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> |
86db1e29 JA |
11 | |
12 | #include "blk.h" | |
13 | ||
6728cb0e | 14 | unsigned long blk_max_low_pfn; |
86db1e29 | 15 | EXPORT_SYMBOL(blk_max_low_pfn); |
6728cb0e JA |
16 | |
17 | unsigned long blk_max_pfn; | |
86db1e29 JA |
18 | |
19 | /** | |
20 | * blk_queue_prep_rq - set a prepare_request function for queue | |
21 | * @q: queue | |
22 | * @pfn: prepare_request function | |
23 | * | |
24 | * It's possible for a queue to register a prepare_request callback which | |
25 | * is invoked before the request is handed to the request_fn. The goal of | |
26 | * the function is to prepare a request for I/O, it can be used to build a | |
27 | * cdb from the request data for instance. | |
28 | * | |
29 | */ | |
30 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) | |
31 | { | |
32 | q->prep_rq_fn = pfn; | |
33 | } | |
86db1e29 JA |
34 | EXPORT_SYMBOL(blk_queue_prep_rq); |
35 | ||
36 | /** | |
37 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
38 | * @q: queue | |
39 | * @mbfn: merge_bvec_fn | |
40 | * | |
41 | * Usually queues have static limitations on the max sectors or segments that | |
42 | * we can put in a request. Stacking drivers may have some settings that | |
43 | * are dynamic, and thus we have to query the queue whether it is ok to | |
44 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
45 | * has such limitations, it needs to register a merge_bvec_fn to control | |
46 | * the size of bio's sent to it. Note that a block device *must* allow a | |
47 | * single page to be added to an empty bio. The block device driver may want | |
48 | * to use the bio_split() function to deal with these bio's. By default | |
49 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
50 | * honored. | |
51 | */ | |
52 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) | |
53 | { | |
54 | q->merge_bvec_fn = mbfn; | |
55 | } | |
86db1e29 JA |
56 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
57 | ||
58 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
59 | { | |
60 | q->softirq_done_fn = fn; | |
61 | } | |
86db1e29 JA |
62 | EXPORT_SYMBOL(blk_queue_softirq_done); |
63 | ||
242f9dcb JA |
64 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
65 | { | |
66 | q->rq_timeout = timeout; | |
67 | } | |
68 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); | |
69 | ||
70 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) | |
71 | { | |
72 | q->rq_timed_out_fn = fn; | |
73 | } | |
74 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); | |
75 | ||
ef9e3fac KU |
76 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) |
77 | { | |
78 | q->lld_busy_fn = fn; | |
79 | } | |
80 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); | |
81 | ||
e475bba2 MP |
82 | /** |
83 | * blk_set_default_limits - reset limits to default values | |
f740f5ca | 84 | * @lim: the queue_limits structure to reset |
e475bba2 MP |
85 | * |
86 | * Description: | |
87 | * Returns a queue_limit struct to its default state. Can be used by | |
88 | * stacking drivers like DM that stage table swaps and reuse an | |
89 | * existing device queue. | |
90 | */ | |
91 | void blk_set_default_limits(struct queue_limits *lim) | |
92 | { | |
93 | lim->max_phys_segments = MAX_PHYS_SEGMENTS; | |
94 | lim->max_hw_segments = MAX_HW_SEGMENTS; | |
95 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; | |
96 | lim->max_segment_size = MAX_SEGMENT_SIZE; | |
5dee2477 MP |
97 | lim->max_sectors = BLK_DEF_MAX_SECTORS; |
98 | lim->max_hw_sectors = INT_MAX; | |
86b37281 MP |
99 | lim->max_discard_sectors = 0; |
100 | lim->discard_granularity = 0; | |
101 | lim->discard_alignment = 0; | |
102 | lim->discard_misaligned = 0; | |
e475bba2 | 103 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
3a02c8e8 | 104 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
