Commit | Line | Data |
---|---|---|
86db1e29 JA |
1 | /* |
2 | * Functions related to setting various queue properties from drivers | |
3 | */ | |
4 | #include <linux/kernel.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/init.h> | |
7 | #include <linux/bio.h> | |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | |
70dd5bf3 | 10 | #include <linux/gcd.h> |
2cda2728 | 11 | #include <linux/lcm.h> |
ad5ebd2f | 12 | #include <linux/jiffies.h> |
5a0e3ad6 | 13 | #include <linux/gfp.h> |
86db1e29 JA |
14 | |
15 | #include "blk.h" | |
16 | ||
6728cb0e | 17 | unsigned long blk_max_low_pfn; |
86db1e29 | 18 | EXPORT_SYMBOL(blk_max_low_pfn); |
6728cb0e JA |
19 | |
20 | unsigned long blk_max_pfn; | |
86db1e29 JA |
21 | |
22 | /** | |
23 | * blk_queue_prep_rq - set a prepare_request function for queue | |
24 | * @q: queue | |
25 | * @pfn: prepare_request function | |
26 | * | |
27 | * It's possible for a queue to register a prepare_request callback which | |
28 | * is invoked before the request is handed to the request_fn. The goal of | |
29 | * the function is to prepare a request for I/O, it can be used to build a | |
30 | * cdb from the request data for instance. | |
31 | * | |
32 | */ | |
33 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) | |
34 | { | |
35 | q->prep_rq_fn = pfn; | |
36 | } | |
86db1e29 JA |
37 | EXPORT_SYMBOL(blk_queue_prep_rq); |
38 | ||
28018c24 JB |
39 | /** |
40 | * blk_queue_unprep_rq - set an unprepare_request function for queue | |
41 | * @q: queue | |
42 | * @ufn: unprepare_request function | |
43 | * | |
44 | * It's possible for a queue to register an unprepare_request callback | |
45 | * which is invoked before the request is finally completed. The goal | |
46 | * of the function is to deallocate any data that was allocated in the | |
47 | * prepare_request callback. | |
48 | * | |
49 | */ | |
50 | void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn) | |
51 | { | |
52 | q->unprep_rq_fn = ufn; | |
53 | } | |
54 | EXPORT_SYMBOL(blk_queue_unprep_rq); | |
55 | ||
86db1e29 JA |
56 | /** |
57 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
58 | * @q: queue | |
59 | * @mbfn: merge_bvec_fn | |
60 | * | |
61 | * Usually queues have static limitations on the max sectors or segments that | |
62 | * we can put in a request. Stacking drivers may have some settings that | |
63 | * are dynamic, and thus we have to query the queue whether it is ok to | |
64 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
65 | * has such limitations, it needs to register a merge_bvec_fn to control | |
66 | * the size of bio's sent to it. Note that a block device *must* allow a | |
67 | * single page to be added to an empty bio. The block device driver may want | |
68 | * to use the bio_split() function to deal with these bio's. By default | |
69 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
70 | * honored. | |
71 | */ | |
72 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) | |
73 | { | |
74 | q->merge_bvec_fn = mbfn; | |
75 | } | |
86db1e29 JA |
76 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
77 | ||
78 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
79 | { | |
80 | q->softirq_done_fn = fn; | |
81 | } | |
86db1e29 JA |
82 | EXPORT_SYMBOL(blk_queue_softirq_done); |
83 | ||
242f9dcb JA |
84 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
85 | { | |
86 | q->rq_timeout = timeout; | |
87 | } | |
88 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); | |
89 | ||
90 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) | |
91 | { | |
92 | q->rq_timed_out_fn = fn; | |
93 | } | |
94 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); | |
95 | ||
ef9e3fac KU |
96 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) |
97 | { | |
98 | q->lld_busy_fn = fn; | |
99 | } | |
100 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); | |
101 | ||
e475bba2 MP |
102 | /** |
103 | * blk_set_default_limits - reset limits to default values | |
f740f5ca | 104 | * @lim: the queue_limits structure to reset |
e475bba2 MP |
105 | * |
106 | * Description: | |
b1bd055d | 107 | * Returns a queue_limit struct to its default state. |
e475bba2 MP |
108 | */ |
109 | void blk_set_default_limits(struct queue_limits *lim) | |
110 | { | |
8a78362c | 111 | lim->max_segments = BLK_MAX_SEGMENTS; |
13f05c8d | 112 | lim->max_integrity_segments = 0; |
e475bba2 | 113 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
eb28d31b | 114 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; |
b1bd055d | 115 | lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; |
762380ad | 116 | lim->chunk_sectors = 0; |
