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: | |
107 | * Returns a queue_limit struct to its default state. Can be used by | |
108 | * stacking drivers like DM that stage table swaps and reuse an | |
109 | * existing device queue. | |
110 | */ | |
111 | void blk_set_default_limits(struct queue_limits *lim) | |
112 | { | |
8a78362c | 113 | lim->max_segments = BLK_MAX_SEGMENTS; |
13f05c8d | 114 | lim->max_integrity_segments = 0; |
e475bba2 | 115 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
eb28d31b | 116 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; |
5dee2477 MP |
117 | lim->max_sectors = BLK_DEF_MAX_SECTORS; |
118 | lim->max_hw_sectors = INT_MAX; | |
86b37281 MP |
119 | lim->max_discard_sectors = 0; |
120 | lim->discard_granularity = 0; | |
121 | lim->discard_alignment = 0; | |
122 | lim->discard_misaligned = 0; | |
98262f27 | 123 | lim->discard_zeroes_data = -1; |
e475bba2 | 124 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
3a02c8e8 | 125 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
e475bba2 MP |
126 | lim->alignment_offset = 0; |
127 | lim->io_opt = 0; | |
128 | lim->misaligned = 0; | |
e692cb66 | 129 | lim->cluster = 1; |
e475bba2 MP |
130 | } |
131 | EXPORT_SYMBOL(blk_set_default_limits); | |
132 | ||
86db1e29 JA |
133 | /** |
134 | * blk_queue_make_request - define an alternate make_request function for a device | |
135 | * @q: the request queue for the device to be affected | |
136 | * @mfn: the alternate make_request function | |
137 | * | |
138 | * Description: | |
139 | * The normal way for &struct bios to be passed to a device | |
140 | * driver is for them to be collected into requests on a request | |
141 | * queue, and then to allow the device driver to select requests | |
142 | * off that queue when it is ready. This works well for many block | |
143 | * devices. However some block devices (typically virtual devices | |
144 | * such as md or lvm) do not benefit from the processing on the | |
145 | * request queue, and are served best by having the requests passed | |
146 | * directly to them. This can be achieved by providing a function | |
147 | * to blk_queue_make_request(). | |
148 | * | |
149 | * Caveat: | |
150 | * The driver that does this *must* be able to deal appropriately | |
151 | * with buffers in "highmemory". This can be accomplished by either calling | |
152 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
153 | * blk_queue_bounce() to create a buffer in normal memory. | |
154 | **/ | |
6728cb0e | 155 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
86db1e29 JA |
156 | { |
157 | /* | |
158 | * set defaults | |
159 | */ | |
160 | q->nr_requests = BLKDEV_MAX_RQ; | |
0e435ac2 | 161 | |
86db1e29 | 162 | q->make_request_fn = mfn; |
86db1e29 JA |
163 | blk_queue_dma_alignment(q, 511); |
164 | blk_queue_congestion_threshold(q); | |
165 | q->nr_batching = BLK_BATCH_REQ; | |
166 | ||
167 | q->unplug_thresh = 4; /* hmm */ | |
ad5ebd2f | 168 | q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */ |
86db1e29 JA |
169 | if (q->unplug_delay == 0) |
170 | q->unplug_delay = 1; | |
171 | ||
86db1e29 JA |
172 | q->unplug_timer.function = blk_unplug_timeout; |
173 | q->unplug_timer.data = (unsigned long)q; | |
174 | ||
e475bba2 | 175 | blk_set_default_limits(&q->limits); |
086fa5ff | 176 | blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS); |
e475bba2 | 177 | |
a4e7d464 JA |
178 | /* |
179 | * If the caller didn't supply a lock, fall back to our embedded | |
180 | * per-queue locks | |
181 | */ | |
182 | if (!q->queue_lock) | |
183 | q->queue_lock = &q->__queue_lock; | |
184 | ||
86db1e29 JA |
185 | /* |
186 | * by default assume old behaviour and bounce for any highmem page | |
187 | */ | |
188 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
189 | } | |
