blob: bd582a7f5310efcbe53e39eb3f1da8abf83d2eb2 [file] [log] [blame]
Jens Axboe86db1e22008-01-29 14:53:40 +01001/*
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
11#include "blk.h"
12
Jens Axboe6728cb02008-01-31 13:03:55 +010013unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010014EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010015
16unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010017
18/**
19 * blk_queue_prep_rq - set a prepare_request function for queue
20 * @q: queue
21 * @pfn: prepare_request function
22 *
23 * It's possible for a queue to register a prepare_request callback which
24 * is invoked before the request is handed to the request_fn. The goal of
25 * the function is to prepare a request for I/O, it can be used to build a
26 * cdb from the request data for instance.
27 *
28 */
29void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
30{
31 q->prep_rq_fn = pfn;
32}
Jens Axboe86db1e22008-01-29 14:53:40 +010033EXPORT_SYMBOL(blk_queue_prep_rq);
34
35/**
David Woodhousefb2dce82008-08-05 18:01:53 +010036 * blk_queue_set_discard - set a discard_sectors function for queue
37 * @q: queue
38 * @dfn: prepare_discard function
39 *
40 * It's possible for a queue to register a discard callback which is used
41 * to transform a discard request into the appropriate type for the
42 * hardware. If none is registered, then discard requests are failed
43 * with %EOPNOTSUPP.
44 *
45 */
46void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
47{
48 q->prepare_discard_fn = dfn;
49}
50EXPORT_SYMBOL(blk_queue_set_discard);
51
52/**
Jens Axboe86db1e22008-01-29 14:53:40 +010053 * blk_queue_merge_bvec - set a merge_bvec function for queue
54 * @q: queue
55 * @mbfn: merge_bvec_fn
56 *
57 * Usually queues have static limitations on the max sectors or segments that
58 * we can put in a request. Stacking drivers may have some settings that
59 * are dynamic, and thus we have to query the queue whether it is ok to
60 * add a new bio_vec to a bio at a given offset or not. If the block device
61 * has such limitations, it needs to register a merge_bvec_fn to control
62 * the size of bio's sent to it. Note that a block device *must* allow a
63 * single page to be added to an empty bio. The block device driver may want
64 * to use the bio_split() function to deal with these bio's. By default
65 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
66 * honored.
67 */
68void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
69{
70 q->merge_bvec_fn = mbfn;
71}
Jens Axboe86db1e22008-01-29 14:53:40 +010072EXPORT_SYMBOL(blk_queue_merge_bvec);
73
74void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
75{
76 q->softirq_done_fn = fn;
77}
Jens Axboe86db1e22008-01-29 14:53:40 +010078EXPORT_SYMBOL(blk_queue_softirq_done);
79
Jens Axboe242f9dc2008-09-14 05:55:09 -070080void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
81{
82 q->rq_timeout = timeout;
83}
84EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
85
86void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
87{
88 q->rq_timed_out_fn = fn;
89}
90EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
91
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020092void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
93{
94 q->lld_busy_fn = fn;
95}
96EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
97
Jens Axboe86db1e22008-01-29 14:53:40 +010098/**
Martin K. Petersene475bba2009-06-16 08:23:52 +020099 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5c2009-06-19 09:18:32 +0200100 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +0200101 *
102 * Description:
103 * Returns a queue_limit struct to its default state. Can be used by
104 * stacking drivers like DM that stage table swaps and reuse an
105 * existing device queue.
106 */
107void blk_set_default_limits(struct queue_limits *lim)
108{
109 lim->max_phys_segments = MAX_PHYS_SEGMENTS;
110 lim->max_hw_segments = MAX_HW_SEGMENTS;
111 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
112 lim->max_segment_size = MAX_SEGMENT_SIZE;
113 lim->max_sectors = lim->max_hw_sectors = SAFE_MAX_SECTORS;
114 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200115 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200116 lim->alignment_offset = 0;
117 lim->io_opt = 0;
118 lim->misaligned = 0;
119 lim->no_cluster = 0;
120}
121EXPORT_SYMBOL(blk_set_default_limits);
122
123/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100124 * blk_queue_make_request - define an alternate make_request function for a device
125 * @q: the request queue for the device to be affected
126 * @mfn: the alternate make_request function
127 *
128 * Description:
129 * The normal way for &struct bios to be passed to a device
130 * driver is for them to be collected into requests on a request
131 * queue, and then to allow the device driver to select requests
132 * off that queue when it is ready. This works well for many block
133 * devices. However some block devices (typically virtual devices
134 * such as md or lvm) do not benefit from the processing on the
135 * request queue, and are served best by having the requests passed
136 * directly to them. This can be achieved by providing a function
137 * to blk_queue_make_request().
