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 
blk_queue_rq_timeout(struct request_queue * q,unsigned int timeout)22 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
23 {
24 	WRITE_ONCE(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  */
blk_set_stacking_limits(struct queue_limits * lim)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_hw_zone_append_sectors = UINT_MAX;
54 	lim->max_user_discard_sectors = UINT_MAX;
55 }
56 EXPORT_SYMBOL(blk_set_stacking_limits);
57 
blk_apply_bdi_limits(struct backing_dev_info * bdi,struct queue_limits * lim)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 	 * There is no hardware limitation for the read-ahead size and the user
66 	 * might have increased the read-ahead size through sysfs, so don't ever
67 	 * decrease it.
68 	 */
69 	bdi->ra_pages = max3(bdi->ra_pages,
70 				lim->io_opt * 2 / PAGE_SIZE,
71 				VM_READAHEAD_PAGES);
72 	bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT;
73 }
74 
blk_validate_zoned_limits(struct queue_limits * lim)75 static int blk_validate_zoned_limits(struct queue_limits *lim)
76 {
77 	if (!(lim->features & BLK_FEAT_ZONED)) {
78 		if (WARN_ON_ONCE(lim->max_open_zones) ||
79 		    WARN_ON_ONCE(lim->max_active_zones) ||
80 		    WARN_ON_ONCE(lim->zone_write_granularity) ||
81 		    WARN_ON_ONCE(lim->max_zone_append_sectors))
82 			return -EINVAL;
83 		return 0;
84 	}
85 
86 	if (WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)))
87 		return -EINVAL;
88 
89 	/*
90 	 * Given that active zones include open zones, the maximum number of
91 	 * open zones cannot be larger than the maximum number of active zones.
92 	 */
93 	if (lim->max_active_zones &&
94 	    lim->max_open_zones > lim->max_active_zones)
95 		return -EINVAL;
96 
97 	if (lim->zone_write_granularity < lim->logical_block_size)
98 		lim->zone_write_granularity = lim->logical_block_size;
99 
100 	/*
101 	 * The Zone Append size is limited by the maximum I/O size and the zone
102 	 * size given that it can't span zones.
103 	 *
104 	 * If no max_hw_zone_append_sectors limit is provided, the block layer
105 	 * will emulated it, else we're also bound by the hardware limit.
106 	 */
107 	lim->max_zone_append_sectors =
108 		min_not_zero(lim->max_hw_zone_append_sectors,
109 			min(lim->chunk_sectors, lim->max_hw_sectors));
110 	return 0;
111 }
112 
blk_validate_integrity_limits(struct queue_limits * lim)113 static int blk_validate_integrity_limits(struct queue_limits *lim)
114 {
115 	struct blk_integrity *bi = &lim->integrity;
116 
117 	if (!bi->tuple_size) {
118 		if (bi->csum_type != BLK_INTEGRITY_CSUM_NONE ||
119 		    bi->tag_size || ((bi->flags & BLK_INTEGRITY_REF_TAG))) {
120 			pr_warn("invalid PI settings.\n");
121 			return -EINVAL;
122 		}
123 		bi->flags |= BLK_INTEGRITY_NOGENERATE | BLK_INTEGRITY_NOVERIFY;
124 		return 0;
125 	}
126 
127 	if (lim->features & BLK_FEAT_BOUNCE_HIGH) {
128 		pr_warn("no bounce buffer support for integrity metadata\n");
129 		return -EINVAL;
130 	}
131 
132 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) {
133 		pr_warn("integrity support disabled.\n");
134 		return -EINVAL;
135 	}
136 
137 	if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE &&
138 	    (bi->flags & BLK_INTEGRITY_REF_TAG)) {
139 		pr_warn("ref tag not support without checksum.\n");
140 		return -EINVAL;
141 	}
142 
143 	if (!bi->interval_exp)
144 		bi->interval_exp = ilog2(lim->logical_block_size);
145 
146 	return 0;
147 }
148 
149 /*
150  * Returns max guaranteed bytes which we can fit in a bio.
151  *
152  * We request that an atomic_write is ITER_UBUF iov_iter (so a single vector),
153  * so we assume that we can fit in at least PAGE_SIZE in a segment, apart from
154  * the first and last segments.
