blob: 476d870650737eda44dc3177675a34afae63fd53 [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 */
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -040010#include <linux/gcd.h>
Jens Axboe86db1e22008-01-29 14:53:40 +010011
12#include "blk.h"
13
Jens Axboe6728cb02008-01-31 13:03:55 +010014unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010015EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010016
17unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010018
19/**
20 * blk_queue_prep_rq - set a prepare_request function for queue
21 * @q: queue
22 * @pfn: prepare_request function
23 *
24 * It's possible for a queue to register a prepare_request callback which
25 * is invoked before the request is handed to the request_fn. The goal of
26 * the function is to prepare a request for I/O, it can be used to build a
27 * cdb from the request data for instance.
28 *
29 */
30void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
31{
32 q->prep_rq_fn = pfn;
33}
Jens Axboe86db1e22008-01-29 14:53:40 +010034EXPORT_SYMBOL(blk_queue_prep_rq);
35
36/**
David Woodhousefb2dce82008-08-05 18:01:53 +010037 * blk_queue_set_discard - set a discard_sectors function for queue
38 * @q: queue
39 * @dfn: prepare_discard function
40 *
41 * It's possible for a queue to register a discard callback which is used
42 * to transform a discard request into the appropriate type for the
43 * hardware. If none is registered, then discard requests are failed
44 * with %EOPNOTSUPP.
45 *
46 */
47void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
48{
49 q->prepare_discard_fn = dfn;
50}
51EXPORT_SYMBOL(blk_queue_set_discard);
52
53/**
Jens Axboe86db1e22008-01-29 14:53:40 +010054 * blk_queue_merge_bvec - set a merge_bvec function for queue
55 * @q: queue
56 * @mbfn: merge_bvec_fn
57 *
58 * Usually queues have static limitations on the max sectors or segments that
59 * we can put in a request. Stacking drivers may have some settings that
60 * are dynamic, and thus we have to query the queue whether it is ok to
61 * add a new bio_vec to a bio at a given offset or not. If the block device
62 * has such limitations, it needs to register a merge_bvec_fn to control
63 * the size of bio's sent to it. Note that a block device *must* allow a
64 * single page to be added to an empty bio. The block device driver may want
65 * to use the bio_split() function to deal with these bio's. By default
66 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
67 * honored.
68 */
69void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
70{
71 q->merge_bvec_fn = mbfn;
72}
Jens Axboe86db1e22008-01-29 14:53:40 +010073EXPORT_SYMBOL(blk_queue_merge_bvec);
74
75void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
76{
77 q->softirq_done_fn = fn;
78}
Jens Axboe86db1e22008-01-29 14:53:40 +010079EXPORT_SYMBOL(blk_queue_softirq_done);
80
Jens Axboe242f9dc2008-09-14 05:55:09 -070081void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
82{
83 q->rq_timeout = timeout;
84}
85EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
86
87void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
88{
89 q->rq_timed_out_fn = fn;
90}
91EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
92
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020093void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
94{
95 q->lld_busy_fn = fn;
96}
97EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
98
Jens Axboe86db1e22008-01-29 14:53:40 +010099/**
Martin K. Petersene475bba2009-06-16 08:23:52 +0200100 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5ca2009-06-19 09:18:32 +0200101 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +0200102 *
103 * Description:
104 * Returns a queue_limit struct to its default state. Can be used by
105 * stacking drivers like DM that stage table swaps and reuse an
106 * existing device queue.
107 */
108void blk_set_default_limits(struct queue_limits *lim)
109{
110 lim->max_phys_segments = MAX_PHYS_SEGMENTS;
111 lim->max_hw_segments = MAX_HW_SEGMENTS;
112 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
113 lim->max_segment_size = MAX_SEGMENT_SIZE;
114 lim->max_sectors = lim->max_hw_sectors = SAFE_MAX_SECTORS;
115 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200116 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200117 lim->alignment_offset = 0;
118 lim->io_opt = 0;
119 lim->misaligned = 0;
120 lim->no_cluster = 0;
121}
122EXPORT_SYMBOL(blk_set_default_limits);
123
124/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100125 * blk_queue_make_request - define an alternate make_request function for a device
126 * @q: the request queue for the device to be affected
127 * @mfn: the alternate make_request function
128 *
129 * Description:
130 * The normal way for &struct bios to be passed to a device
131 * driver is for them to be collected into requests on a request
132 * queue, and then to allow the device driver to select requests
133 * off that queue when it is ready. This works well for many block
134 * devices. However some block devices (typically virtual devices
135 * such as md or lvm) do not benefit from the processing on the
136 * request queue, and are served best by having the requests passed
137 * directly to them. This can be achieved by providing a function
138 * to blk_queue_make_request().