138 *
139 * Caveat:
140 * The driver that does this *must* be able to deal appropriately
141 * with buffers in "highmemory". This can be accomplished by either calling
142 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
143 * blk_queue_bounce() to create a buffer in normal memory.
144 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100145void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100146{
147 /*
148 * set defaults
149 */
150 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100151
Jens Axboe86db1e22008-01-29 14:53:40 +0100152 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100153 blk_queue_dma_alignment(q, 511);
154 blk_queue_congestion_threshold(q);
155 q->nr_batching = BLK_BATCH_REQ;
156
157 q->unplug_thresh = 4; /* hmm */
158 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
159 if (q->unplug_delay == 0)
160 q->unplug_delay = 1;
161
Jens Axboe86db1e22008-01-29 14:53:40 +0100162 q->unplug_timer.function = blk_unplug_timeout;
163 q->unplug_timer.data = (unsigned long)q;
164
Martin K. Petersene475bba2009-06-16 08:23:52 +0200165 blk_set_default_limits(&q->limits);
166
Jens Axboe86db1e22008-01-29 14:53:40 +0100167 /*
168 * by default assume old behaviour and bounce for any highmem page
169 */
170 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
171}
Jens Axboe86db1e22008-01-29 14:53:40 +0100172EXPORT_SYMBOL(blk_queue_make_request);
173
174/**
175 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900176 * @q: the request queue for the device
177 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100178 *
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
Tejun Heocd0aca22009-04-15 22:10:25 +0900183 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100184 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900185void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100186{
Tejun Heocd0aca22009-04-15 22:10:25 +0900187 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100188 int dma = 0;
189
190 q->bounce_gfp = GFP_NOIO;
191#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900192 /*
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))
Jens Axboe86db1e22008-01-29 14:53:40 +0100198 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400199 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100200#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100201 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100202 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400203 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100204#endif
205 if (dma) {
206 init_emergency_isa_pool();
207 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400208 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100209 }
210}
Jens Axboe86db1e22008-01-29 14:53:40 +0100211EXPORT_SYMBOL(blk_queue_bounce_limit);
212
213/**
214 * blk_queue_max_sectors - set max sectors for a request for this queue
215 * @q: the request queue for the device
216 * @max_sectors: max sectors in the usual 512b unit
217 *
218 * Description:
219 * Enables a low level driver to set an upper limit on the size of
220 * received requests.
221 **/
222void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
223{
224 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
225 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700226 printk(KERN_INFO "%s: set to minimum %d\n",
227 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100228 }
229
230 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400231 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100232 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400233 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
234 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100235 }
236}
Jens Axboe86db1e22008-01-29 14:53:40 +0100237EXPORT_SYMBOL(blk_queue_max_sectors);
238
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400239void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
240{
241 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400242 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400243 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400244 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400245}
246EXPORT_SYMBOL(blk_queue_max_hw_sectors);
247
Jens Axboe86db1e22008-01-29 14:53:40 +0100248/**
249 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
250 * @q: the request queue for the device
251 * @max_segments: max number of segments
252 *
253 * Description:
254 * Enables a low level driver to set an upper limit on the number of
255 * physical data segments in a request. This would be the largest sized
256 * scatter list the driver could handle.
257 **/
258void blk_queue_max_phys_segments(struct request_queue *q,
259 unsigned short max_segments)
260{
261 if (!max_segments) {
262 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700263 printk(KERN_INFO "%s: set to minimum %d\n",
264 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100265 }
266
Martin K. Petersen025146e2009-05-22 17:17:51 -0400267 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100268}
Jens Axboe86db1e22008-01-29 14:53:40 +0100269EXPORT_SYMBOL(blk_queue_max_phys_segments);
270
271/**
272 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
273 * @q: the request queue for the device
274 * @max_segments: max number of segments
275 *
276 * Description:
277 * Enables a low level driver to set an upper limit on the number of
278 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200279 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100280 * to the device.
