blob: 1fa7692935976f4e60393e8d7289acf1c56c74c3 [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:
107 * Returns a queue_limit struct to its default state. Can be used by
108 * stacking drivers like DM that stage table swaps and reuse an
109 * existing device queue.
110 */
111void blk_set_default_limits(struct queue_limits *lim)
112{
Martin K. Petersen8a783622010-02-26 00:20:39 -0500113 lim->max_segments = BLK_MAX_SEGMENTS;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200114 lim->max_integrity_segments = 0;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200115 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
Martin K. Peterseneb28d312010-02-26 00:20:37 -0500116 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
Martin K. Petersen5dee2472009-09-21 21:46:05 +0200117 lim->max_sectors = BLK_DEF_MAX_SECTORS;
118 lim->max_hw_sectors = INT_MAX;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100119 lim->max_discard_sectors = 0;
120 lim->discard_granularity = 0;
121 lim->discard_alignment = 0;
122 lim->discard_misaligned = 0;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100123 lim->discard_zeroes_data = -1;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200124 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200125 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200126 lim->alignment_offset = 0;
127 lim->io_opt = 0;
128 lim->misaligned = 0;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100129 lim->cluster = 1;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200130}
131EXPORT_SYMBOL(blk_set_default_limits);
132
133/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100134 * blk_queue_make_request - define an alternate make_request function for a device
135 * @q: the request queue for the device to be affected
136 * @mfn: the alternate make_request function
137 *
138 * Description:
139 * The normal way for &struct bios to be passed to a device
140 * driver is for them to be collected into requests on a request
141 * queue, and then to allow the device driver to select requests
142 * off that queue when it is ready. This works well for many block
143 * devices. However some block devices (typically virtual devices
144 * such as md or lvm) do not benefit from the processing on the
145 * request queue, and are served best by having the requests passed
146 * directly to them. This can be achieved by providing a function
147 * to blk_queue_make_request().
148 *
149 * Caveat:
150 * The driver that does this *must* be able to deal appropriately
151 * with buffers in "highmemory". This can be accomplished by either calling
152 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
153 * blk_queue_bounce() to create a buffer in normal memory.
154 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100155void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100156{
157 /*
158 * set defaults
159 */
160 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100161
Jens Axboe86db1e22008-01-29 14:53:40 +0100162 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100163 blk_queue_dma_alignment(q, 511);
164 blk_queue_congestion_threshold(q);
165 q->nr_batching = BLK_BATCH_REQ;
166
Martin K. Petersene475bba2009-06-16 08:23:52 +0200167 blk_set_default_limits(&q->limits);
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500168 blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200169
Jens Axboe86db1e22008-01-29 14:53:40 +0100170 /*
171 * by default assume old behaviour and bounce for any highmem page
172 */
173 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
174}
Jens Axboe86db1e22008-01-29 14:53:40 +0100175EXPORT_SYMBOL(blk_queue_make_request);
176
177/**
178 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900179 * @q: the request queue for the device
180 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100181 *
182 * Description:
183 * Different hardware can have different requirements as to what pages
184 * it can do I/O directly to. A low level driver can call
185 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900186 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100187 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900188void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100189{
Tejun Heocd0aca22009-04-15 22:10:25 +0900190 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100191 int dma = 0;
192
193 q->bounce_gfp = GFP_NOIO;
194#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900195 /*
196 * Assume anything <= 4GB can be handled by IOMMU. Actually
197 * some IOMMUs can handle everything, but I don't know of a
198 * way to test this here.
