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 | ||
39 | /** | |
40 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
41 | * @q: queue | |
42 | * @mbfn: merge_bvec_fn | |
43 | * | |
44 | * Usually queues have static limitations on the max sectors or segments that | |
45 | * we can put in a request. Stacking drivers may have some settings that | |
46 | * are dynamic, and thus we have to query the queue whether it is ok to | |
47 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
48 | * has such limitations, it needs to register a merge_bvec_fn to control | |
49 | * the size of bio's sent to it. Note that a block device *must* allow a | |
50 | * single page to be added to an empty bio. The block device driver may want | |
51 | * to use the bio_split() function to deal with these bio's. By default | |
52 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
53 | * honored. | |
54 | */ | |
55 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) | |
56 | { | |
57 | q->merge_bvec_fn = mbfn; | |
58 | } | |
86db1e29 JA |
59 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
60 | ||
61 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
62 | { | |
63 | q->softirq_done_fn = fn; | |
64 | } | |
86db1e29 JA |
65 | EXPORT_SYMBOL(blk_queue_softirq_done); |
66 | ||
242f9dcb JA |
67 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
68 | { | |
69 | q->rq_timeout = timeout; | |
70 | } | |
71 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); | |
72 | ||
73 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) | |
74 | { | |
75 | q->rq_timed_out_fn = fn; | |
76 | } | |
77 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); | |
78 | ||
ef9e3fac KU |
79 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) |
80 | { | |
81 | q->lld_busy_fn = fn; | |
82 | } | |
83 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); | |
84 | ||
e475bba2 MP |
85 | /** |
86 | * blk_set_default_limits - reset limits to default values | |
f740f5ca | 87 | * @lim: the queue_limits structure to reset |
e475bba2 MP |
88 | * |
89 | * Description: | |
90 | * Returns a queue_limit struct to its default state. Can be used by | |
91 | * stacking drivers like DM that stage table swaps and reuse an | |
92 | * existing device queue. | |
93 | */ | |
94 | void blk_set_default_limits(struct queue_limits *lim) | |
95 | { | |
8a78362c | 96 | lim->max_segments = BLK_MAX_SEGMENTS; |
e475bba2 | 97 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
eb28d31b | 98 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; |
5dee2477 MP |
99 | lim->max_sectors = BLK_DEF_MAX_SECTORS; |
100 | lim->max_hw_sectors = INT_MAX; | |
86b37281 MP |
101 | lim->max_discard_sectors = 0; |
102 | lim->discard_granularity = 0; | |
103 | lim->discard_alignment = 0; | |
104 | lim->discard_misaligned = 0; | |
98262f27 | 105 | lim->discard_zeroes_data = -1; |
e475bba2 | 106 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
3a02c8e8 | 107 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
e475bba2 MP |
108 | lim->alignment_offset = 0; |
109 | lim->io_opt = 0; | |
110 | lim->misaligned = 0; | |
111 | lim->no_cluster = 0; | |
112 | } | |
113 | EXPORT_SYMBOL(blk_set_default_limits); | |
114 | ||
86db1e29 JA |
115 | /** |
116 | * blk_queue_make_request - define an alternate make_request function for a device | |
117 | * @q: the request queue for the device to be affected | |
118 | * @mfn: the alternate make_request function | |
119 | * | |
120 | * Description: | |
121 | * The normal way for &struct bios to be passed to a device | |
122 | * driver is for them to be collected into requests on a request | |
123 | * queue, and then to allow the device driver to select requests | |
124 | * off that queue when it is ready. This works well for many block | |
125 | * devices. However some block devices (typically virtual devices | |
126 | * such as md or lvm) do not benefit from the processing on the | |
127 | * request queue, and are served best by having the requests passed | |