86db1e29 JA |
190 | EXPORT_SYMBOL(blk_queue_make_request); |
191 | ||
192 | /** | |
193 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
cd0aca2d TH |
194 | * @q: the request queue for the device |
195 | * @dma_mask: the maximum address the device can handle | |
86db1e29 JA |
196 | * |
197 | * Description: | |
198 | * Different hardware can have different requirements as to what pages | |
199 | * it can do I/O directly to. A low level driver can call | |
200 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
cd0aca2d | 201 | * buffers for doing I/O to pages residing above @dma_mask. |
86db1e29 | 202 | **/ |
cd0aca2d | 203 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
86db1e29 | 204 | { |
cd0aca2d | 205 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
86db1e29 JA |
206 | int dma = 0; |
207 | ||
208 | q->bounce_gfp = GFP_NOIO; | |
209 | #if BITS_PER_LONG == 64 | |
cd0aca2d TH |
210 | /* |
211 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
212 | * some IOMMUs can handle everything, but I don't know of a | |
213 | * way to test this here. | |
214 | */ | |
215 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
86db1e29 | 216 | dma = 1; |
efb012b3 | 217 | q->limits.bounce_pfn = max(max_low_pfn, b_pfn); |
86db1e29 | 218 | #else |
6728cb0e | 219 | if (b_pfn < blk_max_low_pfn) |
86db1e29 | 220 | dma = 1; |
c49825fa | 221 | q->limits.bounce_pfn = b_pfn; |
260a67a9 | 222 | #endif |
86db1e29 JA |
223 | if (dma) { |
224 | init_emergency_isa_pool(); | |
225 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
260a67a9 | 226 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
227 | } |
228 | } | |
86db1e29 JA |
229 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
230 | ||
231 | /** | |
72d4cd9f MS |
232 | * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request |
233 | * @limits: the queue limits | |
2800aac1 | 234 | * @max_hw_sectors: max hardware sectors in the usual 512b unit |
86db1e29 JA |
235 | * |
236 | * Description: | |
2800aac1 MP |
237 | * Enables a low level driver to set a hard upper limit, |
238 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by | |
239 | * the device driver based upon the combined capabilities of I/O | |
240 | * controller and storage device. | |
241 | * | |
242 | * max_sectors is a soft limit imposed by the block layer for | |
243 | * filesystem type requests. This value can be overridden on a | |
244 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. | |
245 | * The soft limit can not exceed max_hw_sectors. | |
86db1e29 | 246 | **/ |
72d4cd9f | 247 | void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors) |
86db1e29 | 248 | { |
2800aac1 MP |
249 | if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) { |
250 | max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
24c03d47 | 251 | printk(KERN_INFO "%s: set to minimum %d\n", |
2800aac1 | 252 | __func__, max_hw_sectors); |
86db1e29 JA |
253 | } |
254 | ||
72d4cd9f MS |
255 | limits->max_hw_sectors = max_hw_sectors; |
256 | limits->max_sectors = min_t(unsigned int, max_hw_sectors, | |
257 | BLK_DEF_MAX_SECTORS); | |
258 | } | |
259 | EXPORT_SYMBOL(blk_limits_max_hw_sectors); | |
260 | ||
261 | /** | |
262 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue | |
263 | * @q: the request queue for the device | |
264 | * @max_hw_sectors: max hardware sectors in the usual 512b unit | |
265 | * | |
266 | * Description: | |
267 | * See description for blk_limits_max_hw_sectors(). | |
268 | **/ | |
269 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) | |
270 | { | |
271 | blk_limits_max_hw_sectors(&q->limits, max_hw_sectors); | |
86db1e29 | 272 | } |
086fa5ff | 273 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
86db1e29 | 274 | |
67efc925 CH |
275 | /** |
276 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
