blob: 7f986cafacd58c37d33c855f54b9aa164003fb82 [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>
Jens Axboe86db1e22008-01-29 14:53:40 +010011
12#include "blk.h"
13
Jens Axboe6728cb02008-01-31 13:03:55 +010014unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010015EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010016
17unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010018
19/**
20 * blk_queue_prep_rq - set a prepare_request function for queue
21 * @q: queue
22 * @pfn: prepare_request function
23 *
24 * It's possible for a queue to register a prepare_request callback which
25 * is invoked before the request is handed to the request_fn. The goal of
26 * the function is to prepare a request for I/O, it can be used to build a
27 * cdb from the request data for instance.
28 *
29 */
30void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
31{
32 q->prep_rq_fn = pfn;
33}
Jens Axboe86db1e22008-01-29 14:53:40 +010034EXPORT_SYMBOL(blk_queue_prep_rq);
35
36/**
37 * blk_queue_merge_bvec - set a merge_bvec function for queue
38 * @q: queue
39 * @mbfn: merge_bvec_fn
40 *
41 * Usually queues have static limitations on the max sectors or segments that
42 * we can put in a request. Stacking drivers may have some settings that
43 * are dynamic, and thus we have to query the queue whether it is ok to
44 * add a new bio_vec to a bio at a given offset or not. If the block device
45 * has such limitations, it needs to register a merge_bvec_fn to control
46 * the size of bio's sent to it. Note that a block device *must* allow a
47 * single page to be added to an empty bio. The block device driver may want
48 * to use the bio_split() function to deal with these bio's. By default
49 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
50 * honored.
51 */
52void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
53{
54 q->merge_bvec_fn = mbfn;
55}
Jens Axboe86db1e22008-01-29 14:53:40 +010056EXPORT_SYMBOL(blk_queue_merge_bvec);
57
58void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
59{
60 q->softirq_done_fn = fn;
61}
Jens Axboe86db1e22008-01-29 14:53:40 +010062EXPORT_SYMBOL(blk_queue_softirq_done);
63
Jens Axboe242f9dc2008-09-14 05:55:09 -070064void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
65{
66 q->rq_timeout = timeout;
67}
68EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
69
70void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
71{
72 q->rq_timed_out_fn = fn;
73}
74EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
75
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020076void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
77{
78 q->lld_busy_fn = fn;
79}
80EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
81
Jens Axboe86db1e22008-01-29 14:53:40 +010082/**
Martin K. Petersene475bba2009-06-16 08:23:52 +020083 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5ca2009-06-19 09:18:32 +020084 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +020085 *
86 * Description:
87 * Returns a queue_limit struct to its default state. Can be used by
88 * stacking drivers like DM that stage table swaps and reuse an
89 * existing device queue.
90 */
91void blk_set_default_limits(struct queue_limits *lim)
92{
93 lim->max_phys_segments = MAX_PHYS_SEGMENTS;
94 lim->max_hw_segments = MAX_HW_SEGMENTS;
95 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
96 lim->max_segment_size = MAX_SEGMENT_SIZE;
Martin K. Petersen5dee2472009-09-21 21:46:05 +020097 lim->max_sectors = BLK_DEF_MAX_SECTORS;
98 lim->max_hw_sectors = INT_MAX;
Martin K. Petersen86b37282009-11-10 11:50:21 +010099 lim->max_discard_sectors = 0;
100 lim->discard_granularity = 0;
101 lim->discard_alignment = 0;
102 lim->discard_misaligned = 0;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200103 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200104 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200105 lim->alignment_offset = 0;
106 lim->io_opt = 0;
107 lim->misaligned = 0;
108 lim->no_cluster = 0;
109}
110EXPORT_SYMBOL(blk_set_default_limits);
111
112/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100113 * blk_queue_make_request - define an alternate make_request function for a device
114 * @q: the request queue for the device to be affected
115 * @mfn: the alternate make_request function
116 *
117 * Description:
118 * The normal way for &struct bios to be passed to a device
119 * driver is for them to be collected into requests on a request
120 * queue, and then to allow the device driver to select requests
121 * off that queue when it is ready. This works well for many block
122 * devices. However some block devices (typically virtual devices
123 * such as md or lvm) do not benefit from the processing on the
124 * request queue, and are served best by having the requests passed
125 * directly to them. This can be achieved by providing a function
126 * to blk_queue_make_request().