139 *
140 * Caveat:
141 * The driver that does this *must* be able to deal appropriately
142 * with buffers in "highmemory". This can be accomplished by either calling
143 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
144 * blk_queue_bounce() to create a buffer in normal memory.
145 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100146void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100147{
148 /*
149 * set defaults
150 */
151 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100152
Jens Axboe86db1e22008-01-29 14:53:40 +0100153 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100154 blk_queue_dma_alignment(q, 511);
155 blk_queue_congestion_threshold(q);
156 q->nr_batching = BLK_BATCH_REQ;
157
158 q->unplug_thresh = 4; /* hmm */
159 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
160 if (q->unplug_delay == 0)
161 q->unplug_delay = 1;
162
Jens Axboe86db1e22008-01-29 14:53:40 +0100163 q->unplug_timer.function = blk_unplug_timeout;
164 q->unplug_timer.data = (unsigned long)q;
165
Martin K. Petersene475bba2009-06-16 08:23:52 +0200166 blk_set_default_limits(&q->limits);
167
Jens Axboe86db1e22008-01-29 14:53:40 +0100168 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200169 * If the caller didn't supply a lock, fall back to our embedded
170 * per-queue locks
171 */
172 if (!q->queue_lock)
173 q->queue_lock = &q->__queue_lock;
174
175 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100176 * by default assume old behaviour and bounce for any highmem page
177 */
178 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
179}
Jens Axboe86db1e22008-01-29 14:53:40 +0100180EXPORT_SYMBOL(blk_queue_make_request);
181
182/**
183 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900184 * @q: the request queue for the device
185 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100186 *
187 * Description:
188 * Different hardware can have different requirements as to what pages
189 * it can do I/O directly to. A low level driver can call
190 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900191 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100192 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900193void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100194{
Tejun Heocd0aca22009-04-15 22:10:25 +0900195 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100196 int dma = 0;
197
198 q->bounce_gfp = GFP_NOIO;
199#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900200 /*
201 * Assume anything <= 4GB can be handled by IOMMU. Actually
202 * some IOMMUs can handle everything, but I don't know of a
203 * way to test this here.
204 */
205 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100206 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400207 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100208#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100209 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100210 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400211 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100212#endif
213 if (dma) {
214 init_emergency_isa_pool();
215 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400216 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100217 }
218}
Jens Axboe86db1e22008-01-29 14:53:40 +0100219EXPORT_SYMBOL(blk_queue_bounce_limit);
220
221/**
222 * blk_queue_max_sectors - set max sectors for a request for this queue
223 * @q: the request queue for the device
224 * @max_sectors: max sectors in the usual 512b unit
225 *
226 * Description:
227 * Enables a low level driver to set an upper limit on the size of
228 * received requests.
229 **/
230void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
231{
232 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
233 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700234 printk(KERN_INFO "%s: set to minimum %d\n",
235 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100236 }
237
238 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400239 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100240 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400241 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
242 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100243 }
244}
Jens Axboe86db1e22008-01-29 14:53:40 +0100245EXPORT_SYMBOL(blk_queue_max_sectors);
246
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400247void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
248{
249 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400250 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400251 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400252 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400253}
254EXPORT_SYMBOL(blk_queue_max_hw_sectors);
255
Jens Axboe86db1e22008-01-29 14:53:40 +0100256/**
257 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
258 * @q: the request queue for the device
259 * @max_segments: max number of segments
260 *
261 * Description:
262 * Enables a low level driver to set an upper limit on the number of
263 * physical data segments in a request. This would be the largest sized
264 * scatter list the driver could handle.
265 **/
266void blk_queue_max_phys_segments(struct request_queue *q,
267 unsigned short max_segments)
268{
269 if (!max_segments) {
270 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700271 printk(KERN_INFO "%s: set to minimum %d\n",
272 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100273 }
274
Martin K. Petersen025146e2009-05-22 17:17:51 -0400275 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100276}
Jens Axboe86db1e22008-01-29 14:53:40 +0100277EXPORT_SYMBOL(blk_queue_max_phys_segments);
278
279/**
280 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
281 * @q: the request queue for the device
282 * @max_segments: max number of segments
283 *
284 * Description:
285 * Enables a low level driver to set an upper limit on the number of
286 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200287 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100288 * to the device.
