Loading...
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.
108 */
109void blk_set_default_limits(struct queue_limits *lim)
110{
111 lim->max_segments = BLK_MAX_SEGMENTS;
112 lim->max_integrity_segments = 0;
113 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
114 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
115 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
116 lim->max_write_same_sectors = 0;
117 lim->max_discard_sectors = 0;
118 lim->discard_granularity = 0;
119 lim->discard_alignment = 0;
120 lim->discard_misaligned = 0;
121 lim->discard_zeroes_data = 0;
122 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
123 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
124 lim->alignment_offset = 0;
125 lim->io_opt = 0;
126 lim->misaligned = 0;
127 lim->cluster = 1;
128}
129EXPORT_SYMBOL(blk_set_default_limits);
130
131/**
132 * blk_set_stacking_limits - set default limits for stacking devices
133 * @lim: the queue_limits structure to reset
134 *
135 * Description:
136 * Returns a queue_limit struct to its default state. Should be used
137 * by stacking drivers like DM that have no internal limits.
138 */
139void blk_set_stacking_limits(struct queue_limits *lim)
140{
141 blk_set_default_limits(lim);
142
143 /* Inherit limits from component devices */
144 lim->discard_zeroes_data = 1;
145 lim->max_segments = USHRT_MAX;
146 lim->max_hw_sectors = UINT_MAX;
147 lim->max_segment_size = UINT_MAX;
148 lim->max_sectors = UINT_MAX;
149 lim->max_write_same_sectors = UINT_MAX;
150}
151EXPORT_SYMBOL(blk_set_stacking_limits);
152
153/**
154 * blk_queue_make_request - define an alternate make_request function for a device
155 * @q: the request queue for the device to be affected
156 * @mfn: the alternate make_request function
157 *
158 * Description:
159 * The normal way for &struct bios to be passed to a device
160 * driver is for them to be collected into requests on a request
161 * queue, and then to allow the device driver to select requests
162 * off that queue when it is ready. This works well for many block
163 * devices. However some block devices (typically virtual devices
164 * such as md or lvm) do not benefit from the processing on the
165 * request queue, and are served best by having the requests passed
166 * directly to them. This can be achieved by providing a function
167 * to blk_queue_make_request().
168 *
169 * Caveat:
170 * The driver that does this *must* be able to deal appropriately
171 * with buffers in "highmemory". This can be accomplished by either calling
172 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
173 * blk_queue_bounce() to create a buffer in normal memory.
174 **/
175void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
176{
177 /*
178 * set defaults
179 */
180 q->nr_requests = BLKDEV_MAX_RQ;
181
182 q->make_request_fn = mfn;
183 blk_queue_dma_alignment(q, 511);
184 blk_queue_congestion_threshold(q);
185 q->nr_batching = BLK_BATCH_REQ;
186
187 blk_set_default_limits(&q->limits);
188
189 /*
190 * by default assume old behaviour and bounce for any highmem page
191 */
192 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
193}
194EXPORT_SYMBOL(blk_queue_make_request);
195
196/**
197 * blk_queue_bounce_limit - set bounce buffer limit for queue
198 * @q: the request queue for the device
199 * @max_addr: the maximum address the device can handle
200 *
201 * Description:
202 * Different hardware can have different requirements as to what pages
203 * it can do I/O directly to. A low level driver can call
204 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
205 * buffers for doing I/O to pages residing above @max_addr.
206 **/
207void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
208{
209 unsigned long b_pfn = max_addr >> PAGE_SHIFT;
210 int dma = 0;
211
212 q->bounce_gfp = GFP_NOIO;
213#if BITS_PER_LONG == 64
214 /*
215 * Assume anything <= 4GB can be handled by IOMMU. Actually
216 * some IOMMUs can handle everything, but I don't know of a
217 * way to test this here.
218 */
219 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
220 dma = 1;
221 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
222#else
223 if (b_pfn < blk_max_low_pfn)
224 dma = 1;
225 q->limits.bounce_pfn = b_pfn;
226#endif
227 if (dma) {
228 init_emergency_isa_pool();
229 q->bounce_gfp = GFP_NOIO | GFP_DMA;
230 q->limits.bounce_pfn = b_pfn;
231 }
232}
233EXPORT_SYMBOL(blk_queue_bounce_limit);
234
235/**
236 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
237 * @limits: the queue limits
238 * @max_hw_sectors: max hardware sectors in the usual 512b unit
239 *
240 * Description:
241 * Enables a low level driver to set a hard upper limit,
242 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
243 * the device driver based upon the combined capabilities of I/O
244 * controller and storage device.
245 *
246 * max_sectors is a soft limit imposed by the block layer for
247 * filesystem type requests. This value can be overridden on a
248 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
249 * The soft limit can not exceed max_hw_sectors.
250 **/
251void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
252{
253 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
254 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
255 printk(KERN_INFO "%s: set to minimum %d\n",
256 __func__, max_hw_sectors);
257 }
258
259 limits->max_hw_sectors = max_hw_sectors;
260 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
261 BLK_DEF_MAX_SECTORS);
262}
263EXPORT_SYMBOL(blk_limits_max_hw_sectors);
264
265/**
266 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
267 * @q: the request queue for the device
268 * @max_hw_sectors: max hardware sectors in the usual 512b unit
269 *
270 * Description:
271 * See description for blk_limits_max_hw_sectors().
272 **/
273void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
274{
275 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
276}
277EXPORT_SYMBOL(blk_queue_max_hw_sectors);
278
279/**
280 * blk_queue_max_discard_sectors - set max sectors for a single discard
281 * @q: the request queue for the device
282 * @max_discard_sectors: maximum number of sectors to discard
283 **/
284void blk_queue_max_discard_sectors(struct request_queue *q,
285 unsigned int max_discard_sectors)
286{
287 q->limits.max_discard_sectors = max_discard_sectors;
288}
289EXPORT_SYMBOL(blk_queue_max_discard_sectors);
290
291/**
292 * blk_queue_max_write_same_sectors - set max sectors for a single write same
293 * @q: the request queue for the device
294 * @max_write_same_sectors: maximum number of sectors to write per command
295 **/
296void blk_queue_max_write_same_sectors(struct request_queue *q,
297 unsigned int max_write_same_sectors)
298{
299 q->limits.max_write_same_sectors = max_write_same_sectors;
300}
301EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
302
303/**
304 * blk_queue_max_segments - set max hw segments for a request for this queue
305 * @q: the request queue for the device
306 * @max_segments: max number of segments
307 *
308 * Description:
309 * Enables a low level driver to set an upper limit on the number of
310 * hw data segments in a request.
311 **/
312void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
313{
314 if (!max_segments) {
315 max_segments = 1;
316 printk(KERN_INFO "%s: set to minimum %d\n",
317 __func__, max_segments);
318 }
319
320 q->limits.max_segments = max_segments;
321}
322EXPORT_SYMBOL(blk_queue_max_segments);
323
324/**
325 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
326 * @q: the request queue for the device
327 * @max_size: max size of segment in bytes
328 *
329 * Description:
330 * Enables a low level driver to set an upper limit on the size of a
331 * coalesced segment
332 **/
333void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
334{
335 if (max_size < PAGE_CACHE_SIZE) {
336 max_size = PAGE_CACHE_SIZE;
337 printk(KERN_INFO "%s: set to minimum %d\n",
338 __func__, max_size);
339 }
340
341 q->limits.max_segment_size = max_size;
342}
343EXPORT_SYMBOL(blk_queue_max_segment_size);
344
345/**
346 * blk_queue_logical_block_size - set logical block size for the queue
347 * @q: the request queue for the device
348 * @size: the logical block size, in bytes
349 *
350 * Description:
351 * This should be set to the lowest possible block size that the
352 * storage device can address. The default of 512 covers most
353 * hardware.