128 | * directly to them. This can be achieved by providing a function | |
129 | * to blk_queue_make_request(). | |
130 | * | |
131 | * Caveat: | |
132 | * The driver that does this *must* be able to deal appropriately | |
133 | * with buffers in "highmemory". This can be accomplished by either calling | |
134 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
135 | * blk_queue_bounce() to create a buffer in normal memory. | |
136 | **/ | |
6728cb0e | 137 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
86db1e29 JA |
138 | { |
139 | /* | |
140 | * set defaults | |
141 | */ | |
142 | q->nr_requests = BLKDEV_MAX_RQ; | |
0e435ac2 | 143 | |
86db1e29 | 144 | q->make_request_fn = mfn; |
86db1e29 JA |
145 | blk_queue_dma_alignment(q, 511); |
146 | blk_queue_congestion_threshold(q); | |
147 | q->nr_batching = BLK_BATCH_REQ; | |
148 | ||
149 | q->unplug_thresh = 4; /* hmm */ | |
ad5ebd2f | 150 | q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */ |
86db1e29 JA |
151 | if (q->unplug_delay == 0) |
152 | q->unplug_delay = 1; | |
153 | ||
86db1e29 JA |
154 | q->unplug_timer.function = blk_unplug_timeout; |
155 | q->unplug_timer.data = (unsigned long)q; | |
156 | ||
e475bba2 | 157 | blk_set_default_limits(&q->limits); |
086fa5ff | 158 | blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS); |
e475bba2 | 159 | |
a4e7d464 JA |
160 | /* |
161 | * If the caller didn't supply a lock, fall back to our embedded | |
162 | * per-queue locks | |
163 | */ | |
164 | if (!q->queue_lock) | |
165 | q->queue_lock = &q->__queue_lock; | |
166 | ||
86db1e29 JA |
167 | /* |
168 | * by default assume old behaviour and bounce for any highmem page | |
169 | */ | |
170 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
171 | } | |
86db1e29 JA |
172 | EXPORT_SYMBOL(blk_queue_make_request); |
173 | ||
174 | /** | |
175 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
cd0aca2d TH |
176 | * @q: the request queue for the device |
177 | * @dma_mask: the maximum address the device can handle | |
86db1e29 JA |
178 | * |
179 | * Description: | |
180 | * Different hardware can have different requirements as to what pages | |
181 | * it can do I/O directly to. A low level driver can call | |
182 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
cd0aca2d | 183 | * buffers for doing I/O to pages residing above @dma_mask. |
86db1e29 | 184 | **/ |
cd0aca2d | 185 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
86db1e29 | 186 | { |
cd0aca2d | 187 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
86db1e29 JA |
188 | int dma = 0; |
189 | ||
190 | q->bounce_gfp = GFP_NOIO; | |
191 | #if BITS_PER_LONG == 64 | |
cd0aca2d TH |
192 | /* |
193 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
194 | * some IOMMUs can handle everything, but I don't know of a | |
195 | * way to test this here. | |
196 | */ | |
197 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
86db1e29 | 198 | dma = 1; |
025146e1 | 199 | q->limits.bounce_pfn = max_low_pfn; |
86db1e29 | 200 | #else |
6728cb0e | 201 | if (b_pfn < blk_max_low_pfn) |
86db1e29 | 202 | dma = 1; |
025146e1 | 203 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
204 | #endif |
205 | if (dma) { | |
206 | init_emergency_isa_pool(); | |
207 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
025146e1 | 208 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
209 | } |
210 | } | |
86db1e29 JA |
211 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
212 | ||
213 | /** | |
086fa5ff | 214 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue |
86db1e29 | 215 | * @q: the request queue for the device |
2800aac1 | 216 | * @max_hw_sectors: max hardware sectors in the usual 512b unit |
86db1e29 JA |
217 | * |
218 | * Description: | |
2800aac1 MP |
219 | * Enables a low level driver to set a hard upper limit, |
220 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by | |
221 | * the device driver based upon the combined capabilities of I/O | |
