blob: d3234fc494adcd48ff1dd69c9feb117e93e333c0 [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>
Martin K. Petersen2cda2722010-03-15 12:46:51 +010011#include <linux/lcm.h>
Randy Dunlapad5ebd22009-11-11 13:47:45 +010012#include <linux/jiffies.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/gfp.h>
Jens Axboe86db1e22008-01-29 14:53:40 +010014
15#include "blk.h"
16
Jens Axboe6728cb02008-01-31 13:03:55 +010017unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010018EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010019
20unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010021
22/**
23 * blk_queue_prep_rq - set a prepare_request function for queue
24 * @q: queue
25 * @pfn: prepare_request function
26 *
27 * It's possible for a queue to register a prepare_request callback which
28 * is invoked before the request is handed to the request_fn. The goal of
29 * the function is to prepare a request for I/O, it can be used to build a
30 * cdb from the request data for instance.
31 *
32 */
33void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
34{
35 q->prep_rq_fn = pfn;
36}
Jens Axboe86db1e22008-01-29 14:53:40 +010037EXPORT_SYMBOL(blk_queue_prep_rq);
38
39/**
James Bottomley28018c22010-07-01 19:49:17 +090040 * blk_queue_unprep_rq - set an unprepare_request function for queue
41 * @q: queue
42 * @ufn: unprepare_request function
43 *
44 * It's possible for a queue to register an unprepare_request callback
45 * which is invoked before the request is finally completed. The goal
46 * of the function is to deallocate any data that was allocated in the
47 * prepare_request callback.
48 *
49 */
50void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
51{
52 q->unprep_rq_fn = ufn;
53}
54EXPORT_SYMBOL(blk_queue_unprep_rq);
55
56/**
Jens Axboe86db1e22008-01-29 14:53:40 +010057 * blk_queue_merge_bvec - set a merge_bvec function for queue
58 * @q: queue
59 * @mbfn: merge_bvec_fn
60 *
61 * Usually queues have static limitations on the max sectors or segments that
62 * we can put in a request. Stacking drivers may have some settings that
63 * are dynamic, and thus we have to query the queue whether it is ok to
64 * add a new bio_vec to a bio at a given offset or not. If the block device
65 * has such limitations, it needs to register a merge_bvec_fn to control
66 * the size of bio's sent to it. Note that a block device *must* allow a
67 * single page to be added to an empty bio. The block device driver may want
68 * to use the bio_split() function to deal with these bio's. By default
69 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
70 * honored.
71 */
72void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
73{
74 q->merge_bvec_fn = mbfn;
75}
Jens Axboe86db1e22008-01-29 14:53:40 +010076EXPORT_SYMBOL(blk_queue_merge_bvec);
77
78void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
79{
80 q->softirq_done_fn = fn;
81}
Jens Axboe86db1e22008-01-29 14:53:40 +010082EXPORT_SYMBOL(blk_queue_softirq_done);
83
Jens Axboe242f9dc2008-09-14 05:55:09 -070084void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
85{
86 q->rq_timeout = timeout;
87}
88EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
89
90void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
91{
92 q->rq_timed_out_fn = fn;
93}
94EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
95
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020096void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
97{
98 q->lld_busy_fn = fn;
99}
100EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
101
Jens Axboe86db1e22008-01-29 14:53:40 +0100102/**
Martin K. Petersene475bba2009-06-16 08:23:52 +0200103 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5ca2009-06-19 09:18:32 +0200104 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +0200105 *
106 * Description:
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100107 * Returns a queue_limit struct to its default state.
