Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
86db1e29 JA |
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> | |
4ee60ec1 | 10 | #include <linux/pagemap.h> |
edb0872f | 11 | #include <linux/backing-dev-defs.h> |
70dd5bf3 | 12 | #include <linux/gcd.h> |
2cda2728 | 13 | #include <linux/lcm.h> |
ad5ebd2f | 14 | #include <linux/jiffies.h> |
5a0e3ad6 | 15 | #include <linux/gfp.h> |
45147fb5 | 16 | #include <linux/dma-mapping.h> |
86db1e29 JA |
17 | |
18 | #include "blk.h" | |
0bc65bd4 | 19 | #include "blk-rq-qos.h" |
87760e5e | 20 | #include "blk-wbt.h" |
86db1e29 | 21 | |
242f9dcb JA |
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 | ||
b1bd055d MP |
28 | /** |
29 | * blk_set_stacking_limits - set default limits for stacking devices | |
30 | * @lim: the queue_limits structure to reset | |
31 | * | |
c490f226 CH |
32 | * Prepare queue limits for applying limits from underlying devices using |
33 | * blk_stack_limits(). | |
b1bd055d MP |
34 | */ |
35 | void blk_set_stacking_limits(struct queue_limits *lim) | |
36 | { | |
c490f226 CH |
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; | |
b1bd055d MP |
44 | |
45 | /* Inherit limits from component devices */ | |
b1bd055d | 46 | lim->max_segments = USHRT_MAX; |
42c9cdfe | 47 | lim->max_discard_segments = USHRT_MAX; |
b1bd055d | 48 | lim->max_hw_sectors = UINT_MAX; |
d82ae52e | 49 | lim->max_segment_size = UINT_MAX; |
fe86cdce | 50 | lim->max_sectors = UINT_MAX; |
ca369d51 | 51 | lim->max_dev_sectors = UINT_MAX; |
a6f0788e | 52 | lim->max_write_zeroes_sectors = UINT_MAX; |
0512a75b | 53 | lim->max_zone_append_sectors = UINT_MAX; |
4f563a64 | 54 | lim->max_user_discard_sectors = UINT_MAX; |
b1bd055d MP |
55 | } |
56 | EXPORT_SYMBOL(blk_set_stacking_limits); | |
57 | ||
b9947297 CH |
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 | ||
d690cb8a CH |
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) { | |
038105a2 | 149 | if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE) |
d690cb8a CH |
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 | ||
4f563a64 CH |
165 | lim->max_discard_sectors = |
166 | min(lim->max_hw_discard_sectors, lim->max_user_discard_sectors); | |
167 | ||
d690cb8a CH |
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 | ||
d690cb8a | 184 | /* |
b561ea56 ML |
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. | |
d690cb8a | 190 | */ |
ffd379c1 ML |
191 | if (lim->virt_boundary_mask) { |
192 | if (!lim->max_segment_size) | |
193 | lim->max_segment_size = UINT_MAX; | |
194 | } else { | |
a3911966 CH |
195 | /* |
196 | * The maximum segment size has an odd historic 64k default that | |
197 | * drivers probably should override. Just like the I/O size we | |
198 | * require drivers to at least handle a full page per segment. | |
199 | */ | |
200 | if (!lim->max_segment_size) | |
201 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; | |
202 | if (WARN_ON_ONCE(lim->max_segment_size < PAGE_SIZE)) | |
203 | return -EINVAL; | |
d690cb8a CH |
204 | } |
205 | ||
206 | /* | |
207 | * We require drivers to at least do logical block aligned I/O, but | |
208 | * historically could not check for that due to the separate calls | |
209 | * to set the limits. Once the transition is finished the check | |
210 | * below should be narrowed down to check the logical block size. | |
211 | */ | |
212 | if (!lim->dma_alignment) | |
213 | lim->dma_alignment = SECTOR_SIZE - 1; | |
214 | if (WARN_ON_ONCE(lim->dma_alignment > PAGE_SIZE)) | |
215 | return -EINVAL; | |
216 | ||
217 | if (lim->alignment_offset) { | |
218 | lim->alignment_offset &= (lim->physical_block_size - 1); | |
219 | lim->misaligned = 0; | |
220 | } | |
221 | ||
222 | return blk_validate_zoned_limits(lim); | |
223 | } | |
224 | ||
225 | /* | |
226 | * Set the default limits for a newly allocated queue. @lim contains the | |
227 | * initial limits set by the driver, which could be no limit in which case | |
228 | * all fields are cleared to zero. | |
229 | */ | |
