Commit | Line | Data |
---|---|---|
86db1e29 JA |
1 | /* |
2 | * Functions related to setting various queue properties from drivers | |
3 | */ | |
4 | #include <linux/kernel.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/init.h> | |
7 | #include <linux/bio.h> | |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | |
70dd5bf3 | 10 | #include <linux/gcd.h> |
2cda2728 | 11 | #include <linux/lcm.h> |
ad5ebd2f | 12 | #include <linux/jiffies.h> |
5a0e3ad6 | 13 | #include <linux/gfp.h> |
86db1e29 JA |
14 | |
15 | #include "blk.h" | |
16 | ||
6728cb0e | 17 | unsigned long blk_max_low_pfn; |
86db1e29 | 18 | EXPORT_SYMBOL(blk_max_low_pfn); |
6728cb0e JA |
19 | |
20 | unsigned long blk_max_pfn; | |
86db1e29 JA |
21 | |
22 | /** | |
23 | * blk_queue_prep_rq - set a prepare_request function for queue | |
24 | * @q: queue | |
25 | * @pfn: prepare_request function | |
26 | * | |
27 | * It's possible for a queue to register a prepare_request callback which | |
28 | * is invoked before the request is handed to the request_fn. The goal of | |
29 | * the function is to prepare a request for I/O, it can be used to build a | |
30 | * cdb from the request data for instance. | |
31 | * | |
32 | */ | |
33 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) | |
34 | { | |
35 | q->prep_rq_fn = pfn; | |
36 | } | |
86db1e29 JA |
37 | EXPORT_SYMBOL(blk_queue_prep_rq); |
38 | ||
28018c24 JB |
39 | /** |
40 | * blk_queue_unprep_rq - set an unprepare_request function for queue | |
41 | * @q: queue | |
42 | * @ufn: unprepare_request function | |
43 | * | |
44 | * It's possible for a queue to register an unprepare_request callback | |
45 | * which is invoked before the request is finally completed. The goal | |
46 | * of the function is to deallocate any data that was allocated in the | |
47 | * prepare_request callback. | |
48 | * | |
49 | */ | |
50 | void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn) | |
51 | { | |
52 | q->unprep_rq_fn = ufn; | |
53 | } | |
54 | EXPORT_SYMBOL(blk_queue_unprep_rq); | |
55 | ||
86db1e29 JA |
56 | /** |
57 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
58 | * @q: queue | |
59 | * @mbfn: merge_bvec_fn | |
60 | * | |
61 | * Usually queues have static limitations on the max sectors or segments that | |
62 | * we can put in a request. Stacking drivers may have some settings that | |
63 | * are dynamic, and thus we have to query the queue whether it is ok to | |
64 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
65 | * has such limitations, it needs to register a merge_bvec_fn to control | |
66 | * the size of bio's sent to it. Note that a block device *must* allow a | |
67 | * single page to be added to an empty bio. The block device driver may want | |
68 | * to use the bio_split() function to deal with these bio's. By default | |
69 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
70 | * honored. | |
71 | */ | |
72 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) | |
73 | { | |
74 | q->merge_bvec_fn = mbfn; | |
75 | } | |
86db1e29 JA |
76 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
77 | ||
78 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
79 | { | |
80 | q->softirq_done_fn = fn; | |
81 | } | |
86db1e29 JA |
82 | EXPORT_SYMBOL(blk_queue_softirq_done); |
83 | ||
242f9dcb JA |
84 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
85 | { | |
86 | q->rq_timeout = timeout; | |
87 | } | |
88 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); | |
89 | ||
90 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) | |
91 | { | |
92 | q->rq_timed_out_fn = fn; | |
93 | } | |
94 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); | |
95 | ||
ef9e3fac KU |
96 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) |
97 | { | |
98 | q->lld_busy_fn = fn; | |
99 | } | |
100 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); | |
101 | ||
e475bba2 MP |
102 | /** |
103 | * blk_set_default_limits - reset limits to default values | |
f740f5ca | 104 | * @lim: the queue_limits structure to reset |
e475bba2 MP |
105 | * |
106 | * Description: | |
107 | * Returns a queue_limit struct to its default state. Can be used by | |
108 | * stacking drivers like DM that stage table swaps and reuse an | |
109 | * existing device queue. | |