4363ac7c | 117 | lim->max_write_same_sectors = 0; |
86b37281 MP |
118 | lim->max_discard_sectors = 0; |
119 | lim->discard_granularity = 0; | |
120 | lim->discard_alignment = 0; | |
121 | lim->discard_misaligned = 0; | |
b1bd055d | 122 | lim->discard_zeroes_data = 0; |
e475bba2 | 123 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
3a02c8e8 | 124 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
e475bba2 MP |
125 | lim->alignment_offset = 0; |
126 | lim->io_opt = 0; | |
127 | lim->misaligned = 0; | |
e692cb66 | 128 | lim->cluster = 1; |
e475bba2 MP |
129 | } |
130 | EXPORT_SYMBOL(blk_set_default_limits); | |
131 | ||
b1bd055d MP |
132 | /** |
133 | * blk_set_stacking_limits - set default limits for stacking devices | |
134 | * @lim: the queue_limits structure to reset | |
135 | * | |
136 | * Description: | |
137 | * Returns a queue_limit struct to its default state. Should be used | |
138 | * by stacking drivers like DM that have no internal limits. | |
139 | */ | |
140 | void blk_set_stacking_limits(struct queue_limits *lim) | |
141 | { | |
142 | blk_set_default_limits(lim); | |
143 | ||
144 | /* Inherit limits from component devices */ | |
145 | lim->discard_zeroes_data = 1; | |
146 | lim->max_segments = USHRT_MAX; | |
147 | lim->max_hw_sectors = UINT_MAX; | |
d82ae52e | 148 | lim->max_segment_size = UINT_MAX; |
fe86cdce | 149 | lim->max_sectors = UINT_MAX; |
4363ac7c | 150 | lim->max_write_same_sectors = UINT_MAX; |
b1bd055d MP |
151 | } |
152 | EXPORT_SYMBOL(blk_set_stacking_limits); | |
153 | ||
86db1e29 JA |
154 | /** |
155 | * blk_queue_make_request - define an alternate make_request function for a device | |
156 | * @q: the request queue for the device to be affected | |
157 | * @mfn: the alternate make_request function | |
158 | * | |
159 | * Description: | |
160 | * The normal way for &struct bios to be passed to a device | |
161 | * driver is for them to be collected into requests on a request | |
162 | * queue, and then to allow the device driver to select requests | |
163 | * off that queue when it is ready. This works well for many block | |
164 | * devices. However some block devices (typically virtual devices | |
165 | * such as md or lvm) do not benefit from the processing on the | |
166 | * request queue, and are served best by having the requests passed | |
167 | * directly to them. This can be achieved by providing a function | |
168 | * to blk_queue_make_request(). | |
169 | * | |
170 | * Caveat: | |
171 | * The driver that does this *must* be able to deal appropriately | |
172 | * with buffers in "highmemory". This can be accomplished by either calling | |
173 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
174 | * blk_queue_bounce() to create a buffer in normal memory. | |
175 | **/ | |
6728cb0e | 176 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
86db1e29 JA |
177 | { |
178 | /* | |
179 | * set defaults | |
180 | */ | |
181 | q->nr_requests = BLKDEV_MAX_RQ; | |
0e435ac2 | 182 | |
86db1e29 | 183 | q->make_request_fn = mfn; |
86db1e29 JA |
184 | blk_queue_dma_alignment(q, 511); |
185 | blk_queue_congestion_threshold(q); | |
186 | q->nr_batching = BLK_BATCH_REQ; | |
187 | ||
e475bba2 MP |
188 | blk_set_default_limits(&q->limits); |
189 | ||
86db1e29 JA |
190 | /* |
191 | * by default assume old behaviour and bounce for any highmem page | |
192 | */ | |
193 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
194 | } | |
86db1e29 JA |
195 | EXPORT_SYMBOL(blk_queue_make_request); |
196 | ||
197 | /** | |
198 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
cd0aca2d | 199 | * @q: the request queue for the device |
9f7e45d8 | 200 | * @max_addr: the maximum address the device can handle |
86db1e29 JA |
201 | * |
202 | * Description: | |
203 | * Different hardware can have different requirements as to what pages | |
204 | * it can do I/O directly to. A low level driver can call | |
205 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
9f7e45d8 | 206 | * buffers for doing I/O to pages residing above @max_addr. |
86db1e29 | 207 | **/ |
9f7e45d8 | 208 | void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr) |
86db1e29 | 209 | { |
9f7e45d8 | 210 | unsigned long b_pfn = max_addr >> PAGE_SHIFT; |
86db1e29 JA |
211 | int dma = 0; |
212 | ||
213 | q->bounce_gfp = GFP_NOIO; | |
214 | #if BITS_PER_LONG == 64 | |
cd0aca2d TH |
215 | /* |
216 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
217 | * some IOMMUs can handle everything, but I don't know of a | |
218 | * way to test this here. | |
219 | */ | |
220 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
