blob: 53309333c2f015fff6b6a7ae8a9a01700105b3e8 [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. Petersen4363ac72012-09-18 12:19:27 -0400116 lim->max_write_same_sectors = 0;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100117 lim->max_discard_sectors = 0;
118 lim->discard_granularity = 0;
119 lim->discard_alignment = 0;
120 lim->discard_misaligned = 0;
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100121 lim->discard_zeroes_data = 0;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200122 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200123 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200124 lim->alignment_offset = 0;
125 lim->io_opt = 0;
126 lim->misaligned = 0;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100127 lim->cluster = 1;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200128}
129EXPORT_SYMBOL(blk_set_default_limits);
130
131/**
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100132 * blk_set_stacking_limits - set default limits for stacking devices
133 * @lim: the queue_limits structure to reset
134 *
135 * Description:
136 * Returns a queue_limit struct to its default state. Should be used
137 * by stacking drivers like DM that have no internal limits.
138 */
139void blk_set_stacking_limits(struct queue_limits *lim)
140{
141 blk_set_default_limits(lim);
142
143 /* Inherit limits from component devices */
144 lim->discard_zeroes_data = 1;
145 lim->max_segments = USHRT_MAX;
146 lim->max_hw_sectors = UINT_MAX;
Mike Snitzerd82ae522013-10-18 09:44:49 -0600147 lim->max_segment_size = UINT_MAX;
Mike Snitzerfe86cdc2012-08-01 10:44:28 +0200148 lim->max_sectors = UINT_MAX;
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400149 lim->max_write_same_sectors = UINT_MAX;
Martin K. Petersenb1bd0552012-01-11 16:27:11 +0100150}
151EXPORT_SYMBOL(blk_set_stacking_limits);
152
153/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100154 * blk_queue_make_request - define an alternate make_request function for a device
155 * @q: the request queue for the device to be affected
156 * @mfn: the alternate make_request function
157 *
158 * Description:
159 * The normal way for &struct bios to be passed to a device
160 * driver is for them to be collected into requests on a request
161 * queue, and then to allow the device driver to select requests
162 * off that queue when it is ready. This works well for many block
163 * devices. However some block devices (typically virtual devices
164 * such as md or lvm) do not benefit from the processing on the
165 * request queue, and are served best by having the requests passed
166 * directly to them. This can be achieved by providing a function
167 * to blk_queue_make_request().
168 *
169 * Caveat:
170 * The driver that does this *must* be able to deal appropriately
171 * with buffers in "highmemory". This can be accomplished by either calling
172 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
173 * blk_queue_bounce() to create a buffer in normal memory.
174 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100175void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100176{
177 /*
178 * set defaults
179 */
180 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100181
Jens Axboe86db1e22008-01-29 14:53:40 +0100182 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100183 blk_queue_dma_alignment(q, 511);
184 blk_queue_congestion_threshold(q);
185 q->nr_batching = BLK_BATCH_REQ;
186
Martin K. Petersene475bba2009-06-16 08:23:52 +0200187 blk_set_default_limits(&q->limits);
188
Jens Axboe86db1e22008-01-29 14:53:40 +0100189 /*
190 * by default assume old behaviour and bounce for any highmem page
191 */
192 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
193}
Jens Axboe86db1e22008-01-29 14:53:40 +0100194EXPORT_SYMBOL(blk_queue_make_request);
195
196/**
197 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900198 * @q: the request queue for the device
199 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100200 *
201 * Description:
202 * Different hardware can have different requirements as to what pages
203 * it can do I/O directly to. A low level driver can call
204 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900205 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100206 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900207void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100208{
Tejun Heocd0aca22009-04-15 22:10:25 +0900209 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100210 int dma = 0;
211
212 q->bounce_gfp = GFP_NOIO;
213#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900214 /*
215 * Assume anything <= 4GB can be handled by IOMMU. Actually
216 * some IOMMUs can handle everything, but I don't know of a
217 * way to test this here.
