Loading...
1/*
2 * Functions related to setting various queue properties from drivers
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/init.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
10#include <linux/gcd.h>
11#include <linux/lcm.h>
12#include <linux/jiffies.h>
13#include <linux/gfp.h>
14
15#include "blk.h"
16
17unsigned long blk_max_low_pfn;
18EXPORT_SYMBOL(blk_max_low_pfn);
19
20unsigned long blk_max_pfn;
21
22/**
23 * blk_queue_prep_rq - set a prepare_request function for queue
24 * @q: queue
25 * @pfn: prepare_request function
26 *
27 * It's possible for a queue to register a prepare_request callback which
28 * is invoked before the request is handed to the request_fn. The goal of
29 * the function is to prepare a request for I/O, it can be used to build a
30 * cdb from the request data for instance.
31 *
32 */
33void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
34{
35 q->prep_rq_fn = pfn;
36}
37EXPORT_SYMBOL(blk_queue_prep_rq);
38
39/**
40 * blk_queue_unprep_rq - set an unprepare_request function for queue
41 * @q: queue
42 * @ufn: unprepare_request function
43 *
44 * It's possible for a queue to register an unprepare_request callback
45 * which is invoked before the request is finally completed. The goal
46 * of the function is to deallocate any data that was allocated in the
47 * prepare_request callback.
48 *
49 */
50void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
51{
52 q->unprep_rq_fn = ufn;
53}
54EXPORT_SYMBOL(blk_queue_unprep_rq);
55
56/**
57 * blk_queue_merge_bvec - set a merge_bvec function for queue
58 * @q: queue
59 * @mbfn: merge_bvec_fn
60 *
61 * Usually queues have static limitations on the max sectors or segments that
62 * we can put in a request. Stacking drivers may have some settings that
63 * are dynamic, and thus we have to query the queue whether it is ok to
64 * add a new bio_vec to a bio at a given offset or not. If the block device
65 * has such limitations, it needs to register a merge_bvec_fn to control
66 * the size of bio's sent to it. Note that a block device *must* allow a
67 * single page to be added to an empty bio. The block device driver may want
68 * to use the bio_split() function to deal with these bio's. By default
69 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
70 * honored.
71 */
72void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
73{
74 q->merge_bvec_fn = mbfn;
75}
76EXPORT_SYMBOL(blk_queue_merge_bvec);
77
78void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
79{
80 q->softirq_done_fn = fn;
81}
82EXPORT_SYMBOL(blk_queue_softirq_done);
83
84void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
85{
86 q->rq_timeout = timeout;
87}
88EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
89
90void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
91{
92 q->rq_timed_out_fn = fn;
93}
94EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
95
96void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
97{
98 q->lld_busy_fn = fn;
99}
100EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
101
102/**
103 * blk_set_default_limits - reset limits to default values
104 * @lim: the queue_limits structure to reset
105 *
106 * Description:
107 * Returns a queue_limit struct to its default state.
108 */
109void blk_set_default_limits(struct queue_limits *lim)
110{
111 lim->max_segments = BLK_MAX_SEGMENTS;
112 lim->max_integrity_segments = 0;
113 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
114 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
115 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
116 lim->max_write_same_sectors = 0;
117 lim->max_discard_sectors = 0;
118 lim->discard_granularity = 0;
119 lim->discard_alignment = 0;
120 lim->discard_misaligned = 0;
121 lim->discard_zeroes_data = 0;
122 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
123 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
124 lim->alignment_offset = 0;
125 lim->io_opt = 0;
126 lim->misaligned = 0;
127 lim->cluster = 1;
128}
129EXPORT_SYMBOL(blk_set_default_limits);
130
131/**
132 * blk_set_stacking_limits - set default limits for stacking devices
133 * @lim: the queue_limits structure to reset
134 *
135 * Description:
136 * Returns a queue_limit struct to its default state. Should be used
137 * by stacking drivers like DM that have no internal limits.
