xref: /openbmc/linux/block/blk-settings.c (revision bacf743e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Functions related to setting various queue properties from drivers
4  */
5 #include <linux/kernel.h>
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/bio.h>
9 #include <linux/blkdev.h>
10 #include <linux/pagemap.h>
11 #include <linux/backing-dev-defs.h>
12 #include <linux/gcd.h>
13 #include <linux/lcm.h>
14 #include <linux/jiffies.h>
15 #include <linux/gfp.h>
16 #include <linux/dma-mapping.h>
17 
18 #include "blk.h"
19 #include "blk-wbt.h"
20 
21 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
22 {
23 	q->rq_timeout = timeout;
24 }
25 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
26 
27 /**
28  * blk_set_default_limits - reset limits to default values
29  * @lim:  the queue_limits structure to reset
30  *
31  * Description:
32  *   Returns a queue_limit struct to its default state.
33  */
34 void blk_set_default_limits(struct queue_limits *lim)
35 {
36 	lim->max_segments = BLK_MAX_SEGMENTS;
37 	lim->max_discard_segments = 1;
38 	lim->max_integrity_segments = 0;
39 	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
40 	lim->virt_boundary_mask = 0;
41 	lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
42 	lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
43 	lim->max_dev_sectors = 0;
44 	lim->chunk_sectors = 0;
45 	lim->max_write_zeroes_sectors = 0;
46 	lim->max_zone_append_sectors = 0;
47 	lim->max_discard_sectors = 0;
48 	lim->max_hw_discard_sectors = 0;
49 	lim->discard_granularity = 0;
50 	lim->discard_alignment = 0;
51 	lim->discard_misaligned = 0;
52 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
53 	lim->bounce = BLK_BOUNCE_NONE;
54 	lim->alignment_offset = 0;
55 	lim->io_opt = 0;
56 	lim->misaligned = 0;
57 	lim->zoned = BLK_ZONED_NONE;
58 	lim->zone_write_granularity = 0;
59 }
60 EXPORT_SYMBOL(blk_set_default_limits);
61 
62 /**
63  * blk_set_stacking_limits - set default limits for stacking devices
64  * @lim:  the queue_limits structure to reset
65  *
66  * Description:
67  *   Returns a queue_limit struct to its default state. Should be used
68  *   by stacking drivers like DM that have no internal limits.
69  */
70 void blk_set_stacking_limits(struct queue_limits *lim)
71 {
72 	blk_set_default_limits(lim);
73 
74 	/* Inherit limits from component devices */
75 	lim->max_segments = USHRT_MAX;
76 	lim->max_discard_segments = USHRT_MAX;
77 	lim->max_hw_sectors = UINT_MAX;
78 	lim->max_segment_size = UINT_MAX;
79 	lim->max_sectors = UINT_MAX;
80 	lim->max_dev_sectors = UINT_MAX;
81 	lim->max_write_zeroes_sectors = UINT_MAX;
82 	lim->max_zone_append_sectors = UINT_MAX;
83 }
84 EXPORT_SYMBOL(blk_set_stacking_limits);
85 
86 /**
87  * blk_queue_bounce_limit - set bounce buffer limit for queue
88  * @q: the request queue for the device
89  * @bounce: bounce limit to enforce
90  *
91  * Description:
92  *    Force bouncing for ISA DMA ranges or highmem.
93  *
94  *    DEPRECATED, don't use in new code.
95  **/
96 void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce)
97 {
98 	q->limits.bounce = bounce;
99 }
100 EXPORT_SYMBOL(blk_queue_bounce_limit);
101 
102 /**
103  * blk_queue_max_hw_sectors - set max sectors for a request for this queue
104  * @q:  the request queue for the device
105  * @max_hw_sectors:  max hardware sectors in the usual 512b unit
106  *
107  * Description:
108  *    Enables a low level driver to set a hard upper limit,
109  *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
110  *    the device driver based upon the capabilities of the I/O
111  *    controller.
112  *
113  *    max_dev_sectors is a hard limit imposed by the storage device for
114  *    READ/WRITE requests. It is set by the disk driver.
115  *
116  *    max_sectors is a soft limit imposed by the block layer for
117  *    filesystem type requests.  This value can be overridden on a
118  *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
119  *    The soft limit can not exceed max_hw_sectors.
