blob: 41392fbe19ff9615253278fdf48e63369ad7e856 [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 */
10
11#include "blk.h"
12
Jens Axboe6728cb02008-01-31 13:03:55 +010013unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010014EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010015
16unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010017
18/**
19 * blk_queue_prep_rq - set a prepare_request function for queue
20 * @q: queue
21 * @pfn: prepare_request function
22 *
23 * It's possible for a queue to register a prepare_request callback which
24 * is invoked before the request is handed to the request_fn. The goal of
25 * the function is to prepare a request for I/O, it can be used to build a
26 * cdb from the request data for instance.
27 *
28 */
29void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
30{
31 q->prep_rq_fn = pfn;
32}
Jens Axboe86db1e22008-01-29 14:53:40 +010033EXPORT_SYMBOL(blk_queue_prep_rq);
34
35/**
David Woodhousefb2dce82008-08-05 18:01:53 +010036 * blk_queue_set_discard - set a discard_sectors function for queue
37 * @q: queue
38 * @dfn: prepare_discard function
39 *
40 * It's possible for a queue to register a discard callback which is used
41 * to transform a discard request into the appropriate type for the
42 * hardware. If none is registered, then discard requests are failed
43 * with %EOPNOTSUPP.
44 *
45 */
46void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
47{
48 q->prepare_discard_fn = dfn;
49}
50EXPORT_SYMBOL(blk_queue_set_discard);
51
52/**
Jens Axboe86db1e22008-01-29 14:53:40 +010053 * blk_queue_merge_bvec - set a merge_bvec function for queue
54 * @q: queue
55 * @mbfn: merge_bvec_fn
56 *
57 * Usually queues have static limitations on the max sectors or segments that
58 * we can put in a request. Stacking drivers may have some settings that
59 * are dynamic, and thus we have to query the queue whether it is ok to
60 * add a new bio_vec to a bio at a given offset or not. If the block device
61 * has such limitations, it needs to register a merge_bvec_fn to control
62 * the size of bio's sent to it. Note that a block device *must* allow a
63 * single page to be added to an empty bio. The block device driver may want
64 * to use the bio_split() function to deal with these bio's. By default
65 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
66 * honored.
67 */
68void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
69{
70 q->merge_bvec_fn = mbfn;
71}
Jens Axboe86db1e22008-01-29 14:53:40 +010072EXPORT_SYMBOL(blk_queue_merge_bvec);
73
74void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
75{
76 q->softirq_done_fn = fn;
77}
Jens Axboe86db1e22008-01-29 14:53:40 +010078EXPORT_SYMBOL(blk_queue_softirq_done);
79
Jens Axboe242f9dc2008-09-14 05:55:09 -070080void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
81{
82 q->rq_timeout = timeout;
83}
84EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
85
86void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
87{
88 q->rq_timed_out_fn = fn;
89}
90EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
91
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020092void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
93{
94 q->lld_busy_fn = fn;
95}
96EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
97
Jens Axboe86db1e22008-01-29 14:53:40 +010098/**
99 * blk_queue_make_request - define an alternate make_request function for a device
100 * @q: the request queue for the device to be affected
101 * @mfn: the alternate make_request function
102 *
103 * Description:
104 * The normal way for &struct bios to be passed to a device
105 * driver is for them to be collected into requests on a request
106 * queue, and then to allow the device driver to select requests
107 * off that queue when it is ready. This works well for many block
108 * devices. However some block devices (typically virtual devices
109 * such as md or lvm) do not benefit from the processing on the
110 * request queue, and are served best by having the requests passed
111 * directly to them. This can be achieved by providing a function
112 * to blk_queue_make_request().
113 *
114 * Caveat:
115 * The driver that does this *must* be able to deal appropriately
116 * with buffers in "highmemory". This can be accomplished by either calling
117 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
118 * blk_queue_bounce() to create a buffer in normal memory.
