blob: 1ebc1fdb91445dadabe758c3c5dd9e7b73be9830 [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>
Randy Dunlapad5ebd22009-11-11 13:47:45 +010011#include <linux/jiffies.h>
Jens Axboe86db1e22008-01-29 14:53:40 +010012
13#include "blk.h"
14
Jens Axboe6728cb02008-01-31 13:03:55 +010015unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010016EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010017
18unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010019
20/**
21 * blk_queue_prep_rq - set a prepare_request function for queue
22 * @q: queue
23 * @pfn: prepare_request function
24 *
25 * It's possible for a queue to register a prepare_request callback which
26 * is invoked before the request is handed to the request_fn. The goal of
27 * the function is to prepare a request for I/O, it can be used to build a
28 * cdb from the request data for instance.
29 *
30 */
31void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
32{
33 q->prep_rq_fn = pfn;
34}
Jens Axboe86db1e22008-01-29 14:53:40 +010035EXPORT_SYMBOL(blk_queue_prep_rq);
36
37/**
38 * blk_queue_merge_bvec - set a merge_bvec function for queue
39 * @q: queue
40 * @mbfn: merge_bvec_fn
41 *
42 * Usually queues have static limitations on the max sectors or segments that
43 * we can put in a request. Stacking drivers may have some settings that
44 * are dynamic, and thus we have to query the queue whether it is ok to
45 * add a new bio_vec to a bio at a given offset or not. If the block device
46 * has such limitations, it needs to register a merge_bvec_fn to control
47 * the size of bio's sent to it. Note that a block device *must* allow a
48 * single page to be added to an empty bio. The block device driver may want
49 * to use the bio_split() function to deal with these bio's. By default
50 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
51 * honored.
52 */
53void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
54{
55 q->merge_bvec_fn = mbfn;
56}
Jens Axboe86db1e22008-01-29 14:53:40 +010057EXPORT_SYMBOL(blk_queue_merge_bvec);
58
59void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
60{
61 q->softirq_done_fn = fn;
62}
Jens Axboe86db1e22008-01-29 14:53:40 +010063EXPORT_SYMBOL(blk_queue_softirq_done);
64
Jens Axboe242f9dc2008-09-14 05:55:09 -070065void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
66{
67 q->rq_timeout = timeout;
68}
69EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
70
71void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
72{
73 q->rq_timed_out_fn = fn;
74}
75EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
76
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020077void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
78{
79 q->lld_busy_fn = fn;
80}
81EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
82
Jens Axboe86db1e22008-01-29 14:53:40 +010083/**
Martin K. Petersene475bba2009-06-16 08:23:52 +020084 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5ca2009-06-19 09:18:32 +020085 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +020086 *
87 * Description:
88 * Returns a queue_limit struct to its default state. Can be used by
89 * stacking drivers like DM that stage table swaps and reuse an
90 * existing device queue.
91 */
92void blk_set_default_limits(struct queue_limits *lim)
93{
94 lim->max_phys_segments = MAX_PHYS_SEGMENTS;
95 lim->max_hw_segments = MAX_HW_SEGMENTS;
96 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
97 lim->max_segment_size = MAX_SEGMENT_SIZE;
Martin K. Petersen5dee2472009-09-21 21:46:05 +020098 lim->max_sectors = BLK_DEF_MAX_SECTORS;
99 lim->max_hw_sectors = INT_MAX;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100100 lim->max_discard_sectors = 0;
101 lim->discard_granularity = 0;
102 lim->discard_alignment = 0;
103 lim->discard_misaligned = 0;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200104 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200105 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200106 lim->alignment_offset = 0;
107 lim->io_opt = 0;
108 lim->misaligned = 0;
109 lim->no_cluster = 0;
110}
111EXPORT_SYMBOL(blk_set_default_limits);
112
113/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100114 * blk_queue_make_request - define an alternate make_request function for a device
115 * @q: the request queue for the device to be affected
116 * @mfn: the alternate make_request function
117 *
118 * Description:
119 * The normal way for &struct bios to be passed to a device
120 * driver is for them to be collected into requests on a request
121 * queue, and then to allow the device driver to select requests
122 * off that queue when it is ready. This works well for many block
123 * devices. However some block devices (typically virtual devices
124 * such as md or lvm) do not benefit from the processing on the
125 * request queue, and are served best by having the requests passed
126 * directly to them. This can be achieved by providing a function
127 * to blk_queue_make_request().
