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