138 */
139void blk_set_stacking_limits(struct queue_limits *lim)
140{
141 blk_set_default_limits(lim);
142
143 /* Inherit limits from component devices */
144 lim->discard_zeroes_data = 1;
145 lim->max_segments = USHRT_MAX;
146 lim->max_hw_sectors = UINT_MAX;
147 lim->max_segment_size = UINT_MAX;
148 lim->max_sectors = UINT_MAX;
149 lim->max_write_same_sectors = UINT_MAX;
150}
151EXPORT_SYMBOL(blk_set_stacking_limits);
152
153/**
154 * blk_queue_make_request - define an alternate make_request function for a device
155 * @q: the request queue for the device to be affected
156 * @mfn: the alternate make_request function
157 *
158 * Description:
159 * The normal way for &struct bios to be passed to a device
160 * driver is for them to be collected into requests on a request
161 * queue, and then to allow the device driver to select requests
162 * off that queue when it is ready. This works well for many block
163 * devices. However some block devices (typically virtual devices
164 * such as md or lvm) do not benefit from the processing on the
165 * request queue, and are served best by having the requests passed
166 * directly to them. This can be achieved by providing a function
167 * to blk_queue_make_request().
168 *
169 * Caveat:
170 * The driver that does this *must* be able to deal appropriately
171 * with buffers in "highmemory". This can be accomplished by either calling
172 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
173 * blk_queue_bounce() to create a buffer in normal memory.
174 **/
175void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
176{
177 /*
178 * set defaults
179 */
180 q->nr_requests = BLKDEV_MAX_RQ;
181
182 q->make_request_fn = mfn;
183 blk_queue_dma_alignment(q, 511);
184 blk_queue_congestion_threshold(q);
185 q->nr_batching = BLK_BATCH_REQ;
186
187 blk_set_default_limits(&q->limits);
188
189 /*
190 * by default assume old behaviour and bounce for any highmem page
191 */
192 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
193}
194EXPORT_SYMBOL(blk_queue_make_request);
195
196/**
197 * blk_queue_bounce_limit - set bounce buffer limit for queue
198 * @q: the request queue for the device
199 * @max_addr: the maximum address the device can handle
200 *
201 * Description:
202 * Different hardware can have different requirements as to what pages
203 * it can do I/O directly to. A low level driver can call
204 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
205 * buffers for doing I/O to pages residing above @max_addr.
206 **/
207void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
208{
209 unsigned long b_pfn = max_addr >> PAGE_SHIFT;
210 int dma = 0;
211
212 q->bounce_gfp = GFP_NOIO;
213#if BITS_PER_LONG == 64
214 /*
215 * Assume anything <= 4GB can be handled by IOMMU. Actually
216 * some IOMMUs can handle everything, but I don't know of a
217 * way to test this here.
218 */
219 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
220 dma = 1;
221 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
222#else
223 if (b_pfn < blk_max_low_pfn)
224 dma = 1;
225 q->limits.bounce_pfn = b_pfn;
226#endif
227 if (dma) {
228 init_emergency_isa_pool();
229 q->bounce_gfp = GFP_NOIO | GFP_DMA;
230 q->limits.bounce_pfn = b_pfn;
231 }
232}
233EXPORT_SYMBOL(blk_queue_bounce_limit);
234
235/**
236 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
237 * @limits: the queue limits
238 * @max_hw_sectors: max hardware sectors in the usual 512b unit
239 *
240 * Description:
241 * Enables a low level driver to set a hard upper limit,
242 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
243 * the device driver based upon the combined capabilities of I/O
244 * controller and storage device.
245 *
246 * max_sectors is a soft limit imposed by the block layer for
247 * filesystem type requests. This value can be overridden on a
248 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
249 * The soft limit can not exceed max_hw_sectors.
250 **/
251void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
252{
253 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
254 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
255 printk(KERN_INFO "%s: set to minimum %d\n",
256 __func__, max_hw_sectors);
257 }
258
259 limits->max_hw_sectors = max_hw_sectors;
260 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
261 BLK_DEF_MAX_SECTORS);
262}
263EXPORT_SYMBOL(blk_limits_max_hw_sectors);
264
265/**
266 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
267 * @q: the request queue for the device
268 * @max_hw_sectors: max hardware sectors in the usual 512b unit
269 *
270 * Description:
271 * See description for blk_limits_max_hw_sectors().
272 **/
273void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
274{
275 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
276}
277EXPORT_SYMBOL(blk_queue_max_hw_sectors);
278
279/**
280 * blk_queue_max_discard_sectors - set max sectors for a single discard
281 * @q: the request queue for the device
282 * @max_discard_sectors: maximum number of sectors to discard
283 **/
284void blk_queue_max_discard_sectors(struct request_queue *q,
285 unsigned int max_discard_sectors)
286{
287 q->limits.max_discard_sectors = max_discard_sectors;
288}
289EXPORT_SYMBOL(blk_queue_max_discard_sectors);
290
291/**
292 * blk_queue_max_write_same_sectors - set max sectors for a single write same
293 * @q: the request queue for the device
294 * @max_write_same_sectors: maximum number of sectors to write per command
295 **/
296void blk_queue_max_write_same_sectors(struct request_queue *q,
297 unsigned int max_write_same_sectors)
298{
299 q->limits.max_write_same_sectors = max_write_same_sectors;
300}
301EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
302
303/**
304 * blk_queue_max_segments - set max hw segments for a request for this queue
305 * @q: the request queue for the device
306 * @max_segments: max number of segments
307 *
308 * Description:
309 * Enables a low level driver to set an upper limit on the number of
310 * hw data segments in a request.
