xref: /linux/block/blk-settings.c (revision 5b026e34120766408e76ba19a0e33a9dc996f9f0)
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 
22 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
23 {
24 	q->rq_timeout = timeout;
25 }
26 EXPORT_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  */
35 void 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_zone_append_sectors = UINT_MAX;
54 	lim->max_user_discard_sectors = UINT_MAX;
55 }
56 EXPORT_SYMBOL(blk_set_stacking_limits);
57 
58 void 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 
69 static 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 	if (lim->max_zone_append_sectors) {
95 		/*
96 		 * The Zone Append size is limited by the maximum I/O size
97 		 * and the zone size given that it can't span zones.
98 		 */
99 		lim->max_zone_append_sectors =
100 			min3(lim->max_hw_sectors,
101 			     lim->max_zone_append_sectors,
102 			     lim->chunk_sectors);
103 	}
104 
105 	return 0;
106 }
107 
108 static int blk_validate_integrity_limits(struct queue_limits *lim)
109 {
110 	struct blk_integrity *bi = &lim->integrity;
111 
112 	if (!bi->tuple_size) {
113 		if (bi->csum_type != BLK_INTEGRITY_CSUM_NONE ||
114 		    bi->tag_size || ((bi->flags & BLK_INTEGRITY_REF_TAG))) {
115 			pr_warn("invalid PI settings.\n");
116 			return -EINVAL;
117 		}
118 		return 0;
119 	}
120 
121 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) {
122 		pr_warn("integrity support disabled.\n");
123 		return -EINVAL;
124 	}
125 
126 	if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE &&
127 	    (bi->flags & BLK_INTEGRITY_REF_TAG)) {
128 		pr_warn("ref tag not support without checksum.\n");
129 		return -EINVAL;
130 	}
131 
132 	if (!bi->interval_exp)
133 		bi->interval_exp = ilog2(lim->logical_block_size);
134 
135 	return 0;
136 }
137 
138 /*
139  * Returns max guaranteed bytes which we can fit in a bio.
140  *
141  * We request that an atomic_write is ITER_UBUF iov_iter (so a single vector),
142  * so we assume that we can fit in at least PAGE_SIZE in a segment, apart from
143  * the first and last segments.
144  */
145 static unsigned int blk_queue_max_guaranteed_bio(struct queue_limits *lim)
146 {
147 	unsigned int max_segments = min(BIO_MAX_VECS, lim->max_segments);
148 	unsigned int length;
149 
150 	length = min(max_segments, 2) * lim->logical_block_size;
151 	if (max_segments > 2)
152 		length += (max_segments - 2) * PAGE_SIZE;
153 
154 	return length;
155 }
156 
157 static void blk_atomic_writes_update_limits(struct queue_limits *lim)
158 {
159 	unsigned int unit_limit = min(lim->max_hw_sectors << SECTOR_SHIFT,
160 					blk_queue_max_guaranteed_bio(lim));
161 
162 	unit_limit = rounddown_pow_of_two(unit_limit);
163 
164 	lim->atomic_write_max_sectors =
165 		min(lim->atomic_write_hw_max >> SECTOR_SHIFT,
166 			lim->max_hw_sectors);
167 	lim->atomic_write_unit_min =
168 		min(lim->atomic_write_hw_unit_min, unit_limit);
169 	lim->atomic_write_unit_max =
170 		min(lim->atomic_write_hw_unit_max, unit_limit);
171 	lim->atomic_write_boundary_sectors =
172 		lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
173 }
174 
175 static void blk_validate_atomic_write_limits(struct queue_limits *lim)
176 {
177 	unsigned int boundary_sectors;
178 
179 	if (!lim->atomic_write_hw_max)
180 		goto unsupported;
181 
182 	boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
183 
184 	if (boundary_sectors) {
185 		/*
186 		 * A feature of boundary support is that it disallows bios to
187 		 * be merged which would result in a merged request which
188 		 * crosses either a chunk sector or atomic write HW boundary,
189 		 * even though chunk sectors may be just set for performance.
190 		 * For simplicity, disallow atomic writes for a chunk sector
191 		 * which is non-zero and smaller than atomic write HW boundary.
192 		 * Furthermore, chunk sectors must be a multiple of atomic
193 		 * write HW boundary. Otherwise boundary support becomes
194 		 * complicated.
195 		 * Devices which do not conform to these rules can be dealt
196 		 * with if and when they show up.
