blob: 701859fb9647c31a505f0218800744e3e6d0a775 [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;
129 lim->no_cluster = 0;
130}
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
167 q->unplug_thresh = 4; /* hmm */
Randy Dunlapad5ebd22009-11-11 13:47:45 +0100168 q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */
Jens Axboe86db1e22008-01-29 14:53:40 +0100169 if (q->unplug_delay == 0)
170 q->unplug_delay = 1;
171
Jens Axboe86db1e22008-01-29 14:53:40 +0100172 q->unplug_timer.function = blk_unplug_timeout;
173 q->unplug_timer.data = (unsigned long)q;
174
Martin K. Petersene475bba2009-06-16 08:23:52 +0200175 blk_set_default_limits(&q->limits);
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500176 blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200177
Jens Axboe86db1e22008-01-29 14:53:40 +0100178 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200179 * If the caller didn't supply a lock, fall back to our embedded
180 * per-queue locks
181 */
182 if (!q->queue_lock)
183 q->queue_lock = &q->__queue_lock;
184
185 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100186 * by default assume old behaviour and bounce for any highmem page
187 */
188 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
189}
Jens Axboe86db1e22008-01-29 14:53:40 +0100190EXPORT_SYMBOL(blk_queue_make_request);
191
192/**
193 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900194 * @q: the request queue for the device
195 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100196 *
197 * Description:
198 * Different hardware can have different requirements as to what pages
199 * it can do I/O directly to. A low level driver can call
200 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900201 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100202 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900203void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100204{
Tejun Heocd0aca22009-04-15 22:10:25 +0900205 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100206 int dma = 0;
207
208 q->bounce_gfp = GFP_NOIO;
209#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900210 /*
211 * Assume anything <= 4GB can be handled by IOMMU. Actually
212 * some IOMMUs can handle everything, but I don't know of a
213 * way to test this here.
214 */
215 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100216 dma = 1;
Malahal Naineniefb012b32010-10-01 14:45:27 +0200217 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
Jens Axboe86db1e22008-01-29 14:53:40 +0100218#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100219 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100220 dma = 1;
Malahal Nainenic49825f2010-09-24 20:25:49 +0200221 q->limits.bounce_pfn = b_pfn;
Jens Axboe260a67a2010-10-01 14:42:43 +0200222#endif
Jens Axboe86db1e22008-01-29 14:53:40 +0100223 if (dma) {
224 init_emergency_isa_pool();
225 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Jens Axboe260a67a2010-10-01 14:42:43 +0200226 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100227 }
228}
Jens Axboe86db1e22008-01-29 14:53:40 +0100229EXPORT_SYMBOL(blk_queue_bounce_limit);
230
231/**
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500232 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100233 * @q: the request queue for the device
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500234 * @max_hw_sectors: max hardware sectors in the usual 512b unit
Jens Axboe86db1e22008-01-29 14:53:40 +0100235 *
236 * Description:
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500237 * Enables a low level driver to set a hard upper limit,
238 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
239 * the device driver based upon the combined capabilities of I/O
240 * controller and storage device.
241 *
242 * max_sectors is a soft limit imposed by the block layer for
243 * filesystem type requests. This value can be overridden on a
244 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
245 * The soft limit can not exceed max_hw_sectors.
Jens Axboe86db1e22008-01-29 14:53:40 +0100246 **/
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500247void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
Jens Axboe86db1e22008-01-29 14:53:40 +0100248{
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500249 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
250 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700251 printk(KERN_INFO "%s: set to minimum %d\n",
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500252 __func__, max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100253 }
254
Martin K. Petersen2800aac2010-02-26 00:20:35 -0500255 q->limits.max_hw_sectors = max_hw_sectors;
256 q->limits.max_sectors = min_t(unsigned int, max_hw_sectors,
257 BLK_DEF_MAX_SECTORS);
Jens Axboe86db1e22008-01-29 14:53:40 +0100258}
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500259EXPORT_SYMBOL(blk_queue_max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100260
261/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200262 * blk_queue_max_discard_sectors - set max sectors for a single discard
263 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200264 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200265 **/
266void blk_queue_max_discard_sectors(struct request_queue *q,
267 unsigned int max_discard_sectors)
268{
269 q->limits.max_discard_sectors = max_discard_sectors;
270}
271EXPORT_SYMBOL(blk_queue_max_discard_sectors);
272
273/**
Martin K. Petersen8a783622010-02-26 00:20:39 -0500274 * blk_queue_max_segments - set max hw segments for a request for this queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100275 * @q: the request queue for the device
276 * @max_segments: max number of segments
277 *
278 * Description:
279 * Enables a low level driver to set an upper limit on the number of
Martin K. Petersen8a783622010-02-26 00:20:39 -0500280 * hw data segments in a request.