311 **/
312void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
313{
314 if (!max_segments) {
315 max_segments = 1;
316 printk(KERN_INFO "%s: set to minimum %d\n",
317 __func__, max_segments);
318 }
319
320 q->limits.max_segments = max_segments;
321}
322EXPORT_SYMBOL(blk_queue_max_segments);
323
324/**
325 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
326 * @q: the request queue for the device
327 * @max_size: max size of segment in bytes
328 *
329 * Description:
330 * Enables a low level driver to set an upper limit on the size of a
331 * coalesced segment
332 **/
333void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
334{
335 if (max_size < PAGE_CACHE_SIZE) {
336 max_size = PAGE_CACHE_SIZE;
337 printk(KERN_INFO "%s: set to minimum %d\n",
338 __func__, max_size);
339 }
340
341 q->limits.max_segment_size = max_size;
342}
343EXPORT_SYMBOL(blk_queue_max_segment_size);
344
345/**
346 * blk_queue_logical_block_size - set logical block size for the queue
347 * @q: the request queue for the device
348 * @size: the logical block size, in bytes
349 *
350 * Description:
351 * This should be set to the lowest possible block size that the
352 * storage device can address. The default of 512 covers most
353 * hardware.
354 **/
355void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
356{
357 q->limits.logical_block_size = size;
358
359 if (q->limits.physical_block_size < size)
360 q->limits.physical_block_size = size;
361
362 if (q->limits.io_min < q->limits.physical_block_size)
363 q->limits.io_min = q->limits.physical_block_size;
364}
365EXPORT_SYMBOL(blk_queue_logical_block_size);
366
367/**
368 * blk_queue_physical_block_size - set physical block size for the queue
369 * @q: the request queue for the device
370 * @size: the physical block size, in bytes
371 *
372 * Description:
373 * This should be set to the lowest possible sector size that the
374 * hardware can operate on without reverting to read-modify-write
375 * operations.
376 */
377void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
378{
379 q->limits.physical_block_size = size;
380
381 if (q->limits.physical_block_size < q->limits.logical_block_size)
382 q->limits.physical_block_size = q->limits.logical_block_size;
383
384 if (q->limits.io_min < q->limits.physical_block_size)
385 q->limits.io_min = q->limits.physical_block_size;
386}
387EXPORT_SYMBOL(blk_queue_physical_block_size);
388
389/**
390 * blk_queue_alignment_offset - set physical block alignment offset
391 * @q: the request queue for the device
392 * @offset: alignment offset in bytes
393 *
394 * Description:
395 * Some devices are naturally misaligned to compensate for things like
396 * the legacy DOS partition table 63-sector offset. Low-level drivers
397 * should call this function for devices whose first sector is not
398 * naturally aligned.
399 */
400void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
401{
402 q->limits.alignment_offset =
403 offset & (q->limits.physical_block_size - 1);
404 q->limits.misaligned = 0;
405}
406EXPORT_SYMBOL(blk_queue_alignment_offset);
407
408/**
409 * blk_limits_io_min - set minimum request size for a device
410 * @limits: the queue limits
411 * @min: smallest I/O size in bytes
412 *
413 * Description:
414 * Some devices have an internal block size bigger than the reported
415 * hardware sector size. This function can be used to signal the
416 * smallest I/O the device can perform without incurring a performance
417 * penalty.
418 */
419void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
420{
421 limits->io_min = min;
422
423 if (limits->io_min < limits->logical_block_size)
424 limits->io_min = limits->logical_block_size;
425
426 if (limits->io_min < limits->physical_block_size)
427 limits->io_min = limits->physical_block_size;
428}
429EXPORT_SYMBOL(blk_limits_io_min);
430
431/**
432 * blk_queue_io_min - set minimum request size for the queue
433 * @q: the request queue for the device
434 * @min: smallest I/O size in bytes
435 *
436 * Description:
437 * Storage devices may report a granularity or preferred minimum I/O
438 * size which is the smallest request the device can perform without
439 * incurring a performance penalty. For disk drives this is often the
440 * physical block size. For RAID arrays it is often the stripe chunk
441 * size. A properly aligned multiple of minimum_io_size is the
442 * preferred request size for workloads where a high number of I/O
443 * operations is desired.