e475bba2 MP |
105 | lim->alignment_offset = 0; |
106 | lim->io_opt = 0; | |
107 | lim->misaligned = 0; | |
108 | lim->no_cluster = 0; | |
109 | } | |
110 | EXPORT_SYMBOL(blk_set_default_limits); | |
111 | ||
86db1e29 JA |
112 | /** |
113 | * blk_queue_make_request - define an alternate make_request function for a device | |
114 | * @q: the request queue for the device to be affected | |
115 | * @mfn: the alternate make_request function | |
116 | * | |
117 | * Description: | |
118 | * The normal way for &struct bios to be passed to a device | |
119 | * driver is for them to be collected into requests on a request | |
120 | * queue, and then to allow the device driver to select requests | |
121 | * off that queue when it is ready. This works well for many block | |
122 | * devices. However some block devices (typically virtual devices | |
123 | * such as md or lvm) do not benefit from the processing on the | |
124 | * request queue, and are served best by having the requests passed | |
125 | * directly to them. This can be achieved by providing a function | |
126 | * to blk_queue_make_request(). | |
127 | * | |
128 | * Caveat: | |
129 | * The driver that does this *must* be able to deal appropriately | |
130 | * with buffers in "highmemory". This can be accomplished by either calling | |
131 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
132 | * blk_queue_bounce() to create a buffer in normal memory. | |
133 | **/ | |
6728cb0e | 134 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
86db1e29 JA |
135 | { |
136 | /* | |
137 | * set defaults | |
138 | */ | |
139 | q->nr_requests = BLKDEV_MAX_RQ; | |
0e435ac2 | 140 | |
86db1e29 | 141 | q->make_request_fn = mfn; |
86db1e29 JA |
142 | blk_queue_dma_alignment(q, 511); |
143 | blk_queue_congestion_threshold(q); | |
144 | q->nr_batching = BLK_BATCH_REQ; | |
145 | ||
146 | q->unplug_thresh = 4; /* hmm */ | |
147 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | |
148 | if (q->unplug_delay == 0) | |
149 | q->unplug_delay = 1; | |
150 | ||
86db1e29 JA |
151 | q->unplug_timer.function = blk_unplug_timeout; |
152 | q->unplug_timer.data = (unsigned long)q; | |
153 | ||
e475bba2 | 154 | blk_set_default_limits(&q->limits); |
80ddf247 | 155 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
e475bba2 | 156 | |
a4e7d464 JA |
157 | /* |
158 | * If the caller didn't supply a lock, fall back to our embedded | |
159 | * per-queue locks | |
160 | */ | |
161 | if (!q->queue_lock) | |
162 | q->queue_lock = &q->__queue_lock; | |
163 | ||
86db1e29 JA |
164 | /* |
165 | * by default assume old behaviour and bounce for any highmem page | |
166 | */ | |
167 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
168 | } | |
86db1e29 JA |
169 | EXPORT_SYMBOL(blk_queue_make_request); |
170 | ||
171 | /** | |
172 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
cd0aca2d TH |
173 | * @q: the request queue for the device |
174 | * @dma_mask: the maximum address the device can handle | |
86db1e29 JA |
175 | * |
176 | * Description: | |
177 | * Different hardware can have different requirements as to what pages | |
178 | * it can do I/O directly to. A low level driver can call | |
179 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
cd0aca2d | 180 | * buffers for doing I/O to pages residing above @dma_mask. |
86db1e29 | 181 | **/ |
cd0aca2d | 182 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
86db1e29 | 183 | { |
cd0aca2d | 184 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
86db1e29 JA |
185 | int dma = 0; |
186 | ||
187 | q->bounce_gfp = GFP_NOIO; | |
188 | #if BITS_PER_LONG == 64 | |
cd0aca2d TH |
189 | /* |
190 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
191 | * some IOMMUs can handle everything, but I don't know of a | |
192 | * way to test this here. | |
193 | */ | |