128 *
129 * Caveat:
130 * The driver that does this *must* be able to deal appropriately
131 * with buffers in "highmemory". This can be accomplished by either calling
132 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
133 * blk_queue_bounce() to create a buffer in normal memory.
134 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100135void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100136{
137 /*
138 * set defaults
139 */
140 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100141
Jens Axboe86db1e22008-01-29 14:53:40 +0100142 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100143 blk_queue_dma_alignment(q, 511);
144 blk_queue_congestion_threshold(q);
145 q->nr_batching = BLK_BATCH_REQ;
146
147 q->unplug_thresh = 4; /* hmm */
Randy Dunlapad5ebd22009-11-11 13:47:45 +0100148 q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */
Jens Axboe86db1e22008-01-29 14:53:40 +0100149 if (q->unplug_delay == 0)
150 q->unplug_delay = 1;
151
Jens Axboe86db1e22008-01-29 14:53:40 +0100152 q->unplug_timer.function = blk_unplug_timeout;
153 q->unplug_timer.data = (unsigned long)q;
154
Martin K. Petersene475bba2009-06-16 08:23:52 +0200155 blk_set_default_limits(&q->limits);
Martin K. Petersen80ddf242009-09-18 22:54:37 +0200156 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200157
Jens Axboe86db1e22008-01-29 14:53:40 +0100158 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200159 * If the caller didn't supply a lock, fall back to our embedded
160 * per-queue locks
161 */
162 if (!q->queue_lock)
163 q->queue_lock = &q->__queue_lock;
164
165 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100166 * by default assume old behaviour and bounce for any highmem page
167 */
168 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
169}
Jens Axboe86db1e22008-01-29 14:53:40 +0100170EXPORT_SYMBOL(blk_queue_make_request);
171
172/**
173 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900174 * @q: the request queue for the device
175 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100176 *
177 * Description:
178 * Different hardware can have different requirements as to what pages
179 * it can do I/O directly to. A low level driver can call
180 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900181 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100182 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900183void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100184{
Tejun Heocd0aca22009-04-15 22:10:25 +0900185 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100186 int dma = 0;
187
188 q->bounce_gfp = GFP_NOIO;
189#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900190 /*
191 * Assume anything <= 4GB can be handled by IOMMU. Actually
192 * some IOMMUs can handle everything, but I don't know of a
193 * way to test this here.
194 */
195 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100196 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400197 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100198#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100199 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100200 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400201 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100202#endif
203 if (dma) {
204 init_emergency_isa_pool();
205 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400206 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100207 }
208}
Jens Axboe86db1e22008-01-29 14:53:40 +0100209EXPORT_SYMBOL(blk_queue_bounce_limit);
210
211/**
212 * blk_queue_max_sectors - set max sectors for a request for this queue
213 * @q: the request queue for the device
214 * @max_sectors: max sectors in the usual 512b unit
215 *
216 * Description:
217 * Enables a low level driver to set an upper limit on the size of
218 * received requests.