120  **/
121 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
122 {
123 	struct queue_limits *limits = &q->limits;
124 	unsigned int max_sectors;
125 
126 	if ((max_hw_sectors << 9) < PAGE_SIZE) {
127 		max_hw_sectors = 1 << (PAGE_SHIFT - 9);
128 		printk(KERN_INFO "%s: set to minimum %d\n",
129 		       __func__, max_hw_sectors);
130 	}
131 
132 	max_hw_sectors = round_down(max_hw_sectors,
133 				    limits->logical_block_size >> SECTOR_SHIFT);
134 	limits->max_hw_sectors = max_hw_sectors;
135 
136 	max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
137 	max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
138 	max_sectors = round_down(max_sectors,
139 				 limits->logical_block_size >> SECTOR_SHIFT);
140 	limits->max_sectors = max_sectors;
141 
142 	if (!q->disk)
143 		return;
144 	q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
145 }
146 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
147 
148 /**
149  * blk_queue_chunk_sectors - set size of the chunk for this queue
150  * @q:  the request queue for the device
151  * @chunk_sectors:  chunk sectors in the usual 512b unit
152  *
153  * Description:
154  *    If a driver doesn't want IOs to cross a given chunk size, it can set
155  *    this limit and prevent merging across chunks. Note that the block layer
156  *    must accept a page worth of data at any offset. So if the crossing of
157  *    chunks is a hard limitation in the driver, it must still be prepared
158  *    to split single page bios.
159  **/
160 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
161 {
162 	q->limits.chunk_sectors = chunk_sectors;
163 }
164 EXPORT_SYMBOL(blk_queue_chunk_sectors);
165 
166 /**
167  * blk_queue_max_discard_sectors - set max sectors for a single discard
168  * @q:  the request queue for the device
169  * @max_discard_sectors: maximum number of sectors to discard
170  **/
171 void blk_queue_max_discard_sectors(struct request_queue *q,
172 		unsigned int max_discard_sectors)
173 {
174 	q->limits.max_hw_discard_sectors = max_discard_sectors;
175 	q->limits.max_discard_sectors = max_discard_sectors;
176 }
177 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
178 
179 /**
180  * blk_queue_max_write_zeroes_sectors - set max sectors for a single
181  *                                      write zeroes
182  * @q:  the request queue for the device
183  * @max_write_zeroes_sectors: maximum number of sectors to write per command
184  **/
185 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
186 		unsigned int max_write_zeroes_sectors)
187 {
188 	q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
189 }
190 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
191 
192 /**
193  * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
194  * @q:  the request queue for the device
195  * @max_zone_append_sectors: maximum number of sectors to write per command
196  **/
197 void blk_queue_max_zone_append_sectors(struct request_queue *q,
198 		unsigned int max_zone_append_sectors)
199 {
200 	unsigned int max_sectors;
201 
202 	if (WARN_ON(!blk_queue_is_zoned(q)))
203 		return;
204 
205 	max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
206 	max_sectors = min(q->limits.chunk_sectors, max_sectors);
207 
208 	/*
209 	 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
210 	 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
211 	 * or the max_hw_sectors limit not set.
212 	 */
213 	WARN_ON(!max_sectors);
214 
215 	q->limits.max_zone_append_sectors = max_sectors;
216 }
217 EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
218 
219 /**
220  * blk_queue_max_segments - set max hw segments for a request for this queue
221  * @q:  the request queue for the device
222  * @max_segments:  max number of segments
223  *
224  * Description:
225  *    Enables a low level driver to set an upper limit on the number of
226  *    hw data segments in a request.
227  **/
228 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
229 {
230 	if (!max_segments) {
231 		max_segments = 1;
232 		printk(KERN_INFO "%s: set to minimum %d\n",
233 		       __func__, max_segments);
234 	}
235 
236 	q->limits.max_segments = max_segments;
237 }
238 EXPORT_SYMBOL(blk_queue_max_segments);
239 
240 /**
241  * blk_queue_max_discard_segments - set max segments for discard requests
242  * @q:  the request queue for the device
243  * @max_segments:  max number of segments
244  *
245  * Description:
246  *    Enables a low level driver to set an upper limit on the number of
247  *    segments in a discard request.