444 */
445void blk_queue_io_min(struct request_queue *q, unsigned int min)
446{
447 blk_limits_io_min(&q->limits, min);
448}
449EXPORT_SYMBOL(blk_queue_io_min);
450
451/**
452 * blk_limits_io_opt - set optimal request size for a device
453 * @limits: the queue limits
454 * @opt: smallest I/O size in bytes
455 *
456 * Description:
457 * Storage devices may report an optimal I/O size, which is the
458 * device's preferred unit for sustained I/O. This is rarely reported
459 * for disk drives. For RAID arrays it is usually the stripe width or
460 * the internal track size. A properly aligned multiple of
461 * optimal_io_size is the preferred request size for workloads where
462 * sustained throughput is desired.
463 */
464void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
465{
466 limits->io_opt = opt;
467}
468EXPORT_SYMBOL(blk_limits_io_opt);
469
470/**
471 * blk_queue_io_opt - set optimal request size for the queue
472 * @q: the request queue for the device
473 * @opt: optimal request size in bytes
474 *
475 * Description:
476 * Storage devices may report an optimal I/O size, which is the
477 * device's preferred unit for sustained I/O. This is rarely reported
478 * for disk drives. For RAID arrays it is usually the stripe width or
479 * the internal track size. A properly aligned multiple of
480 * optimal_io_size is the preferred request size for workloads where
481 * sustained throughput is desired.
482 */
483void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
484{
485 blk_limits_io_opt(&q->limits, opt);
486}
487EXPORT_SYMBOL(blk_queue_io_opt);
488
489/**
490 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
491 * @t: the stacking driver (top)
492 * @b: the underlying device (bottom)
493 **/
494void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
495{
496 blk_stack_limits(&t->limits, &b->limits, 0);
497}
498EXPORT_SYMBOL(blk_queue_stack_limits);
499
500/**
501 * blk_stack_limits - adjust queue_limits for stacked devices
502 * @t: the stacking driver limits (top device)
503 * @b: the underlying queue limits (bottom, component device)
504 * @start: first data sector within component device
505 *
506 * Description:
507 * This function is used by stacking drivers like MD and DM to ensure
508 * that all component devices have compatible block sizes and
509 * alignments. The stacking driver must provide a queue_limits
510 * struct (top) and then iteratively call the stacking function for
511 * all component (bottom) devices. The stacking function will
512 * attempt to combine the values and ensure proper alignment.
513 *
514 * Returns 0 if the top and bottom queue_limits are compatible. The
515 * top device's block sizes and alignment offsets may be adjusted to
516 * ensure alignment with the bottom device. If no compatible sizes
517 * and alignments exist, -1 is returned and the resulting top
518 * queue_limits will have the misaligned flag set to indicate that
519 * the alignment_offset is undefined.
520 */
521int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
522 sector_t start)
523{
524 unsigned int top, bottom, alignment, ret = 0;
525
526 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
527 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
528 t->max_write_same_sectors = min(t->max_write_same_sectors,
529 b->max_write_same_sectors);
530 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
531
532 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
533 b->seg_boundary_mask);
534
535 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
536 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
537 b->max_integrity_segments);
538
539 t->max_segment_size = min_not_zero(t->max_segment_size,
540 b->max_segment_size);
541
542 t->misaligned |= b->misaligned;
543
544 alignment = queue_limit_alignment_offset(b, start);
545
546 /* Bottom device has different alignment. Check that it is
547 * compatible with the current top alignment.