277 | * @q: the request queue for the device | |
c7ebf065 | 278 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
279 | **/ |
280 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
281 | unsigned int max_discard_sectors) | |
282 | { | |
283 | q->limits.max_discard_sectors = max_discard_sectors; | |
284 | } | |
285 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
286 | ||
86db1e29 | 287 | /** |
8a78362c | 288 | * blk_queue_max_segments - set max hw segments for a request for this queue |
86db1e29 JA |
289 | * @q: the request queue for the device |
290 | * @max_segments: max number of segments | |
291 | * | |
292 | * Description: | |
293 | * Enables a low level driver to set an upper limit on the number of | |
8a78362c | 294 | * hw data segments in a request. |
86db1e29 | 295 | **/ |
8a78362c | 296 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) |
86db1e29 JA |
297 | { |
298 | if (!max_segments) { | |
299 | max_segments = 1; | |
24c03d47 HH |
300 | printk(KERN_INFO "%s: set to minimum %d\n", |
301 | __func__, max_segments); | |
86db1e29 JA |
302 | } |
303 | ||
8a78362c | 304 | q->limits.max_segments = max_segments; |
86db1e29 | 305 | } |
8a78362c | 306 | EXPORT_SYMBOL(blk_queue_max_segments); |
86db1e29 JA |
307 | |
308 | /** | |
309 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
310 | * @q: the request queue for the device | |
311 | * @max_size: max size of segment in bytes | |
312 | * | |
313 | * Description: | |
314 | * Enables a low level driver to set an upper limit on the size of a | |
315 | * coalesced segment | |
316 | **/ | |
317 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
318 | { | |
319 | if (max_size < PAGE_CACHE_SIZE) { | |
320 | max_size = PAGE_CACHE_SIZE; | |
24c03d47 HH |
321 | printk(KERN_INFO "%s: set to minimum %d\n", |
322 | __func__, max_size); | |
86db1e29 JA |
323 | } |
324 | ||
025146e1 | 325 | q->limits.max_segment_size = max_size; |
86db1e29 | 326 | } |
86db1e29 JA |
327 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
328 | ||
329 | /** | |
e1defc4f | 330 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 331 | * @q: the request queue for the device |
e1defc4f | 332 | * @size: the logical block size, in bytes |
86db1e29 JA |
333 | * |
334 | * Description: | |
e1defc4f MP |
335 | * This should be set to the lowest possible block size that the |
336 | * storage device can address. The default of 512 covers most | |
337 | * hardware. | |
86db1e29 | 338 | **/ |
e1defc4f | 339 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
86db1e29 | 340 | { |
025146e1 | 341 | q->limits.logical_block_size = size; |
c72758f3 MP |
342 | |
343 | if (q->limits.physical_block_size < size) | |
344 | q->limits.physical_block_size = size; | |
345 | ||
346 | if (q->limits.io_min < q->limits.physical_block_size) | |
347 | q->limits.io_min = q->limits.physical_block_size; | |
86db1e29 | 348 | } |
e1defc4f | 349 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 350 | |
c72758f3 MP |
351 | /** |
352 | * blk_queue_physical_block_size - set physical block size for the queue | |
353 | * @q: the request queue for the device | |
354 | * @size: the physical block size, in bytes | |
355 | * | |
356 | * Description: | |
357 | * This should be set to the lowest possible sector size that the | |
358 | * hardware can operate on without reverting to read-modify-write | |
359 | * operations. | |
360 | */ | |
892b6f90 | 361 | void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) |
c72758f3 MP |
362 | { |
363 | q->limits.physical_block_size = size; | |
364 | ||
365 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
366 | q->limits.physical_block_size = q->limits.logical_block_size; | |
367 | ||
368 | if (q->limits.io_min < q->limits.physical_block_size) | |