354 **/
355void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
356{
357 q->limits.logical_block_size = size;
358
359 if (q->limits.physical_block_size < size)
360 q->limits.physical_block_size = size;
361
362 if (q->limits.io_min < q->limits.physical_block_size)
363 q->limits.io_min = q->limits.physical_block_size;
364}
365EXPORT_SYMBOL(blk_queue_logical_block_size);
366
367/**
368 * blk_queue_physical_block_size - set physical block size for the queue
369 * @q: the request queue for the device
370 * @size: the physical block size, in bytes
371 *
372 * Description:
373 * This should be set to the lowest possible sector size that the
374 * hardware can operate on without reverting to read-modify-write
375 * operations.
376 */
377void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
378{
379 q->limits.physical_block_size = size;
380
381 if (q->limits.physical_block_size < q->limits.logical_block_size)
382 q->limits.physical_block_size = q->limits.logical_block_size;
383
384 if (q->limits.io_min < q->limits.physical_block_size)
385 q->limits.io_min = q->limits.physical_block_size;
386}
387EXPORT_SYMBOL(blk_queue_physical_block_size);
388
389/**
390 * blk_queue_alignment_offset - set physical block alignment offset
391 * @q: the request queue for the device
392 * @offset: alignment offset in bytes
393 *
394 * Description:
395 * Some devices are naturally misaligned to compensate for things like
396 * the legacy DOS partition table 63-sector offset. Low-level drivers
397 * should call this function for devices whose first sector is not
398 * naturally aligned.
399 */
400void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
401{
402 q->limits.alignment_offset =
403 offset & (q->limits.physical_block_size - 1);
404 q->limits.misaligned = 0;
405}
406EXPORT_SYMBOL(blk_queue_alignment_offset);
407
408/**
409 * blk_limits_io_min - set minimum request size for a device
410 * @limits: the queue limits
411 * @min: smallest I/O size in bytes
412 *
413 * Description:
414 * Some devices have an internal block size bigger than the reported
415 * hardware sector size. This function can be used to signal the
416 * smallest I/O the device can perform without incurring a performance
417 * penalty.
418 */
419void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
420{
421 limits->io_min = min;
422
423 if (limits->io_min < limits->logical_block_size)
424 limits->io_min = limits->logical_block_size;
425
426 if (limits->io_min < limits->physical_block_size)
427 limits->io_min = limits->physical_block_size;
428}
429EXPORT_SYMBOL(blk_limits_io_min);
430
431/**
432 * blk_queue_io_min - set minimum request size for the queue
433 * @q: the request queue for the device
434 * @min: smallest I/O size in bytes
435 *
436 * Description:
437 * Storage devices may report a granularity or preferred minimum I/O
438 * size which is the smallest request the device can perform without
439 * incurring a performance penalty. For disk drives this is often the
440 * physical block size. For RAID arrays it is often the stripe chunk
441 * size. A properly aligned multiple of minimum_io_size is the
442 * preferred request size for workloads where a high number of I/O
443 * operations is desired.
444 */
445void blk_queue_io_min(struct request_queue *q, unsigned int min)
446{
447 blk_limits_io_min(&q->limits, min);
448}
449EXPORT_SYMBOL(blk_queue_io_min);
450
451/**
452 * blk_limits_io_opt - set optimal request size for a device
453 * @limits: the queue limits
454 * @opt: smallest I/O size in bytes
455 *
456 * Description:
457 * Storage devices may report an optimal I/O size, which is the
458 * device's preferred unit for sustained I/O. This is rarely reported
459 * for disk drives. For RAID arrays it is usually the stripe width or
460 * the internal track size. A properly aligned multiple of
461 * optimal_io_size is the preferred request size for workloads where
462 * sustained throughput is desired.
463 */
464void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
465{
466 limits->io_opt = opt;
467}
468EXPORT_SYMBOL(blk_limits_io_opt);
469
470/**
471 * blk_queue_io_opt - set optimal request size for the queue
472 * @q: the request queue for the device
473 * @opt: optimal request size in bytes
474 *
475 * Description:
476 * Storage devices may report an optimal I/O size, which is the
477 * device's preferred unit for sustained I/O. This is rarely reported
478 * for disk drives. For RAID arrays it is usually the stripe width or
479 * the internal track size. A properly aligned multiple of
480 * optimal_io_size is the preferred request size for workloads where
481 * sustained throughput is desired.
482 */
483void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
484{
485 blk_limits_io_opt(&q->limits, opt);
486}
487EXPORT_SYMBOL(blk_queue_io_opt);
488
489/**
490 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
491 * @t: the stacking driver (top)
492 * @b: the underlying device (bottom)
493 **/
494void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
495{
496 blk_stack_limits(&t->limits, &b->limits, 0);
497}
498EXPORT_SYMBOL(blk_queue_stack_limits);
499
500/**
501 * blk_stack_limits - adjust queue_limits for stacked devices
502 * @t: the stacking driver limits (top device)
503 * @b: the underlying queue limits (bottom, component device)
504 * @start: first data sector within component device
505 *
506 * Description:
507 * This function is used by stacking drivers like MD and DM to ensure
508 * that all component devices have compatible block sizes and
509 * alignments. The stacking driver must provide a queue_limits
510 * struct (top) and then iteratively call the stacking function for
511 * all component (bottom) devices. The stacking function will
512 * attempt to combine the values and ensure proper alignment.
513 *
514 * Returns 0 if the top and bottom queue_limits are compatible. The
515 * top device's block sizes and alignment offsets may be adjusted to
516 * ensure alignment with the bottom device. If no compatible sizes
517 * and alignments exist, -1 is returned and the resulting top
518 * queue_limits will have the misaligned flag set to indicate that
519 * the alignment_offset is undefined.
520 */
521int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
522 sector_t start)
523{
524 unsigned int top, bottom, alignment, ret = 0;
525
526 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
527 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
528 t->max_write_same_sectors = min(t->max_write_same_sectors,
529 b->max_write_same_sectors);
530 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
531
532 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
533 b->seg_boundary_mask);
534
535 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
536 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
537 b->max_integrity_segments);
538
539 t->max_segment_size = min_not_zero(t->max_segment_size,
540 b->max_segment_size);
541
542 t->misaligned |= b->misaligned;
543
544 alignment = queue_limit_alignment_offset(b, start);
545
546 /* Bottom device has different alignment. Check that it is
547 * compatible with the current top alignment.