218 */
219 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100220 dma = 1;
Malahal Naineniefb012b2010-10-01 14:45:27 +0200221 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
Jens Axboe86db1e22008-01-29 14:53:40 +0100222#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100223 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100224 dma = 1;
Malahal Nainenic49825f2010-09-24 20:25:49 +0200225 q->limits.bounce_pfn = b_pfn;
Jens Axboe260a67a2010-10-01 14:42:43 +0200226#endif
Jens Axboe86db1e22008-01-29 14:53:40 +0100227 if (dma) {
228 init_emergency_isa_pool();
229 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Jens Axboe260a67a2010-10-01 14:42:43 +0200230 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100231 }
232}
Jens Axboe86db1e22008-01-29 14:53:40 +0100233EXPORT_SYMBOL(blk_queue_bounce_limit);
234
235/**
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100236 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
237 * @limits: the queue limits
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500238 * @max_hw_sectors: max hardware sectors in the usual 512b unit
Jens Axboe86db1e22008-01-29 14:53:40 +0100239 *
240 * Description:
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500241 * Enables a low level driver to set a hard upper limit,
242 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
243 * the device driver based upon the combined capabilities of I/O
244 * controller and storage device.
245 *
246 * max_sectors is a soft limit imposed by the block layer for
247 * filesystem type requests. This value can be overridden on a
248 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
249 * The soft limit can not exceed max_hw_sectors.
Jens Axboe86db1e22008-01-29 14:53:40 +0100250 **/
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100251void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
Jens Axboe86db1e22008-01-29 14:53:40 +0100252{
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500253 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
254 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700255 printk(KERN_INFO "%s: set to minimum %d\n",
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500256 __func__, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100257 }
258
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100259 limits->max_hw_sectors = max_hw_sectors;
260 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
261 BLK_DEF_MAX_SECTORS);
262}
263EXPORT_SYMBOL(blk_limits_max_hw_sectors);
264
265/**
266 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
267 * @q: the request queue for the device
268 * @max_hw_sectors: max hardware sectors in the usual 512b unit
269 *
270 * Description:
271 * See description for blk_limits_max_hw_sectors().
272 **/
273void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
274{
275 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100276}
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500277EXPORT_SYMBOL(blk_queue_max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100278
279/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200280 * blk_queue_max_discard_sectors - set max sectors for a single discard
281 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200282 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200283 **/
284void blk_queue_max_discard_sectors(struct request_queue *q,
285 unsigned int max_discard_sectors)
286{
287 q->limits.max_discard_sectors = max_discard_sectors;
288}
289EXPORT_SYMBOL(blk_queue_max_discard_sectors);
290
291/**
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400292 * blk_queue_max_write_same_sectors - set max sectors for a single write same
293 * @q: the request queue for the device
294 * @max_write_same_sectors: maximum number of sectors to write per command
295 **/
296void blk_queue_max_write_same_sectors(struct request_queue *q,
297 unsigned int max_write_same_sectors)
298{
299 q->limits.max_write_same_sectors = max_write_same_sectors;
300}
301EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
302
303/**
Martin K. Petersen8a783622010-02-26 00:20:39 -0500304 * blk_queue_max_segments - set max hw segments for a request for this queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100305 * @q: the request queue for the device
306 * @max_segments: max number of segments
307 *
308 * Description:
309 * Enables a low level driver to set an upper limit on the number of
Martin K. Petersen8a783622010-02-26 00:20:39 -0500310 * hw data segments in a request.
