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> |
ad5ebd2f | 11 | #include <linux/jiffies.h> |
86db1e29 JA |
12 | |
13 | #include "blk.h" | |
14 | ||
6728cb0e | 15 | unsigned long blk_max_low_pfn; |
86db1e29 | 16 | EXPORT_SYMBOL(blk_max_low_pfn); |
6728cb0e JA |
17 | |
18 | unsigned long blk_max_pfn; | |
86db1e29 JA |
19 | |
20 | /** | |
21 | * blk_queue_prep_rq - set a prepare_request function for queue | |
22 | * @q: queue | |
23 | * @pfn: prepare_request function | |
24 | * | |
25 | * It's possible for a queue to register a prepare_request callback which | |
26 | * is invoked before the request is handed to the request_fn. The goal of | |
27 | * the function is to prepare a request for I/O, it can be used to build a | |
28 | * cdb from the request data for instance. | |
29 | * | |
30 | */ | |
31 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) | |
32 | { | |
33 | q->prep_rq_fn = pfn; | |
34 | } | |
86db1e29 JA |
35 | EXPORT_SYMBOL(blk_queue_prep_rq); |
36 | ||
37 | /** | |
38 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
39 | * @q: queue | |
40 | * @mbfn: merge_bvec_fn | |
41 | * | |
42 | * Usually queues have static limitations on the max sectors or segments that | |
43 | * we can put in a request. Stacking drivers may have some settings that | |
44 | * are dynamic, and thus we have to query the queue whether it is ok to | |
45 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
46 | * has such limitations, it needs to register a merge_bvec_fn to control | |
47 | * the size of bio's sent to it. Note that a block device *must* allow a | |
48 | * single page to be added to an empty bio. The block device driver may want | |
49 | * to use the bio_split() function to deal with these bio's. By default | |
50 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
51 | * honored. | |
52 | */ | |
53 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) | |
54 | { | |
55 | q->merge_bvec_fn = mbfn; | |
56 | } | |
86db1e29 JA |
57 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
58 | ||
59 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
60 | { | |
61 | q->softirq_done_fn = fn; | |
62 | } | |
86db1e29 JA |
63 | EXPORT_SYMBOL(blk_queue_softirq_done); |
64 | ||
242f9dcb JA |
65 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
66 | { | |
67 | q->rq_timeout = timeout; | |
68 | } | |
69 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); | |
70 | ||
71 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) | |
72 | { | |
73 | q->rq_timed_out_fn = fn; | |
74 | } | |
75 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); | |
76 | ||
ef9e3fac KU |
77 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) |
78 | { | |
79 | q->lld_busy_fn = fn; | |
80 | } | |
81 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); | |
82 | ||
e475bba2 MP |
83 | /** |
84 | * blk_set_default_limits - reset limits to default values | |
f740f5ca | 85 | * @lim: the queue_limits structure to reset |
e475bba2 MP |
86 | * |
87 | * Description: | |
88 | * Returns a queue_limit struct to its default state. Can be used by | |
89 | * stacking drivers like DM that stage table swaps and reuse an | |
90 | * existing device queue. | |
91 | */ | |
92 | void blk_set_default_limits(struct queue_limits *lim) | |
93 | { | |
94 | lim->max_phys_segments = MAX_PHYS_SEGMENTS; | |
95 | lim->max_hw_segments = MAX_HW_SEGMENTS; | |
96 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; | |
97 | lim->max_segment_size = MAX_SEGMENT_SIZE; | |
5dee2477 MP |
98 | lim->max_sectors = BLK_DEF_MAX_SECTORS; |
99 | lim->max_hw_sectors = INT_MAX; | |
86b37281 MP |
100 | lim->max_discard_sectors = 0; |
101 | lim->discard_granularity = 0; | |
102 | lim->discard_alignment = 0; | |
103 | lim->discard_misaligned = 0; | |
98262f27 | 104 | lim->discard_zeroes_data = -1; |
e475bba2 | 105 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
3a02c8e8 | 106 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
e475bba2 MP |
107 | lim->alignment_offset = 0; |
108 | lim->io_opt = 0; | |
109 | lim->misaligned = 0; | |
110 | lim->no_cluster = 0; | |
111 | } | |
112 | EXPORT_SYMBOL(blk_set_default_limits); | |
113 | ||
86db1e29 JA |