548 */
549 if (t->alignment_offset != alignment) {
550
551 top = max(t->physical_block_size, t->io_min)
552 + t->alignment_offset;
553 bottom = max(b->physical_block_size, b->io_min) + alignment;
554
555 /* Verify that top and bottom intervals line up */
556 if (max(top, bottom) & (min(top, bottom) - 1)) {
557 t->misaligned = 1;
558 ret = -1;
559 }
560 }
561
562 t->logical_block_size = max(t->logical_block_size,
563 b->logical_block_size);
564
565 t->physical_block_size = max(t->physical_block_size,
566 b->physical_block_size);
567
568 t->io_min = max(t->io_min, b->io_min);
569 t->io_opt = lcm(t->io_opt, b->io_opt);
570
571 t->cluster &= b->cluster;
572 t->discard_zeroes_data &= b->discard_zeroes_data;
573
574 /* Physical block size a multiple of the logical block size? */
575 if (t->physical_block_size & (t->logical_block_size - 1)) {
576 t->physical_block_size = t->logical_block_size;
577 t->misaligned = 1;
578 ret = -1;
579 }
580
581 /* Minimum I/O a multiple of the physical block size? */
582 if (t->io_min & (t->physical_block_size - 1)) {
583 t->io_min = t->physical_block_size;
584 t->misaligned = 1;
585 ret = -1;
586 }
587
588 /* Optimal I/O a multiple of the physical block size? */
589 if (t->io_opt & (t->physical_block_size - 1)) {
590 t->io_opt = 0;
591 t->misaligned = 1;
592 ret = -1;
593 }
594
595 t->raid_partial_stripes_expensive =
596 max(t->raid_partial_stripes_expensive,
597 b->raid_partial_stripes_expensive);
598
599 /* Find lowest common alignment_offset */
600 t->alignment_offset = lcm(t->alignment_offset, alignment)
601 & (max(t->physical_block_size, t->io_min) - 1);
602
603 /* Verify that new alignment_offset is on a logical block boundary */
604 if (t->alignment_offset & (t->logical_block_size - 1)) {
605 t->misaligned = 1;
606 ret = -1;
607 }
608
609 /* Discard alignment and granularity */
610 if (b->discard_granularity) {
611 alignment = queue_limit_discard_alignment(b, start);
612
613 if (t->discard_granularity != 0 &&
614 t->discard_alignment != alignment) {
615 top = t->discard_granularity + t->discard_alignment;
616 bottom = b->discard_granularity + alignment;
617
618 /* Verify that top and bottom intervals line up */
619 if ((max(top, bottom) % min(top, bottom)) != 0)
620 t->discard_misaligned = 1;
621 }
622
623 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
624 b->max_discard_sectors);
625 t->discard_granularity = max(t->discard_granularity,
626 b->discard_granularity);
627 t->discard_alignment = lcm(t->discard_alignment, alignment) %
628 t->discard_granularity;
629 }
630
631 return ret;
632}
633EXPORT_SYMBOL(blk_stack_limits);
634
635/**
636 * bdev_stack_limits - adjust queue limits for stacked drivers
637 * @t: the stacking driver limits (top device)
638 * @bdev: the component block_device (bottom)
639 * @start: first data sector within component device
640 *
641 * Description:
642 * Merges queue limits for a top device and a block_device. Returns
643 * 0 if alignment didn't change. Returns -1 if adding the bottom
644 * device caused misalignment.
645 */
646int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
647 sector_t start)
648{
649 struct request_queue *bq = bdev_get_queue(bdev);
650
651 start += get_start_sect(bdev);
652
653 return blk_stack_limits(t, &bq->limits, start);
654}
655EXPORT_SYMBOL(bdev_stack_limits);
656
657/**
658 * disk_stack_limits - adjust queue limits for stacked drivers
659 * @disk: MD/DM gendisk (top)
660 * @bdev: the underlying block device (bottom)
661 * @offset: offset to beginning of data within component device
662 *
663 * Description:
664 * Merges the limits for a top level gendisk and a bottom level
665 * block_device.
666 */
667void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
668 sector_t offset)
669{
670 struct request_queue *t = disk->queue;
671
672 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
673 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
674
675 disk_name(disk, 0, top);
676 bdevname(bdev, bottom);
677
678 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
679 top, bottom);
680 }
681}
682EXPORT_SYMBOL(disk_stack_limits);
683
684/**
685 * blk_queue_dma_pad - set pad mask
686 * @q: the request queue for the device
687 * @mask: pad mask
688 *
689 * Set dma pad mask.
690 *
691 * Appending pad buffer to a request modifies the last entry of a
692 * scatter list such that it includes the pad buffer.
693 **/
694void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
695{
696 q->dma_pad_mask = mask;
697}
698EXPORT_SYMBOL(blk_queue_dma_pad);
699
700/**
701 * blk_queue_update_dma_pad - update pad mask
702 * @q: the request queue for the device
703 * @mask: pad mask
704 *
705 * Update dma pad mask.
706 *
707 * Appending pad buffer to a request modifies the last entry of a
708 * scatter list such that it includes the pad buffer.
709 **/
710void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
711{
712 if (mask > q->dma_pad_mask)
713 q->dma_pad_mask = mask;
714}
715EXPORT_SYMBOL(blk_queue_update_dma_pad);
716
717/**
718 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
719 * @q: the request queue for the device
720 * @dma_drain_needed: fn which returns non-zero if drain is necessary
721 * @buf: physically contiguous buffer
722 * @size: size of the buffer in bytes
723 *
724 * Some devices have excess DMA problems and can't simply discard (or
725 * zero fill) the unwanted piece of the transfer. They have to have a
726 * real area of memory to transfer it into. The use case for this is
727 * ATAPI devices in DMA mode. If the packet command causes a transfer
728 * bigger than the transfer size some HBAs will lock up if there
729 * aren't DMA elements to contain the excess transfer. What this API
730 * does is adjust the queue so that the buf is always appended
731 * silently to the scatterlist.
732 *
733 * Note: This routine adjusts max_hw_segments to make room for appending
734 * the drain buffer. If you call blk_queue_max_segments() after calling
735 * this routine, you must set the limit to one fewer than your device
736 * can support otherwise there won't be room for the drain buffer.
737 */
738int blk_queue_dma_drain(struct request_queue *q,
739 dma_drain_needed_fn *dma_drain_needed,
740 void *buf, unsigned int size)
741{
742 if (queue_max_segments(q) < 2)
743 return -EINVAL;
744 /* make room for appending the drain */
745 blk_queue_max_segments(q, queue_max_segments(q) - 1);
746 q->dma_drain_needed = dma_drain_needed;
747 q->dma_drain_buffer = buf;
748 q->dma_drain_size = size;
749
750 return 0;
751}
752EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
753
754/**
755 * blk_queue_segment_boundary - set boundary rules for segment merging
756 * @q: the request queue for the device
757 * @mask: the memory boundary mask
758 **/
759void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
760{
761 if (mask < PAGE_CACHE_SIZE - 1) {
762 mask = PAGE_CACHE_SIZE - 1;
763 printk(KERN_INFO "%s: set to minimum %lx\n",
764 __func__, mask);
765 }
766
767 q->limits.seg_boundary_mask = mask;
768}
769EXPORT_SYMBOL(blk_queue_segment_boundary);
770
771/**
772 * blk_queue_dma_alignment - set dma length and memory alignment
773 * @q: the request queue for the device
774 * @mask: alignment mask
775 *
776 * description:
777 * set required memory and length alignment for direct dma transactions.
778 * this is used when building direct io requests for the queue.
779 *
780 **/
781void blk_queue_dma_alignment(struct request_queue *q, int mask)
782{
783 q->dma_alignment = mask;
784}
785EXPORT_SYMBOL(blk_queue_dma_alignment);
786
787/**
788 * blk_queue_update_dma_alignment - update dma length and memory alignment
789 * @q: the request queue for the device
790 * @mask: alignment mask
791 *
792 * description:
793 * update required memory and length alignment for direct dma transactions.