155  */
blk_queue_max_guaranteed_bio(struct queue_limits * lim)156 static unsigned int blk_queue_max_guaranteed_bio(struct queue_limits *lim)
157 {
158 	unsigned int max_segments = min(BIO_MAX_VECS, lim->max_segments);
159 	unsigned int length;
160 
161 	length = min(max_segments, 2) * lim->logical_block_size;
162 	if (max_segments > 2)
163 		length += (max_segments - 2) * PAGE_SIZE;
164 
165 	return length;
166 }
167 
blk_atomic_writes_update_limits(struct queue_limits * lim)168 static void blk_atomic_writes_update_limits(struct queue_limits *lim)
169 {
170 	unsigned int unit_limit = min(lim->max_hw_sectors << SECTOR_SHIFT,
171 					blk_queue_max_guaranteed_bio(lim));
172 
173 	unit_limit = rounddown_pow_of_two(unit_limit);
174 
175 	lim->atomic_write_max_sectors =
176 		min(lim->atomic_write_hw_max >> SECTOR_SHIFT,
177 			lim->max_hw_sectors);
178 	lim->atomic_write_unit_min =
179 		min(lim->atomic_write_hw_unit_min, unit_limit);
180 	lim->atomic_write_unit_max =
181 		min(lim->atomic_write_hw_unit_max, unit_limit);
182 	lim->atomic_write_boundary_sectors =
183 		lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
184 }
185 
blk_validate_atomic_write_limits(struct queue_limits * lim)186 static void blk_validate_atomic_write_limits(struct queue_limits *lim)
187 {
188 	unsigned int boundary_sectors;
189 
190 	if (!(lim->features & BLK_FEAT_ATOMIC_WRITES))
191 		goto unsupported;
192 
193 	if (!lim->atomic_write_hw_max)
194 		goto unsupported;
195 
196 	if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_min)))
197 		goto unsupported;
198 
199 	if (WARN_ON_ONCE(!is_power_of_2(lim->atomic_write_hw_unit_max)))
200 		goto unsupported;
201 
202 	if (WARN_ON_ONCE(lim->atomic_write_hw_unit_min >
203 			 lim->atomic_write_hw_unit_max))
204 		goto unsupported;
205 
206 	if (WARN_ON_ONCE(lim->atomic_write_hw_unit_max >
207 			 lim->atomic_write_hw_max))
208 		goto unsupported;
209 
210 	boundary_sectors = lim->atomic_write_hw_boundary >> SECTOR_SHIFT;
211 
212 	if (boundary_sectors) {
213 		if (WARN_ON_ONCE(lim->atomic_write_hw_max >
214 				 lim->atomic_write_hw_boundary))
215 			goto unsupported;
216 		/*
217 		 * A feature of boundary support is that it disallows bios to
218 		 * be merged which would result in a merged request which
219 		 * crosses either a chunk sector or atomic write HW boundary,
220 		 * even though chunk sectors may be just set for performance.
221 		 * For simplicity, disallow atomic writes for a chunk sector
222 		 * which is non-zero and smaller than atomic write HW boundary.
223 		 * Furthermore, chunk sectors must be a multiple of atomic
224 		 * write HW boundary. Otherwise boundary support becomes
225 		 * complicated.
226 		 * Devices which do not conform to these rules can be dealt
227 		 * with if and when they show up.
228 		 */
229 		if (WARN_ON_ONCE(lim->chunk_sectors % boundary_sectors))
230 			goto unsupported;
231 
232 		/*
233 		 * The boundary size just needs to be a multiple of unit_max
234 		 * (and not necessarily a power-of-2), so this following check
235 		 * could be relaxed in future.
236 		 * Furthermore, if needed, unit_max could even be reduced so
237 		 * that it is compliant with a !power-of-2 boundary.
238 		 */
239 		if (!is_power_of_2(boundary_sectors))
240 			goto unsupported;
241 	}
242 
243 	blk_atomic_writes_update_limits(lim);
244 	return;
245 
246 unsupported:
247 	lim->atomic_write_max_sectors = 0;
248 	lim->atomic_write_boundary_sectors = 0;
249 	lim->atomic_write_unit_min = 0;
250 	lim->atomic_write_unit_max = 0;
251 }
252 
253 /*
254  * Check that the limits in lim are valid, initialize defaults for unset
255  * values, and cap values based on others where needed.
256  */
blk_validate_limits(struct queue_limits * lim)257 int blk_validate_limits(struct queue_limits *lim)
258 {
259 	unsigned int max_hw_sectors;
260 	unsigned int logical_block_sectors;
261 	unsigned long seg_size;
262 	int err;
263 
264 	/*
265 	 * Unless otherwise specified, default to 512 byte logical blocks and a
266 	 * physical block size equal to the logical block size.
267 	 */
268 	if (!lim->logical_block_size)
269 		lim->logical_block_size = SECTOR_SIZE;
270 	else if (blk_validate_block_size(lim->logical_block_size)) {
271 		pr_warn("Invalid logical block size (%d)\n", lim->logical_block_size);
272 		return -EINVAL;
273 	}
274 	if (lim->physical_block_size < lim->logical_block_size)
275 		lim->physical_block_size = lim->logical_block_size;
276 
277 	/*
278 	 * The minimum I/O size defaults to the physical block size unless
279 	 * explicitly overridden.