230 | int blk_set_default_limits(struct queue_limits *lim) | |
231 | { | |
4f563a64 CH |
232 | /* |
233 | * Most defaults are set by capping the bounds in blk_validate_limits, | |
234 | * but max_user_discard_sectors is special and needs an explicit | |
235 | * initialization to the max value here. | |
236 | */ | |
237 | lim->max_user_discard_sectors = UINT_MAX; | |
d690cb8a CH |
238 | return blk_validate_limits(lim); |
239 | } | |
240 | ||
241 | /** | |
242 | * queue_limits_commit_update - commit an atomic update of queue limits | |
243 | * @q: queue to update | |
244 | * @lim: limits to apply | |
245 | * | |
246 | * Apply the limits in @lim that were obtained from queue_limits_start_update() | |
247 | * and updated by the caller to @q. | |
248 | * | |
249 | * Returns 0 if successful, else a negative error code. | |
250 | */ | |
251 | int queue_limits_commit_update(struct request_queue *q, | |
252 | struct queue_limits *lim) | |
253 | __releases(q->limits_lock) | |
254 | { | |
255 | int error = blk_validate_limits(lim); | |
256 | ||
257 | if (!error) { | |
258 | q->limits = *lim; | |
259 | if (q->disk) | |
260 | blk_apply_bdi_limits(q->disk->bdi, lim); | |
261 | } | |
262 | mutex_unlock(&q->limits_lock); | |
263 | return error; | |
264 | } | |
265 | EXPORT_SYMBOL_GPL(queue_limits_commit_update); | |
266 | ||
631d4efb | 267 | /** |
4c4ab8ae | 268 | * queue_limits_set - apply queue limits to queue |
631d4efb CH |
269 | * @q: queue to update |
270 | * @lim: limits to apply | |
271 | * | |
272 | * Apply the limits in @lim that were freshly initialized to @q. | |
273 | * To update existing limits use queue_limits_start_update() and | |
274 | * queue_limits_commit_update() instead. | |
275 | * | |
276 | * Returns 0 if successful, else a negative error code. | |
277 | */ | |
278 | int queue_limits_set(struct request_queue *q, struct queue_limits *lim) | |
279 | { | |
280 | mutex_lock(&q->limits_lock); | |
281 | return queue_limits_commit_update(q, lim); | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(queue_limits_set); | |
284 | ||
762380ad JA |
285 | /** |
286 | * blk_queue_chunk_sectors - set size of the chunk for this queue | |
287 | * @q: the request queue for the device | |
288 | * @chunk_sectors: chunk sectors in the usual 512b unit | |
289 | * | |
290 | * Description: | |
291 | * If a driver doesn't want IOs to cross a given chunk size, it can set | |
07d098e6 MS |
292 | * this limit and prevent merging across chunks. Note that the block layer |
293 | * must accept a page worth of data at any offset. So if the crossing of | |
294 | * chunks is a hard limitation in the driver, it must still be prepared | |
295 | * to split single page bios. | |
762380ad JA |
296 | **/ |
297 | void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) | |
298 | { | |
762380ad JA |
299 | q->limits.chunk_sectors = chunk_sectors; |
300 | } | |
301 | EXPORT_SYMBOL(blk_queue_chunk_sectors); | |
302 | ||
67efc925 CH |
303 | /** |
304 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
305 | * @q: the request queue for the device | |
c7ebf065 | 306 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
307 | **/ |
308 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
309 | unsigned int max_discard_sectors) | |
310 | { | |
4f563a64 CH |
311 | struct queue_limits *lim = &q->limits; |
312 | ||
313 | lim->max_hw_discard_sectors = max_discard_sectors; | |
314 | lim->max_discard_sectors = | |
315 | min(max_discard_sectors, lim->max_user_discard_sectors); | |
67efc925 CH |
316 | } |
317 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
318 | ||
44abff2c CH |
319 | /** |
320 | * blk_queue_max_secure_erase_sectors - set max sectors for a secure erase | |
321 | * @q: the request queue for the device | |
322 | * @max_sectors: maximum number of sectors to secure_erase | |
323 | **/ | |
324 | void blk_queue_max_secure_erase_sectors(struct request_queue *q, | |
325 | unsigned int max_sectors) | |
326 | { | |
327 | q->limits.max_secure_erase_sectors = max_sectors; | |
