1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Functions related to setting various queue properties from drivers |
4 | */ |
5 | #include <linux/kernel.h> |
6 | #include <linux/module.h> |
7 | #include <linux/init.h> |
8 | #include <linux/bio.h> |
9 | #include <linux/blkdev.h> |
10 | #include <linux/pagemap.h> |
11 | #include <linux/backing-dev-defs.h> |
12 | #include <linux/gcd.h> |
13 | #include <linux/lcm.h> |
14 | #include <linux/jiffies.h> |
15 | #include <linux/gfp.h> |
16 | #include <linux/dma-mapping.h> |
17 | |
18 | #include "blk.h" |
19 | #include "blk-rq-qos.h" |
20 | #include "blk-wbt.h" |
21 | |
22 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
23 | { |
24 | q->rq_timeout = timeout; |
25 | } |
26 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); |
27 | |
28 | /** |
29 | * blk_set_stacking_limits - set default limits for stacking devices |
30 | * @lim: the queue_limits structure to reset |
31 | * |
32 | * Prepare queue limits for applying limits from underlying devices using |
33 | * blk_stack_limits(). |
34 | */ |
35 | void blk_set_stacking_limits(struct queue_limits *lim) |
36 | { |
37 | memset(lim, 0, sizeof(*lim)); |
38 | lim->logical_block_size = SECTOR_SIZE; |
39 | lim->physical_block_size = SECTOR_SIZE; |
40 | lim->io_min = SECTOR_SIZE; |
41 | lim->discard_granularity = SECTOR_SIZE; |
42 | lim->dma_alignment = SECTOR_SIZE - 1; |
43 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
44 | |
45 | /* Inherit limits from component devices */ |
46 | lim->max_segments = USHRT_MAX; |
47 | lim->max_discard_segments = USHRT_MAX; |
48 | lim->max_hw_sectors = UINT_MAX; |
49 | lim->max_segment_size = UINT_MAX; |
50 | lim->max_sectors = UINT_MAX; |
51 | lim->max_dev_sectors = UINT_MAX; |
52 | lim->max_write_zeroes_sectors = UINT_MAX; |
53 | lim->max_zone_append_sectors = UINT_MAX; |
54 | lim->max_user_discard_sectors = UINT_MAX; |
55 | } |
56 | EXPORT_SYMBOL(blk_set_stacking_limits); |
57 | |
58 | static void blk_apply_bdi_limits(struct backing_dev_info *bdi, |
59 | struct queue_limits *lim) |
60 | { |
61 | /* |
62 | * For read-ahead of large files to be effective, we need to read ahead |
63 | * at least twice the optimal I/O size. |
64 | */ |
65 | bdi->ra_pages = max(lim->io_opt * 2 / PAGE_SIZE, VM_READAHEAD_PAGES); |
66 | bdi->io_pages = lim->max_sectors >> PAGE_SECTORS_SHIFT; |
67 | } |
68 | |
69 | static int blk_validate_zoned_limits(struct queue_limits *lim) |
70 | { |
71 | if (!lim->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 | if (lim->zone_write_granularity < lim->logical_block_size) |
84 | lim->zone_write_granularity = lim->logical_block_size; |
85 | |
86 | if (lim->max_zone_append_sectors) { |
87 | /* |
88 | * The Zone Append size is limited by the maximum I/O size |
89 | * and the zone size given that it can't span zones. |
90 | */ |
91 | lim->max_zone_append_sectors = |
92 | min3(lim->max_hw_sectors, |
93 | lim->max_zone_append_sectors, |
94 | lim->chunk_sectors); |
95 | } |
96 | |
97 | return 0; |
98 | } |
99 | |
100 | /* |
101 | * Check that the limits in lim are valid, initialize defaults for unset |
102 | * values, and cap values based on others where needed. |
103 | */ |
104 | static int blk_validate_limits(struct queue_limits *lim) |
105 | { |
106 | unsigned int max_hw_sectors; |
107 | |
108 | /* |
109 | * Unless otherwise specified, default to 512 byte logical blocks and a |
110 | * physical block size equal to the logical block size. |
111 | */ |
112 | if (!lim->logical_block_size) |
113 | lim->logical_block_size = SECTOR_SIZE; |
114 | if (lim->physical_block_size < lim->logical_block_size) |
115 | lim->physical_block_size = lim->logical_block_size; |
116 | |
117 | /* |
118 | * The minimum I/O size defaults to the physical block size unless |
119 | * explicitly overridden. |
120 | */ |
121 | if (lim->io_min < lim->physical_block_size) |
122 | lim->io_min = lim->physical_block_size; |
123 | |
124 | /* |
125 | * max_hw_sectors has a somewhat weird default for historical reason, |
126 | * but driver really should set their own instead of relying on this |
127 | * value. |
128 | * |
129 | * The block layer relies on the fact that every driver can |
130 | * handle at lest a page worth of data per I/O, and needs the value |
131 | * aligned to the logical block size. |
132 | */ |
133 | if (!lim->max_hw_sectors) |
134 | lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; |
135 | if (WARN_ON_ONCE(lim->max_hw_sectors < PAGE_SECTORS)) |
136 | return -EINVAL; |
137 | lim->max_hw_sectors = round_down(lim->max_hw_sectors, |
138 | lim->logical_block_size >> SECTOR_SHIFT); |
139 | |
140 | /* |
141 | * The actual max_sectors value is a complex beast and also takes the |
142 | * max_dev_sectors value (set by SCSI ULPs) and a user configurable |
143 | * value into account. The ->max_sectors value is always calculated |
144 | * from these, so directly setting it won't have any effect. |
145 | */ |
146 | max_hw_sectors = min_not_zero(lim->max_hw_sectors, |
147 | lim->max_dev_sectors); |
148 | if (lim->max_user_sectors) { |
149 | if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE) |
150 | return -EINVAL; |
151 | lim->max_sectors = min(max_hw_sectors, lim->max_user_sectors); |
152 | } else { |
153 | lim->max_sectors = min(max_hw_sectors, BLK_DEF_MAX_SECTORS_CAP); |
154 | } |