114 | /** |
115 | * blk_queue_make_request - define an alternate make_request function for a device | |
116 | * @q: the request queue for the device to be affected | |
117 | * @mfn: the alternate make_request function | |
118 | * | |
119 | * Description: | |
120 | * The normal way for &struct bios to be passed to a device | |
121 | * driver is for them to be collected into requests on a request | |
122 | * queue, and then to allow the device driver to select requests | |
123 | * off that queue when it is ready. This works well for many block | |
124 | * devices. However some block devices (typically virtual devices | |
125 | * such as md or lvm) do not benefit from the processing on the | |
126 | * request queue, and are served best by having the requests passed | |
127 | * directly to them. This can be achieved by providing a function | |
128 | * to blk_queue_make_request(). | |
129 | * | |
130 | * Caveat: | |
131 | * The driver that does this *must* be able to deal appropriately | |
132 | * with buffers in "highmemory". This can be accomplished by either calling | |
133 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
134 | * blk_queue_bounce() to create a buffer in normal memory. | |
135 | **/ | |
6728cb0e | 136 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
86db1e29 JA |
137 | { |
138 | /* | |
139 | * set defaults | |
140 | */ | |
141 | q->nr_requests = BLKDEV_MAX_RQ; | |
0e435ac2 | 142 | |
86db1e29 | 143 | q->make_request_fn = mfn; |
86db1e29 JA |
144 | blk_queue_dma_alignment(q, 511); |
145 | blk_queue_congestion_threshold(q); | |
146 | q->nr_batching = BLK_BATCH_REQ; | |
147 | ||
148 | q->unplug_thresh = 4; /* hmm */ | |
ad5ebd2f | 149 | q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */ |
86db1e29 JA |
150 | if (q->unplug_delay == 0) |
151 | q->unplug_delay = 1; | |
152 | ||
86db1e29 JA |
153 | q->unplug_timer.function = blk_unplug_timeout; |
154 | q->unplug_timer.data = (unsigned long)q; | |
155 | ||
e475bba2 | 156 | blk_set_default_limits(&q->limits); |
80ddf247 | 157 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
e475bba2 | 158 | |
a4e7d464 JA |
159 | /* |
160 | * If the caller didn't supply a lock, fall back to our embedded | |
161 | * per-queue locks | |
162 | */ | |
163 | if (!q->queue_lock) | |
164 | q->queue_lock = &q->__queue_lock; | |
165 | ||
86db1e29 JA |
166 | /* |
167 | * by default assume old behaviour and bounce for any highmem page | |
168 | */ | |
169 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
170 | } | |
86db1e29 JA |
171 | EXPORT_SYMBOL(blk_queue_make_request); |
172 | ||
173 | /** | |
174 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
cd0aca2d TH |
175 | * @q: the request queue for the device |
176 | * @dma_mask: the maximum address the device can handle | |
86db1e29 JA |
177 | * |
178 | * Description: | |
179 | * Different hardware can have different requirements as to what pages | |
180 | * it can do I/O directly to. A low level driver can call | |
181 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
cd0aca2d | 182 | * buffers for doing I/O to pages residing above @dma_mask. |
86db1e29 | 183 | **/ |
cd0aca2d | 184 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
86db1e29 | 185 | { |
cd0aca2d | 186 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
86db1e29 JA |
187 | int dma = 0; |
188 | ||
189 | q->bounce_gfp = GFP_NOIO; | |
190 | #if BITS_PER_LONG == 64 | |
cd0aca2d TH |
191 | /* |
192 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
193 | * some IOMMUs can handle everything, but I don't know of a | |
194 | * way to test this here. | |
195 | */ | |
196 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
86db1e29 | 197 | dma = 1; |
025146e1 | 198 | q->limits.bounce_pfn = max_low_pfn; |
86db1e29 | 199 | #else |
6728cb0e | 200 | if (b_pfn < blk_max_low_pfn) |
86db1e29 | 201 | dma = 1; |
025146e1 | 202 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
203 | #endif |
204 | if (dma) { | |
205 | init_emergency_isa_pool(); | |
206 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
025146e1 | 207 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
208 | } |
209 | } | |
86db1e29 JA |