194 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
86db1e29 | 195 | dma = 1; |
025146e1 | 196 | q->limits.bounce_pfn = max_low_pfn; |
86db1e29 | 197 | #else |
6728cb0e | 198 | if (b_pfn < blk_max_low_pfn) |
86db1e29 | 199 | dma = 1; |
025146e1 | 200 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
201 | #endif |
202 | if (dma) { | |
203 | init_emergency_isa_pool(); | |
204 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
025146e1 | 205 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
206 | } |
207 | } | |
86db1e29 JA |
208 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
209 | ||
210 | /** | |
211 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
212 | * @q: the request queue for the device | |
213 | * @max_sectors: max sectors in the usual 512b unit | |
214 | * | |
215 | * Description: | |
216 | * Enables a low level driver to set an upper limit on the size of | |
217 | * received requests. | |
218 | **/ | |
219 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) | |
220 | { | |
221 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
222 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
24c03d47 HH |
223 | printk(KERN_INFO "%s: set to minimum %d\n", |
224 | __func__, max_sectors); | |
86db1e29 JA |
225 | } |
226 | ||
227 | if (BLK_DEF_MAX_SECTORS > max_sectors) | |
025146e1 | 228 | q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors; |
86db1e29 | 229 | else { |
025146e1 MP |
230 | q->limits.max_sectors = BLK_DEF_MAX_SECTORS; |
231 | q->limits.max_hw_sectors = max_sectors; | |
86db1e29 JA |
232 | } |
233 | } | |
86db1e29 JA |
234 | EXPORT_SYMBOL(blk_queue_max_sectors); |
235 | ||
ae03bf63 MP |
236 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors) |
237 | { | |
238 | if (BLK_DEF_MAX_SECTORS > max_sectors) | |
025146e1 | 239 | q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS; |
ae03bf63 | 240 | else |
025146e1 | 241 | q->limits.max_hw_sectors = max_sectors; |
ae03bf63 MP |
242 | } |
243 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); | |
244 | ||
67efc925 CH |
245 | /** |
246 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
247 | * @q: the request queue for the device | |
c7ebf065 | 248 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
249 | **/ |
250 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
251 | unsigned int max_discard_sectors) | |
252 | { | |
253 | q->limits.max_discard_sectors = max_discard_sectors; | |
254 | } | |
255 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
256 | ||
86db1e29 JA |
257 | /** |
258 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
259 | * @q: the request queue for the device | |
260 | * @max_segments: max number of segments | |
261 | * | |
262 | * Description: | |
263 | * Enables a low level driver to set an upper limit on the number of | |
264 | * physical data segments in a request. This would be the largest sized | |
265 | * scatter list the driver could handle. | |
266 | **/ | |
267 | void blk_queue_max_phys_segments(struct request_queue *q, | |
268 | unsigned short max_segments) | |
269 | { | |
270 | if (!max_segments) { | |
271 | max_segments = 1; | |
24c03d47 HH |
272 | printk(KERN_INFO "%s: set to minimum %d\n", |
273 | __func__, max_segments); | |
86db1e29 JA |
274 | } |
275 | ||
025146e1 | 276 | q->limits.max_phys_segments = max_segments; |
86db1e29 | 277 | } |
86db1e29 JA |
278 | EXPORT_SYMBOL(blk_queue_max_phys_segments); |
279 | ||
280 | /** | |
281 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
282 | * @q: the request queue for the device | |
283 | * @max_segments: max number of segments | |
284 | * | |
285 | * Description: | |
286 | * Enables a low level driver to set an upper limit on the number of | |
287 | * hw data segments in a request. This would be the largest number of | |
710027a4 | 288 | * address/length pairs the host adapter can actually give at once |
