Loading...
1/*
2 * Functions related to setting various queue properties from drivers
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/init.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
10#include <linux/gcd.h>
11#include <linux/lcm.h>
12#include <linux/jiffies.h>
13#include <linux/gfp.h>
14
15#include "blk.h"
16
17unsigned long blk_max_low_pfn;
18EXPORT_SYMBOL(blk_max_low_pfn);
19
20unsigned long blk_max_pfn;
21
22/**
23 * blk_queue_prep_rq - set a prepare_request function for queue
24 * @q: queue
25 * @pfn: prepare_request function
26 *
27 * It's possible for a queue to register a prepare_request callback which
28 * is invoked before the request is handed to the request_fn. The goal of
29 * the function is to prepare a request for I/O, it can be used to build a
30 * cdb from the request data for instance.
31 *
32 */
33void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
34{
35 q->prep_rq_fn = pfn;
36}
37EXPORT_SYMBOL(blk_queue_prep_rq);
38
39/**
40 * blk_queue_unprep_rq - set an unprepare_request function for queue
41 * @q: queue
42 * @ufn: unprepare_request function
43 *
44 * It's possible for a queue to register an unprepare_request callback
45 * which is invoked before the request is finally completed. The goal
46 * of the function is to deallocate any data that was allocated in the
47 * prepare_request callback.
48 *
49 */
50void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
51{
52 q->unprep_rq_fn = ufn;
53}
54EXPORT_SYMBOL(blk_queue_unprep_rq);
55
56/**
57 * blk_queue_merge_bvec - set a merge_bvec function for queue
58 * @q: queue
59 * @mbfn: merge_bvec_fn
60 *
61 * Usually queues have static limitations on the max sectors or segments that
62 * we can put in a request. Stacking drivers may have some settings that
63 * are dynamic, and thus we have to query the queue whether it is ok to
64 * add a new bio_vec to a bio at a given offset or not. If the block device
65 * has such limitations, it needs to register a merge_bvec_fn to control
66 * the size of bio's sent to it. Note that a block device *must* allow a
67 * single page to be added to an empty bio. The block device driver may want
68 * to use the bio_split() function to deal with these bio's. By default
69 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
70 * honored.
71 */
72void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
73{
74 q->merge_bvec_fn = mbfn;
75}
76EXPORT_SYMBOL(blk_queue_merge_bvec);
77
78void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
79{
80 q->softirq_done_fn = fn;
81}
82EXPORT_SYMBOL(blk_queue_softirq_done);
83
84void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
85{
86 q->rq_timeout = timeout;
87}
88EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
89
90void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
91{
92 q->rq_timed_out_fn = fn;
93}
94EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
95
96void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
97{
98 q->lld_busy_fn = fn;
99}
100EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
101
102/**
103 * blk_set_default_limits - reset limits to default values
104 * @lim: the queue_limits structure to reset
105 *
106 * Description:
107 * Returns a queue_limit struct to its default state. Can be used by
108 * stacking drivers like DM that stage table swaps and reuse an
109 * existing device queue.
110 */
111void blk_set_default_limits(struct queue_limits *lim)
112{
113 lim->max_segments = BLK_MAX_SEGMENTS;
114 lim->max_integrity_segments = 0;
115 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
116 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
117 lim->max_sectors = BLK_DEF_MAX_SECTORS;
118 lim->max_hw_sectors = INT_MAX;
119 lim->max_discard_sectors = 0;
120 lim->discard_granularity = 0;
121 lim->discard_alignment = 0;
122 lim->discard_misaligned = 0;
123 lim->discard_zeroes_data = 1;
124 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
125 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
126 lim->alignment_offset = 0;
127 lim->io_opt = 0;
128 lim->misaligned = 0;
129 lim->cluster = 1;
130}
131EXPORT_SYMBOL(blk_set_default_limits);
132
133/**
134 * blk_queue_make_request - define an alternate make_request function for a device
135 * @q: the request queue for the device to be affected
136 * @mfn: the alternate make_request function
137 *
138 * Description:
139 * The normal way for &struct bios to be passed to a device
140 * driver is for them to be collected into requests on a request
141 * queue, and then to allow the device driver to select requests
142 * off that queue when it is ready. This works well for many block
143 * devices. However some block devices (typically virtual devices
144 * such as md or lvm) do not benefit from the processing on the
145 * request queue, and are served best by having the requests passed
146 * directly to them. This can be achieved by providing a function
147 * to blk_queue_make_request().
148 *
149 * Caveat:
150 * The driver that does this *must* be able to deal appropriately
151 * with buffers in "highmemory". This can be accomplished by either calling
152 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
153 * blk_queue_bounce() to create a buffer in normal memory.
154 **/
155void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
156{
157 /*
158 * set defaults
159 */
160 q->nr_requests = BLKDEV_MAX_RQ;
161
162 q->make_request_fn = mfn;
163 blk_queue_dma_alignment(q, 511);
164 blk_queue_congestion_threshold(q);
165 q->nr_batching = BLK_BATCH_REQ;
166
167 blk_set_default_limits(&q->limits);
168 blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS);
169 q->limits.discard_zeroes_data = 0;
170
171 /*
172 * by default assume old behaviour and bounce for any highmem page
173 */
174 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
175}
176EXPORT_SYMBOL(blk_queue_make_request);
177
178/**
179 * blk_queue_bounce_limit - set bounce buffer limit for queue
180 * @q: the request queue for the device
181 * @dma_mask: the maximum address the device can handle
182 *
183 * Description:
184 * Different hardware can have different requirements as to what pages
185 * it can do I/O directly to. A low level driver can call
186 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
187 * buffers for doing I/O to pages residing above @dma_mask.
188 **/
189void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
190{
191 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
192 int dma = 0;
193
194 q->bounce_gfp = GFP_NOIO;
195#if BITS_PER_LONG == 64
196 /*
197 * Assume anything <= 4GB can be handled by IOMMU. Actually
198 * some IOMMUs can handle everything, but I don't know of a
199 * way to test this here.
200 */
201 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
202 dma = 1;
203 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
204#else
205 if (b_pfn < blk_max_low_pfn)
206 dma = 1;
207 q->limits.bounce_pfn = b_pfn;
208#endif
209 if (dma) {
210 init_emergency_isa_pool();
211 q->bounce_gfp = GFP_NOIO | GFP_DMA;
212 q->limits.bounce_pfn = b_pfn;
213 }
214}
215EXPORT_SYMBOL(blk_queue_bounce_limit);
216
217/**
218 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
219 * @limits: the queue limits
220 * @max_hw_sectors: max hardware sectors in the usual 512b unit
221 *
222 * Description:
223 * Enables a low level driver to set a hard upper limit,
224 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
225 * the device driver based upon the combined capabilities of I/O
226 * controller and storage device.
227 *
228 * max_sectors is a soft limit imposed by the block layer for
229 * filesystem type requests. This value can be overridden on a
230 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
231 * The soft limit can not exceed max_hw_sectors.