281 **/
282void blk_queue_max_hw_segments(struct request_queue *q,
283 unsigned short max_segments)
284{
285 if (!max_segments) {
286 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700287 printk(KERN_INFO "%s: set to minimum %d\n",
288 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100289 }
290
Martin K. Petersen025146e2009-05-22 17:17:51 -0400291 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100292}
Jens Axboe86db1e22008-01-29 14:53:40 +0100293EXPORT_SYMBOL(blk_queue_max_hw_segments);
294
295/**
296 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
297 * @q: the request queue for the device
298 * @max_size: max size of segment in bytes
299 *
300 * Description:
301 * Enables a low level driver to set an upper limit on the size of a
302 * coalesced segment
303 **/
304void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
305{
306 if (max_size < PAGE_CACHE_SIZE) {
307 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700308 printk(KERN_INFO "%s: set to minimum %d\n",
309 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100310 }
311
Martin K. Petersen025146e2009-05-22 17:17:51 -0400312 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100313}
Jens Axboe86db1e22008-01-29 14:53:40 +0100314EXPORT_SYMBOL(blk_queue_max_segment_size);
315
316/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400317 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100318 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400319 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100320 *
321 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400322 * This should be set to the lowest possible block size that the
323 * storage device can address. The default of 512 covers most
324 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100325 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400326void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100327{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400328 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400329
330 if (q->limits.physical_block_size < size)
331 q->limits.physical_block_size = size;
332
333 if (q->limits.io_min < q->limits.physical_block_size)
334 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100335}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400336EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100337
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400338/**
339 * blk_queue_physical_block_size - set physical block size for the queue
340 * @q: the request queue for the device
341 * @size: the physical block size, in bytes
342 *
343 * Description:
344 * This should be set to the lowest possible sector size that the
345 * hardware can operate on without reverting to read-modify-write
346 * operations.
347 */
348void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
349{
350 q->limits.physical_block_size = size;
351
352 if (q->limits.physical_block_size < q->limits.logical_block_size)
353 q->limits.physical_block_size = q->limits.logical_block_size;
354
355 if (q->limits.io_min < q->limits.physical_block_size)
356 q->limits.io_min = q->limits.physical_block_size;
357}
358EXPORT_SYMBOL(blk_queue_physical_block_size);
359
360/**
361 * blk_queue_alignment_offset - set physical block alignment offset
362 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700363 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400364 *
365 * Description:
366 * Some devices are naturally misaligned to compensate for things like
367 * the legacy DOS partition table 63-sector offset. Low-level drivers
368 * should call this function for devices whose first sector is not
369 * naturally aligned.
370 */
371void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
372{
373 q->limits.alignment_offset =
374 offset & (q->limits.physical_block_size - 1);
375 q->limits.misaligned = 0;
376}
377EXPORT_SYMBOL(blk_queue_alignment_offset);
378
379/**
380 * blk_queue_io_min - set minimum request size for the queue
381 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700382 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400383 *
384 * Description:
385 * Some devices have an internal block size bigger than the reported
386 * hardware sector size. This function can be used to signal the
387 * smallest I/O the device can perform without incurring a performance
388 * penalty.
389 */
390void blk_queue_io_min(struct request_queue *q, unsigned int min)
391{
392 q->limits.io_min = min;
393
394 if (q->limits.io_min < q->limits.logical_block_size)
395 q->limits.io_min = q->limits.logical_block_size;
396
397 if (q->limits.io_min < q->limits.physical_block_size)
398 q->limits.io_min = q->limits.physical_block_size;
399}
400EXPORT_SYMBOL(blk_queue_io_min);
401
402/**
403 * blk_queue_io_opt - set optimal request size for the queue
404 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700405 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400406 *
407 * Description:
408 * Drivers can call this function to set the preferred I/O request
409 * size for devices that report such a value.
410 */
411void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
412{
413 q->limits.io_opt = opt;
414}
415EXPORT_SYMBOL(blk_queue_io_opt);
416
Jens Axboe86db1e22008-01-29 14:53:40 +0100417/*
418 * Returns the minimum that is _not_ zero, unless both are zero.
