]>
Commit | Line | Data |
---|---|---|
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 */ | |
10 | #include <linux/gcd.h> | |
11 | #include <linux/lcm.h> | |
12 | #include <linux/jiffies.h> | |
13 | #include <linux/gfp.h> | |
14 | ||
15 | #include "blk.h" | |
16 | #include "blk-wbt.h" | |
17 | ||
18 | unsigned long blk_max_low_pfn; | |
19 | EXPORT_SYMBOL(blk_max_low_pfn); | |
20 | ||
21 | unsigned long blk_max_pfn; | |
22 | ||
23 | /** | |
24 | * blk_queue_prep_rq - set a prepare_request function for queue | |
25 | * @q: queue | |
26 | * @pfn: prepare_request function | |
27 | * | |
28 | * It's possible for a queue to register a prepare_request callback which | |
29 | * is invoked before the request is handed to the request_fn. The goal of | |
30 | * the function is to prepare a request for I/O, it can be used to build a | |
31 | * cdb from the request data for instance. | |
32 | * | |
33 | */ | |
34 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) | |
35 | { | |
36 | q->prep_rq_fn = pfn; | |
37 | } | |
38 | EXPORT_SYMBOL(blk_queue_prep_rq); | |
39 | ||
40 | /** | |
41 | * blk_queue_unprep_rq - set an unprepare_request function for queue | |
42 | * @q: queue | |
43 | * @ufn: unprepare_request function | |
44 | * | |
45 | * It's possible for a queue to register an unprepare_request callback | |
46 | * which is invoked before the request is finally completed. The goal | |
47 | * of the function is to deallocate any data that was allocated in the | |
48 | * prepare_request callback. | |
49 | * | |
50 | */ | |
51 | void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn) | |
52 | { | |
53 | q->unprep_rq_fn = ufn; | |
54 | } | |
55 | EXPORT_SYMBOL(blk_queue_unprep_rq); | |
56 | ||
57 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
58 | { | |
59 | q->softirq_done_fn = fn; | |
60 | } | |
61 | EXPORT_SYMBOL(blk_queue_softirq_done); | |
62 | ||
63 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) | |
64 | { | |
65 | q->rq_timeout = timeout; | |
66 | } | |
67 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); | |
68 | ||
69 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) | |
70 | { | |
71 | q->rq_timed_out_fn = fn; | |
72 | } | |
73 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); | |
74 | ||
75 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) | |
76 | { | |
77 | q->lld_busy_fn = fn; | |
78 | } | |
79 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); | |
80 | ||
81 | /** | |
82 | * blk_set_default_limits - reset limits to default values | |
83 | * @lim: the queue_limits structure to reset | |
84 | * | |
85 | * Description: | |
86 | * Returns a queue_limit struct to its default state. | |
87 | */ | |
88 | void blk_set_default_limits(struct queue_limits *lim) | |
89 | { | |
90 | lim->max_segments = BLK_MAX_SEGMENTS; | |
91 | lim->max_discard_segments = 1; | |
92 | lim->max_integrity_segments = 0; | |
93 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; | |
94 | lim->virt_boundary_mask = 0; | |
95 | lim->max_segment_size = BLK_MAX_SEGMENT_SIZE; | |
96 | lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS; | |
97 | lim->max_dev_sectors = 0; | |
98 | lim->chunk_sectors = 0; | |
99 | lim->max_write_same_sectors = 0; | |
100 | lim->max_write_zeroes_sectors = 0; | |
101 | lim->max_discard_sectors = 0; | |
102 | lim->max_hw_discard_sectors = 0; | |
103 | lim->discard_granularity = 0; | |
104 | lim->discard_alignment = 0; | |
105 | lim->discard_misaligned = 0; | |
106 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; | |
107 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); | |
108 | lim->alignment_offset = 0; | |
109 | lim->io_opt = 0; | |
110 | lim->misaligned = 0; | |
111 | lim->cluster = 1; | |
112 | lim->zoned = BLK_ZONED_NONE; | |
113 | } | |
114 | EXPORT_SYMBOL(blk_set_default_limits); | |
115 | ||
116 | /** | |
117 | * blk_set_stacking_limits - set default limits for stacking devices | |
118 | * @lim: the queue_limits structure to reset | |
119 | * | |
120 | * Description: | |
121 | * Returns a queue_limit struct to its default state. Should be used | |
122 | * by stacking drivers like DM that have no internal limits. | |
123 | */ | |
124 | void blk_set_stacking_limits(struct queue_limits *lim) | |
125 | { | |
126 | blk_set_default_limits(lim); | |
127 | ||
128 | /* Inherit limits from component devices */ | |
129 | lim->max_segments = USHRT_MAX; | |
130 | lim->max_discard_segments = 1; | |
131 | lim->max_hw_sectors = UINT_MAX; | |