369 | q->limits.io_min = q->limits.physical_block_size; | |
370 | } | |
371 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
372 | ||
373 | /** | |
374 | * blk_queue_alignment_offset - set physical block alignment offset | |
375 | * @q: the request queue for the device | |
8ebf9756 | 376 | * @offset: alignment offset in bytes |
c72758f3 MP |
377 | * |
378 | * Description: | |
379 | * Some devices are naturally misaligned to compensate for things like | |
380 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
381 | * should call this function for devices whose first sector is not | |
382 | * naturally aligned. | |
383 | */ | |
384 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
385 | { | |
386 | q->limits.alignment_offset = | |
387 | offset & (q->limits.physical_block_size - 1); | |
388 | q->limits.misaligned = 0; | |
389 | } | |
390 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
391 | ||
7c958e32 MP |
392 | /** |
393 | * blk_limits_io_min - set minimum request size for a device | |
394 | * @limits: the queue limits | |
395 | * @min: smallest I/O size in bytes | |
396 | * | |
397 | * Description: | |
398 | * Some devices have an internal block size bigger than the reported | |
399 | * hardware sector size. This function can be used to signal the | |
400 | * smallest I/O the device can perform without incurring a performance | |
401 | * penalty. | |
402 | */ | |
403 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
404 | { | |
405 | limits->io_min = min; | |
406 | ||
407 | if (limits->io_min < limits->logical_block_size) | |
408 | limits->io_min = limits->logical_block_size; | |
409 | ||
410 | if (limits->io_min < limits->physical_block_size) | |
411 | limits->io_min = limits->physical_block_size; | |
412 | } | |
413 | EXPORT_SYMBOL(blk_limits_io_min); | |
414 | ||
c72758f3 MP |
415 | /** |
416 | * blk_queue_io_min - set minimum request size for the queue | |
417 | * @q: the request queue for the device | |
8ebf9756 | 418 | * @min: smallest I/O size in bytes |
c72758f3 MP |
419 | * |
420 | * Description: | |
7e5f5fb0 MP |
421 | * Storage devices may report a granularity or preferred minimum I/O |
422 | * size which is the smallest request the device can perform without | |
423 | * incurring a performance penalty. For disk drives this is often the | |
424 | * physical block size. For RAID arrays it is often the stripe chunk | |
425 | * size. A properly aligned multiple of minimum_io_size is the | |
426 | * preferred request size for workloads where a high number of I/O | |
427 | * operations is desired. | |
c72758f3 MP |
428 | */ |
429 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
430 | { | |
7c958e32 | 431 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
432 | } |
433 | EXPORT_SYMBOL(blk_queue_io_min); | |
434 | ||
3c5820c7 MP |
435 | /** |
436 | * blk_limits_io_opt - set optimal request size for a device | |
437 | * @limits: the queue limits | |
438 | * @opt: smallest I/O size in bytes | |
439 | * | |
440 | * Description: | |
441 | * Storage devices may report an optimal I/O size, which is the | |
442 | * device's preferred unit for sustained I/O. This is rarely reported | |
443 | * for disk drives. For RAID arrays it is usually the stripe width or | |
444 | * the internal track size. A properly aligned multiple of | |
445 | * optimal_io_size is the preferred request size for workloads where | |
446 | * sustained throughput is desired. | |
447 | */ | |
448 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
449 | { | |
450 | limits->io_opt = opt; | |
451 | } | |
452 | EXPORT_SYMBOL(blk_limits_io_opt); | |
453 | ||
c72758f3 MP |
454 | /** |
455 | * blk_queue_io_opt - set optimal request size for the queue | |
456 | * @q: the request queue for the device | |
8ebf9756 | 457 | * @opt: optimal request size in bytes |
c72758f3 MP |