155 | lim->max_sectors = round_down(lim->max_sectors, |
156 | lim->logical_block_size >> SECTOR_SHIFT); |
157 | |
158 | /* |
159 | * Random default for the maximum number of segments. Driver should not |
160 | * rely on this and set their own. |
161 | */ |
162 | if (!lim->max_segments) |
163 | lim->max_segments = BLK_MAX_SEGMENTS; |
164 | |
165 | lim->max_discard_sectors = |
166 | min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors); |
167 | |
168 | if (!lim->max_discard_segments) |
169 | lim->max_discard_segments = 1; |
170 | |
171 | if (lim->discard_granularity < lim->physical_block_size) |
172 | lim->discard_granularity = lim->physical_block_size; |
173 | |
174 | /* |
175 | * By default there is no limit on the segment boundary alignment, |
176 | * but if there is one it can't be smaller than the page size as |
177 | * that would break all the normal I/O patterns. |
178 | */ |
179 | if (!lim->seg_boundary_mask) |
180 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
181 | if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1)) |
182 | return -EINVAL; |
183 | |
184 | /* |
185 | * Stacking device may have both virtual boundary and max segment |
186 | * size limit, so allow this setting now, and long-term the two |
187 | * might need to move out of stacking limits since we have immutable |
188 | * bvec and lower layer bio splitting is supposed to handle the two |
189 | * correctly. |
190 | */ |
191 | if (!lim->virt_boundary_mask) { |
192 | /* |
193 | * The maximum segment size has an odd historic 64k default that |
194 | * drivers probably should override. Just like the I/O size we |
195 | * require drivers to at least handle a full page per segment. |
196 | */ |
197 | if (!lim->max_segment_size) |
198 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; |
199 | if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE)) |
200 | return -EINVAL; |
201 | } |
202 | |
203 | /* |
204 | * We require drivers to at least do logical block aligned I/O, but |
205 | * historically could not check for that due to the separate calls |
206 | * to set the limits. Once the transition is finished the check |
207 | * below should be narrowed down to check the logical block size. |
208 | */ |
209 | if (!lim->dma_alignment) |
210 | lim->dma_alignment = SECTOR_SIZE - 1; |
211 | if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE)) |
212 | return -EINVAL; |
213 | |
214 | if (lim->alignment_offset) { |
215 | lim->alignment_offset &= (lim->physical_block_size - 1); |
216 | lim->misaligned = 0; |
217 | } |
218 | |
219 | return blk_validate_zoned_limits(lim); |
220 | } |
221 | |
222 | /* |
223 | * Set the default limits for a newly allocated queue. @lim contains the |
224 | * initial limits set by the driver, which could be no limit in which case |
225 | * all fields are cleared to zero. |
226 | */ |
227 | int blk_set_default_limits(struct queue_limits *lim) |
228 | { |
229 | /* |
230 | * Most defaults are set by capping the bounds in blk_validate_limits, |
231 | * but max_user_discard_sectors is special and needs an explicit |
232 | * initialization to the max value here. |
233 | */ |
234 | lim->max_user_discard_sectors = UINT_MAX; |
235 | return blk_validate_limits(lim); |
236 | } |
237 | |
238 | /** |
239 | * queue_limits_commit_update - commit an atomic update of queue limits |
240 | * @q: queue to update |
241 | * @lim: limits to apply |
242 | * |
243 | * Apply the limits in @lim that were obtained from queue_limits_start_update() |
244 | * and updated by the caller to @q. |
245 | * |
246 | * Returns 0 if successful, else a negative error code. |
247 | */ |
248 | int queue_limits_commit_update(struct request_queue *q, |
249 | struct queue_limits *lim) |
250 | __releases(q->limits_lock) |
251 | { |
252 | int error = blk_validate_limits(lim); |
253 | |
254 | if (!error) { |
255 | q->limits = *lim; |
256 | if (q->disk) |
257 | blk_apply_bdi_limits(bdi: q->disk->bdi, lim); |
258 | } |
259 | mutex_unlock(lock: &q->limits_lock); |
260 | return error; |
261 | } |
262 | EXPORT_SYMBOL_GPL(queue_limits_commit_update); |
263 | |
264 | /** |
265 | * queue_limits_set - apply queue limits to queue |
266 | * @q: queue to update |
267 | * @lim: limits to apply |
268 | * |
269 | * Apply the limits in @lim that were freshly initialized to @q. |
270 | * To update existing limits use queue_limits_start_update() and |
271 | * queue_limits_commit_update() instead. |
272 | * |
273 | * Returns 0 if successful, else a negative error code. |
274 | */ |
275 | int queue_limits_set(struct request_queue *q, struct queue_limits *lim) |
276 | { |
277 | mutex_lock(&q->limits_lock); |
278 | return queue_limits_commit_update(q, lim); |
279 | } |
280 | EXPORT_SYMBOL_GPL(queue_limits_set); |
281 | |
282 | /** |
283 | * blk_queue_bounce_limit - set bounce buffer limit for queue |
284 | * @q: the request queue for the device |
285 | * @bounce: bounce limit to enforce |
286 | * |
287 | * Description: |
288 | * Force bouncing for ISA DMA ranges or highmem. |
289 | * |
290 | * DEPRECATED, don't use in new code. |
291 | **/ |
292 | void blk_queue_bounce_limit(struct request_queue *q, enum blk_bounce bounce) |
293 | { |
294 | q->limits.bounce = bounce; |
295 | } |
296 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
297 | |
298 | /** |
299 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue |
300 | * @q: the request queue for the device |
301 | * @max_hw_sectors: max hardware sectors in the usual 512b unit |
302 | * |
303 | * Description: |
304 | * Enables a low level driver to set a hard upper limit, |
305 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by |
306 | * the device driver based upon the capabilities of the I/O |
307 | * controller. |
308 | * |
309 | * max_dev_sectors is a hard limit imposed by the storage device for |
310 | * READ/WRITE requests. It is set by the disk driver. |
311 | * |
312 | * max_sectors is a soft limit imposed by the block layer for |
313 | * filesystem type requests. This value can be overridden on a |
314 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. |
315 | * The soft limit can not exceed max_hw_sectors. |
316 | **/ |
317 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) |
318 | { |
319 | struct queue_limits *limits = &q->limits; |
320 | unsigned int max_sectors; |
321 | |
322 | if ((max_hw_sectors << 9) < PAGE_SIZE) { |
323 | max_hw_sectors = 1 << (PAGE_SHIFT - 9); |
324 | pr_info("%s: set to minimum %u\n" , __func__, max_hw_sectors); |
325 | } |
326 | |
327 | max_hw_sectors = round_down(max_hw_sectors, |
328 | limits->logical_block_size >> SECTOR_SHIFT); |
329 | limits->max_hw_sectors = max_hw_sectors; |
330 | |
331 | max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors); |
332 | |
333 | if (limits->max_user_sectors) |
334 | max_sectors = min(max_sectors, limits->max_user_sectors); |
335 | else |
336 | max_sectors = min(max_sectors, BLK_DEF_MAX_SECTORS_CAP); |
337 | |
338 | max_sectors = round_down(max_sectors, |
339 | limits->logical_block_size >> SECTOR_SHIFT); |
340 | limits->max_sectors = max_sectors; |
341 | |
342 | if (!q->disk) |
343 | return; |
344 | q->disk->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9); |
345 | } |
346 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
347 | |
348 | /** |
349 | * blk_queue_chunk_sectors - set size of the chunk for this queue |
350 | * @q: the request queue for the device |
351 | * @chunk_sectors: chunk sectors in the usual 512b unit |
352 | * |
353 | * Description: |
354 | * If a driver doesn't want IOs to cross a given chunk size, it can set |
355 | * this limit and prevent merging across chunks. Note that the block layer |
356 | * must accept a page worth of data at any offset. So if the crossing of |
357 | * chunks is a hard limitation in the driver, it must still be prepared |
358 | * to split single page bios. |
359 | **/ |
360 | void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) |
361 | { |
362 | q->limits.chunk_sectors = chunk_sectors; |
363 | } |
364 | EXPORT_SYMBOL(blk_queue_chunk_sectors); |
365 | |
366 | /** |
367 | * blk_queue_max_discard_sectors - set max sectors for a single discard |
368 | * @q: the request queue for the device |
369 | * @max_discard_sectors: maximum number of sectors to discard |
370 | **/ |
371 | void blk_queue_max_discard_sectors(struct request_queue *q, |
372 | unsigned int max_discard_sectors) |
373 | { |
374 | struct queue_limits *lim = &q->limits; |
375 | |
376 | lim->max_hw_discard_sectors = max_discard_sectors; |
377 | lim->max_discard_sectors = |
378 | min(max_discard_sectors, lim->max_user_discard_sectors); |
379 | } |
380 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); |
381 | |
382 | /** |
383 | * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase |
384 | * @q: the request queue for the device |
385 | * @max_sectors: maximum number of sectors to secure_erase |
386 | **/ |
387 | void blk_queue_max_secure_erase_sectors(struct request_queue *q, |
388 | unsigned int max_sectors) |
389 | { |
390 | q->limits.max_secure_erase_sectors = max_sectors; |
391 | } |
392 | EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors); |
393 | |
394 | /** |
395 | * blk_queue_max_write_zeroes_sectors - set max sectors for a single |
396 | * write zeroes |
397 | * @q: the request queue for the device |
398 | * @max_write_zeroes_sectors: maximum number of sectors to write per command |
399 | **/ |
400 | void blk_queue_max_write_zeroes_sectors(struct request_queue *q, |
401 | unsigned int max_write_zeroes_sectors) |
402 | { |
403 | q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors; |
404 | } |
405 | EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors); |
406 | |
407 | /** |
408 | * blk_queue_max_zone_append_sectors - set max sectors for a single zone append |
409 | * @q: the request queue for the device |
410 | * @max_zone_append_sectors: maximum number of sectors to write per command |
411 | **/ |
412 | void blk_queue_max_zone_append_sectors(struct request_queue *q, |
413 | unsigned int max_zone_append_sectors) |
414 | { |
415 | unsigned int max_sectors; |
416 | |
417 | if (WARN_ON(!blk_queue_is_zoned(q))) |
418 | return; |
419 | |
420 | max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors); |
421 | max_sectors = min(q->limits.chunk_sectors, max_sectors); |
422 | |
423 | /* |
424 | * Signal eventual driver bugs resulting in the max_zone_append sectors limit |
425 | * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set, |
426 | * or the max_hw_sectors limit not set. |
427 | */ |
428 | WARN_ON(!max_sectors); |
429 | |
430 | q->limits.max_zone_append_sectors = max_sectors; |
431 | } |
432 | EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors); |