219 **/
220void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
221{
222 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
223 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700224 printk(KERN_INFO "%s: set to minimum %d\n",
225 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100226 }
227
228 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400229 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100230 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400231 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
232 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100233 }
234}
Jens Axboe86db1e22008-01-29 14:53:40 +0100235EXPORT_SYMBOL(blk_queue_max_sectors);
236
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400237void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
238{
239 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400240 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400241 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400242 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400243}
244EXPORT_SYMBOL(blk_queue_max_hw_sectors);
245
Jens Axboe86db1e22008-01-29 14:53:40 +0100246/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200247 * blk_queue_max_discard_sectors - set max sectors for a single discard
248 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200249 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200250 **/
251void blk_queue_max_discard_sectors(struct request_queue *q,
252 unsigned int max_discard_sectors)
253{
254 q->limits.max_discard_sectors = max_discard_sectors;
255}
256EXPORT_SYMBOL(blk_queue_max_discard_sectors);
257
258/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100259 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
260 * @q: the request queue for the device
261 * @max_segments: max number of segments
262 *
263 * Description:
264 * Enables a low level driver to set an upper limit on the number of
265 * physical data segments in a request. This would be the largest sized
266 * scatter list the driver could handle.
267 **/
268void blk_queue_max_phys_segments(struct request_queue *q,
269 unsigned short max_segments)
270{
271 if (!max_segments) {
272 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700273 printk(KERN_INFO "%s: set to minimum %d\n",
274 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100275 }
276
Martin K. Petersen025146e2009-05-22 17:17:51 -0400277 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100278}
Jens Axboe86db1e22008-01-29 14:53:40 +0100279EXPORT_SYMBOL(blk_queue_max_phys_segments);
280
281/**
282 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
283 * @q: the request queue for the device
284 * @max_segments: max number of segments
285 *
286 * Description:
287 * Enables a low level driver to set an upper limit on the number of
288 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200289 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100290 * to the device.
291 **/
292void blk_queue_max_hw_segments(struct request_queue *q,
293 unsigned short max_segments)
294{
295 if (!max_segments) {
296 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700297 printk(KERN_INFO "%s: set to minimum %d\n",
298 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100299 }
300
Martin K. Petersen025146e2009-05-22 17:17:51 -0400301 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100302}
Jens Axboe86db1e22008-01-29 14:53:40 +0100303EXPORT_SYMBOL(blk_queue_max_hw_segments);
304
305/**
306 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
307 * @q: the request queue for the device
308 * @max_size: max size of segment in bytes
309 *
310 * Description:
311 * Enables a low level driver to set an upper limit on the size of a
312 * coalesced segment
313 **/
314void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
315{
316 if (max_size < PAGE_CACHE_SIZE) {
317 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700318 printk(KERN_INFO "%s: set to minimum %d\n",
319 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100320 }
321
Martin K. Petersen025146e2009-05-22 17:17:51 -0400322 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100323}
Jens Axboe86db1e22008-01-29 14:53:40 +0100324EXPORT_SYMBOL(blk_queue_max_segment_size);
325
326/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400327 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100328 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400329 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100330 *
331 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400332 * This should be set to the lowest possible block size that the
333 * storage device can address. The default of 512 covers most
334 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100335 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400336void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100337{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400338 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400339
340 if (q->limits.physical_block_size < size)
341 q->limits.physical_block_size = size;
342
343 if (q->limits.io_min < q->limits.physical_block_size)
344 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100345}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400346EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100347
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400348/**
349 * blk_queue_physical_block_size - set physical block size for the queue
350 * @q: the request queue for the device
351 * @size: the physical block size, in bytes
352 *
353 * Description:
354 * This should be set to the lowest possible sector size that the
355 * hardware can operate on without reverting to read-modify-write
356 * operations.
357 */
358void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
359{
360 q->limits.physical_block_size = size;
361
362 if (q->limits.physical_block_size < q->limits.logical_block_size)
363 q->limits.physical_block_size = q->limits.logical_block_size;
364
365 if (q->limits.io_min < q->limits.physical_block_size)
366 q->limits.io_min = q->limits.physical_block_size;
367}
368EXPORT_SYMBOL(blk_queue_physical_block_size);
369
370/**
371 * blk_queue_alignment_offset - set physical block alignment offset
372 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700373 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400374 *
375 * Description:
376 * Some devices are naturally misaligned to compensate for things like
377 * the legacy DOS partition table 63-sector offset. Low-level drivers
378 * should call this function for devices whose first sector is not
379 * naturally aligned.