328 | } | |
329 | EXPORT_SYMBOL(blk_queue_max_secure_erase_sectors); | |
330 | ||
a6f0788e CK |
331 | /** |
332 | * blk_queue_max_write_zeroes_sectors - set max sectors for a single | |
333 | * write zeroes | |
334 | * @q: the request queue for the device | |
335 | * @max_write_zeroes_sectors: maximum number of sectors to write per command | |
336 | **/ | |
337 | void blk_queue_max_write_zeroes_sectors(struct request_queue *q, | |
338 | unsigned int max_write_zeroes_sectors) | |
339 | { | |
340 | q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors; | |
341 | } | |
342 | EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors); | |
343 | ||
0512a75b KB |
344 | /** |
345 | * blk_queue_max_zone_append_sectors - set max sectors for a single zone append | |
346 | * @q: the request queue for the device | |
347 | * @max_zone_append_sectors: maximum number of sectors to write per command | |
ccdbf0aa DLM |
348 | * |
349 | * Sets the maximum number of sectors allowed for zone append commands. If | |
350 | * Specifying 0 for @max_zone_append_sectors indicates that the queue does | |
351 | * not natively support zone append operations and that the block layer must | |
352 | * emulate these operations using regular writes. | |
0512a75b KB |
353 | **/ |
354 | void blk_queue_max_zone_append_sectors(struct request_queue *q, | |
355 | unsigned int max_zone_append_sectors) | |
356 | { | |
ccdbf0aa | 357 | unsigned int max_sectors = 0; |
0512a75b KB |
358 | |
359 | if (WARN_ON(!blk_queue_is_zoned(q))) | |
360 | return; | |
361 | ||
ccdbf0aa DLM |
362 | if (max_zone_append_sectors) { |
363 | max_sectors = min(q->limits.max_hw_sectors, | |
364 | max_zone_append_sectors); | |
365 | max_sectors = min(q->limits.chunk_sectors, max_sectors); | |
0512a75b | 366 | |
ccdbf0aa DLM |
367 | /* |
368 | * Signal eventual driver bugs resulting in the max_zone_append | |
369 | * sectors limit being 0 due to the chunk_sectors limit (zone | |
370 | * size) not set or the max_hw_sectors limit not set. | |
371 | */ | |
372 | WARN_ON_ONCE(!max_sectors); | |
373 | } | |
0512a75b KB |
374 | |
375 | q->limits.max_zone_append_sectors = max_sectors; | |
376 | } | |
377 | EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors); | |
378 | ||
86db1e29 | 379 | /** |
e1defc4f | 380 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 381 | * @q: the request queue for the device |
e1defc4f | 382 | * @size: the logical block size, in bytes |
86db1e29 JA |
383 | * |
384 | * Description: | |
e1defc4f MP |
385 | * This should be set to the lowest possible block size that the |
386 | * storage device can address. The default of 512 covers most | |
387 | * hardware. | |
86db1e29 | 388 | **/ |
ad6bf88a | 389 | void blk_queue_logical_block_size(struct request_queue *q, unsigned int size) |
86db1e29 | 390 | { |
817046ec DLM |
391 | struct queue_limits *limits = &q->limits; |
392 | ||
393 | limits->logical_block_size = size; | |
394 | ||
3c407dc7 CH |
395 | if (limits->discard_granularity < limits->logical_block_size) |
396 | limits->discard_granularity = limits->logical_block_size; | |
397 | ||
817046ec DLM |
398 | if (limits->physical_block_size < size) |
399 | limits->physical_block_size = size; | |
c72758f3 | 400 | |
817046ec DLM |
401 | if (limits->io_min < limits->physical_block_size) |
402 | limits->io_min = limits->physical_block_size; | |
c72758f3 | 403 | |
817046ec DLM |
404 | limits->max_hw_sectors = |
405 | round_down(limits->max_hw_sectors, size >> SECTOR_SHIFT); | |
406 | limits->max_sectors = | |
407 | round_down(limits->max_sectors, size >> SECTOR_SHIFT); | |
86db1e29 | 408 | } |
e1defc4f | 409 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 410 | |
c72758f3 MP |
411 | /** |
412 | * blk_queue_physical_block_size - set physical block size for the queue | |
413 | * @q: the request queue for the device | |
414 | * @size: the physical block size, in bytes | |
415 | * | |
416 | * Description: | |