Martin K. Petersene475bba2009-06-16 08:23:52 +0200108 */
109void blk_set_default_limits(struct queue_limits *lim)
110{
Martin K. Petersen8a783622010-02-26 00:20:39 -0500111 lim->max_segments = BLK_MAX_SEGMENTS;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200112 lim->max_integrity_segments = 0;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200113 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
Martin K. Peterseneb28d312010-02-26 00:20:37 -0500114 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100115 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100116 lim->max_discard_sectors = 0;
117 lim->discard_granularity = 0;
118 lim->discard_alignment = 0;
119 lim->discard_misaligned = 0;
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100120 lim->discard_zeroes_data = 0;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200121 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200122 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200123 lim->alignment_offset = 0;
124 lim->io_opt = 0;
125 lim->misaligned = 0;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100126 lim->cluster = 1;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200127}
128EXPORT_SYMBOL(blk_set_default_limits);
129
130/**
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100131 * blk_set_stacking_limits - set default limits for stacking devices
132 * @lim: the queue_limits structure to reset
133 *
134 * Description:
135 * Returns a queue_limit struct to its default state. Should be used
136 * by stacking drivers like DM that have no internal limits.
137 */
138void blk_set_stacking_limits(struct queue_limits *lim)
139{
140 blk_set_default_limits(lim);
141
142 /* Inherit limits from component devices */
143 lim->discard_zeroes_data = 1;
144 lim->max_segments = USHRT_MAX;
145 lim->max_hw_sectors = UINT_MAX;
146
147 lim->max_sectors = BLK_DEF_MAX_SECTORS;
148}
149EXPORT_SYMBOL(blk_set_stacking_limits);
150
151/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100152 * blk_queue_make_request - define an alternate make_request function for a device
153 * @q: the request queue for the device to be affected
154 * @mfn: the alternate make_request function
155 *
156 * Description:
157 * The normal way for &struct bios to be passed to a device
158 * driver is for them to be collected into requests on a request
159 * queue, and then to allow the device driver to select requests
160 * off that queue when it is ready. This works well for many block
161 * devices. However some block devices (typically virtual devices
162 * such as md or lvm) do not benefit from the processing on the
163 * request queue, and are served best by having the requests passed
164 * directly to them. This can be achieved by providing a function
165 * to blk_queue_make_request().
166 *
167 * Caveat:
168 * The driver that does this *must* be able to deal appropriately
169 * with buffers in "highmemory". This can be accomplished by either calling
170 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
171 * blk_queue_bounce() to create a buffer in normal memory.
172 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100173void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100174{
175 /*
176 * set defaults
177 */
178 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100179
Jens Axboe86db1e22008-01-29 14:53:40 +0100180 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100181 blk_queue_dma_alignment(q, 511);
182 blk_queue_congestion_threshold(q);
183 q->nr_batching = BLK_BATCH_REQ;
184
Martin K. Petersene475bba2009-06-16 08:23:52 +0200185 blk_set_default_limits(&q->limits);
186
Jens Axboe86db1e22008-01-29 14:53:40 +0100187 /*
188 * by default assume old behaviour and bounce for any highmem page
189 */
190 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
191}
Jens Axboe86db1e22008-01-29 14:53:40 +0100192EXPORT_SYMBOL(blk_queue_make_request);
193
194/**
195 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900196 * @q: the request queue for the device
197 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100198 *
199 * Description:
200 * Different hardware can have different requirements as to what pages
201 * it can do I/O directly to. A low level driver can call
202 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900203 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100204 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900205void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100206{
Tejun Heocd0aca22009-04-15 22:10:25 +0900207 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100208 int dma = 0;
209
210 q->bounce_gfp = GFP_NOIO;
211#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900212 /*
213 * Assume anything <= 4GB can be handled by IOMMU. Actually
214 * some IOMMUs can handle everything, but I don't know of a
215 * way to test this here.