Jens Axboe86db1e22008-01-29 14:53:40 +0100311 **/
Martin K. Petersen8a783622010-02-26 00:20:39 -0500312void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
Jens Axboe86db1e22008-01-29 14:53:40 +0100313{
314 if (!max_segments) {
315 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700316 printk(KERN_INFO "%s: set to minimum %d\n",
317 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100318 }
319
Martin K. Petersen8a783622010-02-26 00:20:39 -0500320 q->limits.max_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100321}
Martin K. Petersen8a783622010-02-26 00:20:39 -0500322EXPORT_SYMBOL(blk_queue_max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100323
324/**
325 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
326 * @q: the request queue for the device
327 * @max_size: max size of segment in bytes
328 *
329 * Description:
330 * Enables a low level driver to set an upper limit on the size of a
331 * coalesced segment
332 **/
333void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
334{
335 if (max_size < PAGE_CACHE_SIZE) {
336 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700337 printk(KERN_INFO "%s: set to minimum %d\n",
338 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100339 }
340
Martin K. Petersen025146e2009-05-22 17:17:51 -0400341 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100342}
Jens Axboe86db1e22008-01-29 14:53:40 +0100343EXPORT_SYMBOL(blk_queue_max_segment_size);
344
345/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400346 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100347 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400348 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100349 *
350 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400351 * This should be set to the lowest possible block size that the
352 * storage device can address. The default of 512 covers most
353 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100354 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400355void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100356{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400357 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400358
359 if (q->limits.physical_block_size < size)
360 q->limits.physical_block_size = size;
361
362 if (q->limits.io_min < q->limits.physical_block_size)
363 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100364}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400365EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100366
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400367/**
368 * blk_queue_physical_block_size - set physical block size for the queue
369 * @q: the request queue for the device
370 * @size: the physical block size, in bytes
371 *
372 * Description:
373 * This should be set to the lowest possible sector size that the
374 * hardware can operate on without reverting to read-modify-write
375 * operations.
376 */
Martin K. Petersen892b6f92010-10-13 21:18:03 +0200377void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400378{
379 q->limits.physical_block_size = size;
380
381 if (q->limits.physical_block_size < q->limits.logical_block_size)
382 q->limits.physical_block_size = q->limits.logical_block_size;
383
384 if (q->limits.io_min < q->limits.physical_block_size)
385 q->limits.io_min = q->limits.physical_block_size;
386}
387EXPORT_SYMBOL(blk_queue_physical_block_size);
388
389/**
390 * blk_queue_alignment_offset - set physical block alignment offset
391 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700392 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400393 *
394 * Description:
395 * Some devices are naturally misaligned to compensate for things like
396 * the legacy DOS partition table 63-sector offset. Low-level drivers
397 * should call this function for devices whose first sector is not
398 * naturally aligned.
399 */
400void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
401{
402 q->limits.alignment_offset =
403 offset & (q->limits.physical_block_size - 1);
404 q->limits.misaligned = 0;
405}
406EXPORT_SYMBOL(blk_queue_alignment_offset);
407
408/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400409 * blk_limits_io_min - set minimum request size for a device
410 * @limits: the queue limits
411 * @min: smallest I/O size in bytes
412 *
413 * Description:
414 * Some devices have an internal block size bigger than the reported
415 * hardware sector size. This function can be used to signal the
416 * smallest I/O the device can perform without incurring a performance
417 * penalty.
418 */
419void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
420{
421 limits->io_min = min;
422
423 if (limits->io_min < limits->logical_block_size)
424 limits->io_min = limits->logical_block_size;
425
426 if (limits->io_min < limits->physical_block_size)
427 limits->io_min = limits->physical_block_size;
428}
429EXPORT_SYMBOL(blk_limits_io_min);
430
431/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400432 * blk_queue_io_min - set minimum request size for the queue
433 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700434 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400435 *
436 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400437 * Storage devices may report a granularity or preferred minimum I/O
438 * size which is the smallest request the device can perform without
439 * incurring a performance penalty. For disk drives this is often the
440 * physical block size. For RAID arrays it is often the stripe chunk
441 * size. A properly aligned multiple of minimum_io_size is the
442 * preferred request size for workloads where a high number of I/O
443 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400444 */
445void blk_queue_io_min(struct request_queue *q, unsigned int min)
446{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400447 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400448}
449EXPORT_SYMBOL(blk_queue_io_min);
450
451/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200452 * blk_limits_io_opt - set optimal request size for a device
453 * @limits: the queue limits
454 * @opt: smallest I/O size in bytes
455 *
456 * Description:
457 * 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.