417 | * This should be set to the lowest possible sector size that the | |
418 | * hardware can operate on without reverting to read-modify-write | |
419 | * operations. | |
420 | */ | |
892b6f90 | 421 | void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) |
c72758f3 MP |
422 | { |
423 | q->limits.physical_block_size = size; | |
424 | ||
425 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
426 | q->limits.physical_block_size = q->limits.logical_block_size; | |
427 | ||
458aa1a0 CH |
428 | if (q->limits.discard_granularity < q->limits.physical_block_size) |
429 | q->limits.discard_granularity = q->limits.physical_block_size; | |
430 | ||
c72758f3 MP |
431 | if (q->limits.io_min < q->limits.physical_block_size) |
432 | q->limits.io_min = q->limits.physical_block_size; | |
433 | } | |
434 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
435 | ||
a805a4fa DLM |
436 | /** |
437 | * blk_queue_zone_write_granularity - set zone write granularity for the queue | |
438 | * @q: the request queue for the zoned device | |
439 | * @size: the zone write granularity size, in bytes | |
440 | * | |
441 | * Description: | |
442 | * This should be set to the lowest possible size allowing to write in | |
443 | * sequential zones of a zoned block device. | |
444 | */ | |
445 | void blk_queue_zone_write_granularity(struct request_queue *q, | |
446 | unsigned int size) | |
447 | { | |
448 | if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) | |
449 | return; | |
450 | ||
451 | q->limits.zone_write_granularity = size; | |
452 | ||
453 | if (q->limits.zone_write_granularity < q->limits.logical_block_size) | |
454 | q->limits.zone_write_granularity = q->limits.logical_block_size; | |
455 | } | |
456 | EXPORT_SYMBOL_GPL(blk_queue_zone_write_granularity); | |
457 | ||
c72758f3 MP |
458 | /** |
459 | * blk_queue_alignment_offset - set physical block alignment offset | |
460 | * @q: the request queue for the device | |
8ebf9756 | 461 | * @offset: alignment offset in bytes |
c72758f3 MP |
462 | * |
463 | * Description: | |
464 | * Some devices are naturally misaligned to compensate for things like | |
465 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
466 | * should call this function for devices whose first sector is not | |
467 | * naturally aligned. | |
468 | */ | |
469 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
470 | { | |
471 | q->limits.alignment_offset = | |
472 | offset & (q->limits.physical_block_size - 1); | |
473 | q->limits.misaligned = 0; | |
474 | } | |
475 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
476 | ||
471aa704 | 477 | void disk_update_readahead(struct gendisk *disk) |
c2e4cd57 | 478 | { |
b9947297 | 479 | blk_apply_bdi_limits(disk->bdi, &disk->queue->limits); |
c2e4cd57 | 480 | } |
471aa704 | 481 | EXPORT_SYMBOL_GPL(disk_update_readahead); |
c2e4cd57 | 482 | |
7c958e32 MP |
483 | /** |
484 | * blk_limits_io_min - set minimum request size for a device | |
485 | * @limits: the queue limits | |
486 | * @min: smallest I/O size in bytes | |
487 | * | |
488 | * Description: | |
489 | * Some devices have an internal block size bigger than the reported | |
490 | * hardware sector size. This function can be used to signal the | |
491 | * smallest I/O the device can perform without incurring a performance | |
492 | * penalty. | |
493 | */ | |
494 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
495 | { | |
496 | limits->io_min = min; | |
497 | ||
498 | if (limits->io_min < limits->logical_block_size) | |
499 | limits->io_min = limits->logical_block_size; | |
500 | ||
501 | if (limits->io_min < limits->physical_block_size) | |
502 | limits->io_min = limits->physical_block_size; | |
503 | } | |
504 | EXPORT_SYMBOL(blk_limits_io_min); | |
505 | ||
c72758f3 MP |
506 | /** |
507 | * blk_queue_io_min - set minimum request size for the queue | |
508 | * @q: the request queue for the device | |
8ebf9756 | 509 | * @min: smallest I/O size in bytes |
c72758f3 MP |
510 | * |
511 | * Description: | |
7e5f5fb0 MP |