Jens Axboe86db1e22008-01-29 14:53:40 +0100281 **/
Martin K. Petersen8a783622010-02-26 00:20:39 -0500282void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
Jens Axboe86db1e22008-01-29 14:53:40 +0100283{
284 if (!max_segments) {
285 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700286 printk(KERN_INFO "%s: set to minimum %d\n",
287 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100288 }
289
Martin K. Petersen8a783622010-02-26 00:20:39 -0500290 q->limits.max_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100291}
Martin K. Petersen8a783622010-02-26 00:20:39 -0500292EXPORT_SYMBOL(blk_queue_max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100293
294/**
295 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
296 * @q: the request queue for the device
297 * @max_size: max size of segment in bytes
298 *
299 * Description:
300 * Enables a low level driver to set an upper limit on the size of a
301 * coalesced segment
302 **/
303void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
304{
305 if (max_size < PAGE_CACHE_SIZE) {
306 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700307 printk(KERN_INFO "%s: set to minimum %d\n",
308 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100309 }
310
Martin K. Petersen025146e2009-05-22 17:17:51 -0400311 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100312}
Jens Axboe86db1e22008-01-29 14:53:40 +0100313EXPORT_SYMBOL(blk_queue_max_segment_size);
314
315/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400316 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100317 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400318 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100319 *
320 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400321 * This should be set to the lowest possible block size that the
322 * storage device can address. The default of 512 covers most
323 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100324 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400325void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100326{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400327 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400328
329 if (q->limits.physical_block_size < size)
330 q->limits.physical_block_size = size;
331
332 if (q->limits.io_min < q->limits.physical_block_size)
333 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100334}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400335EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100336
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400337/**
338 * blk_queue_physical_block_size - set physical block size for the queue
339 * @q: the request queue for the device
340 * @size: the physical block size, in bytes
341 *
342 * Description:
343 * This should be set to the lowest possible sector size that the
344 * hardware can operate on without reverting to read-modify-write
345 * operations.
346 */
Martin K. Petersen892b6f92010-10-13 21:18:03 +0200347void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400348{
349 q->limits.physical_block_size = size;
350
351 if (q->limits.physical_block_size < q->limits.logical_block_size)
352 q->limits.physical_block_size = q->limits.logical_block_size;
353
354 if (q->limits.io_min < q->limits.physical_block_size)
355 q->limits.io_min = q->limits.physical_block_size;
356}
357EXPORT_SYMBOL(blk_queue_physical_block_size);
358
359/**
360 * blk_queue_alignment_offset - set physical block alignment offset
361 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700362 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400363 *
364 * Description:
365 * Some devices are naturally misaligned to compensate for things like
366 * the legacy DOS partition table 63-sector offset. Low-level drivers
367 * should call this function for devices whose first sector is not
368 * naturally aligned.
369 */
370void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
371{
372 q->limits.alignment_offset =
373 offset & (q->limits.physical_block_size - 1);
374 q->limits.misaligned = 0;
375}
376EXPORT_SYMBOL(blk_queue_alignment_offset);
377
378/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400379 * blk_limits_io_min - set minimum request size for a device
380 * @limits: the queue limits
381 * @min: smallest I/O size in bytes
382 *
383 * Description:
384 * Some devices have an internal block size bigger than the reported
385 * hardware sector size. This function can be used to signal the
386 * smallest I/O the device can perform without incurring a performance
387 * penalty.
388 */
389void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
390{
391 limits->io_min = min;
392
393 if (limits->io_min < limits->logical_block_size)
394 limits->io_min = limits->logical_block_size;
395
396 if (limits->io_min < limits->physical_block_size)
397 limits->io_min = limits->physical_block_size;
398}
399EXPORT_SYMBOL(blk_limits_io_min);
400
401/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400402 * blk_queue_io_min - set minimum request size for the queue
403 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700404 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400405 *
406 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400407 * Storage devices may report a granularity or preferred minimum I/O
408 * size which is the smallest request the device can perform without
409 * incurring a performance penalty. For disk drives this is often the
410 * physical block size. For RAID arrays it is often the stripe chunk
411 * size. A properly aligned multiple of minimum_io_size is the
412 * preferred request size for workloads where a high number of I/O
413 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400414 */
415void blk_queue_io_min(struct request_queue *q, unsigned int min)
416{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400417 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400418}
419EXPORT_SYMBOL(blk_queue_io_min);
420
421/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200422 * blk_limits_io_opt - set optimal request size for a device
423 * @limits: the queue limits
424 * @opt: smallest I/O size in bytes
425 *
426 * Description:
427 * Storage devices may report an optimal I/O size, which is the
428 * device's preferred unit for sustained I/O. This is rarely reported
429 * for disk drives. For RAID arrays it is usually the stripe width or
430 * the internal track size. A properly aligned multiple of
431 * optimal_io_size is the preferred request size for workloads where
432 * sustained throughput is desired.