119 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100120void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100121{
122 /*
123 * set defaults
124 */
125 q->nr_requests = BLKDEV_MAX_RQ;
126 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
127 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
128 q->make_request_fn = mfn;
Jens Axboe6728cb02008-01-31 13:03:55 +0100129 q->backing_dev_info.ra_pages =
130 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
Jens Axboe86db1e22008-01-29 14:53:40 +0100131 q->backing_dev_info.state = 0;
132 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
133 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
134 blk_queue_hardsect_size(q, 512);
135 blk_queue_dma_alignment(q, 511);
136 blk_queue_congestion_threshold(q);
137 q->nr_batching = BLK_BATCH_REQ;
138
139 q->unplug_thresh = 4; /* hmm */
140 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
141 if (q->unplug_delay == 0)
142 q->unplug_delay = 1;
143
Jens Axboe86db1e22008-01-29 14:53:40 +0100144 q->unplug_timer.function = blk_unplug_timeout;
145 q->unplug_timer.data = (unsigned long)q;
146
147 /*
148 * by default assume old behaviour and bounce for any highmem page
149 */
150 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
151}
Jens Axboe86db1e22008-01-29 14:53:40 +0100152EXPORT_SYMBOL(blk_queue_make_request);
153
154/**
155 * blk_queue_bounce_limit - set bounce buffer limit for queue
156 * @q: the request queue for the device
157 * @dma_addr: bus address limit
158 *
159 * Description:
160 * Different hardware can have different requirements as to what pages
161 * it can do I/O directly to. A low level driver can call
162 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Randy Dunlap710027a2008-08-19 20:13:11 +0200163 * buffers for doing I/O to pages residing above @dma_addr.
Jens Axboe86db1e22008-01-29 14:53:40 +0100164 **/
165void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
166{
Jens Axboe6728cb02008-01-31 13:03:55 +0100167 unsigned long b_pfn = dma_addr >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100168 int dma = 0;
169
170 q->bounce_gfp = GFP_NOIO;
171#if BITS_PER_LONG == 64
172 /* Assume anything <= 4GB can be handled by IOMMU.
173 Actually some IOMMUs can handle everything, but I don't
174 know of a way to test this here. */
Andrea Arcangeli00d61e32008-04-02 09:06:44 +0200175 if (b_pfn < (min_t(u64, 0x100000000UL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100176 dma = 1;
177 q->bounce_pfn = max_low_pfn;
178#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100179 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100180 dma = 1;
Jens Axboe6728cb02008-01-31 13:03:55 +0100181 q->bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100182#endif
183 if (dma) {
184 init_emergency_isa_pool();
185 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Jens Axboe6728cb02008-01-31 13:03:55 +0100186 q->bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100187 }
188}
Jens Axboe86db1e22008-01-29 14:53:40 +0100189EXPORT_SYMBOL(blk_queue_bounce_limit);
190
191/**
192 * blk_queue_max_sectors - set max sectors for a request for this queue
193 * @q: the request queue for the device
194 * @max_sectors: max sectors in the usual 512b unit
195 *
196 * Description:
197 * Enables a low level driver to set an upper limit on the size of
198 * received requests.
199 **/
200void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
201{
202 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
203 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700204 printk(KERN_INFO "%s: set to minimum %d\n",
205 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100206 }
207
208 if (BLK_DEF_MAX_SECTORS > max_sectors)
209 q->max_hw_sectors = q->max_sectors = max_sectors;
210 else {
211 q->max_sectors = BLK_DEF_MAX_SECTORS;
212 q->max_hw_sectors = max_sectors;
213 }
214}
Jens Axboe86db1e22008-01-29 14:53:40 +0100215EXPORT_SYMBOL(blk_queue_max_sectors);
216
217/**
218 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
219 * @q: the request queue for the device
220 * @max_segments: max number of segments
221 *
222 * Description:
223 * Enables a low level driver to set an upper limit on the number of
224 * physical data segments in a request. This would be the largest sized
225 * scatter list the driver could handle.