199 */
200 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100201 dma = 1;
Malahal Naineniefb012b32010-10-01 14:45:27 +0200202 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
Jens Axboe86db1e22008-01-29 14:53:40 +0100203#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100204 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100205 dma = 1;
Malahal Nainenic49825f2010-09-24 20:25:49 +0200206 q->limits.bounce_pfn = b_pfn;
Jens Axboe260a67a2010-10-01 14:42:43 +0200207#endif
Jens Axboe86db1e22008-01-29 14:53:40 +0100208 if (dma) {
209 init_emergency_isa_pool();
210 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Jens Axboe260a67a2010-10-01 14:42:43 +0200211 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100212 }
213}
Jens Axboe86db1e22008-01-29 14:53:40 +0100214EXPORT_SYMBOL(blk_queue_bounce_limit);
215
216/**
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100217 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
218 * @limits: the queue limits
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500219 * @max_hw_sectors: max hardware sectors in the usual 512b unit
Jens Axboe86db1e22008-01-29 14:53:40 +0100220 *
221 * Description:
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500222 * Enables a low level driver to set a hard upper limit,
223 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
224 * the device driver based upon the combined capabilities of I/O
225 * controller and storage device.
226 *
227 * max_sectors is a soft limit imposed by the block layer for
228 * filesystem type requests. This value can be overridden on a
229 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
230 * The soft limit can not exceed max_hw_sectors.
Jens Axboe86db1e22008-01-29 14:53:40 +0100231 **/
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100232void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
Jens Axboe86db1e22008-01-29 14:53:40 +0100233{
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500234 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
235 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700236 printk(KERN_INFO "%s: set to minimum %d\n",
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500237 __func__, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100238 }
239
Mike Snitzer72d4cd92010-12-17 08:34:20 +0100240 limits->max_hw_sectors = max_hw_sectors;
241 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
242 BLK_DEF_MAX_SECTORS);
243}
244EXPORT_SYMBOL(blk_limits_max_hw_sectors);
245
246/**
247 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
248 * @q: the request queue for the device
249 * @max_hw_sectors: max hardware sectors in the usual 512b unit
250 *
251 * Description:
252 * See description for blk_limits_max_hw_sectors().
253 **/
254void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
255{
256 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100257}
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500258EXPORT_SYMBOL(blk_queue_max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100259
260/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200261 * blk_queue_max_discard_sectors - set max sectors for a single discard
262 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200263 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200264 **/
265void blk_queue_max_discard_sectors(struct request_queue *q,
266 unsigned int max_discard_sectors)
267{
268 q->limits.max_discard_sectors = max_discard_sectors;
269}
270EXPORT_SYMBOL(blk_queue_max_discard_sectors);
271
272/**
Martin K. Petersen8a783622010-02-26 00:20:39 -0500273 * blk_queue_max_segments - set max hw segments for a request for this queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100274 * @q: the request queue for the device
275 * @max_segments: max number of segments
276 *
277 * Description:
278 * Enables a low level driver to set an upper limit on the number of
Martin K. Petersen8a783622010-02-26 00:20:39 -0500279 * hw data segments in a request.
Jens Axboe86db1e22008-01-29 14:53:40 +0100280 **/
Martin K. Petersen8a783622010-02-26 00:20:39 -0500281void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
Jens Axboe86db1e22008-01-29 14:53:40 +0100282{
283 if (!max_segments) {
284 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700285 printk(KERN_INFO "%s: set to minimum %d\n",
286 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100287 }
288
Martin K. Petersen8a783622010-02-26 00:20:39 -0500289 q->limits.max_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100290}
Martin K. Petersen8a783622010-02-26 00:20:39 -0500291EXPORT_SYMBOL(blk_queue_max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100292
293/**
294 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
295 * @q: the request queue for the device
296 * @max_size: max size of segment in bytes
297 *
298 * Description:
299 * Enables a low level driver to set an upper limit on the size of a
300 * coalesced segment
301 **/
302void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
303{
304 if (max_size < PAGE_CACHE_SIZE) {
305 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700306 printk(KERN_INFO "%s: set to minimum %d\n",
307 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100308 }
309
Martin K. Petersen025146e2009-05-22 17:17:51 -0400310 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100311}
Jens Axboe86db1e22008-01-29 14:53:40 +0100312EXPORT_SYMBOL(blk_queue_max_segment_size);
313
314/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400315 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100316 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400317 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100318 *
319 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400320 * This should be set to the lowest possible block size that the
321 * storage device can address. The default of 512 covers most
322 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100323 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400324void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100325{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400326 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400327
328 if (q->limits.physical_block_size < size)
329 q->limits.physical_block_size = size;
330
331 if (q->limits.io_min < q->limits.physical_block_size)
332 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100333}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400334EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100335
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400336/**
337 * blk_queue_physical_block_size - set physical block size for the queue
338 * @q: the request queue for the device
339 * @size: the physical block size, in bytes
340 *
341 * Description:
342 * This should be set to the lowest possible sector size that the
343 * hardware can operate on without reverting to read-modify-write
344 * operations.