463 */
464void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
465{
466 limits->io_opt = opt;
467}
468EXPORT_SYMBOL(blk_limits_io_opt);
469
470/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400471 * blk_queue_io_opt - set optimal request size for the queue
472 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700473 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400474 *
475 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400476 * Storage devices may report an optimal I/O size, which is the
477 * device's preferred unit for sustained I/O. This is rarely reported
478 * for disk drives. For RAID arrays it is usually the stripe width or
479 * the internal track size. A properly aligned multiple of
480 * optimal_io_size is the preferred request size for workloads where
481 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400482 */
483void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
484{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200485 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400486}
487EXPORT_SYMBOL(blk_queue_io_opt);
488
Jens Axboe86db1e22008-01-29 14:53:40 +0100489/**
490 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
491 * @t: the stacking driver (top)
492 * @b: the underlying device (bottom)
493 **/
494void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
495{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400496 blk_stack_limits(&t->limits, &b->limits, 0);
Jens Axboe86db1e22008-01-29 14:53:40 +0100497}
Jens Axboe86db1e22008-01-29 14:53:40 +0100498EXPORT_SYMBOL(blk_queue_stack_limits);
499
500/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400501 * blk_stack_limits - adjust queue_limits for stacked devices
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100502 * @t: the stacking driver limits (top device)
503 * @b: the underlying queue limits (bottom, component device)
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500504 * @start: first data sector within component device
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400505 *
506 * Description:
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100507 * This function is used by stacking drivers like MD and DM to ensure
508 * that all component devices have compatible block sizes and
509 * alignments. The stacking driver must provide a queue_limits
510 * struct (top) and then iteratively call the stacking function for
511 * all component (bottom) devices. The stacking function will
512 * attempt to combine the values and ensure proper alignment.
513 *
514 * Returns 0 if the top and bottom queue_limits are compatible. The
515 * top device's block sizes and alignment offsets may be adjusted to
516 * ensure alignment with the bottom device. If no compatible sizes
517 * and alignments exist, -1 is returned and the resulting top
518 * queue_limits will have the misaligned flag set to indicate that
519 * the alignment_offset is undefined.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400520 */
521int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500522 sector_t start)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400523{
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500524 unsigned int top, bottom, alignment, ret = 0;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100525
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400526 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
527 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400528 t->max_write_same_sectors = min(t->max_write_same_sectors,
529 b->max_write_same_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200530 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400531
532 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
533 b->seg_boundary_mask);
534
Martin K. Petersen8a783622010-02-26 00:20:39 -0500535 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200536 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
537 b->max_integrity_segments);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400538
539 t->max_segment_size = min_not_zero(t->max_segment_size,
540 b->max_segment_size);
541
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500542 t->misaligned |= b->misaligned;
543
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500544 alignment = queue_limit_alignment_offset(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100545
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100546 /* Bottom device has different alignment. Check that it is
547 * compatible with the current top alignment.
548 */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100549 if (t->alignment_offset != alignment) {
550
551 top = max(t->physical_block_size, t->io_min)
552 + t->alignment_offset;
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100553 bottom = max(b->physical_block_size, b->io_min) + alignment;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100554
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100555 /* Verify that top and bottom intervals line up */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500556 if (max(top, bottom) & (min(top, bottom) - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100557 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500558 ret = -1;
559 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100560 }
561
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400562 t->logical_block_size = max(t->logical_block_size,
563 b->logical_block_size);
564
565 t->physical_block_size = max(t->physical_block_size,
566 b->physical_block_size);
567
568 t->io_min = max(t->io_min, b->io_min);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100569 t->io_opt = lcm(t->io_opt, b->io_opt);
570
Martin K. Petersene692cb62010-12-01 19:41:49 +0100571 t->cluster &= b->cluster;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100572 t->discard_zeroes_data &= b->discard_zeroes_data;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400573
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100574 /* Physical block size a multiple of the logical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100575 if (t->physical_block_size & (t->logical_block_size - 1)) {
576 t->physical_block_size = t->logical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400577 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500578 ret = -1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100579 }
580
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100581 /* Minimum I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100582 if (t->io_min & (t->physical_block_size - 1)) {