380 */
381void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
382{
383 q->limits.alignment_offset =
384 offset & (q->limits.physical_block_size - 1);
385 q->limits.misaligned = 0;
386}
387EXPORT_SYMBOL(blk_queue_alignment_offset);
388
389/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400390 * blk_limits_io_min - set minimum request size for a device
391 * @limits: the queue limits
392 * @min: smallest I/O size in bytes
393 *
394 * Description:
395 * Some devices have an internal block size bigger than the reported
396 * hardware sector size. This function can be used to signal the
397 * smallest I/O the device can perform without incurring a performance
398 * penalty.
399 */
400void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
401{
402 limits->io_min = min;
403
404 if (limits->io_min < limits->logical_block_size)
405 limits->io_min = limits->logical_block_size;
406
407 if (limits->io_min < limits->physical_block_size)
408 limits->io_min = limits->physical_block_size;
409}
410EXPORT_SYMBOL(blk_limits_io_min);
411
412/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400413 * blk_queue_io_min - set minimum request size for the queue
414 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700415 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400416 *
417 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400418 * Storage devices may report a granularity or preferred minimum I/O
419 * size which is the smallest request the device can perform without
420 * incurring a performance penalty. For disk drives this is often the
421 * physical block size. For RAID arrays it is often the stripe chunk
422 * size. A properly aligned multiple of minimum_io_size is the
423 * preferred request size for workloads where a high number of I/O
424 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400425 */
426void blk_queue_io_min(struct request_queue *q, unsigned int min)
427{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400428 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400429}
430EXPORT_SYMBOL(blk_queue_io_min);
431
432/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200433 * blk_limits_io_opt - set optimal request size for a device
434 * @limits: the queue limits
435 * @opt: smallest I/O size in bytes
436 *
437 * Description:
438 * Storage devices may report an optimal I/O size, which is the
439 * device's preferred unit for sustained I/O. This is rarely reported
440 * for disk drives. For RAID arrays it is usually the stripe width or
441 * the internal track size. A properly aligned multiple of
442 * optimal_io_size is the preferred request size for workloads where
443 * sustained throughput is desired.
444 */
445void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
446{
447 limits->io_opt = opt;
448}
449EXPORT_SYMBOL(blk_limits_io_opt);
450
451/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400452 * blk_queue_io_opt - set optimal request size for the queue
453 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700454 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400455 *
456 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400457 * Storage devices may report an optimal I/O size, which is the
458 * device's preferred unit for sustained I/O. This is rarely reported
459 * for disk drives. For RAID arrays it is usually the stripe width or
460 * the internal track size. A properly aligned multiple of
461 * optimal_io_size is the preferred request size for workloads where
462 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400463 */
464void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
465{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200466 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400467}
468EXPORT_SYMBOL(blk_queue_io_opt);
469
Jens Axboe86db1e22008-01-29 14:53:40 +0100470/*
471 * Returns the minimum that is _not_ zero, unless both are zero.
472 */
473#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
474
475/**
476 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
477 * @t: the stacking driver (top)
478 * @b: the underlying device (bottom)
479 **/
480void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
481{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400482 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400483
Neil Browne7e72bf2008-05-14 16:05:54 -0700484 if (!t->queue_lock)
485 WARN_ON_ONCE(1);
486 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
487 unsigned long flags;
488 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200489 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700490 spin_unlock_irqrestore(t->queue_lock, flags);
491 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100492}
Jens Axboe86db1e22008-01-29 14:53:40 +0100493EXPORT_SYMBOL(blk_queue_stack_limits);
494
Martin K. Petersen86b37282009-11-10 11:50:21 +0100495static unsigned int lcm(unsigned int a, unsigned int b)
496{
497 if (a && b)
498 return (a * b) / gcd(a, b);
499 else if (b)
500 return b;
501
502 return a;
503}
504
Jens Axboe86db1e22008-01-29 14:53:40 +0100505/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400506 * blk_stack_limits - adjust queue_limits for stacked devices
507 * @t: the stacking driver limits (top)
Martin K. Petersen77634f32009-06-09 06:23:22 +0200508 * @b: the underlying queue limits (bottom)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400509 * @offset: offset to beginning of data within component device
510 *
511 * Description:
512 * Merges two queue_limit structs. Returns 0 if alignment didn't
513 * change. Returns -1 if adding the bottom device caused
514 * misalignment.