132 | lim->max_segment_size = UINT_MAX; | |
133 | lim->max_sectors = UINT_MAX; | |
134 | lim->max_dev_sectors = UINT_MAX; | |
135 | lim->max_write_same_sectors = UINT_MAX; | |
136 | lim->max_write_zeroes_sectors = UINT_MAX; | |
137 | } | |
138 | EXPORT_SYMBOL(blk_set_stacking_limits); | |
139 | ||
140 | /** | |
141 | * blk_queue_make_request - define an alternate make_request function for a device | |
142 | * @q: the request queue for the device to be affected | |
143 | * @mfn: the alternate make_request function | |
144 | * | |
145 | * Description: | |
146 | * The normal way for &struct bios to be passed to a device | |
147 | * driver is for them to be collected into requests on a request | |
148 | * queue, and then to allow the device driver to select requests | |
149 | * off that queue when it is ready. This works well for many block | |
150 | * devices. However some block devices (typically virtual devices | |
151 | * such as md or lvm) do not benefit from the processing on the | |
152 | * request queue, and are served best by having the requests passed | |
153 | * directly to them. This can be achieved by providing a function | |
154 | * to blk_queue_make_request(). | |
155 | * | |
156 | * Caveat: | |
157 | * The driver that does this *must* be able to deal appropriately | |
158 | * with buffers in "highmemory". This can be accomplished by either calling | |
159 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
160 | * blk_queue_bounce() to create a buffer in normal memory. | |
161 | **/ | |
162 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) | |
163 | { | |
164 | /* | |
165 | * set defaults | |
166 | */ | |
167 | q->nr_requests = BLKDEV_MAX_RQ; | |
168 | ||
169 | q->make_request_fn = mfn; | |
170 | blk_queue_dma_alignment(q, 511); | |
171 | blk_queue_congestion_threshold(q); | |
172 | q->nr_batching = BLK_BATCH_REQ; | |
173 | ||
174 | blk_set_default_limits(&q->limits); | |
175 | } | |
176 | EXPORT_SYMBOL(blk_queue_make_request); | |
177 | ||
178 | /** | |
179 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
180 | * @q: the request queue for the device | |
181 | * @max_addr: the maximum address the device can handle | |
182 | * | |
183 | * Description: | |
184 | * Different hardware can have different requirements as to what pages | |
185 | * it can do I/O directly to. A low level driver can call | |
186 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
187 | * buffers for doing I/O to pages residing above @max_addr. | |
188 | **/ | |
189 | void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr) | |
190 | { | |
191 | unsigned long b_pfn = max_addr >> PAGE_SHIFT; | |
192 | int dma = 0; | |
193 | ||
194 | q->bounce_gfp = GFP_NOIO; | |
195 | #if BITS_PER_LONG == 64 | |
196 | /* | |
197 | * Assume anything <= 4GB can be handled by IOMMU. Actually | |
198 | * some IOMMUs can handle everything, but I don't know of a | |
199 | * way to test this here. | |
200 | */ | |
201 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
202 | dma = 1; | |
203 | q->limits.bounce_pfn = max(max_low_pfn, b_pfn); | |
204 | #else | |
205 | if (b_pfn < blk_max_low_pfn) | |
206 | dma = 1; | |
207 | q->limits.bounce_pfn = b_pfn; | |
208 | #endif | |
209 | if (dma) { | |
210 | init_emergency_isa_pool(); | |
211 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
212 | q->limits.bounce_pfn = b_pfn; | |
213 | } | |
214 | } | |
215 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
216 | ||
217 | /** | |
218 | * blk_queue_max_hw_sectors - set max sectors for a request for this queue | |
219 | * @q: the request queue for the device | |
220 | * @max_hw_sectors: max hardware sectors in the usual 512b unit | |
221 | * | |
222 | * Description: | |
223 | * Enables a low level driver to set a hard upper limit, | |
224 | * max_hw_sectors, on the size of requests. max_hw_sectors is set by | |
225 | * the device driver based upon the capabilities of the I/O | |
226 | * controller. | |
227 | * | |
228 | * max_dev_sectors is a hard limit imposed by the storage device for | |
229 | * READ/WRITE requests. It is set by the disk driver. | |
230 | * | |
231 | * max_sectors is a soft limit imposed by the block layer for | |
232 | * filesystem type requests. This value can be overridden on a | |
233 | * per-device basis in /sys/block/<device>/queue/max_sectors_kb. | |
234 | * The soft limit can not exceed max_hw_sectors. | |
235 | **/ | |
236 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors) | |
237 | { | |