222 | * controller and storage device. | |
223 | * | |
224 | * max_sectors is a soft limit imposed by the block layer for | |
225 | * filesystem type requests. This value can be overridden on a | |
226 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. | |
227 | * The soft limit can not exceed max_hw_sectors. | |
86db1e29 | 228 | **/ |
086fa5ff | 229 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) |
86db1e29 | 230 | { |
2800aac1 MP |
231 | if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) { |
232 | max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
24c03d47 | 233 | printk(KERN_INFO "%s: set to minimum %d\n", |
2800aac1 | 234 | __func__, max_hw_sectors); |
86db1e29 JA |
235 | } |
236 | ||
2800aac1 MP |
237 | q->limits.max_hw_sectors = max_hw_sectors; |
238 | q->limits.max_sectors = min_t(unsigned int, max_hw_sectors, | |
239 | BLK_DEF_MAX_SECTORS); | |
86db1e29 | 240 | } |
086fa5ff | 241 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
86db1e29 | 242 | |
67efc925 CH |
243 | /** |
244 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
245 | * @q: the request queue for the device | |
c7ebf065 | 246 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
247 | **/ |
248 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
249 | unsigned int max_discard_sectors) | |
250 | { | |
251 | q->limits.max_discard_sectors = max_discard_sectors; | |
252 | } | |
253 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
254 | ||
86db1e29 | 255 | /** |
8a78362c | 256 | * blk_queue_max_segments - set max hw segments for a request for this queue |
86db1e29 JA |
257 | * @q: the request queue for the device |
258 | * @max_segments: max number of segments | |
259 | * | |
260 | * Description: | |
261 | * Enables a low level driver to set an upper limit on the number of | |
8a78362c | 262 | * hw data segments in a request. |
86db1e29 | 263 | **/ |
8a78362c | 264 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) |
86db1e29 JA |
265 | { |
266 | if (!max_segments) { | |
267 | max_segments = 1; | |
24c03d47 HH |
268 | printk(KERN_INFO "%s: set to minimum %d\n", |
269 | __func__, max_segments); | |
86db1e29 JA |
270 | } |
271 | ||
8a78362c | 272 | q->limits.max_segments = max_segments; |
86db1e29 | 273 | } |
8a78362c | 274 | EXPORT_SYMBOL(blk_queue_max_segments); |
86db1e29 JA |
275 | |
276 | /** | |
277 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
278 | * @q: the request queue for the device | |
279 | * @max_size: max size of segment in bytes | |
280 | * | |
281 | * Description: | |
282 | * Enables a low level driver to set an upper limit on the size of a | |
283 | * coalesced segment | |
284 | **/ | |
285 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
286 | { | |
287 | if (max_size < PAGE_CACHE_SIZE) { | |
288 | max_size = PAGE_CACHE_SIZE; | |
24c03d47 HH |
289 | printk(KERN_INFO "%s: set to minimum %d\n", |
290 | __func__, max_size); | |
86db1e29 JA |
291 | } |
292 | ||
025146e1 | 293 | q->limits.max_segment_size = max_size; |
86db1e29 | 294 | } |
86db1e29 JA |
295 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
296 | ||
297 | /** | |
e1defc4f | 298 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 299 | * @q: the request queue for the device |
e1defc4f | 300 | * @size: the logical block size, in bytes |
86db1e29 JA |
301 | * |
302 | * Description: | |
e1defc4f MP |
303 | * This should be set to the lowest possible block size that the |
304 | * storage device can address. The default of 512 covers most | |
305 | * hardware. | |
86db1e29 | 306 | **/ |
e1defc4f | 307 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
86db1e29 | 308 | { |
025146e1 | 309 | q->limits.logical_block_size = size; |
c72758f3 MP |
310 | |
311 | if (q->limits.physical_block_size < size) | |
312 | q->limits.physical_block_size = size; | |
313 | ||
314 | if (q->limits.io_min < q->limits.physical_block_size) | |
315 | q->limits.io_min = q->limits.physical_block_size; | |
86db1e29 | 316 | } |