197 		 */
198 		if (WARN_ON_ONCE(lim->chunk_sectors % boundary_sectors))
199 			goto unsupported;
200 
201 		/*
202 		 * The boundary size just needs to be a multiple of unit_max
203 		 * (and not necessarily a power-of-2), so this following check
204 		 * could be relaxed in future.
205 		 * Furthermore, if needed, unit_max could even be reduced so
206 		 * that it is compliant with a !power-of-2 boundary.
207 		 */
208 		if (!is_power_of_2(boundary_sectors))
209 			goto unsupported;
210 	}
211 
212 	blk_atomic_writes_update_limits(lim);
213 	return;
214 
215 unsupported:
216 	lim->atomic_write_max_sectors = 0;
217 	lim->atomic_write_boundary_sectors = 0;
218 	lim->atomic_write_unit_min = 0;
219 	lim->atomic_write_unit_max = 0;
220 }
221 
222 /*
223  * Check that the limits in lim are valid, initialize defaults for unset
224  * values, and cap values based on others where needed.
225  */
226 static int blk_validate_limits(struct queue_limits *lim)
227 {
228 	unsigned int max_hw_sectors;
229 	unsigned int logical_block_sectors;
230 	int err;
231 
232 	/*
233 	 * Unless otherwise specified, default to 512 byte logical blocks and a
234 	 * physical block size equal to the logical block size.
235 	 */
236 	if (!lim->logical_block_size)
237 		lim->logical_block_size = SECTOR_SIZE;
238 	if (lim->physical_block_size < lim->logical_block_size)
239 		lim->physical_block_size = lim->logical_block_size;
240 
241 	/*
242 	 * The minimum I/O size defaults to the physical block size unless
243 	 * explicitly overridden.
244 	 */
245 	if (lim->io_min < lim->physical_block_size)
246 		lim->io_min = lim->physical_block_size;
247 
248 	/*
249 	 * max_hw_sectors has a somewhat weird default for historical reason,
250 	 * but driver really should set their own instead of relying on this
251 	 * value.
252 	 *
253 	 * The block layer relies on the fact that every driver can
254 	 * handle at lest a page worth of data per I/O, and needs the value
255 	 * aligned to the logical block size.
256 	 */
257 	if (!lim->max_hw_sectors)
258 		lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
259 	if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
260 		return -EINVAL;
261 	logical_block_sectors = lim->logical_block_size >> SECTOR_SHIFT;
262 	if (WARN_ON_ONCE(logical_block_sectors > lim->max_hw_sectors))
263 		return -EINVAL;
264 	lim->max_hw_sectors = round_down(lim->max_hw_sectors,
265 			logical_block_sectors);
266 
267 	/*
268 	 * The actual max_sectors value is a complex beast and also takes the
269 	 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
270 	 * value into account.  The ->max_sectors value is always calculated
271 	 * from these, so directly setting it won't have any effect.
272 	 */
273 	max_hw_sectors = min_not_zero(lim->max_hw_sectors,
274 				lim->max_dev_sectors);
275 	if (lim->max_user_sectors) {
276 		if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
277 			return -EINVAL;
278 		lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
279 	} else if (lim->io_opt) {
280 		lim->max_sectors =
281 			min(max_hw_sectors, lim->io_opt >> SECTOR_SHIFT);
282 	} else if (lim->io_min &&
283 		   lim->io_min > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
284 		lim->max_sectors =
285 			min(max_hw_sectors, lim->io_min >> SECTOR_SHIFT);
286 	} else {
287 		lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
288 	}
289 	lim->max_sectors = round_down(lim->max_sectors,
290 			logical_block_sectors);
291 
292 	/*
293 	 * Random default for the maximum number of segments.  Driver should not
294 	 * rely on this and set their own.
295 	 */
296 	if (!lim->max_segments)
297 		lim->max_segments = BLK_MAX_SEGMENTS;
298 
299 	lim->max_discard_sectors =
300 		min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
301 
302 	if (!lim->max_discard_segments)
303 		lim->max_discard_segments = 1;
304 
305 	if (lim->discard_granularity < lim->physical_block_size)
306 		lim->discard_granularity = lim->physical_block_size;
307 
308 	/*
309 	 * By default there is no limit on the segment boundary alignment,
310 	 * but if there is one it can't be smaller than the page size as
311 	 * that would break all the normal I/O patterns.