289 **/
290void blk_queue_max_hw_segments(struct request_queue *q,
291 unsigned short max_segments)
292{
293 if (!max_segments) {
294 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700295 printk(KERN_INFO "%s: set to minimum %d\n",
296 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100297 }
298
Martin K. Petersen025146e2009-05-22 17:17:51 -0400299 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100300}
Jens Axboe86db1e22008-01-29 14:53:40 +0100301EXPORT_SYMBOL(blk_queue_max_hw_segments);
302
303/**
304 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
305 * @q: the request queue for the device
306 * @max_size: max size of segment in bytes
307 *
308 * Description:
309 * Enables a low level driver to set an upper limit on the size of a
310 * coalesced segment
311 **/
312void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
313{
314 if (max_size < PAGE_CACHE_SIZE) {
315 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700316 printk(KERN_INFO "%s: set to minimum %d\n",
317 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100318 }
319
Martin K. Petersen025146e2009-05-22 17:17:51 -0400320 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100321}
Jens Axboe86db1e22008-01-29 14:53:40 +0100322EXPORT_SYMBOL(blk_queue_max_segment_size);
323
324/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400325 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100326 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400327 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100328 *
329 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400330 * This should be set to the lowest possible block size that the
331 * storage device can address. The default of 512 covers most
332 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100333 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400334void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100335{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400336 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400337
338 if (q->limits.physical_block_size < size)
339 q->limits.physical_block_size = size;
340
341 if (q->limits.io_min < q->limits.physical_block_size)
342 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100343}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400344EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100345
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400346/**
347 * blk_queue_physical_block_size - set physical block size for the queue
348 * @q: the request queue for the device
349 * @size: the physical block size, in bytes
350 *
351 * Description:
352 * This should be set to the lowest possible sector size that the
353 * hardware can operate on without reverting to read-modify-write
354 * operations.
355 */
356void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
357{
358 q->limits.physical_block_size = size;
359
360 if (q->limits.physical_block_size < q->limits.logical_block_size)
361 q->limits.physical_block_size = q->limits.logical_block_size;
362
363 if (q->limits.io_min < q->limits.physical_block_size)
364 q->limits.io_min = q->limits.physical_block_size;
365}
366EXPORT_SYMBOL(blk_queue_physical_block_size);
367
368/**
369 * blk_queue_alignment_offset - set physical block alignment offset
370 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700371 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400372 *
373 * Description:
374 * Some devices are naturally misaligned to compensate for things like
375 * the legacy DOS partition table 63-sector offset. Low-level drivers
376 * should call this function for devices whose first sector is not
377 * naturally aligned.
378 */
379void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
380{
381 q->limits.alignment_offset =
382 offset & (q->limits.physical_block_size - 1);
383 q->limits.misaligned = 0;
384}
385EXPORT_SYMBOL(blk_queue_alignment_offset);
386
387/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400388 * blk_limits_io_min - set minimum request size for a device
389 * @limits: the queue limits
390 * @min: smallest I/O size in bytes
391 *
392 * Description:
393 * Some devices have an internal block size bigger than the reported
394 * hardware sector size. This function can be used to signal the
395 * smallest I/O the device can perform without incurring a performance
396 * penalty.
397 */
398void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
399{
400 limits->io_min = min;
401
402 if (limits->io_min < limits->logical_block_size)
403 limits->io_min = limits->logical_block_size;
404
405 if (limits->io_min < limits->physical_block_size)
406 limits->io_min = limits->physical_block_size;
407}
408EXPORT_SYMBOL(blk_limits_io_min);
409
410/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400411 * blk_queue_io_min - set minimum request size for the queue
412 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700413 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400414 *
415 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400416 * Storage devices may report a granularity or preferred minimum I/O
417 * size which is the smallest request the device can perform without
418 * incurring a performance penalty. For disk drives this is often the
419 * physical block size. For RAID arrays it is often the stripe chunk
420 * size. A properly aligned multiple of minimum_io_size is the
421 * preferred request size for workloads where a high number of I/O
422 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400423 */
424void blk_queue_io_min(struct request_queue *q, unsigned int min)
425{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400426 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400427}
428EXPORT_SYMBOL(blk_queue_io_min);
429
430/**
431 * blk_queue_io_opt - set optimal request size for the queue
432 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700433 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400434 *
435 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400436 * Storage devices may report an optimal I/O size, which is the
437 * device's preferred unit for sustained I/O. This is rarely reported
438 * for disk drives. For RAID arrays it is usually the stripe width or
439 * the internal track size. A properly aligned multiple of
440 * optimal_io_size is the preferred request size for workloads where
441 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400442 */
443void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
444{
445 q->limits.io_opt = opt;
446}
447EXPORT_SYMBOL(blk_queue_io_opt);
448
Jens Axboe86db1e22008-01-29 14:53:40 +0100449/*
450 * Returns the minimum that is _not_ zero, unless both are zero.