433 */
434void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
435{
436 limits->io_opt = opt;
437}
438EXPORT_SYMBOL(blk_limits_io_opt);
439
440/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400441 * blk_queue_io_opt - set optimal request size for the queue
442 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700443 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400444 *
445 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400446 * Storage devices may report an optimal I/O size, which is the
447 * device's preferred unit for sustained I/O. This is rarely reported
448 * for disk drives. For RAID arrays it is usually the stripe width or
449 * the internal track size. A properly aligned multiple of
450 * optimal_io_size is the preferred request size for workloads where
451 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400452 */
453void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
454{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200455 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400456}
457EXPORT_SYMBOL(blk_queue_io_opt);
458
Jens Axboe86db1e22008-01-29 14:53:40 +0100459/**
460 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
461 * @t: the stacking driver (top)
462 * @b: the underlying device (bottom)
463 **/
464void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
465{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400466 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400467
Neil Browne7e72bf2008-05-14 16:05:54 -0700468 if (!t->queue_lock)
469 WARN_ON_ONCE(1);
470 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
471 unsigned long flags;
472 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200473 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700474 spin_unlock_irqrestore(t->queue_lock, flags);
475 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100476}
Jens Axboe86db1e22008-01-29 14:53:40 +0100477EXPORT_SYMBOL(blk_queue_stack_limits);
478
479/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400480 * blk_stack_limits - adjust queue_limits for stacked devices
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100481 * @t: the stacking driver limits (top device)
482 * @b: the underlying queue limits (bottom, component device)
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500483 * @start: first data sector within component device
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400484 *
485 * Description:
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100486 * This function is used by stacking drivers like MD and DM to ensure
487 * that all component devices have compatible block sizes and
488 * alignments. The stacking driver must provide a queue_limits
489 * struct (top) and then iteratively call the stacking function for
490 * all component (bottom) devices. The stacking function will
491 * attempt to combine the values and ensure proper alignment.
492 *
493 * Returns 0 if the top and bottom queue_limits are compatible. The
494 * top device's block sizes and alignment offsets may be adjusted to
495 * ensure alignment with the bottom device. If no compatible sizes
496 * and alignments exist, -1 is returned and the resulting top
497 * queue_limits will have the misaligned flag set to indicate that
498 * the alignment_offset is undefined.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400499 */
500int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500501 sector_t start)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400502{
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500503 unsigned int top, bottom, alignment, ret = 0;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100504
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400505 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
506 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200507 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400508
509 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
510 b->seg_boundary_mask);
511
Martin K. Petersen8a783622010-02-26 00:20:39 -0500512 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200513 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
514 b->max_integrity_segments);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400515
516 t->max_segment_size = min_not_zero(t->max_segment_size,
517 b->max_segment_size);
518
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500519 t->misaligned |= b->misaligned;
520
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500521 alignment = queue_limit_alignment_offset(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100522
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100523 /* Bottom device has different alignment. Check that it is
524 * compatible with the current top alignment.