127 *
128 * Caveat:
129 * The driver that does this *must* be able to deal appropriately
130 * with buffers in "highmemory". This can be accomplished by either calling
131 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
132 * blk_queue_bounce() to create a buffer in normal memory.
133 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100134void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100135{
136 /*
137 * set defaults
138 */
139 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100140
Jens Axboe86db1e22008-01-29 14:53:40 +0100141 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100142 blk_queue_dma_alignment(q, 511);
143 blk_queue_congestion_threshold(q);
144 q->nr_batching = BLK_BATCH_REQ;
145
146 q->unplug_thresh = 4; /* hmm */
147 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
148 if (q->unplug_delay == 0)
149 q->unplug_delay = 1;
150
Jens Axboe86db1e22008-01-29 14:53:40 +0100151 q->unplug_timer.function = blk_unplug_timeout;
152 q->unplug_timer.data = (unsigned long)q;
153
Martin K. Petersene475bba2009-06-16 08:23:52 +0200154 blk_set_default_limits(&q->limits);
Martin K. Petersen80ddf242009-09-18 22:54:37 +0200155 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200156
Jens Axboe86db1e22008-01-29 14:53:40 +0100157 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200158 * If the caller didn't supply a lock, fall back to our embedded
159 * per-queue locks
160 */
161 if (!q->queue_lock)
162 q->queue_lock = &q->__queue_lock;
163
164 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100165 * by default assume old behaviour and bounce for any highmem page
166 */
167 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
168}
Jens Axboe86db1e22008-01-29 14:53:40 +0100169EXPORT_SYMBOL(blk_queue_make_request);
170
171/**
172 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900173 * @q: the request queue for the device
174 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100175 *
176 * Description:
177 * Different hardware can have different requirements as to what pages
178 * it can do I/O directly to. A low level driver can call
179 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900180 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100181 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900182void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100183{
Tejun Heocd0aca22009-04-15 22:10:25 +0900184 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100185 int dma = 0;
186
187 q->bounce_gfp = GFP_NOIO;
188#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900189 /*
190 * Assume anything <= 4GB can be handled by IOMMU. Actually
191 * some IOMMUs can handle everything, but I don't know of a
192 * way to test this here.
193 */
194 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100195 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400196 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100197#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100198 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100199 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400200 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100201#endif
202 if (dma) {
203 init_emergency_isa_pool();
204 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400205 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100206 }
207}
Jens Axboe86db1e22008-01-29 14:53:40 +0100208EXPORT_SYMBOL(blk_queue_bounce_limit);
209
210/**
211 * blk_queue_max_sectors - set max sectors for a request for this queue
212 * @q: the request queue for the device
213 * @max_sectors: max sectors in the usual 512b unit
214 *
215 * Description:
216 * Enables a low level driver to set an upper limit on the size of
217 * received requests.
218 **/
219void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
220{
221 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
222 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700223 printk(KERN_INFO "%s: set to minimum %d\n",
224 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100225 }
226
227 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400228 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100229 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400230 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
231 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100232 }
233}
Jens Axboe86db1e22008-01-29 14:53:40 +0100234EXPORT_SYMBOL(blk_queue_max_sectors);
235
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400236void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
237{
238 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400239 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400240 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400241 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400242}
243EXPORT_SYMBOL(blk_queue_max_hw_sectors);
244
Jens Axboe86db1e22008-01-29 14:53:40 +0100245/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200246 * blk_queue_max_discard_sectors - set max sectors for a single discard
247 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200248 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200249 **/
250void blk_queue_max_discard_sectors(struct request_queue *q,
251 unsigned int max_discard_sectors)
252{
253 q->limits.max_discard_sectors = max_discard_sectors;
254}
255EXPORT_SYMBOL(blk_queue_max_discard_sectors);
256
257/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100258 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
259 * @q: the request queue for the device
260 * @max_segments: max number of segments
261 *
262 * Description:
263 * Enables a low level driver to set an upper limit on the number of
264 * physical data segments in a request. This would be the largest sized
265 * scatter list the driver could handle.