312 	 */
313 	if (!lim->seg_boundary_mask)
314 		lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
315 	if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1))
316 		return -EINVAL;
317 
318 	/*
319 	 * Stacking device may have both virtual boundary and max segment
320 	 * size limit, so allow this setting now, and long-term the two
321 	 * might need to move out of stacking limits since we have immutable
322 	 * bvec and lower layer bio splitting is supposed to handle the two
323 	 * correctly.
324 	 */
325 	if (lim->virt_boundary_mask) {
326 		if (!lim->max_segment_size)
327 			lim->max_segment_size = UINT_MAX;
328 	} else {
329 		/*
330 		 * The maximum segment size has an odd historic 64k default that
331 		 * drivers probably should override.  Just like the I/O size we
332 		 * require drivers to at least handle a full page per segment.
333 		 */
334 		if (!lim->max_segment_size)
335 			lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
336 		if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE))
337 			return -EINVAL;
338 	}
339 
340 	/*
341 	 * We require drivers to at least do logical block aligned I/O, but
342 	 * historically could not check for that due to the separate calls
343 	 * to set the limits.  Once the transition is finished the check
344 	 * below should be narrowed down to check the logical block size.
345 	 */
346 	if (!lim->dma_alignment)
347 		lim->dma_alignment = SECTOR_SIZE - 1;
348 	if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
349 		return -EINVAL;
350 
351 	if (lim->alignment_offset) {
352 		lim->alignment_offset &= (lim->physical_block_size - 1);
353 		lim->flags &= ~BLK_FLAG_MISALIGNED;
354 	}
355 
356 	if (!(lim->features & BLK_FEAT_WRITE_CACHE))
357 		lim->features &= ~BLK_FEAT_FUA;
358 
359 	blk_validate_atomic_write_limits(lim);
360 
361 	err = blk_validate_integrity_limits(lim);
362 	if (err)
363 		return err;
364 	return blk_validate_zoned_limits(lim);
365 }
366 
367 /*
368  * Set the default limits for a newly allocated queue.  @lim contains the
369  * initial limits set by the driver, which could be no limit in which case
370  * all fields are cleared to zero.
371  */
372 int blk_set_default_limits(struct queue_limits *lim)
373 {
374 	/*
375 	 * Most defaults are set by capping the bounds in blk_validate_limits,
376 	 * but max_user_discard_sectors is special and needs an explicit
377 	 * initialization to the max value here.
378 	 */
379 	lim->max_user_discard_sectors = UINT_MAX;
380 	return blk_validate_limits(lim);
381 }
382 
383 /**
384  * queue_limits_commit_update - commit an atomic update of queue limits
385  * @q:		queue to update
386  * @lim:	limits to apply
387  *
388  * Apply the limits in @lim that were obtained from queue_limits_start_update()
389  * and updated by the caller to @q.
390  *
391  * Returns 0 if successful, else a negative error code.
392  */
393 int queue_limits_commit_update(struct request_queue *q,
394 		struct queue_limits *lim)
395 {
396 	int error;
397 
398 	error = blk_validate_limits(lim);
399 	if (error)
400 		goto out_unlock;
401 
402 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
403 	if (q->crypto_profile && lim->integrity.tag_size) {
404 		pr_warn("blk-integrity: Integrity and hardware inline encryption are not supported together.\n");
405 		error = -EINVAL;
406 		goto out_unlock;
407 	}
408 #endif
409 
410 	q->limits = *lim;
411 	if (q->disk)
412 		blk_apply_bdi_limits(q->disk->bdi, lim);
413 out_unlock:
414 	mutex_unlock(&q->limits_lock);
415 	return error;
416 }
417 EXPORT_SYMBOL_GPL(queue_limits_commit_update);
418 
419 /**
420  * queue_limits_set - apply queue limits to queue
421  * @q:		queue to update
422  * @lim:	limits to apply
423  *
424  * Apply the limits in @lim that were freshly initialized to @q.
425  * To update existing limits use queue_limits_start_update() and
426  * queue_limits_commit_update() instead.
427  *
428  * Returns 0 if successful, else a negative error code.
429  */
430 int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
431 {
432 	mutex_lock(&q->limits_lock);
433 	return queue_limits_commit_update(q, lim);
434 }
435 EXPORT_SYMBOL_GPL(queue_limits_set);
436 
437 /**
438  * blk_limits_io_min - set minimum request size for a device
439  * @limits: the queue limits
440  * @min:  smallest I/O size in bytes
441  *
442  * Description:
443  *   Some devices have an internal block size bigger than the reported
444  *   hardware sector size.  This function can be used to signal the
445  *   smallest I/O the device can perform without incurring a performance
446  *   penalty.