210 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
211 | ||
212 | /** | |
213 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
214 | * @q: the request queue for the device | |
215 | * @max_sectors: max sectors in the usual 512b unit | |
216 | * | |
217 | * Description: | |
218 | * Enables a low level driver to set an upper limit on the size of | |
219 | * received requests. | |
220 | **/ | |
221 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) | |
222 | { | |
223 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
224 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
24c03d47 HH |
225 | printk(KERN_INFO "%s: set to minimum %d\n", |
226 | __func__, max_sectors); | |
86db1e29 JA |
227 | } |
228 | ||
229 | if (BLK_DEF_MAX_SECTORS > max_sectors) | |
025146e1 | 230 | q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors; |
86db1e29 | 231 | else { |
025146e1 MP |
232 | q->limits.max_sectors = BLK_DEF_MAX_SECTORS; |
233 | q->limits.max_hw_sectors = max_sectors; | |
86db1e29 JA |
234 | } |
235 | } | |
86db1e29 JA |
236 | EXPORT_SYMBOL(blk_queue_max_sectors); |
237 | ||
ae03bf63 MP |
238 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors) |
239 | { | |
240 | if (BLK_DEF_MAX_SECTORS > max_sectors) | |
025146e1 | 241 | q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS; |
ae03bf63 | 242 | else |
025146e1 | 243 | q->limits.max_hw_sectors = max_sectors; |
ae03bf63 MP |
244 | } |
245 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); | |
246 | ||
67efc925 CH |
247 | /** |
248 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
249 | * @q: the request queue for the device | |
c7ebf065 | 250 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
251 | **/ |
252 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
253 | unsigned int max_discard_sectors) | |
254 | { | |
255 | q->limits.max_discard_sectors = max_discard_sectors; | |
256 | } | |
257 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
258 | ||
86db1e29 JA |
259 | /** |
260 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
261 | * @q: the request queue for the device | |
262 | * @max_segments: max number of segments | |
263 | * | |
264 | * Description: | |
265 | * Enables a low level driver to set an upper limit on the number of | |
266 | * physical data segments in a request. This would be the largest sized | |
267 | * scatter list the driver could handle. | |
268 | **/ | |
269 | void blk_queue_max_phys_segments(struct request_queue *q, | |
270 | unsigned short max_segments) | |
271 | { | |
272 | if (!max_segments) { | |
273 | max_segments = 1; | |
24c03d47 HH |
274 | printk(KERN_INFO "%s: set to minimum %d\n", |
275 | __func__, max_segments); | |
86db1e29 JA |
276 | } |
277 | ||
025146e1 | 278 | q->limits.max_phys_segments = max_segments; |
86db1e29 | 279 | } |
86db1e29 JA |
280 | EXPORT_SYMBOL(blk_queue_max_phys_segments); |
281 | ||
282 | /** | |
283 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
284 | * @q: the request queue for the device | |
285 | * @max_segments: max number of segments | |
286 | * | |
287 | * Description: | |
288 | * Enables a low level driver to set an upper limit on the number of | |
289 | * hw data segments in a request. This would be the largest number of | |
710027a4 | 290 | * address/length pairs the host adapter can actually give at once |
86db1e29 JA |
291 | * to the device. |
292 | **/ | |
293 | void blk_queue_max_hw_segments(struct request_queue *q, | |
294 | unsigned short max_segments) | |
295 | { | |
296 | if (!max_segments) { | |
297 | max_segments = 1; | |
24c03d47 HH |
298 | printk(KERN_INFO "%s: set to minimum %d\n", |
299 | __func__, max_segments); | |
86db1e29 JA |
300 | } |
301 | ||
025146e1 | 302 | q->limits.max_hw_segments = max_segments; |
86db1e29 | 303 | } |
86db1e29 JA |
304 | EXPORT_SYMBOL(blk_queue_max_hw_segments); |
305 | ||
306 | /** | |
307 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
308 | * @q: the request queue for the device | |
309 | * @max_size: max size of segment in bytes | |
310 | * | |
311 | * Description: | |
312 | * Enables a low level driver to set an upper limit on the size of a | |
313 | * coalesced segment | |
314 | **/ | |