86db1e29 | 221 | dma = 1; |
efb012b3 | 222 | q->limits.bounce_pfn = max(max_low_pfn, b_pfn); |
86db1e29 | 223 | #else |
6728cb0e | 224 | if (b_pfn < blk_max_low_pfn) |
86db1e29 | 225 | dma = 1; |
c49825fa | 226 | q->limits.bounce_pfn = b_pfn; |
260a67a9 | 227 | #endif |
86db1e29 JA |
228 | if (dma) { |
229 | init_emergency_isa_pool(); | |
230 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
260a67a9 | 231 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
232 | } |
233 | } | |
86db1e29 JA |
234 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
235 | ||
236 | /** | |
72d4cd9f MS |
237 | * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request |
238 | * @limits: the queue limits | |
2800aac1 | 239 | * @max_hw_sectors: max hardware sectors in the usual 512b unit |
86db1e29 JA |
240 | * |
241 | * Description: | |
2800aac1 MP |
242 | * Enables a low level driver to set a hard upper limit, |
243 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by | |
244 | * the device driver based upon the combined capabilities of I/O | |
245 | * controller and storage device. | |
246 | * | |
247 | * max_sectors is a soft limit imposed by the block layer for | |
248 | * filesystem type requests. This value can be overridden on a | |
249 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. | |
250 | * The soft limit can not exceed max_hw_sectors. | |
86db1e29 | 251 | **/ |
72d4cd9f | 252 | void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors) |
86db1e29 | 253 | { |
2800aac1 MP |
254 | if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) { |
255 | max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
24c03d47 | 256 | printk(KERN_INFO "%s: set to minimum %d\n", |
2800aac1 | 257 | __func__, max_hw_sectors); |
86db1e29 JA |
258 | } |
259 | ||
72d4cd9f MS |
260 | limits->max_hw_sectors = max_hw_sectors; |
261 | limits->max_sectors = min_t(unsigned int, max_hw_sectors, | |
262 | BLK_DEF_MAX_SECTORS); | |
263 | } | |
264 | EXPORT_SYMBOL(blk_limits_max_hw_sectors); | |
265 | ||
266 | /** | |
267 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue | |
268 | * @q: the request queue for the device | |
269 | * @max_hw_sectors: max hardware sectors in the usual 512b unit | |
270 | * | |
271 | * Description: | |
272 | * See description for blk_limits_max_hw_sectors(). | |
273 | **/ | |
274 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) | |
275 | { | |
276 | blk_limits_max_hw_sectors(&q->limits, max_hw_sectors); | |
86db1e29 | 277 | } |
086fa5ff | 278 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
86db1e29 | 279 | |
762380ad JA |
280 | /** |
281 | * blk_queue_chunk_sectors - set size of the chunk for this queue | |
282 | * @q: the request queue for the device | |
283 | * @chunk_sectors: chunk sectors in the usual 512b unit | |
284 | * | |
285 | * Description: | |
286 | * If a driver doesn't want IOs to cross a given chunk size, it can set | |
287 | * this limit and prevent merging across chunks. Note that the chunk size | |
58a4915a JA |
288 | * must currently be a power-of-2 in sectors. Also note that the block |
289 | * layer must accept a page worth of data at any offset. So if the | |
290 | * crossing of chunks is a hard limitation in the driver, it must still be | |
291 | * prepared to split single page bios. | |
762380ad JA |
292 | **/ |
293 | void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) | |
294 | { | |
295 | BUG_ON(!is_power_of_2(chunk_sectors)); | |
296 | q->limits.chunk_sectors = chunk_sectors; | |
297 | } | |
298 | EXPORT_SYMBOL(blk_queue_chunk_sectors); | |
299 | ||
67efc925 CH |
300 | /** |
301 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
302 | * @q: the request queue for the device | |
c7ebf065 | 303 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
304 | **/ |
305 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
306 | unsigned int max_discard_sectors) | |
307 | { | |
308 | q->limits.max_discard_sectors = max_discard_sectors; | |
309 | } | |
310 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
311 | ||
4363ac7c MP |
312 | /** |
313 | * blk_queue_max_write_same_sectors - set max sectors for a single write same | |
314 | * @q: the request queue for the device | |
315 | * @max_write_same_sectors: maximum number of sectors to write per command | |
316 | **/ | |
317 | void blk_queue_max_write_same_sectors(struct request_queue *q, | |
318 | unsigned int max_write_same_sectors) | |
319 | { | |
320 | q->limits.max_write_same_sectors = max_write_same_sectors; | |
321 | } | |
322 | EXPORT_SYMBOL(blk_queue_max_write_same_sectors); | |
