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