216 */
217 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100218 dma = 1;
Malahal Naineniefb012b2010-10-01 14:45:27 +0200219 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
Jens Axboe86db1e22008-01-29 14:53:40 +0100220#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100221 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100222 dma = 1;
Malahal Nainenic49825f2010-09-24 20:25:49 +0200223 q->limits.bounce_pfn = b_pfn;
Jens Axboe260a67a2010-10-01 14:42:43 +0200224#endif
Jens Axboe86db1e22008-01-29 14:53:40 +0100225 if (dma) {
226 init_emergency_isa_pool();
227 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Jens Axboe260a67a2010-10-01 14:42:43 +0200228 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100229 }
230}
Jens Axboe86db1e22008-01-29 14:53:40 +0100231EXPORT_SYMBOL(blk_queue_bounce_limit);
232
233/**
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100234 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
235 * @limits: the queue limits
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500236 * @max_hw_sectors: max hardware sectors in the usual 512b unit
Jens Axboe86db1e22008-01-29 14:53:40 +0100237 *
238 * Description:
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500239 * Enables a low level driver to set a hard upper limit,
240 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
241 * the device driver based upon the combined capabilities of I/O
242 * controller and storage device.
243 *
244 * max_sectors is a soft limit imposed by the block layer for
245 * filesystem type requests. This value can be overridden on a
246 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
247 * The soft limit can not exceed max_hw_sectors.
Jens Axboe86db1e22008-01-29 14:53:40 +0100248 **/
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100249void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
Jens Axboe86db1e22008-01-29 14:53:40 +0100250{
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500251 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
252 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700253 printk(KERN_INFO "%s: set to minimum %d\n",
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500254 __func__, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100255 }
256
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100257 limits->max_hw_sectors = max_hw_sectors;
258 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
259 BLK_DEF_MAX_SECTORS);
260}
261EXPORT_SYMBOL(blk_limits_max_hw_sectors);
262
263/**
264 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
265 * @q: the request queue for the device
266 * @max_hw_sectors: max hardware sectors in the usual 512b unit
267 *
268 * Description:
269 * See description for blk_limits_max_hw_sectors().
270 **/
271void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
272{
273 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100274}
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500275EXPORT_SYMBOL(blk_queue_max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100276
277/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200278 * blk_queue_max_discard_sectors - set max sectors for a single discard
279 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200280 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200281 **/
282void blk_queue_max_discard_sectors(struct request_queue *q,
283 unsigned int max_discard_sectors)
284{
285 q->limits.max_discard_sectors = max_discard_sectors;
286}
287EXPORT_SYMBOL(blk_queue_max_discard_sectors);
288
289/**
Martin K. Petersen8a783622010-02-26 00:20:39 -0500290 * blk_queue_max_segments - set max hw segments for a request for this queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100291 * @q: the request queue for the device
292 * @max_segments: max number of segments
293 *
294 * Description:
295 * Enables a low level driver to set an upper limit on the number of
Martin K. Petersen8a783622010-02-26 00:20:39 -0500296 * hw data segments in a request.
Jens Axboe86db1e22008-01-29 14:53:40 +0100297 **/
Martin K. Petersen8a783622010-02-26 00:20:39 -0500298void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
Jens Axboe86db1e22008-01-29 14:53:40 +0100299{
300 if (!max_segments) {
301 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700302 printk(KERN_INFO "%s: set to minimum %d\n",
303 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100304 }
305
Martin K. Petersen8a783622010-02-26 00:20:39 -0500306 q->limits.max_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100307}
Martin K. Petersen8a783622010-02-26 00:20:39 -0500308EXPORT_SYMBOL(blk_queue_max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100309
310/**
311 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
312 * @q: the request queue for the device
313 * @max_size: max size of segment in bytes
314 *
315 * Description:
316 * Enables a low level driver to set an upper limit on the size of a
317 * coalesced segment
318 **/
319void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
320{
321 if (max_size < PAGE_CACHE_SIZE) {
322 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700323 printk(KERN_INFO "%s: set to minimum %d\n",
324 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100325 }
326
Martin K. Petersen025146e2009-05-22 17:17:51 -0400327 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100328}
Jens Axboe86db1e22008-01-29 14:53:40 +0100329EXPORT_SYMBOL(blk_queue_max_segment_size);
330
331/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400332 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100333 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400334 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100335 *
336 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400337 * This should be set to the lowest possible block size that the
338 * storage device can address. The default of 512 covers most
339 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100340 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400341void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100342{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400343 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400344
345 if (q->limits.physical_block_size < size)
346 q->limits.physical_block_size = size;
347
348 if (q->limits.io_min < q->limits.physical_block_size)
349 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100350}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400351EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100352
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400353/**
354 * blk_queue_physical_block_size - set physical block size for the queue
355 * @q: the request queue for the device
356 * @size: the physical block size, in bytes
357 *
358 * Description:
359 * This should be set to the lowest possible sector size that the
360 * hardware can operate on without reverting to read-modify-write
361 * operations.