433 | |
434 | /** |
435 | * blk_queue_max_segments - set max hw segments for a request for this queue |
436 | * @q: the request queue for the device |
437 | * @max_segments: max number of segments |
438 | * |
439 | * Description: |
440 | * Enables a low level driver to set an upper limit on the number of |
441 | * hw data segments in a request. |
442 | **/ |
443 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) |
444 | { |
445 | if (!max_segments) { |
446 | max_segments = 1; |
447 | pr_info("%s: set to minimum %u\n" , __func__, max_segments); |
448 | } |
449 | |
450 | q->limits.max_segments = max_segments; |
451 | } |
452 | EXPORT_SYMBOL(blk_queue_max_segments); |
453 | |
454 | /** |
455 | * blk_queue_max_discard_segments - set max segments for discard requests |
456 | * @q: the request queue for the device |
457 | * @max_segments: max number of segments |
458 | * |
459 | * Description: |
460 | * Enables a low level driver to set an upper limit on the number of |
461 | * segments in a discard request. |
462 | **/ |
463 | void blk_queue_max_discard_segments(struct request_queue *q, |
464 | unsigned short max_segments) |
465 | { |
466 | q->limits.max_discard_segments = max_segments; |
467 | } |
468 | EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments); |
469 | |
470 | /** |
471 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg |
472 | * @q: the request queue for the device |
473 | * @max_size: max size of segment in bytes |
474 | * |
475 | * Description: |
476 | * Enables a low level driver to set an upper limit on the size of a |
477 | * coalesced segment |
478 | **/ |
479 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) |
480 | { |
481 | if (max_size < PAGE_SIZE) { |
482 | max_size = PAGE_SIZE; |
483 | pr_info("%s: set to minimum %u\n" , __func__, max_size); |
484 | } |
485 | |
486 | /* see blk_queue_virt_boundary() for the explanation */ |
487 | WARN_ON_ONCE(q->limits.virt_boundary_mask); |
488 | |
489 | q->limits.max_segment_size = max_size; |
490 | } |
491 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
492 | |
493 | /** |
494 | * blk_queue_logical_block_size - set logical block size for the queue |
495 | * @q: the request queue for the device |
496 | * @size: the logical block size, in bytes |
497 | * |
498 | * Description: |
499 | * This should be set to the lowest possible block size that the |
500 | * storage device can address. The default of 512 covers most |
501 | * hardware. |
502 | **/ |
503 | void blk_queue_logical_block_size(struct request_queue *q, unsigned int size) |
504 | { |
505 | struct queue_limits *limits = &q->limits; |
506 | |
507 | limits->logical_block_size = size; |
508 | |
509 | if (limits->discard_granularity < limits->logical_block_size) |
510 | limits->discard_granularity = limits->logical_block_size; |
511 | |
512 | if (limits->physical_block_size < size) |
513 | limits->physical_block_size = size; |
514 | |
515 | if (limits->io_min < limits->physical_block_size) |
516 | limits->io_min = limits->physical_block_size; |
517 | |
518 | limits->max_hw_sectors = |
519 | round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT); |
520 | limits->max_sectors = |
521 | round_down(limits->max_sectors, size >> SECTOR_SHIFT); |
522 | } |
523 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
524 | |
525 | /** |
526 | * blk_queue_physical_block_size - set physical block size for the queue |
527 | * @q: the request queue for the device |
528 | * @size: the physical block size, in bytes |
529 | * |
530 | * Description: |
531 | * This should be set to the lowest possible sector size that the |
532 | * hardware can operate on without reverting to read-modify-write |
533 | * operations. |
534 | */ |
535 | void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) |
536 | { |
537 | q->limits.physical_block_size = size; |
538 | |
539 | if (q->limits.physical_block_size < q->limits.logical_block_size) |
540 | q->limits.physical_block_size = q->limits.logical_block_size; |
541 | |
542 | if (q->limits.discard_granularity < q->limits.physical_block_size) |
543 | q->limits.discard_granularity = q->limits.physical_block_size; |
544 | |
545 | if (q->limits.io_min < q->limits.physical_block_size) |
546 | q->limits.io_min = q->limits.physical_block_size; |
547 | } |
548 | EXPORT_SYMBOL(blk_queue_physical_block_size); |
549 | |
550 | /** |
551 | * blk_queue_zone_write_granularity - set zone write granularity for the queue |
552 | * @q: the request queue for the zoned device |
553 | * @size: the zone write granularity size, in bytes |
554 | * |
555 | * Description: |
556 | * This should be set to the lowest possible size allowing to write in |
557 | * sequential zones of a zoned block device. |
558 | */ |
559 | void blk_queue_zone_write_granularity(struct request_queue *q, |
560 | unsigned int size) |
561 | { |
562 | if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) |
563 | return; |
564 | |
565 | q->limits.zone_write_granularity = size; |
566 | |
567 | if (q->limits.zone_write_granularity < q->limits.logical_block_size) |
568 | q->limits.zone_write_granularity = q->limits.logical_block_size; |
569 | } |
570 | EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity); |
571 | |
572 | /** |
573 | * blk_queue_alignment_offset - set physical block alignment offset |