548 */
549 if (t->alignment_offset != alignment) {
550
551 top = max(t->physical_block_size, t->io_min)
552 + t->alignment_offset;
553 bottom = max(b->physical_block_size, b->io_min) + alignment;
554
555 /* Verify that top and bottom intervals line up */
556 if (max(top, bottom) & (min(top, bottom) - 1)) {
557 t->misaligned = 1;
558 ret = -1;
559 }
560 }
561
562 t->logical_block_size = max(t->logical_block_size,
563 b->logical_block_size);
564
565 t->physical_block_size = max(t->physical_block_size,
566 b->physical_block_size);
567
568 t->io_min = max(t->io_min, b->io_min);
569 t->io_opt = lcm(t->io_opt, b->io_opt);
570
571 t->cluster &= b->cluster;
572 t->discard_zeroes_data &= b->discard_zeroes_data;
573
574 /* Physical block size a multiple of the logical block size? */
575 if (t->physical_block_size & (t->logical_block_size - 1)) {
576 t->physical_block_size = t->logical_block_size;
577 t->misaligned = 1;
578 ret = -1;
579 }
580
581 /* Minimum I/O a multiple of the physical block size? */
582 if (t->io_min & (t->physical_block_size - 1)) {
583 t->io_min = t->physical_block_size;
584 t->misaligned = 1;
585 ret = -1;
586 }
587
588 /* Optimal I/O a multiple of the physical block size? */
589 if (t->io_opt & (t->physical_block_size - 1)) {
590 t->io_opt = 0;
591 t->misaligned = 1;
592 ret = -1;
593 }
594
595 t->raid_partial_stripes_expensive =
596 max(t->raid_partial_stripes_expensive,
597 b->raid_partial_stripes_expensive);
598
599 /* Find lowest common alignment_offset */
600 t->alignment_offset = lcm(t->alignment_offset, alignment)
601 & (max(t->physical_block_size, t->io_min) - 1);
602
603 /* Verify that new alignment_offset is on a logical block boundary */
604 if (t->alignment_offset & (t->logical_block_size - 1)) {
605 t->misaligned = 1;
606 ret = -1;
607 }
608
609 /* Discard alignment and granularity */
610 if (b->discard_granularity) {
611 alignment = queue_limit_discard_alignment(b, start);
612
613 if (t->discard_granularity != 0 &&
614 t->discard_alignment != alignment) {
615 top = t->discard_granularity + t->discard_alignment;
616 bottom = b->discard_granularity + alignment;
617
618 /* Verify that top and bottom intervals line up */
619 if ((max(top, bottom) % min(top, bottom)) != 0)
620 t->discard_misaligned = 1;
621 }
622
623 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
624 b->max_discard_sectors);
625 t->discard_granularity = max(t->discard_granularity,
626 b->discard_granularity);
627 t->discard_alignment = lcm(t->discard_alignment, alignment) %
628 t->discard_granularity;
629 }
630
631 return ret;
632}
633EXPORT_SYMBOL(blk_stack_limits);
634
635/**
636 * bdev_stack_limits - adjust queue limits for stacked drivers
637 * @t: the stacking driver limits (top device)
638 * @bdev: the component block_device (bottom)
639 * @start: first data sector within component device
640 *
641 * Description:
642 * Merges queue limits for a top device and a block_device. Returns
643 * 0 if alignment didn't change. Returns -1 if adding the bottom
644 * device caused misalignment.
645 */
646int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
647 sector_t start)
648{
649 struct request_queue *bq = bdev_get_queue(bdev);
650
651 start += get_start_sect(bdev);
652
653 return blk_stack_limits(t, &bq->limits, start);
654}
655EXPORT_SYMBOL(bdev_stack_limits);
656
657/**
658 * disk_stack_limits - adjust queue limits for stacked drivers
659 * @disk: MD/DM gendisk (top)
660 * @bdev: the underlying block device (bottom)
661 * @offset: offset to beginning of data within component device
662 *
663 * Description:
664 * Merges the limits for a top level gendisk and a bottom level
665 * block_device.
666 */
667void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
668 sector_t offset)
669{
670 struct request_queue *t = disk->queue;
671
672 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
673 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
674
675 disk_name(disk, 0, top);
676 bdevname(bdev, bottom);
677
678 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
679 top, bottom);
680 }
681}
682EXPORT_SYMBOL(disk_stack_limits);
683
684/**
685 * blk_queue_dma_pad - set pad mask
686 * @q: the request queue for the device
687 * @mask: pad mask
688 *
689 * Set dma pad mask.
690 *
691 * Appending pad buffer to a request modifies the last entry of a
692 * scatter list such that it includes the pad buffer.
693 **/
694void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
695{
696 q->dma_pad_mask = mask;
697}
698EXPORT_SYMBOL(blk_queue_dma_pad);
699
700/**
701 * blk_queue_update_dma_pad - update pad mask
702 * @q: the request queue for the device
703 * @mask: pad mask
704 *
705 * Update dma pad mask.
706 *
707 * Appending pad buffer to a request modifies the last entry of a
708 * scatter list such that it includes the pad buffer.