266 **/
267void blk_queue_max_phys_segments(struct request_queue *q,
268 unsigned short max_segments)
269{
270 if (!max_segments) {
271 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700272 printk(KERN_INFO "%s: set to minimum %d\n",
273 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100274 }
275
Martin K. Petersen025146e2009-05-22 17:17:51 -0400276 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100277}
Jens Axboe86db1e22008-01-29 14:53:40 +0100278EXPORT_SYMBOL(blk_queue_max_phys_segments);
279
280/**
281 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
282 * @q: the request queue for the device
283 * @max_segments: max number of segments
284 *
285 * Description:
286 * Enables a low level driver to set an upper limit on the number of
287 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200288 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100289 * to the device.
290 **/
291void blk_queue_max_hw_segments(struct request_queue *q,
292 unsigned short max_segments)
293{
294 if (!max_segments) {
295 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700296 printk(KERN_INFO "%s: set to minimum %d\n",
297 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100298 }
299
Martin K. Petersen025146e2009-05-22 17:17:51 -0400300 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100301}
Jens Axboe86db1e22008-01-29 14:53:40 +0100302EXPORT_SYMBOL(blk_queue_max_hw_segments);
303
304/**
305 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
306 * @q: the request queue for the device
307 * @max_size: max size of segment in bytes
308 *
309 * Description:
310 * Enables a low level driver to set an upper limit on the size of a
311 * coalesced segment
312 **/
313void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
314{
315 if (max_size < PAGE_CACHE_SIZE) {
316 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700317 printk(KERN_INFO "%s: set to minimum %d\n",
318 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100319 }
320
Martin K. Petersen025146e2009-05-22 17:17:51 -0400321 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100322}
Jens Axboe86db1e22008-01-29 14:53:40 +0100323EXPORT_SYMBOL(blk_queue_max_segment_size);
324
325/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400326 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100327 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400328 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100329 *
330 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400331 * This should be set to the lowest possible block size that the
332 * storage device can address. The default of 512 covers most
333 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100334 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400335void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100336{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400337 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400338
339 if (q->limits.physical_block_size < size)
340 q->limits.physical_block_size = size;
341
342 if (q->limits.io_min < q->limits.physical_block_size)
343 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100344}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400345EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100346
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400347/**
348 * blk_queue_physical_block_size - set physical block size for the queue
349 * @q: the request queue for the device
350 * @size: the physical block size, in bytes
351 *
352 * Description:
353 * This should be set to the lowest possible sector size that the
354 * hardware can operate on without reverting to read-modify-write
355 * operations.
356 */
357void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
358{
359 q->limits.physical_block_size = size;
360
361 if (q->limits.physical_block_size < q->limits.logical_block_size)
362 q->limits.physical_block_size = q->limits.logical_block_size;
363
364 if (q->limits.io_min < q->limits.physical_block_size)
365 q->limits.io_min = q->limits.physical_block_size;
366}
367EXPORT_SYMBOL(blk_queue_physical_block_size);
368
369/**
370 * blk_queue_alignment_offset - set physical block alignment offset
371 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700372 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400373 *
374 * Description:
375 * Some devices are naturally misaligned to compensate for things like
376 * the legacy DOS partition table 63-sector offset. Low-level drivers
377 * should call this function for devices whose first sector is not
378 * naturally aligned.