512 | * Storage devices may report a granularity or preferred minimum I/O |
513 | * size which is the smallest request the device can perform without | |
514 | * incurring a performance penalty. For disk drives this is often the | |
515 | * physical block size. For RAID arrays it is often the stripe chunk | |
516 | * size. A properly aligned multiple of minimum_io_size is the | |
517 | * preferred request size for workloads where a high number of I/O | |
518 | * operations is desired. | |
c72758f3 MP |
519 | */ |
520 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
521 | { | |
7c958e32 | 522 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
523 | } |
524 | EXPORT_SYMBOL(blk_queue_io_min); | |
525 | ||
3c5820c7 MP |
526 | /** |
527 | * blk_limits_io_opt - set optimal request size for a device | |
528 | * @limits: the queue limits | |
529 | * @opt: smallest I/O size in bytes | |
530 | * | |
531 | * Description: | |
532 | * Storage devices may report an optimal I/O size, which is the | |
533 | * device's preferred unit for sustained I/O. This is rarely reported | |
534 | * for disk drives. For RAID arrays it is usually the stripe width or | |
535 | * the internal track size. A properly aligned multiple of | |
536 | * optimal_io_size is the preferred request size for workloads where | |
537 | * sustained throughput is desired. | |
538 | */ | |
539 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
540 | { | |
541 | limits->io_opt = opt; | |
542 | } | |
543 | EXPORT_SYMBOL(blk_limits_io_opt); | |
544 | ||
aa261f20 | 545 | static int queue_limit_alignment_offset(const struct queue_limits *lim, |
89098b07 CH |
546 | sector_t sector) |
547 | { | |
548 | unsigned int granularity = max(lim->physical_block_size, lim->io_min); | |
549 | unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT) | |
550 | << SECTOR_SHIFT; | |
551 | ||
552 | return (granularity + lim->alignment_offset - alignment) % granularity; | |
553 | } | |
554 | ||
aa261f20 BVA |
555 | static unsigned int queue_limit_discard_alignment( |
556 | const struct queue_limits *lim, sector_t sector) | |
5c4b4a5c CH |
557 | { |
558 | unsigned int alignment, granularity, offset; | |
559 | ||
560 | if (!lim->max_discard_sectors) | |
561 | return 0; | |
562 | ||
563 | /* Why are these in bytes, not sectors? */ | |
564 | alignment = lim->discard_alignment >> SECTOR_SHIFT; | |
565 | granularity = lim->discard_granularity >> SECTOR_SHIFT; | |
566 | if (!granularity) | |
567 | return 0; | |
568 | ||
569 | /* Offset of the partition start in 'granularity' sectors */ | |
570 | offset = sector_div(sector, granularity); | |
571 | ||
572 | /* And why do we do this modulus *again* in blkdev_issue_discard()? */ | |
573 | offset = (granularity + alignment - offset) % granularity; | |
574 | ||
575 | /* Turn it back into bytes, gaah */ | |
576 | return offset << SECTOR_SHIFT; | |
577 | } | |
578 | ||
97f433c3 MP |
579 | static unsigned int blk_round_down_sectors(unsigned int sectors, unsigned int lbs) |
580 | { | |
581 | sectors = round_down(sectors, lbs >> SECTOR_SHIFT); | |
582 | if (sectors < PAGE_SIZE >> SECTOR_SHIFT) | |
583 | sectors = PAGE_SIZE >> SECTOR_SHIFT; | |
584 | return sectors; | |
585 | } | |
586 | ||
c72758f3 MP |
587 | /** |
588 | * blk_stack_limits - adjust queue_limits for stacked devices | |
81744ee4 MP |
589 | * @t: the stacking driver limits (top device) |
590 | * @b: the underlying queue limits (bottom, component device) | |
e03a72e1 | 591 | * @start: first data sector within component device |
c72758f3 MP |
592 | * |
593 | * Description: | |
81744ee4 MP |
594 | * This function is used by stacking drivers like MD and DM to ensure |
595 | * that all component devices have compatible block sizes and | |
596 | * alignments. The stacking driver must provide a queue_limits | |
597 | * struct (top) and then iteratively call the stacking function for | |
598 | * all component (bottom) devices. The stacking function will | |