232 **/
233void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
234{
235 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
236 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
237 printk(KERN_INFO "%s: set to minimum %d\n",
238 __func__, max_hw_sectors);
239 }
240
241 limits->max_hw_sectors = max_hw_sectors;
242 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
243 BLK_DEF_MAX_SECTORS);
244}
245EXPORT_SYMBOL(blk_limits_max_hw_sectors);
246
247/**
248 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
249 * @q: the request queue for the device
250 * @max_hw_sectors: max hardware sectors in the usual 512b unit
251 *
252 * Description:
253 * See description for blk_limits_max_hw_sectors().
254 **/
255void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
256{
257 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
258}
259EXPORT_SYMBOL(blk_queue_max_hw_sectors);
260
261/**
262 * blk_queue_max_discard_sectors - set max sectors for a single discard
263 * @q: the request queue for the device
264 * @max_discard_sectors: maximum number of sectors to discard
265 **/
266void blk_queue_max_discard_sectors(struct request_queue *q,
267 unsigned int max_discard_sectors)
268{
269 q->limits.max_discard_sectors = max_discard_sectors;
270}
271EXPORT_SYMBOL(blk_queue_max_discard_sectors);
272
273/**
274 * blk_queue_max_segments - set max hw segments for a request for this queue
275 * @q: the request queue for the device
276 * @max_segments: max number of segments
277 *
278 * Description:
279 * Enables a low level driver to set an upper limit on the number of
280 * hw data segments in a request.
281 **/
282void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
283{
284 if (!max_segments) {
285 max_segments = 1;
286 printk(KERN_INFO "%s: set to minimum %d\n",
287 __func__, max_segments);
288 }
289
290 q->limits.max_segments = max_segments;
291}
292EXPORT_SYMBOL(blk_queue_max_segments);
293
294/**
295 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
296 * @q: the request queue for the device
297 * @max_size: max size of segment in bytes
298 *
299 * Description:
300 * Enables a low level driver to set an upper limit on the size of a
301 * coalesced segment
302 **/
303void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
304{
305 if (max_size < PAGE_CACHE_SIZE) {
306 max_size = PAGE_CACHE_SIZE;
307 printk(KERN_INFO "%s: set to minimum %d\n",
308 __func__, max_size);
309 }
310
311 q->limits.max_segment_size = max_size;
312}
313EXPORT_SYMBOL(blk_queue_max_segment_size);
314
315/**
316 * blk_queue_logical_block_size - set logical block size for the queue
317 * @q: the request queue for the device
318 * @size: the logical block size, in bytes
319 *
320 * Description:
321 * This should be set to the lowest possible block size that the
322 * storage device can address. The default of 512 covers most
323 * hardware.
324 **/
325void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
326{
327 q->limits.logical_block_size = size;
328
329 if (q->limits.physical_block_size < size)
330 q->limits.physical_block_size = size;
331
332 if (q->limits.io_min < q->limits.physical_block_size)
333 q->limits.io_min = q->limits.physical_block_size;
334}
335EXPORT_SYMBOL(blk_queue_logical_block_size);
336
337/**
338 * blk_queue_physical_block_size - set physical block size for the queue
339 * @q: the request queue for the device
340 * @size: the physical block size, in bytes
341 *
342 * Description:
343 * This should be set to the lowest possible sector size that the
344 * hardware can operate on without reverting to read-modify-write
345 * operations.
346 */
347void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
348{
349 q->limits.physical_block_size = size;
350
351 if (q->limits.physical_block_size < q->limits.logical_block_size)
352 q->limits.physical_block_size = q->limits.logical_block_size;
353
354 if (q->limits.io_min < q->limits.physical_block_size)
355 q->limits.io_min = q->limits.physical_block_size;
356}
357EXPORT_SYMBOL(blk_queue_physical_block_size);
358
359/**
360 * blk_queue_alignment_offset - set physical block alignment offset
361 * @q: the request queue for the device
362 * @offset: alignment offset in bytes
363 *
364 * Description:
365 * Some devices are naturally misaligned to compensate for things like
366 * the legacy DOS partition table 63-sector offset. Low-level drivers
367 * should call this function for devices whose first sector is not
368 * naturally aligned.
369 */
370void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
371{
372 q->limits.alignment_offset =
373 offset & (q->limits.physical_block_size - 1);
374 q->limits.misaligned = 0;
375}
376EXPORT_SYMBOL(blk_queue_alignment_offset);
377
378/**
379 * blk_limits_io_min - set minimum request size for a device
380 * @limits: the queue limits
381 * @min: smallest I/O size in bytes
382 *
383 * Description:
384 * Some devices have an internal block size bigger than the reported
385 * hardware sector size. This function can be used to signal the
386 * smallest I/O the device can perform without incurring a performance
387 * penalty.
388 */
389void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
390{
391 limits->io_min = min;
392
393 if (limits->io_min < limits->logical_block_size)
394 limits->io_min = limits->logical_block_size;
395
396 if (limits->io_min < limits->physical_block_size)
397 limits->io_min = limits->physical_block_size;
398}
399EXPORT_SYMBOL(blk_limits_io_min);
400
401/**
402 * blk_queue_io_min - set minimum request size for the queue
403 * @q: the request queue for the device
404 * @min: smallest I/O size in bytes
405 *
406 * Description:
407 * Storage devices may report a granularity or preferred minimum I/O
408 * size which is the smallest request the device can perform without
409 * incurring a performance penalty. For disk drives this is often the
410 * physical block size. For RAID arrays it is often the stripe chunk
411 * size. A properly aligned multiple of minimum_io_size is the
412 * preferred request size for workloads where a high number of I/O
413 * operations is desired.
414 */
415void blk_queue_io_min(struct request_queue *q, unsigned int min)
416{
417 blk_limits_io_min(&q->limits, min);
418}
419EXPORT_SYMBOL(blk_queue_io_min);
420
421/**
422 * blk_limits_io_opt - set optimal request size for a device
423 * @limits: the queue limits
424 * @opt: smallest I/O size in bytes
425 *
426 * Description:
427 * Storage devices may report an optimal I/O size, which is the
428 * device's preferred unit for sustained I/O. This is rarely reported
429 * for disk drives. For RAID arrays it is usually the stripe width or
430 * the internal track size. A properly aligned multiple of
431 * optimal_io_size is the preferred request size for workloads where
432 * sustained throughput is desired.
433 */
434void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
435{
436 limits->io_opt = opt;
437}
438EXPORT_SYMBOL(blk_limits_io_opt);
439
440/**
441 * blk_queue_io_opt - set optimal request size for the queue
442 * @q: the request queue for the device
443 * @opt: optimal request size in bytes
444 *
445 * Description:
446 * Storage devices may report an optimal I/O size, which is the
447 * device's preferred unit for sustained I/O. This is rarely reported
448 * for disk drives. For RAID arrays it is usually the stripe width or
449 * the internal track size. A properly aligned multiple of
450 * optimal_io_size is the preferred request size for workloads where
451 * sustained throughput is desired.