794 * If the requested alignment is larger than the current alignment, then
795 * the current queue alignment is updated to the new value, otherwise it
796 * is left alone. The design of this is to allow multiple objects
797 * (driver, device, transport etc) to set their respective
798 * alignments without having them interfere.
799 *
800 **/
801void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
802{
803 BUG_ON(mask > PAGE_SIZE);
804
805 if (mask > q->dma_alignment)
806 q->dma_alignment = mask;
807}
808EXPORT_SYMBOL(blk_queue_update_dma_alignment);
809
810/**
811 * blk_queue_flush - configure queue's cache flush capability
812 * @q: the request queue for the device
813 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
814 *
815 * Tell block layer cache flush capability of @q. If it supports
816 * flushing, REQ_FLUSH should be set. If it supports bypassing
817 * write cache for individual writes, REQ_FUA should be set.
818 */
819void blk_queue_flush(struct request_queue *q, unsigned int flush)
820{
821 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
822
823 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
824 flush &= ~REQ_FUA;
825
826 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
827}
828EXPORT_SYMBOL_GPL(blk_queue_flush);
829
830void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
831{
832 q->flush_not_queueable = !queueable;
833}
834EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
835
836static int __init blk_settings_init(void)
837{
838 blk_max_low_pfn = max_low_pfn - 1;
839 blk_max_pfn = max_pfn - 1;
840 return 0;
841}
842subsys_initcall(blk_settings_init);
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-rq-qos.h"
20#include "blk-wbt.h"
21
22void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
23{
24 q->rq_timeout = timeout;
25}
26EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
27
28/**
29 * blk_set_stacking_limits - set default limits for stacking devices
30 * @lim: the queue_limits structure to reset
31 *
32 * Prepare queue limits for applying limits from underlying devices using
33 * blk_stack_limits().
34 */
35void blk_set_stacking_limits(struct queue_limits *lim)
36{
37 memset(lim, 0, sizeof(*lim));
38 lim->logical_block_size = SECTOR_SIZE;
39 lim->physical_block_size = SECTOR_SIZE;
40 lim->io_min = SECTOR_SIZE;
41 lim->discard_granularity = SECTOR_SIZE;
42 lim->dma_alignment = SECTOR_SIZE - 1;
43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
44
45 /* Inherit limits from component devices */
46 lim->max_segments = USHRT_MAX;
47 lim->max_discard_segments = USHRT_MAX;
48 lim->max_hw_sectors = UINT_MAX;
49 lim->max_segment_size = UINT_MAX;
50 lim->max_sectors = UINT_MAX;
51 lim->max_dev_sectors = UINT_MAX;
52 lim->max_write_zeroes_sectors = UINT_MAX;
53 lim->max_zone_append_sectors = UINT_MAX;
54 lim->max_user_discard_sectors = UINT_MAX;
55}
56EXPORT_SYMBOL(blk_set_stacking_limits);
57
58static void blk_apply_bdi_limits(struct backing_dev_info *bdi,
59 struct queue_limits *lim)
60{
61 /*
62 * For read-ahead of large files to be effective, we need to read ahead
63 * at least twice the optimal I/O size.
64 */
65 bdi->ra_pages = max(lim->io_opt * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
66 bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT;
67}
68
69static int blk_validate_zoned_limits(struct queue_limits *lim)
70{
71 if (!lim->zoned) {
72 if (WARN_ON_ONCE(lim->max_open_zones) ||
73 WARN_ON_ONCE(lim->max_active_zones) ||
74 WARN_ON_ONCE(lim->zone_write_granularity) ||
75 WARN_ON_ONCE(lim->max_zone_append_sectors))
76 return -EINVAL;
77 return 0;
78 }
79
80 if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)))
81 return -EINVAL;
82
83 if (lim->zone_write_granularity < lim->logical_block_size)
84 lim->zone_write_granularity = lim->logical_block_size;
85
86 if (lim->max_zone_append_sectors) {
87 /*
88 * The Zone Append size is limited by the maximum I/O size
89 * and the zone size given that it can't span zones.
90 */
91 lim->max_zone_append_sectors =
92 min3(lim->max_hw_sectors,
93 lim->max_zone_append_sectors,
94 lim->chunk_sectors);
95 }
96
97 return 0;
98}
99
100/*
101 * Check that the limits in lim are valid, initialize defaults for unset
102 * values, and cap values based on others where needed.
103 */
104static int blk_validate_limits(struct queue_limits *lim)
105{
106 unsigned int max_hw_sectors;
107
108 /*
109 * Unless otherwise specified, default to 512 byte logical blocks and a
110 * physical block size equal to the logical block size.
111 */
112 if (!lim->logical_block_size)
113 lim->logical_block_size = SECTOR_SIZE;
114 if (lim->physical_block_size < lim->logical_block_size)
115 lim->physical_block_size = lim->logical_block_size;
116
117 /*
118 * The minimum I/O size defaults to the physical block size unless
119 * explicitly overridden.
120 */
121 if (lim->io_min < lim->physical_block_size)
122 lim->io_min = lim->physical_block_size;
123
124 /*
125 * max_hw_sectors has a somewhat weird default for historical reason,
126 * but driver really should set their own instead of relying on this
127 * value.
128 *
129 * The block layer relies on the fact that every driver can
130 * handle at lest a page worth of data per I/O, and needs the value
131 * aligned to the logical block size.
132 */
133 if (!lim->max_hw_sectors)
134 lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
135 if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
136 return -EINVAL;
137 lim->max_hw_sectors = round_down(lim->max_hw_sectors,
138 lim->logical_block_size >> SECTOR_SHIFT);
139
140 /*
141 * The actual max_sectors value is a complex beast and also takes the
142 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
143 * value into account. The ->max_sectors value is always calculated
144 * from these, so directly setting it won't have any effect.
145 */
146 max_hw_sectors = min_not_zero(lim->max_hw_sectors,
147 lim->max_dev_sectors);
148 if (lim->max_user_sectors) {
149 if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
150 return -EINVAL;
151 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
152 } else {
153 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
154 }
155 lim->max_sectors = round_down(lim->max_sectors,
156 lim->logical_block_size >> SECTOR_SHIFT);
157
158 /*
159 * Random default for the maximum number of segments. Driver should not
160 * rely on this and set their own.
161 */
162 if (!lim->max_segments)
163 lim->max_segments = BLK_MAX_SEGMENTS;
164
165 lim->max_discard_sectors =
166 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
167
168 if (!lim->max_discard_segments)
169 lim->max_discard_segments = 1;
170
171 if (lim->discard_granularity < lim->physical_block_size)
172 lim->discard_granularity = lim->physical_block_size;
173
174 /*
175 * By default there is no limit on the segment boundary alignment,
176 * but if there is one it can't be smaller than the page size as
177 * that would break all the normal I/O patterns.
178 */
179 if (!lim->seg_boundary_mask)
180 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
181 if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
182 return -EINVAL;
183
184 /*
185 * Stacking device may have both virtual boundary and max segment
186 * size limit, so allow this setting now, and long-term the two
187 * might need to move out of stacking limits since we have immutable
188 * bvec and lower layer bio splitting is supposed to handle the two
189 * correctly.