458 | * |
459 | * Description: | |
7e5f5fb0 MP |
460 | * Storage devices may report an optimal I/O size, which is the |
461 | * device's preferred unit for sustained I/O. This is rarely reported | |
462 | * for disk drives. For RAID arrays it is usually the stripe width or | |
463 | * the internal track size. A properly aligned multiple of | |
464 | * optimal_io_size is the preferred request size for workloads where | |
465 | * sustained throughput is desired. | |
c72758f3 MP |
466 | */ |
467 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
468 | { | |
3c5820c7 | 469 | blk_limits_io_opt(&q->limits, opt); |
c72758f3 MP |
470 | } |
471 | EXPORT_SYMBOL(blk_queue_io_opt); | |
472 | ||
86db1e29 JA |
473 | /** |
474 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
475 | * @t: the stacking driver (top) | |
476 | * @b: the underlying device (bottom) | |
477 | **/ | |
478 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
479 | { | |
fef24667 | 480 | blk_stack_limits(&t->limits, &b->limits, 0); |
86db1e29 | 481 | } |
86db1e29 JA |
482 | EXPORT_SYMBOL(blk_queue_stack_limits); |
483 | ||
c72758f3 MP |
484 | /** |
485 | * blk_stack_limits - adjust queue_limits for stacked devices | |
81744ee4 MP |
486 | * @t: the stacking driver limits (top device) |
487 | * @b: the underlying queue limits (bottom, component device) | |
e03a72e1 | 488 | * @start: first data sector within component device |
c72758f3 MP |
489 | * |
490 | * Description: | |
81744ee4 MP |
491 | * This function is used by stacking drivers like MD and DM to ensure |
492 | * that all component devices have compatible block sizes and | |
493 | * alignments. The stacking driver must provide a queue_limits | |
494 | * struct (top) and then iteratively call the stacking function for | |
495 | * all component (bottom) devices. The stacking function will | |
496 | * attempt to combine the values and ensure proper alignment. | |
497 | * | |
498 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
499 | * top device's block sizes and alignment offsets may be adjusted to | |
500 | * ensure alignment with the bottom device. If no compatible sizes | |
501 | * and alignments exist, -1 is returned and the resulting top | |
502 | * queue_limits will have the misaligned flag set to indicate that | |
503 | * the alignment_offset is undefined. | |
c72758f3 MP |
504 | */ |
505 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
e03a72e1 | 506 | sector_t start) |
c72758f3 | 507 | { |
e03a72e1 | 508 | unsigned int top, bottom, alignment, ret = 0; |
86b37281 | 509 | |
c72758f3 MP |
510 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
511 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
77634f33 | 512 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
c72758f3 MP |
513 | |
514 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
515 | b->seg_boundary_mask); | |
516 | ||
8a78362c | 517 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
13f05c8d MP |
518 | t->max_integrity_segments = min_not_zero(t->max_integrity_segments, |
519 | b->max_integrity_segments); | |
c72758f3 MP |
520 | |
521 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
522 | b->max_segment_size); | |
523 | ||
fe0b393f MP |
524 | t->misaligned |= b->misaligned; |
525 | ||
e03a72e1 | 526 | alignment = queue_limit_alignment_offset(b, start); |
9504e086 | 527 | |
81744ee4 MP |
528 | /* Bottom device has different alignment. Check that it is |
529 | * compatible with the current top alignment. | |
530 | */ | |
9504e086 MP |
531 | if (t->alignment_offset != alignment) { |
532 | ||
533 | top = max(t->physical_block_size, t->io_min) | |
534 | + t->alignment_offset; | |
81744ee4 | 535 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
9504e086 | 536 | |
81744ee4 | 537 | /* Verify that top and bottom intervals line up */ |