248  **/
249 void blk_queue_max_discard_segments(struct request_queue *q,
250 		unsigned short max_segments)
251 {
252 	q->limits.max_discard_segments = max_segments;
253 }
254 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
255 
256 /**
257  * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
258  * @q:  the request queue for the device
259  * @max_size:  max size of segment in bytes
260  *
261  * Description:
262  *    Enables a low level driver to set an upper limit on the size of a
263  *    coalesced segment
264  **/
265 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
266 {
267 	if (max_size < PAGE_SIZE) {
268 		max_size = PAGE_SIZE;
269 		printk(KERN_INFO "%s: set to minimum %d\n",
270 		       __func__, max_size);
271 	}
272 
273 	/* see blk_queue_virt_boundary() for the explanation */
274 	WARN_ON_ONCE(q->limits.virt_boundary_mask);
275 
276 	q->limits.max_segment_size = max_size;
277 }
278 EXPORT_SYMBOL(blk_queue_max_segment_size);
279 
280 /**
281  * blk_queue_logical_block_size - set logical block size for the queue
282  * @q:  the request queue for the device
283  * @size:  the logical block size, in bytes
284  *
285  * Description:
286  *   This should be set to the lowest possible block size that the
287  *   storage device can address.  The default of 512 covers most
288  *   hardware.
289  **/
290 void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
291 {
292 	struct queue_limits *limits = &q->limits;
293 
294 	limits->logical_block_size = size;
295 
296 	if (limits->physical_block_size < size)
297 		limits->physical_block_size = size;
298 
299 	if (limits->io_min < limits->physical_block_size)
300 		limits->io_min = limits->physical_block_size;
301 
302 	limits->max_hw_sectors =
303 		round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT);
304 	limits->max_sectors =
305 		round_down(limits->max_sectors, size >> SECTOR_SHIFT);
306 }
307 EXPORT_SYMBOL(blk_queue_logical_block_size);
308 
309 /**
310  * blk_queue_physical_block_size - set physical block size for the queue
311  * @q:  the request queue for the device
312  * @size:  the physical block size, in bytes
313  *
314  * Description:
315  *   This should be set to the lowest possible sector size that the
316  *   hardware can operate on without reverting to read-modify-write
317  *   operations.
318  */
319 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
320 {
321 	q->limits.physical_block_size = size;
322 
323 	if (q->limits.physical_block_size < q->limits.logical_block_size)
324 		q->limits.physical_block_size = q->limits.logical_block_size;
325 
326 	if (q->limits.io_min < q->limits.physical_block_size)
327 		q->limits.io_min = q->limits.physical_block_size;
328 }
329 EXPORT_SYMBOL(blk_queue_physical_block_size);
330 
331 /**
332  * blk_queue_zone_write_granularity - set zone write granularity for the queue
333  * @q:  the request queue for the zoned device
334  * @size:  the zone write granularity size, in bytes
335  *
336  * Description:
337  *   This should be set to the lowest possible size allowing to write in
338  *   sequential zones of a zoned block device.
339  */
340 void blk_queue_zone_write_granularity(struct request_queue *q,
341 				      unsigned int size)
342 {
343 	if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
344 		return;
345 
346 	q->limits.zone_write_granularity = size;
347 
348 	if (q->limits.zone_write_granularity < q->limits.logical_block_size)
349 		q->limits.zone_write_granularity = q->limits.logical_block_size;
350 }
351 EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity);
352 
353 /**
354  * blk_queue_alignment_offset - set physical block alignment offset
355  * @q:	the request queue for the device
356  * @offset: alignment offset in bytes
357  *
358  * Description:
359  *   Some devices are naturally misaligned to compensate for things like
360  *   the legacy DOS partition table 63-sector offset.  Low-level drivers
361  *   should call this function for devices whose first sector is not
362  *   naturally aligned.
363  */
364 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
365 {
366 	q->limits.alignment_offset =
367 		offset & (q->limits.physical_block_size - 1);
368 	q->limits.misaligned = 0;
369 }
370 EXPORT_SYMBOL(blk_queue_alignment_offset);
371 
372 void disk_update_readahead(struct gendisk *disk)
373 {
374 	struct request_queue *q = disk->queue;
375 
376 	/*
377 	 * For read-ahead of large files to be effective, we need to read ahead
378 	 * at least twice the optimal I/O size.