190 */
191 if (lim->virt_boundary_mask) {
192 if (!lim->max_segment_size)
193 lim->max_segment_size = UINT_MAX;
194 } else {
195 /*
196 * The maximum segment size has an odd historic 64k default that
197 * drivers probably should override. Just like the I/O size we
198 * require drivers to at least handle a full page per segment.
199 */
200 if (!lim->max_segment_size)
201 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
202 if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
203 return -EINVAL;
204 }
205
206 /*
207 * We require drivers to at least do logical block aligned I/O, but
208 * historically could not check for that due to the separate calls
209 * to set the limits. Once the transition is finished the check
210 * below should be narrowed down to check the logical block size.
211 */
212 if (!lim->dma_alignment)
213 lim->dma_alignment = SECTOR_SIZE - 1;
214 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
215 return -EINVAL;
216
217 if (lim->alignment_offset) {
218 lim->alignment_offset &= (lim->physical_block_size - 1);
219 lim->misaligned = 0;
220 }
221
222 return blk_validate_zoned_limits(lim);
223}
224
225/*
226 * Set the default limits for a newly allocated queue. @lim contains the
227 * initial limits set by the driver, which could be no limit in which case
228 * all fields are cleared to zero.
229 */
230int blk_set_default_limits(struct queue_limits *lim)
231{
232 /*
233 * Most defaults are set by capping the bounds in blk_validate_limits,
234 * but max_user_discard_sectors is special and needs an explicit
235 * initialization to the max value here.
236 */
237 lim->max_user_discard_sectors = UINT_MAX;
238 return blk_validate_limits(lim);
239}
240
241/**
242 * queue_limits_commit_update - commit an atomic update of queue limits
243 * @q: queue to update
244 * @lim: limits to apply
245 *
246 * Apply the limits in @lim that were obtained from queue_limits_start_update()
247 * and updated by the caller to @q.
248 *
249 * Returns 0 if successful, else a negative error code.
250 */
251int queue_limits_commit_update(struct request_queue *q,
252 struct queue_limits *lim)
253 __releases(q->limits_lock)
254{
255 int error = blk_validate_limits(lim);
256
257 if (!error) {
258 q->limits = *lim;
259 if (q->disk)
260 blk_apply_bdi_limits(q->disk->bdi, lim);
261 }
262 mutex_unlock(&q->limits_lock);
263 return error;
264}
265EXPORT_SYMBOL_GPL(queue_limits_commit_update);
266
267/**
268 * queue_limits_set - apply queue limits to queue
269 * @q: queue to update
270 * @lim: limits to apply
271 *
272 * Apply the limits in @lim that were freshly initialized to @q.
273 * To update existing limits use queue_limits_start_update() and
274 * queue_limits_commit_update() instead.
275 *
276 * Returns 0 if successful, else a negative error code.
277 */
278int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
279{
280 mutex_lock(&q->limits_lock);
281 return queue_limits_commit_update(q, lim);
282}
283EXPORT_SYMBOL_GPL(queue_limits_set);
284
285/**
286 * blk_queue_bounce_limit - set bounce buffer limit for queue
287 * @q: the request queue for the device
288 * @bounce: bounce limit to enforce
289 *
290 * Description:
291 * Force bouncing for ISA DMA ranges or highmem.
292 *
293 * DEPRECATED, don't use in new code.
294 **/
295void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
296{
297 q->limits.bounce = bounce;
298}
299EXPORT_SYMBOL(blk_queue_bounce_limit);
300
301/**
302 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
303 * @q: the request queue for the device
304 * @max_hw_sectors: max hardware sectors in the usual 512b unit
305 *
306 * Description:
307 * Enables a low level driver to set a hard upper limit,
308 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
309 * the device driver based upon the capabilities of the I/O
310 * controller.
311 *
312 * max_dev_sectors is a hard limit imposed by the storage device for
313 * READ/WRITE requests. It is set by the disk driver.
314 *
315 * max_sectors is a soft limit imposed by the block layer for
316 * filesystem type requests. This value can be overridden on a
317 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
318 * The soft limit can not exceed max_hw_sectors.
319 **/
320void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
321{
322 struct queue_limits *limits = &q->limits;
323 unsigned int max_sectors;
324
325 if ((max_hw_sectors << 9) < PAGE_SIZE) {
326 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
327 pr_info("%s: set to minimum %u\n", __func__, max_hw_sectors);
328 }
329
330 max_hw_sectors = round_down(max_hw_sectors,
331 limits->logical_block_size >> SECTOR_SHIFT);
332 limits->max_hw_sectors = max_hw_sectors;
333
334 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
335
336 if (limits->max_user_sectors)
337 max_sectors = min(max_sectors, limits->max_user_sectors);
338 else
339 max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS_CAP);
340
341 max_sectors = round_down(max_sectors,
342 limits->logical_block_size >> SECTOR_SHIFT);
343 limits->max_sectors = max_sectors;
344
345 if (!q->disk)
346 return;
347 q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
348}
349EXPORT_SYMBOL(blk_queue_max_hw_sectors);
350
351/**
352 * blk_queue_chunk_sectors - set size of the chunk for this queue
353 * @q: the request queue for the device
354 * @chunk_sectors: chunk sectors in the usual 512b unit
355 *
356 * Description:
357 * If a driver doesn't want IOs to cross a given chunk size, it can set
358 * this limit and prevent merging across chunks. Note that the block layer
359 * must accept a page worth of data at any offset. So if the crossing of
360 * chunks is a hard limitation in the driver, it must still be prepared
361 * to split single page bios.
362 **/
363void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
364{
365 q->limits.chunk_sectors = chunk_sectors;
366}
367EXPORT_SYMBOL(blk_queue_chunk_sectors);
368
369/**
370 * blk_queue_max_discard_sectors - set max sectors for a single discard
371 * @q: the request queue for the device
372 * @max_discard_sectors: maximum number of sectors to discard
373 **/
374void blk_queue_max_discard_sectors(struct request_queue *q,
375 unsigned int max_discard_sectors)
376{
377 struct queue_limits *lim = &q->limits;
378
379 lim->max_hw_discard_sectors = max_discard_sectors;
380 lim->max_discard_sectors =
381 min(max_discard_sectors, lim->max_user_discard_sectors);
382}
383EXPORT_SYMBOL(blk_queue_max_discard_sectors);
384
385/**
386 * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase
387 * @q: the request queue for the device
388 * @max_sectors: maximum number of sectors to secure_erase
389 **/
390void blk_queue_max_secure_erase_sectors(struct request_queue *q,
391 unsigned int max_sectors)
392{
393 q->limits.max_secure_erase_sectors = max_sectors;
394}
395EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors);
396
397/**
398 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
399 * write zeroes
400 * @q: the request queue for the device
401 * @max_write_zeroes_sectors: maximum number of sectors to write per command
402 **/
403void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
404 unsigned int max_write_zeroes_sectors)
405{
406 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
407}
408EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
409
410/**
411 * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
412 * @q: the request queue for the device
413 * @max_zone_append_sectors: maximum number of sectors to write per command
414 **/
415void blk_queue_max_zone_append_sectors(struct request_queue *q,
416 unsigned int max_zone_append_sectors)
417{
418 unsigned int max_sectors;
419
420 if (WARN_ON(!blk_queue_is_zoned(q)))
421 return;
422
423 max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
424 max_sectors = min(q->limits.chunk_sectors, max_sectors);
425
426 /*
427 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
428 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
429 * or the max_hw_sectors limit not set.