583 t->io_min = t->physical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400584 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500585 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400586 }
587
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100588 /* Optimal I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100589 if (t->io_opt & (t->physical_block_size - 1)) {
590 t->io_opt = 0;
591 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500592 ret = -1;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100593 }
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400594
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100595 /* Find lowest common alignment_offset */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100596 t->alignment_offset = lcm(t->alignment_offset, alignment)
597 & (max(t->physical_block_size, t->io_min) - 1);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400598
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100599 /* Verify that new alignment_offset is on a logical block boundary */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500600 if (t->alignment_offset & (t->logical_block_size - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100601 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500602 ret = -1;
603 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100604
605 /* Discard alignment and granularity */
606 if (b->discard_granularity) {
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500607 alignment = queue_limit_discard_alignment(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100608
609 if (t->discard_granularity != 0 &&
610 t->discard_alignment != alignment) {
611 top = t->discard_granularity + t->discard_alignment;
612 bottom = b->discard_granularity + alignment;
613
614 /* Verify that top and bottom intervals line up */
Shaohua Li8dd2cb72012-12-14 11:15:36 +0800615 if ((max(top, bottom) % min(top, bottom)) != 0)
Martin K. Petersen9504e082009-12-21 15:55:51 +0100616 t->discard_misaligned = 1;
617 }
618
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100619 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
620 b->max_discard_sectors);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100621 t->discard_granularity = max(t->discard_granularity,
622 b->discard_granularity);
Shaohua Li8dd2cb72012-12-14 11:15:36 +0800623 t->discard_alignment = lcm(t->discard_alignment, alignment) %
624 t->discard_granularity;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100625 }
626
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500627 return ret;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400628}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200629EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400630
631/**
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500632 * bdev_stack_limits - adjust queue limits for stacked drivers
633 * @t: the stacking driver limits (top device)
634 * @bdev: the component block_device (bottom)
635 * @start: first data sector within component device
636 *
637 * Description:
638 * Merges queue limits for a top device and a block_device. Returns
639 * 0 if alignment didn't change. Returns -1 if adding the bottom
640 * device caused misalignment.
641 */
642int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
643 sector_t start)
644{
645 struct request_queue *bq = bdev_get_queue(bdev);
646
647 start += get_start_sect(bdev);
648
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500649 return blk_stack_limits(t, &bq->limits, start);
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500650}
651EXPORT_SYMBOL(bdev_stack_limits);
652
653/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400654 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200655 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400656 * @bdev: the underlying block device (bottom)
657 * @offset: offset to beginning of data within component device
658 *
659 * Description:
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500660 * Merges the limits for a top level gendisk and a bottom level
661 * block_device.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400662 */
663void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
664 sector_t offset)
665{
666 struct request_queue *t = disk->queue;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400667
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500668 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400669 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
670
671 disk_name(disk, 0, top);
672 bdevname(bdev, bottom);
673
674 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
675 top, bottom);
676 }
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400677}
678EXPORT_SYMBOL(disk_stack_limits);
679
680/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100681 * blk_queue_dma_pad - set pad mask
682 * @q: the request queue for the device
683 * @mask: pad mask
684 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200685 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100686 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200687 * Appending pad buffer to a request modifies the last entry of a
688 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100689 **/
690void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
691{
692 q->dma_pad_mask = mask;
693}
694EXPORT_SYMBOL(blk_queue_dma_pad);
695
696/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200697 * blk_queue_update_dma_pad - update pad mask
698 * @q: the request queue for the device
699 * @mask: pad mask
700 *
701 * Update dma pad mask.
702 *
703 * Appending pad buffer to a request modifies the last entry of a
704 * scatter list such that it includes the pad buffer.
705 **/
706void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
707{
708 if (mask > q->dma_pad_mask)
709 q->dma_pad_mask = mask;
710}
711EXPORT_SYMBOL(blk_queue_update_dma_pad);
712
713/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100714 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100715 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100716 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100717 * @buf: physically contiguous buffer
718 * @size: size of the buffer in bytes
719 *
720 * Some devices have excess DMA problems and can't simply discard (or
721 * zero fill) the unwanted piece of the transfer. They have to have a
722 * real area of memory to transfer it into. The use case for this is
723 * ATAPI devices in DMA mode. If the packet command causes a transfer
724 * bigger than the transfer size some HBAs will lock up if there
725 * aren't DMA elements to contain the excess transfer. What this API
726 * does is adjust the queue so that the buf is always appended
727 * silently to the scatterlist.