379 	 */
380 	disk->bdi->ra_pages =
381 		max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
382 	disk->bdi->io_pages = queue_max_sectors(q) >> (PAGE_SHIFT - 9);
383 }
384 EXPORT_SYMBOL_GPL(disk_update_readahead);
385 
386 /**
387  * blk_limits_io_min - set minimum request size for a device
388  * @limits: the queue limits
389  * @min:  smallest I/O size in bytes
390  *
391  * Description:
392  *   Some devices have an internal block size bigger than the reported
393  *   hardware sector size.  This function can be used to signal the
394  *   smallest I/O the device can perform without incurring a performance
395  *   penalty.
396  */
397 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
398 {
399 	limits->io_min = min;
400 
401 	if (limits->io_min < limits->logical_block_size)
402 		limits->io_min = limits->logical_block_size;
403 
404 	if (limits->io_min < limits->physical_block_size)
405 		limits->io_min = limits->physical_block_size;
406 }
407 EXPORT_SYMBOL(blk_limits_io_min);
408 
409 /**
410  * blk_queue_io_min - set minimum request size for the queue
411  * @q:	the request queue for the device
412  * @min:  smallest I/O size in bytes
413  *
414  * Description:
415  *   Storage devices may report a granularity or preferred minimum I/O
416  *   size which is the smallest request the device can perform without
417  *   incurring a performance penalty.  For disk drives this is often the
418  *   physical block size.  For RAID arrays it is often the stripe chunk
419  *   size.  A properly aligned multiple of minimum_io_size is the
420  *   preferred request size for workloads where a high number of I/O
421  *   operations is desired.
422  */
423 void blk_queue_io_min(struct request_queue *q, unsigned int min)
424 {
425 	blk_limits_io_min(&q->limits, min);
426 }
427 EXPORT_SYMBOL(blk_queue_io_min);
428 
429 /**
430  * blk_limits_io_opt - set optimal request size for a device
431  * @limits: the queue limits
432  * @opt:  smallest I/O size in bytes
433  *
434  * Description:
435  *   Storage devices may report an optimal I/O size, which is the
436  *   device's preferred unit for sustained I/O.  This is rarely reported
437  *   for disk drives.  For RAID arrays it is usually the stripe width or
438  *   the internal track size.  A properly aligned multiple of
439  *   optimal_io_size is the preferred request size for workloads where
440  *   sustained throughput is desired.
441  */
442 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
443 {
444 	limits->io_opt = opt;
445 }
446 EXPORT_SYMBOL(blk_limits_io_opt);
447 
448 /**
449  * blk_queue_io_opt - set optimal request size for the queue
450  * @q:	the request queue for the device
451  * @opt:  optimal request size in bytes
452  *
453  * Description:
454  *   Storage devices may report an optimal I/O size, which is the
455  *   device's preferred unit for sustained I/O.  This is rarely reported
456  *   for disk drives.  For RAID arrays it is usually the stripe width or
457  *   the internal track size.  A properly aligned multiple of
458  *   optimal_io_size is the preferred request size for workloads where
459  *   sustained throughput is desired.
460  */
461 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
462 {
463 	blk_limits_io_opt(&q->limits, opt);
464 	if (!q->disk)
465 		return;
466 	q->disk->bdi->ra_pages =
467 		max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
468 }
469 EXPORT_SYMBOL(blk_queue_io_opt);
470 
471 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
472 {
473 	sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
474 	if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
475 		sectors = PAGE_SIZE >> SECTOR_SHIFT;
476 	return sectors;
477 }
478 
479 /**
480  * blk_stack_limits - adjust queue_limits for stacked devices
481  * @t:	the stacking driver limits (top device)
482  * @b:  the underlying queue limits (bottom, component device)
483  * @start:  first data sector within component device
484  *
485  * Description:
486  *    This function is used by stacking drivers like MD and DM to ensure
487  *    that all component devices have compatible block sizes and
488  *    alignments.  The stacking driver must provide a queue_limits
489  *    struct (top) and then iteratively call the stacking function for
490  *    all component (bottom) devices.  The stacking function will
491  *    attempt to combine the values and ensure proper alignment.
492  *
493  *    Returns 0 if the top and bottom queue_limits are compatible.  The
494  *    top device's block sizes and alignment offsets may be adjusted to
495  *    ensure alignment with the bottom device. If no compatible sizes
496  *    and alignments exist, -1 is returned and the resulting top
497  *    queue_limits will have the misaligned flag set to indicate that
498  *    the alignment_offset is undefined.