452 */
453void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
454{
455 blk_limits_io_opt(&q->limits, opt);
456}
457EXPORT_SYMBOL(blk_queue_io_opt);
458
459/**
460 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
461 * @t: the stacking driver (top)
462 * @b: the underlying device (bottom)
463 **/
464void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
465{
466 blk_stack_limits(&t->limits, &b->limits, 0);
467}
468EXPORT_SYMBOL(blk_queue_stack_limits);
469
470/**
471 * blk_stack_limits - adjust queue_limits for stacked devices
472 * @t: the stacking driver limits (top device)
473 * @b: the underlying queue limits (bottom, component device)
474 * @start: first data sector within component device
475 *
476 * Description:
477 * This function is used by stacking drivers like MD and DM to ensure
478 * that all component devices have compatible block sizes and
479 * alignments. The stacking driver must provide a queue_limits
480 * struct (top) and then iteratively call the stacking function for
481 * all component (bottom) devices. The stacking function will
482 * attempt to combine the values and ensure proper alignment.
483 *
484 * Returns 0 if the top and bottom queue_limits are compatible. The
485 * top device's block sizes and alignment offsets may be adjusted to
486 * ensure alignment with the bottom device. If no compatible sizes
487 * and alignments exist, -1 is returned and the resulting top
488 * queue_limits will have the misaligned flag set to indicate that
489 * the alignment_offset is undefined.
490 */
491int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
492 sector_t start)
493{
494 unsigned int top, bottom, alignment, ret = 0;
495
496 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
497 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
498 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
499
500 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
501 b->seg_boundary_mask);
502
503 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
504 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
505 b->max_integrity_segments);
506
507 t->max_segment_size = min_not_zero(t->max_segment_size,
508 b->max_segment_size);
509
510 t->misaligned |= b->misaligned;
511
512 alignment = queue_limit_alignment_offset(b, start);
513
514 /* Bottom device has different alignment. Check that it is
515 * compatible with the current top alignment.
516 */
517 if (t->alignment_offset != alignment) {
518
519 top = max(t->physical_block_size, t->io_min)
520 + t->alignment_offset;
521 bottom = max(b->physical_block_size, b->io_min) + alignment;
522
523 /* Verify that top and bottom intervals line up */
524 if (max(top, bottom) & (min(top, bottom) - 1)) {
525 t->misaligned = 1;
526 ret = -1;
527 }
528 }
529
530 t->logical_block_size = max(t->logical_block_size,
531 b->logical_block_size);
532
533 t->physical_block_size = max(t->physical_block_size,
534 b->physical_block_size);
535
536 t->io_min = max(t->io_min, b->io_min);
537 t->io_opt = lcm(t->io_opt, b->io_opt);
538
539 t->cluster &= b->cluster;
540 t->discard_zeroes_data &= b->discard_zeroes_data;
541
542 /* Physical block size a multiple of the logical block size? */
543 if (t->physical_block_size & (t->logical_block_size - 1)) {
544 t->physical_block_size = t->logical_block_size;
545 t->misaligned = 1;
546 ret = -1;
547 }
548
549 /* Minimum I/O a multiple of the physical block size? */
550 if (t->io_min & (t->physical_block_size - 1)) {
551 t->io_min = t->physical_block_size;
552 t->misaligned = 1;
553 ret = -1;
554 }
555
556 /* Optimal I/O a multiple of the physical block size? */
557 if (t->io_opt & (t->physical_block_size - 1)) {
558 t->io_opt = 0;
559 t->misaligned = 1;
560 ret = -1;
561 }
562
563 /* Find lowest common alignment_offset */
564 t->alignment_offset = lcm(t->alignment_offset, alignment)
565 & (max(t->physical_block_size, t->io_min) - 1);
566
567 /* Verify that new alignment_offset is on a logical block boundary */
568 if (t->alignment_offset & (t->logical_block_size - 1)) {
569 t->misaligned = 1;
570 ret = -1;
571 }
572
573 /* Discard alignment and granularity */
574 if (b->discard_granularity) {
575 alignment = queue_limit_discard_alignment(b, start);
576
577 if (t->discard_granularity != 0 &&
578 t->discard_alignment != alignment) {
579 top = t->discard_granularity + t->discard_alignment;
580 bottom = b->discard_granularity + alignment;
581
582 /* Verify that top and bottom intervals line up */
583 if (max(top, bottom) & (min(top, bottom) - 1))
584 t->discard_misaligned = 1;
585 }
586
587 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
588 b->max_discard_sectors);
589 t->discard_granularity = max(t->discard_granularity,
590 b->discard_granularity);
591 t->discard_alignment = lcm(t->discard_alignment, alignment) &
592 (t->discard_granularity - 1);
593 }
594
595 return ret;
596}
597EXPORT_SYMBOL(blk_stack_limits);
598
599/**
600 * bdev_stack_limits - adjust queue limits for stacked drivers
601 * @t: the stacking driver limits (top device)
602 * @bdev: the component block_device (bottom)
603 * @start: first data sector within component device
604 *
605 * Description:
606 * Merges queue limits for a top device and a block_device. Returns
607 * 0 if alignment didn't change. Returns -1 if adding the bottom
608 * device caused misalignment.
609 */
610int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
611 sector_t start)
612{
613 struct request_queue *bq = bdev_get_queue(bdev);
614
615 start += get_start_sect(bdev);
616
617 return blk_stack_limits(t, &bq->limits, start);
618}
619EXPORT_SYMBOL(bdev_stack_limits);
620
621/**
622 * disk_stack_limits - adjust queue limits for stacked drivers
623 * @disk: MD/DM gendisk (top)
624 * @bdev: the underlying block device (bottom)
625 * @offset: offset to beginning of data within component device
626 *
627 * Description:
628 * Merges the limits for a top level gendisk and a bottom level
629 * block_device.
630 */
631void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
632 sector_t offset)
633{
634 struct request_queue *t = disk->queue;
635
636 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
637 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
638
639 disk_name(disk, 0, top);
640 bdevname(bdev, bottom);
641
642 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
643 top, bottom);
644 }
645}
646EXPORT_SYMBOL(disk_stack_limits);
647
648/**
649 * blk_queue_dma_pad - set pad mask
650 * @q: the request queue for the device
651 * @mask: pad mask
652 *
653 * Set dma pad mask.
654 *
655 * Appending pad buffer to a request modifies the last entry of a
656 * scatter list such that it includes the pad buffer.
657 **/
658void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
659{
660 q->dma_pad_mask = mask;
661}
662EXPORT_SYMBOL(blk_queue_dma_pad);
663
664/**
665 * blk_queue_update_dma_pad - update pad mask
666 * @q: the request queue for the device
667 * @mask: pad mask
668 *
669 * Update dma pad mask.
670 *
671 * Appending pad buffer to a request modifies the last entry of a
672 * scatter list such that it includes the pad buffer.
673 **/
674void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
675{
676 if (mask > q->dma_pad_mask)
677 q->dma_pad_mask = mask;
678}
679EXPORT_SYMBOL(blk_queue_update_dma_pad);
680
681/**
682 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
683 * @q: the request queue for the device
684 * @dma_drain_needed: fn which returns non-zero if drain is necessary
685 * @buf: physically contiguous buffer
686 * @size: size of the buffer in bytes
687 *
688 * Some devices have excess DMA problems and can't simply discard (or
689 * zero fill) the unwanted piece of the transfer. They have to have a
690 * real area of memory to transfer it into. The use case for this is
691 * ATAPI devices in DMA mode. If the packet command causes a transfer
692 * bigger than the transfer size some HBAs will lock up if there
693 * aren't DMA elements to contain the excess transfer. What this API
694 * does is adjust the queue so that the buf is always appended
695 * silently to the scatterlist.