315 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
316 | { | |
317 | if (max_size < PAGE_CACHE_SIZE) { | |
318 | max_size = PAGE_CACHE_SIZE; | |
24c03d47 HH |
319 | printk(KERN_INFO "%s: set to minimum %d\n", |
320 | __func__, max_size); | |
86db1e29 JA |
321 | } |
322 | ||
025146e1 | 323 | q->limits.max_segment_size = max_size; |
86db1e29 | 324 | } |
86db1e29 JA |
325 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
326 | ||
327 | /** | |
e1defc4f | 328 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 329 | * @q: the request queue for the device |
e1defc4f | 330 | * @size: the logical block size, in bytes |
86db1e29 JA |
331 | * |
332 | * Description: | |
e1defc4f MP |
333 | * This should be set to the lowest possible block size that the |
334 | * storage device can address. The default of 512 covers most | |
335 | * hardware. | |
86db1e29 | 336 | **/ |
e1defc4f | 337 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
86db1e29 | 338 | { |
025146e1 | 339 | q->limits.logical_block_size = size; |
c72758f3 MP |
340 | |
341 | if (q->limits.physical_block_size < size) | |
342 | q->limits.physical_block_size = size; | |
343 | ||
344 | if (q->limits.io_min < q->limits.physical_block_size) | |
345 | q->limits.io_min = q->limits.physical_block_size; | |
86db1e29 | 346 | } |
e1defc4f | 347 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 348 | |
c72758f3 MP |
349 | /** |
350 | * blk_queue_physical_block_size - set physical block size for the queue | |
351 | * @q: the request queue for the device | |
352 | * @size: the physical block size, in bytes | |
353 | * | |
354 | * Description: | |
355 | * This should be set to the lowest possible sector size that the | |
356 | * hardware can operate on without reverting to read-modify-write | |
357 | * operations. | |
358 | */ | |
359 | void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) | |
360 | { | |
361 | q->limits.physical_block_size = size; | |
362 | ||
363 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
364 | q->limits.physical_block_size = q->limits.logical_block_size; | |
365 | ||
366 | if (q->limits.io_min < q->limits.physical_block_size) | |
367 | q->limits.io_min = q->limits.physical_block_size; | |
368 | } | |
369 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
370 | ||
371 | /** | |
372 | * blk_queue_alignment_offset - set physical block alignment offset | |
373 | * @q: the request queue for the device | |
8ebf9756 | 374 | * @offset: alignment offset in bytes |
c72758f3 MP |
375 | * |
376 | * Description: | |
377 | * Some devices are naturally misaligned to compensate for things like | |
378 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
379 | * should call this function for devices whose first sector is not | |
380 | * naturally aligned. | |
381 | */ | |
382 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
383 | { | |
384 | q->limits.alignment_offset = | |
385 | offset & (q->limits.physical_block_size - 1); | |
386 | q->limits.misaligned = 0; | |
387 | } | |
388 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
389 | ||
7c958e32 MP |
390 | /** |
391 | * blk_limits_io_min - set minimum request size for a device | |
392 | * @limits: the queue limits | |
393 | * @min: smallest I/O size in bytes | |
394 | * | |
395 | * Description: | |
396 | * Some devices have an internal block size bigger than the reported | |
397 | * hardware sector size. This function can be used to signal the | |
398 | * smallest I/O the device can perform without incurring a performance | |
399 | * penalty. | |
400 | */ | |
401 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
402 | { | |
403 | limits->io_min = min; | |
404 | ||
405 | if (limits->io_min < limits->logical_block_size) | |
406 | limits->io_min = limits->logical_block_size; | |
407 | ||
408 | if (limits->io_min < limits->physical_block_size) | |
409 | limits->io_min = limits->physical_block_size; | |
410 | } | |
411 | EXPORT_SYMBOL(blk_limits_io_min); | |
412 | ||
c72758f3 MP |
413 | /** |
414 | * blk_queue_io_min - set minimum request size for the queue | |
415 | * @q: the request queue for the device | |
8ebf9756 | 416 | * @min: smallest I/O size in bytes |