226 **/
227void blk_queue_max_phys_segments(struct request_queue *q,
228 unsigned short max_segments)
229{
230 if (!max_segments) {
231 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700232 printk(KERN_INFO "%s: set to minimum %d\n",
233 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100234 }
235
236 q->max_phys_segments = max_segments;
237}
Jens Axboe86db1e22008-01-29 14:53:40 +0100238EXPORT_SYMBOL(blk_queue_max_phys_segments);
239
240/**
241 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
242 * @q: the request queue for the device
243 * @max_segments: max number of segments
244 *
245 * Description:
246 * Enables a low level driver to set an upper limit on the number of
247 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200248 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100249 * to the device.
250 **/
251void blk_queue_max_hw_segments(struct request_queue *q,
252 unsigned short max_segments)
253{
254 if (!max_segments) {
255 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700256 printk(KERN_INFO "%s: set to minimum %d\n",
257 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100258 }
259
260 q->max_hw_segments = max_segments;
261}
Jens Axboe86db1e22008-01-29 14:53:40 +0100262EXPORT_SYMBOL(blk_queue_max_hw_segments);
263
264/**
265 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
266 * @q: the request queue for the device
267 * @max_size: max size of segment in bytes
268 *
269 * Description:
270 * Enables a low level driver to set an upper limit on the size of a
271 * coalesced segment
272 **/
273void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
274{
275 if (max_size < PAGE_CACHE_SIZE) {
276 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700277 printk(KERN_INFO "%s: set to minimum %d\n",
278 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100279 }
280
281 q->max_segment_size = max_size;
282}
Jens Axboe86db1e22008-01-29 14:53:40 +0100283EXPORT_SYMBOL(blk_queue_max_segment_size);
284
285/**
286 * blk_queue_hardsect_size - set hardware sector size for the queue
287 * @q: the request queue for the device
288 * @size: the hardware sector size, in bytes
289 *
290 * Description:
291 * This should typically be set to the lowest possible sector size
292 * that the hardware can operate on (possible without reverting to
293 * even internal read-modify-write operations). Usually the default
294 * of 512 covers most hardware.
295 **/
296void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
297{
298 q->hardsect_size = size;
299}
Jens Axboe86db1e22008-01-29 14:53:40 +0100300EXPORT_SYMBOL(blk_queue_hardsect_size);
301
302/*
303 * Returns the minimum that is _not_ zero, unless both are zero.
304 */
305#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
306
307/**
308 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
309 * @t: the stacking driver (top)
310 * @b: the underlying device (bottom)
311 **/
312void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
313{
314 /* zero is "infinity" */
Jens Axboe6728cb02008-01-31 13:03:55 +0100315 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
316 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100317
Jens Axboe6728cb02008-01-31 13:03:55 +0100318 t->max_phys_segments = min(t->max_phys_segments, b->max_phys_segments);
319 t->max_hw_segments = min(t->max_hw_segments, b->max_hw_segments);
320 t->max_segment_size = min(t->max_segment_size, b->max_segment_size);
321 t->hardsect_size = max(t->hardsect_size, b->hardsect_size);
Neil Browne7e72bf2008-05-14 16:05:54 -0700322 if (!t->queue_lock)
323 WARN_ON_ONCE(1);
324 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
325 unsigned long flags;
326 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200327 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700328 spin_unlock_irqrestore(t->queue_lock, flags);
329 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100330}
Jens Axboe86db1e22008-01-29 14:53:40 +0100331EXPORT_SYMBOL(blk_queue_stack_limits);
332
333/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100334 * blk_queue_dma_pad - set pad mask
335 * @q: the request queue for the device
336 * @mask: pad mask
337 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200338 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100339 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200340 * Appending pad buffer to a request modifies the last entry of a
341 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100342 **/
343void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
344{
345 q->dma_pad_mask = mask;
346}
347EXPORT_SYMBOL(blk_queue_dma_pad);
348
349/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200350 * blk_queue_update_dma_pad - update pad mask
351 * @q: the request queue for the device
352 * @mask: pad mask
353 *
354 * Update dma pad mask.