525 */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100526 if (t->alignment_offset != alignment) {
527
528 top = max(t->physical_block_size, t->io_min)
529 + t->alignment_offset;
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100530 bottom = max(b->physical_block_size, b->io_min) + alignment;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100531
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100532 /* Verify that top and bottom intervals line up */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500533 if (max(top, bottom) & (min(top, bottom) - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100534 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500535 ret = -1;
536 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100537 }
538
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400539 t->logical_block_size = max(t->logical_block_size,
540 b->logical_block_size);
541
542 t->physical_block_size = max(t->physical_block_size,
543 b->physical_block_size);
544
545 t->io_min = max(t->io_min, b->io_min);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100546 t->io_opt = lcm(t->io_opt, b->io_opt);
547
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400548 t->no_cluster |= b->no_cluster;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100549 t->discard_zeroes_data &= b->discard_zeroes_data;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400550
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100551 /* Physical block size a multiple of the logical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100552 if (t->physical_block_size & (t->logical_block_size - 1)) {
553 t->physical_block_size = t->logical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400554 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500555 ret = -1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100556 }
557
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100558 /* Minimum I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100559 if (t->io_min & (t->physical_block_size - 1)) {
560 t->io_min = t->physical_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. Petersenc72758f2009-05-22 17:17:53 -0400563 }
564
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100565 /* Optimal I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100566 if (t->io_opt & (t->physical_block_size - 1)) {
567 t->io_opt = 0;
568 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500569 ret = -1;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100570 }
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400571
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100572 /* Find lowest common alignment_offset */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100573 t->alignment_offset = lcm(t->alignment_offset, alignment)
574 & (max(t->physical_block_size, t->io_min) - 1);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400575
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100576 /* Verify that new alignment_offset is on a logical block boundary */
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500577 if (t->alignment_offset & (t->logical_block_size - 1)) {
Martin K. Petersen9504e082009-12-21 15:55:51 +0100578 t->misaligned = 1;
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500579 ret = -1;
580 }
Martin K. Petersen9504e082009-12-21 15:55:51 +0100581
582 /* Discard alignment and granularity */
583 if (b->discard_granularity) {
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500584 alignment = queue_limit_discard_alignment(b, start);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100585
586 if (t->discard_granularity != 0 &&
587 t->discard_alignment != alignment) {
588 top = t->discard_granularity + t->discard_alignment;
589 bottom = b->discard_granularity + alignment;
590
591 /* Verify that top and bottom intervals line up */
592 if (max(top, bottom) & (min(top, bottom) - 1))
593 t->discard_misaligned = 1;
594 }
595
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100596 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
597 b->max_discard_sectors);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100598 t->discard_granularity = max(t->discard_granularity,
599 b->discard_granularity);
600 t->discard_alignment = lcm(t->discard_alignment, alignment) &
601 (t->discard_granularity - 1);
602 }
603
Martin K. Petersenfe0b3932010-01-11 03:21:47 -0500604 return ret;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400605}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200606EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400607
608/**
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500609 * bdev_stack_limits - adjust queue limits for stacked drivers
610 * @t: the stacking driver limits (top device)
611 * @bdev: the component block_device (bottom)
612 * @start: first data sector within component device
613 *
614 * Description:
615 * Merges queue limits for a top device and a block_device. Returns
616 * 0 if alignment didn't change. Returns -1 if adding the bottom
617 * device caused misalignment.
618 */
619int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
620 sector_t start)
621{
622 struct request_queue *bq = bdev_get_queue(bdev);
623
624 start += get_start_sect(bdev);
625
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500626 return blk_stack_limits(t, &bq->limits, start);
Martin K. Petersen17be8c22010-01-11 03:21:49 -0500627}
628EXPORT_SYMBOL(bdev_stack_limits);
629
630/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400631 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200632 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400633 * @bdev: the underlying block device (bottom)
634 * @offset: offset to beginning of data within component device
635 *
636 * Description:
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500637 * Merges the limits for a top level gendisk and a bottom level
638 * block_device.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400639 */
640void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
641 sector_t offset)
642{
643 struct request_queue *t = disk->queue;
644 struct request_queue *b = bdev_get_queue(bdev);
645
Martin K. Petersene03a72e2010-01-11 03:21:51 -0500646 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400647 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
648
649 disk_name(disk, 0, top);
650 bdevname(bdev, bottom);
651
652 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
653 top, bottom);
654 }
655
656 if (!t->queue_lock)
657 WARN_ON_ONCE(1);
658 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
659 unsigned long flags;
660
661 spin_lock_irqsave(t->queue_lock, flags);
662 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
663 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
664 spin_unlock_irqrestore(t->queue_lock, flags);
665 }
666}
667EXPORT_SYMBOL(disk_stack_limits);
668
669/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100670 * blk_queue_dma_pad - set pad mask
671 * @q: the request queue for the device
672 * @mask: pad mask
673 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200674 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100675 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200676 * Appending pad buffer to a request modifies the last entry of a
677 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100678 **/
679void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
680{
681 q->dma_pad_mask = mask;
682}
683EXPORT_SYMBOL(blk_queue_dma_pad);
684
685/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200686 * blk_queue_update_dma_pad - update pad mask
687 * @q: the request queue for the device
688 * @mask: pad mask
689 *
690 * Update dma pad mask.
691 *
692 * Appending pad buffer to a request modifies the last entry of a
693 * scatter list such that it includes the pad buffer.