86db1e29 JA |
289 | * to the device. |
290 | **/ | |
291 | void blk_queue_max_hw_segments(struct request_queue *q, | |
292 | unsigned short max_segments) | |
293 | { | |
294 | if (!max_segments) { | |
295 | max_segments = 1; | |
24c03d47 HH |
296 | printk(KERN_INFO "%s: set to minimum %d\n", |
297 | __func__, max_segments); | |
86db1e29 JA |
298 | } |
299 | ||
025146e1 | 300 | q->limits.max_hw_segments = max_segments; |
86db1e29 | 301 | } |
86db1e29 JA |
302 | EXPORT_SYMBOL(blk_queue_max_hw_segments); |
303 | ||
304 | /** | |
305 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
306 | * @q: the request queue for the device | |
307 | * @max_size: max size of segment in bytes | |
308 | * | |
309 | * Description: | |
310 | * Enables a low level driver to set an upper limit on the size of a | |
311 | * coalesced segment | |
312 | **/ | |
313 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
314 | { | |
315 | if (max_size < PAGE_CACHE_SIZE) { | |
316 | max_size = PAGE_CACHE_SIZE; | |
24c03d47 HH |
317 | printk(KERN_INFO "%s: set to minimum %d\n", |
318 | __func__, max_size); | |
86db1e29 JA |
319 | } |
320 | ||
025146e1 | 321 | q->limits.max_segment_size = max_size; |
86db1e29 | 322 | } |
86db1e29 JA |
323 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
324 | ||
325 | /** | |
e1defc4f | 326 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 327 | * @q: the request queue for the device |
e1defc4f | 328 | * @size: the logical block size, in bytes |
86db1e29 JA |
329 | * |
330 | * Description: | |
e1defc4f MP |
331 | * This should be set to the lowest possible block size that the |
332 | * storage device can address. The default of 512 covers most | |
333 | * hardware. | |
86db1e29 | 334 | **/ |
e1defc4f | 335 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
86db1e29 | 336 | { |
025146e1 | 337 | q->limits.logical_block_size = size; |
c72758f3 MP |
338 | |
339 | if (q->limits.physical_block_size < size) | |
340 | q->limits.physical_block_size = size; | |
341 | ||
342 | if (q->limits.io_min < q->limits.physical_block_size) | |
343 | q->limits.io_min = q->limits.physical_block_size; | |
86db1e29 | 344 | } |
e1defc4f | 345 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 346 | |
c72758f3 MP |
347 | /** |
348 | * blk_queue_physical_block_size - set physical block size for the queue | |
349 | * @q: the request queue for the device | |
350 | * @size: the physical block size, in bytes | |
351 | * | |
352 | * Description: | |
353 | * This should be set to the lowest possible sector size that the | |
354 | * hardware can operate on without reverting to read-modify-write | |
355 | * operations. | |
356 | */ | |
357 | void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) | |
358 | { | |
359 | q->limits.physical_block_size = size; | |
360 | ||
361 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
362 | q->limits.physical_block_size = q->limits.logical_block_size; | |
363 | ||
364 | if (q->limits.io_min < q->limits.physical_block_size) | |
365 | q->limits.io_min = q->limits.physical_block_size; | |
366 | } | |
367 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
368 | ||
369 | /** | |
370 | * blk_queue_alignment_offset - set physical block alignment offset | |
371 | * @q: the request queue for the device | |
8ebf9756 | 372 | * @offset: alignment offset in bytes |
c72758f3 MP |
373 | * |
374 | * Description: | |
375 | * Some devices are naturally misaligned to compensate for things like | |
376 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
377 | * should call this function for devices whose first sector is not | |
378 | * naturally aligned. | |
379 | */ | |
380 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
381 | { | |
382 | q->limits.alignment_offset = | |
383 | offset & (q->limits.physical_block_size - 1); | |