c72758f3 MP |
417 | * |
418 | * Description: | |
7e5f5fb0 MP |
419 | * Storage devices may report a granularity or preferred minimum I/O |
420 | * size which is the smallest request the device can perform without | |
421 | * incurring a performance penalty. For disk drives this is often the | |
422 | * physical block size. For RAID arrays it is often the stripe chunk | |
423 | * size. A properly aligned multiple of minimum_io_size is the | |
424 | * preferred request size for workloads where a high number of I/O | |
425 | * operations is desired. | |
c72758f3 MP |
426 | */ |
427 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
428 | { | |
7c958e32 | 429 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
430 | } |
431 | EXPORT_SYMBOL(blk_queue_io_min); | |
432 | ||
3c5820c7 MP |
433 | /** |
434 | * blk_limits_io_opt - set optimal request size for a device | |
435 | * @limits: the queue limits | |
436 | * @opt: smallest I/O size in bytes | |
437 | * | |
438 | * Description: | |
439 | * Storage devices may report an optimal I/O size, which is the | |
440 | * device's preferred unit for sustained I/O. This is rarely reported | |
441 | * for disk drives. For RAID arrays it is usually the stripe width or | |
442 | * the internal track size. A properly aligned multiple of | |
443 | * optimal_io_size is the preferred request size for workloads where | |
444 | * sustained throughput is desired. | |
445 | */ | |
446 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
447 | { | |
448 | limits->io_opt = opt; | |
449 | } | |
450 | EXPORT_SYMBOL(blk_limits_io_opt); | |
451 | ||
c72758f3 MP |
452 | /** |
453 | * blk_queue_io_opt - set optimal request size for the queue | |
454 | * @q: the request queue for the device | |
8ebf9756 | 455 | * @opt: optimal request size in bytes |
c72758f3 MP |
456 | * |
457 | * Description: | |
7e5f5fb0 MP |
458 | * Storage devices may report an optimal I/O size, which is the |
459 | * device's preferred unit for sustained I/O. This is rarely reported | |
460 | * for disk drives. For RAID arrays it is usually the stripe width or | |
461 | * the internal track size. A properly aligned multiple of | |
462 | * optimal_io_size is the preferred request size for workloads where | |
463 | * sustained throughput is desired. | |
c72758f3 MP |
464 | */ |
465 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
466 | { | |
3c5820c7 | 467 | blk_limits_io_opt(&q->limits, opt); |
c72758f3 MP |
468 | } |
469 | EXPORT_SYMBOL(blk_queue_io_opt); | |
470 | ||
86db1e29 JA |
471 | /* |
472 | * Returns the minimum that is _not_ zero, unless both are zero. | |
473 | */ | |
474 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
475 | ||
476 | /** | |
477 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
478 | * @t: the stacking driver (top) | |
479 | * @b: the underlying device (bottom) | |
480 | **/ | |
481 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
482 | { | |
fef24667 | 483 | blk_stack_limits(&t->limits, &b->limits, 0); |
025146e1 | 484 | |
e7e72bf6 NB |
485 | if (!t->queue_lock) |
486 | WARN_ON_ONCE(1); | |
487 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
488 | unsigned long flags; | |
489 | spin_lock_irqsave(t->queue_lock, flags); | |
75ad23bc | 490 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
e7e72bf6 NB |
491 | spin_unlock_irqrestore(t->queue_lock, flags); |
492 | } | |
86db1e29 | 493 | } |
86db1e29 JA |
494 | EXPORT_SYMBOL(blk_queue_stack_limits); |
495 | ||
86b37281 MP |
496 | static unsigned int lcm(unsigned int a, unsigned int b) |
497 | { | |
498 | if (a && b) | |
499 | return (a * b) / gcd(a, b); | |
500 | else if (b) | |
501 | return b; | |
502 | ||
503 | return a; | |
504 | } | |
505 | ||
c72758f3 MP |
506 | /** |
507 | * blk_stack_limits - adjust queue_limits for stacked devices | |
81744ee4 MP |
508 | * @t: the stacking driver limits (top device) |
509 | * @b: the underlying queue limits (bottom, component device) | |
c72758f3 MP |
510 | * @offset: offset to beginning of data within component device |
511 | * | |
512 | * Description: | |
81744ee4 MP |
513 | * This function is used by stacking drivers like MD and DM to ensure |