430 */
431 WARN_ON(!max_sectors);
432
433 q->limits.max_zone_append_sectors = max_sectors;
434}
435EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
436
437/**
438 * blk_queue_max_segments - set max hw segments for a request for this queue
439 * @q: the request queue for the device
440 * @max_segments: max number of segments
441 *
442 * Description:
443 * Enables a low level driver to set an upper limit on the number of
444 * hw data segments in a request.
445 **/
446void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
447{
448 if (!max_segments) {
449 max_segments = 1;
450 pr_info("%s: set to minimum %u\n", __func__, max_segments);
451 }
452
453 q->limits.max_segments = max_segments;
454}
455EXPORT_SYMBOL(blk_queue_max_segments);
456
457/**
458 * blk_queue_max_discard_segments - set max segments for discard requests
459 * @q: the request queue for the device
460 * @max_segments: max number of segments
461 *
462 * Description:
463 * Enables a low level driver to set an upper limit on the number of
464 * segments in a discard request.
465 **/
466void blk_queue_max_discard_segments(struct request_queue *q,
467 unsigned short max_segments)
468{
469 q->limits.max_discard_segments = max_segments;
470}
471EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
472
473/**
474 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
475 * @q: the request queue for the device
476 * @max_size: max size of segment in bytes
477 *
478 * Description:
479 * Enables a low level driver to set an upper limit on the size of a
480 * coalesced segment
481 **/
482void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
483{
484 if (max_size < PAGE_SIZE) {
485 max_size = PAGE_SIZE;
486 pr_info("%s: set to minimum %u\n", __func__, max_size);
487 }
488
489 /* see blk_queue_virt_boundary() for the explanation */
490 WARN_ON_ONCE(q->limits.virt_boundary_mask);
491
492 q->limits.max_segment_size = max_size;
493}
494EXPORT_SYMBOL(blk_queue_max_segment_size);
495
496/**
497 * blk_queue_logical_block_size - set logical block size for the queue
498 * @q: the request queue for the device
499 * @size: the logical block size, in bytes
500 *
501 * Description:
502 * This should be set to the lowest possible block size that the
503 * storage device can address. The default of 512 covers most
504 * hardware.
505 **/
506void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
507{
508 struct queue_limits *limits = &q->limits;
509
510 limits->logical_block_size = size;
511
512 if (limits->discard_granularity < limits->logical_block_size)
513 limits->discard_granularity = limits->logical_block_size;
514
515 if (limits->physical_block_size < size)
516 limits->physical_block_size = size;
517
518 if (limits->io_min < limits->physical_block_size)
519 limits->io_min = limits->physical_block_size;
520
521 limits->max_hw_sectors =
522 round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
523 limits->max_sectors =
524 round_down(limits->max_sectors, size >> SECTOR_SHIFT);
525}
526EXPORT_SYMBOL(blk_queue_logical_block_size);
527
528/**
529 * blk_queue_physical_block_size - set physical block size for the queue
530 * @q: the request queue for the device
531 * @size: the physical block size, in bytes
532 *
533 * Description:
534 * This should be set to the lowest possible sector size that the
535 * hardware can operate on without reverting to read-modify-write
536 * operations.
537 */
538void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
539{
540 q->limits.physical_block_size = size;
541
542 if (q->limits.physical_block_size < q->limits.logical_block_size)
543 q->limits.physical_block_size = q->limits.logical_block_size;
544
545 if (q->limits.discard_granularity < q->limits.physical_block_size)
546 q->limits.discard_granularity = q->limits.physical_block_size;
547
548 if (q->limits.io_min < q->limits.physical_block_size)
549 q->limits.io_min = q->limits.physical_block_size;
550}
551EXPORT_SYMBOL(blk_queue_physical_block_size);
552
553/**
554 * blk_queue_zone_write_granularity - set zone write granularity for the queue
555 * @q: the request queue for the zoned device
556 * @size: the zone write granularity size, in bytes
557 *
558 * Description:
559 * This should be set to the lowest possible size allowing to write in
560 * sequential zones of a zoned block device.
561 */
562void blk_queue_zone_write_granularity(struct request_queue *q,
563 unsigned int size)
564{
565 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
566 return;
567
568 q->limits.zone_write_granularity = size;
569
570 if (q->limits.zone_write_granularity < q->limits.logical_block_size)
571 q->limits.zone_write_granularity = q->limits.logical_block_size;
572}
573EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
574
575/**
576 * blk_queue_alignment_offset - set physical block alignment offset
577 * @q: the request queue for the device
578 * @offset: alignment offset in bytes
579 *
580 * Description:
581 * Some devices are naturally misaligned to compensate for things like
582 * the legacy DOS partition table 63-sector offset. Low-level drivers
583 * should call this function for devices whose first sector is not
584 * naturally aligned.
585 */
586void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
587{
588 q->limits.alignment_offset =
589 offset & (q->limits.physical_block_size - 1);
590 q->limits.misaligned = 0;
591}
592EXPORT_SYMBOL(blk_queue_alignment_offset);
593
594void disk_update_readahead(struct gendisk *disk)
595{
596 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
597}
598EXPORT_SYMBOL_GPL(disk_update_readahead);
599
600/**
601 * blk_limits_io_min - set minimum request size for a device
602 * @limits: the queue limits
603 * @min: smallest I/O size in bytes
604 *
605 * Description:
606 * Some devices have an internal block size bigger than the reported
607 * hardware sector size. This function can be used to signal the
608 * smallest I/O the device can perform without incurring a performance
609 * penalty.
610 */
611void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
612{
613 limits->io_min = min;
614
615 if (limits->io_min < limits->logical_block_size)
616 limits->io_min = limits->logical_block_size;
617
618 if (limits->io_min < limits->physical_block_size)
619 limits->io_min = limits->physical_block_size;
620}
621EXPORT_SYMBOL(blk_limits_io_min);
622
623/**
624 * blk_queue_io_min - set minimum request size for the queue
625 * @q: the request queue for the device
626 * @min: smallest I/O size in bytes
627 *
628 * Description:
629 * Storage devices may report a granularity or preferred minimum I/O
630 * size which is the smallest request the device can perform without
631 * incurring a performance penalty. For disk drives this is often the
632 * physical block size. For RAID arrays it is often the stripe chunk
633 * size. A properly aligned multiple of minimum_io_size is the
634 * preferred request size for workloads where a high number of I/O
635 * operations is desired.
636 */
637void blk_queue_io_min(struct request_queue *q, unsigned int min)
638{
639 blk_limits_io_min(&q->limits, min);
640}
641EXPORT_SYMBOL(blk_queue_io_min);
642
643/**
644 * blk_limits_io_opt - set optimal request size for a device
645 * @limits: the queue limits
646 * @opt: smallest I/O size in bytes
647 *
648 * Description:
649 * Storage devices may report an optimal I/O size, which is the
650 * device's preferred unit for sustained I/O. This is rarely reported
651 * for disk drives. For RAID arrays it is usually the stripe width or
652 * the internal track size. A properly aligned multiple of
653 * optimal_io_size is the preferred request size for workloads where
654 * sustained throughput is desired.