384 | q->limits.misaligned = 0; | |
385 | } | |
386 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
387 | ||
7c958e32 MP |
388 | /** |
389 | * blk_limits_io_min - set minimum request size for a device | |
390 | * @limits: the queue limits | |
391 | * @min: smallest I/O size in bytes | |
392 | * | |
393 | * Description: | |
394 | * Some devices have an internal block size bigger than the reported | |
395 | * hardware sector size. This function can be used to signal the | |
396 | * smallest I/O the device can perform without incurring a performance | |
397 | * penalty. | |
398 | */ | |
399 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
400 | { | |
401 | limits->io_min = min; | |
402 | ||
403 | if (limits->io_min < limits->logical_block_size) | |
404 | limits->io_min = limits->logical_block_size; | |
405 | ||
406 | if (limits->io_min < limits->physical_block_size) | |
407 | limits->io_min = limits->physical_block_size; | |
408 | } | |
409 | EXPORT_SYMBOL(blk_limits_io_min); | |
410 | ||
c72758f3 MP |
411 | /** |
412 | * blk_queue_io_min - set minimum request size for the queue | |
413 | * @q: the request queue for the device | |
8ebf9756 | 414 | * @min: smallest I/O size in bytes |
c72758f3 MP |
415 | * |
416 | * Description: | |
7e5f5fb0 MP |
417 | * Storage devices may report a granularity or preferred minimum I/O |
418 | * size which is the smallest request the device can perform without | |
419 | * incurring a performance penalty. For disk drives this is often the | |
420 | * physical block size. For RAID arrays it is often the stripe chunk | |
421 | * size. A properly aligned multiple of minimum_io_size is the | |
422 | * preferred request size for workloads where a high number of I/O | |
423 | * operations is desired. | |
c72758f3 MP |
424 | */ |
425 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
426 | { | |
7c958e32 | 427 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
428 | } |
429 | EXPORT_SYMBOL(blk_queue_io_min); | |
430 | ||
3c5820c7 MP |
431 | /** |
432 | * blk_limits_io_opt - set optimal request size for a device | |
433 | * @limits: the queue limits | |
434 | * @opt: smallest I/O size in bytes | |
435 | * | |
436 | * Description: | |
437 | * Storage devices may report an optimal I/O size, which is the | |
438 | * device's preferred unit for sustained I/O. This is rarely reported | |
439 | * for disk drives. For RAID arrays it is usually the stripe width or | |
440 | * the internal track size. A properly aligned multiple of | |
441 | * optimal_io_size is the preferred request size for workloads where | |
442 | * sustained throughput is desired. | |
443 | */ | |
444 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
445 | { | |
446 | limits->io_opt = opt; | |
447 | } | |
448 | EXPORT_SYMBOL(blk_limits_io_opt); | |
449 | ||
c72758f3 MP |
450 | /** |
451 | * blk_queue_io_opt - set optimal request size for the queue | |
452 | * @q: the request queue for the device | |
8ebf9756 | 453 | * @opt: optimal request size in bytes |
c72758f3 MP |
454 | * |
455 | * Description: | |
7e5f5fb0 MP |
456 | * Storage devices may report an optimal I/O size, which is the |
457 | * device's preferred unit for sustained I/O. This is rarely reported | |
458 | * for disk drives. For RAID arrays it is usually the stripe width or | |
459 | * the internal track size. A properly aligned multiple of | |
460 | * optimal_io_size is the preferred request size for workloads where | |
461 | * sustained throughput is desired. | |
c72758f3 MP |
462 | */ |
463 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
464 | { | |
3c5820c7 | 465 | blk_limits_io_opt(&q->limits, opt); |
c72758f3 MP |
466 | } |
467 | EXPORT_SYMBOL(blk_queue_io_opt); | |
468 | ||
86db1e29 JA |
469 | /* |
470 | * Returns the minimum that is _not_ zero, unless both are zero. | |