238 | struct queue_limits *limits = &q->limits; | |
239 | unsigned int max_sectors; | |
240 | ||
241 | if ((max_hw_sectors << 9) < PAGE_SIZE) { | |
242 | max_hw_sectors = 1 << (PAGE_SHIFT - 9); | |
243 | printk(KERN_INFO "%s: set to minimum %d\n", | |
244 | __func__, max_hw_sectors); | |
245 | } | |
246 | ||
247 | limits->max_hw_sectors = max_hw_sectors; | |
248 | max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors); | |
249 | max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS); | |
250 | limits->max_sectors = max_sectors; | |
251 | q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9); | |
252 | } | |
253 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); | |
254 | ||
255 | /** | |
256 | * blk_queue_chunk_sectors - set size of the chunk for this queue | |
257 | * @q: the request queue for the device | |
258 | * @chunk_sectors: chunk sectors in the usual 512b unit | |
259 | * | |
260 | * Description: | |
261 | * If a driver doesn't want IOs to cross a given chunk size, it can set | |
262 | * this limit and prevent merging across chunks. Note that the chunk size | |
263 | * must currently be a power-of-2 in sectors. Also note that the block | |
264 | * layer must accept a page worth of data at any offset. So if the | |
265 | * crossing of chunks is a hard limitation in the driver, it must still be | |
266 | * prepared to split single page bios. | |
267 | **/ | |
268 | void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors) | |
269 | { | |
270 | BUG_ON(!is_power_of_2(chunk_sectors)); | |
271 | q->limits.chunk_sectors = chunk_sectors; | |
272 | } | |
273 | EXPORT_SYMBOL(blk_queue_chunk_sectors); | |
274 | ||
275 | /** | |
276 | * blk_queue_max_discard_sectors - set max sectors for a single discard | |
277 | * @q: the request queue for the device | |
278 | * @max_discard_sectors: maximum number of sectors to discard | |
279 | **/ | |
280 | void blk_queue_max_discard_sectors(struct request_queue *q, | |
281 | unsigned int max_discard_sectors) | |
282 | { | |
283 | q->limits.max_hw_discard_sectors = max_discard_sectors; | |
284 | q->limits.max_discard_sectors = max_discard_sectors; | |
285 | } | |
286 | EXPORT_SYMBOL(blk_queue_max_discard_sectors); | |
287 | ||
288 | /** | |
289 | * blk_queue_max_write_same_sectors - set max sectors for a single write same | |
290 | * @q: the request queue for the device | |
291 | * @max_write_same_sectors: maximum number of sectors to write per command | |
292 | **/ | |
293 | void blk_queue_max_write_same_sectors(struct request_queue *q, | |
294 | unsigned int max_write_same_sectors) | |
295 | { | |
296 | q->limits.max_write_same_sectors = max_write_same_sectors; | |
297 | } | |
298 | EXPORT_SYMBOL(blk_queue_max_write_same_sectors); | |
299 | ||
300 | /** | |
301 | * blk_queue_max_write_zeroes_sectors - set max sectors for a single | |
302 | * write zeroes | |
303 | * @q: the request queue for the device | |
304 | * @max_write_zeroes_sectors: maximum number of sectors to write per command | |
305 | **/ | |
306 | void blk_queue_max_write_zeroes_sectors(struct request_queue *q, | |
307 | unsigned int max_write_zeroes_sectors) | |
308 | { | |
309 | q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors; | |
310 | } | |
311 | EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors); | |
312 | ||
313 | /** | |
314 | * blk_queue_max_segments - set max hw segments for a request for this queue | |
315 | * @q: the request queue for the device | |
316 | * @max_segments: max number of segments | |
317 | * | |
318 | * Description: | |
319 | * Enables a low level driver to set an upper limit on the number of | |
320 | * hw data segments in a request. | |
321 | **/ | |
322 | void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments) | |
323 | { | |
324 | if (!max_segments) { | |
325 | max_segments = 1; | |
326 | printk(KERN_INFO "%s: set to minimum %d\n", | |
327 | __func__, max_segments); | |
328 | } | |
329 | ||
330 | q->limits.max_segments = max_segments; | |
331 | } | |
332 | EXPORT_SYMBOL(blk_queue_max_segments); | |
333 | ||
334 | /** | |
335 | * blk_queue_max_discard_segments - set max segments for discard requests | |
336 | * @q: the request queue for the device | |
337 | * @max_segments: max number of segments | |
338 | * | |
339 | * Description: | |
340 | * Enables a low level driver to set an upper limit on the number of | |
341 | * segments in a discard request. | |
342 | **/ | |
343 | void blk_queue_max_discard_segments(struct request_queue *q, | |