499  */
500 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
501 		     sector_t start)
502 {
503 	unsigned int top, bottom, alignment, ret = 0;
504 
505 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
506 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
507 	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
508 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
509 					b->max_write_zeroes_sectors);
510 	t->max_zone_append_sectors = min(t->max_zone_append_sectors,
511 					b->max_zone_append_sectors);
512 	t->bounce = max(t->bounce, b->bounce);
513 
514 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
515 					    b->seg_boundary_mask);
516 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
517 					    b->virt_boundary_mask);
518 
519 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
520 	t->max_discard_segments = min_not_zero(t->max_discard_segments,
521 					       b->max_discard_segments);
522 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
523 						 b->max_integrity_segments);
524 
525 	t->max_segment_size = min_not_zero(t->max_segment_size,
526 					   b->max_segment_size);
527 
528 	t->misaligned |= b->misaligned;
529 
530 	alignment = queue_limit_alignment_offset(b, start);
531 
532 	/* Bottom device has different alignment.  Check that it is
533 	 * compatible with the current top alignment.
534 	 */
535 	if (t->alignment_offset != alignment) {
536 
537 		top = max(t->physical_block_size, t->io_min)
538 			+ t->alignment_offset;
539 		bottom = max(b->physical_block_size, b->io_min) + alignment;
540 
541 		/* Verify that top and bottom intervals line up */
542 		if (max(top, bottom) % min(top, bottom)) {
543 			t->misaligned = 1;
544 			ret = -1;
545 		}
546 	}
547 
548 	t->logical_block_size = max(t->logical_block_size,
549 				    b->logical_block_size);
550 
551 	t->physical_block_size = max(t->physical_block_size,
552 				     b->physical_block_size);
553 
554 	t->io_min = max(t->io_min, b->io_min);
555 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
556 
557 	/* Set non-power-of-2 compatible chunk_sectors boundary */
558 	if (b->chunk_sectors)
559 		t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
560 
561 	/* Physical block size a multiple of the logical block size? */
562 	if (t->physical_block_size & (t->logical_block_size - 1)) {
563 		t->physical_block_size = t->logical_block_size;
564 		t->misaligned = 1;
565 		ret = -1;
566 	}
567 
568 	/* Minimum I/O a multiple of the physical block size? */
569 	if (t->io_min & (t->physical_block_size - 1)) {
570 		t->io_min = t->physical_block_size;
571 		t->misaligned = 1;
572 		ret = -1;
573 	}
574 
575 	/* Optimal I/O a multiple of the physical block size? */
576 	if (t->io_opt & (t->physical_block_size - 1)) {
577 		t->io_opt = 0;
578 		t->misaligned = 1;
579 		ret = -1;
580 	}
581 
582 	/* chunk_sectors a multiple of the physical block size? */
583 	if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
584 		t->chunk_sectors = 0;
585 		t->misaligned = 1;
586 		ret = -1;
587 	}
588 
589 	t->raid_partial_stripes_expensive =
590 		max(t->raid_partial_stripes_expensive,
591 		    b->raid_partial_stripes_expensive);
592 
593 	/* Find lowest common alignment_offset */
594 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
595 		% max(t->physical_block_size, t->io_min);
596 
597 	/* Verify that new alignment_offset is on a logical block boundary */
598 	if (t->alignment_offset & (t->logical_block_size - 1)) {
599 		t->misaligned = 1;
600 		ret = -1;
601 	}
602 
603 	t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
604 	t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
605 	t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
606 
607 	/* Discard alignment and granularity */
608 	if (b->discard_granularity) {
609 		alignment = queue_limit_discard_alignment(b, start);
610 
611 		if (t->discard_granularity != 0 &&
612 		    t->discard_alignment != alignment) {
613 			top = t->discard_granularity + t->discard_alignment;
614 			bottom = b->discard_granularity + alignment;
615 
616 			/* Verify that top and bottom intervals line up */
617 			if ((max(top, bottom) % min(top, bottom)) != 0)
618 				t->discard_misaligned = 1;
619 		}
620 
621 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
622 						      b->max_discard_sectors);
623 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
624 							 b->max_hw_discard_sectors);
625 		t->discard_granularity = max(t->discard_granularity,
626 					     b->discard_granularity);
627 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
628 			t->discard_granularity;
629 	}
630 
631 	t->zone_write_granularity = max(t->zone_write_granularity,
632 					b->zone_write_granularity);
633 	t->zoned = max(t->zoned, b->zoned);
634 	return ret;
635 }
636 EXPORT_SYMBOL(blk_stack_limits);
637 
638 /**
639  * disk_stack_limits - adjust queue limits for stacked drivers
640  * @disk:  MD/DM gendisk (top)
641  * @bdev:  the underlying block device (bottom)
642  * @offset:  offset to beginning of data within component device
643  *
644  * Description:
645  *    Merges the limits for a top level gendisk and a bottom level
646  *    block_device.