471 | */ | |
472 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
473 | ||
474 | /** | |
475 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
476 | * @t: the stacking driver (top) | |
477 | * @b: the underlying device (bottom) | |
478 | **/ | |
479 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
480 | { | |
fef24667 | 481 | blk_stack_limits(&t->limits, &b->limits, 0); |
025146e1 | 482 | |
e7e72bf6 NB |
483 | if (!t->queue_lock) |
484 | WARN_ON_ONCE(1); | |
485 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
486 | unsigned long flags; | |
487 | spin_lock_irqsave(t->queue_lock, flags); | |
75ad23bc | 488 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
e7e72bf6 NB |
489 | spin_unlock_irqrestore(t->queue_lock, flags); |
490 | } | |
86db1e29 | 491 | } |
86db1e29 JA |
492 | EXPORT_SYMBOL(blk_queue_stack_limits); |
493 | ||
86b37281 MP |
494 | static unsigned int lcm(unsigned int a, unsigned int b) |
495 | { | |
496 | if (a && b) | |
497 | return (a * b) / gcd(a, b); | |
498 | else if (b) | |
499 | return b; | |
500 | ||
501 | return a; | |
502 | } | |
503 | ||
c72758f3 MP |
504 | /** |
505 | * blk_stack_limits - adjust queue_limits for stacked devices | |
506 | * @t: the stacking driver limits (top) | |
77634f33 | 507 | * @b: the underlying queue limits (bottom) |
c72758f3 MP |
508 | * @offset: offset to beginning of data within component device |
509 | * | |
510 | * Description: | |
511 | * Merges two queue_limit structs. Returns 0 if alignment didn't | |
512 | * change. Returns -1 if adding the bottom device caused | |
513 | * misalignment. | |
514 | */ | |
515 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
516 | sector_t offset) | |
517 | { | |
86b37281 MP |
518 | int ret; |
519 | ||
520 | ret = 0; | |
521 | ||
c72758f3 MP |
522 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
523 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
77634f33 | 524 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
c72758f3 MP |
525 | |
526 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
527 | b->seg_boundary_mask); | |
528 | ||
529 | t->max_phys_segments = min_not_zero(t->max_phys_segments, | |
530 | b->max_phys_segments); | |
531 | ||
532 | t->max_hw_segments = min_not_zero(t->max_hw_segments, | |
533 | b->max_hw_segments); | |
534 | ||
535 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
536 | b->max_segment_size); | |
537 | ||
538 | t->logical_block_size = max(t->logical_block_size, | |
539 | b->logical_block_size); | |
540 | ||
541 | t->physical_block_size = max(t->physical_block_size, | |
542 | b->physical_block_size); | |
543 | ||
544 | t->io_min = max(t->io_min, b->io_min); | |
545 | t->no_cluster |= b->no_cluster; | |
546 | ||
547 | /* Bottom device offset aligned? */ | |
548 | if (offset && | |
549 | (offset & (b->physical_block_size - 1)) != b->alignment_offset) { | |
550 | t->misaligned = 1; | |
86b37281 MP |
551 | ret = -1; |
552 | } | |
553 | ||
554 | if (offset && | |
555 | (offset & (b->discard_granularity - 1)) != b->discard_alignment) { | |
556 | t->discard_misaligned = 1; | |
557 | ret = -1; | |
c72758f3 MP |
558 | } |
559 | ||
560 | /* If top has no alignment offset, inherit from bottom */ | |
561 | if (!t->alignment_offset) | |
562 | t->alignment_offset = | |
563 | b->alignment_offset & (b->physical_block_size - 1); | |
564 | ||
86b37281 MP |
565 | if (!t->discard_alignment) |
566 | t->discard_alignment = | |
567 | b->discard_alignment & (b->discard_granularity - 1); | |
568 | ||
c72758f3 MP |
569 | /* Top device aligned on logical block boundary? */ |
570 | if (t->alignment_offset & (t->logical_block_size - 1)) { | |
571 | t->misaligned = 1; | |
86b37281 | 572 | ret = -1; |
c72758f3 MP |
573 | } |
574 | ||
86b37281 MP |
575 | /* Find lcm() of optimal I/O size and granularity */ |