345 */
Martin K. Petersen892b6f92010-10-13 21:18:03 +0200346void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400347{
348 q->limits.physical_block_size = size;
349
350 if (q->limits.physical_block_size < q->limits.logical_block_size)
351 q->limits.physical_block_size = q->limits.logical_block_size;
352
353 if (q->limits.io_min < q->limits.physical_block_size)
354 q->limits.io_min = q->limits.physical_block_size;
355}
356EXPORT_SYMBOL(blk_queue_physical_block_size);
357
358/**
359 * blk_queue_alignment_offset - set physical block alignment offset
360 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700361 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400362 *
363 * Description:
364 * Some devices are naturally misaligned to compensate for things like
365 * the legacy DOS partition table 63-sector offset. Low-level drivers
366 * should call this function for devices whose first sector is not
367 * naturally aligned.
368 */
369void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
370{
371 q->limits.alignment_offset =
372 offset & (q->limits.physical_block_size - 1);
373 q->limits.misaligned = 0;
374}
375EXPORT_SYMBOL(blk_queue_alignment_offset);
376
377/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400378 * blk_limits_io_min - set minimum request size for a device
379 * @limits: the queue limits
380 * @min: smallest I/O size in bytes
381 *
382 * Description:
383 * Some devices have an internal block size bigger than the reported
384 * hardware sector size. This function can be used to signal the
385 * smallest I/O the device can perform without incurring a performance
386 * penalty.
387 */
388void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
389{
390 limits->io_min = min;
391
392 if (limits->io_min < limits->logical_block_size)
393 limits->io_min = limits->logical_block_size;
394
395 if (limits->io_min < limits->physical_block_size)
396 limits->io_min = limits->physical_block_size;
397}
398EXPORT_SYMBOL(blk_limits_io_min);
399
400/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400401 * blk_queue_io_min - set minimum request size for the queue
402 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700403 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400404 *
405 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400406 * Storage devices may report a granularity or preferred minimum I/O
407 * size which is the smallest request the device can perform without
408 * incurring a performance penalty. For disk drives this is often the
409 * physical block size. For RAID arrays it is often the stripe chunk
410 * size. A properly aligned multiple of minimum_io_size is the
411 * preferred request size for workloads where a high number of I/O
412 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400413 */
414void blk_queue_io_min(struct request_queue *q, unsigned int min)
415{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400416 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400417}
418EXPORT_SYMBOL(blk_queue_io_min);
419
420/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200421 * blk_limits_io_opt - set optimal request size for a device
422 * @limits: the queue limits
423 * @opt: smallest I/O size in bytes
424 *
425 * Description:
426 * Storage devices may report an optimal I/O size, which is the
427 * device's preferred unit for sustained I/O. This is rarely reported
428 * for disk drives. For RAID arrays it is usually the stripe width or
429 * the internal track size. A properly aligned multiple of
430 * optimal_io_size is the preferred request size for workloads where
431 * sustained throughput is desired.
