blob: eaf122ff5f16b67d2339acdd2186a4544795213a [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;
Martin K. Petersen5dee2472009-09-21 21:46:05 +0200114 lim->max_sectors = BLK_DEF_MAX_SECTORS;
115 lim->max_hw_sectors = INT_MAX;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200116 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200117 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200118 lim->alignment_offset = 0;
119 lim->io_opt = 0;
120 lim->misaligned = 0;
121 lim->no_cluster = 0;
122}
123EXPORT_SYMBOL(blk_set_default_limits);
124
125/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100126 * blk_queue_make_request - define an alternate make_request function for a device
127 * @q: the request queue for the device to be affected
128 * @mfn: the alternate make_request function
129 *
130 * Description:
131 * The normal way for &struct bios to be passed to a device
132 * driver is for them to be collected into requests on a request
133 * queue, and then to allow the device driver to select requests
134 * off that queue when it is ready. This works well for many block
135 * devices. However some block devices (typically virtual devices
136 * such as md or lvm) do not benefit from the processing on the
137 * request queue, and are served best by having the requests passed
138 * directly to them. This can be achieved by providing a function
139 * to blk_queue_make_request().
140 *
141 * Caveat:
142 * The driver that does this *must* be able to deal appropriately
143 * with buffers in "highmemory". This can be accomplished by either calling
144 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
145 * blk_queue_bounce() to create a buffer in normal memory.
146 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100147void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100148{
149 /*
150 * set defaults
151 */
152 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100153
Jens Axboe86db1e22008-01-29 14:53:40 +0100154 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100155 blk_queue_dma_alignment(q, 511);
156 blk_queue_congestion_threshold(q);
157 q->nr_batching = BLK_BATCH_REQ;
158
159 q->unplug_thresh = 4; /* hmm */
160 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
161 if (q->unplug_delay == 0)
162 q->unplug_delay = 1;
163
Jens Axboe86db1e22008-01-29 14:53:40 +0100164 q->unplug_timer.function = blk_unplug_timeout;
165 q->unplug_timer.data = (unsigned long)q;
166
Martin K. Petersene475bba2009-06-16 08:23:52 +0200167 blk_set_default_limits(&q->limits);
Martin K. Petersen80ddf242009-09-18 22:54:37 +0200168 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200169
Jens Axboe86db1e22008-01-29 14:53:40 +0100170 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200171 * If the caller didn't supply a lock, fall back to our embedded
172 * per-queue locks
173 */
174 if (!q->queue_lock)
175 q->queue_lock = &q->__queue_lock;
176
177 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100178 * by default assume old behaviour and bounce for any highmem page
179 */
180 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
181}
Jens Axboe86db1e22008-01-29 14:53:40 +0100182EXPORT_SYMBOL(blk_queue_make_request);
183
184/**
185 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900186 * @q: the request queue for the device
187 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100188 *
189 * Description:
190 * Different hardware can have different requirements as to what pages
191 * it can do I/O directly to. A low level driver can call
192 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900193 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100194 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900195void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100196{
Tejun Heocd0aca22009-04-15 22:10:25 +0900197 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100198 int dma = 0;
199
200 q->bounce_gfp = GFP_NOIO;
201#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900202 /*
203 * Assume anything <= 4GB can be handled by IOMMU. Actually
204 * some IOMMUs can handle everything, but I don't know of a
205 * way to test this here.
206 */
207 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100208 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400209 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100210#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100211 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100212 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400213 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100214#endif
215 if (dma) {
216 init_emergency_isa_pool();
217 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400218 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100219 }
220}
Jens Axboe86db1e22008-01-29 14:53:40 +0100221EXPORT_SYMBOL(blk_queue_bounce_limit);
222
223/**
224 * blk_queue_max_sectors - set max sectors for a request for this queue
225 * @q: the request queue for the device
226 * @max_sectors: max sectors in the usual 512b unit
227 *
228 * Description:
229 * Enables a low level driver to set an upper limit on the size of
230 * received requests.
231 **/
232void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
233{
234 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
235 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700236 printk(KERN_INFO "%s: set to minimum %d\n",
237 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100238 }
239
240 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400241 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100242 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400243 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
244 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100245 }
246}
Jens Axboe86db1e22008-01-29 14:53:40 +0100247EXPORT_SYMBOL(blk_queue_max_sectors);
248
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400249void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
250{
251 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400252 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400253 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400254 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400255}
256EXPORT_SYMBOL(blk_queue_max_hw_sectors);
257
Jens Axboe86db1e22008-01-29 14:53:40 +0100258/**
259 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
260 * @q: the request queue for the device
261 * @max_segments: max number of segments
262 *
263 * Description:
264 * Enables a low level driver to set an upper limit on the number of
265 * physical data segments in a request. This would be the largest sized
266 * scatter list the driver could handle.