280 	 */
281 	if (lim->io_min < lim->physical_block_size)
282 		lim->io_min = lim->physical_block_size;
283 
284 	/*
285 	 * The optimal I/O size may not be aligned to physical block size
286 	 * (because it may be limited by dma engines which have no clue about
287 	 * block size of the disks attached to them), so we round it down here.
288 	 */
289 	lim->io_opt = round_down(lim->io_opt, lim->physical_block_size);
290 
291 	/*
292 	 * max_hw_sectors has a somewhat weird default for historical reason,
293 	 * but driver really should set their own instead of relying on this
294 	 * value.
295 	 *
296 	 * The block layer relies on the fact that every driver can
297 	 * handle at lest a page worth of data per I/O, and needs the value
298 	 * aligned to the logical block size.
299 	 */
300 	if (!lim->max_hw_sectors)
301 		lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
302 	if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS))
303 		return -EINVAL;
304 	logical_block_sectors = lim->logical_block_size >> SECTOR_SHIFT;
305 	if (WARN_ON_ONCE(logical_block_sectors > lim->max_hw_sectors))
306 		return -EINVAL;
307 	lim->max_hw_sectors = round_down(lim->max_hw_sectors,
308 			logical_block_sectors);
309 
310 	/*
311 	 * The actual max_sectors value is a complex beast and also takes the
312 	 * max_dev_sectors value (set by SCSI ULPs) and a user configurable
313 	 * value into account.  The ->max_sectors value is always calculated
314 	 * from these, so directly setting it won't have any effect.
315 	 */
316 	max_hw_sectors = min_not_zero(lim->max_hw_sectors,
317 				lim->max_dev_sectors);
318 	if (lim->max_user_sectors) {
319 		if (lim->max_user_sectors < BLK_MIN_SEGMENT_SIZE / SECTOR_SIZE)
320 			return -EINVAL;
321 		lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors);
322 	} else if (lim->io_opt > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
323 		lim->max_sectors =
324 			min(max_hw_sectors, lim->io_opt >> SECTOR_SHIFT);
325 	} else if (lim->io_min > (BLK_DEF_MAX_SECTORS_CAP << SECTOR_SHIFT)) {
326 		lim->max_sectors =
327 			min(max_hw_sectors, lim->io_min >> SECTOR_SHIFT);
328 	} else {
329 		lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP);
330 	}
331 	lim->max_sectors = round_down(lim->max_sectors,
332 			logical_block_sectors);
333 
334 	/*
335 	 * Random default for the maximum number of segments.  Driver should not
336 	 * rely on this and set their own.
337 	 */
338 	if (!lim->max_segments)
339 		lim->max_segments = BLK_MAX_SEGMENTS;
340 
341 	lim->max_discard_sectors =
342 		min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors);
343 
344 	if (!lim->max_discard_segments)
345 		lim->max_discard_segments = 1;
346 
347 	if (lim->discard_granularity < lim->physical_block_size)
348 		lim->discard_granularity = lim->physical_block_size;
349 
350 	/*
351 	 * By default there is no limit on the segment boundary alignment,
352 	 * but if there is one it can't be smaller than the page size as
353 	 * that would break all the normal I/O patterns.
354 	 */
355 	if (!lim->seg_boundary_mask)
356 		lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
357 	if (WARN_ON_ONCE(lim->seg_boundary_mask < BLK_MIN_SEGMENT_SIZE - 1))
358 		return -EINVAL;
359 
360 	/*
361 	 * Stacking device may have both virtual boundary and max segment
362 	 * size limit, so allow this setting now, and long-term the two
363 	 * might need to move out of stacking limits since we have immutable
364 	 * bvec and lower layer bio splitting is supposed to handle the two
365 	 * correctly.
366 	 */
367 	if (lim->virt_boundary_mask) {
368 		if (!lim->max_segment_size)
369 			lim->max_segment_size = UINT_MAX;
370 	} else {
371 		/*
372 		 * The maximum segment size has an odd historic 64k default that
373 		 * drivers probably should override.  Just like the I/O size we
374 		 * require drivers to at least handle a full page per segment.