432 */
433void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
434{
435 limits->io_opt = opt;
436}
437EXPORT_SYMBOL(blk_limits_io_opt);
438
439/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400440 * blk_queue_io_opt - set optimal request size for the queue
441 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700442 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400443 *
444 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400445 * Storage devices may report an optimal I/O size, which is the
446 * device's preferred unit for sustained I/O. This is rarely reported
447 * for disk drives. For RAID arrays it is usually the stripe width or
448 * the internal track size. A properly aligned multiple of
449 * optimal_io_size is the preferred request size for workloads where
450 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400451 */
452void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
453{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200454 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400455}
456EXPORT_SYMBOL(blk_queue_io_opt);
457
Jens Axboe86db1e22008-01-29 14:53:40 +0100458/**
459 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
460 * @t: the stacking driver (top)
461 * @b: the underlying device (bottom)
462 **/
463void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
464{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400465 blk_stack_limits(&t->limits, &b->limits, 0);
Jens Axboe86db1e22008-01-29 14:53:40 +0100466}
Jens Axboe86db1e22008-01-29 14:53:40 +0100467EXPORT_SYMBOL(blk_queue_stack_limits);
468
469/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400470 * blk_stack_limits - adjust queue_limits for stacked devices
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100471 * @t: the stacking driver limits (top device)
472 * @b: the underlying queue limits (bottom, component device)
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500473 * @start: first data sector within component device
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400474 *
475 * Description:
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100476 * This function is used by stacking drivers like MD and DM to ensure
477 * that all component devices have compatible block sizes and
478 * alignments. The stacking driver must provide a queue_limits
479 * struct (top) and then iteratively call the stacking function for
480 * all component (bottom) devices. The stacking function will
481 * attempt to combine the values and ensure proper alignment.
482 *
483 * Returns 0 if the top and bottom queue_limits are compatible. The
484 * top device's block sizes and alignment offsets may be adjusted to
485 * ensure alignment with the bottom device. If no compatible sizes
486 * and alignments exist, -1 is returned and the resulting top
487 * queue_limits will have the misaligned flag set to indicate that
488 * the alignment_offset is undefined.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400489 */
490int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500491 sector_t start)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400492{
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500493 unsigned int top, bottom, alignment, ret = 0;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100494
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400495 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
496 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200497 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400498
499 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
500 b->seg_boundary_mask);
501
Martin K. Petersen8a783622010-02-26 00:20:39 -0500502 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200503 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
504 b->max_integrity_segments);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400505
506 t->max_segment_size = min_not_zero(t->max_segment_size,
507 b->max_segment_size);
508
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500509 t->misaligned |= b->misaligned;
510
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500511 alignment = queue_limit_alignment_offset(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100512
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100513 /* Bottom device has different alignment. Check that it is
514 * compatible with the current top alignment.
515 */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100516 if (t->alignment_offset != alignment) {
517
518 top = max(t->physical_block_size, t->io_min)
519 + t->alignment_offset;
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100520 bottom = max(b->physical_block_size, b->io_min) + alignment;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100521
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100522 /* Verify that top and bottom intervals line up */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500523 if (max(top, bottom) & (min(top, bottom) - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100524 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500525 ret = -1;
526 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100527 }
528
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400529 t->logical_block_size = max(t->logical_block_size,
530 b->logical_block_size);
531
532 t->physical_block_size = max(t->physical_block_size,
533 b->physical_block_size);
534
535 t->io_min = max(t->io_min, b->io_min);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100536 t->io_opt = lcm(t->io_opt, b->io_opt);
537
Martin K. Petersene692cb62010-12-01 19:41:49 +0100538 t->cluster &= b->cluster;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100539 t->discard_zeroes_data &= b->discard_zeroes_data;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400540
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100541 /* Physical block size a multiple of the logical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100542 if (t->physical_block_size & (t->logical_block_size - 1)) {
543 t->physical_block_size = t->logical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400544 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500545 ret = -1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100546 }
547
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100548 /* Minimum I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100549 if (t->io_min & (t->physical_block_size - 1)) {