267 **/
268void blk_queue_max_phys_segments(struct request_queue *q,
269 unsigned short max_segments)
270{
271 if (!max_segments) {
272 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700273 printk(KERN_INFO "%s: set to minimum %d\n",
274 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100275 }
276
Martin K. Petersen025146e2009-05-22 17:17:51 -0400277 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100278}
Jens Axboe86db1e22008-01-29 14:53:40 +0100279EXPORT_SYMBOL(blk_queue_max_phys_segments);
280
281/**
282 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
283 * @q: the request queue for the device
284 * @max_segments: max number of segments
285 *
286 * Description:
287 * Enables a low level driver to set an upper limit on the number of
288 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200289 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100290 * to the device.
291 **/
292void blk_queue_max_hw_segments(struct request_queue *q,
293 unsigned short max_segments)
294{
295 if (!max_segments) {
296 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700297 printk(KERN_INFO "%s: set to minimum %d\n",
298 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100299 }
300
Martin K. Petersen025146e2009-05-22 17:17:51 -0400301 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100302}
Jens Axboe86db1e22008-01-29 14:53:40 +0100303EXPORT_SYMBOL(blk_queue_max_hw_segments);
304
305/**
306 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
307 * @q: the request queue for the device
308 * @max_size: max size of segment in bytes
309 *
310 * Description:
311 * Enables a low level driver to set an upper limit on the size of a
312 * coalesced segment
313 **/
314void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
315{
316 if (max_size < PAGE_CACHE_SIZE) {
317 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700318 printk(KERN_INFO "%s: set to minimum %d\n",
319 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100320 }
321
Martin K. Petersen025146e2009-05-22 17:17:51 -0400322 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100323}
Jens Axboe86db1e22008-01-29 14:53:40 +0100324EXPORT_SYMBOL(blk_queue_max_segment_size);
325
326/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400327 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100328 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400329 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100330 *
331 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400332 * This should be set to the lowest possible block size that the
333 * storage device can address. The default of 512 covers most
334 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100335 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400336void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100337{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400338 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400339
340 if (q->limits.physical_block_size < size)
341 q->limits.physical_block_size = size;
342
343 if (q->limits.io_min < q->limits.physical_block_size)
344 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100345}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400346EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100347
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400348/**
349 * blk_queue_physical_block_size - set physical block size for the queue
350 * @q: the request queue for the device
351 * @size: the physical block size, in bytes
352 *
353 * Description:
354 * This should be set to the lowest possible sector size that the
355 * hardware can operate on without reverting to read-modify-write
356 * operations.
357 */
358void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
359{
360 q->limits.physical_block_size = size;
361
362 if (q->limits.physical_block_size < q->limits.logical_block_size)
363 q->limits.physical_block_size = q->limits.logical_block_size;
364
365 if (q->limits.io_min < q->limits.physical_block_size)
366 q->limits.io_min = q->limits.physical_block_size;
367}
368EXPORT_SYMBOL(blk_queue_physical_block_size);
369
370/**
371 * blk_queue_alignment_offset - set physical block alignment offset
372 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700373 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400374 *
375 * Description:
376 * Some devices are naturally misaligned to compensate for things like
377 * the legacy DOS partition table 63-sector offset. Low-level drivers
378 * should call this function for devices whose first sector is not
379 * naturally aligned.
380 */
381void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
382{
383 q->limits.alignment_offset =
384 offset & (q->limits.physical_block_size - 1);
385 q->limits.misaligned = 0;
386}
387EXPORT_SYMBOL(blk_queue_alignment_offset);
388
389/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400390 * blk_limits_io_min - set minimum request size for a device
391 * @limits: the queue limits
392 * @min: smallest I/O size in bytes
393 *
394 * Description:
395 * Some devices have an internal block size bigger than the reported
396 * hardware sector size. This function can be used to signal the
397 * smallest I/O the device can perform without incurring a performance
398 * penalty.
399 */
400void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
401{
402 limits->io_min = min;
403
404 if (limits->io_min < limits->logical_block_size)
405 limits->io_min = limits->logical_block_size;
406
407 if (limits->io_min < limits->physical_block_size)
408 limits->io_min = limits->physical_block_size;
409}
410EXPORT_SYMBOL(blk_limits_io_min);
411
412/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400413 * blk_queue_io_min - set minimum request size for the queue
414 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700415 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400416 *
417 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400418 * Storage devices may report a granularity or preferred minimum I/O
419 * size which is the smallest request the device can perform without
420 * incurring a performance penalty. For disk drives this is often the
421 * physical block size. For RAID arrays it is often the stripe chunk
422 * size. A properly aligned multiple of minimum_io_size is the
423 * preferred request size for workloads where a high number of I/O
424 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400425 */
426void blk_queue_io_min(struct request_queue *q, unsigned int min)
427{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400428 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400429}
430EXPORT_SYMBOL(blk_queue_io_min);
431
432/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200433 * blk_limits_io_opt - set optimal request size for a device
434 * @limits: the queue limits
435 * @opt: smallest I/O size in bytes
436 *
437 * Description:
438 * Storage devices may report an optimal I/O size, which is the
439 * device's preferred unit for sustained I/O. This is rarely reported
440 * for disk drives. For RAID arrays it is usually the stripe width or
441 * the internal track size. A properly aligned multiple of
442 * optimal_io_size is the preferred request size for workloads where
443 * sustained throughput is desired.