355 *
356 * Appending pad buffer to a request modifies the last entry of a
357 * scatter list such that it includes the pad buffer.
358 **/
359void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
360{
361 if (mask > q->dma_pad_mask)
362 q->dma_pad_mask = mask;
363}
364EXPORT_SYMBOL(blk_queue_update_dma_pad);
365
366/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100367 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100368 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100369 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100370 * @buf: physically contiguous buffer
371 * @size: size of the buffer in bytes
372 *
373 * Some devices have excess DMA problems and can't simply discard (or
374 * zero fill) the unwanted piece of the transfer. They have to have a
375 * real area of memory to transfer it into. The use case for this is
376 * ATAPI devices in DMA mode. If the packet command causes a transfer
377 * bigger than the transfer size some HBAs will lock up if there
378 * aren't DMA elements to contain the excess transfer. What this API
379 * does is adjust the queue so that the buf is always appended
380 * silently to the scatterlist.
381 *
382 * Note: This routine adjusts max_hw_segments to make room for
383 * appending the drain buffer. If you call
384 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
385 * calling this routine, you must set the limit to one fewer than your
386 * device can support otherwise there won't be room for the drain
387 * buffer.
388 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100389int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100390 dma_drain_needed_fn *dma_drain_needed,
391 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100392{
393 if (q->max_hw_segments < 2 || q->max_phys_segments < 2)
394 return -EINVAL;
395 /* make room for appending the drain */
396 --q->max_hw_segments;
397 --q->max_phys_segments;
Tejun Heo2fb98e82008-02-19 11:36:53 +0100398 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100399 q->dma_drain_buffer = buf;
400 q->dma_drain_size = size;
401
402 return 0;
403}
Jens Axboe86db1e22008-01-29 14:53:40 +0100404EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
405
406/**
407 * blk_queue_segment_boundary - set boundary rules for segment merging
408 * @q: the request queue for the device
409 * @mask: the memory boundary mask
410 **/
411void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
412{
413 if (mask < PAGE_CACHE_SIZE - 1) {
414 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700415 printk(KERN_INFO "%s: set to minimum %lx\n",
416 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100417 }
418
419 q->seg_boundary_mask = mask;
420}
Jens Axboe86db1e22008-01-29 14:53:40 +0100421EXPORT_SYMBOL(blk_queue_segment_boundary);
422
423/**
424 * blk_queue_dma_alignment - set dma length and memory alignment
425 * @q: the request queue for the device
426 * @mask: alignment mask
427 *
428 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200429 * set required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100430 * this is used when buiding direct io requests for the queue.
431 *
432 **/
433void blk_queue_dma_alignment(struct request_queue *q, int mask)
434{
435 q->dma_alignment = mask;
436}
Jens Axboe86db1e22008-01-29 14:53:40 +0100437EXPORT_SYMBOL(blk_queue_dma_alignment);
438
439/**
440 * blk_queue_update_dma_alignment - update dma length and memory alignment
441 * @q: the request queue for the device
442 * @mask: alignment mask
443 *
444 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200445 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100446 * If the requested alignment is larger than the current alignment, then
447 * the current queue alignment is updated to the new value, otherwise it
448 * is left alone. The design of this is to allow multiple objects
449 * (driver, device, transport etc) to set their respective
450 * alignments without having them interfere.
451 *
452 **/
453void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
454{
455 BUG_ON(mask > PAGE_SIZE);
456
457 if (mask > q->dma_alignment)
458 q->dma_alignment = mask;
459}
Jens Axboe86db1e22008-01-29 14:53:40 +0100460EXPORT_SYMBOL(blk_queue_update_dma_alignment);
461
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200462static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100463{
464 blk_max_low_pfn = max_low_pfn - 1;
465 blk_max_pfn = max_pfn - 1;
466 return 0;
467}
468subsys_initcall(blk_settings_init);