514 | * that all component devices have compatible block sizes and | |
515 | * alignments. The stacking driver must provide a queue_limits | |
516 | * struct (top) and then iteratively call the stacking function for | |
517 | * all component (bottom) devices. The stacking function will | |
518 | * attempt to combine the values and ensure proper alignment. | |
519 | * | |
520 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
521 | * top device's block sizes and alignment offsets may be adjusted to | |
522 | * ensure alignment with the bottom device. If no compatible sizes | |
523 | * and alignments exist, -1 is returned and the resulting top | |
524 | * queue_limits will have the misaligned flag set to indicate that | |
525 | * the alignment_offset is undefined. | |
c72758f3 MP |
526 | */ |
527 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
528 | sector_t offset) | |
529 | { | |
9504e086 | 530 | sector_t alignment; |
81744ee4 | 531 | unsigned int top, bottom; |
86b37281 | 532 | |
c72758f3 MP |
533 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
534 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
77634f33 | 535 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
c72758f3 MP |
536 | |
537 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
538 | b->seg_boundary_mask); | |
539 | ||
540 | t->max_phys_segments = min_not_zero(t->max_phys_segments, | |
541 | b->max_phys_segments); | |
542 | ||
543 | t->max_hw_segments = min_not_zero(t->max_hw_segments, | |
544 | b->max_hw_segments); | |
545 | ||
546 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
547 | b->max_segment_size); | |
548 | ||
81744ee4 | 549 | alignment = queue_limit_alignment_offset(b, offset); |
9504e086 | 550 | |
81744ee4 MP |
551 | /* Bottom device has different alignment. Check that it is |
552 | * compatible with the current top alignment. | |
553 | */ | |
9504e086 MP |
554 | if (t->alignment_offset != alignment) { |
555 | ||
556 | top = max(t->physical_block_size, t->io_min) | |
557 | + t->alignment_offset; | |
81744ee4 | 558 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
9504e086 | 559 | |
81744ee4 | 560 | /* Verify that top and bottom intervals line up */ |
9504e086 MP |
561 | if (max(top, bottom) & (min(top, bottom) - 1)) |
562 | t->misaligned = 1; | |
563 | } | |
564 | ||
c72758f3 MP |
565 | t->logical_block_size = max(t->logical_block_size, |
566 | b->logical_block_size); | |
567 | ||
568 | t->physical_block_size = max(t->physical_block_size, | |
569 | b->physical_block_size); | |
570 | ||
571 | t->io_min = max(t->io_min, b->io_min); | |
9504e086 MP |
572 | t->io_opt = lcm(t->io_opt, b->io_opt); |
573 | ||
c72758f3 | 574 | t->no_cluster |= b->no_cluster; |
98262f27 | 575 | t->discard_zeroes_data &= b->discard_zeroes_data; |
c72758f3 | 576 | |
81744ee4 | 577 | /* Physical block size a multiple of the logical block size? */ |
9504e086 MP |
578 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
579 | t->physical_block_size = t->logical_block_size; | |
c72758f3 | 580 | t->misaligned = 1; |
86b37281 MP |
581 | } |
582 | ||
81744ee4 | 583 | /* Minimum I/O a multiple of the physical block size? */ |
9504e086 MP |
584 | if (t->io_min & (t->physical_block_size - 1)) { |
585 | t->io_min = t->physical_block_size; | |
586 | t->misaligned = 1; | |
c72758f3 MP |
587 | } |
588 | ||
81744ee4 | 589 | /* Optimal I/O a multiple of the physical block size? */ |
9504e086 MP |
590 | if (t->io_opt & (t->physical_block_size - 1)) { |
591 | t->io_opt = 0; | |
592 | t->misaligned = 1; | |
593 | } | |
c72758f3 | 594 | |
81744ee4 | 595 | /* Find lowest common alignment_offset */ |
9504e086 MP |
596 | t->alignment_offset = lcm(t->alignment_offset, alignment) |
597 | & (max(t->physical_block_size, t->io_min) - 1); | |
86b37281 | 598 | |
81744ee4 | 599 | /* Verify that new alignment_offset is on a logical block boundary */ |
9504e086 | 600 | if (t->alignment_offset & (t->logical_block_size - 1)) |
c72758f3 | 601 | t->misaligned = 1; |
c72758f3 | 602 | |
9504e086 MP |
603 | /* Discard alignment and granularity */ |
604 | if (b->discard_granularity) { | |
81744ee4 MP |
605 | unsigned int granularity = b->discard_granularity; |