362 */
Martin K. Petersen892b6f92010-10-13 21:18:03 +0200363void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400364{
365 q->limits.physical_block_size = size;
366
367 if (q->limits.physical_block_size < q->limits.logical_block_size)
368 q->limits.physical_block_size = q->limits.logical_block_size;
369
370 if (q->limits.io_min < q->limits.physical_block_size)
371 q->limits.io_min = q->limits.physical_block_size;
372}
373EXPORT_SYMBOL(blk_queue_physical_block_size);
374
375/**
376 * blk_queue_alignment_offset - set physical block alignment offset
377 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700378 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400379 *
380 * Description:
381 * Some devices are naturally misaligned to compensate for things like
382 * the legacy DOS partition table 63-sector offset. Low-level drivers
383 * should call this function for devices whose first sector is not
384 * naturally aligned.
385 */
386void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
387{
388 q->limits.alignment_offset =
389 offset & (q->limits.physical_block_size - 1);
390 q->limits.misaligned = 0;
391}
392EXPORT_SYMBOL(blk_queue_alignment_offset);
393
394/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400395 * blk_limits_io_min - set minimum request size for a device
396 * @limits: the queue limits
397 * @min: smallest I/O size in bytes
398 *
399 * Description:
400 * Some devices have an internal block size bigger than the reported
401 * hardware sector size. This function can be used to signal the
402 * smallest I/O the device can perform without incurring a performance
403 * penalty.
404 */
405void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
406{
407 limits->io_min = min;
408
409 if (limits->io_min < limits->logical_block_size)
410 limits->io_min = limits->logical_block_size;
411
412 if (limits->io_min < limits->physical_block_size)
413 limits->io_min = limits->physical_block_size;
414}
415EXPORT_SYMBOL(blk_limits_io_min);
416
417/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400418 * blk_queue_io_min - set minimum request size for the queue
419 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700420 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400421 *
422 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400423 * Storage devices may report a granularity or preferred minimum I/O
424 * size which is the smallest request the device can perform without
425 * incurring a performance penalty. For disk drives this is often the
426 * physical block size. For RAID arrays it is often the stripe chunk
427 * size. A properly aligned multiple of minimum_io_size is the
428 * preferred request size for workloads where a high number of I/O
429 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400430 */
431void blk_queue_io_min(struct request_queue *q, unsigned int min)
432{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400433 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400434}
435EXPORT_SYMBOL(blk_queue_io_min);
436
437/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200438 * blk_limits_io_opt - set optimal request size for a device
439 * @limits: the queue limits
440 * @opt: smallest I/O size in bytes
441 *
442 * Description:
443 * Storage devices may report an optimal I/O size, which is the
444 * device's preferred unit for sustained I/O. This is rarely reported
445 * for disk drives. For RAID arrays it is usually the stripe width or
446 * the internal track size. A properly aligned multiple of
447 * optimal_io_size is the preferred request size for workloads where
448 * sustained throughput is desired.