696 *
697 * Note: This routine adjusts max_hw_segments to make room for appending
698 * the drain buffer. If you call blk_queue_max_segments() after calling
699 * this routine, you must set the limit to one fewer than your device
700 * can support otherwise there won't be room for the drain buffer.
701 */
702int blk_queue_dma_drain(struct request_queue *q,
703 dma_drain_needed_fn *dma_drain_needed,
704 void *buf, unsigned int size)
705{
706 if (queue_max_segments(q) < 2)
707 return -EINVAL;
708 /* make room for appending the drain */
709 blk_queue_max_segments(q, queue_max_segments(q) - 1);
710 q->dma_drain_needed = dma_drain_needed;
711 q->dma_drain_buffer = buf;
712 q->dma_drain_size = size;
713
714 return 0;
715}
716EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
717
718/**
719 * blk_queue_segment_boundary - set boundary rules for segment merging
720 * @q: the request queue for the device
721 * @mask: the memory boundary mask
722 **/
723void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
724{
725 if (mask < PAGE_CACHE_SIZE - 1) {
726 mask = PAGE_CACHE_SIZE - 1;
727 printk(KERN_INFO "%s: set to minimum %lx\n",
728 __func__, mask);
729 }
730
731 q->limits.seg_boundary_mask = mask;
732}
733EXPORT_SYMBOL(blk_queue_segment_boundary);
734
735/**
736 * blk_queue_dma_alignment - set dma length and memory alignment
737 * @q: the request queue for the device
738 * @mask: alignment mask
739 *
740 * description:
741 * set required memory and length alignment for direct dma transactions.
742 * this is used when building direct io requests for the queue.
743 *
744 **/
745void blk_queue_dma_alignment(struct request_queue *q, int mask)
746{
747 q->dma_alignment = mask;
748}
749EXPORT_SYMBOL(blk_queue_dma_alignment);
750
751/**
752 * blk_queue_update_dma_alignment - update dma length and memory alignment
753 * @q: the request queue for the device
754 * @mask: alignment mask
755 *
756 * description:
757 * update required memory and length alignment for direct dma transactions.
758 * If the requested alignment is larger than the current alignment, then
759 * the current queue alignment is updated to the new value, otherwise it
760 * is left alone. The design of this is to allow multiple objects
761 * (driver, device, transport etc) to set their respective
762 * alignments without having them interfere.
763 *
764 **/
765void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
766{
767 BUG_ON(mask > PAGE_SIZE);
768
769 if (mask > q->dma_alignment)
770 q->dma_alignment = mask;
771}
772EXPORT_SYMBOL(blk_queue_update_dma_alignment);
773
774/**
775 * blk_queue_flush - configure queue's cache flush capability
776 * @q: the request queue for the device
777 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
778 *
779 * Tell block layer cache flush capability of @q. If it supports
780 * flushing, REQ_FLUSH should be set. If it supports bypassing
781 * write cache for individual writes, REQ_FUA should be set.
782 */
783void blk_queue_flush(struct request_queue *q, unsigned int flush)
784{
785 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
786
787 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
788 flush &= ~REQ_FUA;
789
790 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
791}
792EXPORT_SYMBOL_GPL(blk_queue_flush);
793
794void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
795{
796 q->flush_not_queueable = !queueable;
797}
798EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
799
800static int __init blk_settings_init(void)
801{
802 blk_max_low_pfn = max_low_pfn - 1;
803 blk_max_pfn = max_pfn - 1;
804 return 0;
805}
806subsys_initcall(blk_settings_init);
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/blk-integrity.h>
10#include <linux/pagemap.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/gcd.h>
13#include <linux/lcm.h>
14#include <linux/jiffies.h>
15#include <linux/gfp.h>
16#include <linux/dma-mapping.h>
17
18#include "blk.h"
19#include "blk-rq-qos.h"
20#include "blk-wbt.h"
21
22void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
23{
24 q->rq_timeout = timeout;
25}
26EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
27
28/**
29 * blk_set_stacking_limits - set default limits for stacking devices
30 * @lim: the queue_limits structure to reset
31 *
32 * Prepare queue limits for applying limits from underlying devices using
33 * blk_stack_limits().
34 */
35void blk_set_stacking_limits(struct queue_limits *lim)
36{
37 memset(lim, 0, sizeof(*lim));
38 lim->logical_block_size = SECTOR_SIZE;
39 lim->physical_block_size = SECTOR_SIZE;
40 lim->io_min = SECTOR_SIZE;
41 lim->discard_granularity = SECTOR_SIZE;
42 lim->dma_alignment = SECTOR_SIZE - 1;
43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
44
45 /* Inherit limits from component devices */
46 lim->max_segments = USHRT_MAX;
47 lim->max_discard_segments = USHRT_MAX;
48 lim->max_hw_sectors = UINT_MAX;
49 lim->max_segment_size = UINT_MAX;
50 lim->max_sectors = UINT_MAX;
51 lim->max_dev_sectors = UINT_MAX;
52 lim->max_write_zeroes_sectors = UINT_MAX;
53 lim->max_hw_zone_append_sectors = UINT_MAX;
54 lim->max_user_discard_sectors = UINT_MAX;
55}
56EXPORT_SYMBOL(blk_set_stacking_limits);
57
58void blk_apply_bdi_limits(struct backing_dev_info *bdi,
59 struct queue_limits *lim)
60{
61 /*
62 * For read-ahead of large files to be effective, we need to read ahead
63 * at least twice the optimal I/O size.
64 */
65 bdi->ra_pages = max(lim->io_opt * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
66 bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT;
67}
68
69static int blk_validate_zoned_limits(struct queue_limits *lim)
70{
71 if (!(lim->features & BLK_FEAT_ZONED)) {
72 if (WARN_ON_ONCE(lim->max_open_zones) ||
73 WARN_ON_ONCE(lim->max_active_zones) ||
74 WARN_ON_ONCE(lim->zone_write_granularity) ||
75 WARN_ON_ONCE(lim->max_zone_append_sectors))
76 return -EINVAL;
77 return 0;
78 }
79
80 if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)))
81 return -EINVAL;
82
83 /*
84 * Given that active zones include open zones, the maximum number of
85 * open zones cannot be larger than the maximum number of active zones.
86 */
87 if (lim->max_active_zones &&
88 lim->max_open_zones > lim->max_active_zones)
89 return -EINVAL;
90
91 if (lim->zone_write_granularity < lim->logical_block_size)
92 lim->zone_write_granularity = lim->logical_block_size;
93
94 /*
95 * The Zone Append size is limited by the maximum I/O size and the zone
96 * size given that it can't span zones.
97 *
98 * If no max_hw_zone_append_sectors limit is provided, the block layer
99 * will emulated it, else we're also bound by the hardware limit.