323 | ||
86db1e29 | 324 | /** |
8a78362c | 325 | * blk_queue_max_segments - set max hw segments for a request for this queue |
86db1e29 JA |
326 | * @q: the request queue for the device |
327 | * @max_segments: max number of segments | |
328 | * | |
329 | * Description: | |
330 | * Enables a low level driver to set an upper limit on the number of | |
8a78362c | 331 | * hw data segments in a request. |
86db1e29 | 332 | **/ |
8a78362c | 333 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) |
86db1e29 JA |
334 | { |
335 | if (!max_segments) { | |
336 | max_segments = 1; | |
24c03d47 HH |
337 | printk(KERN_INFO "%s: set to minimum %d\n", |
338 | __func__, max_segments); | |
86db1e29 JA |
339 | } |
340 | ||
8a78362c | 341 | q->limits.max_segments = max_segments; |
86db1e29 | 342 | } |
8a78362c | 343 | EXPORT_SYMBOL(blk_queue_max_segments); |
86db1e29 JA |
344 | |
345 | /** | |
346 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
347 | * @q: the request queue for the device | |
348 | * @max_size: max size of segment in bytes | |
349 | * | |
350 | * Description: | |
351 | * Enables a low level driver to set an upper limit on the size of a | |
352 | * coalesced segment | |
353 | **/ | |
354 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
355 | { | |
356 | if (max_size < PAGE_CACHE_SIZE) { | |
357 | max_size = PAGE_CACHE_SIZE; | |
24c03d47 HH |
358 | printk(KERN_INFO "%s: set to minimum %d\n", |
359 | __func__, max_size); | |
86db1e29 JA |
360 | } |
361 | ||
025146e1 | 362 | q->limits.max_segment_size = max_size; |
86db1e29 | 363 | } |
86db1e29 JA |
364 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
365 | ||
366 | /** | |
e1defc4f | 367 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 368 | * @q: the request queue for the device |
e1defc4f | 369 | * @size: the logical block size, in bytes |
86db1e29 JA |
370 | * |
371 | * Description: | |
e1defc4f MP |
372 | * This should be set to the lowest possible block size that the |
373 | * storage device can address. The default of 512 covers most | |
374 | * hardware. | |
86db1e29 | 375 | **/ |
e1defc4f | 376 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
86db1e29 | 377 | { |
025146e1 | 378 | q->limits.logical_block_size = size; |
c72758f3 MP |
379 | |
380 | if (q->limits.physical_block_size < size) | |
381 | q->limits.physical_block_size = size; | |
382 | ||
383 | if (q->limits.io_min < q->limits.physical_block_size) | |
384 | q->limits.io_min = q->limits.physical_block_size; | |
86db1e29 | 385 | } |
e1defc4f | 386 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 387 | |
c72758f3 MP |
388 | /** |
389 | * blk_queue_physical_block_size - set physical block size for the queue | |
390 | * @q: the request queue for the device | |
391 | * @size: the physical block size, in bytes | |
392 | * | |
393 | * Description: | |
394 | * This should be set to the lowest possible sector size that the | |
395 | * hardware can operate on without reverting to read-modify-write | |
396 | * operations. | |
397 | */ | |
892b6f90 | 398 | void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) |
c72758f3 MP |
399 | { |
400 | q->limits.physical_block_size = size; | |
401 | ||
402 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
403 | q->limits.physical_block_size = q->limits.logical_block_size; | |
404 | ||
405 | if (q->limits.io_min < q->limits.physical_block_size) | |
406 | q->limits.io_min = q->limits.physical_block_size; | |
407 | } | |
408 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
409 | ||
410 | /** | |
411 | * blk_queue_alignment_offset - set physical block alignment offset | |
412 | * @q: the request queue for the device | |
8ebf9756 | 413 | * @offset: alignment offset in bytes |
c72758f3 MP |
414 | * |
415 | * Description: | |
416 | * Some devices are naturally misaligned to compensate for things like | |
417 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
418 | * should call this function for devices whose first sector is not | |
419 | * naturally aligned. | |
420 | */ | |
421 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
422 | { | |
423 | q->limits.alignment_offset = | |
424 | offset & (q->limits.physical_block_size - 1); | |
425 | q->limits.misaligned = 0; | |
426 | } | |
427 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
428 | ||
7c958e32 MP |
429 | /** |
430 | * blk_limits_io_min - set minimum request size for a device | |
431 | * @limits: the queue limits | |
432 | * @min: smallest I/O size in bytes | |
433 | * | |
434 | * Description: | |
435 | * Some devices have an internal block size bigger than the reported | |