fe0b393f | 538 | if (max(top, bottom) & (min(top, bottom) - 1)) { |
9504e086 | 539 | t->misaligned = 1; |
fe0b393f MP |
540 | ret = -1; |
541 | } | |
9504e086 MP |
542 | } |
543 | ||
c72758f3 MP |
544 | t->logical_block_size = max(t->logical_block_size, |
545 | b->logical_block_size); | |
546 | ||
547 | t->physical_block_size = max(t->physical_block_size, | |
548 | b->physical_block_size); | |
549 | ||
550 | t->io_min = max(t->io_min, b->io_min); | |
9504e086 MP |
551 | t->io_opt = lcm(t->io_opt, b->io_opt); |
552 | ||
e692cb66 | 553 | t->cluster &= b->cluster; |
98262f27 | 554 | t->discard_zeroes_data &= b->discard_zeroes_data; |
c72758f3 | 555 | |
81744ee4 | 556 | /* Physical block size a multiple of the logical block size? */ |
9504e086 MP |
557 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
558 | t->physical_block_size = t->logical_block_size; | |
c72758f3 | 559 | t->misaligned = 1; |
fe0b393f | 560 | ret = -1; |
86b37281 MP |
561 | } |
562 | ||
81744ee4 | 563 | /* Minimum I/O a multiple of the physical block size? */ |
9504e086 MP |
564 | if (t->io_min & (t->physical_block_size - 1)) { |
565 | t->io_min = t->physical_block_size; | |
566 | t->misaligned = 1; | |
fe0b393f | 567 | ret = -1; |
c72758f3 MP |
568 | } |
569 | ||
81744ee4 | 570 | /* Optimal I/O a multiple of the physical block size? */ |
9504e086 MP |
571 | if (t->io_opt & (t->physical_block_size - 1)) { |
572 | t->io_opt = 0; | |
573 | t->misaligned = 1; | |
fe0b393f | 574 | ret = -1; |
9504e086 | 575 | } |
c72758f3 | 576 | |
81744ee4 | 577 | /* Find lowest common alignment_offset */ |
9504e086 MP |
578 | t->alignment_offset = lcm(t->alignment_offset, alignment) |
579 | & (max(t->physical_block_size, t->io_min) - 1); | |
86b37281 | 580 | |
81744ee4 | 581 | /* Verify that new alignment_offset is on a logical block boundary */ |
fe0b393f | 582 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
c72758f3 | 583 | t->misaligned = 1; |
fe0b393f MP |
584 | ret = -1; |
585 | } | |
c72758f3 | 586 | |
9504e086 MP |
587 | /* Discard alignment and granularity */ |
588 | if (b->discard_granularity) { | |
e03a72e1 | 589 | alignment = queue_limit_discard_alignment(b, start); |
9504e086 MP |
590 | |
591 | if (t->discard_granularity != 0 && | |
592 | t->discard_alignment != alignment) { | |
593 | top = t->discard_granularity + t->discard_alignment; | |
594 | bottom = b->discard_granularity + alignment; | |
70dd5bf3 | 595 | |
9504e086 MP |
596 | /* Verify that top and bottom intervals line up */ |
597 | if (max(top, bottom) & (min(top, bottom) - 1)) | |
598 | t->discard_misaligned = 1; | |
599 | } | |
600 | ||
81744ee4 MP |
601 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
602 | b->max_discard_sectors); | |
9504e086 MP |
603 | t->discard_granularity = max(t->discard_granularity, |
604 | b->discard_granularity); | |
605 | t->discard_alignment = lcm(t->discard_alignment, alignment) & | |
606 | (t->discard_granularity - 1); | |
607 | } | |
70dd5bf3 | 608 | |
fe0b393f | 609 | return ret; |
c72758f3 | 610 | } |
5d85d324 | 611 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 | 612 | |
17be8c24 MP |
613 | /** |
614 | * bdev_stack_limits - adjust queue limits for stacked drivers | |
615 | * @t: the stacking driver limits (top device) | |
616 | * @bdev: the component block_device (bottom) | |
617 | * @start: first data sector within component device | |
618 | * | |
619 | * Description: | |
620 | * Merges queue limits for a top device and a block_device. Returns | |
621 | * 0 if alignment didn't change. Returns -1 if adding the bottom | |
622 | * device caused misalignment. | |
623 | */ | |
624 | int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, | |
625 | sector_t start) | |
626 | { | |
627 | struct request_queue *bq = bdev_get_queue(bdev); | |
628 | ||