694 **/
695void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
696{
697 if (mask > q->dma_pad_mask)
698 q->dma_pad_mask = mask;
699}
700EXPORT_SYMBOL(blk_queue_update_dma_pad);
701
702/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100703 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100704 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100705 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100706 * @buf: physically contiguous buffer
707 * @size: size of the buffer in bytes
708 *
709 * Some devices have excess DMA problems and can't simply discard (or
710 * zero fill) the unwanted piece of the transfer. They have to have a
711 * real area of memory to transfer it into. The use case for this is
712 * ATAPI devices in DMA mode. If the packet command causes a transfer
713 * bigger than the transfer size some HBAs will lock up if there
714 * aren't DMA elements to contain the excess transfer. What this API
715 * does is adjust the queue so that the buf is always appended
716 * silently to the scatterlist.
717 *
Martin K. Petersen8a783622010-02-26 00:20:39 -0500718 * Note: This routine adjusts max_hw_segments to make room for appending
719 * the drain buffer. If you call blk_queue_max_segments() after calling
720 * this routine, you must set the limit to one fewer than your device
721 * can support otherwise there won't be room for the drain buffer.
Jens Axboe86db1e22008-01-29 14:53:40 +0100722 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100723int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100724 dma_drain_needed_fn *dma_drain_needed,
725 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100726{
Martin K. Petersen8a783622010-02-26 00:20:39 -0500727 if (queue_max_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100728 return -EINVAL;
729 /* make room for appending the drain */
Martin K. Petersen8a783622010-02-26 00:20:39 -0500730 blk_queue_max_segments(q, queue_max_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100731 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100732 q->dma_drain_buffer = buf;
733 q->dma_drain_size = size;
734
735 return 0;
736}
Jens Axboe86db1e22008-01-29 14:53:40 +0100737EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
738
739/**
740 * blk_queue_segment_boundary - set boundary rules for segment merging
741 * @q: the request queue for the device
742 * @mask: the memory boundary mask
743 **/
744void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
745{
746 if (mask < PAGE_CACHE_SIZE - 1) {
747 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700748 printk(KERN_INFO "%s: set to minimum %lx\n",
749 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100750 }
751
Martin K. Petersen025146e2009-05-22 17:17:51 -0400752 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100753}
Jens Axboe86db1e22008-01-29 14:53:40 +0100754EXPORT_SYMBOL(blk_queue_segment_boundary);
755
756/**
757 * blk_queue_dma_alignment - set dma length and memory alignment
758 * @q: the request queue for the device
759 * @mask: alignment mask
760 *
761 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200762 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100763 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100764 *
765 **/
766void blk_queue_dma_alignment(struct request_queue *q, int mask)
767{
768 q->dma_alignment = mask;
769}
Jens Axboe86db1e22008-01-29 14:53:40 +0100770EXPORT_SYMBOL(blk_queue_dma_alignment);
771
772/**
773 * blk_queue_update_dma_alignment - update dma length and memory alignment
774 * @q: the request queue for the device
775 * @mask: alignment mask
776 *
777 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200778 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100779 * If the requested alignment is larger than the current alignment, then
780 * the current queue alignment is updated to the new value, otherwise it
781 * is left alone. The design of this is to allow multiple objects
782 * (driver, device, transport etc) to set their respective
783 * alignments without having them interfere.
784 *
785 **/
786void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
787{
788 BUG_ON(mask > PAGE_SIZE);
789
790 if (mask > q->dma_alignment)
791 q->dma_alignment = mask;
792}
Jens Axboe86db1e22008-01-29 14:53:40 +0100793EXPORT_SYMBOL(blk_queue_update_dma_alignment);
794
Tejun Heo4913efe2010-09-03 11:56:16 +0200795/**
796 * blk_queue_flush - configure queue's cache flush capability
797 * @q: the request queue for the device
798 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
799 *
800 * Tell block layer cache flush capability of @q. If it supports
801 * flushing, REQ_FLUSH should be set. If it supports bypassing
802 * write cache for individual writes, REQ_FUA should be set.
803 */
804void blk_queue_flush(struct request_queue *q, unsigned int flush)
805{
806 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
807
808 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
809 flush &= ~REQ_FUA;
810
811 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
812}
813EXPORT_SYMBOL_GPL(blk_queue_flush);
814
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200815static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100816{
817 blk_max_low_pfn = max_low_pfn - 1;
818 blk_max_pfn = max_pfn - 1;
819 return 0;
820}
821subsys_initcall(blk_settings_init);