419 */
420#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
421
422/**
423 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
424 * @t: the stacking driver (top)
425 * @b: the underlying device (bottom)
426 **/
427void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
428{
429 /* zero is "infinity" */
Martin K. Petersen025146e2009-05-22 17:17:51 -0400430 t->limits.max_sectors = min_not_zero(queue_max_sectors(t),
431 queue_max_sectors(b));
Jens Axboe86db1e22008-01-29 14:53:40 +0100432
Martin K. Petersen025146e2009-05-22 17:17:51 -0400433 t->limits.max_hw_sectors = min_not_zero(queue_max_hw_sectors(t),
434 queue_max_hw_sectors(b));
435
436 t->limits.seg_boundary_mask = min_not_zero(queue_segment_boundary(t),
437 queue_segment_boundary(b));
438
439 t->limits.max_phys_segments = min_not_zero(queue_max_phys_segments(t),
440 queue_max_phys_segments(b));
441
442 t->limits.max_hw_segments = min_not_zero(queue_max_hw_segments(t),
443 queue_max_hw_segments(b));
444
445 t->limits.max_segment_size = min_not_zero(queue_max_segment_size(t),
446 queue_max_segment_size(b));
447
448 t->limits.logical_block_size = max(queue_logical_block_size(t),
449 queue_logical_block_size(b));
450
Neil Browne7e72bf2008-05-14 16:05:54 -0700451 if (!t->queue_lock)
452 WARN_ON_ONCE(1);
453 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
454 unsigned long flags;
455 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200456 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700457 spin_unlock_irqrestore(t->queue_lock, flags);
458 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100459}
Jens Axboe86db1e22008-01-29 14:53:40 +0100460EXPORT_SYMBOL(blk_queue_stack_limits);
461
462/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400463 * blk_stack_limits - adjust queue_limits for stacked devices
464 * @t: the stacking driver limits (top)
Martin K. Petersen77634f32009-06-09 06:23:22 +0200465 * @b: the underlying queue limits (bottom)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400466 * @offset: offset to beginning of data within component device
467 *
468 * Description:
469 * Merges two queue_limit structs. Returns 0 if alignment didn't
470 * change. Returns -1 if adding the bottom device caused
471 * misalignment.
472 */
473int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
474 sector_t offset)
475{
476 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
477 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200478 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400479
480 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
481 b->seg_boundary_mask);
482
483 t->max_phys_segments = min_not_zero(t->max_phys_segments,
484 b->max_phys_segments);
485
486 t->max_hw_segments = min_not_zero(t->max_hw_segments,
487 b->max_hw_segments);
488
489 t->max_segment_size = min_not_zero(t->max_segment_size,
490 b->max_segment_size);
491
492 t->logical_block_size = max(t->logical_block_size,
493 b->logical_block_size);
494
495 t->physical_block_size = max(t->physical_block_size,
496 b->physical_block_size);
497
498 t->io_min = max(t->io_min, b->io_min);
499 t->no_cluster |= b->no_cluster;
500
501 /* Bottom device offset aligned? */
502 if (offset &&
503 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
504 t->misaligned = 1;
505 return -1;
506 }
507
508 /* If top has no alignment offset, inherit from bottom */
509 if (!t->alignment_offset)
510 t->alignment_offset =
511 b->alignment_offset & (b->physical_block_size - 1);
512
513 /* Top device aligned on logical block boundary? */
514 if (t->alignment_offset & (t->logical_block_size - 1)) {
515 t->misaligned = 1;
516 return -1;
517 }
518
519 return 0;
520}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200521EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400522
523/**
524 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200525 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400526 * @bdev: the underlying block device (bottom)
527 * @offset: offset to beginning of data within component device
528 *
529 * Description:
530 * Merges the limits for two queues. Returns 0 if alignment
531 * didn't change. Returns -1 if adding the bottom device caused
532 * misalignment.
