 b->max_write_zeroes_sectors);
509 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
510
511 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
512 b->seg_boundary_mask);
513 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
514 b->virt_boundary_mask);
515
516 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
517 t->max_discard_segments = min_not_zero(t->max_discard_segments,
518 b->max_discard_segments);
519 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
520 b->max_integrity_segments);
521
522 t->max_segment_size = min_not_zero(t->max_segment_size,
523 b->max_segment_size);
524
525 t->misaligned |= b->misaligned;
526
527 alignment = queue_limit_alignment_offset(b, start);
528
529 /* Bottom device has different alignment. Check that it is
530 * compatible with the current top alignment.
531 */
532 if (t->alignment_offset != alignment) {
533
534 top = max(t->physical_block_size, t->io_min)
535 + t->alignment_offset;
536 bottom = max(b->physical_block_size, b->io_min) + alignment;
537
538 /* Verify that top and bottom intervals line up */
539 if (max(top, bottom) % min(top, bottom)) {
540 t->misaligned = 1;
541 ret = -1;
542 }
543 }
544
545 t->logical_block_size = max(t->logical_block_size,
546 b->logical_block_size);
547
548 t->physical_block_size = max(t->physical_block_size,
549 b->physical_block_size);
550
551 t->io_min = max(t->io_min, b->io_min);
552 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
553
554 /* Physical block size a multiple of the logical block size? */
555 if (t->physical_block_size & (t->logical_block_size - 1)) {
556 t->physical_block_size = t->logical_block_size;
557 t->misaligned = 1;
558 ret = -1;
559 }
560
561 /* Minimum I/O a multiple of the physical block size? */
562 if (t->io_min & (t->physical_block_size - 1)) {
563 t->io_min = t->physical_block_size;
564 t->misaligned = 1;
565 ret = -1;
566 }
567
568 /* Optimal I/O a multiple of the physical block size? */
569 if (t->io_opt & (t->physical_block_size - 1)) {
570 t->io_opt = 0;
571 t->misaligned = 1;
572 ret = -1;
573 }
574
575 t->raid_partial_stripes_expensive =
576 max(t->raid_partial_stripes_expensive,
577 b->raid_partial_stripes_expensive);
578
579 /* Find lowest common alignment_offset */
580 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
581 % max(t->physical_block_size, t->io_min);
582
583 /* Verify that new alignment_offset is on a logical block boundary */
584 if (t->alignment_offset & (t->logical_block_size - 1)) {
585 t->misaligned = 1;
586 ret = -1;
587 }
588
589 /* Discard alignment and granularity */
590 if (b->discard_granularity) {
591 alignment = queue_limit_discard_alignment(b, start);
592
593 if (t->discard_granularity != 0 &&
594 t->discard_alignment != alignment) {
595 top = t->discard_granularity + t->discard_alignment;
596 bottom = b->discard_granularity + alignment;
597
598 /* Verify that top and bottom intervals line up */
599 if ((max(top, bottom) % min(top, bottom)) != 0)
600 t->discard_misaligned = 1;
601 }
602
603 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
604 b->max_discard_sectors);
605 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
606 b->max_hw_discard_sectors);
607 t->discard_granularity = max(t->discard_granularity,
608 b->discard_granularity);
609 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
610 t->discard_granularity;
611 }
612
613 if (b->chunk_sectors)
614 t->chunk_sectors = min_not_zero(t->chunk_sectors,
615 b->chunk_sectors);
616
617 return ret;
618}
619EXPORT_SYMBOL(blk_stack_limits);
620
621/**
622 * bdev_stack_limits - adjust queue limits for stacked drivers
623 * @t: the stacking driver limits (top device)
624 * @bdev: the component block_device (bottom)
625 * @start: first data sector within component device
626 *
627 * Description:
628 * Merges queue limits for a top device and a block_device. Returns
629 * 0 if alignment didn't change. Returns -1 if adding the bottom
630 * device caused misalignment.
631 */
632int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
633 sector_t start)
634{
635 struct request_queue *bq = bdev_get_queue(bdev);
636
637 start += get_start_sect(bdev);
638
639 return blk_stack_limits(t, &bq->limits, start);
640}
641EXPORT_SYMBOL(bdev_stack_limits);
642
643/**
644 * disk_stack_limits - adjust queue limits for stacked drivers
645 * @disk: MD/DM gendisk (top)
646 * @bdev: the underlying block device (bottom)
647 * @offset: offset to beginning of data within component device
648 *
649 * Description:
650 * Merges the limits for a top level gendisk and a bottom level
651 * block_device.
652 */
653void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
654 sector_t offset)
655{
656 struct request_queue *t = disk->queue;
657
658 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
659 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
660
661 disk_name(disk, 0, top);
662 bdevname(bdev, bottom);
663
664 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
665 top, bottom);
666 }
667}
668EXPORT_SYMBOL(disk_stack_limits);
669
670/**
671 * blk_queue_update_dma_pad - update pad mask
672 * @q: the request queue for the device
673 * @mask: pad mask
674 *
675 * Update dma pad mask.
676 *
677 * Appending pad buffer to a request modifies the last entry of a
678 * scatter list such that it includes the pad buffer.
679 **/
680void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
681{
682 if (mask > q->dma_pad_mask)
683 q->dma_pad_mask = mask;
684}
685EXPORT_SYMBOL(blk_queue_update_dma_pad);
686
687/**
688 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
689 * @q: the request queue for the device
690 * @dma_drain_needed: fn which returns non-zero if drain is necessary
691 * @buf: physically contiguous buffer
692 * @size: size of the buffer in bytes
693 *
694 * Some devices have excess DMA problems and can't simply discard (or
695 * zero fill) the unwanted piece of the transfer. They have to have a
696 * real area of memory to transfer it into. The use case for this is
697 * ATAPI devices in DMA mode. If the packet command causes a transfer
698 * bigger than the transfer size some HBAs will lock up if there
699 * aren't DMA elements to contain the excess transfer. What this API
700 * does is adjust the queue so that the buf is always appended
701 * silently to the scatterlist.