709 **/
710void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
711{
712 if (mask > q->dma_pad_mask)
713 q->dma_pad_mask = mask;
714}
715EXPORT_SYMBOL(blk_queue_update_dma_pad);
716
717/**
718 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
719 * @q: the request queue for the device
720 * @dma_drain_needed: fn which returns non-zero if drain is necessary
721 * @buf: physically contiguous buffer
722 * @size: size of the buffer in bytes
723 *
724 * Some devices have excess DMA problems and can't simply discard (or
725 * zero fill) the unwanted piece of the transfer. They have to have a
726 * real area of memory to transfer it into. The use case for this is
727 * ATAPI devices in DMA mode. If the packet command causes a transfer
728 * bigger than the transfer size some HBAs will lock up if there
729 * aren't DMA elements to contain the excess transfer. What this API
730 * does is adjust the queue so that the buf is always appended
731 * silently to the scatterlist.
732 *
733 * Note: This routine adjusts max_hw_segments to make room for appending
734 * the drain buffer. If you call blk_queue_max_segments() after calling
735 * this routine, you must set the limit to one fewer than your device
736 * can support otherwise there won't be room for the drain buffer.
737 */
738int blk_queue_dma_drain(struct request_queue *q,
739 dma_drain_needed_fn *dma_drain_needed,
740 void *buf, unsigned int size)
741{
742 if (queue_max_segments(q) < 2)
743 return -EINVAL;
744 /* make room for appending the drain */
745 blk_queue_max_segments(q, queue_max_segments(q) - 1);
746 q->dma_drain_needed = dma_drain_needed;
747 q->dma_drain_buffer = buf;
748 q->dma_drain_size = size;
749
750 return 0;
751}
752EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
753
754/**
755 * blk_queue_segment_boundary - set boundary rules for segment merging
756 * @q: the request queue for the device
757 * @mask: the memory boundary mask
758 **/
759void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
760{
761 if (mask < PAGE_CACHE_SIZE - 1) {
762 mask = PAGE_CACHE_SIZE - 1;
763 printk(KERN_INFO "%s: set to minimum %lx\n",
764 __func__, mask);
765 }
766
767 q->limits.seg_boundary_mask = mask;
768}
769EXPORT_SYMBOL(blk_queue_segment_boundary);
770
771/**
772 * blk_queue_dma_alignment - set dma length and memory alignment
773 * @q: the request queue for the device
774 * @mask: alignment mask
775 *
776 * description:
777 * set required memory and length alignment for direct dma transactions.
778 * this is used when building direct io requests for the queue.
779 *
780 **/
781void blk_queue_dma_alignment(struct request_queue *q, int mask)
782{
783 q->dma_alignment = mask;
784}
785EXPORT_SYMBOL(blk_queue_dma_alignment);
786
787/**
788 * blk_queue_update_dma_alignment - update dma length and memory alignment
789 * @q: the request queue for the device
790 * @mask: alignment mask
791 *
792 * description:
793 * update required memory and length alignment for direct dma transactions.
794 * If the requested alignment is larger than the current alignment, then
795 * the current queue alignment is updated to the new value, otherwise it
796 * is left alone. The design of this is to allow multiple objects
797 * (driver, device, transport etc) to set their respective
798 * alignments without having them interfere.
799 *
800 **/
801void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
802{
803 BUG_ON(mask > PAGE_SIZE);
804
805 if (mask > q->dma_alignment)
806 q->dma_alignment = mask;
807}
808EXPORT_SYMBOL(blk_queue_update_dma_alignment);
809
810/**
811 * blk_queue_flush - configure queue's cache flush capability
812 * @q: the request queue for the device
813 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
814 *
815 * Tell block layer cache flush capability of @q. If it supports
816 * flushing, REQ_FLUSH should be set. If it supports bypassing
817 * write cache for individual writes, REQ_FUA should be set.
818 */
819void blk_queue_flush(struct request_queue *q, unsigned int flush)
820{
821 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
822
823 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
824 flush &= ~REQ_FUA;
825
826 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
827}
828EXPORT_SYMBOL_GPL(blk_queue_flush);
829
830void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
831{
832 q->flush_not_queueable = !queueable;
833}
834EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
835
836static int __init blk_settings_init(void)
837{
838 blk_max_low_pfn = max_low_pfn - 1;
839 blk_max_pfn = max_pfn - 1;
840 return 0;
841}
842subsys_initcall(blk_settings_init);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/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);