100 */
101 lim->max_zone_append_sectors =
102 min_not_zero(lim->max_hw_zone_append_sectors,
103 min(lim->chunk_sectors, lim->max_hw_sectors));
104 return 0;
105}
106
107static int blk_validate_integrity_limits(struct queue_limits *lim)
108{
109 struct blk_integrity *bi = &lim->integrity;
110
111 if (!bi->tuple_size) {
112 if (bi->csum_type != BLK_INTEGRITY_CSUM_NONE ||
113 bi->tag_size || ((bi->flags & BLK_INTEGRITY_REF_TAG))) {
114 pr_warn("invalid PI settings.\n");
115 return -EINVAL;
116 }
117 return 0;
118 }
119
120 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) {
121 pr_warn("integrity support disabled.\n");
122 return -EINVAL;
123 }
124
125 if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE &&
126 (bi->flags & BLK_INTEGRITY_REF_TAG)) {
127 pr_warn("ref tag not support without checksum.\n");
128 return -EINVAL;
129 }
130
131 if (!bi->interval_exp)
132 bi->interval_exp = ilog2(lim->logical_block_size);
133
134 return 0;
135}
136
137/*
138 * Returns max guaranteed bytes which we can fit in a bio.
139 *
140 * We request that an atomic_write is ITER_UBUF iov_iter (so a single vector),
141 * so we assume that we can fit in at least PAGE_SIZE in a segment, apart from
142 * the first and last segments.
143 */
144static unsigned int blk_queue_max_guaranteed_bio(struct queue_limits *lim)
145{
146 unsigned int max_segments = min(BIO_MAX_VECS, lim->max_segments);
147 unsigned int length;
148
149 length = min(max_segments, 2) * lim->logical_block_size;
150 if (max_segments > 2)
151 length += (max_segments - 2) * PAGE_SIZE;
152
153 return length;
154}
155
156static void blk_atomic_writes_update_limits(struct queue_limits *lim)
157{
158 unsigned int unit_limit = min(lim->max_hw_sectors << SECTOR_SHIFT,
159 blk_queue_max_guaranteed_bio(lim));
160
161 unit_limit = rounddown_pow_of_two(unit_limit);
162
163 lim->atomic_write_max_sectors =
164 min(lim->atomic_write_hw_max >> SECTOR_SHIFT,
165 lim->max_hw_sectors);
166 lim->atomic_write_unit_min =
167 min(lim->atomic_write_hw_unit_min, unit_limit);
168 lim->atomic_write_unit_max =
169 min(lim->atomic_write_hw_unit_max, unit_limit);
170 lim->atomic_write_boundary_sectors =
171 lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
172}
173
174static void blk_validate_atomic_write_limits(struct queue_limits *lim)
175{
176 unsigned int boundary_sectors;
177
178 if (!lim->atomic_write_hw_max)
179 goto unsupported;
180
181 if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_min)))
182 goto unsupported;
183
184 if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_max)))
185 goto unsupported;
186
187 if (WARN_ON_ONCE(lim->atomic_write_hw_unit_min >
188 lim->atomic_write_hw_unit_max))
189 goto unsupported;
190
191 if (WARN_ON_ONCE(lim->atomic_write_hw_unit_max >
192 lim->atomic_write_hw_max))
193 goto unsupported;
194
195 boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
196
197 if (boundary_sectors) {
198 if (WARN_ON_ONCE(lim->atomic_write_hw_max >
199 lim->atomic_write_hw_boundary))
200 goto unsupported;
201 /*
202 * A feature of boundary support is that it disallows bios to
203 * be merged which would result in a merged request which
204 * crosses either a chunk sector or atomic write HW boundary,
205 * even though chunk sectors may be just set for performance.
206 * For simplicity, disallow atomic writes for a chunk sector
207 * which is non-zero and smaller than atomic write HW boundary.
208 * Furthermore, chunk sectors must be a multiple of atomic
209 * write HW boundary. Otherwise boundary support becomes
210 * complicated.
211 * Devices which do not conform to these rules can be dealt
212 * with if and when they show up.
213 */
214 if (WARN_ON_ONCE(lim->chunk_sectors % boundary_sectors))
215 goto unsupported;
216
217 /*
218 * The boundary size just needs to be a multiple of unit_max
219 * (and not necessarily a power-of-2), so this following check
220 * could be relaxed in future.
221 * Furthermore, if needed, unit_max could even be reduced so
222 * that it is compliant with a !power-of-2 boundary.
223 */
224 if (!is_power_of_2(boundary_sectors))
225 goto unsupported;
226 }
227
228 blk_atomic_writes_update_limits(lim);
229 return;
230
231unsupported:
232 lim->atomic_write_max_sectors = 0;
233 lim->atomic_write_boundary_sectors = 0;
234 lim->atomic_write_unit_min = 0;
235 lim->atomic_write_unit_max = 0;
236}
237
238/*
239 * Check that the limits in lim are valid, initialize defaults for unset
240 * values, and cap values based on others where needed.
241 */
242int blk_validate_limits(struct queue_limits *lim)
243{
244 unsigned int max_hw_sectors;
245 unsigned int logical_block_sectors;
246 int err;
247
248 /*
249 * Unless otherwise specified, default to 512 byte logical blocks and a
250 * physical block size equal to the logical block size.
251 */
252 if (!lim->logical_block_size)
253 lim->logical_block_size = SECTOR_SIZE;
254 else if (blk_validate_block_size(lim->logical_block_size)) {
255 pr_warn("Invalid logical block size (%d)\n", lim->logical_block_size);
256 return -EINVAL;
257 }
258 if (lim->physical_block_size < lim->logical_block_size)
259 lim->physical_block_size = lim->logical_block_size;
260
261 /*
262 * The minimum I/O size defaults to the physical block size unless
263 * explicitly overridden.
264 */
265 if (lim->io_min < lim->physical_block_size)
266 lim->io_min = lim->physical_block_size;
267
268 /*
269 * The optimal I/O size may not be aligned to physical block size
270 * (because it may be limited by dma engines which have no clue about
271 * block size of the disks attached to them), so we round it down here.
272 */
273 lim->io_opt = round_down(lim->io_opt, lim->physical_block_size);
274
275 /*
276 * max_hw_sectors has a somewhat weird default for historical reason,
277 * but driver really should set their own instead of relying on this
278 * value.
279 *
280 * The block layer relies on the fact that every driver can
281 * handle at lest a page worth of data per I/O, and needs the value
282 * aligned to the logical block size.
283 */
284 if (!lim->max_hw_sectors)
285 lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
286 if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
287 return -EINVAL;
288 logical_block_sectors = lim->logical_block_size >> SECTOR_SHIFT;
289 if (WARN_ON_ONCE(logical_block_sectors > lim->max_hw_sectors))
290 return -EINVAL;
291 lim->max_hw_sectors = round_down(lim->max_hw_sectors,
292 logical_block_sectors);
293
294 /*
295 * The actual max_sectors value is a complex beast and also takes the
296 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
297 * value into account. The ->max_sectors value is always calculated
298 * from these, so directly setting it won't have any effect.
