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