110 | */ | |
111 | void blk_set_default_limits(struct queue_limits *lim) | |
112 | { | |
8a78362c | 113 | lim->max_segments = BLK_MAX_SEGMENTS; |
e475bba2 | 114 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
eb28d31b | 115 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; |
5dee2477 MP |
116 | lim->max_sectors = BLK_DEF_MAX_SECTORS; |
117 | lim->max_hw_sectors = INT_MAX; | |
86b37281 MP |
118 | lim->max_discard_sectors = 0; |
119 | lim->discard_granularity = 0; | |
120 | lim->discard_alignment = 0; | |
121 | lim->discard_misaligned = 0; | |
98262f27 | 122 | lim->discard_zeroes_data = -1; |
e475bba2 | 123 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
3a02c8e8 | 124 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
e475bba2 MP |
125 | lim->alignment_offset = 0; |
126 | lim->io_opt = 0; | |
127 | lim->misaligned = 0; | |
128 | lim->no_cluster = 0; | |
129 | } | |
130 | EXPORT_SYMBOL(blk_set_default_limits); | |
131 | ||
86db1e29 JA |
132 | /** |
133 | * blk_queue_make_request - define an alternate make_request function for a device | |
134 | * @q: the request queue for the device to be affected | |
135 | * @mfn: the alternate make_request function | |
136 | * | |
137 | * Description: | |
138 | * The normal way for &struct bios to be passed to a device | |
139 | * driver is for them to be collected into requests on a request | |
140 | * queue, and then to allow the device driver to select requests | |
141 | * off that queue when it is ready. This works well for many block | |
142 | * devices. However some block devices (typically virtual devices | |
143 | * such as md or lvm) do not benefit from the processing on the | |
144 | * request queue, and are served best by having the requests passed | |
145 | * directly to them. This can be achieved by providing a function | |
146 | * to blk_queue_make_request(). | |
147 | * | |
148 | * Caveat: | |
149 | * The driver that does this *must* be able to deal appropriately | |
150 | * with buffers in "highmemory". This can be accomplished by either calling | |
151 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
152 | * blk_queue_bounce() to create a buffer in normal memory. | |
153 | **/ | |
6728cb0e | 154 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
86db1e29 JA |
155 | { |
156 | /* | |
157 | * set defaults | |
158 | */ | |
159 | q->nr_requests = BLKDEV_MAX_RQ; | |
0e435ac2 | 160 | |
86db1e29 | 161 | q->make_request_fn = mfn; |
86db1e29 JA |
162 | blk_queue_dma_alignment(q, 511); |
163 | blk_queue_congestion_threshold(q); | |
164 | q->nr_batching = BLK_BATCH_REQ; | |
165 | ||
166 | q->unplug_thresh = 4; /* hmm */ | |
ad5ebd2f | 167 | q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */ |
86db1e29 JA |
168 | if (q->unplug_delay == 0) |
169 | q->unplug_delay = 1; | |
170 | ||
86db1e29 JA |
171 | q->unplug_timer.function = blk_unplug_timeout; |
172 | q->unplug_timer.data = (unsigned long)q; | |
173 | ||
e475bba2 | 174 | blk_set_default_limits(&q->limits); |
086fa5ff | 175 | blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS); |
e475bba2 | 176 | |
a4e7d464 JA |
177 | /* |
178 | * If the caller didn't supply a lock, fall back to our embedded | |
179 | * per-queue locks | |
180 | */ | |
181 | if (!q->queue_lock) | |
182 | q->queue_lock = &q->__queue_lock; | |
183 | ||
86db1e29 JA |
184 | /* |
185 | * by default assume old behaviour and bounce for any highmem page | |
186 | */ | |
187 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
188 | } | |
86db1e29 JA |
189 | EXPORT_SYMBOL(blk_queue_make_request); |
190 | ||
191 | /** | |
192 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
cd0aca2d TH |
193 | * @q: the request queue for the device |
194 | * @dma_mask: the maximum address the device can handle | |
86db1e29 JA |
195 | * |
196 | * Description: | |
197 | * Different hardware can have different requirements as to what pages | |
198 | * it can do I/O directly to. A low level driver can call | |
199 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
cd0aca2d | 200 | * buffers for doing I/O to pages residing above @dma_mask. |
86db1e29 | 201 | **/ |
cd0aca2d | 202 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
86db1e29 | 203 | { |