533 */
534void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
535 sector_t offset)
536{
537 struct request_queue *t = disk->queue;
538 struct request_queue *b = bdev_get_queue(bdev);
539
540 offset += get_start_sect(bdev) << 9;
541
542 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
543 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
544
545 disk_name(disk, 0, top);
546 bdevname(bdev, bottom);
547
548 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
549 top, bottom);
550 }
551
552 if (!t->queue_lock)
553 WARN_ON_ONCE(1);
554 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
555 unsigned long flags;
556
557 spin_lock_irqsave(t->queue_lock, flags);
558 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
559 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
560 spin_unlock_irqrestore(t->queue_lock, flags);
561 }
562}
563EXPORT_SYMBOL(disk_stack_limits);
564
565/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100566 * blk_queue_dma_pad - set pad mask
567 * @q: the request queue for the device
568 * @mask: pad mask
569 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200570 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100571 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200572 * Appending pad buffer to a request modifies the last entry of a
573 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100574 **/
575void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
576{
577 q->dma_pad_mask = mask;
578}
579EXPORT_SYMBOL(blk_queue_dma_pad);
580
581/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200582 * blk_queue_update_dma_pad - update pad mask
583 * @q: the request queue for the device
584 * @mask: pad mask
585 *
586 * Update dma pad mask.
587 *
588 * Appending pad buffer to a request modifies the last entry of a
589 * scatter list such that it includes the pad buffer.
590 **/
591void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
592{
593 if (mask > q->dma_pad_mask)
594 q->dma_pad_mask = mask;
595}
596EXPORT_SYMBOL(blk_queue_update_dma_pad);
597
598/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100599 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100600 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100601 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100602 * @buf: physically contiguous buffer
603 * @size: size of the buffer in bytes
604 *
605 * Some devices have excess DMA problems and can't simply discard (or
606 * zero fill) the unwanted piece of the transfer. They have to have a
607 * real area of memory to transfer it into. The use case for this is
608 * ATAPI devices in DMA mode. If the packet command causes a transfer
609 * bigger than the transfer size some HBAs will lock up if there
610 * aren't DMA elements to contain the excess transfer. What this API
611 * does is adjust the queue so that the buf is always appended
612 * silently to the scatterlist.
613 *
614 * Note: This routine adjusts max_hw_segments to make room for
615 * appending the drain buffer. If you call
616 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
617 * calling this routine, you must set the limit to one fewer than your
618 * device can support otherwise there won't be room for the drain
619 * buffer.
620 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100621int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100622 dma_drain_needed_fn *dma_drain_needed,
623 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100624{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400625 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100626 return -EINVAL;
627 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400628 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
629 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100630 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100631 q->dma_drain_buffer = buf;
632 q->dma_drain_size = size;
633
634 return 0;
635}
Jens Axboe86db1e22008-01-29 14:53:40 +0100636EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
637
638/**
639 * blk_queue_segment_boundary - set boundary rules for segment merging
640 * @q: the request queue for the device
641 * @mask: the memory boundary mask
642 **/
643void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
644{
645 if (mask < PAGE_CACHE_SIZE - 1) {
646 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700647 printk(KERN_INFO "%s: set to minimum %lx\n",
648 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100649 }
650
Martin K. Petersen025146e2009-05-22 17:17:51 -0400651 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100652}
Jens Axboe86db1e22008-01-29 14:53:40 +0100653EXPORT_SYMBOL(blk_queue_segment_boundary);
654
655/**
656 * blk_queue_dma_alignment - set dma length and memory alignment
657 * @q: the request queue for the device
658 * @mask: alignment mask
659 *
660 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200661 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100662 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100663 *
664 **/
665void blk_queue_dma_alignment(struct request_queue *q, int mask)
666{
667 q->dma_alignment = mask;
668}
Jens Axboe86db1e22008-01-29 14:53:40 +0100669EXPORT_SYMBOL(blk_queue_dma_alignment);
670
671/**
672 * blk_queue_update_dma_alignment - update dma length and memory alignment
673 * @q: the request queue for the device
674 * @mask: alignment mask
675 *
676 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200677 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100678 * If the requested alignment is larger than the current alignment, then
679 * the current queue alignment is updated to the new value, otherwise it
680 * is left alone. The design of this is to allow multiple objects
681 * (driver, device, transport etc) to set their respective
682 * alignments without having them interfere.
683 *
684 **/
685void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
686{
687 BUG_ON(mask > PAGE_SIZE);
688
689 if (mask > q->dma_alignment)
690 q->dma_alignment = mask;
691}
Jens Axboe86db1e22008-01-29 14:53:40 +0100692EXPORT_SYMBOL(blk_queue_update_dma_alignment);
693
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200694static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100695{
696 blk_max_low_pfn = max_low_pfn - 1;
697 blk_max_pfn = max_pfn - 1;
698 return 0;
699}
700subsys_initcall(blk_settings_init);