436 | * hardware sector size. This function can be used to signal the | |
437 | * smallest I/O the device can perform without incurring a performance | |
438 | * penalty. | |
439 | */ | |
440 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
441 | { | |
442 | limits->io_min = min; | |
443 | ||
444 | if (limits->io_min < limits->logical_block_size) | |
445 | limits->io_min = limits->logical_block_size; | |
446 | ||
447 | if (limits->io_min < limits->physical_block_size) | |
448 | limits->io_min = limits->physical_block_size; | |
449 | } | |
450 | EXPORT_SYMBOL(blk_limits_io_min); | |
451 | ||
c72758f3 MP |
452 | /** |
453 | * blk_queue_io_min - set minimum request size for the queue | |
454 | * @q: the request queue for the device | |
8ebf9756 | 455 | * @min: smallest I/O size in bytes |
c72758f3 MP |
456 | * |
457 | * Description: | |
7e5f5fb0 MP |
458 | * Storage devices may report a granularity or preferred minimum I/O |
459 | * size which is the smallest request the device can perform without | |
460 | * incurring a performance penalty. For disk drives this is often the | |
461 | * physical block size. For RAID arrays it is often the stripe chunk | |
462 | * size. A properly aligned multiple of minimum_io_size is the | |
463 | * preferred request size for workloads where a high number of I/O | |
464 | * operations is desired. | |
c72758f3 MP |
465 | */ |
466 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
467 | { | |
7c958e32 | 468 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
469 | } |
470 | EXPORT_SYMBOL(blk_queue_io_min); | |
471 | ||
3c5820c7 MP |
472 | /** |
473 | * blk_limits_io_opt - set optimal request size for a device | |
474 | * @limits: the queue limits | |
475 | * @opt: smallest I/O size in bytes | |
476 | * | |
477 | * Description: | |
478 | * Storage devices may report an optimal I/O size, which is the | |
479 | * device's preferred unit for sustained I/O. This is rarely reported | |
480 | * for disk drives. For RAID arrays it is usually the stripe width or | |
481 | * the internal track size. A properly aligned multiple of | |
482 | * optimal_io_size is the preferred request size for workloads where | |
483 | * sustained throughput is desired. | |
484 | */ | |
485 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
486 | { | |
487 | limits->io_opt = opt; | |
488 | } | |
489 | EXPORT_SYMBOL(blk_limits_io_opt); | |
490 | ||
c72758f3 MP |
491 | /** |
492 | * blk_queue_io_opt - set optimal request size for the queue | |
493 | * @q: the request queue for the device | |
8ebf9756 | 494 | * @opt: optimal request size in bytes |
c72758f3 MP |
495 | * |
496 | * Description: | |
7e5f5fb0 MP |
497 | * Storage devices may report an optimal I/O size, which is the |
498 | * device's preferred unit for sustained I/O. This is rarely reported | |
499 | * for disk drives. For RAID arrays it is usually the stripe width or | |
500 | * the internal track size. A properly aligned multiple of | |
501 | * optimal_io_size is the preferred request size for workloads where | |
502 | * sustained throughput is desired. | |
c72758f3 MP |
503 | */ |
504 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
505 | { | |
3c5820c7 | 506 | blk_limits_io_opt(&q->limits, opt); |
c72758f3 MP |
507 | } |
508 | EXPORT_SYMBOL(blk_queue_io_opt); | |
509 | ||
86db1e29 JA |
510 | /** |
511 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
512 | * @t: the stacking driver (top) | |
513 | * @b: the underlying device (bottom) | |
514 | **/ | |
515 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
516 | { | |
fef24667 | 517 | blk_stack_limits(&t->limits, &b->limits, 0); |
86db1e29 | 518 | } |
86db1e29 JA |
519 | EXPORT_SYMBOL(blk_queue_stack_limits); |
520 | ||
c72758f3 MP |
521 | /** |
522 | * blk_stack_limits - adjust queue_limits for stacked devices | |
81744ee4 MP |
523 | * @t: the stacking driver limits (top device) |
524 | * @b: the underlying queue limits (bottom, component device) | |
e03a72e1 | 525 | * @start: first data sector within component device |
c72758f3 MP |
526 | * |
527 | * Description: | |
81744ee4 MP |
528 | * This function is used by stacking drivers like MD and DM to ensure |
529 | * that all component devices have compatible block sizes and | |
530 | * alignments. The stacking driver must provide a queue_limits | |
531 | * struct (top) and then iteratively call the stacking function for | |
532 | * all component (bottom) devices. The stacking function will | |
533 | * attempt to combine the values and ensure proper alignment. | |
534 | * | |
535 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