447  */
448 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
449 {
450 	limits->io_min = min;
451 
452 	if (limits->io_min < limits->logical_block_size)
453 		limits->io_min = limits->logical_block_size;
454 
455 	if (limits->io_min < limits->physical_block_size)
456 		limits->io_min = limits->physical_block_size;
457 }
458 EXPORT_SYMBOL(blk_limits_io_min);
459 
460 /**
461  * blk_limits_io_opt - set optimal request size for a device
462  * @limits: the queue limits
463  * @opt:  smallest I/O size in bytes
464  *
465  * Description:
466  *   Storage devices may report an optimal I/O size, which is the
467  *   device's preferred unit for sustained I/O.  This is rarely reported
468  *   for disk drives.  For RAID arrays it is usually the stripe width or
469  *   the internal track size.  A properly aligned multiple of
470  *   optimal_io_size is the preferred request size for workloads where
471  *   sustained throughput is desired.
472  */
473 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
474 {
475 	limits->io_opt = opt;
476 }
477 EXPORT_SYMBOL(blk_limits_io_opt);
478 
479 static int queue_limit_alignment_offset(const struct queue_limits *lim,
480 		sector_t sector)
481 {
482 	unsigned int granularity = max(lim->physical_block_size, lim->io_min);
483 	unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
484 		<< SECTOR_SHIFT;
485 
486 	return (granularity + lim->alignment_offset - alignment) % granularity;
487 }
488 
489 static unsigned int queue_limit_discard_alignment(
490 		const struct queue_limits *lim, sector_t sector)
491 {
492 	unsigned int alignment, granularity, offset;
493 
494 	if (!lim->max_discard_sectors)
495 		return 0;
496 
497 	/* Why are these in bytes, not sectors? */
498 	alignment = lim->discard_alignment >> SECTOR_SHIFT;
499 	granularity = lim->discard_granularity >> SECTOR_SHIFT;
500 	if (!granularity)
501 		return 0;
502 
503 	/* Offset of the partition start in 'granularity' sectors */
504 	offset = sector_div(sector, granularity);
505 
506 	/* And why do we do this modulus *again* in blkdev_issue_discard()? */
507 	offset = (granularity + alignment - offset) % granularity;
508 
509 	/* Turn it back into bytes, gaah */
510 	return offset << SECTOR_SHIFT;
511 }
512 
513 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
514 {
515 	sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
516 	if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
517 		sectors = PAGE_SIZE >> SECTOR_SHIFT;
518 	return sectors;
519 }
520 
521 /**
522  * blk_stack_limits - adjust queue_limits for stacked devices
523  * @t:	the stacking driver limits (top device)
524  * @b:  the underlying queue limits (bottom, component device)
525  * @start:  first data sector within component device
526  *
527  * Description:
528  *    This function is used by stacking drivers like MD and DM to ensure
529  *    that all component devices have compatible block sizes and
530  *    alignments.  The stacking driver must provide a queue_limits
531  *    struct (top) and then iteratively call the stacking function for
532  *    all component (bottom) devices.  The stacking function will
533  *    attempt to combine the values and ensure proper alignment.
534  *
535  *    Returns 0 if the top and bottom queue_limits are compatible.  The
536  *    top device's block sizes and alignment offsets may be adjusted to
537  *    ensure alignment with the bottom device. If no compatible sizes
538  *    and alignments exist, -1 is returned and the resulting top
539  *    queue_limits will have the misaligned flag set to indicate that
540  *    the alignment_offset is undefined.
541  */
542 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
543 		     sector_t start)
544 {
545 	unsigned int top, bottom, alignment, ret = 0;
546 
547 	t->features |= (b->features & BLK_FEAT_INHERIT_MASK);
548 
549 	/*
550 	 * BLK_FEAT_NOWAIT and BLK_FEAT_POLL need to be supported both by the
551 	 * stacking driver and all underlying devices.  The stacking driver sets
552 	 * the flags before stacking the limits, and this will clear the flags
553 	 * if any of the underlying devices does not support it.