576 | t->io_opt = lcm(t->io_opt, b->io_opt); | |
577 | t->discard_granularity = lcm(t->discard_granularity, | |
578 | b->discard_granularity); | |
70dd5bf3 MP |
579 | |
580 | /* Verify that optimal I/O size is a multiple of io_min */ | |
581 | if (t->io_min && t->io_opt % t->io_min) | |
86b37281 | 582 | ret = -1; |
70dd5bf3 | 583 | |
86b37281 | 584 | return ret; |
c72758f3 | 585 | } |
5d85d324 | 586 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 MP |
587 | |
588 | /** | |
589 | * disk_stack_limits - adjust queue limits for stacked drivers | |
77634f33 | 590 | * @disk: MD/DM gendisk (top) |
c72758f3 MP |
591 | * @bdev: the underlying block device (bottom) |
592 | * @offset: offset to beginning of data within component device | |
593 | * | |
594 | * Description: | |
595 | * Merges the limits for two queues. Returns 0 if alignment | |
596 | * didn't change. Returns -1 if adding the bottom device caused | |
597 | * misalignment. | |
598 | */ | |
599 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
600 | sector_t offset) | |
601 | { | |
602 | struct request_queue *t = disk->queue; | |
603 | struct request_queue *b = bdev_get_queue(bdev); | |
604 | ||
605 | offset += get_start_sect(bdev) << 9; | |
606 | ||
607 | if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) { | |
608 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; | |
609 | ||
610 | disk_name(disk, 0, top); | |
611 | bdevname(bdev, bottom); | |
612 | ||
613 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
614 | top, bottom); | |
615 | } | |
616 | ||
617 | if (!t->queue_lock) | |
618 | WARN_ON_ONCE(1); | |
619 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
620 | unsigned long flags; | |
621 | ||
622 | spin_lock_irqsave(t->queue_lock, flags); | |
623 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) | |
624 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); | |
625 | spin_unlock_irqrestore(t->queue_lock, flags); | |
626 | } | |
627 | } | |
628 | EXPORT_SYMBOL(disk_stack_limits); | |
629 | ||
e3790c7d TH |
630 | /** |
631 | * blk_queue_dma_pad - set pad mask | |
632 | * @q: the request queue for the device | |
633 | * @mask: pad mask | |
634 | * | |
27f8221a | 635 | * Set dma pad mask. |
e3790c7d | 636 | * |
27f8221a FT |
637 | * Appending pad buffer to a request modifies the last entry of a |
638 | * scatter list such that it includes the pad buffer. | |
e3790c7d TH |
639 | **/ |
640 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
641 | { | |
642 | q->dma_pad_mask = mask; | |
643 | } | |
644 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
645 | ||
27f8221a FT |
646 | /** |
647 | * blk_queue_update_dma_pad - update pad mask | |
648 | * @q: the request queue for the device | |
649 | * @mask: pad mask | |
650 | * | |
651 | * Update dma pad mask. | |
652 | * | |
653 | * Appending pad buffer to a request modifies the last entry of a | |
654 | * scatter list such that it includes the pad buffer. | |
655 | **/ | |
656 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
657 | { | |
658 | if (mask > q->dma_pad_mask) | |
659 | q->dma_pad_mask = mask; | |
660 | } | |
661 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
662 | ||
86db1e29 JA |
663 | /** |
664 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
86db1e29 | 665 | * @q: the request queue for the device |
2fb98e84 | 666 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
86db1e29 JA |
667 | * @buf: physically contiguous buffer |
668 | * @size: size of the buffer in bytes | |
669 | * | |
670 | * Some devices have excess DMA problems and can't simply discard (or | |
671 | * zero fill) the unwanted piece of the transfer. They have to have a | |
672 | * real area of memory to transfer it into. The use case for this is | |
673 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
674 | * bigger than the transfer size some HBAs will lock up if there | |