cd0aca2d | 204 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
86db1e29 JA |
205 | int dma = 0; |
206 | ||
207 | q->bounce_gfp = GFP_NOIO; | |
208 | #if BITS_PER_LONG == 64 | |
cd0aca2d TH |
209 | /* |
210 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
211 | * some IOMMUs can handle everything, but I don't know of a | |
212 | * way to test this here. | |
213 | */ | |
214 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
86db1e29 | 215 | dma = 1; |
025146e1 | 216 | q->limits.bounce_pfn = max_low_pfn; |
86db1e29 | 217 | #else |
6728cb0e | 218 | if (b_pfn < blk_max_low_pfn) |
86db1e29 | 219 | dma = 1; |
025146e1 | 220 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
221 | #endif |
222 | if (dma) { | |
223 | init_emergency_isa_pool(); | |
224 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
025146e1 | 225 | q->limits.bounce_pfn = b_pfn; |
86db1e29 JA |
226 | } |
227 | } | |
86db1e29 JA |
228 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
229 | ||
230 | /** | |
086fa5ff | 231 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue |
86db1e29 | 232 | * @q: the request queue for the device |
2800aac1 | 233 | * @max_hw_sectors: max hardware sectors in the usual 512b unit |
86db1e29 JA |
234 | * |
235 | * Description: | |
2800aac1 MP |
236 | * Enables a low level driver to set a hard upper limit, |
237 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by | |
238 | * the device driver based upon the combined capabilities of I/O | |
239 | * controller and storage device. | |
240 | * | |
241 | * max_sectors is a soft limit imposed by the block layer for | |
242 | * filesystem type requests. This value can be overridden on a | |
243 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. | |
244 | * The soft limit can not exceed max_hw_sectors. | |
86db1e29 | 245 | **/ |
086fa5ff | 246 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) |
86db1e29 | 247 | { |
2800aac1 MP |
248 | if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) { |
249 | max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
24c03d47 | 250 | printk(KERN_INFO "%s: set to minimum %d\n", |
2800aac1 | 251 | __func__, max_hw_sectors); |
86db1e29 JA |
252 | } |
253 | ||
2800aac1 MP |
254 | q->limits.max_hw_sectors = max_hw_sectors; |
255 | q->limits.max_sectors = min_t(unsigned int, max_hw_sectors, | |
256 | BLK_DEF_MAX_SECTORS); | |
86db1e29 | 257 | } |
086fa5ff | 258 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
86db1e29 | 259 | |
67efc925 CH |
260 | /** |
261 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
262 | * @q: the request queue for the device | |
c7ebf065 | 263 | * @max_discard_sectors: maximum number of sectors to discard |
67efc925 CH |
264 | **/ |
265 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
266 | unsigned int max_discard_sectors) | |
267 | { | |
268 | q->limits.max_discard_sectors = max_discard_sectors; | |
269 | } | |
270 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
271 | ||
86db1e29 | 272 | /** |
8a78362c | 273 | * blk_queue_max_segments - set max hw segments for a request for this queue |
86db1e29 JA |
274 | * @q: the request queue for the device |
275 | * @max_segments: max number of segments | |
276 | * | |
277 | * Description: | |
278 | * Enables a low level driver to set an upper limit on the number of | |
8a78362c | 279 | * hw data segments in a request. |
86db1e29 | 280 | **/ |
8a78362c | 281 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) |
86db1e29 JA |
282 | { |
283 | if (!max_segments) { | |
284 | max_segments = 1; | |
24c03d47 HH |
285 | printk(KERN_INFO "%s: set to minimum %d\n", |
286 | __func__, max_segments); | |
86db1e29 JA |
287 | } |
288 | ||
8a78362c | 289 | q->limits.max_segments = max_segments; |
86db1e29 | 290 | } |
8a78362c | 291 | EXPORT_SYMBOL(blk_queue_max_segments); |
86db1e29 JA |
292 | |
293 | /** | |
294 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
295 | * @q: the request queue for the device | |
296 | * @max_size: max size of segment in bytes | |
297 | * | |
298 | * Description: | |
299 | * Enables a low level driver to set an upper limit on the size of a | |
300 | * coalesced segment | |
301 | **/ | |
302 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
303 | { | |