449 */
450void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
451{
452 limits->io_opt = opt;
453}
454EXPORT_SYMBOL(blk_limits_io_opt);
455
456/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400457 * blk_queue_io_opt - set optimal request size for the queue
458 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700459 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400460 *
461 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400462 * Storage devices may report an optimal I/O size, which is the
463 * device's preferred unit for sustained I/O. This is rarely reported
464 * for disk drives. For RAID arrays it is usually the stripe width or
465 * the internal track size. A properly aligned multiple of
466 * optimal_io_size is the preferred request size for workloads where
467 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400468 */
469void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
470{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200471 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400472}
473EXPORT_SYMBOL(blk_queue_io_opt);
474
Jens Axboe86db1e22008-01-29 14:53:40 +0100475/**
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);
Jens Axboe86db1e22008-01-29 14:53:40 +0100483}
Jens Axboe86db1e22008-01-29 14:53:40 +0100484EXPORT_SYMBOL(blk_queue_stack_limits);
485
486/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400487 * blk_stack_limits - adjust queue_limits for stacked devices
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100488 * @t: the stacking driver limits (top device)
489 * @b: the underlying queue limits (bottom, component device)
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500490 * @start: first data sector within component device
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400491 *
492 * Description:
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100493 * This function is used by stacking drivers like MD and DM to ensure
494 * that all component devices have compatible block sizes and
495 * alignments. The stacking driver must provide a queue_limits
496 * struct (top) and then iteratively call the stacking function for
497 * all component (bottom) devices. The stacking function will
498 * attempt to combine the values and ensure proper alignment.
499 *
500 * Returns 0 if the top and bottom queue_limits are compatible. The
501 * top device's block sizes and alignment offsets may be adjusted to
502 * ensure alignment with the bottom device. If no compatible sizes
503 * and alignments exist, -1 is returned and the resulting top
504 * queue_limits will have the misaligned flag set to indicate that
505 * the alignment_offset is undefined.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400506 */
507int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500508 sector_t start)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400509{
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500510 unsigned int top, bottom, alignment, ret = 0;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100511
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400512 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
513 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200514 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400515
516 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
517 b->seg_boundary_mask);
518
Martin K. Petersen8a783622010-02-26 00:20:39 -0500519 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200520 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
521 b->max_integrity_segments);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400522
523 t->max_segment_size = min_not_zero(t->max_segment_size,
524 b->max_segment_size);
525
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500526 t->misaligned |= b->misaligned;
527
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500528 alignment = queue_limit_alignment_offset(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100529
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100530 /* Bottom device has different alignment. Check that it is
531 * compatible with the current top alignment.
532 */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100533 if (t->alignment_offset != alignment) {
534
535 top = max(t->physical_block_size, t->io_min)
536 + t->alignment_offset;
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100537 bottom = max(b->physical_block_size, b->io_min) + alignment;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100538
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100539 /* Verify that top and bottom intervals line up */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500540 if (max(top, bottom) & (min(top, bottom) - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100541 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500542 ret = -1;
543 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100544 }
545
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400546 t->logical_block_size = max(t->logical_block_size,
547 b->logical_block_size);
548
549 t->physical_block_size = max(t->physical_block_size,
550 b->physical_block_size);
551
552 t->io_min = max(t->io_min, b->io_min);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100553 t->io_opt = lcm(t->io_opt, b->io_opt);
554
Martin K. Petersene692cb62010-12-01 19:41:49 +0100555 t->cluster &= b->cluster;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100556 t->discard_zeroes_data &= b->discard_zeroes_data;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400557
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100558 /* Physical block size a multiple of the logical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100559 if (t->physical_block_size & (t->logical_block_size - 1)) {
560 t->physical_block_size = t->logical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400561 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500562 ret = -1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100563 }
564
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100565 /* Minimum I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100566 if (t->io_min & (t->physical_block_size - 1)) {
567 t->io_min = t->physical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400568 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500569 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400570 }
571