675 | * aren't DMA elements to contain the excess transfer. What this API | |
676 | * does is adjust the queue so that the buf is always appended | |
677 | * silently to the scatterlist. | |
678 | * | |
679 | * Note: This routine adjusts max_hw_segments to make room for | |
680 | * appending the drain buffer. If you call | |
681 | * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after | |
682 | * calling this routine, you must set the limit to one fewer than your | |
683 | * device can support otherwise there won't be room for the drain | |
684 | * buffer. | |
685 | */ | |
448da4d2 | 686 | int blk_queue_dma_drain(struct request_queue *q, |
2fb98e84 TH |
687 | dma_drain_needed_fn *dma_drain_needed, |
688 | void *buf, unsigned int size) | |
86db1e29 | 689 | { |
ae03bf63 | 690 | if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2) |
86db1e29 JA |
691 | return -EINVAL; |
692 | /* make room for appending the drain */ | |
ae03bf63 MP |
693 | blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1); |
694 | blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1); | |
2fb98e84 | 695 | q->dma_drain_needed = dma_drain_needed; |
86db1e29 JA |
696 | q->dma_drain_buffer = buf; |
697 | q->dma_drain_size = size; | |
698 | ||
699 | return 0; | |
700 | } | |
86db1e29 JA |
701 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
702 | ||
703 | /** | |
704 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
705 | * @q: the request queue for the device | |
706 | * @mask: the memory boundary mask | |
707 | **/ | |
708 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
709 | { | |
710 | if (mask < PAGE_CACHE_SIZE - 1) { | |
711 | mask = PAGE_CACHE_SIZE - 1; | |
24c03d47 HH |
712 | printk(KERN_INFO "%s: set to minimum %lx\n", |
713 | __func__, mask); | |
86db1e29 JA |
714 | } |
715 | ||
025146e1 | 716 | q->limits.seg_boundary_mask = mask; |
86db1e29 | 717 | } |
86db1e29 JA |
718 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
719 | ||
720 | /** | |
721 | * blk_queue_dma_alignment - set dma length and memory alignment | |
722 | * @q: the request queue for the device | |
723 | * @mask: alignment mask | |
724 | * | |
725 | * description: | |
710027a4 | 726 | * set required memory and length alignment for direct dma transactions. |
8feb4d20 | 727 | * this is used when building direct io requests for the queue. |
86db1e29 JA |
728 | * |
729 | **/ | |
730 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
731 | { | |
732 | q->dma_alignment = mask; | |
733 | } | |
86db1e29 JA |
734 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
735 | ||
736 | /** | |
737 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
738 | * @q: the request queue for the device | |
739 | * @mask: alignment mask | |
740 | * | |
741 | * description: | |
710027a4 | 742 | * update required memory and length alignment for direct dma transactions. |
86db1e29 JA |
743 | * If the requested alignment is larger than the current alignment, then |
744 | * the current queue alignment is updated to the new value, otherwise it | |
745 | * is left alone. The design of this is to allow multiple objects | |
746 | * (driver, device, transport etc) to set their respective | |
747 | * alignments without having them interfere. | |
748 | * | |
749 | **/ | |
750 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
751 | { | |
752 | BUG_ON(mask > PAGE_SIZE); | |
753 | ||
754 | if (mask > q->dma_alignment) | |
755 | q->dma_alignment = mask; | |
756 | } | |
86db1e29 JA |
757 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
758 | ||
aeb3d3a8 | 759 | static int __init blk_settings_init(void) |
86db1e29 JA |
760 | { |
761 | blk_max_low_pfn = max_low_pfn - 1; | |
762 | blk_max_pfn = max_pfn - 1; | |
763 | return 0; | |
764 | } | |
765 | subsys_initcall(blk_settings_init); |