536 | * top device's block sizes and alignment offsets may be adjusted to | |
537 | * ensure alignment with the bottom device. If no compatible sizes | |
538 | * and alignments exist, -1 is returned and the resulting top | |
539 | * queue_limits will have the misaligned flag set to indicate that | |
540 | * the alignment_offset is undefined. | |
c72758f3 MP |
541 | */ |
542 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
e03a72e1 | 543 | sector_t start) |
c72758f3 | 544 | { |
e03a72e1 | 545 | unsigned int top, bottom, alignment, ret = 0; |
86b37281 | 546 | |
c72758f3 MP |
547 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
548 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
4363ac7c MP |
549 | t->max_write_same_sectors = min(t->max_write_same_sectors, |
550 | b->max_write_same_sectors); | |
77634f33 | 551 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
c72758f3 MP |
552 | |
553 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
554 | b->seg_boundary_mask); | |
555 | ||
8a78362c | 556 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
13f05c8d MP |
557 | t->max_integrity_segments = min_not_zero(t->max_integrity_segments, |
558 | b->max_integrity_segments); | |
c72758f3 MP |
559 | |
560 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
561 | b->max_segment_size); | |
562 | ||
fe0b393f MP |
563 | t->misaligned |= b->misaligned; |
564 | ||
e03a72e1 | 565 | alignment = queue_limit_alignment_offset(b, start); |
9504e086 | 566 | |
81744ee4 MP |
567 | /* Bottom device has different alignment. Check that it is |
568 | * compatible with the current top alignment. | |
569 | */ | |
9504e086 MP |
570 | if (t->alignment_offset != alignment) { |
571 | ||
572 | top = max(t->physical_block_size, t->io_min) | |
573 | + t->alignment_offset; | |
81744ee4 | 574 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
9504e086 | 575 | |
81744ee4 | 576 | /* Verify that top and bottom intervals line up */ |
fe0b393f | 577 | if (max(top, bottom) & (min(top, bottom) - 1)) { |
9504e086 | 578 | t->misaligned = 1; |
fe0b393f MP |
579 | ret = -1; |
580 | } | |
9504e086 MP |
581 | } |
582 | ||
c72758f3 MP |
583 | t->logical_block_size = max(t->logical_block_size, |
584 | b->logical_block_size); | |
585 | ||
586 | t->physical_block_size = max(t->physical_block_size, | |
587 | b->physical_block_size); | |
588 | ||
589 | t->io_min = max(t->io_min, b->io_min); | |
9504e086 MP |
590 | t->io_opt = lcm(t->io_opt, b->io_opt); |
591 | ||
e692cb66 | 592 | t->cluster &= b->cluster; |
98262f27 | 593 | t->discard_zeroes_data &= b->discard_zeroes_data; |
c72758f3 | 594 | |
81744ee4 | 595 | /* Physical block size a multiple of the logical block size? */ |
9504e086 MP |
596 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
597 | t->physical_block_size = t->logical_block_size; | |
c72758f3 | 598 | t->misaligned = 1; |
fe0b393f | 599 | ret = -1; |
86b37281 MP |
600 | } |
601 | ||
81744ee4 | 602 | /* Minimum I/O a multiple of the physical block size? */ |
9504e086 MP |
603 | if (t->io_min & (t->physical_block_size - 1)) { |
604 | t->io_min = t->physical_block_size; | |
605 | t->misaligned = 1; | |
fe0b393f | 606 | ret = -1; |
c72758f3 MP |
607 | } |
608 | ||
81744ee4 | 609 | /* Optimal I/O a multiple of the physical block size? */ |
9504e086 MP |
610 | if (t->io_opt & (t->physical_block_size - 1)) { |
611 | t->io_opt = 0; | |
612 | t->misaligned = 1; | |
fe0b393f | 613 | ret = -1; |
9504e086 | 614 | } |
c72758f3 | 615 | |
c78afc62 KO |
616 | t->raid_partial_stripes_expensive = |
617 | max(t->raid_partial_stripes_expensive, | |
618 | b->raid_partial_stripes_expensive); | |
619 | ||
81744ee4 | 620 | /* Find lowest common alignment_offset */ |
9504e086 MP |
621 | t->alignment_offset = lcm(t->alignment_offset, alignment) |
622 | & (max(t->physical_block_size, t->io_min) - 1); | |
86b37281 | 623 | |
81744ee4 | 624 | /* Verify that new alignment_offset is on a logical block boundary */ |
fe0b393f | 625 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
c72758f3 | 626 | t->misaligned = 1; |
fe0b393f MP |
627 | ret = -1; |
628 | } | |
c72758f3 | 629 | |
9504e086 MP |
630 | /* Discard alignment and granularity */ |
631 | if (b->discard_granularity) { | |
e03a72e1 | 632 | alignment = queue_limit_discard_alignment(b, start); |
9504e086 MP |
633 | |
634 | if (t->discard_granularity != 0 && | |
635 | t->discard_alignment != alignment) { | |
636 | top = t->discard_granularity + t->discard_alignment; | |
637 | bottom = b->discard_granularity + alignment; | |
70dd5bf3 | 638 | |
9504e086 | 639 | /* Verify that top and bottom intervals line up */ |