629 | start += get_start_sect(bdev); | |
630 | ||
e03a72e1 | 631 | return blk_stack_limits(t, &bq->limits, start); |
17be8c24 MP |
632 | } |
633 | EXPORT_SYMBOL(bdev_stack_limits); | |
634 | ||
c72758f3 MP |
635 | /** |
636 | * disk_stack_limits - adjust queue limits for stacked drivers | |
77634f33 | 637 | * @disk: MD/DM gendisk (top) |
c72758f3 MP |
638 | * @bdev: the underlying block device (bottom) |
639 | * @offset: offset to beginning of data within component device | |
640 | * | |
641 | * Description: | |
e03a72e1 MP |
642 | * Merges the limits for a top level gendisk and a bottom level |
643 | * block_device. | |
c72758f3 MP |
644 | */ |
645 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
646 | sector_t offset) | |
647 | { | |
648 | struct request_queue *t = disk->queue; | |
c72758f3 | 649 | |
e03a72e1 | 650 | if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { |
c72758f3 MP |
651 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; |
652 | ||
653 | disk_name(disk, 0, top); | |
654 | bdevname(bdev, bottom); | |
655 | ||
656 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
657 | top, bottom); | |
658 | } | |
c72758f3 MP |
659 | } |
660 | EXPORT_SYMBOL(disk_stack_limits); | |
661 | ||
e3790c7d TH |
662 | /** |
663 | * blk_queue_dma_pad - set pad mask | |
664 | * @q: the request queue for the device | |
665 | * @mask: pad mask | |
666 | * | |
27f8221a | 667 | * Set dma pad mask. |
e3790c7d | 668 | * |
27f8221a FT |
669 | * Appending pad buffer to a request modifies the last entry of a |
670 | * scatter list such that it includes the pad buffer. | |
e3790c7d TH |
671 | **/ |
672 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
673 | { | |
674 | q->dma_pad_mask = mask; | |
675 | } | |
676 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
677 | ||
27f8221a FT |
678 | /** |
679 | * blk_queue_update_dma_pad - update pad mask | |
680 | * @q: the request queue for the device | |
681 | * @mask: pad mask | |
682 | * | |
683 | * Update dma pad mask. | |
684 | * | |
685 | * Appending pad buffer to a request modifies the last entry of a | |
686 | * scatter list such that it includes the pad buffer. | |
687 | **/ | |
688 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
689 | { | |
690 | if (mask > q->dma_pad_mask) | |
691 | q->dma_pad_mask = mask; | |
692 | } | |
693 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
694 | ||
86db1e29 JA |
695 | /** |
696 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
86db1e29 | 697 | * @q: the request queue for the device |
2fb98e84 | 698 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
86db1e29 JA |
699 | * @buf: physically contiguous buffer |
700 | * @size: size of the buffer in bytes | |
701 | * | |
702 | * Some devices have excess DMA problems and can't simply discard (or | |
703 | * zero fill) the unwanted piece of the transfer. They have to have a | |
704 | * real area of memory to transfer it into. The use case for this is | |
705 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
706 | * bigger than the transfer size some HBAs will lock up if there | |
707 | * aren't DMA elements to contain the excess transfer. What this API | |
708 | * does is adjust the queue so that the buf is always appended | |
709 | * silently to the scatterlist. | |
710 | * | |
8a78362c MP |
711 | * Note: This routine adjusts max_hw_segments to make room for appending |
712 | * the drain buffer. If you call blk_queue_max_segments() after calling | |
713 | * this routine, you must set the limit to one fewer than your device | |
714 | * can support otherwise there won't be room for the drain buffer. | |
86db1e29 | 715 | */ |
448da4d2 | 716 | int blk_queue_dma_drain(struct request_queue *q, |
2fb98e84 TH |
717 | dma_drain_needed_fn *dma_drain_needed, |
718 | void *buf, unsigned int size) | |