515 */
516int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
517 sector_t offset)
518{
Martin K. Petersen86b37282009-11-10 11:50:21 +0100519 int ret;
520
521 ret = 0;
522
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400523 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
524 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200525 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400526
527 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
528 b->seg_boundary_mask);
529
530 t->max_phys_segments = min_not_zero(t->max_phys_segments,
531 b->max_phys_segments);
532
533 t->max_hw_segments = min_not_zero(t->max_hw_segments,
534 b->max_hw_segments);
535
536 t->max_segment_size = min_not_zero(t->max_segment_size,
537 b->max_segment_size);
538
539 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);
546 t->no_cluster |= b->no_cluster;
547
548 /* Bottom device offset aligned? */
549 if (offset &&
550 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
551 t->misaligned = 1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100552 ret = -1;
553 }
554
555 if (offset &&
556 (offset & (b->discard_granularity - 1)) != b->discard_alignment) {
557 t->discard_misaligned = 1;
558 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400559 }
560
561 /* If top has no alignment offset, inherit from bottom */
562 if (!t->alignment_offset)
563 t->alignment_offset =
564 b->alignment_offset & (b->physical_block_size - 1);
565
Martin K. Petersen86b37282009-11-10 11:50:21 +0100566 if (!t->discard_alignment)
567 t->discard_alignment =
568 b->discard_alignment & (b->discard_granularity - 1);
569
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400570 /* Top device aligned on logical block boundary? */
571 if (t->alignment_offset & (t->logical_block_size - 1)) {
572 t->misaligned = 1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100573 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400574 }
575
Martin K. Petersen86b37282009-11-10 11:50:21 +0100576 /* Find lcm() of optimal I/O size and granularity */
577 t->io_opt = lcm(t->io_opt, b->io_opt);
578 t->discard_granularity = lcm(t->discard_granularity,
579 b->discard_granularity);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400580
581 /* Verify that optimal I/O size is a multiple of io_min */
582 if (t->io_min && t->io_opt % t->io_min)
Martin K. Petersen86b37282009-11-10 11:50:21 +0100583 ret = -1;
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400584
Martin K. Petersen86b37282009-11-10 11:50:21 +0100585 return ret;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400586}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200587EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400588
589/**
590 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200591 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400592 * @bdev: the underlying block device (bottom)
593 * @offset: offset to beginning of data within component device
594 *
595 * Description:
596 * Merges the limits for two queues. Returns 0 if alignment
597 * didn't change. Returns -1 if adding the bottom device caused
598 * misalignment.
599 */
600void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
601 sector_t offset)
602{
603 struct request_queue *t = disk->queue;
604 struct request_queue *b = bdev_get_queue(bdev);
605
606 offset += get_start_sect(bdev) << 9;
607
608 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
609 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
610
611 disk_name(disk, 0, top);
612 bdevname(bdev, bottom);
613
614 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
615 top, bottom);
616 }
617
618 if (!t->queue_lock)
619 WARN_ON_ONCE(1);
620 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
621 unsigned long flags;
622
623 spin_lock_irqsave(t->queue_lock, flags);
624 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
625 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
626 spin_unlock_irqrestore(t->queue_lock, flags);
627 }
628}
629EXPORT_SYMBOL(disk_stack_limits);
630
631/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100632 * blk_queue_dma_pad - set pad mask
633 * @q: the request queue for the device
634 * @mask: pad mask
635 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200636 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100637 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200638 * Appending pad buffer to a request modifies the last entry of a
639 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100640 **/
641void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
642{
643 q->dma_pad_mask = mask;
644}
645EXPORT_SYMBOL(blk_queue_dma_pad);
646
647/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200648 * blk_queue_update_dma_pad - update pad mask
649 * @q: the request queue for the device
650 * @mask: pad mask
651 *
652 * Update dma pad mask.