379 */
380void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
381{
382 q->limits.alignment_offset =
383 offset & (q->limits.physical_block_size - 1);
384 q->limits.misaligned = 0;
385}
386EXPORT_SYMBOL(blk_queue_alignment_offset);
387
388/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400389 * blk_limits_io_min - set minimum request size for a device
390 * @limits: the queue limits
391 * @min: smallest I/O size in bytes
392 *
393 * Description:
394 * Some devices have an internal block size bigger than the reported
395 * hardware sector size. This function can be used to signal the
396 * smallest I/O the device can perform without incurring a performance
397 * penalty.
398 */
399void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
400{
401 limits->io_min = min;
402
403 if (limits->io_min < limits->logical_block_size)
404 limits->io_min = limits->logical_block_size;
405
406 if (limits->io_min < limits->physical_block_size)
407 limits->io_min = limits->physical_block_size;
408}
409EXPORT_SYMBOL(blk_limits_io_min);
410
411/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400412 * blk_queue_io_min - set minimum request size for the queue
413 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700414 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400415 *
416 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400417 * Storage devices may report a granularity or preferred minimum I/O
418 * size which is the smallest request the device can perform without
419 * incurring a performance penalty. For disk drives this is often the
420 * physical block size. For RAID arrays it is often the stripe chunk
421 * size. A properly aligned multiple of minimum_io_size is the
422 * preferred request size for workloads where a high number of I/O
423 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400424 */
425void blk_queue_io_min(struct request_queue *q, unsigned int min)
426{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400427 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400428}
429EXPORT_SYMBOL(blk_queue_io_min);
430
431/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200432 * blk_limits_io_opt - set optimal request size for a device
433 * @limits: the queue limits
434 * @opt: smallest I/O size in bytes
435 *
436 * Description:
437 * Storage devices may report an optimal I/O size, which is the
438 * device's preferred unit for sustained I/O. This is rarely reported
439 * for disk drives. For RAID arrays it is usually the stripe width or
440 * the internal track size. A properly aligned multiple of
441 * optimal_io_size is the preferred request size for workloads where
442 * sustained throughput is desired.
443 */
444void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
445{
446 limits->io_opt = opt;
447}
448EXPORT_SYMBOL(blk_limits_io_opt);
449
450/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400451 * blk_queue_io_opt - set optimal request size for the queue
452 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700453 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400454 *
455 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400456 * Storage devices may report an optimal I/O size, which is the
457 * device's preferred unit for sustained I/O. This is rarely reported
458 * for disk drives. For RAID arrays it is usually the stripe width or
459 * the internal track size. A properly aligned multiple of
460 * optimal_io_size is the preferred request size for workloads where
461 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400462 */
463void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
464{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200465 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400466}
467EXPORT_SYMBOL(blk_queue_io_opt);
468
Jens Axboe86db1e22008-01-29 14:53:40 +0100469/*
470 * Returns the minimum that is _not_ zero, unless both are zero.
471 */
472#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
473
474/**
475 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
476 * @t: the stacking driver (top)
477 * @b: the underlying device (bottom)
478 **/
479void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
480{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400481 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400482
Neil Browne7e72bf2008-05-14 16:05:54 -0700483 if (!t->queue_lock)
484 WARN_ON_ONCE(1);
485 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
486 unsigned long flags;
487 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200488 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700489 spin_unlock_irqrestore(t->queue_lock, flags);
490 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100491}
Jens Axboe86db1e22008-01-29 14:53:40 +0100492EXPORT_SYMBOL(blk_queue_stack_limits);
493
Martin K. Petersen86b37282009-11-10 11:50:21 +0100494static unsigned int lcm(unsigned int a, unsigned int b)
495{
496 if (a && b)
497 return (a * b) / gcd(a, b);
498 else if (b)
499 return b;
500
501 return a;
502}
503
Jens Axboe86db1e22008-01-29 14:53:40 +0100504/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400505 * blk_stack_limits - adjust queue_limits for stacked devices
506 * @t: the stacking driver limits (top)
Martin K. Petersen77634f32009-06-09 06:23:22 +0200507 * @b: the underlying queue limits (bottom)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400508 * @offset: offset to beginning of data within component device
509 *
510 * Description:
511 * Merges two queue_limit structs. Returns 0 if alignment didn't
512 * change. Returns -1 if adding the bottom device caused
513 * misalignment.