299 */
300 max_hw_sectors = min_not_zero(lim->max_hw_sectors,
301 lim->max_dev_sectors);
302 if (lim->max_user_sectors) {
303 if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
304 return -EINVAL;
305 lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
306 } else if (lim->io_opt > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
307 lim->max_sectors =
308 min(max_hw_sectors, lim->io_opt >> SECTOR_SHIFT);
309 } else if (lim->io_min > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
310 lim->max_sectors =
311 min(max_hw_sectors, lim->io_min >> SECTOR_SHIFT);
312 } else {
313 lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
314 }
315 lim->max_sectors = round_down(lim->max_sectors,
316 logical_block_sectors);
317
318 /*
319 * Random default for the maximum number of segments. Driver should not
320 * rely on this and set their own.
321 */
322 if (!lim->max_segments)
323 lim->max_segments = BLK_MAX_SEGMENTS;
324
325 lim->max_discard_sectors =
326 min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
327
328 if (!lim->max_discard_segments)
329 lim->max_discard_segments = 1;
330
331 if (lim->discard_granularity < lim->physical_block_size)
332 lim->discard_granularity = lim->physical_block_size;
333
334 /*
335 * By default there is no limit on the segment boundary alignment,
336 * but if there is one it can't be smaller than the page size as
337 * that would break all the normal I/O patterns.
338 */
339 if (!lim->seg_boundary_mask)
340 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
341 if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
342 return -EINVAL;
343
344 /*
345 * Stacking device may have both virtual boundary and max segment
346 * size limit, so allow this setting now, and long-term the two
347 * might need to move out of stacking limits since we have immutable
348 * bvec and lower layer bio splitting is supposed to handle the two
349 * correctly.
350 */
351 if (lim->virt_boundary_mask) {
352 if (!lim->max_segment_size)
353 lim->max_segment_size = UINT_MAX;
354 } else {
355 /*
356 * The maximum segment size has an odd historic 64k default that
357 * drivers probably should override. Just like the I/O size we
358 * require drivers to at least handle a full page per segment.
359 */
360 if (!lim->max_segment_size)
361 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
362 if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
363 return -EINVAL;
364 }
365
366 /*
367 * We require drivers to at least do logical block aligned I/O, but
368 * historically could not check for that due to the separate calls
369 * to set the limits. Once the transition is finished the check
370 * below should be narrowed down to check the logical block size.
371 */
372 if (!lim->dma_alignment)
373 lim->dma_alignment = SECTOR_SIZE - 1;
374 if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
375 return -EINVAL;
376
377 if (lim->alignment_offset) {
378 lim->alignment_offset &= (lim->physical_block_size - 1);
379 lim->flags &= ~BLK_FLAG_MISALIGNED;
380 }
381
382 if (!(lim->features & BLK_FEAT_WRITE_CACHE))
383 lim->features &= ~BLK_FEAT_FUA;
384
385 blk_validate_atomic_write_limits(lim);
386
387 err = blk_validate_integrity_limits(lim);
388 if (err)
389 return err;
390 return blk_validate_zoned_limits(lim);
391}
392EXPORT_SYMBOL_GPL(blk_validate_limits);
393
394/*
395 * Set the default limits for a newly allocated queue. @lim contains the
396 * initial limits set by the driver, which could be no limit in which case
397 * all fields are cleared to zero.
398 */
399int blk_set_default_limits(struct queue_limits *lim)
400{
401 /*
402 * Most defaults are set by capping the bounds in blk_validate_limits,
403 * but max_user_discard_sectors is special and needs an explicit
404 * initialization to the max value here.
405 */
406 lim->max_user_discard_sectors = UINT_MAX;
407 return blk_validate_limits(lim);
408}
409
410/**
411 * queue_limits_commit_update - commit an atomic update of queue limits
412 * @q: queue to update
413 * @lim: limits to apply
414 *
415 * Apply the limits in @lim that were obtained from queue_limits_start_update()
416 * and updated by the caller to @q.
417 *
418 * Returns 0 if successful, else a negative error code.
419 */
420int queue_limits_commit_update(struct request_queue *q,
421 struct queue_limits *lim)
422{
423 int error;
424
425 error = blk_validate_limits(lim);
426 if (error)
427 goto out_unlock;
428
429#ifdef CONFIG_BLK_INLINE_ENCRYPTION
430 if (q->crypto_profile && lim->integrity.tag_size) {
431 pr_warn("blk-integrity: Integrity and hardware inline encryption are not supported together.\n");
432 error = -EINVAL;
433 goto out_unlock;
434 }
435#endif
436
437 q->limits = *lim;
438 if (q->disk)
439 blk_apply_bdi_limits(q->disk->bdi, lim);
440out_unlock:
441 mutex_unlock(&q->limits_lock);
442 return error;
443}
444EXPORT_SYMBOL_GPL(queue_limits_commit_update);
445
446/**
447 * queue_limits_commit_update_frozen - commit an atomic update of queue limits
448 * @q: queue to update
449 * @lim: limits to apply
450 *
451 * Apply the limits in @lim that were obtained from queue_limits_start_update()
452 * and updated with the new values by the caller to @q. Freezes the queue
453 * before the update and unfreezes it after.
454 *
455 * Returns 0 if successful, else a negative error code.
456 */
457int queue_limits_commit_update_frozen(struct request_queue *q,
458 struct queue_limits *lim)
459{
460 int ret;
461
462 blk_mq_freeze_queue(q);
463 ret = queue_limits_commit_update(q, lim);
464 blk_mq_unfreeze_queue(q);
465
466 return ret;
467}
468EXPORT_SYMBOL_GPL(queue_limits_commit_update_frozen);
469
470/**
471 * queue_limits_set - apply queue limits to queue
472 * @q: queue to update
473 * @lim: limits to apply
474 *
475 * Apply the limits in @lim that were freshly initialized to @q.
476 * To update existing limits use queue_limits_start_update() and
477 * queue_limits_commit_update() instead.
478 *
479 * Returns 0 if successful, else a negative error code.
480 */
481int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
482{
483 mutex_lock(&q->limits_lock);
484 return queue_limits_commit_update(q, lim);
485}
486EXPORT_SYMBOL_GPL(queue_limits_set);
487
488static int queue_limit_alignment_offset(const struct queue_limits *lim,
489 sector_t sector)
490{
491 unsigned int granularity = max(lim->physical_block_size, lim->io_min);
492 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
493 << SECTOR_SHIFT;
494
495 return (granularity + lim->alignment_offset - alignment) % granularity;
496}
497
498static unsigned int queue_limit_discard_alignment(
499 const struct queue_limits *lim, sector_t sector)
500{
501 unsigned int alignment, granularity, offset;
502
503 if (!lim->max_discard_sectors)
504 return 0;
505
506 /* Why are these in bytes, not sectors? */
507 alignment = lim->discard_alignment >> SECTOR_SHIFT;
508 granularity = lim->discard_granularity >> SECTOR_SHIFT;
509
510 /* Offset of the partition start in 'granularity' sectors */
511 offset = sector_div(sector, granularity);
512
513 /* And why do we do this modulus *again* in blkdev_issue_discard()? */
514 offset = (granularity + alignment - offset) % granularity;
515
516 /* Turn it back into bytes, gaah */
517 return offset << SECTOR_SHIFT;
518}
519
520static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
521{
522 sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
523 if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
524 sectors = PAGE_SIZE >> SECTOR_SHIFT;
525 return sectors;
526}
527
528/* Check if second and later bottom devices are compliant */
529static bool blk_stack_atomic_writes_tail(struct queue_limits *t,
530 struct queue_limits *b)
531{
532 /* We're not going to support different boundary sizes.. yet */
533 if (t->atomic_write_hw_boundary != b->atomic_write_hw_boundary)
534 return false;
535
536 /* Can't support this */
537 if (t->atomic_write_hw_unit_min > b->atomic_write_hw_unit_max)
538 return false;
539
540 /* Or this */
541 if (t->atomic_write_hw_unit_max < b->atomic_write_hw_unit_min)
542 return false;
543
544 t->atomic_write_hw_max = min(t->atomic_write_hw_max,
545 b->atomic_write_hw_max);
546 t->atomic_write_hw_unit_min = max(t->atomic_write_hw_unit_min,
547 b->atomic_write_hw_unit_min);
548 t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max,
549 b->atomic_write_hw_unit_max);
550 return true;
551}
552
553/* Check for valid boundary of first bottom device */
554static bool blk_stack_atomic_writes_boundary_head(struct queue_limits *t,
555 struct queue_limits *b)
556{
557 /*
558 * Ensure atomic write boundary is aligned with chunk sectors. Stacked
559 * devices store chunk sectors in t->io_min.