86db1e29 | 719 | { |
8a78362c | 720 | if (queue_max_segments(q) < 2) |
86db1e29 JA |
721 | return -EINVAL; |
722 | /* make room for appending the drain */ | |
8a78362c | 723 | blk_queue_max_segments(q, queue_max_segments(q) - 1); |
2fb98e84 | 724 | q->dma_drain_needed = dma_drain_needed; |
86db1e29 JA |
725 | q->dma_drain_buffer = buf; |
726 | q->dma_drain_size = size; | |
727 | ||
728 | return 0; | |
729 | } | |
86db1e29 JA |
730 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
731 | ||
732 | /** | |
733 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
734 | * @q: the request queue for the device | |
735 | * @mask: the memory boundary mask | |
736 | **/ | |
737 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
738 | { | |
739 | if (mask < PAGE_CACHE_SIZE - 1) { | |
740 | mask = PAGE_CACHE_SIZE - 1; | |
24c03d47 HH |
741 | printk(KERN_INFO "%s: set to minimum %lx\n", |
742 | __func__, mask); | |
86db1e29 JA |
743 | } |
744 | ||
025146e1 | 745 | q->limits.seg_boundary_mask = mask; |
86db1e29 | 746 | } |
86db1e29 JA |
747 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
748 | ||
749 | /** | |
750 | * blk_queue_dma_alignment - set dma length and memory alignment | |
751 | * @q: the request queue for the device | |
752 | * @mask: alignment mask | |
753 | * | |
754 | * description: | |
710027a4 | 755 | * set required memory and length alignment for direct dma transactions. |
8feb4d20 | 756 | * this is used when building direct io requests for the queue. |
86db1e29 JA |
757 | * |
758 | **/ | |
759 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
760 | { | |
761 | q->dma_alignment = mask; | |
762 | } | |
86db1e29 JA |
763 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
764 | ||
765 | /** | |
766 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
767 | * @q: the request queue for the device | |
768 | * @mask: alignment mask | |
769 | * | |
770 | * description: | |
710027a4 | 771 | * update required memory and length alignment for direct dma transactions. |
86db1e29 JA |
772 | * If the requested alignment is larger than the current alignment, then |
773 | * the current queue alignment is updated to the new value, otherwise it | |
774 | * is left alone. The design of this is to allow multiple objects | |
775 | * (driver, device, transport etc) to set their respective | |
776 | * alignments without having them interfere. | |
777 | * | |
778 | **/ | |
779 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
780 | { | |
781 | BUG_ON(mask > PAGE_SIZE); | |
782 | ||
783 | if (mask > q->dma_alignment) | |
784 | q->dma_alignment = mask; | |
785 | } | |
86db1e29 JA |
786 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
787 | ||
4913efe4 TH |
788 | /** |
789 | * blk_queue_flush - configure queue's cache flush capability | |
790 | * @q: the request queue for the device | |
791 | * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA | |
792 | * | |
793 | * Tell block layer cache flush capability of @q. If it supports | |
794 | * flushing, REQ_FLUSH should be set. If it supports bypassing | |
795 | * write cache for individual writes, REQ_FUA should be set. | |
796 | */ | |
797 | void blk_queue_flush(struct request_queue *q, unsigned int flush) | |
798 | { | |
799 | WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA)); | |
800 | ||
801 | if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA))) | |
802 | flush &= ~REQ_FUA; | |
803 | ||
804 | q->flush_flags = flush & (REQ_FLUSH | REQ_FUA); | |
805 | } | |
806 | EXPORT_SYMBOL_GPL(blk_queue_flush); | |
807 | ||
aeb3d3a8 | 808 | static int __init blk_settings_init(void) |
86db1e29 JA |
809 | { |
810 | blk_max_low_pfn = max_low_pfn - 1; | |
811 | blk_max_pfn = max_pfn - 1; | |
812 | return 0; | |
813 | } | |
814 | subsys_initcall(blk_settings_init); |