375 		 */
376 		if (!lim->max_segment_size)
377 			lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
378 		if (WARN_ON_ONCE(lim->max_segment_size < BLK_MIN_SEGMENT_SIZE))
379 			return -EINVAL;
380 	}
381 
382 	/* setup min segment size for building new segment in fast path */
383 	if (lim->seg_boundary_mask > lim->max_segment_size - 1)
384 		seg_size = lim->max_segment_size;
385 	else
386 		seg_size = lim->seg_boundary_mask + 1;
387 	lim->min_segment_size = min_t(unsigned int, seg_size, PAGE_SIZE);
388 
389 	/*
390 	 * We require drivers to at least do logical block aligned I/O, but
391 	 * historically could not check for that due to the separate calls
392 	 * to set the limits.  Once the transition is finished the check
393 	 * below should be narrowed down to check the logical block size.
394 	 */
395 	if (!lim->dma_alignment)
396 		lim->dma_alignment = SECTOR_SIZE - 1;
397 	if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE))
398 		return -EINVAL;
399 
400 	if (lim->alignment_offset) {
401 		lim->alignment_offset &= (lim->physical_block_size - 1);
402 		lim->flags &= ~BLK_FLAG_MISALIGNED;
403 	}
404 
405 	if (!(lim->features & BLK_FEAT_WRITE_CACHE))
406 		lim->features &= ~BLK_FEAT_FUA;
407 
408 	blk_validate_atomic_write_limits(lim);
409 
410 	err = blk_validate_integrity_limits(lim);
411 	if (err)
412 		return err;
413 	return blk_validate_zoned_limits(lim);
414 }
415 EXPORT_SYMBOL_GPL(blk_validate_limits);
416 
417 /*
418  * Set the default limits for a newly allocated queue.  @lim contains the
419  * initial limits set by the driver, which could be no limit in which case
420  * all fields are cleared to zero.
421  */
blk_set_default_limits(struct queue_limits * lim)422 int blk_set_default_limits(struct queue_limits *lim)
423 {
424 	/*
425 	 * Most defaults are set by capping the bounds in blk_validate_limits,
426 	 * but max_user_discard_sectors is special and needs an explicit
427 	 * initialization to the max value here.
428 	 */
429 	lim->max_user_discard_sectors = UINT_MAX;
430 	return blk_validate_limits(lim);
431 }
432 
433 /**
434  * queue_limits_commit_update - commit an atomic update of queue limits
435  * @q:		queue to update
436  * @lim:	limits to apply
437  *
438  * Apply the limits in @lim that were obtained from queue_limits_start_update()
439  * and updated by the caller to @q.  The caller must have frozen the queue or
440  * ensure that there are no outstanding I/Os by other means.
441  *
442  * Returns 0 if successful, else a negative error code.
443  */
queue_limits_commit_update(struct request_queue * q,struct queue_limits * lim)444 int queue_limits_commit_update(struct request_queue *q,
445 		struct queue_limits *lim)
446 {
447 	int error;
448 
449 	error = blk_validate_limits(lim);
450 	if (error)
451 		goto out_unlock;
452 
453 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
454 	if (q->crypto_profile && lim->integrity.tag_size) {
455 		pr_warn("blk-integrity: Integrity and hardware inline encryption are not supported together.\n");
456 		error = -EINVAL;
457 		goto out_unlock;
458 	}
459 #endif
460 
461 	q->limits = *lim;
462 	if (q->disk)
463 		blk_apply_bdi_limits(q->disk->bdi, lim);
464 out_unlock:
465 	mutex_unlock(&q->limits_lock);
466 	return error;
467 }
468 EXPORT_SYMBOL_GPL(queue_limits_commit_update);
469 
470 /**
471  * queue_limits_commit_update_frozen - commit an atomic update of queue limits
472  * @q:		queue to update
473  * @lim:	limits to apply
474  *
475  * Apply the limits in @lim that were obtained from queue_limits_start_update()
476  * and updated with the new values by the caller to @q.  Freezes the queue
477  * before the update and unfreezes it after.
478  *
479  * Returns 0 if successful, else a negative error code.
480  */
queue_limits_commit_update_frozen(struct request_queue * q,struct queue_limits * lim)481 int queue_limits_commit_update_frozen(struct request_queue *q,
482 		struct queue_limits *lim)
483 {
484 	unsigned int memflags;
485 	int ret;
486 
487 	memflags = blk_mq_freeze_queue(q);
488 	ret = queue_limits_commit_update(q, lim);
489 	blk_mq_unfreeze_queue(q, memflags);
490 
491 	return ret;
492 }
493 EXPORT_SYMBOL_GPL(queue_limits_commit_update_frozen);
494 
495 /**
496  * queue_limits_set - apply queue limits to queue
497  * @q:		queue to update
498  * @lim:	limits to apply
499  *
500  * Apply the limits in @lim that were freshly initialized to @q.
501  * To update existing limits use queue_limits_start_update() and
502  * queue_limits_commit_update() instead.