344 | unsigned short max_segments) | |
345 | { | |
346 | q->limits.max_discard_segments = max_segments; | |
347 | } | |
348 | EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments); | |
349 | ||
350 | /** | |
351 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
352 | * @q: the request queue for the device | |
353 | * @max_size: max size of segment in bytes | |
354 | * | |
355 | * Description: | |
356 | * Enables a low level driver to set an upper limit on the size of a | |
357 | * coalesced segment | |
358 | **/ | |
359 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
360 | { | |
361 | if (max_size < PAGE_SIZE) { | |
362 | max_size = PAGE_SIZE; | |
363 | printk(KERN_INFO "%s: set to minimum %d\n", | |
364 | __func__, max_size); | |
365 | } | |
366 | ||
367 | q->limits.max_segment_size = max_size; | |
368 | } | |
369 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
370 | ||
371 | /** | |
372 | * blk_queue_logical_block_size - set logical block size for the queue | |
373 | * @q: the request queue for the device | |
374 | * @size: the logical block size, in bytes | |
375 | * | |
376 | * Description: | |
377 | * This should be set to the lowest possible block size that the | |
378 | * storage device can address. The default of 512 covers most | |
379 | * hardware. | |
380 | **/ | |
381 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) | |
382 | { | |
383 | q->limits.logical_block_size = size; | |
384 | ||
385 | if (q->limits.physical_block_size < size) | |
386 | q->limits.physical_block_size = size; | |
387 | ||
388 | if (q->limits.io_min < q->limits.physical_block_size) | |
389 | q->limits.io_min = q->limits.physical_block_size; | |
390 | } | |
391 | EXPORT_SYMBOL(blk_queue_logical_block_size); | |
392 | ||
393 | /** | |
394 | * blk_queue_physical_block_size - set physical block size for the queue | |
395 | * @q: the request queue for the device | |
396 | * @size: the physical block size, in bytes | |
397 | * | |
398 | * Description: | |
399 | * This should be set to the lowest possible sector size that the | |
400 | * hardware can operate on without reverting to read-modify-write | |
401 | * operations. | |
402 | */ | |
403 | void blk_queue_physical_block_size(struct request_queue *q, unsigned int size) | |
404 | { | |
405 | q->limits.physical_block_size = size; | |
406 | ||
407 | if (q->limits.physical_block_size < q->limits.logical_block_size) | |
408 | q->limits.physical_block_size = q->limits.logical_block_size; | |
409 | ||
410 | if (q->limits.io_min < q->limits.physical_block_size) | |
411 | q->limits.io_min = q->limits.physical_block_size; | |
412 | } | |
413 | EXPORT_SYMBOL(blk_queue_physical_block_size); | |
414 | ||
415 | /** | |
416 | * blk_queue_alignment_offset - set physical block alignment offset | |
417 | * @q: the request queue for the device | |
418 | * @offset: alignment offset in bytes | |
419 | * | |
420 | * Description: | |
421 | * Some devices are naturally misaligned to compensate for things like | |
422 | * the legacy DOS partition table 63-sector offset. Low-level drivers | |
423 | * should call this function for devices whose first sector is not | |
424 | * naturally aligned. | |
425 | */ | |
426 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) | |
427 | { | |
428 | q->limits.alignment_offset = | |
429 | offset & (q->limits.physical_block_size - 1); | |
430 | q->limits.misaligned = 0; | |
431 | } | |
432 | EXPORT_SYMBOL(blk_queue_alignment_offset); | |
433 | ||
434 | /** | |
435 | * blk_limits_io_min - set minimum request size for a device | |
436 | * @limits: the queue limits | |
437 | * @min: smallest I/O size in bytes | |
438 | * | |
439 | * Description: | |
440 | * Some devices have an internal block size bigger than the reported | |
441 | * hardware sector size. This function can be used to signal the | |
442 | * smallest I/O the device can perform without incurring a performance | |
443 | * penalty. | |
444 | */ | |
445 | void blk_limits_io_min(struct queue_limits *limits, unsigned int min) | |
446 | { | |
447 | limits->io_min = min; | |
448 | ||
449 | if (limits->io_min < limits->logical_block_size) | |
450 | limits->io_min = limits->logical_block_size; | |
451 | ||
452 | if (limits->io_min < limits->physical_block_size) | |
453 | limits->io_min = limits->physical_block_size; | |
454 | } | |
455 | EXPORT_SYMBOL(blk_limits_io_min); | |
456 | ||
457 | /** | |
458 | * blk_queue_io_min - set minimum request size for the queue | |
459 | * @q: the request queue for the device | |