655 */
656void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
657{
658 limits->io_opt = opt;
659}
660EXPORT_SYMBOL(blk_limits_io_opt);
661
662/**
663 * blk_queue_io_opt - set optimal request size for the queue
664 * @q: the request queue for the device
665 * @opt: optimal request size in bytes
666 *
667 * Description:
668 * Storage devices may report an optimal I/O size, which is the
669 * device's preferred unit for sustained I/O. This is rarely reported
670 * for disk drives. For RAID arrays it is usually the stripe width or
671 * the internal track size. A properly aligned multiple of
672 * optimal_io_size is the preferred request size for workloads where
673 * sustained throughput is desired.
674 */
675void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
676{
677 blk_limits_io_opt(&q->limits, opt);
678 if (!q->disk)
679 return;
680 q->disk->bdi->ra_pages =
681 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
682}
683EXPORT_SYMBOL(blk_queue_io_opt);
684
685static int queue_limit_alignment_offset(const struct queue_limits *lim,
686 sector_t sector)
687{
688 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
689 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
690 << SECTOR_SHIFT;
691
692 return (granularity + lim->alignment_offset - alignment) % granularity;
693}
694
695static unsigned int queue_limit_discard_alignment(
696 const struct queue_limits *lim, sector_t sector)
697{
698 unsigned int alignment, granularity, offset;
699
700 if (!lim->max_discard_sectors)
701 return 0;
702
703 /* Why are these in bytes, not sectors? */
704 alignment = lim->discard_alignment >> SECTOR_SHIFT;
705 granularity = lim->discard_granularity >> SECTOR_SHIFT;
706 if (!granularity)
707 return 0;
708
709 /* Offset of the partition start in 'granularity' sectors */
710 offset = sector_div(sector, granularity);
711
712 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
713 offset = (granularity + alignment - offset) % granularity;
714
715 /* Turn it back into bytes, gaah */
716 return offset << SECTOR_SHIFT;
717}
718
719static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
720{
721 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
722 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
723 sectors = PAGE_SIZE >> SECTOR_SHIFT;
724 return sectors;
725}
726
727/**
728 * blk_stack_limits - adjust queue_limits for stacked devices
729 * @t: the stacking driver limits (top device)
730 * @b: the underlying queue limits (bottom, component device)
731 * @start: first data sector within component device
732 *
733 * Description:
734 * This function is used by stacking drivers like MD and DM to ensure
735 * that all component devices have compatible block sizes and
736 * alignments. The stacking driver must provide a queue_limits
737 * struct (top) and then iteratively call the stacking function for
738 * all component (bottom) devices. The stacking function will
739 * attempt to combine the values and ensure proper alignment.
740 *
741 * Returns 0 if the top and bottom queue_limits are compatible. The
742 * top device's block sizes and alignment offsets may be adjusted to
743 * ensure alignment with the bottom device. If no compatible sizes
744 * and alignments exist, -1 is returned and the resulting top
745 * queue_limits will have the misaligned flag set to indicate that
746 * the alignment_offset is undefined.
747 */
748int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
749 sector_t start)
750{
751 unsigned int top, bottom, alignment, ret = 0;
752
753 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
754 t->max_user_sectors = min_not_zero(t->max_user_sectors,
755 b->max_user_sectors);
756 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
757 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
758 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
759 b->max_write_zeroes_sectors);
760 t->max_zone_append_sectors = min(t->max_zone_append_sectors,
761 b->max_zone_append_sectors);
762 t->bounce = max(t->bounce, b->bounce);
763
764 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
765 b->seg_boundary_mask);
766 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
767 b->virt_boundary_mask);
768
769 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
770 t->max_discard_segments = min_not_zero(t->max_discard_segments,
771 b->max_discard_segments);
772 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
773 b->max_integrity_segments);
774
775 t->max_segment_size = min_not_zero(t->max_segment_size,
776 b->max_segment_size);
777
778 t->misaligned |= b->misaligned;
779
780 alignment = queue_limit_alignment_offset(b, start);
781
782 /* Bottom device has different alignment. Check that it is
783 * compatible with the current top alignment.
784 */
785 if (t->alignment_offset != alignment) {
786
787 top = max(t->physical_block_size, t->io_min)
788 + t->alignment_offset;
789 bottom = max(b->physical_block_size, b->io_min) + alignment;
790
791 /* Verify that top and bottom intervals line up */
792 if (max(top, bottom) % min(top, bottom)) {
793 t->misaligned = 1;
794 ret = -1;
795 }
796 }
797
798 t->logical_block_size = max(t->logical_block_size,
799 b->logical_block_size);
800
801 t->physical_block_size = max(t->physical_block_size,
802 b->physical_block_size);
803
804 t->io_min = max(t->io_min, b->io_min);
805 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
806 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
807
808 /* Set non-power-of-2 compatible chunk_sectors boundary */
809 if (b->chunk_sectors)
810 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
811
812 /* Physical block size a multiple of the logical block size? */
813 if (t->physical_block_size & (t->logical_block_size - 1)) {
814 t->physical_block_size = t->logical_block_size;
815 t->misaligned = 1;
816 ret = -1;
817 }
818
819 /* Minimum I/O a multiple of the physical block size? */
820 if (t->io_min & (t->physical_block_size - 1)) {
821 t->io_min = t->physical_block_size;
822 t->misaligned = 1;
823 ret = -1;
824 }
825
826 /* Optimal I/O a multiple of the physical block size? */
827 if (t->io_opt & (t->physical_block_size - 1)) {
828 t->io_opt = 0;
829 t->misaligned = 1;
830 ret = -1;
831 }
832
833 /* chunk_sectors a multiple of the physical block size? */
834 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
835 t->chunk_sectors = 0;
836 t->misaligned = 1;
837 ret = -1;
838 }
839
840 t->raid_partial_stripes_expensive =
841 max(t->raid_partial_stripes_expensive,
842 b->raid_partial_stripes_expensive);
843
844 /* Find lowest common alignment_offset */
845 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
846 % max(t->physical_block_size, t->io_min);
847
848 /* Verify that new alignment_offset is on a logical block boundary */
849 if (t->alignment_offset & (t->logical_block_size - 1)) {
850 t->misaligned = 1;
851 ret = -1;
852 }
853
854 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
855 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
856 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
857
858 /* Discard alignment and granularity */
859 if (b->discard_granularity) {
860 alignment = queue_limit_discard_alignment(b, start);
861
862 if (t->discard_granularity != 0 &&
863 t->discard_alignment != alignment) {
864 top = t->discard_granularity + t->discard_alignment;
865 bottom = b->discard_granularity + alignment;
866
867 /* Verify that top and bottom intervals line up */
868 if ((max(top, bottom) % min(top, bottom)) != 0)
869 t->discard_misaligned = 1;
870 }
871
872 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
873 b->max_discard_sectors);
874 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
875 b->max_hw_discard_sectors);
876 t->discard_granularity = max(t->discard_granularity,
877 b->discard_granularity);
878 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
879 t->discard_granularity;
880 }
881 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
882 b->max_secure_erase_sectors);
883 t->zone_write_granularity = max(t->zone_write_granularity,
884 b->zone_write_granularity);
885 t->zoned = max(t->zoned, b->zoned);
886 if (!t->zoned) {
887 t->zone_write_granularity = 0;
888 t->max_zone_append_sectors = 0;
889 }
890 return ret;
891}
892EXPORT_SYMBOL(blk_stack_limits);
893
894/**
895 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
896 * @t: the stacking driver limits (top device)
897 * @bdev: the underlying block device (bottom)
898 * @offset: offset to beginning of data within component device
899 * @pfx: prefix to use for warnings logged
900 *
901 * Description:
902 * This function is used by stacking drivers like MD and DM to ensure
903 * that all component devices have compatible block sizes and
904 * alignments. The stacking driver must provide a queue_limits
905 * struct (top) and then iteratively call the stacking function for
906 * all component (bottom) devices. The stacking function will
907 * attempt to combine the values and ensure proper alignment.