554 	 */
555 	if (!(b->features & BLK_FEAT_NOWAIT))
556 		t->features &= ~BLK_FEAT_NOWAIT;
557 	if (!(b->features & BLK_FEAT_POLL))
558 		t->features &= ~BLK_FEAT_POLL;
559 
560 	t->flags |= (b->flags & BLK_FLAG_MISALIGNED);
561 
562 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
563 	t->max_user_sectors = min_not_zero(t->max_user_sectors,
564 			b->max_user_sectors);
565 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
566 	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
567 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
568 					b->max_write_zeroes_sectors);
569 	t->max_zone_append_sectors = min(queue_limits_max_zone_append_sectors(t),
570 					 queue_limits_max_zone_append_sectors(b));
571 
572 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
573 					    b->seg_boundary_mask);
574 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
575 					    b->virt_boundary_mask);
576 
577 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
578 	t->max_discard_segments = min_not_zero(t->max_discard_segments,
579 					       b->max_discard_segments);
580 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
581 						 b->max_integrity_segments);
582 
583 	t->max_segment_size = min_not_zero(t->max_segment_size,
584 					   b->max_segment_size);
585 
586 	alignment = queue_limit_alignment_offset(b, start);
587 
588 	/* Bottom device has different alignment.  Check that it is
589 	 * compatible with the current top alignment.
590 	 */
591 	if (t->alignment_offset != alignment) {
592 
593 		top = max(t->physical_block_size, t->io_min)
594 			+ t->alignment_offset;
595 		bottom = max(b->physical_block_size, b->io_min) + alignment;
596 
597 		/* Verify that top and bottom intervals line up */
598 		if (max(top, bottom) % min(top, bottom)) {
599 			t->flags |= BLK_FLAG_MISALIGNED;
600 			ret = -1;
601 		}
602 	}
603 
604 	t->logical_block_size = max(t->logical_block_size,
605 				    b->logical_block_size);
606 
607 	t->physical_block_size = max(t->physical_block_size,
608 				     b->physical_block_size);
609 
610 	t->io_min = max(t->io_min, b->io_min);
611 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
612 	t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
613 
614 	/* Set non-power-of-2 compatible chunk_sectors boundary */
615 	if (b->chunk_sectors)
616 		t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
617 
618 	/* Physical block size a multiple of the logical block size? */
619 	if (t->physical_block_size & (t->logical_block_size - 1)) {
620 		t->physical_block_size = t->logical_block_size;
621 		t->flags |= BLK_FLAG_MISALIGNED;
622 		ret = -1;
623 	}
624 
625 	/* Minimum I/O a multiple of the physical block size? */
626 	if (t->io_min & (t->physical_block_size - 1)) {
627 		t->io_min = t->physical_block_size;
628 		t->flags |= BLK_FLAG_MISALIGNED;
629 		ret = -1;
630 	}
631 
632 	/* Optimal I/O a multiple of the physical block size? */
633 	if (t->io_opt & (t->physical_block_size - 1)) {
634 		t->io_opt = 0;
635 		t->flags |= BLK_FLAG_MISALIGNED;
636 		ret = -1;
637 	}
638 
639 	/* chunk_sectors a multiple of the physical block size? */
640 	if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
641 		t->chunk_sectors = 0;
642 		t->flags |= BLK_FLAG_MISALIGNED;
643 		ret = -1;
644 	}
645 
646 	/* Find lowest common alignment_offset */
647 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
648 		% max(t->physical_block_size, t->io_min);
649 
650 	/* Verify that new alignment_offset is on a logical block boundary */
651 	if (t->alignment_offset & (t->logical_block_size - 1)) {
652 		t->flags |= BLK_FLAG_MISALIGNED;
653 		ret = -1;
654 	}
655 
656 	t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
657 	t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
658 	t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
659 
660 	/* Discard alignment and granularity */
661 	if (b->discard_granularity) {
662 		alignment = queue_limit_discard_alignment(b, start);
663 
664 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
665 						      b->max_discard_sectors);
666 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
667 							 b->max_hw_discard_sectors);
668 		t->discard_granularity = max(t->discard_granularity,
669 					     b->discard_granularity);
670 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
671 			t->discard_granularity;
672 	}
673 	t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
674 						   b->max_secure_erase_sectors);
675 	t->zone_write_granularity = max(t->zone_write_granularity,
676 					b->zone_write_granularity);
677 	if (!(t->features & BLK_FEAT_ZONED)) {
678 		t->zone_write_granularity = 0;
679 		t->max_zone_append_sectors = 0;
680 	}
681 	return ret;
682 }
683 EXPORT_SYMBOL(blk_stack_limits);
684 
685 /**
686  * queue_limits_stack_bdev - adjust queue_limits for stacked devices
687  * @t:	the stacking driver limits (top device)
688  * @bdev:  the underlying block device (bottom)
689  * @offset:  offset to beginning of data within component device
690  * @pfx: prefix to use for warnings logged
691  *
692  * Description:
693  *    This function is used by stacking drivers like MD and DM to ensure
694  *    that all component devices have compatible block sizes and
695  *    alignments.  The stacking driver must provide a queue_limits
696  *    struct (top) and then iteratively call the stacking function for
697  *    all component (bottom) devices.  The stacking function will
698  *    attempt to combine the values and ensure proper alignment.