550 t->io_min = t->physical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400551 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500552 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400553 }
554
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100555 /* Optimal I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100556 if (t->io_opt & (t->physical_block_size - 1)) {
557 t->io_opt = 0;
558 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500559 ret = -1;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100560 }
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400561
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100562 /* Find lowest common alignment_offset */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100563 t->alignment_offset = lcm(t->alignment_offset, alignment)
564 & (max(t->physical_block_size, t->io_min) - 1);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400565
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100566 /* Verify that new alignment_offset is on a logical block boundary */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500567 if (t->alignment_offset & (t->logical_block_size - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100568 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500569 ret = -1;
570 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100571
572 /* Discard alignment and granularity */
573 if (b->discard_granularity) {
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500574 alignment = queue_limit_discard_alignment(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100575
576 if (t->discard_granularity != 0 &&
577 t->discard_alignment != alignment) {
578 top = t->discard_granularity + t->discard_alignment;
579 bottom = b->discard_granularity + alignment;
580
581 /* Verify that top and bottom intervals line up */
582 if (max(top, bottom) & (min(top, bottom) - 1))
583 t->discard_misaligned = 1;
584 }
585
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100586 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
587 b->max_discard_sectors);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100588 t->discard_granularity = max(t->discard_granularity,
589 b->discard_granularity);
590 t->discard_alignment = lcm(t->discard_alignment, alignment) &
591 (t->discard_granularity - 1);
592 }
593
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500594 return ret;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400595}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200596EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400597
598/**
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500599 * bdev_stack_limits - adjust queue limits for stacked drivers
600 * @t: the stacking driver limits (top device)
601 * @bdev: the component block_device (bottom)
602 * @start: first data sector within component device
603 *
604 * Description:
605 * Merges queue limits for a top device and a block_device. Returns
606 * 0 if alignment didn't change. Returns -1 if adding the bottom
607 * device caused misalignment.
608 */
609int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
610 sector_t start)
611{
612 struct request_queue *bq = bdev_get_queue(bdev);
613
614 start += get_start_sect(bdev);
615
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500616 return blk_stack_limits(t, &bq->limits, start);
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500617}
618EXPORT_SYMBOL(bdev_stack_limits);
619
620/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400621 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200622 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400623 * @bdev: the underlying block device (bottom)
624 * @offset: offset to beginning of data within component device
625 *
626 * Description:
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500627 * Merges the limits for a top level gendisk and a bottom level
628 * block_device.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400629 */
630void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
631 sector_t offset)
632{
633 struct request_queue *t = disk->queue;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400634
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500635 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400636 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
637
638 disk_name(disk, 0, top);
639 bdevname(bdev, bottom);
640
641 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
642 top, bottom);
643 }
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400644}
645EXPORT_SYMBOL(disk_stack_limits);
646
647/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100648 * blk_queue_dma_pad - set pad mask
649 * @q: the request queue for the device
650 * @mask: pad mask
651 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200652 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100653 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200654 * Appending pad buffer to a request modifies the last entry of a
655 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100656 **/
657void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
658{
659 q->dma_pad_mask = mask;
660}
661EXPORT_SYMBOL(blk_queue_dma_pad);
662
663/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200664 * blk_queue_update_dma_pad - update pad mask
665 * @q: the request queue for the device
666 * @mask: pad mask
667 *
668 * Update dma pad mask.
669 *
670 * Appending pad buffer to a request modifies the last entry of a
671 * scatter list such that it includes the pad buffer.
672 **/
673void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
674{
675 if (mask > q->dma_pad_mask)
676 q->dma_pad_mask = mask;
677}
678EXPORT_SYMBOL(blk_queue_update_dma_pad);
679
680/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100681 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100682 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100683 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100684 * @buf: physically contiguous buffer
685 * @size: size of the buffer in bytes
686 *
687 * Some devices have excess DMA problems and can't simply discard (or
688 * zero fill) the unwanted piece of the transfer. They have to have a
689 * real area of memory to transfer it into. The use case for this is
690 * ATAPI devices in DMA mode. If the packet command causes a transfer
691 * bigger than the transfer size some HBAs will lock up if there
692 * aren't DMA elements to contain the excess transfer. What this API
693 * does is adjust the queue so that the buf is always appended
694 * silently to the scatterlist.