574 | * @q: the request queue for the device |
575 | * @offset: alignment offset in bytes |
576 | * |
577 | * Description: |
578 | * Some devices are naturally misaligned to compensate for things like |
579 | * the legacy DOS partition table 63-sector offset. Low-level drivers |
580 | * should call this function for devices whose first sector is not |
581 | * naturally aligned. |
582 | */ |
583 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) |
584 | { |
585 | q->limits.alignment_offset = |
586 | offset & (q->limits.physical_block_size - 1); |
587 | q->limits.misaligned = 0; |
588 | } |
589 | EXPORT_SYMBOL(blk_queue_alignment_offset); |
590 | |
591 | void disk_update_readahead(struct gendisk *disk) |
592 | { |
593 | blk_apply_bdi_limits(bdi: disk->bdi, lim: &disk->queue->limits); |
594 | } |
595 | EXPORT_SYMBOL_GPL(disk_update_readahead); |
596 | |
597 | /** |
598 | * blk_limits_io_min - set minimum request size for a device |
599 | * @limits: the queue limits |
600 | * @min: smallest I/O size in bytes |
601 | * |
602 | * Description: |
603 | * Some devices have an internal block size bigger than the reported |
604 | * hardware sector size. This function can be used to signal the |
605 | * smallest I/O the device can perform without incurring a performance |
606 | * penalty. |
607 | */ |
608 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) |
609 | { |
610 | limits->io_min = min; |
611 | |
612 | if (limits->io_min < limits->logical_block_size) |
613 | limits->io_min = limits->logical_block_size; |
614 | |
615 | if (limits->io_min < limits->physical_block_size) |
616 | limits->io_min = limits->physical_block_size; |
617 | } |
618 | EXPORT_SYMBOL(blk_limits_io_min); |
619 | |
620 | /** |
621 | * blk_queue_io_min - set minimum request size for the queue |
622 | * @q: the request queue for the device |
623 | * @min: smallest I/O size in bytes |
624 | * |
625 | * Description: |
626 | * Storage devices may report a granularity or preferred minimum I/O |
627 | * size which is the smallest request the device can perform without |
628 | * incurring a performance penalty. For disk drives this is often the |
629 | * physical block size. For RAID arrays it is often the stripe chunk |
630 | * size. A properly aligned multiple of minimum_io_size is the |
631 | * preferred request size for workloads where a high number of I/O |
632 | * operations is desired. |
633 | */ |
634 | void blk_queue_io_min(struct request_queue *q, unsigned int min) |
635 | { |
636 | blk_limits_io_min(&q->limits, min); |
637 | } |
638 | EXPORT_SYMBOL(blk_queue_io_min); |
639 | |
640 | /** |
641 | * blk_limits_io_opt - set optimal request size for a device |
642 | * @limits: the queue limits |
643 | * @opt: smallest I/O size in bytes |
644 | * |
645 | * Description: |
646 | * Storage devices may report an optimal I/O size, which is the |
647 | * device's preferred unit for sustained I/O. This is rarely reported |
648 | * for disk drives. For RAID arrays it is usually the stripe width or |
649 | * the internal track size. A properly aligned multiple of |
650 | * optimal_io_size is the preferred request size for workloads where |
651 | * sustained throughput is desired. |
652 | */ |
653 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) |
654 | { |
655 | limits->io_opt = opt; |
656 | } |
657 | EXPORT_SYMBOL(blk_limits_io_opt); |
658 | |
659 | /** |
660 | * blk_queue_io_opt - set optimal request size for the queue |
661 | * @q: the request queue for the device |
662 | * @opt: optimal request size in bytes |
663 | * |
664 | * Description: |
665 | * Storage devices may report an optimal I/O size, which is the |
666 | * device's preferred unit for sustained I/O. This is rarely reported |
667 | * for disk drives. For RAID arrays it is usually the stripe width or |
668 | * the internal track size. A properly aligned multiple of |
669 | * optimal_io_size is the preferred request size for workloads where |
670 | * sustained throughput is desired. |
671 | */ |
672 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) |
673 | { |
674 | blk_limits_io_opt(&q->limits, opt); |
675 | if (!q->disk) |
676 | return; |
677 | q->disk->bdi->ra_pages = |
678 | max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES); |
679 | } |
680 | EXPORT_SYMBOL(blk_queue_io_opt); |
681 | |
682 | static int queue_limit_alignment_offset(const struct queue_limits *lim, |
683 | sector_t sector) |
684 | { |
685 | unsigned int granularity = max(lim->physical_block_size, lim->io_min); |
686 | unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT) |
687 | << SECTOR_SHIFT; |
688 | |
689 | return (granularity + lim->alignment_offset - alignment) % granularity; |
690 | } |
691 | |
692 | static unsigned int queue_limit_discard_alignment( |
693 | const struct queue_limits *lim, sector_t sector) |
694 | { |
695 | unsigned int alignment, granularity, offset; |
696 | |
697 | if (!lim->max_discard_sectors) |
698 | return 0; |
699 | |
700 | /* Why are these in bytes, not sectors? */ |
701 | alignment = lim->discard_alignment >> SECTOR_SHIFT; |
702 | granularity = lim->discard_granularity >> SECTOR_SHIFT; |
703 | if (!granularity) |
704 | return 0; |
705 | |
706 | /* Offset of the partition start in 'granularity' sectors */ |
707 | offset = sector_div(sector, granularity); |
708 | |
709 | /* And why do we do this modulus *again* in blkdev_issue_discard()? */ |