599 | * attempt to combine the values and ensure proper alignment. | |
600 | * | |
601 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
602 | * top device's block sizes and alignment offsets may be adjusted to | |
603 | * ensure alignment with the bottom device. If no compatible sizes | |
604 | * and alignments exist, -1 is returned and the resulting top | |
605 | * queue_limits will have the misaligned flag set to indicate that | |
606 | * the alignment_offset is undefined. | |
c72758f3 MP |
607 | */ |
608 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
e03a72e1 | 609 | sector_t start) |
c72758f3 | 610 | { |
e03a72e1 | 611 | unsigned int top, bottom, alignment, ret = 0; |
86b37281 | 612 | |
c72758f3 MP |
613 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
614 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
ca369d51 | 615 | t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); |
a6f0788e CK |
616 | t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, |
617 | b->max_write_zeroes_sectors); | |
ccdbf0aa DLM |
618 | t->max_zone_append_sectors = min(queue_limits_max_zone_append_sectors(t), |
619 | queue_limits_max_zone_append_sectors(b)); | |
9bb33f24 | 620 | t->bounce = max(t->bounce, b->bounce); |
c72758f3 MP |
621 | |
622 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
623 | b->seg_boundary_mask); | |
03100aad KB |
624 | t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, |
625 | b->virt_boundary_mask); | |
c72758f3 | 626 | |
8a78362c | 627 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
1e739730 CH |
628 | t->max_discard_segments = min_not_zero(t->max_discard_segments, |
629 | b->max_discard_segments); | |
13f05c8d MP |
630 | t->max_integrity_segments = min_not_zero(t->max_integrity_segments, |
631 | b->max_integrity_segments); | |
c72758f3 MP |
632 | |
633 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
634 | b->max_segment_size); | |
635 | ||
fe0b393f MP |
636 | t->misaligned |= b->misaligned; |
637 | ||
e03a72e1 | 638 | alignment = queue_limit_alignment_offset(b, start); |
9504e086 | 639 | |
81744ee4 MP |
640 | /* Bottom device has different alignment. Check that it is |
641 | * compatible with the current top alignment. | |
642 | */ | |
9504e086 MP |
643 | if (t->alignment_offset != alignment) { |
644 | ||
645 | top = max(t->physical_block_size, t->io_min) | |
646 | + t->alignment_offset; | |
81744ee4 | 647 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
9504e086 | 648 | |
81744ee4 | 649 | /* Verify that top and bottom intervals line up */ |
b8839b8c | 650 | if (max(top, bottom) % min(top, bottom)) { |
9504e086 | 651 | t->misaligned = 1; |
fe0b393f MP |
652 | ret = -1; |
653 | } | |
9504e086 MP |
654 | } |
655 | ||
c72758f3 MP |
656 | t->logical_block_size = max(t->logical_block_size, |
657 | b->logical_block_size); | |
658 | ||
659 | t->physical_block_size = max(t->physical_block_size, | |
660 | b->physical_block_size); | |
661 | ||
662 | t->io_min = max(t->io_min, b->io_min); | |
e9637415 | 663 | t->io_opt = lcm_not_zero(t->io_opt, b->io_opt); |
c964d62f | 664 | t->dma_alignment = max(t->dma_alignment, b->dma_alignment); |
7e7986f9 MS |
665 | |
666 | /* Set non-power-of-2 compatible chunk_sectors boundary */ | |
667 | if (b->chunk_sectors) | |
668 | t->chunk_sectors = gcd(t->chunk_sectors, b->chunk_sectors); | |
9504e086 | 669 | |
81744ee4 | 670 | /* Physical block size a multiple of the logical block size? */ |
9504e086 MP |
671 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
672 | t->physical_block_size = t->logical_block_size; | |
c72758f3 | 673 | t->misaligned = 1; |
fe0b393f | 674 | ret = -1; |
86b37281 MP |
675 | } |
676 | ||
81744ee4 | 677 | /* Minimum I/O a multiple of the physical block size? */ |
9504e086 MP |
678 | if (t->io_min & (t->physical_block_size - 1)) { |
679 | t->io_min = t->physical_block_size; | |
680 | t->misaligned = 1; | |
fe0b393f | 681 | ret = -1; |
c72758f3 MP |
682 | } |
683 | ||
81744ee4 | 684 | /* Optimal I/O a multiple of the physical block size? */ |