695 *
Martin K. Petersen8a783622010-02-26 00:20:39 -0500696 * Note: This routine adjusts max_hw_segments to make room for appending
697 * the drain buffer. If you call blk_queue_max_segments() after calling
698 * this routine, you must set the limit to one fewer than your device
699 * can support otherwise there won't be room for the drain buffer.
Jens Axboe86db1e22008-01-29 14:53:40 +0100700 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100701int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100702 dma_drain_needed_fn *dma_drain_needed,
703 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100704{
Martin K. Petersen8a783622010-02-26 00:20:39 -0500705 if (queue_max_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100706 return -EINVAL;
707 /* make room for appending the drain */
Martin K. Petersen8a783622010-02-26 00:20:39 -0500708 blk_queue_max_segments(q, queue_max_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100709 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100710 q->dma_drain_buffer = buf;
711 q->dma_drain_size = size;
712
713 return 0;
714}
Jens Axboe86db1e22008-01-29 14:53:40 +0100715EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
716
717/**
718 * blk_queue_segment_boundary - set boundary rules for segment merging
719 * @q: the request queue for the device
720 * @mask: the memory boundary mask
721 **/
722void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
723{
724 if (mask < PAGE_CACHE_SIZE - 1) {
725 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700726 printk(KERN_INFO "%s: set to minimum %lx\n",
727 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100728 }
729
Martin K. Petersen025146e2009-05-22 17:17:51 -0400730 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100731}
Jens Axboe86db1e22008-01-29 14:53:40 +0100732EXPORT_SYMBOL(blk_queue_segment_boundary);
733
734/**
735 * blk_queue_dma_alignment - set dma length and memory alignment
736 * @q: the request queue for the device
737 * @mask: alignment mask
738 *
739 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200740 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100741 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100742 *
743 **/
744void blk_queue_dma_alignment(struct request_queue *q, int mask)
745{
746 q->dma_alignment = mask;
747}
Jens Axboe86db1e22008-01-29 14:53:40 +0100748EXPORT_SYMBOL(blk_queue_dma_alignment);
749
750/**
751 * blk_queue_update_dma_alignment - update dma length and memory alignment
752 * @q: the request queue for the device
753 * @mask: alignment mask
754 *
755 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200756 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100757 * If the requested alignment is larger than the current alignment, then
758 * the current queue alignment is updated to the new value, otherwise it
759 * is left alone. The design of this is to allow multiple objects
760 * (driver, device, transport etc) to set their respective
761 * alignments without having them interfere.
762 *
763 **/
764void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
765{
766 BUG_ON(mask > PAGE_SIZE);
767
768 if (mask > q->dma_alignment)
769 q->dma_alignment = mask;
770}
Jens Axboe86db1e22008-01-29 14:53:40 +0100771EXPORT_SYMBOL(blk_queue_update_dma_alignment);
772
Tejun Heo4913efe2010-09-03 11:56:16 +0200773/**
774 * blk_queue_flush - configure queue's cache flush capability
775 * @q: the request queue for the device
776 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
777 *
778 * Tell block layer cache flush capability of @q. If it supports
779 * flushing, REQ_FLUSH should be set. If it supports bypassing
780 * write cache for individual writes, REQ_FUA should be set.
781 */
782void blk_queue_flush(struct request_queue *q, unsigned int flush)
783{
784 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
785
786 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
787 flush &= ~REQ_FUA;
788
789 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
790}
791EXPORT_SYMBOL_GPL(blk_queue_flush);
792
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200793static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100794{
795 blk_max_low_pfn = max_low_pfn - 1;
796 blk_max_pfn = max_pfn - 1;
797 return 0;
798}
799subsys_initcall(blk_settings_init);