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