710 | offset = (granularity + alignment - offset) % granularity; |
711 | |
712 | /* Turn it back into bytes, gaah */ |
713 | return offset << SECTOR_SHIFT; |
714 | } |
715 | |
716 | static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs) |
717 | { |
718 | sectors = round_down(sectors, lbs >> SECTOR_SHIFT); |
719 | if (sectors < PAGE_SIZE >> SECTOR_SHIFT) |
720 | sectors = PAGE_SIZE >> SECTOR_SHIFT; |
721 | return sectors; |
722 | } |
723 | |
724 | /** |
725 | * blk_stack_limits - adjust queue_limits for stacked devices |
726 | * @t: the stacking driver limits (top device) |
727 | * @b: the underlying queue limits (bottom, component device) |
728 | * @start: first data sector within component device |
729 | * |
730 | * Description: |
731 | * This function is used by stacking drivers like MD and DM to ensure |
732 | * that all component devices have compatible block sizes and |
733 | * alignments. The stacking driver must provide a queue_limits |
734 | * struct (top) and then iteratively call the stacking function for |
735 | * all component (bottom) devices. The stacking function will |
736 | * attempt to combine the values and ensure proper alignment. |
737 | * |
738 | * Returns 0 if the top and bottom queue_limits are compatible. The |
739 | * top device's block sizes and alignment offsets may be adjusted to |
740 | * ensure alignment with the bottom device. If no compatible sizes |
741 | * and alignments exist, -1 is returned and the resulting top |
742 | * queue_limits will have the misaligned flag set to indicate that |
743 | * the alignment_offset is undefined. |
744 | */ |
745 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, |
746 | sector_t start) |
747 | { |
748 | unsigned int top, bottom, alignment, ret = 0; |
749 | |
750 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
751 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); |
752 | t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); |
753 | t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, |
754 | b->max_write_zeroes_sectors); |
755 | t->max_zone_append_sectors = min(t->max_zone_append_sectors, |
756 | b->max_zone_append_sectors); |
757 | t->bounce = max(t->bounce, b->bounce); |
758 | |
759 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, |
760 | b->seg_boundary_mask); |
761 | t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, |
762 | b->virt_boundary_mask); |
763 | |
764 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
765 | t->max_discard_segments = min_not_zero(t->max_discard_segments, |
766 | b->max_discard_segments); |
767 | t->max_integrity_segments = min_not_zero(t->max_integrity_segments, |
768 | b->max_integrity_segments); |
769 | |
770 | t->max_segment_size = min_not_zero(t->max_segment_size, |
771 | b->max_segment_size); |
772 | |
773 | t->misaligned |= b->misaligned; |
774 | |
775 | alignment = queue_limit_alignment_offset(lim: b, sector: start); |
776 | |
777 | /* Bottom device has different alignment. Check that it is |
778 | * compatible with the current top alignment. |
779 | */ |
780 | if (t->alignment_offset != alignment) { |
781 | |
782 | top = max(t->physical_block_size, t->io_min) |
783 | + t->alignment_offset; |
784 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
785 | |
786 | /* Verify that top and bottom intervals line up */ |
787 | if (max(top, bottom) % min(top, bottom)) { |
788 | t->misaligned = 1; |
789 | ret = -1; |
790 | } |
791 | } |
792 | |
793 | t->logical_block_size = max(t->logical_block_size, |
794 | b->logical_block_size); |
795 | |
796 | t->physical_block_size = max(t->physical_block_size, |
797 | b->physical_block_size); |
798 | |
799 | t->io_min = max(t->io_min, b->io_min); |
800 | t->io_opt = lcm_not_zero(a: t->io_opt, b: b->io_opt); |
801 | t->dma_alignment = max(t->dma_alignment, b->dma_alignment); |
802 | |
803 | /* Set non-power-of-2 compatible chunk_sectors boundary */ |
804 | if (b->chunk_sectors) |
805 | t->chunk_sectors = gcd(a: t->chunk_sectors, b: b->chunk_sectors); |
806 | |
807 | /* Physical block size a multiple of the logical block size? */ |
808 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
809 | t->physical_block_size = t->logical_block_size; |
810 | t->misaligned = 1; |
811 | ret = -1; |
812 | } |
813 | |
814 | /* Minimum I/O a multiple of the physical block size? */ |
815 | if (t->io_min & (t->physical_block_size - 1)) { |
816 | t->io_min = t->physical_block_size; |
817 | t->misaligned = 1; |
818 | ret = -1; |
819 | } |
820 | |
821 | /* Optimal I/O a multiple of the physical block size? */ |
822 | if (t->io_opt & (t->physical_block_size - 1)) { |
823 | t->io_opt = 0; |
824 | t->misaligned = 1; |
825 | ret = -1; |
826 | } |
827 | |
828 | /* chunk_sectors a multiple of the physical block size? */ |
829 | if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) { |
830 | t->chunk_sectors = 0; |
831 | t->misaligned = 1; |
832 | ret = -1; |
833 | } |
834 | |
835 | t->raid_partial_stripes_expensive = |
836 | max(t->raid_partial_stripes_expensive, |
837 | b->raid_partial_stripes_expensive); |
838 | |
839 | /* Find lowest common alignment_offset */ |
840 | t->alignment_offset = lcm_not_zero(a: t->alignment_offset, b: alignment) |
841 | % max(t->physical_block_size, t->io_min); |
842 | |
843 | /* Verify that new alignment_offset is on a logical block boundary */ |