908 */
909void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
910 sector_t offset, const char *pfx)
911{
912 if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits,
913 get_start_sect(bdev) + offset))
914 pr_notice("%s: Warning: Device %pg is misaligned\n",
915 pfx, bdev);
916}
917EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
918
919/**
920 * blk_queue_update_dma_pad - update pad mask
921 * @q: the request queue for the device
922 * @mask: pad mask
923 *
924 * Update dma pad mask.
925 *
926 * Appending pad buffer to a request modifies the last entry of a
927 * scatter list such that it includes the pad buffer.
928 **/
929void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
930{
931 if (mask > q->dma_pad_mask)
932 q->dma_pad_mask = mask;
933}
934EXPORT_SYMBOL(blk_queue_update_dma_pad);
935
936/**
937 * blk_queue_segment_boundary - set boundary rules for segment merging
938 * @q: the request queue for the device
939 * @mask: the memory boundary mask
940 **/
941void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
942{
943 if (mask < PAGE_SIZE - 1) {
944 mask = PAGE_SIZE - 1;
945 pr_info("%s: set to minimum %lx\n", __func__, mask);
946 }
947
948 q->limits.seg_boundary_mask = mask;
949}
950EXPORT_SYMBOL(blk_queue_segment_boundary);
951
952/**
953 * blk_queue_virt_boundary - set boundary rules for bio merging
954 * @q: the request queue for the device
955 * @mask: the memory boundary mask
956 **/
957void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
958{
959 q->limits.virt_boundary_mask = mask;
960
961 /*
962 * Devices that require a virtual boundary do not support scatter/gather
963 * I/O natively, but instead require a descriptor list entry for each
964 * page (which might not be idential to the Linux PAGE_SIZE). Because
965 * of that they are not limited by our notion of "segment size".
966 */
967 if (mask)
968 q->limits.max_segment_size = UINT_MAX;
969}
970EXPORT_SYMBOL(blk_queue_virt_boundary);
971
972/**
973 * blk_queue_dma_alignment - set dma length and memory alignment
974 * @q: the request queue for the device
975 * @mask: alignment mask
976 *
977 * description:
978 * set required memory and length alignment for direct dma transactions.
979 * this is used when building direct io requests for the queue.
980 *
981 **/
982void blk_queue_dma_alignment(struct request_queue *q, int mask)
983{
984 q->limits.dma_alignment = mask;
985}
986EXPORT_SYMBOL(blk_queue_dma_alignment);
987
988/**
989 * blk_queue_update_dma_alignment - update dma length and memory alignment
990 * @q: the request queue for the device
991 * @mask: alignment mask
992 *
993 * description:
994 * update required memory and length alignment for direct dma transactions.
995 * If the requested alignment is larger than the current alignment, then
996 * the current queue alignment is updated to the new value, otherwise it
997 * is left alone. The design of this is to allow multiple objects
998 * (driver, device, transport etc) to set their respective
999 * alignments without having them interfere.
1000 *
1001 **/
1002void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
1003{
1004 BUG_ON(mask > PAGE_SIZE);
1005
1006 if (mask > q->limits.dma_alignment)
1007 q->limits.dma_alignment = mask;
1008}
1009EXPORT_SYMBOL(blk_queue_update_dma_alignment);
1010
1011/**
1012 * blk_set_queue_depth - tell the block layer about the device queue depth
1013 * @q: the request queue for the device
1014 * @depth: queue depth
1015 *
1016 */
1017void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
1018{
1019 q->queue_depth = depth;
1020 rq_qos_queue_depth_changed(q);
1021}
1022EXPORT_SYMBOL(blk_set_queue_depth);
1023
1024/**
1025 * blk_queue_write_cache - configure queue's write cache
1026 * @q: the request queue for the device
1027 * @wc: write back cache on or off
1028 * @fua: device supports FUA writes, if true
1029 *
1030 * Tell the block layer about the write cache of @q.
1031 */
1032void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
1033{
1034 if (wc) {
1035 blk_queue_flag_set(QUEUE_FLAG_HW_WC, q);
1036 blk_queue_flag_set(QUEUE_FLAG_WC, q);
1037 } else {
1038 blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q);
1039 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
1040 }
1041 if (fua)
1042 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
1043 else
1044 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
1045}
1046EXPORT_SYMBOL_GPL(blk_queue_write_cache);
1047
1048/**
1049 * blk_queue_required_elevator_features - Set a queue required elevator features
1050 * @q: the request queue for the target device
1051 * @features: Required elevator features OR'ed together
1052 *
1053 * Tell the block layer that for the device controlled through @q, only the
1054 * only elevators that can be used are those that implement at least the set of
1055 * features specified by @features.
1056 */
1057void blk_queue_required_elevator_features(struct request_queue *q,
1058 unsigned int features)
1059{
1060 q->required_elevator_features = features;
1061}
1062EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
1063
1064/**
1065 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
1066 * @q: the request queue for the device
1067 * @dev: the device pointer for dma
1068 *
1069 * Tell the block layer about merging the segments by dma map of @q.
1070 */
1071bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
1072 struct device *dev)
1073{
1074 unsigned long boundary = dma_get_merge_boundary(dev);
1075
1076 if (!boundary)
1077 return false;
1078
1079 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
1080 blk_queue_virt_boundary(q, boundary);
1081
1082 return true;
1083}
1084EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
1085
1086/**
1087 * disk_set_zoned - inidicate a zoned device
1088 * @disk: gendisk to configure
1089 */
1090void disk_set_zoned(struct gendisk *disk)
1091{
1092 struct request_queue *q = disk->queue;
1093
1094 WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
1095
1096 /*
1097 * Set the zone write granularity to the device logical block
1098 * size by default. The driver can change this value if needed.
1099 */
1100 q->limits.zoned = true;
1101 blk_queue_zone_write_granularity(q, queue_logical_block_size(q));
1102}
1103EXPORT_SYMBOL_GPL(disk_set_zoned);
1104
1105int bdev_alignment_offset(struct block_device *bdev)
1106{
1107 struct request_queue *q = bdev_get_queue(bdev);
1108
1109 if (q->limits.misaligned)
1110 return -1;
1111 if (bdev_is_partition(bdev))
1112 return queue_limit_alignment_offset(&q->limits,
1113 bdev->bd_start_sect);
1114 return q->limits.alignment_offset;
1115}
1116EXPORT_SYMBOL_GPL(bdev_alignment_offset);
1117
1118unsigned int bdev_discard_alignment(struct block_device *bdev)
1119{
1120 struct request_queue *q = bdev_get_queue(bdev);
1121
1122 if (bdev_is_partition(bdev))
1123 return queue_limit_discard_alignment(&q->limits,
1124 bdev->bd_start_sect);
1125 return q->limits.discard_alignment;
1126}
1127EXPORT_SYMBOL_GPL(bdev_discard_alignment);