e1defc4f | 317 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 318 | |
c72758f3 MP |
319 | /** |
320 | * blk_queue_physical_block_size - set physical block size for the queue | |
321 | * @q: the request queue for the device | |
322 | * @size: the physical block size, in bytes | |
323 | * | |
324 | * Description: | |
325 | * This should be set to the lowest possible sector size that the | |
326 | * hardware can operate on without reverting to read-modify-write | |
327 | * operations. | |
328 | */ | |
329 | void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) | |
330 | { | |
331 | q->limits.physical_block_size = size; | |
332 | ||
333 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
334 | q->limits.physical_block_size = q->limits.logical_block_size; | |
335 | ||
336 | if (q->limits.io_min < q->limits.physical_block_size) | |
337 | q->limits.io_min = q->limits.physical_block_size; | |
338 | } | |
339 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
340 | ||
341 | /** | |
342 | * blk_queue_alignment_offset - set physical block alignment offset | |
343 | * @q: the request queue for the device | |
8ebf9756 | 344 | * @offset: alignment offset in bytes |
c72758f3 MP |
345 | * |
346 | * Description: | |
347 | * Some devices are naturally misaligned to compensate for things like | |
348 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
349 | * should call this function for devices whose first sector is not | |
350 | * naturally aligned. | |
351 | */ | |
352 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
353 | { | |
354 | q->limits.alignment_offset = | |
355 | offset & (q->limits.physical_block_size - 1); | |
356 | q->limits.misaligned = 0; | |
357 | } | |
358 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
359 | ||
7c958e32 MP |
360 | /** |
361 | * blk_limits_io_min - set minimum request size for a device | |
362 | * @limits: the queue limits | |
363 | * @min: smallest I/O size in bytes | |
364 | * | |
365 | * Description: | |
366 | * Some devices have an internal block size bigger than the reported | |
367 | * hardware sector size. This function can be used to signal the | |
368 | * smallest I/O the device can perform without incurring a performance | |
369 | * penalty. | |
370 | */ | |
371 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
372 | { | |
373 | limits->io_min = min; | |
374 | ||
375 | if (limits->io_min < limits->logical_block_size) | |
376 | limits->io_min = limits->logical_block_size; | |
377 | ||
378 | if (limits->io_min < limits->physical_block_size) | |
379 | limits->io_min = limits->physical_block_size; | |
380 | } | |
381 | EXPORT_SYMBOL(blk_limits_io_min); | |
382 | ||
c72758f3 MP |
383 | /** |
384 | * blk_queue_io_min - set minimum request size for the queue | |
385 | * @q: the request queue for the device | |
8ebf9756 | 386 | * @min: smallest I/O size in bytes |
c72758f3 MP |
387 | * |
388 | * Description: | |
7e5f5fb0 MP |
389 | * Storage devices may report a granularity or preferred minimum I/O |
390 | * size which is the smallest request the device can perform without | |
391 | * incurring a performance penalty. For disk drives this is often the | |
392 | * physical block size. For RAID arrays it is often the stripe chunk | |
393 | * size. A properly aligned multiple of minimum_io_size is the | |
394 | * preferred request size for workloads where a high number of I/O | |
395 | * operations is desired. | |
c72758f3 MP |
396 | */ |
397 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
398 | { | |
7c958e32 | 399 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
400 | } |
401 | EXPORT_SYMBOL(blk_queue_io_min); | |
402 | ||
3c5820c7 MP |
403 | /** |
404 | * blk_limits_io_opt - set optimal request size for a device | |
405 | * @limits: the queue limits | |
406 | * @opt: smallest I/O size in bytes | |
407 | * | |
408 | * Description: | |
409 | * Storage devices may report an optimal I/O size, which is the | |
410 | * device's preferred unit for sustained I/O. This is rarely reported | |