451 */
452#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
453
454/**
455 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
456 * @t: the stacking driver (top)
457 * @b: the underlying device (bottom)
458 **/
459void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
460{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400461 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400462
Neil Browne7e72bf2008-05-14 16:05:54 -0700463 if (!t->queue_lock)
464 WARN_ON_ONCE(1);
465 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
466 unsigned long flags;
467 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200468 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700469 spin_unlock_irqrestore(t->queue_lock, flags);
470 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100471}
Jens Axboe86db1e22008-01-29 14:53:40 +0100472EXPORT_SYMBOL(blk_queue_stack_limits);
473
474/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400475 * blk_stack_limits - adjust queue_limits for stacked devices
476 * @t: the stacking driver limits (top)
Martin K. Petersen77634f32009-06-09 06:23:22 +0200477 * @b: the underlying queue limits (bottom)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400478 * @offset: offset to beginning of data within component device
479 *
480 * Description:
481 * Merges two queue_limit structs. Returns 0 if alignment didn't
482 * change. Returns -1 if adding the bottom device caused
483 * misalignment.
484 */
485int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
486 sector_t offset)
487{
488 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
489 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200490 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400491
492 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
493 b->seg_boundary_mask);
494
495 t->max_phys_segments = min_not_zero(t->max_phys_segments,
496 b->max_phys_segments);
497
498 t->max_hw_segments = min_not_zero(t->max_hw_segments,
499 b->max_hw_segments);
500
501 t->max_segment_size = min_not_zero(t->max_segment_size,
502 b->max_segment_size);
503
504 t->logical_block_size = max(t->logical_block_size,
505 b->logical_block_size);
506
507 t->physical_block_size = max(t->physical_block_size,
508 b->physical_block_size);
509
510 t->io_min = max(t->io_min, b->io_min);
511 t->no_cluster |= b->no_cluster;
512
513 /* Bottom device offset aligned? */
514 if (offset &&
515 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
516 t->misaligned = 1;
517 return -1;
518 }
519
520 /* If top has no alignment offset, inherit from bottom */
521 if (!t->alignment_offset)
522 t->alignment_offset =
523 b->alignment_offset & (b->physical_block_size - 1);
524
525 /* Top device aligned on logical block boundary? */
526 if (t->alignment_offset & (t->logical_block_size - 1)) {
527 t->misaligned = 1;
528 return -1;
529 }
530
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400531 /* Find lcm() of optimal I/O size */
532 if (t->io_opt && b->io_opt)
533 t->io_opt = (t->io_opt * b->io_opt) / gcd(t->io_opt, b->io_opt);
534 else if (b->io_opt)
535 t->io_opt = b->io_opt;
536
537 /* Verify that optimal I/O size is a multiple of io_min */
538 if (t->io_min && t->io_opt % t->io_min)
539 return -1;
540
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400541 return 0;
542}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200543EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400544
545/**
546 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200547 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400548 * @bdev: the underlying block device (bottom)
549 * @offset: offset to beginning of data within component device
550 *
551 * Description:
552 * Merges the limits for two queues. Returns 0 if alignment
553 * didn't change. Returns -1 if adding the bottom device caused
554 * misalignment.