8dd2cb7e | 640 | if ((max(top, bottom) % min(top, bottom)) != 0) |
9504e086 MP |
641 | t->discard_misaligned = 1; |
642 | } | |
643 | ||
81744ee4 MP |
644 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
645 | b->max_discard_sectors); | |
9504e086 MP |
646 | t->discard_granularity = max(t->discard_granularity, |
647 | b->discard_granularity); | |
8dd2cb7e SL |
648 | t->discard_alignment = lcm(t->discard_alignment, alignment) % |
649 | t->discard_granularity; | |
9504e086 | 650 | } |
70dd5bf3 | 651 | |
fe0b393f | 652 | return ret; |
c72758f3 | 653 | } |
5d85d324 | 654 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 | 655 | |
17be8c24 MP |
656 | /** |
657 | * bdev_stack_limits - adjust queue limits for stacked drivers | |
658 | * @t: the stacking driver limits (top device) | |
659 | * @bdev: the component block_device (bottom) | |
660 | * @start: first data sector within component device | |
661 | * | |
662 | * Description: | |
663 | * Merges queue limits for a top device and a block_device. Returns | |
664 | * 0 if alignment didn't change. Returns -1 if adding the bottom | |
665 | * device caused misalignment. | |
666 | */ | |
667 | int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, | |
668 | sector_t start) | |
669 | { | |
670 | struct request_queue *bq = bdev_get_queue(bdev); | |
671 | ||
672 | start += get_start_sect(bdev); | |
673 | ||
e03a72e1 | 674 | return blk_stack_limits(t, &bq->limits, start); |
17be8c24 MP |
675 | } |
676 | EXPORT_SYMBOL(bdev_stack_limits); | |
677 | ||
c72758f3 MP |
678 | /** |
679 | * disk_stack_limits - adjust queue limits for stacked drivers | |
77634f33 | 680 | * @disk: MD/DM gendisk (top) |
c72758f3 MP |
681 | * @bdev: the underlying block device (bottom) |
682 | * @offset: offset to beginning of data within component device | |
683 | * | |
684 | * Description: | |
e03a72e1 MP |
685 | * Merges the limits for a top level gendisk and a bottom level |
686 | * block_device. | |
c72758f3 MP |
687 | */ |
688 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
689 | sector_t offset) | |
690 | { | |
691 | struct request_queue *t = disk->queue; | |
c72758f3 | 692 | |
e03a72e1 | 693 | if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { |
c72758f3 MP |
694 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; |
695 | ||
696 | disk_name(disk, 0, top); | |
697 | bdevname(bdev, bottom); | |
698 | ||
699 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
700 | top, bottom); | |
701 | } | |
c72758f3 MP |
702 | } |
703 | EXPORT_SYMBOL(disk_stack_limits); | |
704 | ||
e3790c7d TH |
705 | /** |
706 | * blk_queue_dma_pad - set pad mask | |
707 | * @q: the request queue for the device | |
708 | * @mask: pad mask | |
709 | * | |
27f8221a | 710 | * Set dma pad mask. |
e3790c7d | 711 | * |
27f8221a FT |
712 | * Appending pad buffer to a request modifies the last entry of a |
713 | * scatter list such that it includes the pad buffer. | |
e3790c7d TH |
714 | **/ |
715 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
716 | { | |
717 | q->dma_pad_mask = mask; | |
718 | } | |
719 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
720 | ||
27f8221a FT |
721 | /** |
722 | * blk_queue_update_dma_pad - update pad mask | |
723 | * @q: the request queue for the device | |
724 | * @mask: pad mask | |
725 | * | |
726 | * Update dma pad mask. | |
727 | * | |
728 | * Appending pad buffer to a request modifies the last entry of a | |
729 | * scatter list such that it includes the pad buffer. | |
730 | **/ | |
731 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
732 | { | |
733 | if (mask > q->dma_pad_mask) | |
734 | q->dma_pad_mask = mask; | |
735 | } | |
736 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
737 | ||
86db1e29 JA |
738 | /** |
739 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
86db1e29 | 740 | * @q: the request queue for the device |
2fb98e84 | 741 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
86db1e29 JA |
742 | * @buf: physically contiguous buffer |
743 | * @size: size of the buffer in bytes | |
744 | * | |
745 | * Some devices have excess DMA problems and can't simply discard (or | |
746 | * zero fill) the unwanted piece of the transfer. They have to have a | |
747 | * real area of memory to transfer it into. The use case for this is | |
748 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
749 | * bigger than the transfer size some HBAs will lock up if there | |
750 | * aren't DMA elements to contain the excess transfer. What this API | |
751 | * does is adjust the queue so that the buf is always appended | |