304 | if (max_size < PAGE_CACHE_SIZE) { | |
305 | max_size = PAGE_CACHE_SIZE; | |
24c03d47 HH |
306 | printk(KERN_INFO "%s: set to minimum %d\n", |
307 | __func__, max_size); | |
86db1e29 JA |
308 | } |
309 | ||
025146e1 | 310 | q->limits.max_segment_size = max_size; |
86db1e29 | 311 | } |
86db1e29 JA |
312 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
313 | ||
314 | /** | |
e1defc4f | 315 | * blk_queue_logical_block_size - set logical block size for the queue |
86db1e29 | 316 | * @q: the request queue for the device |
e1defc4f | 317 | * @size: the logical block size, in bytes |
86db1e29 JA |
318 | * |
319 | * Description: | |
e1defc4f MP |
320 | * This should be set to the lowest possible block size that the |
321 | * storage device can address. The default of 512 covers most | |
322 | * hardware. | |
86db1e29 | 323 | **/ |
e1defc4f | 324 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
86db1e29 | 325 | { |
025146e1 | 326 | q->limits.logical_block_size = size; |
c72758f3 MP |
327 | |
328 | if (q->limits.physical_block_size < size) | |
329 | q->limits.physical_block_size = size; | |
330 | ||
331 | if (q->limits.io_min < q->limits.physical_block_size) | |
332 | q->limits.io_min = q->limits.physical_block_size; | |
86db1e29 | 333 | } |
e1defc4f | 334 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
86db1e29 | 335 | |
c72758f3 MP |
336 | /** |
337 | * blk_queue_physical_block_size - set physical block size for the queue | |
338 | * @q: the request queue for the device | |
339 | * @size: the physical block size, in bytes | |
340 | * | |
341 | * Description: | |
342 | * This should be set to the lowest possible sector size that the | |
343 | * hardware can operate on without reverting to read-modify-write | |
344 | * operations. | |
345 | */ | |
346 | void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) | |
347 | { | |
348 | q->limits.physical_block_size = size; | |
349 | ||
350 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
351 | q->limits.physical_block_size = q->limits.logical_block_size; | |
352 | ||
353 | if (q->limits.io_min < q->limits.physical_block_size) | |
354 | q->limits.io_min = q->limits.physical_block_size; | |
355 | } | |
356 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
357 | ||
358 | /** | |
359 | * blk_queue_alignment_offset - set physical block alignment offset | |
360 | * @q: the request queue for the device | |
8ebf9756 | 361 | * @offset: alignment offset in bytes |
c72758f3 MP |
362 | * |
363 | * Description: | |
364 | * Some devices are naturally misaligned to compensate for things like | |
365 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
366 | * should call this function for devices whose first sector is not | |
367 | * naturally aligned. | |
368 | */ | |
369 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
370 | { | |
371 | q->limits.alignment_offset = | |
372 | offset & (q->limits.physical_block_size - 1); | |
373 | q->limits.misaligned = 0; | |
374 | } | |
375 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
376 | ||
7c958e32 MP |
377 | /** |
378 | * blk_limits_io_min - set minimum request size for a device | |
379 | * @limits: the queue limits | |
380 | * @min: smallest I/O size in bytes | |
381 | * | |
382 | * Description: | |
383 | * Some devices have an internal block size bigger than the reported | |
384 | * hardware sector size. This function can be used to signal the | |
385 | * smallest I/O the device can perform without incurring a performance | |
386 | * penalty. | |
387 | */ | |
388 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
389 | { | |
390 | limits->io_min = min; | |
391 | ||
392 | if (limits->io_min < limits->logical_block_size) | |
393 | limits->io_min = limits->logical_block_size; | |
394 | ||
395 | if (limits->io_min < limits->physical_block_size) | |
396 | limits->io_min = limits->physical_block_size; | |
397 | } | |
398 | EXPORT_SYMBOL(blk_limits_io_min); | |
399 | ||
c72758f3 MP |
400 | /** |
401 | * blk_queue_io_min - set minimum request size for the queue | |
402 | * @q: the request queue for the device | |
8ebf9756 | 403 | * @min: smallest I/O size in bytes |
c72758f3 MP |
404 | * |
405 | * Description: | |
7e5f5fb0 MP |
406 | * Storage devices may report a granularity or preferred minimum I/O |