460 | * @min: smallest I/O size in bytes | |
461 | * | |
462 | * Description: | |
463 | * Storage devices may report a granularity or preferred minimum I/O | |
464 | * size which is the smallest request the device can perform without | |
465 | * incurring a performance penalty. For disk drives this is often the | |
466 | * physical block size. For RAID arrays it is often the stripe chunk | |
467 | * size. A properly aligned multiple of minimum_io_size is the | |
468 | * preferred request size for workloads where a high number of I/O | |
469 | * operations is desired. | |
470 | */ | |
471 | void blk_queue_io_min(struct request_queue *q, unsigned int min) | |
472 | { | |
473 | blk_limits_io_min(&q->limits, min); | |
474 | } | |
475 | EXPORT_SYMBOL(blk_queue_io_min); | |
476 | ||
477 | /** | |
478 | * blk_limits_io_opt - set optimal request size for a device | |
479 | * @limits: the queue limits | |
480 | * @opt: smallest I/O size in bytes | |
481 | * | |
482 | * Description: | |
483 | * Storage devices may report an optimal I/O size, which is the | |
484 | * device's preferred unit for sustained I/O. This is rarely reported | |
485 | * for disk drives. For RAID arrays it is usually the stripe width or | |
486 | * the internal track size. A properly aligned multiple of | |
487 | * optimal_io_size is the preferred request size for workloads where | |
488 | * sustained throughput is desired. | |
489 | */ | |
490 | void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt) | |
491 | { | |
492 | limits->io_opt = opt; | |
493 | } | |
494 | EXPORT_SYMBOL(blk_limits_io_opt); | |
495 | ||
496 | /** | |
497 | * blk_queue_io_opt - set optimal request size for the queue | |
498 | * @q: the request queue for the device | |
499 | * @opt: optimal request size in bytes | |
500 | * | |
501 | * Description: | |
502 | * Storage devices may report an optimal I/O size, which is the | |
503 | * device's preferred unit for sustained I/O. This is rarely reported | |
504 | * for disk drives. For RAID arrays it is usually the stripe width or | |
505 | * the internal track size. A properly aligned multiple of | |
506 | * optimal_io_size is the preferred request size for workloads where | |
507 | * sustained throughput is desired. | |
508 | */ | |
509 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) | |
510 | { | |
511 | blk_limits_io_opt(&q->limits, opt); | |
512 | } | |
513 | EXPORT_SYMBOL(blk_queue_io_opt); | |
514 | ||
515 | /** | |
516 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
517 | * @t: the stacking driver (top) | |
518 | * @b: the underlying device (bottom) | |
519 | **/ | |
520 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
521 | { | |
522 | blk_stack_limits(&t->limits, &b->limits, 0); | |
523 | } | |
524 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
525 | ||
526 | /** | |
527 | * blk_stack_limits - adjust queue_limits for stacked devices | |
528 | * @t: the stacking driver limits (top device) | |
529 | * @b: the underlying queue limits (bottom, component device) | |
530 | * @start: first data sector within component device | |
531 | * | |
532 | * Description: | |
533 | * This function is used by stacking drivers like MD and DM to ensure | |
534 | * that all component devices have compatible block sizes and | |
535 | * alignments. The stacking driver must provide a queue_limits | |
536 | * struct (top) and then iteratively call the stacking function for | |
537 | * all component (bottom) devices. The stacking function will | |
538 | * attempt to combine the values and ensure proper alignment. | |
539 | * | |
540 | * Returns 0 if the top and bottom queue_limits are compatible. The | |
541 | * top device's block sizes and alignment offsets may be adjusted to | |
542 | * ensure alignment with the bottom device. If no compatible sizes | |
543 | * and alignments exist, -1 is returned and the resulting top | |
544 | * queue_limits will have the misaligned flag set to indicate that | |
545 | * the alignment_offset is undefined. | |
546 | */ | |
547 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |
548 | sector_t start) | |
549 | { | |
550 | unsigned int top, bottom, alignment, ret = 0; | |
551 | ||
552 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); | |
553 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | |
554 | t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors); | |
555 | t->max_write_same_sectors = min(t->max_write_same_sectors, | |
556 | b->max_write_same_sectors); | |