647  */
648 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
649 		       sector_t offset)
650 {
651 	struct request_queue *t = disk->queue;
652 
653 	if (blk_stack_limits(&t->limits, &bdev_get_queue(bdev)->limits,
654 			get_start_sect(bdev) + (offset >> 9)) < 0)
655 		pr_notice("%s: Warning: Device %pg is misaligned\n",
656 			disk->disk_name, bdev);
657 
658 	disk_update_readahead(disk);
659 }
660 EXPORT_SYMBOL(disk_stack_limits);
661 
662 /**
663  * blk_queue_update_dma_pad - update pad mask
664  * @q:     the request queue for the device
665  * @mask:  pad mask
666  *
667  * Update dma pad mask.
668  *
669  * Appending pad buffer to a request modifies the last entry of a
670  * scatter list such that it includes the pad buffer.
671  **/
672 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
673 {
674 	if (mask > q->dma_pad_mask)
675 		q->dma_pad_mask = mask;
676 }
677 EXPORT_SYMBOL(blk_queue_update_dma_pad);
678 
679 /**
680  * blk_queue_segment_boundary - set boundary rules for segment merging
681  * @q:  the request queue for the device
682  * @mask:  the memory boundary mask
683  **/
684 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
685 {
686 	if (mask < PAGE_SIZE - 1) {
687 		mask = PAGE_SIZE - 1;
688 		printk(KERN_INFO "%s: set to minimum %lx\n",
689 		       __func__, mask);
690 	}
691 
692 	q->limits.seg_boundary_mask = mask;
693 }
694 EXPORT_SYMBOL(blk_queue_segment_boundary);
695 
696 /**
697  * blk_queue_virt_boundary - set boundary rules for bio merging
698  * @q:  the request queue for the device
699  * @mask:  the memory boundary mask
700  **/
701 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
702 {
703 	q->limits.virt_boundary_mask = mask;
704 
705 	/*
706 	 * Devices that require a virtual boundary do not support scatter/gather
707 	 * I/O natively, but instead require a descriptor list entry for each
708 	 * page (which might not be idential to the Linux PAGE_SIZE).  Because
709 	 * of that they are not limited by our notion of "segment size".
710 	 */
711 	if (mask)
712 		q->limits.max_segment_size = UINT_MAX;
713 }
714 EXPORT_SYMBOL(blk_queue_virt_boundary);
715 
716 /**
717  * blk_queue_dma_alignment - set dma length and memory alignment
718  * @q:     the request queue for the device
719  * @mask:  alignment mask
720  *
721  * description:
722  *    set required memory and length alignment for direct dma transactions.
723  *    this is used when building direct io requests for the queue.
724  *
725  **/
726 void blk_queue_dma_alignment(struct request_queue *q, int mask)
727 {
728 	q->dma_alignment = mask;
729 }
730 EXPORT_SYMBOL(blk_queue_dma_alignment);
731 
732 /**
733  * blk_queue_update_dma_alignment - update dma length and memory alignment
734  * @q:     the request queue for the device
735  * @mask:  alignment mask
736  *
737  * description:
738  *    update required memory and length alignment for direct dma transactions.
739  *    If the requested alignment is larger than the current alignment, then
740  *    the current queue alignment is updated to the new value, otherwise it
741  *    is left alone.  The design of this is to allow multiple objects
742  *    (driver, device, transport etc) to set their respective
743  *    alignments without having them interfere.
744  *
745  **/
746 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
747 {
748 	BUG_ON(mask > PAGE_SIZE);
749 
750 	if (mask > q->dma_alignment)
751 		q->dma_alignment = mask;
752 }
753 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
754 
755 /**
756  * blk_set_queue_depth - tell the block layer about the device queue depth
757  * @q:		the request queue for the device
758  * @depth:		queue depth
759  *
760  */
761 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
762 {
763 	q->queue_depth = depth;
764 	rq_qos_queue_depth_changed(q);
765 }
766 EXPORT_SYMBOL(blk_set_queue_depth);
767 
768 /**
769  * blk_queue_write_cache - configure queue's write cache
770  * @q:		the request queue for the device
771  * @wc:		write back cache on or off
772  * @fua:	device supports FUA writes, if true
773  *
774  * Tell the block layer about the write cache of @q.