702 *
703 * Note: This routine adjusts max_hw_segments to make room for appending
704 * the drain buffer. If you call blk_queue_max_segments() after calling
705 * this routine, you must set the limit to one fewer than your device
706 * can support otherwise there won't be room for the drain buffer.
707 */
708int blk_queue_dma_drain(struct request_queue *q,
709 dma_drain_needed_fn *dma_drain_needed,
710 void *buf, unsigned int size)
711{
712 if (queue_max_segments(q) < 2)
713 return -EINVAL;
714 /* make room for appending the drain */
715 blk_queue_max_segments(q, queue_max_segments(q) - 1);
716 q->dma_drain_needed = dma_drain_needed;
717 q->dma_drain_buffer = buf;
718 q->dma_drain_size = size;
719
720 return 0;
721}
722EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
723
724/**
725 * blk_queue_segment_boundary - set boundary rules for segment merging
726 * @q: the request queue for the device
727 * @mask: the memory boundary mask
728 **/
729void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
730{
731 if (mask < PAGE_SIZE - 1) {
732 mask = PAGE_SIZE - 1;
733 printk(KERN_INFO "%s: set to minimum %lx\n",
734 __func__, mask);
735 }
736
737 q->limits.seg_boundary_mask = mask;
738}
739EXPORT_SYMBOL(blk_queue_segment_boundary);
740
741/**
742 * blk_queue_virt_boundary - set boundary rules for bio merging
743 * @q: the request queue for the device
744 * @mask: the memory boundary mask
745 **/
746void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
747{
748 q->limits.virt_boundary_mask = mask;
749
750 /*
751 * Devices that require a virtual boundary do not support scatter/gather
752 * I/O natively, but instead require a descriptor list entry for each
753 * page (which might not be idential to the Linux PAGE_SIZE). Because
754 * of that they are not limited by our notion of "segment size".
755 */
756 if (mask)
757 q->limits.max_segment_size = UINT_MAX;
758}
759EXPORT_SYMBOL(blk_queue_virt_boundary);
760
761/**
762 * blk_queue_dma_alignment - set dma length and memory alignment
763 * @q: the request queue for the device
764 * @mask: alignment mask
765 *
766 * description:
767 * set required memory and length alignment for direct dma transactions.
768 * this is used when building direct io requests for the queue.
769 *
770 **/
771void blk_queue_dma_alignment(struct request_queue *q, int mask)
772{
773 q->dma_alignment = mask;
774}
775EXPORT_SYMBOL(blk_queue_dma_alignment);
776
777/**
778 * blk_queue_update_dma_alignment - update dma length and memory alignment
779 * @q: the request queue for the device
780 * @mask: alignment mask
781 *
782 * description:
783 * update required memory and length alignment for direct dma transactions.
784 * If the requested alignment is larger than the current alignment, then
785 * the current queue alignment is updated to the new value, otherwise it
786 * is left alone. The design of this is to allow multiple objects
787 * (driver, device, transport etc) to set their respective
788 * alignments without having them interfere.
789 *
790 **/
791void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
792{
793 BUG_ON(mask > PAGE_SIZE);
794
795 if (mask > q->dma_alignment)
796 q->dma_alignment = mask;
797}
798EXPORT_SYMBOL(blk_queue_update_dma_alignment);
799
800/**
801 * blk_set_queue_depth - tell the block layer about the device queue depth
802 * @q: the request queue for the device
803 * @depth: queue depth
804 *
805 */
806void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
807{
808 q->queue_depth = depth;
809 rq_qos_queue_depth_changed(q);
810}
811EXPORT_SYMBOL(blk_set_queue_depth);
812
813/**
814 * blk_queue_write_cache - configure queue's write cache
815 * @q: the request queue for the device
816 * @wc: write back cache on or off
817 * @fua: device supports FUA writes, if true
818 *
819 * Tell the block layer about the write cache of @q.
820 */
821void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
822{
823 if (wc)
824 blk_queue_flag_set(QUEUE_FLAG_WC, q);
825 else
826 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
827 if (fua)
828 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
829 else
830 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
831
832 wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
833}
834EXPORT_SYMBOL_GPL(blk_queue_write_cache);
835
836/**
837 * blk_queue_required_elevator_features - Set a queue required elevator features
838 * @q: the request queue for the target device
839 * @features: Required elevator features OR'ed together
840 *
841 * Tell the block layer that for the device controlled through @q, only the
842 * only elevators that can be used are those that implement at least the set of
843 * features specified by @features.
844 */
845void blk_queue_required_elevator_features(struct request_queue *q,
846 unsigned int features)
847{
848 q->required_elevator_features = features;
849}
850EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
851
852/**
853 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
854 * @q: the request queue for the device
855 * @dev: the device pointer for dma
856 *
857 * Tell the block layer about merging the segments by dma map of @q.
858 */
859bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
860 struct device *dev)
861{
862 unsigned long boundary = dma_get_merge_boundary(dev);
863
864 if (!boundary)
865 return false;
866
867 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
868 blk_queue_virt_boundary(q, boundary);
869
870 return true;
871}
872EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
873
874static int __init blk_settings_init(void)
875{
876 blk_max_low_pfn = max_low_pfn - 1;
877 blk_max_pfn = max_pfn - 1;
878 return 0;
879}
880subsys_initcall(blk_settings_init);