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