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