775  */
776 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
777 {
778 	if (wc)
779 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
780 	else
781 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
782 	if (fua)
783 		blk_queue_flag_set(QUEUE_FLAG_FUA, q);
784 	else
785 		blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
786 
787 	wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
788 }
789 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
790 
791 /**
792  * blk_queue_required_elevator_features - Set a queue required elevator features
793  * @q:		the request queue for the target device
794  * @features:	Required elevator features OR'ed together
795  *
796  * Tell the block layer that for the device controlled through @q, only the
797  * only elevators that can be used are those that implement at least the set of
798  * features specified by @features.
799  */
800 void blk_queue_required_elevator_features(struct request_queue *q,
801 					  unsigned int features)
802 {
803 	q->required_elevator_features = features;
804 }
805 EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
806 
807 /**
808  * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
809  * @q:		the request queue for the device
810  * @dev:	the device pointer for dma
811  *
812  * Tell the block layer about merging the segments by dma map of @q.
813  */
814 bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
815 				       struct device *dev)
816 {
817 	unsigned long boundary = dma_get_merge_boundary(dev);
818 
819 	if (!boundary)
820 		return false;
821 
822 	/* No need to update max_segment_size. see blk_queue_virt_boundary() */
823 	blk_queue_virt_boundary(q, boundary);
824 
825 	return true;
826 }
827 EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
828 
829 static bool disk_has_partitions(struct gendisk *disk)
830 {
831 	unsigned long idx;
832 	struct block_device *part;
833 	bool ret = false;
834 
835 	rcu_read_lock();
836 	xa_for_each(&disk->part_tbl, idx, part) {
837 		if (bdev_is_partition(part)) {
838 			ret = true;
839 			break;
840 		}
841 	}
842 	rcu_read_unlock();
843 
844 	return ret;
845 }
846 
847 /**
848  * blk_queue_set_zoned - configure a disk queue zoned model.
849  * @disk:	the gendisk of the queue to configure
850  * @model:	the zoned model to set
851  *
852  * Set the zoned model of the request queue of @disk according to @model.
853  * When @model is BLK_ZONED_HM (host managed), this should be called only
854  * if zoned block device support is enabled (CONFIG_BLK_DEV_ZONED option).
855  * If @model specifies BLK_ZONED_HA (host aware), the effective model used
856  * depends on CONFIG_BLK_DEV_ZONED settings and on the existence of partitions
857  * on the disk.
858  */
859 void blk_queue_set_zoned(struct gendisk *disk, enum blk_zoned_model model)
860 {
861 	struct request_queue *q = disk->queue;
862 
863 	switch (model) {
864 	case BLK_ZONED_HM:
865 		/*
866 		 * Host managed devices are supported only if
867 		 * CONFIG_BLK_DEV_ZONED is enabled.
868 		 */
869 		WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED));
870 		break;
871 	case BLK_ZONED_HA:
872 		/*
873 		 * Host aware devices can be treated either as regular block
874 		 * devices (similar to drive managed devices) or as zoned block
875 		 * devices to take advantage of the zone command set, similarly
876 		 * to host managed devices. We try the latter if there are no
877 		 * partitions and zoned block device support is enabled, else
878 		 * we do nothing special as far as the block layer is concerned.
879 		 */
880 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) ||
881 		    disk_has_partitions(disk))
882 			model = BLK_ZONED_NONE;
883 		break;
884 	case BLK_ZONED_NONE:
885 	default:
886 		if (WARN_ON_ONCE(model != BLK_ZONED_NONE))
887 			model = BLK_ZONED_NONE;
888 		break;
889 	}
890 
891 	q->limits.zoned = model;
892 	if (model != BLK_ZONED_NONE) {
893 		/*
894 		 * Set the zone write granularity to the device logical block
895 		 * size by default. The driver can change this value if needed.
896 		 */
897 		blk_queue_zone_write_granularity(q,
898 						queue_logical_block_size(q));
899 	} else {
900 		blk_queue_clear_zone_settings(q);
901 	}
902 }
903 EXPORT_SYMBOL_GPL(blk_queue_set_zoned);
904