514 */
515int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
516 sector_t offset)
517{
Martin K. Petersen86b37282009-11-10 11:50:21 +0100518 int ret;
519
520 ret = 0;
521
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400522 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
523 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200524 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400525
526 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
527 b->seg_boundary_mask);
528
529 t->max_phys_segments = min_not_zero(t->max_phys_segments,
530 b->max_phys_segments);
531
532 t->max_hw_segments = min_not_zero(t->max_hw_segments,
533 b->max_hw_segments);
534
535 t->max_segment_size = min_not_zero(t->max_segment_size,
536 b->max_segment_size);
537
538 t->logical_block_size = max(t->logical_block_size,
539 b->logical_block_size);
540
541 t->physical_block_size = max(t->physical_block_size,
542 b->physical_block_size);
543
544 t->io_min = max(t->io_min, b->io_min);
545 t->no_cluster |= b->no_cluster;
546
547 /* Bottom device offset aligned? */
548 if (offset &&
549 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
550 t->misaligned = 1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100551 ret = -1;
552 }
553
554 if (offset &&
555 (offset & (b->discard_granularity - 1)) != b->discard_alignment) {
556 t->discard_misaligned = 1;
557 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400558 }
559
560 /* If top has no alignment offset, inherit from bottom */
561 if (!t->alignment_offset)
562 t->alignment_offset =
563 b->alignment_offset & (b->physical_block_size - 1);
564
Martin K. Petersen86b37282009-11-10 11:50:21 +0100565 if (!t->discard_alignment)
566 t->discard_alignment =
567 b->discard_alignment & (b->discard_granularity - 1);
568
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400569 /* Top device aligned on logical block boundary? */
570 if (t->alignment_offset & (t->logical_block_size - 1)) {
571 t->misaligned = 1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100572 ret = -1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400573 }
574
Martin K. Petersen86b37282009-11-10 11:50:21 +0100575 /* Find lcm() of optimal I/O size and granularity */
576 t->io_opt = lcm(t->io_opt, b->io_opt);
577 t->discard_granularity = lcm(t->discard_granularity,
578 b->discard_granularity);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400579
580 /* Verify that optimal I/O size is a multiple of io_min */
581 if (t->io_min && t->io_opt % t->io_min)
Martin K. Petersen86b37282009-11-10 11:50:21 +0100582 ret = -1;
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400583
Martin K. Petersen86b37282009-11-10 11:50:21 +0100584 return ret;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400585}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200586EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400587
588/**
589 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200590 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400591 * @bdev: the underlying block device (bottom)
592 * @offset: offset to beginning of data within component device
593 *
594 * Description:
595 * Merges the limits for two queues. Returns 0 if alignment
596 * didn't change. Returns -1 if adding the bottom device caused
597 * misalignment.
598 */
599void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
600 sector_t offset)
601{
602 struct request_queue *t = disk->queue;
603 struct request_queue *b = bdev_get_queue(bdev);
604
605 offset += get_start_sect(bdev) << 9;
606
607 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
608 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
609
610 disk_name(disk, 0, top);
611 bdevname(bdev, bottom);
612
613 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
614 top, bottom);
615 }
616
617 if (!t->queue_lock)
618 WARN_ON_ONCE(1);
619 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
620 unsigned long flags;
621
622 spin_lock_irqsave(t->queue_lock, flags);
623 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
624 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
625 spin_unlock_irqrestore(t->queue_lock, flags);
626 }
627}
628EXPORT_SYMBOL(disk_stack_limits);
629
630/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100631 * blk_queue_dma_pad - set pad mask
632 * @q: the request queue for the device
633 * @mask: pad mask
634 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200635 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100636 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200637 * Appending pad buffer to a request modifies the last entry of a
638 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100639 **/
640void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
641{
642 q->dma_pad_mask = mask;
643}
644EXPORT_SYMBOL(blk_queue_dma_pad);
645
646/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200647 * blk_queue_update_dma_pad - update pad mask
648 * @q: the request queue for the device
649 * @mask: pad mask
650 *
651 * Update dma pad mask.