844 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
845 | t->misaligned = 1; |
846 | ret = -1; |
847 | } |
848 | |
849 | t->max_sectors = blk_round_down_sectors(sectors: t->max_sectors, lbs: t->logical_block_size); |
850 | t->max_hw_sectors = blk_round_down_sectors(sectors: t->max_hw_sectors, lbs: t->logical_block_size); |
851 | t->max_dev_sectors = blk_round_down_sectors(sectors: t->max_dev_sectors, lbs: t->logical_block_size); |
852 | |
853 | /* Discard alignment and granularity */ |
854 | if (b->discard_granularity) { |
855 | alignment = queue_limit_discard_alignment(lim: b, sector: start); |
856 | |
857 | if (t->discard_granularity != 0 && |
858 | t->discard_alignment != alignment) { |
859 | top = t->discard_granularity + t->discard_alignment; |
860 | bottom = b->discard_granularity + alignment; |
861 | |
862 | /* Verify that top and bottom intervals line up */ |
863 | if ((max(top, bottom) % min(top, bottom)) != 0) |
864 | t->discard_misaligned = 1; |
865 | } |
866 | |
867 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
868 | b->max_discard_sectors); |
869 | t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, |
870 | b->max_hw_discard_sectors); |
871 | t->discard_granularity = max(t->discard_granularity, |
872 | b->discard_granularity); |
873 | t->discard_alignment = lcm_not_zero(a: t->discard_alignment, b: alignment) % |
874 | t->discard_granularity; |
875 | } |
876 | t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors, |
877 | b->max_secure_erase_sectors); |
878 | t->zone_write_granularity = max(t->zone_write_granularity, |
879 | b->zone_write_granularity); |
880 | t->zoned = max(t->zoned, b->zoned); |
881 | if (!t->zoned) { |
882 | t->zone_write_granularity = 0; |
883 | t->max_zone_append_sectors = 0; |
884 | } |
885 | return ret; |
886 | } |
887 | EXPORT_SYMBOL(blk_stack_limits); |
888 | |
889 | /** |
890 | * queue_limits_stack_bdev - adjust queue_limits for stacked devices |
891 | * @t: the stacking driver limits (top device) |
892 | * @bdev: the underlying block device (bottom) |
893 | * @offset: offset to beginning of data within component device |
894 | * @pfx: prefix to use for warnings logged |
895 | * |
896 | * Description: |
897 | * This function is used by stacking drivers like MD and DM to ensure |
898 | * that all component devices have compatible block sizes and |
899 | * alignments. The stacking driver must provide a queue_limits |
900 | * struct (top) and then iteratively call the stacking function for |
901 | * all component (bottom) devices. The stacking function will |
902 | * attempt to combine the values and ensure proper alignment. |
903 | */ |
904 | void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev, |
905 | sector_t offset, const char *pfx) |
906 | { |
907 | if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits, |
908 | get_start_sect(bdev) + offset)) |
909 | pr_notice("%s: Warning: Device %pg is misaligned\n" , |
910 | pfx, bdev); |
911 | } |
912 | EXPORT_SYMBOL_GPL(queue_limits_stack_bdev); |
913 | |
914 | /** |
915 | * blk_queue_update_dma_pad - update pad mask |
916 | * @q: the request queue for the device |
917 | * @mask: pad mask |
918 | * |
919 | * Update dma pad mask. |
920 | * |
921 | * Appending pad buffer to a request modifies the last entry of a |
922 | * scatter list such that it includes the pad buffer. |
923 | **/ |
924 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) |
925 | { |
926 | if (mask > q->dma_pad_mask) |
927 | q->dma_pad_mask = mask; |
928 | } |
929 | EXPORT_SYMBOL(blk_queue_update_dma_pad); |
930 | |
931 | /** |
932 | * blk_queue_segment_boundary - set boundary rules for segment merging |
933 | * @q: the request queue for the device |
934 | * @mask: the memory boundary mask |
935 | **/ |
936 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) |
937 | { |
938 | if (mask < PAGE_SIZE - 1) { |
939 | mask = PAGE_SIZE - 1; |
940 | pr_info("%s: set to minimum %lx\n" , __func__, mask); |
941 | } |
942 | |
943 | q->limits.seg_boundary_mask = mask; |
944 | } |
945 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
946 | |
947 | /** |
948 | * blk_queue_virt_boundary - set boundary rules for bio merging |
949 | * @q: the request queue for the device |
950 | * @mask: the memory boundary mask |
951 | **/ |
952 | void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask) |
953 | { |
954 | q->limits.virt_boundary_mask = mask; |
955 | |
956 | /* |
957 | * Devices that require a virtual boundary do not support scatter/gather |
958 | * I/O natively, but instead require a descriptor list entry for each |
959 | * page (which might not be idential to the Linux PAGE_SIZE). Because |
960 | * of that they are not limited by our notion of "segment size". |
961 | */ |
962 | if (mask) |
963 | q->limits.max_segment_size = UINT_MAX; |
964 | } |
965 | EXPORT_SYMBOL(blk_queue_virt_boundary); |
966 | |
967 | /** |
968 | * blk_queue_dma_alignment - set dma length and memory alignment |
969 | * @q: the request queue for the device |
970 | * @mask: alignment mask |
971 | * |
972 | * description: |
973 | * set required memory and length alignment for direct dma transactions. |
974 | * this is used when building direct io requests for the queue. |
975 | * |
976 | **/ |
977 | void blk_queue_dma_alignment(struct request_queue *q, int mask) |
978 | { |