411 | * for disk drives. For RAID arrays it is usually the stripe width or | |
412 | * the internal track size. A properly aligned multiple of | |
413 | * optimal_io_size is the preferred request size for workloads where | |
414 | * sustained throughput is desired. | |
415 | */ | |
416 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
417 | { | |
418 | limits->io_opt = opt; | |
419 | } | |
420 | EXPORT_SYMBOL(blk_limits_io_opt); | |
421 | ||
c72758f3 MP |
422 | /** |
423 | * blk_queue_io_opt - set optimal request size for the queue | |
424 | * @q: the request queue for the device | |
8ebf9756 | 425 | * @opt: optimal request size in bytes |
c72758f3 MP |
426 | * |
427 | * Description: | |
7e5f5fb0 MP |
428 | * Storage devices may report an optimal I/O size, which is the |
429 | * device's preferred unit for sustained I/O. This is rarely reported | |
430 | * for disk drives. For RAID arrays it is usually the stripe width or | |
431 | * the internal track size. A properly aligned multiple of | |
432 | * optimal_io_size is the preferred request size for workloads where | |
433 | * sustained throughput is desired. | |
c72758f3 MP |
434 | */ |
435 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
436 | { | |
3c5820c7 | 437 | blk_limits_io_opt(&q->limits, opt); |
c72758f3 MP |
438 | } |
439 | EXPORT_SYMBOL(blk_queue_io_opt); | |
440 | ||
86db1e29 JA |
441 | /* |
442 | * Returns the minimum that is _not_ zero, unless both are zero. | |
443 | */ | |
444 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
445 | ||
446 | /** | |
447 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
448 | * @t: the stacking driver (top) | |
449 | * @b: the underlying device (bottom) | |
450 | **/ | |
451 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
452 | { | |
fef24667 | 453 | blk_stack_limits(&t->limits, &b->limits, 0); |
025146e1 | 454 | |
e7e72bf6 NB |
455 | if (!t->queue_lock) |
456 | WARN_ON_ONCE(1); | |
457 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
458 | unsigned long flags; | |
459 | spin_lock_irqsave(t->queue_lock, flags); | |
75ad23bc | 460 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
e7e72bf6 NB |
461 | spin_unlock_irqrestore(t->queue_lock, flags); |
462 | } | |
86db1e29 | 463 | } |
86db1e29 JA |
464 | EXPORT_SYMBOL(blk_queue_stack_limits); |
465 | ||
c72758f3 MP |
466 | /** |
467 | * blk_stack_limits - adjust queue_limits for stacked devices | |
81744ee4 MP |
468 | * @t: the stacking driver limits (top device) |
469 | * @b: the underlying queue limits (bottom, component device) | |
e03a72e1 | 470 | * @start: first data sector within component device |
c72758f3 MP |
471 | * |
472 | * Description: | |
81744ee4 MP |
473 | * This function is used by stacking drivers like MD and DM to ensure |
474 | * that all component devices have compatible block sizes and | |
475 | * alignments. The stacking driver must provide a queue_limits | |
476 | * struct (top) and then iteratively call the stacking function for | |
477 | * all component (bottom) devices. The stacking function will | |
478 | * attempt to combine the values and ensure proper alignment. | |
479 | * | |
480 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
481 | * top device's block sizes and alignment offsets may be adjusted to | |
482 | * ensure alignment with the bottom device. If no compatible sizes | |
483 | * and alignments exist, -1 is returned and the resulting top | |
484 | * queue_limits will have the misaligned flag set to indicate that | |
485 | * the alignment_offset is undefined. | |
c72758f3 MP |
486 | */ |
487 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
e03a72e1 | 488 | sector_t start) |
c72758f3 | 489 | { |
e03a72e1 | 490 | unsigned int top, bottom, alignment, ret = 0; |
86b37281 | 491 | |
c72758f3 MP |
492 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
493 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
77634f33 | 494 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