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100572 /* Optimal I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100573 if (t->io_opt & (t->physical_block_size - 1)) {
574 t->io_opt = 0;
575 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500576 ret = -1;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100577 }
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400578
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100579 /* Find lowest common alignment_offset */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100580 t->alignment_offset = lcm(t->alignment_offset, alignment)
581 & (max(t->physical_block_size, t->io_min) - 1);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400582
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100583 /* Verify that new alignment_offset is on a logical block boundary */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500584 if (t->alignment_offset & (t->logical_block_size - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100585 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500586 ret = -1;
587 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100588
589 /* Discard alignment and granularity */
590 if (b->discard_granularity) {
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500591 alignment = queue_limit_discard_alignment(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100592
593 if (t->discard_granularity != 0 &&
594 t->discard_alignment != alignment) {
595 top = t->discard_granularity + t->discard_alignment;
596 bottom = b->discard_granularity + alignment;
597
598 /* Verify that top and bottom intervals line up */
599 if (max(top, bottom) & (min(top, bottom) - 1))
600 t->discard_misaligned = 1;
601 }
602
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100603 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
604 b->max_discard_sectors);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100605 t->discard_granularity = max(t->discard_granularity,
606 b->discard_granularity);
607 t->discard_alignment = lcm(t->discard_alignment, alignment) &
608 (t->discard_granularity - 1);
609 }
610
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500611 return ret;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400612}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200613EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400614
615/**
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500616 * bdev_stack_limits - adjust queue limits for stacked drivers
617 * @t: the stacking driver limits (top device)
618 * @bdev: the component block_device (bottom)
619 * @start: first data sector within component device
620 *
621 * Description:
622 * Merges queue limits for a top device and a block_device. Returns
623 * 0 if alignment didn't change. Returns -1 if adding the bottom
624 * device caused misalignment.
625 */
626int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
627 sector_t start)
628{
629 struct request_queue *bq = bdev_get_queue(bdev);
630
631 start += get_start_sect(bdev);
632
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500633 return blk_stack_limits(t, &bq->limits, start);
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500634}
635EXPORT_SYMBOL(bdev_stack_limits);
636
637/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400638 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200639 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400640 * @bdev: the underlying block device (bottom)
641 * @offset: offset to beginning of data within component device
642 *
643 * Description:
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500644 * Merges the limits for a top level gendisk and a bottom level
645 * block_device.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400646 */
647void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
648 sector_t offset)
649{
650 struct request_queue *t = disk->queue;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400651
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500652 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400653 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
654
655 disk_name(disk, 0, top);
656 bdevname(bdev, bottom);
657
658 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
659 top, bottom);
660 }
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400661}
662EXPORT_SYMBOL(disk_stack_limits);
663
664/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100665 * blk_queue_dma_pad - set pad mask
666 * @q: the request queue for the device
667 * @mask: pad mask
668 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200669 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100670 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200671 * Appending pad buffer to a request modifies the last entry of a
672 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100673 **/
674void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
675{
676 q->dma_pad_mask = mask;
677}
678EXPORT_SYMBOL(blk_queue_dma_pad);
679
680/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200681 * blk_queue_update_dma_pad - update pad mask
682 * @q: the request queue for the device
683 * @mask: pad mask
684 *
685 * Update dma pad mask.
686 *
687 * Appending pad buffer to a request modifies the last entry of a
688 * scatter list such that it includes the pad buffer.
689 **/
690void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
691{
692 if (mask > q->dma_pad_mask)
693 q->dma_pad_mask = mask;
694}
695EXPORT_SYMBOL(blk_queue_update_dma_pad);
696
697/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100698 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100699 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100700 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100701 * @buf: physically contiguous buffer
702 * @size: size of the buffer in bytes
703 *
704 * Some devices have excess DMA problems and can't simply discard (or
705 * zero fill) the unwanted piece of the transfer. They have to have a
706 * real area of memory to transfer it into. The use case for this is
707 * ATAPI devices in DMA mode. If the packet command causes a transfer
708 * bigger than the transfer size some HBAs will lock up if there
709 * aren't DMA elements to contain the excess transfer. What this API
710 * does is adjust the queue so that the buf is always appended
711 * silently to the scatterlist.