557 | t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors, | |
558 | b->max_write_zeroes_sectors); | |
559 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); | |
560 | ||
561 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, | |
562 | b->seg_boundary_mask); | |
563 | t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask, | |
564 | b->virt_boundary_mask); | |
565 | ||
566 | t->max_segments = min_not_zero(t->max_segments, b->max_segments); | |
567 | t->max_discard_segments = min_not_zero(t->max_discard_segments, | |
568 | b->max_discard_segments); | |
569 | t->max_integrity_segments = min_not_zero(t->max_integrity_segments, | |
570 | b->max_integrity_segments); | |
571 | ||
572 | t->max_segment_size = min_not_zero(t->max_segment_size, | |
573 | b->max_segment_size); | |
574 | ||
575 | t->misaligned |= b->misaligned; | |
576 | ||
577 | alignment = queue_limit_alignment_offset(b, start); | |
578 | ||
579 | /* Bottom device has different alignment. Check that it is | |
580 | * compatible with the current top alignment. | |
581 | */ | |
582 | if (t->alignment_offset != alignment) { | |
583 | ||
584 | top = max(t->physical_block_size, t->io_min) | |
585 | + t->alignment_offset; | |
586 | bottom = max(b->physical_block_size, b->io_min) + alignment; | |
587 | ||
588 | /* Verify that top and bottom intervals line up */ | |
589 | if (max(top, bottom) % min(top, bottom)) { | |
590 | t->misaligned = 1; | |
591 | ret = -1; | |
592 | } | |
593 | } | |
594 | ||
595 | t->logical_block_size = max(t->logical_block_size, | |
596 | b->logical_block_size); | |
597 | ||
598 | t->physical_block_size = max(t->physical_block_size, | |
599 | b->physical_block_size); | |
600 | ||
601 | t->io_min = max(t->io_min, b->io_min); | |
602 | t->io_opt = lcm_not_zero(t->io_opt, b->io_opt); | |
603 | ||
604 | t->cluster &= b->cluster; | |
605 | ||
606 | /* Physical block size a multiple of the logical block size? */ | |
607 | if (t->physical_block_size & (t->logical_block_size - 1)) { | |
608 | t->physical_block_size = t->logical_block_size; | |
609 | t->misaligned = 1; | |
610 | ret = -1; | |
611 | } | |
612 | ||
613 | /* Minimum I/O a multiple of the physical block size? */ | |
614 | if (t->io_min & (t->physical_block_size - 1)) { | |
615 | t->io_min = t->physical_block_size; | |
616 | t->misaligned = 1; | |
617 | ret = -1; | |
618 | } | |
619 | ||
620 | /* Optimal I/O a multiple of the physical block size? */ | |
621 | if (t->io_opt & (t->physical_block_size - 1)) { | |
622 | t->io_opt = 0; | |
623 | t->misaligned = 1; | |
624 | ret = -1; | |
625 | } | |
626 | ||
627 | t->raid_partial_stripes_expensive = | |
628 | max(t->raid_partial_stripes_expensive, | |
629 | b->raid_partial_stripes_expensive); | |
630 | ||
631 | /* Find lowest common alignment_offset */ | |
632 | t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment) | |
633 | % max(t->physical_block_size, t->io_min); | |
634 | ||
635 | /* Verify that new alignment_offset is on a logical block boundary */ | |
636 | if (t->alignment_offset & (t->logical_block_size - 1)) { | |
637 | t->misaligned = 1; | |
638 | ret = -1; | |
639 | } | |
640 | ||
641 | /* Discard alignment and granularity */ | |
642 | if (b->discard_granularity) { | |
643 | alignment = queue_limit_discard_alignment(b, start); | |
644 | ||
645 | if (t->discard_granularity != 0 && | |
646 | t->discard_alignment != alignment) { | |
647 | top = t->discard_granularity + t->discard_alignment; | |
648 | bottom = b->discard_granularity + alignment; | |
649 | ||
650 | /* Verify that top and bottom intervals line up */ | |
651 | if ((max(top, bottom) % min(top, bottom)) != 0) | |
652 | t->discard_misaligned = 1; | |
653 | } | |
654 | ||
655 | t->max_discard_sectors = min_not_zero(t->max_discard_sectors, | |
656 | b->max_discard_sectors); | |
657 | t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors, | |
658 | b->max_hw_discard_sectors); | |
659 | t->discard_granularity = max(t->discard_granularity, | |
660 | b->discard_granularity); | |
661 | t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) % | |
662 | t->discard_granularity; | |
663 | } | |
664 | ||
665 | if (b->chunk_sectors) | |
666 | t->chunk_sectors = min_not_zero(t->chunk_sectors, | |
667 | b->chunk_sectors); | |
668 | ||
669 | return ret; | |
670 | } | |
671 | EXPORT_SYMBOL(blk_stack_limits); | |
672 | ||
673 | /** | |
674 | * bdev_stack_limits - adjust queue limits for stacked drivers | |
675 | * @t: the stacking driver limits (top device) | |