652 *
653 * Appending pad buffer to a request modifies the last entry of a
654 * scatter list such that it includes the pad buffer.
655 **/
656void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
657{
658 if (mask > q->dma_pad_mask)
659 q->dma_pad_mask = mask;
660}
661EXPORT_SYMBOL(blk_queue_update_dma_pad);
662
663/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100664 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100665 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100666 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100667 * @buf: physically contiguous buffer
668 * @size: size of the buffer in bytes
669 *
670 * Some devices have excess DMA problems and can't simply discard (or
671 * zero fill) the unwanted piece of the transfer. They have to have a
672 * real area of memory to transfer it into. The use case for this is
673 * ATAPI devices in DMA mode. If the packet command causes a transfer
674 * bigger than the transfer size some HBAs will lock up if there
675 * aren't DMA elements to contain the excess transfer. What this API
676 * does is adjust the queue so that the buf is always appended
677 * silently to the scatterlist.
678 *
679 * Note: This routine adjusts max_hw_segments to make room for
680 * appending the drain buffer. If you call
681 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
682 * calling this routine, you must set the limit to one fewer than your
683 * device can support otherwise there won't be room for the drain
684 * buffer.
685 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100686int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100687 dma_drain_needed_fn *dma_drain_needed,
688 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100689{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400690 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100691 return -EINVAL;
692 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400693 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
694 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100695 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100696 q->dma_drain_buffer = buf;
697 q->dma_drain_size = size;
698
699 return 0;
700}
Jens Axboe86db1e22008-01-29 14:53:40 +0100701EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
702
703/**
704 * blk_queue_segment_boundary - set boundary rules for segment merging
705 * @q: the request queue for the device
706 * @mask: the memory boundary mask
707 **/
708void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
709{
710 if (mask < PAGE_CACHE_SIZE - 1) {
711 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700712 printk(KERN_INFO "%s: set to minimum %lx\n",
713 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100714 }
715
Martin K. Petersen025146e2009-05-22 17:17:51 -0400716 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100717}
Jens Axboe86db1e22008-01-29 14:53:40 +0100718EXPORT_SYMBOL(blk_queue_segment_boundary);
719
720/**
721 * blk_queue_dma_alignment - set dma length and memory alignment
722 * @q: the request queue for the device
723 * @mask: alignment mask
724 *
725 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200726 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100727 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100728 *
729 **/
730void blk_queue_dma_alignment(struct request_queue *q, int mask)
731{
732 q->dma_alignment = mask;
733}
Jens Axboe86db1e22008-01-29 14:53:40 +0100734EXPORT_SYMBOL(blk_queue_dma_alignment);
735
736/**
737 * blk_queue_update_dma_alignment - update dma length and memory alignment
738 * @q: the request queue for the device
739 * @mask: alignment mask
740 *
741 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200742 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100743 * If the requested alignment is larger than the current alignment, then
744 * the current queue alignment is updated to the new value, otherwise it
745 * is left alone. The design of this is to allow multiple objects
746 * (driver, device, transport etc) to set their respective
747 * alignments without having them interfere.
748 *
749 **/
750void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
751{
752 BUG_ON(mask > PAGE_SIZE);
753
754 if (mask > q->dma_alignment)
755 q->dma_alignment = mask;
756}
Jens Axboe86db1e22008-01-29 14:53:40 +0100757EXPORT_SYMBOL(blk_queue_update_dma_alignment);
758
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200759static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100760{
761 blk_max_low_pfn = max_low_pfn - 1;
762 blk_max_pfn = max_pfn - 1;
763 return 0;
764}
765subsys_initcall(blk_settings_init);