503  *
504  * Returns 0 if successful, else a negative error code.
505  */
queue_limits_set(struct request_queue * q,struct queue_limits * lim)506 int queue_limits_set(struct request_queue *q, struct queue_limits *lim)
507 {
508 	mutex_lock(&q->limits_lock);
509 	return queue_limits_commit_update(q, lim);
510 }
511 EXPORT_SYMBOL_GPL(queue_limits_set);
512 
queue_limit_alignment_offset(const struct queue_limits * lim,sector_t sector)513 static int queue_limit_alignment_offset(const struct queue_limits *lim,
514 		sector_t sector)
515 {
516 	unsigned int granularity = max(lim->physical_block_size, lim->io_min);
517 	unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT)
518 		<< SECTOR_SHIFT;
519 
520 	return (granularity + lim->alignment_offset - alignment) % granularity;
521 }
522 
queue_limit_discard_alignment(const struct queue_limits * lim,sector_t sector)523 static unsigned int queue_limit_discard_alignment(
524 		const struct queue_limits *lim, sector_t sector)
525 {
526 	unsigned int alignment, granularity, offset;
527 
528 	if (!lim->max_discard_sectors)
529 		return 0;
530 
531 	/* Why are these in bytes, not sectors? */
532 	alignment = lim->discard_alignment >> SECTOR_SHIFT;
533 	granularity = lim->discard_granularity >> SECTOR_SHIFT;
534 
535 	/* Offset of the partition start in 'granularity' sectors */
536 	offset = sector_div(sector, granularity);
537 
538 	/* And why do we do this modulus *again* in blkdev_issue_discard()? */
539 	offset = (granularity + alignment - offset) % granularity;
540 
541 	/* Turn it back into bytes, gaah */
542 	return offset << SECTOR_SHIFT;
543 }
544 
blk_round_down_sectors(unsigned int sectors,unsigned int lbs)545 static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs)
546 {
547 	sectors = round_down(sectors, lbs >> SECTOR_SHIFT);
548 	if (sectors < PAGE_SIZE >> SECTOR_SHIFT)
549 		sectors = PAGE_SIZE >> SECTOR_SHIFT;
550 	return sectors;
551 }
552 
553 /* Check if second and later bottom devices are compliant */
blk_stack_atomic_writes_tail(struct queue_limits * t,struct queue_limits * b)554 static bool blk_stack_atomic_writes_tail(struct queue_limits *t,
555 				struct queue_limits *b)
556 {
557 	/* We're not going to support different boundary sizes.. yet */
558 	if (t->atomic_write_hw_boundary != b->atomic_write_hw_boundary)
559 		return false;
560 
561 	/* Can't support this */
562 	if (t->atomic_write_hw_unit_min > b->atomic_write_hw_unit_max)
563 		return false;
564 
565 	/* Or this */
566 	if (t->atomic_write_hw_unit_max < b->atomic_write_hw_unit_min)
567 		return false;
568 
569 	t->atomic_write_hw_max = min(t->atomic_write_hw_max,
570 				b->atomic_write_hw_max);
571 	t->atomic_write_hw_unit_min = max(t->atomic_write_hw_unit_min,
572 				b->atomic_write_hw_unit_min);
573 	t->atomic_write_hw_unit_max = min(t->atomic_write_hw_unit_max,
574 				b->atomic_write_hw_unit_max);
575 	return true;
576 }
577 
578 /* Check for valid boundary of first bottom device */
blk_stack_atomic_writes_boundary_head(struct queue_limits * t,struct queue_limits * b)579 static bool blk_stack_atomic_writes_boundary_head(struct queue_limits *t,
580 				struct queue_limits *b)
581 {
582 	/*
583 	 * Ensure atomic write boundary is aligned with chunk sectors. Stacked
584 	 * devices store chunk sectors in t->io_min.
585 	 */
586 	if (b->atomic_write_hw_boundary > t->io_min &&
587 	    b->atomic_write_hw_boundary % t->io_min)
588 		return false;
589 	if (t->io_min > b->atomic_write_hw_boundary &&
590 	    t->io_min % b->atomic_write_hw_boundary)
591 		return false;
592 
593 	t->atomic_write_hw_boundary = b->atomic_write_hw_boundary;
594 	return true;
595 }
596 
597 
598 /* Check stacking of first bottom device */
blk_stack_atomic_writes_head(struct queue_limits * t,struct queue_limits * b)599 static bool blk_stack_atomic_writes_head(struct queue_limits *t,
600 				struct queue_limits *b)
601 {
602 	if (b->atomic_write_hw_boundary &&
603 	    !blk_stack_atomic_writes_boundary_head(t, b))
604 		return false;
605 
606 	if (t->io_min <= SECTOR_SIZE) {
607 		/* No chunk sectors, so use bottom device values directly */
608 		t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
609 		t->atomic_write_hw_unit_min = b->atomic_write_hw_unit_min;
610 		t->atomic_write_hw_max = b->atomic_write_hw_max;
611 		return true;
612 	}
613 
614 	/*
615 	 * Find values for limits which work for chunk size.