653 *
654 * Appending pad buffer to a request modifies the last entry of a
655 * scatter list such that it includes the pad buffer.
656 **/
657void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
658{
659 if (mask > q->dma_pad_mask)
660 q->dma_pad_mask = mask;
661}
662EXPORT_SYMBOL(blk_queue_update_dma_pad);
663
664/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100665 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100666 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100667 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100668 * @buf: physically contiguous buffer
669 * @size: size of the buffer in bytes
670 *
671 * Some devices have excess DMA problems and can't simply discard (or
672 * zero fill) the unwanted piece of the transfer. They have to have a
673 * real area of memory to transfer it into. The use case for this is
674 * ATAPI devices in DMA mode. If the packet command causes a transfer
675 * bigger than the transfer size some HBAs will lock up if there
676 * aren't DMA elements to contain the excess transfer. What this API
677 * does is adjust the queue so that the buf is always appended
678 * silently to the scatterlist.
679 *
680 * Note: This routine adjusts max_hw_segments to make room for
681 * appending the drain buffer. If you call
682 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
683 * calling this routine, you must set the limit to one fewer than your
684 * device can support otherwise there won't be room for the drain
685 * buffer.
686 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100687int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100688 dma_drain_needed_fn *dma_drain_needed,
689 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100690{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400691 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100692 return -EINVAL;
693 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400694 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
695 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100696 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100697 q->dma_drain_buffer = buf;
698 q->dma_drain_size = size;
699
700 return 0;
701}
Jens Axboe86db1e22008-01-29 14:53:40 +0100702EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
703
704/**
705 * blk_queue_segment_boundary - set boundary rules for segment merging
706 * @q: the request queue for the device
707 * @mask: the memory boundary mask
708 **/
709void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
710{
711 if (mask < PAGE_CACHE_SIZE - 1) {
712 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700713 printk(KERN_INFO "%s: set to minimum %lx\n",
714 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100715 }
716
Martin K. Petersen025146e2009-05-22 17:17:51 -0400717 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100718}
Jens Axboe86db1e22008-01-29 14:53:40 +0100719EXPORT_SYMBOL(blk_queue_segment_boundary);
720
721/**
722 * blk_queue_dma_alignment - set dma length and memory alignment
723 * @q: the request queue for the device
724 * @mask: alignment mask
725 *
726 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200727 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100728 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100729 *
730 **/
731void blk_queue_dma_alignment(struct request_queue *q, int mask)
732{
733 q->dma_alignment = mask;
734}
Jens Axboe86db1e22008-01-29 14:53:40 +0100735EXPORT_SYMBOL(blk_queue_dma_alignment);
736
737/**
738 * blk_queue_update_dma_alignment - update dma length and memory alignment
739 * @q: the request queue for the device
740 * @mask: alignment mask
741 *
742 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200743 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100744 * If the requested alignment is larger than the current alignment, then
745 * the current queue alignment is updated to the new value, otherwise it
746 * is left alone. The design of this is to allow multiple objects
747 * (driver, device, transport etc) to set their respective
748 * alignments without having them interfere.
749 *
750 **/
751void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
752{
753 BUG_ON(mask > PAGE_SIZE);
754
755 if (mask > q->dma_alignment)
756 q->dma_alignment = mask;
757}
Jens Axboe86db1e22008-01-29 14:53:40 +0100758EXPORT_SYMBOL(blk_queue_update_dma_alignment);
759
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200760static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100761{
762 blk_max_low_pfn = max_low_pfn - 1;
763 blk_max_pfn = max_pfn - 1;
764 return 0;
765}
766subsys_initcall(blk_settings_init);