407 | * size which is the smallest request the device can perform without | |
408 | * incurring a performance penalty. For disk drives this is often the | |
409 | * physical block size. For RAID arrays it is often the stripe chunk | |
410 | * size. A properly aligned multiple of minimum_io_size is the | |
411 | * preferred request size for workloads where a high number of I/O | |
412 | * operations is desired. | |
c72758f3 MP |
413 | */ |
414 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
415 | { | |
7c958e32 | 416 | blk_limits_io_min(&q->limits, min); |
c72758f3 MP |
417 | } |
418 | EXPORT_SYMBOL(blk_queue_io_min); | |
419 | ||
3c5820c7 MP |
420 | /** |
421 | * blk_limits_io_opt - set optimal request size for a device | |
422 | * @limits: the queue limits | |
423 | * @opt: smallest I/O size in bytes | |
424 | * | |
425 | * Description: | |
426 | * Storage devices may report an optimal I/O size, which is the | |
427 | * device's preferred unit for sustained I/O. This is rarely reported | |
428 | * for disk drives. For RAID arrays it is usually the stripe width or | |
429 | * the internal track size. A properly aligned multiple of | |
430 | * optimal_io_size is the preferred request size for workloads where | |
431 | * sustained throughput is desired. | |
432 | */ | |
433 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
434 | { | |
435 | limits->io_opt = opt; | |
436 | } | |
437 | EXPORT_SYMBOL(blk_limits_io_opt); | |
438 | ||
c72758f3 MP |
439 | /** |
440 | * blk_queue_io_opt - set optimal request size for the queue | |
441 | * @q: the request queue for the device | |
8ebf9756 | 442 | * @opt: optimal request size in bytes |
c72758f3 MP |
443 | * |
444 | * Description: | |
7e5f5fb0 MP |
445 | * Storage devices may report an optimal I/O size, which is the |
446 | * device's preferred unit for sustained I/O. This is rarely reported | |
447 | * for disk drives. For RAID arrays it is usually the stripe width or | |
448 | * the internal track size. A properly aligned multiple of | |
449 | * optimal_io_size is the preferred request size for workloads where | |
450 | * sustained throughput is desired. | |
c72758f3 MP |
451 | */ |
452 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
453 | { | |
3c5820c7 | 454 | blk_limits_io_opt(&q->limits, opt); |
c72758f3 MP |
455 | } |
456 | EXPORT_SYMBOL(blk_queue_io_opt); | |
457 | ||
86db1e29 JA |
458 | /* |
459 | * Returns the minimum that is _not_ zero, unless both are zero. | |
460 | */ | |
461 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
462 | ||
463 | /** | |
464 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
465 | * @t: the stacking driver (top) | |
466 | * @b: the underlying device (bottom) | |
467 | **/ | |
468 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
469 | { | |
fef24667 | 470 | blk_stack_limits(&t->limits, &b->limits, 0); |
025146e1 | 471 | |
e7e72bf6 NB |
472 | if (!t->queue_lock) |
473 | WARN_ON_ONCE(1); | |
474 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
475 | unsigned long flags; | |
476 | spin_lock_irqsave(t->queue_lock, flags); | |
75ad23bc | 477 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
e7e72bf6 NB |
478 | spin_unlock_irqrestore(t->queue_lock, flags); |
479 | } | |
86db1e29 | 480 | } |
86db1e29 JA |
481 | EXPORT_SYMBOL(blk_queue_stack_limits); |
482 | ||
c72758f3 MP |
483 | /** |
484 | * blk_stack_limits - adjust queue_limits for stacked devices | |
81744ee4 MP |
485 | * @t: the stacking driver limits (top device) |
486 | * @b: the underlying queue limits (bottom, component device) | |
e03a72e1 | 487 | * @start: first data sector within component device |
c72758f3 MP |
488 | * |
489 | * Description: | |
81744ee4 MP |
490 | * This function is used by stacking drivers like MD and DM to ensure |
491 | * that all component devices have compatible block sizes and | |
492 | * alignments. The stacking driver must provide a queue_limits | |
493 | * struct (top) and then iteratively call the stacking function for | |
494 | * all component (bottom) devices. The stacking function will | |
495 | * attempt to combine the values and ensure proper alignment. | |
496 | * | |
497 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