c72758f3 MP |
495 | |
496 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
497 | b->seg_boundary_mask); | |
498 | ||
8a78362c | 499 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
c72758f3 MP |
500 | |
501 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
502 | b->max_segment_size); | |
503 | ||
fe0b393f MP |
504 | t->misaligned |= b->misaligned; |
505 | ||
e03a72e1 | 506 | alignment = queue_limit_alignment_offset(b, start); |
9504e086 | 507 | |
81744ee4 MP |
508 | /* Bottom device has different alignment. Check that it is |
509 | * compatible with the current top alignment. | |
510 | */ | |
9504e086 MP |
511 | if (t->alignment_offset != alignment) { |
512 | ||
513 | top = max(t->physical_block_size, t->io_min) | |
514 | + t->alignment_offset; | |
81744ee4 | 515 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
9504e086 | 516 | |
81744ee4 | 517 | /* Verify that top and bottom intervals line up */ |
fe0b393f | 518 | if (max(top, bottom) & (min(top, bottom) - 1)) { |
9504e086 | 519 | t->misaligned = 1; |
fe0b393f MP |
520 | ret = -1; |
521 | } | |
9504e086 MP |
522 | } |
523 | ||
c72758f3 MP |
524 | t->logical_block_size = max(t->logical_block_size, |
525 | b->logical_block_size); | |
526 | ||
527 | t->physical_block_size = max(t->physical_block_size, | |
528 | b->physical_block_size); | |
529 | ||
530 | t->io_min = max(t->io_min, b->io_min); | |
9504e086 MP |
531 | t->io_opt = lcm(t->io_opt, b->io_opt); |
532 | ||
c72758f3 | 533 | t->no_cluster |= b->no_cluster; |
98262f27 | 534 | t->discard_zeroes_data &= b->discard_zeroes_data; |
c72758f3 | 535 | |
81744ee4 | 536 | /* Physical block size a multiple of the logical block size? */ |
9504e086 MP |
537 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
538 | t->physical_block_size = t->logical_block_size; | |
c72758f3 | 539 | t->misaligned = 1; |
fe0b393f | 540 | ret = -1; |
86b37281 MP |
541 | } |
542 | ||
81744ee4 | 543 | /* Minimum I/O a multiple of the physical block size? */ |
9504e086 MP |
544 | if (t->io_min & (t->physical_block_size - 1)) { |
545 | t->io_min = t->physical_block_size; | |
546 | t->misaligned = 1; | |
fe0b393f | 547 | ret = -1; |
c72758f3 MP |
548 | } |
549 | ||
81744ee4 | 550 | /* Optimal I/O a multiple of the physical block size? */ |
9504e086 MP |
551 | if (t->io_opt & (t->physical_block_size - 1)) { |
552 | t->io_opt = 0; | |
553 | t->misaligned = 1; | |
fe0b393f | 554 | ret = -1; |
9504e086 | 555 | } |
c72758f3 | 556 | |
81744ee4 | 557 | /* Find lowest common alignment_offset */ |
9504e086 MP |
558 | t->alignment_offset = lcm(t->alignment_offset, alignment) |
559 | & (max(t->physical_block_size, t->io_min) - 1); | |
86b37281 | 560 | |
81744ee4 | 561 | /* Verify that new alignment_offset is on a logical block boundary */ |
fe0b393f | 562 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
c72758f3 | 563 | t->misaligned = 1; |
fe0b393f MP |
564 | ret = -1; |
565 | } | |
c72758f3 | 566 | |
9504e086 MP |
567 | /* Discard alignment and granularity */ |
568 | if (b->discard_granularity) { | |
e03a72e1 | 569 | alignment = queue_limit_discard_alignment(b, start); |
9504e086 MP |
570 | |
571 | if (t->discard_granularity != 0 && | |
572 | t->discard_alignment != alignment) { | |
573 | top = t->discard_granularity + t->discard_alignment; | |
574 | bottom = b->discard_granularity + alignment; | |
70dd5bf3 | 575 | |
9504e086 MP |
576 | /* Verify that top and bottom intervals line up */ |
577 | if (max(top, bottom) & (min(top, bottom) - 1)) | |
578 | t->discard_misaligned = 1; | |
579 | } | |
580 | ||
81744ee4 MP |
581 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
582 | b->max_discard_sectors); | |
9504e086 MP |
583 | t->discard_granularity = max(t->discard_granularity, |
584 | b->discard_granularity); | |
585 | t->discard_alignment = lcm(t->discard_alignment, alignment) & | |
586 | (t->discard_granularity - 1); | |
587 | } | |
70dd5bf3 | 588 | |
fe0b393f | 589 | return ret; |
c72758f3 | 590 | } |