9504e086 MP |
685 | if (t->io_opt & (t->physical_block_size - 1)) { |
686 | t->io_opt = 0; | |
687 | t->misaligned = 1; | |
fe0b393f | 688 | ret = -1; |
9504e086 | 689 | } |
c72758f3 | 690 | |
22ada802 MS |
691 | /* chunk_sectors a multiple of the physical block size? */ |
692 | if ((t->chunk_sectors << 9) & (t->physical_block_size - 1)) { | |
693 | t->chunk_sectors = 0; | |
694 | t->misaligned = 1; | |
695 | ret = -1; | |
696 | } | |
697 | ||
c78afc62 KO |
698 | t->raid_partial_stripes_expensive = |
699 | max(t->raid_partial_stripes_expensive, | |
700 | b->raid_partial_stripes_expensive); | |
701 | ||
81744ee4 | 702 | /* Find lowest common alignment_offset */ |
e9637415 | 703 | t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment) |
b8839b8c | 704 | % max(t->physical_block_size, t->io_min); |
86b37281 | 705 | |
81744ee4 | 706 | /* Verify that new alignment_offset is on a logical block boundary */ |
fe0b393f | 707 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
c72758f3 | 708 | t->misaligned = 1; |
fe0b393f MP |
709 | ret = -1; |
710 | } | |
c72758f3 | 711 | |
97f433c3 MP |
712 | t->max_sectors = blk_round_down_sectors(t->max_sectors, t->logical_block_size); |
713 | t->max_hw_sectors = blk_round_down_sectors(t->max_hw_sectors, t->logical_block_size); | |
714 | t->max_dev_sectors = blk_round_down_sectors(t->max_dev_sectors, t->logical_block_size); | |
715 | ||
9504e086 MP |
716 | /* Discard alignment and granularity */ |
717 | if (b->discard_granularity) { | |
e03a72e1 | 718 | alignment = queue_limit_discard_alignment(b, start); |
9504e086 MP |
719 | |
720 | if (t->discard_granularity != 0 && | |
721 | t->discard_alignment != alignment) { | |
722 | top = t->discard_granularity + t->discard_alignment; | |
723 | bottom = b->discard_granularity + alignment; | |
70dd5bf3 | 724 | |
9504e086 | 725 | /* Verify that top and bottom intervals line up */ |
8dd2cb7e | 726 | if ((max(top, bottom) % min(top, bottom)) != 0) |
9504e086 MP |
727 | t->discard_misaligned = 1; |
728 | } | |
729 | ||
81744ee4 MP |
730 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
731 | b->max_discard_sectors); | |
0034af03 JA |
732 | t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, |
733 | b->max_hw_discard_sectors); | |
9504e086 MP |
734 | t->discard_granularity = max(t->discard_granularity, |
735 | b->discard_granularity); | |
e9637415 | 736 | t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) % |
8dd2cb7e | 737 | t->discard_granularity; |
9504e086 | 738 | } |
44abff2c CH |
739 | t->max_secure_erase_sectors = min_not_zero(t->max_secure_erase_sectors, |
740 | b->max_secure_erase_sectors); | |
a805a4fa DLM |
741 | t->zone_write_granularity = max(t->zone_write_granularity, |
742 | b->zone_write_granularity); | |
3093a479 | 743 | t->zoned = max(t->zoned, b->zoned); |
c8f6f88d DLM |
744 | if (!t->zoned) { |
745 | t->zone_write_granularity = 0; | |
746 | t->max_zone_append_sectors = 0; | |
747 | } | |
fe0b393f | 748 | return ret; |
c72758f3 | 749 | } |
5d85d324 | 750 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 | 751 | |
c1373f1c CH |
752 | /** |
753 | * queue_limits_stack_bdev - adjust queue_limits for stacked devices | |
754 | * @t: the stacking driver limits (top device) | |
755 | * @bdev: the underlying block device (bottom) | |
756 | * @offset: offset to beginning of data within component device | |
757 | * @pfx: prefix to use for warnings logged | |
758 | * | |
759 | * Description: | |
760 | * This function is used by stacking drivers like MD and DM to ensure | |
761 | * that all component devices have compatible block sizes and | |
762 | * alignments. The stacking driver must provide a queue_limits | |
763 | * struct (top) and then iteratively call the stacking function for | |
764 | * all component (bottom) devices. The stacking function will | |
765 | * attempt to combine the values and ensure proper alignment. | |
766 | */ | |
767 | void queue_limits_stack_bdev(struct queue_limits *t, struct block_device *bdev, | |