676 | * @bdev: the component block_device (bottom) | |
677 | * @start: first data sector within component device | |
678 | * | |
679 | * Description: | |
680 | * Merges queue limits for a top device and a block_device. Returns | |
681 | * 0 if alignment didn't change. Returns -1 if adding the bottom | |
682 | * device caused misalignment. | |
683 | */ | |
684 | int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, | |
685 | sector_t start) | |
686 | { | |
687 | struct request_queue *bq = bdev_get_queue(bdev); | |
688 | ||
689 | start += get_start_sect(bdev); | |
690 | ||
691 | return blk_stack_limits(t, &bq->limits, start); | |
692 | } | |
693 | EXPORT_SYMBOL(bdev_stack_limits); | |
694 | ||
695 | /** | |
696 | * disk_stack_limits - adjust queue limits for stacked drivers | |
697 | * @disk: MD/DM gendisk (top) | |
698 | * @bdev: the underlying block device (bottom) | |
699 | * @offset: offset to beginning of data within component device | |
700 | * | |
701 | * Description: | |
702 | * Merges the limits for a top level gendisk and a bottom level | |
703 | * block_device. | |
704 | */ | |
705 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, | |
706 | sector_t offset) | |
707 | { | |
708 | struct request_queue *t = disk->queue; | |
709 | ||
710 | if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { | |
711 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; | |
712 | ||
713 | disk_name(disk, 0, top); | |
714 | bdevname(bdev, bottom); | |
715 | ||
716 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", | |
717 | top, bottom); | |
718 | } | |
719 | } | |
720 | EXPORT_SYMBOL(disk_stack_limits); | |
721 | ||
722 | /** | |
723 | * blk_queue_dma_pad - set pad mask | |
724 | * @q: the request queue for the device | |
725 | * @mask: pad mask | |
726 | * | |
727 | * Set dma pad mask. | |
728 | * | |
729 | * Appending pad buffer to a request modifies the last entry of a | |
730 | * scatter list such that it includes the pad buffer. | |
731 | **/ | |
732 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) | |
733 | { | |
734 | q->dma_pad_mask = mask; | |
735 | } | |
736 | EXPORT_SYMBOL(blk_queue_dma_pad); | |
737 | ||
738 | /** | |
739 | * blk_queue_update_dma_pad - update pad mask | |
740 | * @q: the request queue for the device | |
741 | * @mask: pad mask | |
742 | * | |
743 | * Update dma pad mask. | |
744 | * | |
745 | * Appending pad buffer to a request modifies the last entry of a | |
746 | * scatter list such that it includes the pad buffer. | |
747 | **/ | |
748 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) | |
749 | { | |
750 | if (mask > q->dma_pad_mask) | |
751 | q->dma_pad_mask = mask; | |
752 | } | |
753 | EXPORT_SYMBOL(blk_queue_update_dma_pad); | |
754 | ||
755 | /** | |
756 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. | |
757 | * @q: the request queue for the device | |
758 | * @dma_drain_needed: fn which returns non-zero if drain is necessary | |
759 | * @buf: physically contiguous buffer | |
760 | * @size: size of the buffer in bytes | |
761 | * | |
762 | * Some devices have excess DMA problems and can't simply discard (or | |
763 | * zero fill) the unwanted piece of the transfer. They have to have a | |
764 | * real area of memory to transfer it into. The use case for this is | |
765 | * ATAPI devices in DMA mode. If the packet command causes a transfer | |
766 | * bigger than the transfer size some HBAs will lock up if there | |
767 | * aren't DMA elements to contain the excess transfer. What this API | |
768 | * does is adjust the queue so that the buf is always appended | |
769 | * silently to the scatterlist. | |
770 | * | |
771 | * Note: This routine adjusts max_hw_segments to make room for appending | |
772 | * the drain buffer. If you call blk_queue_max_segments() after calling | |
773 | * this routine, you must set the limit to one fewer than your device | |
774 | * can support otherwise there won't be room for the drain buffer. | |
775 | */ | |
776 | int blk_queue_dma_drain(struct request_queue *q, | |
777 | dma_drain_needed_fn *dma_drain_needed, | |
778 | void *buf, unsigned int size) | |
779 | { | |
780 | if (queue_max_segments(q) < 2) | |
781 | return -EINVAL; | |
782 | /* make room for appending the drain */ | |
783 | blk_queue_max_segments(q, queue_max_segments(q) - 1); | |
784 | q->dma_drain_needed = dma_drain_needed; | |
785 | q->dma_drain_buffer = buf; | |
786 | q->dma_drain_size = size; | |
787 | ||
788 | return 0; | |
789 | } | |