498 | * top device's block sizes and alignment offsets may be adjusted to | |
499 | * ensure alignment with the bottom device. If no compatible sizes | |
500 | * and alignments exist, -1 is returned and the resulting top | |
501 | * queue_limits will have the misaligned flag set to indicate that | |
502 | * the alignment_offset is undefined. | |
c72758f3 MP |
503 | */ |
504 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
e03a72e1 | 505 | sector_t start) |
c72758f3 | 506 | { |
e03a72e1 | 507 | unsigned int top, bottom, alignment, ret = 0; |
86b37281 | 508 | |
c72758f3 MP |
509 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
510 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
77634f33 | 511 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
c72758f3 MP |
512 | |
513 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
514 | b->seg_boundary_mask); | |
515 | ||
8a78362c | 516 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); |
c72758f3 MP |
517 | |
518 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
519 | b->max_segment_size); | |
520 | ||
fe0b393f MP |
521 | t->misaligned |= b->misaligned; |
522 | ||
e03a72e1 | 523 | alignment = queue_limit_alignment_offset(b, start); |
9504e086 | 524 | |
81744ee4 MP |
525 | /* Bottom device has different alignment. Check that it is |
526 | * compatible with the current top alignment. | |
527 | */ | |
9504e086 MP |
528 | if (t->alignment_offset != alignment) { |
529 | ||
530 | top = max(t->physical_block_size, t->io_min) | |
531 | + t->alignment_offset; | |
81744ee4 | 532 | bottom = max(b->physical_block_size, b->io_min) + alignment; |
9504e086 | 533 | |
81744ee4 | 534 | /* Verify that top and bottom intervals line up */ |
fe0b393f | 535 | if (max(top, bottom) & (min(top, bottom) - 1)) { |
9504e086 | 536 | t->misaligned = 1; |
fe0b393f MP |
537 | ret = -1; |
538 | } | |
9504e086 MP |
539 | } |
540 | ||
c72758f3 MP |
541 | t->logical_block_size = max(t->logical_block_size, |
542 | b->logical_block_size); | |
543 | ||
544 | t->physical_block_size = max(t->physical_block_size, | |
545 | b->physical_block_size); | |
546 | ||
547 | t->io_min = max(t->io_min, b->io_min); | |
9504e086 MP |
548 | t->io_opt = lcm(t->io_opt, b->io_opt); |
549 | ||
c72758f3 | 550 | t->no_cluster |= b->no_cluster; |
98262f27 | 551 | t->discard_zeroes_data &= b->discard_zeroes_data; |
c72758f3 | 552 | |
81744ee4 | 553 | /* Physical block size a multiple of the logical block size? */ |
9504e086 MP |
554 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
555 | t->physical_block_size = t->logical_block_size; | |
c72758f3 | 556 | t->misaligned = 1; |
fe0b393f | 557 | ret = -1; |
86b37281 MP |
558 | } |
559 | ||
81744ee4 | 560 | /* Minimum I/O a multiple of the physical block size? */ |
9504e086 MP |
561 | if (t->io_min & (t->physical_block_size - 1)) { |
562 | t->io_min = t->physical_block_size; | |
563 | t->misaligned = 1; | |
fe0b393f | 564 | ret = -1; |
c72758f3 MP |
565 | } |
566 | ||
81744ee4 | 567 | /* Optimal I/O a multiple of the physical block size? */ |
9504e086 MP |
568 | if (t->io_opt & (t->physical_block_size - 1)) { |
569 | t->io_opt = 0; | |
570 | t->misaligned = 1; | |
fe0b393f | 571 | ret = -1; |
9504e086 | 572 | } |
c72758f3 | 573 | |
81744ee4 | 574 | /* Find lowest common alignment_offset */ |
9504e086 MP |
575 | t->alignment_offset = lcm(t->alignment_offset, alignment) |
576 | & (max(t->physical_block_size, t->io_min) - 1); | |
86b37281 | 577 | |
81744ee4 | 578 | /* Verify that new alignment_offset is on a logical block boundary */ |
fe0b393f | 579 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
c72758f3 | 580 | t->misaligned = 1; |
fe0b393f MP |
581 | ret = -1; |
582 | } | |
c72758f3 | 583 | |
9504e086 MP |
584 | /* Discard alignment and granularity */ |
585 | if (b->discard_granularity) { | |
e03a72e1 | 586 | alignment = queue_limit_discard_alignment(b, start); |
9504e086 MP |
587 | |
588 | if (t->discard_granularity != 0 && | |
589 | t->discard_alignment != alignment) { | |
590 | top = t->discard_granularity + t->discard_alignment; | |
591 | bottom = b->discard_granularity + alignment; | |
70dd5bf3 | 592 | |
9504e086 MP |
593 | /* Verify that top and bottom intervals line up */ |