616 	 * b->atomic_write_hw_unit_{min, max} may not be aligned with chunk
617 	 * size (t->io_min), as chunk size is not restricted to a power-of-2.
618 	 * So we need to find highest power-of-2 which works for the chunk
619 	 * size.
620 	 * As an example scenario, we could have b->unit_max = 16K and
621 	 * t->io_min = 24K. For this case, reduce t->unit_max to a value
622 	 * aligned with both limits, i.e. 8K in this example.
623 	 */
624 	t->atomic_write_hw_unit_max = b->atomic_write_hw_unit_max;
625 	while (t->io_min % t->atomic_write_hw_unit_max)
626 		t->atomic_write_hw_unit_max /= 2;
627 
628 	t->atomic_write_hw_unit_min = min(b->atomic_write_hw_unit_min,
629 					  t->atomic_write_hw_unit_max);
630 	t->atomic_write_hw_max = min(b->atomic_write_hw_max, t->io_min);
631 
632 	return true;
633 }
634 
blk_stack_atomic_writes_limits(struct queue_limits * t,struct queue_limits * b,sector_t start)635 static void blk_stack_atomic_writes_limits(struct queue_limits *t,
636 				struct queue_limits *b, sector_t start)
637 {
638 	if (!(b->features & BLK_FEAT_ATOMIC_WRITES))
639 		goto unsupported;
640 
641 	if (!b->atomic_write_hw_unit_min)
642 		goto unsupported;
643 
644 	if (!blk_atomic_write_start_sect_aligned(start, b))
645 		goto unsupported;
646 
647 	/*
648 	 * If atomic_write_hw_max is set, we have already stacked 1x bottom
649 	 * device, so check for compliance.
650 	 */
651 	if (t->atomic_write_hw_max) {
652 		if (!blk_stack_atomic_writes_tail(t, b))
653 			goto unsupported;
654 		return;
655 	}
656 
657 	if (!blk_stack_atomic_writes_head(t, b))
658 		goto unsupported;
659 	return;
660 
661 unsupported:
662 	t->atomic_write_hw_max = 0;
663 	t->atomic_write_hw_unit_max = 0;
664 	t->atomic_write_hw_unit_min = 0;
665 	t->atomic_write_hw_boundary = 0;
666 }
667 
668 /**
669  * blk_stack_limits - adjust queue_limits for stacked devices
670  * @t:	the stacking driver limits (top device)
671  * @b:  the underlying queue limits (bottom, component device)
672  * @start:  first data sector within component device
673  *
674  * Description:
675  *    This function is used by stacking drivers like MD and DM to ensure
676  *    that all component devices have compatible block sizes and
677  *    alignments.  The stacking driver must provide a queue_limits
678  *    struct (top) and then iteratively call the stacking function for
679  *    all component (bottom) devices.  The stacking function will
680  *    attempt to combine the values and ensure proper alignment.
681  *
682  *    Returns 0 if the top and bottom queue_limits are compatible.  The
683  *    top device's block sizes and alignment offsets may be adjusted to
684  *    ensure alignment with the bottom device. If no compatible sizes
685  *    and alignments exist, -1 is returned and the resulting top
686  *    queue_limits will have the misaligned flag set to indicate that
687  *    the alignment_offset is undefined.
688  */
blk_stack_limits(struct queue_limits * t,struct queue_limits * b,sector_t start)689 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
690 		     sector_t start)
691 {
692 	unsigned int top, bottom, alignment, ret = 0;
693 
694 	t->features |= (b->features & BLK_FEAT_INHERIT_MASK);
695 
696 	/*
697 	 * Some feaures need to be supported both by the stacking driver and all
698 	 * underlying devices.  The stacking driver sets these flags before
699 	 * stacking the limits, and this will clear the flags if any of the
700 	 * underlying devices does not support it.