560 */
561 if (b->atomic_write_hw_boundary > t->io_min &&
562 b->atomic_write_hw_boundary % t->io_min)
563 return false;
564 if (t->io_min > b->atomic_write_hw_boundary &&
565 t->io_min % b->atomic_write_hw_boundary)
566 return false;
567
568 t->atomic_write_hw_boundary = b->atomic_write_hw_boundary;
569 return true;
570}
571
572
573/* Check stacking of first bottom device */
574static bool blk_stack_atomic_writes_head(struct queue_limits *t,
575 struct queue_limits *b)
576{
577 if (b->atomic_write_hw_boundary &&
578 !blk_stack_atomic_writes_boundary_head(t, b))
579 return false;
580
581 if (t->io_min <= SECTOR_SIZE) {
582 /* No chunk sectors, so use bottom device values directly */
583 t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
584 t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
585 t->atomic_write_hw_max = b->atomic_write_hw_max;
586 return true;
587 }
588
589 /*
590 * Find values for limits which work for chunk size.
591 * b->atomic_write_hw_unit_{min, max} may not be aligned with chunk
592 * size (t->io_min), as chunk size is not restricted to a power-of-2.
593 * So we need to find highest power-of-2 which works for the chunk
594 * size.
595 * As an example scenario, we could have b->unit_max = 16K and
596 * t->io_min = 24K. For this case, reduce t->unit_max to a value
597 * aligned with both limits, i.e. 8K in this example.
598 */
599 t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
600 while (t->io_min % t->atomic_write_hw_unit_max)
601 t->atomic_write_hw_unit_max /= 2;
602
603 t->atomic_write_hw_unit_min = min(b->atomic_write_hw_unit_min,
604 t->atomic_write_hw_unit_max);
605 t->atomic_write_hw_max = min(b->atomic_write_hw_max, t->io_min);
606
607 return true;
608}
609
610static void blk_stack_atomic_writes_limits(struct queue_limits *t,
611 struct queue_limits *b, sector_t start)
612{
613 if (!(t->features & BLK_FEAT_ATOMIC_WRITES_STACKED))
614 goto unsupported;
615
616 if (!b->atomic_write_unit_min)
617 goto unsupported;
618
619 if (!blk_atomic_write_start_sect_aligned(start, b))
620 goto unsupported;
621
622 /*
623 * If atomic_write_hw_max is set, we have already stacked 1x bottom
624 * device, so check for compliance.
625 */
626 if (t->atomic_write_hw_max) {
627 if (!blk_stack_atomic_writes_tail(t, b))
628 goto unsupported;
629 return;
630 }
631
632 if (!blk_stack_atomic_writes_head(t, b))
633 goto unsupported;
634 return;
635
636unsupported:
637 t->atomic_write_hw_max = 0;
638 t->atomic_write_hw_unit_max = 0;
639 t->atomic_write_hw_unit_min = 0;
640 t->atomic_write_hw_boundary = 0;
641 t->features &= ~BLK_FEAT_ATOMIC_WRITES_STACKED;
642}
643
644/**
645 * blk_stack_limits - adjust queue_limits for stacked devices
646 * @t: the stacking driver limits (top device)
647 * @b: the underlying queue limits (bottom, component device)
648 * @start: first data sector within component device
649 *
650 * Description:
651 * This function is used by stacking drivers like MD and DM to ensure
652 * that all component devices have compatible block sizes and
653 * alignments. The stacking driver must provide a queue_limits
654 * struct (top) and then iteratively call the stacking function for
655 * all component (bottom) devices. The stacking function will
656 * attempt to combine the values and ensure proper alignment.
657 *
658 * Returns 0 if the top and bottom queue_limits are compatible. The
659 * top device's block sizes and alignment offsets may be adjusted to
660 * ensure alignment with the bottom device. If no compatible sizes
661 * and alignments exist, -1 is returned and the resulting top
662 * queue_limits will have the misaligned flag set to indicate that
663 * the alignment_offset is undefined.
664 */
665int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
666 sector_t start)
667{
668 unsigned int top, bottom, alignment, ret = 0;
669
670 t->features |= (b->features & BLK_FEAT_INHERIT_MASK);
671
672 /*
673 * Some feaures need to be supported both by the stacking driver and all
674 * underlying devices. The stacking driver sets these flags before
675 * stacking the limits, and this will clear the flags if any of the
676 * underlying devices does not support it.
677 */
678 if (!(b->features & BLK_FEAT_NOWAIT))
679 t->features &= ~BLK_FEAT_NOWAIT;
680 if (!(b->features & BLK_FEAT_POLL))
681 t->features &= ~BLK_FEAT_POLL;
682
683 t->flags |= (b->flags & BLK_FLAG_MISALIGNED);
684
685 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
686 t->max_user_sectors = min_not_zero(t->max_user_sectors,
687 b->max_user_sectors);
688 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
689 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
690 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
691 b->max_write_zeroes_sectors);
692 t->max_hw_zone_append_sectors = min(t->max_hw_zone_append_sectors,
693 b->max_hw_zone_append_sectors);
694
695 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
696 b->seg_boundary_mask);
697 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
698 b->virt_boundary_mask);
699
700 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
701 t->max_discard_segments = min_not_zero(t->max_discard_segments,
702 b->max_discard_segments);
703 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
704 b->max_integrity_segments);
705
706 t->max_segment_size = min_not_zero(t->max_segment_size,
707 b->max_segment_size);
708
709 alignment = queue_limit_alignment_offset(b, start);
710
711 /* Bottom device has different alignment. Check that it is
712 * compatible with the current top alignment.