979 | q->limits.dma_alignment = mask; |
980 | } |
981 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
982 | |
983 | /** |
984 | * blk_queue_update_dma_alignment - update dma length and memory alignment |
985 | * @q: the request queue for the device |
986 | * @mask: alignment mask |
987 | * |
988 | * description: |
989 | * update required memory and length alignment for direct dma transactions. |
990 | * If the requested alignment is larger than the current alignment, then |
991 | * the current queue alignment is updated to the new value, otherwise it |
992 | * is left alone. The design of this is to allow multiple objects |
993 | * (driver, device, transport etc) to set their respective |
994 | * alignments without having them interfere. |
995 | * |
996 | **/ |
997 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) |
998 | { |
999 | BUG_ON(mask > PAGE_SIZE); |
1000 | |
1001 | if (mask > q->limits.dma_alignment) |
1002 | q->limits.dma_alignment = mask; |
1003 | } |
1004 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
1005 | |
1006 | /** |
1007 | * blk_set_queue_depth - tell the block layer about the device queue depth |
1008 | * @q: the request queue for the device |
1009 | * @depth: queue depth |
1010 | * |
1011 | */ |
1012 | void blk_set_queue_depth(struct request_queue *q, unsigned int depth) |
1013 | { |
1014 | q->queue_depth = depth; |
1015 | rq_qos_queue_depth_changed(q); |
1016 | } |
1017 | EXPORT_SYMBOL(blk_set_queue_depth); |
1018 | |
1019 | /** |
1020 | * blk_queue_write_cache - configure queue's write cache |
1021 | * @q: the request queue for the device |
1022 | * @wc: write back cache on or off |
1023 | * @fua: device supports FUA writes, if true |
1024 | * |
1025 | * Tell the block layer about the write cache of @q. |
1026 | */ |
1027 | void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua) |
1028 | { |
1029 | if (wc) { |
1030 | blk_queue_flag_set(QUEUE_FLAG_HW_WC, q); |
1031 | blk_queue_flag_set(QUEUE_FLAG_WC, q); |
1032 | } else { |
1033 | blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q); |
1034 | blk_queue_flag_clear(QUEUE_FLAG_WC, q); |
1035 | } |
1036 | if (fua) |
1037 | blk_queue_flag_set(QUEUE_FLAG_FUA, q); |
1038 | else |
1039 | blk_queue_flag_clear(QUEUE_FLAG_FUA, q); |
1040 | } |
1041 | EXPORT_SYMBOL_GPL(blk_queue_write_cache); |
1042 | |
1043 | /** |
1044 | * blk_queue_required_elevator_features - Set a queue required elevator features |
1045 | * @q: the request queue for the target device |
1046 | * @features: Required elevator features OR'ed together |
1047 | * |
1048 | * Tell the block layer that for the device controlled through @q, only the |
1049 | * only elevators that can be used are those that implement at least the set of |
1050 | * features specified by @features. |
1051 | */ |
1052 | void blk_queue_required_elevator_features(struct request_queue *q, |
1053 | unsigned int features) |
1054 | { |
1055 | q->required_elevator_features = features; |
1056 | } |
1057 | EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features); |
1058 | |
1059 | /** |
1060 | * blk_queue_can_use_dma_map_merging - configure queue for merging segments. |
1061 | * @q: the request queue for the device |
1062 | * @dev: the device pointer for dma |
1063 | * |
1064 | * Tell the block layer about merging the segments by dma map of @q. |
1065 | */ |
1066 | bool blk_queue_can_use_dma_map_merging(struct request_queue *q, |
1067 | struct device *dev) |
1068 | { |
1069 | unsigned long boundary = dma_get_merge_boundary(dev); |
1070 | |
1071 | if (!boundary) |
1072 | return false; |
1073 | |
1074 | /* No need to update max_segment_size. see blk_queue_virt_boundary() */ |
1075 | blk_queue_virt_boundary(q, boundary); |
1076 | |
1077 | return true; |
1078 | } |
1079 | EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging); |
1080 | |
1081 | /** |
1082 | * disk_set_zoned - inidicate a zoned device |
1083 | * @disk: gendisk to configure |
1084 | */ |
1085 | void disk_set_zoned(struct gendisk *disk) |
1086 | { |
1087 | struct request_queue *q = disk->queue; |
1088 | |
1089 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)); |
1090 | |
1091 | /* |
1092 | * Set the zone write granularity to the device logical block |
1093 | * size by default. The driver can change this value if needed. |
1094 | */ |
1095 | q->limits.zoned = true; |
1096 | blk_queue_zone_write_granularity(q, queue_logical_block_size(q)); |
1097 | } |
1098 | EXPORT_SYMBOL_GPL(disk_set_zoned); |
1099 | |
1100 | int bdev_alignment_offset(struct block_device *bdev) |
1101 | { |
1102 | struct request_queue *q = bdev_get_queue(bdev); |
1103 | |
1104 | if (q->limits.misaligned) |
1105 | return -1; |
1106 | if (bdev_is_partition(bdev)) |
1107 | return queue_limit_alignment_offset(lim: &q->limits, |
1108 | sector: bdev->bd_start_sect); |
1109 | return q->limits.alignment_offset; |
1110 | } |
1111 | EXPORT_SYMBOL_GPL(bdev_alignment_offset); |
1112 | |
1113 | unsigned int bdev_discard_alignment(struct block_device *bdev) |
1114 | { |
1115 | struct request_queue *q = bdev_get_queue(bdev); |
1116 | |
1117 | if (bdev_is_partition(bdev)) |
1118 | return queue_limit_discard_alignment(lim: &q->limits, |
1119 | sector: bdev->bd_start_sect); |
1120 | return q->limits.discard_alignment; |
1121 | } |
1122 | EXPORT_SYMBOL_GPL(bdev_discard_alignment); |
1123 | |