594 | if (max(top, bottom) & (min(top, bottom) - 1)) | |
595 | t->discard_misaligned = 1; | |
596 | } | |
597 | ||
81744ee4 MP |
598 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, |
599 | b->max_discard_sectors); | |
9504e086 MP |
600 | t->discard_granularity = max(t->discard_granularity, |
601 | b->discard_granularity); | |
602 | t->discard_alignment = lcm(t->discard_alignment, alignment) & | |
603 | (t->discard_granularity - 1); | |
604 | } | |
70dd5bf3 | 605 | |
fe0b393f | 606 | return ret; |
c72758f3 | 607 | } |
5d85d324 | 608 | EXPORT_SYMBOL(blk_stack_limits); |
c72758f3 | 609 | |
17be8c24 MP |
610 | /** |
611 | * bdev_stack_limits - adjust queue limits for stacked drivers | |
612 | * @t: the stacking driver limits (top device) | |
613 | * @bdev: the component block_device (bottom) | |
614 | * @start: first data sector within component device | |
615 | * | |
616 | * Description: | |
617 | * Merges queue limits for a top device and a block_device. Returns | |
618 | * 0 if alignment didn't change. Returns -1 if adding the bottom | |
619 | * device caused misalignment. | |
620 | */ | |
621 | int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, | |
622 | sector_t start) | |
623 | { | |
624 | struct request_queue *bq = bdev_get_queue(bdev); | |
625 | ||
626 | start += get_start_sect(bdev); | |
627 | ||
e03a72e1 | 628 | return blk_stack_limits(t, &bq->limits, start); |
17be8c24 MP |
629 | } |
630 | EXPORT_SYMBOL(bdev_stack_limits); | |
631 | ||
c72758f3 MP |
632 | /** |
633 | * disk_stack_limits - adjust queue limits for stacked drivers | |
77634f33 | 634 | * @disk: MD/DM gendisk (top) |
c72758f3 MP |
635 | * @bdev: the underlying block device (bottom) |
636 | * @offset: offset to beginning of data within component device | |
637 | * | |
638 | * Description: | |
e03a72e1 MP |
639 | * Merges the limits for a top level gendisk and a bottom level |
640 | * block_device. | |
c72758f3 MP |
641 | */ |
642 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
643 | sector_t offset) | |
644 | { | |
645 | struct request_queue *t = disk->queue; | |
646 | struct request_queue *b = bdev_get_queue(bdev); | |
647 | ||
e03a72e1 | 648 | if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { |
c72758f3 MP |
649 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; |
650 | ||
651 | disk_name(disk, 0, top); | |
652 | bdevname(bdev, bottom); | |
653 | ||
654 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
655 | top, bottom); | |
656 | } | |
657 | ||
658 | if (!t->queue_lock) | |
659 | WARN_ON_ONCE(1); | |
660 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { | |
661 | unsigned long flags; | |
662 | ||
663 | spin_lock_irqsave(t->queue_lock, flags); | |
664 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) | |
665 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); | |
666 | spin_unlock_irqrestore(t->queue_lock, flags); | |
667 | } | |
668 | } | |
669 | EXPORT_SYMBOL(disk_stack_limits); | |
670 | ||
e3790c7d TH |
671 | /** |
672 | * blk_queue_dma_pad - set pad mask | |
673 | * @q: the request queue for the device | |
674 | * @mask: pad mask | |
675 | * | |
27f8221a | 676 | * Set dma pad mask. |
e3790c7d | 677 | * |
27f8221a FT |
678 | * Appending pad buffer to a request modifies the last entry of a |
679 | * scatter list such that it includes the pad buffer. | |
e3790c7d TH |
680 | **/ |
681 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
682 | { | |
683 | q->dma_pad_mask = mask; | |
684 | } | |
685 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
686 | ||
27f8221a FT |
687 | /** |
688 | * blk_queue_update_dma_pad - update pad mask | |
689 | * @q: the request queue for the device | |
690 | * @mask: pad mask | |
691 | * | |
692 | * Update dma pad mask. | |
693 | * | |
694 | * Appending pad buffer to a request modifies the last entry of a | |
695 | * scatter list such that it includes the pad buffer. | |
696 | **/ | |
697 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
698 | { | |
699 | if (mask > q->dma_pad_mask) | |
700 | q->dma_pad_mask = mask; | |
701 | } | |
702 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
703 | ||
86db1e29 JA |
704 | /** |
705 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
86db1e29 | 706 | * @q: the request queue for the device |