701 	 */
702 	if (!(b->features & BLK_FEAT_NOWAIT))
703 		t->features &= ~BLK_FEAT_NOWAIT;
704 	if (!(b->features & BLK_FEAT_POLL))
705 		t->features &= ~BLK_FEAT_POLL;
706 
707 	t->flags |= (b->flags & BLK_FLAG_MISALIGNED);
708 
709 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
710 	t->max_user_sectors = min_not_zero(t->max_user_sectors,
711 			b->max_user_sectors);
712 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
713 	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
714 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
715 					b->max_write_zeroes_sectors);
716 	t->max_hw_zone_append_sectors = min(t->max_hw_zone_append_sectors,
717 					b->max_hw_zone_append_sectors);
718 
719 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
720 					    b->seg_boundary_mask);
721 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
722 					    b->virt_boundary_mask);
723 
724 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
725 	t->max_discard_segments = min_not_zero(t->max_discard_segments,
726 					       b->max_discard_segments);
727 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
728 						 b->max_integrity_segments);
729 
730 	t->max_segment_size = min_not_zero(t->max_segment_size,
731 					   b->max_segment_size);
732 
733 	alignment = queue_limit_alignment_offset(b, start);
734 
735 	/* Bottom device has different alignment.  Check that it is
736 	 * compatible with the current top alignment.
737 	 */
738 	if (t->alignment_offset != alignment) {
739 
740 		top = max(t->physical_block_size, t->io_min)
741 			+ t->alignment_offset;
742 		bottom = max(b->physical_block_size, b->io_min) + alignment;
743 
744 		/* Verify that top and bottom intervals line up */
745 		if (max(top, bottom) % min(top, bottom)) {
746 			t->flags |= BLK_FLAG_MISALIGNED;
747 			ret = -1;
748 		}
749 	}
750 
751 	t->logical_block_size = max(t->logical_block_size,
752 				    b->logical_block_size);
753 
754 	t->physical_block_size = max(t->physical_block_size,
755 				     b->physical_block_size);
756 
757 	t->io_min = max(t->io_min, b->io_min);
758 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
759 	t->dma_alignment = max(t->dma_alignment, b->dma_alignment);
760 
761 	/* Set non-power-of-2 compatible chunk_sectors boundary */
762 	if (b->chunk_sectors)
763 		t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors);
764 
765 	/* Physical block size a multiple of the logical block size? */
766 	if (t->physical_block_size & (t->logical_block_size - 1)) {
767 		t->physical_block_size = t->logical_block_size;
768 		t->flags |= BLK_FLAG_MISALIGNED;
769 		ret = -1;
770 	}
771 
772 	/* Minimum I/O a multiple of the physical block size? */
773 	if (t->io_min & (t->physical_block_size - 1)) {
774 		t->io_min = t->physical_block_size;
775 		t->flags |= BLK_FLAG_MISALIGNED;
776 		ret = -1;
777 	}
778 
779 	/* Optimal I/O a multiple of the physical block size? */
780 	if (t->io_opt & (t->physical_block_size - 1)) {
781 		t->io_opt = 0;
782 		t->flags |= BLK_FLAG_MISALIGNED;
783 		ret = -1;
784 	}
785 
786 	/* chunk_sectors a multiple of the physical block size? */
787 	if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) {
788 		t->chunk_sectors = 0;
789 		t->flags |= BLK_FLAG_MISALIGNED;
790 		ret = -1;
791 	}
792 
793 	/* Find lowest common alignment_offset */
794 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
795 		% max(t->physical_block_size, t->io_min);
796 
797 	/* Verify that new alignment_offset is on a logical block boundary */
798 	if (t->alignment_offset & (t->logical_block_size - 1)) {
799 		t->flags |= BLK_FLAG_MISALIGNED;
800 		ret = -1;
801 	}
802 
803 	t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size);
804 	t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size);
805 	t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size);
806 
807 	/* Discard alignment and granularity */
808 	if (b->discard_granularity) {
809 		alignment = queue_limit_discard_alignment(b, start);
810 
811 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
812 						      b->max_discard_sectors);
813 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
814 							 b->max_hw_discard_sectors);
815 		t->discard_granularity = max(t->discard_granularity,
816 					     b->discard_granularity);
817 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
818 			t->discard_granularity;
819 	}
820 	t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors,
821 						   b->max_secure_erase_sectors);
822 	t->zone_write_granularity = max(t->zone_write_granularity,
823 					b->zone_write_granularity);
824 	if (!(t->features & BLK_FEAT_ZONED)) {
825 		t->zone_write_granularity = 0;
826 		t->max_zone_append_sectors = 0;
827 	}
828 	blk_stack_atomic_writes_limits(t, b, start);
829 
830 	return ret;
831 }
832 EXPORT_SYMBOL(blk_stack_limits);
833 
834 /**
835  * queue_limits_stack_bdev - adjust queue_limits for stacked devices
836  * @t:	the stacking driver limits (top device)
837  * @bdev:  the underlying block device (bottom)
838  * @offset:  offset to beginning of data within component device
839  * @pfx: prefix to use for warnings logged
840  *
841  * Description:
842  *    This function is used by stacking drivers like MD and DM to ensure
843  *    that all component devices have compatible block sizes and
844  *    alignments.  The stacking driver must provide a queue_limits
845  *    struct (top) and then iteratively call the stacking function for
846  *    all component (bottom) devices.  The stacking function will
847  *    attempt to combine the values and ensure proper alignment.