444 */
445void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
446{
447 limits->io_opt = opt;
448}
449EXPORT_SYMBOL(blk_limits_io_opt);
450
451/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400452 * blk_queue_io_opt - set optimal request size for the queue
453 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700454 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400455 *
456 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400457 * Storage devices may report an optimal I/O size, which is the
458 * device's preferred unit for sustained I/O. This is rarely reported
459 * for disk drives. For RAID arrays it is usually the stripe width or
460 * the internal track size. A properly aligned multiple of
461 * optimal_io_size is the preferred request size for workloads where
462 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400463 */
464void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
465{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200466 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400467}
468EXPORT_SYMBOL(blk_queue_io_opt);
469
Jens Axboe86db1e22008-01-29 14:53:40 +0100470/*
471 * Returns the minimum that is _not_ zero, unless both are zero.
472 */
473#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
474
475/**
476 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
477 * @t: the stacking driver (top)
478 * @b: the underlying device (bottom)
479 **/
480void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
481{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400482 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400483
Neil Browne7e72bf2008-05-14 16:05:54 -0700484 if (!t->queue_lock)
485 WARN_ON_ONCE(1);
486 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
487 unsigned long flags;
488 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200489 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700490 spin_unlock_irqrestore(t->queue_lock, flags);
491 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100492}
Jens Axboe86db1e22008-01-29 14:53:40 +0100493EXPORT_SYMBOL(blk_queue_stack_limits);
494
495/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400496 * blk_stack_limits - adjust queue_limits for stacked devices
497 * @t: the stacking driver limits (top)
Martin K. Petersen77634f32009-06-09 06:23:22 +0200498 * @b: the underlying queue limits (bottom)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400499 * @offset: offset to beginning of data within component device
500 *
501 * Description:
502 * Merges two queue_limit structs. Returns 0 if alignment didn't
503 * change. Returns -1 if adding the bottom device caused
504 * misalignment.
505 */
506int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
507 sector_t offset)
508{
509 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
510 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200511 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400512
513 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
514 b->seg_boundary_mask);
515
516 t->max_phys_segments = min_not_zero(t->max_phys_segments,
517 b->max_phys_segments);
518
519 t->max_hw_segments = min_not_zero(t->max_hw_segments,
520 b->max_hw_segments);
521
522 t->max_segment_size = min_not_zero(t->max_segment_size,
523 b->max_segment_size);
524
525 t->logical_block_size = max(t->logical_block_size,
526 b->logical_block_size);
527
528 t->physical_block_size = max(t->physical_block_size,
529 b->physical_block_size);
530
531 t->io_min = max(t->io_min, b->io_min);
532 t->no_cluster |= b->no_cluster;
533
534 /* Bottom device offset aligned? */
535 if (offset &&
536 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
537 t->misaligned = 1;
538 return -1;
539 }
540
541 /* If top has no alignment offset, inherit from bottom */
542 if (!t->alignment_offset)
543 t->alignment_offset =
544 b->alignment_offset & (b->physical_block_size - 1);
545
546 /* Top device aligned on logical block boundary? */
547 if (t->alignment_offset & (t->logical_block_size - 1)) {
548 t->misaligned = 1;
549 return -1;
550 }
551
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400552 /* Find lcm() of optimal I/O size */
553 if (t->io_opt && b->io_opt)
554 t->io_opt = (t->io_opt * b->io_opt) / gcd(t->io_opt, b->io_opt);
555 else if (b->io_opt)
556 t->io_opt = b->io_opt;
557
558 /* Verify that optimal I/O size is a multiple of io_min */
559 if (t->io_min && t->io_opt % t->io_min)
560 return -1;
561
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400562 return 0;
563}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200564EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400565
566/**
567 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200568 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400569 * @bdev: the underlying block device (bottom)
570 * @offset: offset to beginning of data within component device
571 *
572 * Description:
573 * Merges the limits for two queues. Returns 0 if alignment
574 * didn't change. Returns -1 if adding the bottom device caused
575 * misalignment.