2fb98e84 | 707 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
86db1e29 JA |
708 | * @buf: physically contiguous buffer |
709 | * @size: size of the buffer in bytes | |
710 | * | |
711 | * Some devices have excess DMA problems and can't simply discard (or | |
712 | * zero fill) the unwanted piece of the transfer. They have to have a | |
713 | * real area of memory to transfer it into. The use case for this is | |
714 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
715 | * bigger than the transfer size some HBAs will lock up if there | |
716 | * aren't DMA elements to contain the excess transfer. What this API | |
717 | * does is adjust the queue so that the buf is always appended | |
718 | * silently to the scatterlist. | |
719 | * | |
8a78362c MP |
720 | * Note: This routine adjusts max_hw_segments to make room for appending |
721 | * the drain buffer. If you call blk_queue_max_segments() after calling | |
722 | * this routine, you must set the limit to one fewer than your device | |
723 | * can support otherwise there won't be room for the drain buffer. | |
86db1e29 | 724 | */ |
448da4d2 | 725 | int blk_queue_dma_drain(struct request_queue *q, |
2fb98e84 TH |
726 | dma_drain_needed_fn *dma_drain_needed, |
727 | void *buf, unsigned int size) | |
86db1e29 | 728 | { |
8a78362c | 729 | if (queue_max_segments(q) < 2) |
86db1e29 JA |
730 | return -EINVAL; |
731 | /* make room for appending the drain */ | |
8a78362c | 732 | blk_queue_max_segments(q, queue_max_segments(q) - 1); |
2fb98e84 | 733 | q->dma_drain_needed = dma_drain_needed; |
86db1e29 JA |
734 | q->dma_drain_buffer = buf; |
735 | q->dma_drain_size = size; | |
736 | ||
737 | return 0; | |
738 | } | |
86db1e29 JA |
739 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
740 | ||
741 | /** | |
742 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
743 | * @q: the request queue for the device | |
744 | * @mask: the memory boundary mask | |
745 | **/ | |
746 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
747 | { | |
748 | if (mask < PAGE_CACHE_SIZE - 1) { | |
749 | mask = PAGE_CACHE_SIZE - 1; | |
24c03d47 HH |
750 | printk(KERN_INFO "%s: set to minimum %lx\n", |
751 | __func__, mask); | |
86db1e29 JA |
752 | } |
753 | ||
025146e1 | 754 | q->limits.seg_boundary_mask = mask; |
86db1e29 | 755 | } |
86db1e29 JA |
756 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
757 | ||
758 | /** | |
759 | * blk_queue_dma_alignment - set dma length and memory alignment | |
760 | * @q: the request queue for the device | |
761 | * @mask: alignment mask | |
762 | * | |
763 | * description: | |
710027a4 | 764 | * set required memory and length alignment for direct dma transactions. |
8feb4d20 | 765 | * this is used when building direct io requests for the queue. |
86db1e29 JA |
766 | * |
767 | **/ | |
768 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
769 | { | |
770 | q->dma_alignment = mask; | |
771 | } | |
86db1e29 JA |
772 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
773 | ||
774 | /** | |
775 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
776 | * @q: the request queue for the device | |
777 | * @mask: alignment mask | |
778 | * | |
779 | * description: | |
710027a4 | 780 | * update required memory and length alignment for direct dma transactions. |
86db1e29 JA |
781 | * If the requested alignment is larger than the current alignment, then |
782 | * the current queue alignment is updated to the new value, otherwise it | |
783 | * is left alone. The design of this is to allow multiple objects | |
784 | * (driver, device, transport etc) to set their respective | |
785 | * alignments without having them interfere. | |
786 | * | |
787 | **/ | |
788 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
789 | { | |
790 | BUG_ON(mask > PAGE_SIZE); | |
791 | ||
792 | if (mask > q->dma_alignment) | |
793 | q->dma_alignment = mask; | |
794 | } | |
86db1e29 JA |
795 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
796 | ||
aeb3d3a8 | 797 | static int __init blk_settings_init(void) |
86db1e29 JA |
798 | { |
799 | blk_max_low_pfn = max_low_pfn - 1; | |
800 | blk_max_pfn = max_pfn - 1; | |
801 | return 0; | |
802 | } | |
803 | subsys_initcall(blk_settings_init); |