713 */
714 if (t->alignment_offset != alignment) {
715
716 top = max(t->physical_block_size, t->io_min)
717 + t->alignment_offset;
718 bottom = max(b->physical_block_size, b->io_min) + alignment;
719
720 /* Verify that top and bottom intervals line up */
721 if (max(top, bottom) % min(top, bottom)) {
722 t->flags |= BLK_FLAG_MISALIGNED;
723 ret = -1;
724 }
725 }
726
727 t->logical_block_size = max(t->logical_block_size,
728 b->logical_block_size);
729
730 t->physical_block_size = max(t->physical_block_size,
731 b->physical_block_size);
732
733 t->io_min = max(t->io_min, b->io_min);
734 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
735 t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
736
737 /* Set non-power-of-2 compatible chunk_sectors boundary */
738 if (b->chunk_sectors)
739 t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
740
741 /* Physical block size a multiple of the logical block size? */
742 if (t->physical_block_size & (t->logical_block_size - 1)) {
743 t->physical_block_size = t->logical_block_size;
744 t->flags |= BLK_FLAG_MISALIGNED;
745 ret = -1;
746 }
747
748 /* Minimum I/O a multiple of the physical block size? */
749 if (t->io_min & (t->physical_block_size - 1)) {
750 t->io_min = t->physical_block_size;
751 t->flags |= BLK_FLAG_MISALIGNED;
752 ret = -1;
753 }
754
755 /* Optimal I/O a multiple of the physical block size? */
756 if (t->io_opt & (t->physical_block_size - 1)) {
757 t->io_opt = 0;
758 t->flags |= BLK_FLAG_MISALIGNED;
759 ret = -1;
760 }
761
762 /* chunk_sectors a multiple of the physical block size? */
763 if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
764 t->chunk_sectors = 0;
765 t->flags |= BLK_FLAG_MISALIGNED;
766 ret = -1;
767 }
768
769 /* Find lowest common alignment_offset */
770 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
771 % max(t->physical_block_size, t->io_min);
772
773 /* Verify that new alignment_offset is on a logical block boundary */
774 if (t->alignment_offset & (t->logical_block_size - 1)) {
775 t->flags |= BLK_FLAG_MISALIGNED;
776 ret = -1;
777 }
778
779 t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
780 t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
781 t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
782
783 /* Discard alignment and granularity */
784 if (b->discard_granularity) {
785 alignment = queue_limit_discard_alignment(b, start);
786
787 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
788 b->max_discard_sectors);
789 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
790 b->max_hw_discard_sectors);
791 t->discard_granularity = max(t->discard_granularity,
792 b->discard_granularity);
793 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
794 t->discard_granularity;
795 }
796 t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
797 b->max_secure_erase_sectors);
798 t->zone_write_granularity = max(t->zone_write_granularity,
799 b->zone_write_granularity);
800 if (!(t->features & BLK_FEAT_ZONED)) {
801 t->zone_write_granularity = 0;
802 t->max_zone_append_sectors = 0;
803 }
804 blk_stack_atomic_writes_limits(t, b, start);
805
806 return ret;
807}
808EXPORT_SYMBOL(blk_stack_limits);
809
810/**
811 * queue_limits_stack_bdev - adjust queue_limits for stacked devices
812 * @t: the stacking driver limits (top device)
813 * @bdev: the underlying block device (bottom)
814 * @offset: offset to beginning of data within component device
815 * @pfx: prefix to use for warnings logged
816 *
817 * Description:
818 * This function is used by stacking drivers like MD and DM to ensure
819 * that all component devices have compatible block sizes and
820 * alignments. The stacking driver must provide a queue_limits
821 * struct (top) and then iteratively call the stacking function for
822 * all component (bottom) devices. The stacking function will
823 * attempt to combine the values and ensure proper alignment.
824 */
825void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
826 sector_t offset, const char *pfx)
827{
828 if (blk_stack_limits(t, bdev_limits(bdev),
829 get_start_sect(bdev) + offset))
830 pr_notice("%s: Warning: Device %pg is misaligned\n",
831 pfx, bdev);
832}
833EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
834
835/**
836 * queue_limits_stack_integrity - stack integrity profile
837 * @t: target queue limits
838 * @b: base queue limits
839 *
840 * Check if the integrity profile in the @b can be stacked into the
841 * target @t. Stacking is possible if either:
842 *
843 * a) does not have any integrity information stacked into it yet
844 * b) the integrity profile in @b is identical to the one in @t
845 *
846 * If @b can be stacked into @t, return %true. Else return %false and clear the
847 * integrity information in @t.
848 */
849bool queue_limits_stack_integrity(struct queue_limits *t,
850 struct queue_limits *b)
851{
852 struct blk_integrity *ti = &t->integrity;
853 struct blk_integrity *bi = &b->integrity;
854
855 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
856 return true;
857
858 if (!ti->tuple_size) {
859 /* inherit the settings from the first underlying device */
860 if (!(ti->flags & BLK_INTEGRITY_STACKED)) {
861 ti->flags = BLK_INTEGRITY_DEVICE_CAPABLE |
862 (bi->flags & BLK_INTEGRITY_REF_TAG);
863 ti->csum_type = bi->csum_type;
864 ti->tuple_size = bi->tuple_size;
865 ti->pi_offset = bi->pi_offset;
866 ti->interval_exp = bi->interval_exp;
867 ti->tag_size = bi->tag_size;
868 goto done;
869 }
870 if (!bi->tuple_size)
871 goto done;
872 }
873
874 if (ti->tuple_size != bi->tuple_size)
875 goto incompatible;
876 if (ti->interval_exp != bi->interval_exp)
877 goto incompatible;
878 if (ti->tag_size != bi->tag_size)
879 goto incompatible;
880 if (ti->csum_type != bi->csum_type)
881 goto incompatible;
882 if ((ti->flags & BLK_INTEGRITY_REF_TAG) !=
883 (bi->flags & BLK_INTEGRITY_REF_TAG))
884 goto incompatible;
885
886done:
887 ti->flags |= BLK_INTEGRITY_STACKED;
888 return true;
889
890incompatible:
891 memset(ti, 0, sizeof(*ti));
892 return false;
893}
894EXPORT_SYMBOL_GPL(queue_limits_stack_integrity);
895
896/**
897 * blk_set_queue_depth - tell the block layer about the device queue depth
898 * @q: the request queue for the device
899 * @depth: queue depth
900 *
901 */
902void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
903{
904 q->queue_depth = depth;
905 rq_qos_queue_depth_changed(q);
906}
907EXPORT_SYMBOL(blk_set_queue_depth);
908
909int bdev_alignment_offset(struct block_device *bdev)
910{
911 struct request_queue *q = bdev_get_queue(bdev);
912
913 if (q->limits.flags & BLK_FLAG_MISALIGNED)
914 return -1;
915 if (bdev_is_partition(bdev))
916 return queue_limit_alignment_offset(&q->limits,
917 bdev->bd_start_sect);
918 return q->limits.alignment_offset;
919}
920EXPORT_SYMBOL_GPL(bdev_alignment_offset);
921
922unsigned int bdev_discard_alignment(struct block_device *bdev)
923{
924 struct request_queue *q = bdev_get_queue(bdev);
925
926 if (bdev_is_partition(bdev))
927 return queue_limit_discard_alignment(&q->limits,
928 bdev->bd_start_sect);
929 return q->limits.discard_alignment;
930}
931EXPORT_SYMBOL_GPL(bdev_discard_alignment);