848  */
queue_limits_stack_bdev(struct queue_limits * t,struct block_device * bdev,sector_t offset,const char * pfx)849 void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev,
850 		sector_t offset, const char *pfx)
851 {
852 	if (blk_stack_limits(t, bdev_limits(bdev),
853 			get_start_sect(bdev) + offset))
854 		pr_notice("%s: Warning: Device %pg is misaligned\n",
855 			pfx, bdev);
856 }
857 EXPORT_SYMBOL_GPL(queue_limits_stack_bdev);
858 
859 /**
860  * queue_limits_stack_integrity - stack integrity profile
861  * @t: target queue limits
862  * @b: base queue limits
863  *
864  * Check if the integrity profile in the @b can be stacked into the
865  * target @t.  Stacking is possible if either:
866  *
867  *   a) does not have any integrity information stacked into it yet
868  *   b) the integrity profile in @b is identical to the one in @t
869  *
870  * If @b can be stacked into @t, return %true.  Else return %false and clear the
871  * integrity information in @t.
872  */
queue_limits_stack_integrity(struct queue_limits * t,struct queue_limits * b)873 bool queue_limits_stack_integrity(struct queue_limits *t,
874 		struct queue_limits *b)
875 {
876 	struct blk_integrity *ti = &t->integrity;
877 	struct blk_integrity *bi = &b->integrity;
878 
879 	if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
880 		return true;
881 
882 	if (ti->flags & BLK_INTEGRITY_STACKED) {
883 		if (ti->tuple_size != bi->tuple_size)
884 			goto incompatible;
885 		if (ti->interval_exp != bi->interval_exp)
886 			goto incompatible;
887 		if (ti->tag_size != bi->tag_size)
888 			goto incompatible;
889 		if (ti->csum_type != bi->csum_type)
890 			goto incompatible;
891 		if ((ti->flags & BLK_INTEGRITY_REF_TAG) !=
892 		    (bi->flags & BLK_INTEGRITY_REF_TAG))
893 			goto incompatible;
894 	} else {
895 		ti->flags = BLK_INTEGRITY_STACKED;
896 		ti->flags |= (bi->flags & BLK_INTEGRITY_DEVICE_CAPABLE) |
897 			     (bi->flags & BLK_INTEGRITY_REF_TAG);
898 		ti->csum_type = bi->csum_type;
899 		ti->tuple_size = bi->tuple_size;
900 		ti->pi_offset = bi->pi_offset;
901 		ti->interval_exp = bi->interval_exp;
902 		ti->tag_size = bi->tag_size;
903 	}
904 	return true;
905 
906 incompatible:
907 	memset(ti, 0, sizeof(*ti));
908 	return false;
909 }
910 EXPORT_SYMBOL_GPL(queue_limits_stack_integrity);
911 
912 /**
913  * blk_set_queue_depth - tell the block layer about the device queue depth
914  * @q:		the request queue for the device
915  * @depth:		queue depth
916  *
917  */
blk_set_queue_depth(struct request_queue * q,unsigned int depth)918 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
919 {
920 	q->queue_depth = depth;
921 	rq_qos_queue_depth_changed(q);
922 }
923 EXPORT_SYMBOL(blk_set_queue_depth);
924 
bdev_alignment_offset(struct block_device * bdev)925 int bdev_alignment_offset(struct block_device *bdev)
926 {
927 	struct request_queue *q = bdev_get_queue(bdev);
928 
929 	if (q->limits.flags & BLK_FLAG_MISALIGNED)
930 		return -1;
931 	if (bdev_is_partition(bdev))
932 		return queue_limit_alignment_offset(&q->limits,
933 				bdev->bd_start_sect);
934 	return q->limits.alignment_offset;
935 }
936 EXPORT_SYMBOL_GPL(bdev_alignment_offset);
937 
bdev_discard_alignment(struct block_device * bdev)938 unsigned int bdev_discard_alignment(struct block_device *bdev)
939 {
940 	struct request_queue *q = bdev_get_queue(bdev);
941 
942 	if (bdev_is_partition(bdev))
943 		return queue_limit_discard_alignment(&q->limits,
944 				bdev->bd_start_sect);
945 	return q->limits.discard_alignment;
946 }
947 EXPORT_SYMBOL_GPL(bdev_discard_alignment);
948