728 *
Martin K. Petersen8a783622010-02-26 00:20:39 -0500729 * Note: This routine adjusts max_hw_segments to make room for appending
730 * the drain buffer. If you call blk_queue_max_segments() after calling
731 * this routine, you must set the limit to one fewer than your device
732 * can support otherwise there won't be room for the drain buffer.
Jens Axboe86db1e22008-01-29 14:53:40 +0100733 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100734int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100735 dma_drain_needed_fn *dma_drain_needed,
736 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100737{
Martin K. Petersen8a783622010-02-26 00:20:39 -0500738 if (queue_max_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100739 return -EINVAL;
740 /* make room for appending the drain */
Martin K. Petersen8a783622010-02-26 00:20:39 -0500741 blk_queue_max_segments(q, queue_max_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100742 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100743 q->dma_drain_buffer = buf;
744 q->dma_drain_size = size;
745
746 return 0;
747}
Jens Axboe86db1e22008-01-29 14:53:40 +0100748EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
749
750/**
751 * blk_queue_segment_boundary - set boundary rules for segment merging
752 * @q: the request queue for the device
753 * @mask: the memory boundary mask
754 **/
755void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
756{
757 if (mask < PAGE_CACHE_SIZE - 1) {
758 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700759 printk(KERN_INFO "%s: set to minimum %lx\n",
760 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100761 }
762
Martin K. Petersen025146e2009-05-22 17:17:51 -0400763 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100764}
Jens Axboe86db1e22008-01-29 14:53:40 +0100765EXPORT_SYMBOL(blk_queue_segment_boundary);
766
767/**
768 * blk_queue_dma_alignment - set 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 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100774 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100775 *
776 **/
777void blk_queue_dma_alignment(struct request_queue *q, int mask)
778{
779 q->dma_alignment = mask;
780}
Jens Axboe86db1e22008-01-29 14:53:40 +0100781EXPORT_SYMBOL(blk_queue_dma_alignment);
782
783/**
784 * blk_queue_update_dma_alignment - update dma length and memory alignment
785 * @q: the request queue for the device
786 * @mask: alignment mask
787 *
788 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200789 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100790 * If the requested alignment is larger than the current alignment, then
791 * the current queue alignment is updated to the new value, otherwise it
792 * is left alone. The design of this is to allow multiple objects
793 * (driver, device, transport etc) to set their respective
794 * alignments without having them interfere.
795 *
796 **/
797void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
798{
799 BUG_ON(mask > PAGE_SIZE);
800
801 if (mask > q->dma_alignment)
802 q->dma_alignment = mask;
803}
Jens Axboe86db1e22008-01-29 14:53:40 +0100804EXPORT_SYMBOL(blk_queue_update_dma_alignment);
805
Tejun Heo4913efe2010-09-03 11:56:16 +0200806/**
807 * blk_queue_flush - configure queue's cache flush capability
808 * @q: the request queue for the device
809 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
810 *
811 * Tell block layer cache flush capability of @q. If it supports
812 * flushing, REQ_FLUSH should be set. If it supports bypassing
813 * write cache for individual writes, REQ_FUA should be set.
814 */
815void blk_queue_flush(struct request_queue *q, unsigned int flush)
816{
817 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
818
819 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
820 flush &= ~REQ_FUA;
821
822 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
823}
824EXPORT_SYMBOL_GPL(blk_queue_flush);
825
shaohua.li@intel.comf3876932011-05-06 11:34:32 -0600826void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
827{
828 q->flush_not_queueable = !queueable;
829}
830EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
831
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200832static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100833{
834 blk_max_low_pfn = max_low_pfn - 1;
835 blk_max_pfn = max_pfn - 1;
836 return 0;
837}
838subsys_initcall(blk_settings_init);