752 | * silently to the scatterlist. | |
753 | * | |
8a78362c MP |
754 | * Note: This routine adjusts max_hw_segments to make room for appending |
755 | * the drain buffer. If you call blk_queue_max_segments() after calling | |
756 | * this routine, you must set the limit to one fewer than your device | |
757 | * can support otherwise there won't be room for the drain buffer. | |
86db1e29 | 758 | */ |
448da4d2 | 759 | int blk_queue_dma_drain(struct request_queue *q, |
2fb98e84 TH |
760 | dma_drain_needed_fn *dma_drain_needed, |
761 | void *buf, unsigned int size) | |
86db1e29 | 762 | { |
8a78362c | 763 | if (queue_max_segments(q) < 2) |
86db1e29 JA |
764 | return -EINVAL; |
765 | /* make room for appending the drain */ | |
8a78362c | 766 | blk_queue_max_segments(q, queue_max_segments(q) - 1); |
2fb98e84 | 767 | q->dma_drain_needed = dma_drain_needed; |
86db1e29 JA |
768 | q->dma_drain_buffer = buf; |
769 | q->dma_drain_size = size; | |
770 | ||
771 | return 0; | |
772 | } | |
86db1e29 JA |
773 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
774 | ||
775 | /** | |
776 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
777 | * @q: the request queue for the device | |
778 | * @mask: the memory boundary mask | |
779 | **/ | |
780 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
781 | { | |
782 | if (mask < PAGE_CACHE_SIZE - 1) { | |
783 | mask = PAGE_CACHE_SIZE - 1; | |
24c03d47 HH |
784 | printk(KERN_INFO "%s: set to minimum %lx\n", |
785 | __func__, mask); | |
86db1e29 JA |
786 | } |
787 | ||
025146e1 | 788 | q->limits.seg_boundary_mask = mask; |
86db1e29 | 789 | } |
86db1e29 JA |
790 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
791 | ||
792 | /** | |
793 | * blk_queue_dma_alignment - set dma length and memory alignment | |
794 | * @q: the request queue for the device | |
795 | * @mask: alignment mask | |
796 | * | |
797 | * description: | |
710027a4 | 798 | * set required memory and length alignment for direct dma transactions. |
8feb4d20 | 799 | * this is used when building direct io requests for the queue. |
86db1e29 JA |
800 | * |
801 | **/ | |
802 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
803 | { | |
804 | q->dma_alignment = mask; | |
805 | } | |
86db1e29 JA |
806 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
807 | ||
808 | /** | |
809 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
810 | * @q: the request queue for the device | |
811 | * @mask: alignment mask | |
812 | * | |
813 | * description: | |
710027a4 | 814 | * update required memory and length alignment for direct dma transactions. |
86db1e29 JA |
815 | * If the requested alignment is larger than the current alignment, then |
816 | * the current queue alignment is updated to the new value, otherwise it | |
817 | * is left alone. The design of this is to allow multiple objects | |
818 | * (driver, device, transport etc) to set their respective | |
819 | * alignments without having them interfere. | |
820 | * | |
821 | **/ | |
822 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
823 | { | |
824 | BUG_ON(mask > PAGE_SIZE); | |
825 | ||
826 | if (mask > q->dma_alignment) | |
827 | q->dma_alignment = mask; | |
828 | } | |
86db1e29 JA |
829 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
830 | ||
4913efe4 TH |
831 | /** |
832 | * blk_queue_flush - configure queue's cache flush capability | |
833 | * @q: the request queue for the device | |
834 | * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA | |
835 | * | |
836 | * Tell block layer cache flush capability of @q. If it supports | |
837 | * flushing, REQ_FLUSH should be set. If it supports bypassing | |
838 | * write cache for individual writes, REQ_FUA should be set. | |
839 | */ | |
840 | void blk_queue_flush(struct request_queue *q, unsigned int flush) | |
841 | { | |
842 | WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA)); | |
843 | ||
844 | if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA))) | |
845 | flush &= ~REQ_FUA; | |
846 | ||
847 | q->flush_flags = flush & (REQ_FLUSH | REQ_FUA); | |
848 | } | |
849 | EXPORT_SYMBOL_GPL(blk_queue_flush); | |
850 | ||
f3876930 | 851 | void blk_queue_flush_queueable(struct request_queue *q, bool queueable) |
852 | { | |
853 | q->flush_not_queueable = !queueable; | |
854 | } | |
855 | EXPORT_SYMBOL_GPL(blk_queue_flush_queueable); | |
856 | ||
aeb3d3a8 | 857 | static int __init blk_settings_init(void) |
86db1e29 JA |
858 | { |
859 | blk_max_low_pfn = max_low_pfn - 1; | |
860 | blk_max_pfn = max_pfn - 1; | |
861 | return 0; | |
862 | } | |
863 | subsys_initcall(blk_settings_init); |