606 | offset &= granularity - 1; | |
9504e086 | 607 | |
81744ee4 MP |
608 | alignment = (granularity + b->discard_alignment - offset) |
609 | & (granularity - 1); | |
9504e086 MP |
610 | |
611 | if (t->discard_granularity != 0 && | |
612 | t->discard_alignment != alignment) { | |
613 | top = t->discard_granularity + t->discard_alignment; | |
614 | bottom = b->discard_granularity + alignment; | |
70dd5bf3 | 615 | |
9504e086 MP |
616 | /* Verify that top and bottom intervals line up */ |
617 | if (max(top, bottom) & (min(top, bottom) - 1)) | |
618 | t->discard_misaligned = 1; | |
619 | } | |
620 | ||
81744ee4 MP |
621 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
622 | b->max_discard_sectors); | |
9504e086 MP |
623 | t->discard_granularity = max(t->discard_granularity, |
624 | b->discard_granularity); | |
625 | t->discard_alignment = lcm(t->discard_alignment, alignment) & | |
626 | (t->discard_granularity - 1); | |
627 | } | |
70dd5bf3 | 628 | |
9504e086 | 629 | return t->misaligned ? -1 : 0; |
c72758f3 | 630 | } |
5d85d324 | 631 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 MP |
632 | |
633 | /** | |
634 | * disk_stack_limits - adjust queue limits for stacked drivers | |
77634f33 | 635 | * @disk: MD/DM gendisk (top) |
c72758f3 MP |
636 | * @bdev: the underlying block device (bottom) |
637 | * @offset: offset to beginning of data within component device | |
638 | * | |
639 | * Description: | |
640 | * Merges the limits for two queues. Returns 0 if alignment | |
641 | * didn't change. Returns -1 if adding the bottom device caused | |
642 | * misalignment. | |
643 | */ | |
644 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
645 | sector_t offset) | |
646 | { | |
647 | struct request_queue *t = disk->queue; | |
648 | struct request_queue *b = bdev_get_queue(bdev); | |
649 | ||
650 | offset += get_start_sect(bdev) << 9; | |
651 | ||
652 | if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) { | |
653 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; | |
654 | ||
655 | disk_name(disk, 0, top); | |
656 | bdevname(bdev, bottom); | |
657 | ||
658 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
659 | top, bottom); | |
660 | } | |
661 | ||
662 | if (!t->queue_lock) | |
663 | WARN_ON_ONCE(1); | |
664 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
665 | unsigned long flags; | |
666 | ||
667 | spin_lock_irqsave(t->queue_lock, flags); | |
668 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) | |
669 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); | |
670 | spin_unlock_irqrestore(t->queue_lock, flags); | |
671 | } | |
672 | } | |
673 | EXPORT_SYMBOL(disk_stack_limits); | |
674 | ||
e3790c7d TH |
675 | /** |
676 | * blk_queue_dma_pad - set pad mask | |
677 | * @q: the request queue for the device | |
678 | * @mask: pad mask | |
679 | * | |
27f8221a | 680 | * Set dma pad mask. |
e3790c7d | 681 | * |
27f8221a FT |
682 | * Appending pad buffer to a request modifies the last entry of a |
683 | * scatter list such that it includes the pad buffer. | |
e3790c7d TH |
684 | **/ |
685 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
686 | { | |
687 | q->dma_pad_mask = mask; | |
688 | } | |
689 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
690 | ||
27f8221a FT |
691 | /** |
692 | * blk_queue_update_dma_pad - update pad mask | |
693 | * @q: the request queue for the device | |
694 | * @mask: pad mask | |
695 | * | |
696 | * Update dma pad mask. | |
697 | * | |
698 | * Appending pad buffer to a request modifies the last entry of a | |
699 | * scatter list such that it includes the pad buffer. | |
700 | **/ | |
701 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
702 | { | |
703 | if (mask > q->dma_pad_mask) | |
704 | q->dma_pad_mask = mask; | |
705 | } | |
706 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
707 | ||
86db1e29 JA |
708 | /** |
709 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
86db1e29 | 710 | * @q: the request queue for the device |
2fb98e84 | 711 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
86db1e29 JA |
712 | * @buf: physically contiguous buffer |
713 | * @size: size of the buffer in bytes | |
714 | * | |