699  */
700 void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
701 		sector_t offset, const char *pfx)
702 {
703 	if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits,
704 			get_start_sect(bdev) + offset))
705 		pr_notice("%s: Warning: Device %pg is misaligned\n",
706 			pfx, bdev);
707 }
708 EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
709 
710 /**
711  * queue_limits_stack_integrity - stack integrity profile
712  * @t: target queue limits
713  * @b: base queue limits
714  *
715  * Check if the integrity profile in the @b can be stacked into the
716  * target @t.  Stacking is possible if either:
717  *
718  *   a) does not have any integrity information stacked into it yet
719  *   b) the integrity profile in @b is identical to the one in @t
720  *
721  * If @b can be stacked into @t, return %true.  Else return %false and clear the
722  * integrity information in @t.
723  */
724 bool queue_limits_stack_integrity(struct queue_limits *t,
725 		struct queue_limits *b)
726 {
727 	struct blk_integrity *ti = &t->integrity;
728 	struct blk_integrity *bi = &b->integrity;
729 
730 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
731 		return true;
732 
733 	if (!ti->tuple_size) {
734 		/* inherit the settings from the first underlying device */
735 		if (!(ti->flags & BLK_INTEGRITY_STACKED)) {
736 			ti->flags = BLK_INTEGRITY_DEVICE_CAPABLE |
737 				(bi->flags & BLK_INTEGRITY_REF_TAG);
738 			ti->csum_type = bi->csum_type;
739 			ti->tuple_size = bi->tuple_size;
740 			ti->pi_offset = bi->pi_offset;
741 			ti->interval_exp = bi->interval_exp;
742 			ti->tag_size = bi->tag_size;
743 			goto done;
744 		}
745 		if (!bi->tuple_size)
746 			goto done;
747 	}
748 
749 	if (ti->tuple_size != bi->tuple_size)
750 		goto incompatible;
751 	if (ti->interval_exp != bi->interval_exp)
752 		goto incompatible;
753 	if (ti->tag_size != bi->tag_size)
754 		goto incompatible;
755 	if (ti->csum_type != bi->csum_type)
756 		goto incompatible;
757 	if ((ti->flags & BLK_INTEGRITY_REF_TAG) !=
758 	    (bi->flags & BLK_INTEGRITY_REF_TAG))
759 		goto incompatible;
760 
761 done:
762 	ti->flags |= BLK_INTEGRITY_STACKED;
763 	return true;
764 
765 incompatible:
766 	memset(ti, 0, sizeof(*ti));
767 	return false;
768 }
769 EXPORT_SYMBOL_GPL(queue_limits_stack_integrity);
770 
771 /**
772  * blk_set_queue_depth - tell the block layer about the device queue depth
773  * @q:		the request queue for the device
774  * @depth:		queue depth
775  *
776  */
777 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
778 {
779 	q->queue_depth = depth;
780 	rq_qos_queue_depth_changed(q);
781 }
782 EXPORT_SYMBOL(blk_set_queue_depth);
783 
784 int bdev_alignment_offset(struct block_device *bdev)
785 {
786 	struct request_queue *q = bdev_get_queue(bdev);
787 
788 	if (q->limits.flags & BLK_FLAG_MISALIGNED)
789 		return -1;
790 	if (bdev_is_partition(bdev))
791 		return queue_limit_alignment_offset(&q->limits,
792 				bdev->bd_start_sect);
793 	return q->limits.alignment_offset;
794 }
795 EXPORT_SYMBOL_GPL(bdev_alignment_offset);
796 
797 unsigned int bdev_discard_alignment(struct block_device *bdev)
798 {
799 	struct request_queue *q = bdev_get_queue(bdev);
800 
801 	if (bdev_is_partition(bdev))
802 		return queue_limit_discard_alignment(&q->limits,
803 				bdev->bd_start_sect);
804 	return q->limits.discard_alignment;
805 }
806 EXPORT_SYMBOL_GPL(bdev_discard_alignment);
807