768 | sector_t offset, const char *pfx) | |
769 | { | |
770 | if (blk_stack_limits(t, &bdev_get_queue(bdev)->limits, | |
771 | get_start_sect(bdev) + offset)) | |
772 | pr_notice("%s: Warning: Device %pg is misaligned\n", | |
773 | pfx, bdev); | |
774 | } | |
775 | EXPORT_SYMBOL_GPL(queue_limits_stack_bdev); | |
776 | ||
27f8221a FT |
777 | /** |
778 | * blk_queue_update_dma_pad - update pad mask | |
779 | * @q: the request queue for the device | |
780 | * @mask: pad mask | |
781 | * | |
782 | * Update dma pad mask. | |
783 | * | |
784 | * Appending pad buffer to a request modifies the last entry of a | |
785 | * scatter list such that it includes the pad buffer. | |
786 | **/ | |
787 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
788 | { | |
789 | if (mask > q->dma_pad_mask) | |
790 | q->dma_pad_mask = mask; | |
791 | } | |
792 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
793 | ||
d278d4a8 JA |
794 | /** |
795 | * blk_set_queue_depth - tell the block layer about the device queue depth | |
796 | * @q: the request queue for the device | |
797 | * @depth: queue depth | |
798 | * | |
799 | */ | |
800 | void blk_set_queue_depth(struct request_queue *q, unsigned int depth) | |
801 | { | |
802 | q->queue_depth = depth; | |
9677a3e0 | 803 | rq_qos_queue_depth_changed(q); |
d278d4a8 JA |
804 | } |
805 | EXPORT_SYMBOL(blk_set_queue_depth); | |
806 | ||
93e9d8e8 JA |
807 | /** |
808 | * blk_queue_write_cache - configure queue's write cache | |
809 | * @q: the request queue for the device | |
810 | * @wc: write back cache on or off | |
811 | * @fua: device supports FUA writes, if true | |
812 | * | |
813 | * Tell the block layer about the write cache of @q. | |
814 | */ | |
815 | void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua) | |
816 | { | |
43c9835b CH |
817 | if (wc) { |
818 | blk_queue_flag_set(QUEUE_FLAG_HW_WC, q); | |
57d74df9 | 819 | blk_queue_flag_set(QUEUE_FLAG_WC, q); |
43c9835b CH |
820 | } else { |
821 | blk_queue_flag_clear(QUEUE_FLAG_HW_WC, q); | |
57d74df9 | 822 | blk_queue_flag_clear(QUEUE_FLAG_WC, q); |
43c9835b | 823 | } |
c888a8f9 | 824 | if (fua) |
57d74df9 | 825 | blk_queue_flag_set(QUEUE_FLAG_FUA, q); |
c888a8f9 | 826 | else |
57d74df9 | 827 | blk_queue_flag_clear(QUEUE_FLAG_FUA, q); |
93e9d8e8 JA |
828 | } |
829 | EXPORT_SYMBOL_GPL(blk_queue_write_cache); | |
830 | ||
27ba3e8f | 831 | /** |
d73e93b4 CH |
832 | * disk_set_zoned - inidicate a zoned device |
833 | * @disk: gendisk to configure | |
27ba3e8f | 834 | */ |
d73e93b4 | 835 | void disk_set_zoned(struct gendisk *disk) |
27ba3e8f | 836 | { |
a805a4fa DLM |
837 | struct request_queue *q = disk->queue; |
838 | ||
d73e93b4 CH |
839 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_BLK_DEV_ZONED)); |
840 | ||
841 | /* | |
842 | * Set the zone write granularity to the device logical block | |
843 | * size by default. The driver can change this value if needed. | |
844 | */ | |
845 | q->limits.zoned = true; | |
846 | blk_queue_zone_write_granularity(q, queue_logical_block_size(q)); | |
27ba3e8f | 847 | } |
6b2bd274 | 848 | EXPORT_SYMBOL_GPL(disk_set_zoned); |
89098b07 CH |
849 | |
850 | int bdev_alignment_offset(struct block_device *bdev) | |
851 | { | |
852 | struct request_queue *q = bdev_get_queue(bdev); | |
853 | ||
854 | if (q->limits.misaligned) | |
855 | return -1; | |
856 | if (bdev_is_partition(bdev)) | |
857 | return queue_limit_alignment_offset(&q->limits, | |
858 | bdev->bd_start_sect); | |
859 | return q->limits.alignment_offset; | |
860 | } | |
861 | EXPORT_SYMBOL_GPL(bdev_alignment_offset); | |
5c4b4a5c CH |
862 | |
863 | unsigned int bdev_discard_alignment(struct block_device *bdev) | |
864 | { | |
865 | struct request_queue *q = bdev_get_queue(bdev); | |
866 | ||
867 | if (bdev_is_partition(bdev)) | |
868 | return queue_limit_discard_alignment(&q->limits, | |
869 | bdev->bd_start_sect); | |
870 | return q->limits.discard_alignment; | |
871 | } | |
872 | EXPORT_SYMBOL_GPL(bdev_discard_alignment); |