715 | * Some devices have excess DMA problems and can't simply discard (or | |
716 | * zero fill) the unwanted piece of the transfer. They have to have a | |
717 | * real area of memory to transfer it into. The use case for this is | |
718 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
719 | * bigger than the transfer size some HBAs will lock up if there | |
720 | * aren't DMA elements to contain the excess transfer. What this API | |
721 | * does is adjust the queue so that the buf is always appended | |
722 | * silently to the scatterlist. | |
723 | * | |
724 | * Note: This routine adjusts max_hw_segments to make room for | |
725 | * appending the drain buffer. If you call | |
726 | * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after | |
727 | * calling this routine, you must set the limit to one fewer than your | |
728 | * device can support otherwise there won't be room for the drain | |
729 | * buffer. | |
730 | */ | |
448da4d2 | 731 | int blk_queue_dma_drain(struct request_queue *q, |
2fb98e84 TH |
732 | dma_drain_needed_fn *dma_drain_needed, |
733 | void *buf, unsigned int size) | |
86db1e29 | 734 | { |
ae03bf63 | 735 | if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2) |
86db1e29 JA |
736 | return -EINVAL; |
737 | /* make room for appending the drain */ | |
ae03bf63 MP |
738 | blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1); |
739 | blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1); | |
2fb98e84 | 740 | q->dma_drain_needed = dma_drain_needed; |
86db1e29 JA |
741 | q->dma_drain_buffer = buf; |
742 | q->dma_drain_size = size; | |
743 | ||
744 | return 0; | |
745 | } | |
86db1e29 JA |
746 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
747 | ||
748 | /** | |
749 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
750 | * @q: the request queue for the device | |
751 | * @mask: the memory boundary mask | |
752 | **/ | |
753 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
754 | { | |
755 | if (mask < PAGE_CACHE_SIZE - 1) { | |
756 | mask = PAGE_CACHE_SIZE - 1; | |
24c03d47 HH |
757 | printk(KERN_INFO "%s: set to minimum %lx\n", |
758 | __func__, mask); | |
86db1e29 JA |
759 | } |
760 | ||
025146e1 | 761 | q->limits.seg_boundary_mask = mask; |
86db1e29 | 762 | } |
86db1e29 JA |
763 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
764 | ||
765 | /** | |
766 | * blk_queue_dma_alignment - set dma length and memory alignment | |
767 | * @q: the request queue for the device | |
768 | * @mask: alignment mask | |
769 | * | |
770 | * description: | |
710027a4 | 771 | * set required memory and length alignment for direct dma transactions. |
8feb4d20 | 772 | * this is used when building direct io requests for the queue. |
86db1e29 JA |
773 | * |
774 | **/ | |
775 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
776 | { | |
777 | q->dma_alignment = mask; | |
778 | } | |
86db1e29 JA |
779 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
780 | ||
781 | /** | |
782 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
783 | * @q: the request queue for the device | |
784 | * @mask: alignment mask | |
785 | * | |
786 | * description: | |
710027a4 | 787 | * update required memory and length alignment for direct dma transactions. |
86db1e29 JA |
788 | * If the requested alignment is larger than the current alignment, then |
789 | * the current queue alignment is updated to the new value, otherwise it | |
790 | * is left alone. The design of this is to allow multiple objects | |
791 | * (driver, device, transport etc) to set their respective | |
792 | * alignments without having them interfere. | |
793 | * | |
794 | **/ | |
795 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
796 | { | |
797 | BUG_ON(mask > PAGE_SIZE); | |
798 | ||
799 | if (mask > q->dma_alignment) | |
800 | q->dma_alignment = mask; | |
801 | } | |
86db1e29 JA |
802 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
803 | ||
aeb3d3a8 | 804 | static int __init blk_settings_init(void) |
86db1e29 JA |
805 | { |
806 | blk_max_low_pfn = max_low_pfn - 1; | |
807 | blk_max_pfn = max_pfn - 1; | |
808 | return 0; | |
809 | } | |
810 | subsys_initcall(blk_settings_init); |