712 *
Martin K. Petersen8a783622010-02-26 00:20:39 -0500713 * Note: This routine adjusts max_hw_segments to make room for appending
714 * the drain buffer. If you call blk_queue_max_segments() after calling
715 * this routine, you must set the limit to one fewer than your device
716 * can support otherwise there won't be room for the drain buffer.
Jens Axboe86db1e22008-01-29 14:53:40 +0100717 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100718int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100719 dma_drain_needed_fn *dma_drain_needed,
720 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100721{
Martin K. Petersen8a783622010-02-26 00:20:39 -0500722 if (queue_max_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100723 return -EINVAL;
724 /* make room for appending the drain */
Martin K. Petersen8a783622010-02-26 00:20:39 -0500725 blk_queue_max_segments(q, queue_max_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100726 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100727 q->dma_drain_buffer = buf;
728 q->dma_drain_size = size;
729
730 return 0;
731}
Jens Axboe86db1e22008-01-29 14:53:40 +0100732EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
733
734/**
735 * blk_queue_segment_boundary - set boundary rules for segment merging
736 * @q: the request queue for the device
737 * @mask: the memory boundary mask
738 **/
739void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
740{
741 if (mask < PAGE_CACHE_SIZE - 1) {
742 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700743 printk(KERN_INFO "%s: set to minimum %lx\n",
744 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100745 }
746
Martin K. Petersen025146e2009-05-22 17:17:51 -0400747 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100748}
Jens Axboe86db1e22008-01-29 14:53:40 +0100749EXPORT_SYMBOL(blk_queue_segment_boundary);
750
751/**
752 * blk_queue_dma_alignment - set dma length and memory alignment
753 * @q: the request queue for the device
754 * @mask: alignment mask
755 *
756 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200757 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100758 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100759 *
760 **/
761void blk_queue_dma_alignment(struct request_queue *q, int mask)
762{
763 q->dma_alignment = mask;
764}
Jens Axboe86db1e22008-01-29 14:53:40 +0100765EXPORT_SYMBOL(blk_queue_dma_alignment);
766
767/**
768 * blk_queue_update_dma_alignment - update dma length and memory alignment
769 * @q: the request queue for the device
770 * @mask: alignment mask
771 *
772 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200773 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100774 * If the requested alignment is larger than the current alignment, then
775 * the current queue alignment is updated to the new value, otherwise it
776 * is left alone. The design of this is to allow multiple objects
777 * (driver, device, transport etc) to set their respective
778 * alignments without having them interfere.
779 *
780 **/
781void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
782{
783 BUG_ON(mask > PAGE_SIZE);
784
785 if (mask > q->dma_alignment)
786 q->dma_alignment = mask;
787}
Jens Axboe86db1e22008-01-29 14:53:40 +0100788EXPORT_SYMBOL(blk_queue_update_dma_alignment);
789
Tejun Heo4913efe2010-09-03 11:56:16 +0200790/**
791 * blk_queue_flush - configure queue's cache flush capability
792 * @q: the request queue for the device
793 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
794 *
795 * Tell block layer cache flush capability of @q. If it supports
796 * flushing, REQ_FLUSH should be set. If it supports bypassing
797 * write cache for individual writes, REQ_FUA should be set.
798 */
799void blk_queue_flush(struct request_queue *q, unsigned int flush)
800{
801 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
802
803 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
804 flush &= ~REQ_FUA;
805
806 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
807}
808EXPORT_SYMBOL_GPL(blk_queue_flush);
809
shaohua.li@intel.comf3876932011-05-06 11:34:32 -0600810void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
811{
812 q->flush_not_queueable = !queueable;
813}
814EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
815
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200816static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100817{
818 blk_max_low_pfn = max_low_pfn - 1;
819 blk_max_pfn = max_pfn - 1;
820 return 0;
821}
822subsys_initcall(blk_settings_init);