576 */
577void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
578 sector_t offset)
579{
580 struct request_queue *t = disk->queue;
581 struct request_queue *b = bdev_get_queue(bdev);
582
583 offset += get_start_sect(bdev) << 9;
584
585 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
586 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
587
588 disk_name(disk, 0, top);
589 bdevname(bdev, bottom);
590
591 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
592 top, bottom);
593 }
594
595 if (!t->queue_lock)
596 WARN_ON_ONCE(1);
597 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
598 unsigned long flags;
599
600 spin_lock_irqsave(t->queue_lock, flags);
601 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
602 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
603 spin_unlock_irqrestore(t->queue_lock, flags);
604 }
605}
606EXPORT_SYMBOL(disk_stack_limits);
607
608/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100609 * blk_queue_dma_pad - set pad mask
610 * @q: the request queue for the device
611 * @mask: pad mask
612 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200613 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100614 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200615 * Appending pad buffer to a request modifies the last entry of a
616 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100617 **/
618void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
619{
620 q->dma_pad_mask = mask;
621}
622EXPORT_SYMBOL(blk_queue_dma_pad);
623
624/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200625 * blk_queue_update_dma_pad - update pad mask
626 * @q: the request queue for the device
627 * @mask: pad mask
628 *
629 * Update dma pad mask.
630 *
631 * Appending pad buffer to a request modifies the last entry of a
632 * scatter list such that it includes the pad buffer.
633 **/
634void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
635{
636 if (mask > q->dma_pad_mask)
637 q->dma_pad_mask = mask;
638}
639EXPORT_SYMBOL(blk_queue_update_dma_pad);
640
641/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100642 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100643 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100644 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100645 * @buf: physically contiguous buffer
646 * @size: size of the buffer in bytes
647 *
648 * Some devices have excess DMA problems and can't simply discard (or
649 * zero fill) the unwanted piece of the transfer. They have to have a
650 * real area of memory to transfer it into. The use case for this is
651 * ATAPI devices in DMA mode. If the packet command causes a transfer
652 * bigger than the transfer size some HBAs will lock up if there
653 * aren't DMA elements to contain the excess transfer. What this API
654 * does is adjust the queue so that the buf is always appended
655 * silently to the scatterlist.
656 *
657 * Note: This routine adjusts max_hw_segments to make room for
658 * appending the drain buffer. If you call
659 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
660 * calling this routine, you must set the limit to one fewer than your
661 * device can support otherwise there won't be room for the drain
662 * buffer.
663 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100664int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100665 dma_drain_needed_fn *dma_drain_needed,
666 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100667{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400668 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100669 return -EINVAL;
670 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400671 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
672 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100673 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100674 q->dma_drain_buffer = buf;
675 q->dma_drain_size = size;
676
677 return 0;
678}
Jens Axboe86db1e22008-01-29 14:53:40 +0100679EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
680
681/**
682 * blk_queue_segment_boundary - set boundary rules for segment merging
683 * @q: the request queue for the device
684 * @mask: the memory boundary mask
685 **/
686void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
687{
688 if (mask < PAGE_CACHE_SIZE - 1) {
689 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700690 printk(KERN_INFO "%s: set to minimum %lx\n",
691 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100692 }
693
Martin K. Petersen025146e2009-05-22 17:17:51 -0400694 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100695}
Jens Axboe86db1e22008-01-29 14:53:40 +0100696EXPORT_SYMBOL(blk_queue_segment_boundary);
697
698/**
699 * blk_queue_dma_alignment - set dma length and memory alignment
700 * @q: the request queue for the device
701 * @mask: alignment mask
702 *
703 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200704 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100705 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100706 *
707 **/
708void blk_queue_dma_alignment(struct request_queue *q, int mask)
709{
710 q->dma_alignment = mask;
711}
Jens Axboe86db1e22008-01-29 14:53:40 +0100712EXPORT_SYMBOL(blk_queue_dma_alignment);
713
714/**
715 * blk_queue_update_dma_alignment - update dma length and memory alignment
716 * @q: the request queue for the device
717 * @mask: alignment mask
718 *
719 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200720 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100721 * If the requested alignment is larger than the current alignment, then
722 * the current queue alignment is updated to the new value, otherwise it
723 * is left alone. The design of this is to allow multiple objects
724 * (driver, device, transport etc) to set their respective
725 * alignments without having them interfere.
726 *
727 **/
728void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
729{
730 BUG_ON(mask > PAGE_SIZE);
731
732 if (mask > q->dma_alignment)
733 q->dma_alignment = mask;
734}
Jens Axboe86db1e22008-01-29 14:53:40 +0100735EXPORT_SYMBOL(blk_queue_update_dma_alignment);
736
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200737static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100738{
739 blk_max_low_pfn = max_low_pfn - 1;
740 blk_max_pfn = max_pfn - 1;
741 return 0;
742}
743subsys_initcall(blk_settings_init);