555 */
556void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
557 sector_t offset)
558{
559 struct request_queue *t = disk->queue;
560 struct request_queue *b = bdev_get_queue(bdev);
561
562 offset += get_start_sect(bdev) << 9;
563
564 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
565 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
566
567 disk_name(disk, 0, top);
568 bdevname(bdev, bottom);
569
570 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
571 top, bottom);
572 }
573
574 if (!t->queue_lock)
575 WARN_ON_ONCE(1);
576 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
577 unsigned long flags;
578
579 spin_lock_irqsave(t->queue_lock, flags);
580 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
581 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
582 spin_unlock_irqrestore(t->queue_lock, flags);
583 }
584}
585EXPORT_SYMBOL(disk_stack_limits);
586
587/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100588 * blk_queue_dma_pad - set pad mask
589 * @q: the request queue for the device
590 * @mask: pad mask
591 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200592 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100593 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200594 * Appending pad buffer to a request modifies the last entry of a
595 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100596 **/
597void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
598{
599 q->dma_pad_mask = mask;
600}
601EXPORT_SYMBOL(blk_queue_dma_pad);
602
603/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200604 * blk_queue_update_dma_pad - update pad mask
605 * @q: the request queue for the device
606 * @mask: pad mask
607 *
608 * Update dma pad mask.
609 *
610 * Appending pad buffer to a request modifies the last entry of a
611 * scatter list such that it includes the pad buffer.
612 **/
613void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
614{
615 if (mask > q->dma_pad_mask)
616 q->dma_pad_mask = mask;
617}
618EXPORT_SYMBOL(blk_queue_update_dma_pad);
619
620/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100621 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100622 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100623 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100624 * @buf: physically contiguous buffer
625 * @size: size of the buffer in bytes
626 *
627 * Some devices have excess DMA problems and can't simply discard (or
628 * zero fill) the unwanted piece of the transfer. They have to have a
629 * real area of memory to transfer it into. The use case for this is
630 * ATAPI devices in DMA mode. If the packet command causes a transfer
631 * bigger than the transfer size some HBAs will lock up if there
632 * aren't DMA elements to contain the excess transfer. What this API
633 * does is adjust the queue so that the buf is always appended
634 * silently to the scatterlist.
635 *
636 * Note: This routine adjusts max_hw_segments to make room for
637 * appending the drain buffer. If you call
638 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
639 * calling this routine, you must set the limit to one fewer than your
640 * device can support otherwise there won't be room for the drain
641 * buffer.
642 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100643int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100644 dma_drain_needed_fn *dma_drain_needed,
645 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100646{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400647 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100648 return -EINVAL;
649 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400650 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
651 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100652 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100653 q->dma_drain_buffer = buf;
654 q->dma_drain_size = size;
655
656 return 0;
657}
Jens Axboe86db1e22008-01-29 14:53:40 +0100658EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
659
660/**
661 * blk_queue_segment_boundary - set boundary rules for segment merging
662 * @q: the request queue for the device
663 * @mask: the memory boundary mask
664 **/
665void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
666{
667 if (mask < PAGE_CACHE_SIZE - 1) {
668 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700669 printk(KERN_INFO "%s: set to minimum %lx\n",
670 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100671 }
672
Martin K. Petersen025146e2009-05-22 17:17:51 -0400673 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100674}
Jens Axboe86db1e22008-01-29 14:53:40 +0100675EXPORT_SYMBOL(blk_queue_segment_boundary);
676
677/**
678 * blk_queue_dma_alignment - set dma length and memory alignment
679 * @q: the request queue for the device
680 * @mask: alignment mask
681 *
682 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200683 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100684 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100685 *
686 **/
687void blk_queue_dma_alignment(struct request_queue *q, int mask)
688{
689 q->dma_alignment = mask;
690}
Jens Axboe86db1e22008-01-29 14:53:40 +0100691EXPORT_SYMBOL(blk_queue_dma_alignment);
692
693/**
694 * blk_queue_update_dma_alignment - update dma length and memory alignment
695 * @q: the request queue for the device
696 * @mask: alignment mask
697 *
698 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200699 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100700 * If the requested alignment is larger than the current alignment, then
701 * the current queue alignment is updated to the new value, otherwise it
702 * is left alone. The design of this is to allow multiple objects
703 * (driver, device, transport etc) to set their respective
704 * alignments without having them interfere.
705 *
706 **/
707void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
708{
709 BUG_ON(mask > PAGE_SIZE);
710
711 if (mask > q->dma_alignment)
712 q->dma_alignment = mask;
713}
Jens Axboe86db1e22008-01-29 14:53:40 +0100714EXPORT_SYMBOL(blk_queue_update_dma_alignment);
715
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200716static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100717{
718 blk_max_low_pfn = max_low_pfn - 1;
719 blk_max_pfn = max_pfn - 1;
720 return 0;
721}
722subsys_initcall(blk_settings_init);