5d85d324 | 591 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 | 592 | |
17be8c24 MP |
593 | /** |
594 | * bdev_stack_limits - adjust queue limits for stacked drivers | |
595 | * @t: the stacking driver limits (top device) | |
596 | * @bdev: the component block_device (bottom) | |
597 | * @start: first data sector within component device | |
598 | * | |
599 | * Description: | |
600 | * Merges queue limits for a top device and a block_device. Returns | |
601 | * 0 if alignment didn't change. Returns -1 if adding the bottom | |
602 | * device caused misalignment. | |
603 | */ | |
604 | int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, | |
605 | sector_t start) | |
606 | { | |
607 | struct request_queue *bq = bdev_get_queue(bdev); | |
608 | ||
609 | start += get_start_sect(bdev); | |
610 | ||
e03a72e1 | 611 | return blk_stack_limits(t, &bq->limits, start); |
17be8c24 MP |
612 | } |
613 | EXPORT_SYMBOL(bdev_stack_limits); | |
614 | ||
c72758f3 MP |
615 | /** |
616 | * disk_stack_limits - adjust queue limits for stacked drivers | |
77634f33 | 617 | * @disk: MD/DM gendisk (top) |
c72758f3 MP |
618 | * @bdev: the underlying block device (bottom) |
619 | * @offset: offset to beginning of data within component device | |
620 | * | |
621 | * Description: | |
e03a72e1 MP |
622 | * Merges the limits for a top level gendisk and a bottom level |
623 | * block_device. | |
c72758f3 MP |
624 | */ |
625 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
626 | sector_t offset) | |
627 | { | |
628 | struct request_queue *t = disk->queue; | |
629 | struct request_queue *b = bdev_get_queue(bdev); | |
630 | ||
e03a72e1 | 631 | if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { |
c72758f3 MP |
632 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; |
633 | ||
634 | disk_name(disk, 0, top); | |
635 | bdevname(bdev, bottom); | |
636 | ||
637 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
638 | top, bottom); | |
639 | } | |
640 | ||
641 | if (!t->queue_lock) | |
642 | WARN_ON_ONCE(1); | |
643 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
644 | unsigned long flags; | |
645 | ||
646 | spin_lock_irqsave(t->queue_lock, flags); | |
647 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) | |
648 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); | |
649 | spin_unlock_irqrestore(t->queue_lock, flags); | |
650 | } | |
651 | } | |
652 | EXPORT_SYMBOL(disk_stack_limits); | |
653 | ||
e3790c7d TH |
654 | /** |
655 | * blk_queue_dma_pad - set pad mask | |
656 | * @q: the request queue for the device | |
657 | * @mask: pad mask | |
658 | * | |
27f8221a | 659 | * Set dma pad mask. |
e3790c7d | 660 | * |
27f8221a FT |
661 | * Appending pad buffer to a request modifies the last entry of a |
662 | * scatter list such that it includes the pad buffer. | |
e3790c7d TH |
663 | **/ |
664 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
665 | { | |
666 | q->dma_pad_mask = mask; | |
667 | } | |
668 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
669 | ||
27f8221a FT |
670 | /** |
671 | * blk_queue_update_dma_pad - update pad mask | |
672 | * @q: the request queue for the device | |
673 | * @mask: pad mask | |
674 | * | |
675 | * Update dma pad mask. | |
676 | * | |
677 | * Appending pad buffer to a request modifies the last entry of a | |
678 | * scatter list such that it includes the pad buffer. | |
679 | **/ | |
680 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
681 | { | |
682 | if (mask > q->dma_pad_mask) | |
683 | q->dma_pad_mask = mask; | |
684 | } | |
685 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
686 | ||
86db1e29 JA |
687 | /** |
688 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
86db1e29 | 689 | * @q: the request queue for the device |
2fb98e84 | 690 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
86db1e29 JA |
691 | * @buf: physically contiguous buffer |
692 | * @size: size of the buffer in bytes | |
693 | * | |
694 | * Some devices have excess DMA problems and can't simply discard (or | |