790 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); | |
791 | ||
792 | /** | |
793 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
794 | * @q: the request queue for the device | |
795 | * @mask: the memory boundary mask | |
796 | **/ | |
797 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
798 | { | |
799 | if (mask < PAGE_SIZE - 1) { | |
800 | mask = PAGE_SIZE - 1; | |
801 | printk(KERN_INFO "%s: set to minimum %lx\n", | |
802 | __func__, mask); | |
803 | } | |
804 | ||
805 | q->limits.seg_boundary_mask = mask; | |
806 | } | |
807 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
808 | ||
809 | /** | |
810 | * blk_queue_virt_boundary - set boundary rules for bio merging | |
811 | * @q: the request queue for the device | |
812 | * @mask: the memory boundary mask | |
813 | **/ | |
814 | void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask) | |
815 | { | |
816 | q->limits.virt_boundary_mask = mask; | |
817 | } | |
818 | EXPORT_SYMBOL(blk_queue_virt_boundary); | |
819 | ||
820 | /** | |
821 | * blk_queue_dma_alignment - set dma length and memory alignment | |
822 | * @q: the request queue for the device | |
823 | * @mask: alignment mask | |
824 | * | |
825 | * description: | |
826 | * set required memory and length alignment for direct dma transactions. | |
827 | * this is used when building direct io requests for the queue. | |
828 | * | |
829 | **/ | |
830 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
831 | { | |
832 | q->dma_alignment = mask; | |
833 | } | |
834 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
835 | ||
836 | /** | |
837 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
838 | * @q: the request queue for the device | |
839 | * @mask: alignment mask | |
840 | * | |
841 | * description: | |
842 | * update required memory and length alignment for direct dma transactions. | |
843 | * If the requested alignment is larger than the current alignment, then | |
844 | * the current queue alignment is updated to the new value, otherwise it | |
845 | * is left alone. The design of this is to allow multiple objects | |
846 | * (driver, device, transport etc) to set their respective | |
847 | * alignments without having them interfere. | |
848 | * | |
849 | **/ | |
850 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
851 | { | |
852 | BUG_ON(mask > PAGE_SIZE); | |
853 | ||
854 | if (mask > q->dma_alignment) | |
855 | q->dma_alignment = mask; | |
856 | } | |
857 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); | |
858 | ||
859 | void blk_queue_flush_queueable(struct request_queue *q, bool queueable) | |
860 | { | |
861 | spin_lock_irq(q->queue_lock); | |
862 | if (queueable) | |
863 | clear_bit(QUEUE_FLAG_FLUSH_NQ, &q->queue_flags); | |
864 | else | |
865 | set_bit(QUEUE_FLAG_FLUSH_NQ, &q->queue_flags); | |
866 | spin_unlock_irq(q->queue_lock); | |
867 | } | |
868 | EXPORT_SYMBOL_GPL(blk_queue_flush_queueable); | |
869 | ||
870 | /** | |
871 | * blk_set_queue_depth - tell the block layer about the device queue depth | |
872 | * @q: the request queue for the device | |
873 | * @depth: queue depth | |
874 | * | |
875 | */ | |
876 | void blk_set_queue_depth(struct request_queue *q, unsigned int depth) | |
877 | { | |
878 | q->queue_depth = depth; | |
879 | wbt_set_queue_depth(q->rq_wb, depth); | |
880 | } | |
881 | EXPORT_SYMBOL(blk_set_queue_depth); | |
882 | ||
883 | /** | |
884 | * blk_queue_write_cache - configure queue's write cache | |
885 | * @q: the request queue for the device | |
886 | * @wc: write back cache on or off | |
887 | * @fua: device supports FUA writes, if true | |
888 | * | |
889 | * Tell the block layer about the write cache of @q. | |
890 | */ | |
891 | void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua) | |
892 | { | |
893 | spin_lock_irq(q->queue_lock); | |
894 | if (wc) | |
895 | queue_flag_set(QUEUE_FLAG_WC, q); | |
896 | else | |
897 | queue_flag_clear(QUEUE_FLAG_WC, q); | |
898 | if (fua) | |
899 | queue_flag_set(QUEUE_FLAG_FUA, q); | |
900 | else | |
901 | queue_flag_clear(QUEUE_FLAG_FUA, q); | |
902 | spin_unlock_irq(q->queue_lock); | |
903 | ||
904 | wbt_set_write_cache(q->rq_wb, test_bit(QUEUE_FLAG_WC, &q->queue_flags)); | |
905 | } | |
906 | EXPORT_SYMBOL_GPL(blk_queue_write_cache); | |
907 | ||
908 | static int __init blk_settings_init(void) | |
909 | { | |
910 | blk_max_low_pfn = max_low_pfn - 1; | |
911 | blk_max_pfn = max_pfn - 1; | |
912 | return 0; | |
913 | } | |
914 | subsys_initcall(blk_settings_init); |