695 | * zero fill) the unwanted piece of the transfer. They have to have a | |
696 | * real area of memory to transfer it into. The use case for this is | |
697 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
698 | * bigger than the transfer size some HBAs will lock up if there | |
699 | * aren't DMA elements to contain the excess transfer. What this API | |
700 | * does is adjust the queue so that the buf is always appended | |
701 | * silently to the scatterlist. | |
702 | * | |
8a78362c MP |
703 | * Note: This routine adjusts max_hw_segments to make room for appending |
704 | * the drain buffer. If you call blk_queue_max_segments() after calling | |
705 | * this routine, you must set the limit to one fewer than your device | |
706 | * can support otherwise there won't be room for the drain buffer. | |
86db1e29 | 707 | */ |
448da4d2 | 708 | int blk_queue_dma_drain(struct request_queue *q, |
2fb98e84 TH |
709 | dma_drain_needed_fn *dma_drain_needed, |
710 | void *buf, unsigned int size) | |
86db1e29 | 711 | { |
8a78362c | 712 | if (queue_max_segments(q) < 2) |
86db1e29 JA |
713 | return -EINVAL; |
714 | /* make room for appending the drain */ | |
8a78362c | 715 | blk_queue_max_segments(q, queue_max_segments(q) - 1); |
2fb98e84 | 716 | q->dma_drain_needed = dma_drain_needed; |
86db1e29 JA |
717 | q->dma_drain_buffer = buf; |
718 | q->dma_drain_size = size; | |
719 | ||
720 | return 0; | |
721 | } | |
86db1e29 JA |
722 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
723 | ||
724 | /** | |
725 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
726 | * @q: the request queue for the device | |
727 | * @mask: the memory boundary mask | |
728 | **/ | |
729 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
730 | { | |
731 | if (mask < PAGE_CACHE_SIZE - 1) { | |
732 | mask = PAGE_CACHE_SIZE - 1; | |
24c03d47 HH |
733 | printk(KERN_INFO "%s: set to minimum %lx\n", |
734 | __func__, mask); | |
86db1e29 JA |
735 | } |
736 | ||
025146e1 | 737 | q->limits.seg_boundary_mask = mask; |
86db1e29 | 738 | } |
86db1e29 JA |
739 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
740 | ||
741 | /** | |
742 | * blk_queue_dma_alignment - set dma length and memory alignment | |
743 | * @q: the request queue for the device | |
744 | * @mask: alignment mask | |
745 | * | |
746 | * description: | |
710027a4 | 747 | * set required memory and length alignment for direct dma transactions. |
8feb4d20 | 748 | * this is used when building direct io requests for the queue. |
86db1e29 JA |
749 | * |
750 | **/ | |
751 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
752 | { | |
753 | q->dma_alignment = mask; | |
754 | } | |
86db1e29 JA |
755 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
756 | ||
757 | /** | |
758 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
759 | * @q: the request queue for the device | |
760 | * @mask: alignment mask | |
761 | * | |
762 | * description: | |
710027a4 | 763 | * update required memory and length alignment for direct dma transactions. |
86db1e29 JA |
764 | * If the requested alignment is larger than the current alignment, then |
765 | * the current queue alignment is updated to the new value, otherwise it | |
766 | * is left alone. The design of this is to allow multiple objects | |
767 | * (driver, device, transport etc) to set their respective | |
768 | * alignments without having them interfere. | |
769 | * | |
770 | **/ | |
771 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
772 | { | |
773 | BUG_ON(mask > PAGE_SIZE); | |
774 | ||
775 | if (mask > q->dma_alignment) | |
776 | q->dma_alignment = mask; | |
777 | } | |
86db1e29 JA |
778 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
779 | ||
aeb3d3a8 | 780 | static int __init blk_settings_init(void) |
86db1e29 JA |
781 | { |
782 | blk_max_low_pfn = max_low_pfn - 1; | |
783 | blk_max_pfn = max_pfn - 1; | |
784 | return 0; | |
785 | } | |
786 | subsys_initcall(blk_settings_init); |