blob: 539d873c820dc3781314a6161d8664d438a7ff8c [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
80/**
81 * blk_queue_make_request - define an alternate make_request function for a device
82 * @q: the request queue for the device to be affected
83 * @mfn: the alternate make_request function
84 *
85 * Description:
86 * The normal way for &struct bios to be passed to a device
87 * driver is for them to be collected into requests on a request
88 * queue, and then to allow the device driver to select requests
89 * off that queue when it is ready. This works well for many block
90 * devices. However some block devices (typically virtual devices
91 * such as md or lvm) do not benefit from the processing on the
92 * request queue, and are served best by having the requests passed
93 * directly to them. This can be achieved by providing a function
94 * to blk_queue_make_request().
95 *
96 * Caveat:
97 * The driver that does this *must* be able to deal appropriately
98 * with buffers in "highmemory". This can be accomplished by either calling
99 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
100 * blk_queue_bounce() to create a buffer in normal memory.
101 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100102void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100103{
104 /*
105 * set defaults
106 */
107 q->nr_requests = BLKDEV_MAX_RQ;
108 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
109 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
110 q->make_request_fn = mfn;
Jens Axboe6728cb02008-01-31 13:03:55 +0100111 q->backing_dev_info.ra_pages =
112 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
Jens Axboe86db1e22008-01-29 14:53:40 +0100113 q->backing_dev_info.state = 0;
114 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
115 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
116 blk_queue_hardsect_size(q, 512);
117 blk_queue_dma_alignment(q, 511);
118 blk_queue_congestion_threshold(q);
119 q->nr_batching = BLK_BATCH_REQ;
120
121 q->unplug_thresh = 4; /* hmm */
122 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
123 if (q->unplug_delay == 0)
124 q->unplug_delay = 1;
125
126 INIT_WORK(&q->unplug_work, blk_unplug_work);
127
128 q->unplug_timer.function = blk_unplug_timeout;
129 q->unplug_timer.data = (unsigned long)q;
130
131 /*
132 * by default assume old behaviour and bounce for any highmem page
133 */
134 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
135}
Jens Axboe86db1e22008-01-29 14:53:40 +0100136EXPORT_SYMBOL(blk_queue_make_request);
137
138/**
139 * blk_queue_bounce_limit - set bounce buffer limit for queue
140 * @q: the request queue for the device
141 * @dma_addr: bus address limit
142 *
143 * Description:
144 * Different hardware can have different requirements as to what pages
145 * it can do I/O directly to. A low level driver can call
146 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
147 * buffers for doing I/O to pages residing above @page.
148 **/
149void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
150{
Jens Axboe6728cb02008-01-31 13:03:55 +0100151 unsigned long b_pfn = dma_addr >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100152 int dma = 0;
153
154 q->bounce_gfp = GFP_NOIO;
155#if BITS_PER_LONG == 64
156 /* Assume anything <= 4GB can be handled by IOMMU.
157 Actually some IOMMUs can handle everything, but I don't
158 know of a way to test this here. */
Andrea Arcangeli00d61e32008-04-02 09:06:44 +0200159 if (b_pfn < (min_t(u64, 0x100000000UL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100160 dma = 1;
161 q->bounce_pfn = max_low_pfn;
162#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100163 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100164 dma = 1;
Jens Axboe6728cb02008-01-31 13:03:55 +0100165 q->bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100166#endif
167 if (dma) {
168 init_emergency_isa_pool();
169 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Jens Axboe6728cb02008-01-31 13:03:55 +0100170 q->bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100171 }
172}
Jens Axboe86db1e22008-01-29 14:53:40 +0100173EXPORT_SYMBOL(blk_queue_bounce_limit);
174
175/**
176 * blk_queue_max_sectors - set max sectors for a request for this queue
177 * @q: the request queue for the device
178 * @max_sectors: max sectors in the usual 512b unit
179 *
180 * Description:
181 * Enables a low level driver to set an upper limit on the size of
182 * received requests.
183 **/
184void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
185{
186 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
187 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700188 printk(KERN_INFO "%s: set to minimum %d\n",
189 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100190 }
191
192 if (BLK_DEF_MAX_SECTORS > max_sectors)
193 q->max_hw_sectors = q->max_sectors = max_sectors;
194 else {
195 q->max_sectors = BLK_DEF_MAX_SECTORS;
196 q->max_hw_sectors = max_sectors;
197 }
198}
Jens Axboe86db1e22008-01-29 14:53:40 +0100199EXPORT_SYMBOL(blk_queue_max_sectors);
200
201/**
202 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
203 * @q: the request queue for the device
204 * @max_segments: max number of segments
205 *
206 * Description:
207 * Enables a low level driver to set an upper limit on the number of
208 * physical data segments in a request. This would be the largest sized
209 * scatter list the driver could handle.
210 **/
211void blk_queue_max_phys_segments(struct request_queue *q,
212 unsigned short max_segments)
213{
214 if (!max_segments) {
215 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700216 printk(KERN_INFO "%s: set to minimum %d\n",
217 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100218 }
219
220 q->max_phys_segments = max_segments;
221}
Jens Axboe86db1e22008-01-29 14:53:40 +0100222EXPORT_SYMBOL(blk_queue_max_phys_segments);
223
224/**
225 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
226 * @q: the request queue for the device
227 * @max_segments: max number of segments
228 *
229 * Description:
230 * Enables a low level driver to set an upper limit on the number of
231 * hw data segments in a request. This would be the largest number of
232 * address/length pairs the host adapter can actually give as once
233 * to the device.
234 **/
235void blk_queue_max_hw_segments(struct request_queue *q,
236 unsigned short max_segments)
237{
238 if (!max_segments) {
239 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700240 printk(KERN_INFO "%s: set to minimum %d\n",
241 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100242 }
243
244 q->max_hw_segments = max_segments;
245}
Jens Axboe86db1e22008-01-29 14:53:40 +0100246EXPORT_SYMBOL(blk_queue_max_hw_segments);
247
248/**
249 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
250 * @q: the request queue for the device
251 * @max_size: max size of segment in bytes
252 *
253 * Description:
254 * Enables a low level driver to set an upper limit on the size of a
255 * coalesced segment
256 **/
257void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
258{
259 if (max_size < PAGE_CACHE_SIZE) {
260 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700261 printk(KERN_INFO "%s: set to minimum %d\n",
262 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100263 }
264
265 q->max_segment_size = max_size;
266}
Jens Axboe86db1e22008-01-29 14:53:40 +0100267EXPORT_SYMBOL(blk_queue_max_segment_size);
268
269/**
270 * blk_queue_hardsect_size - set hardware sector size for the queue
271 * @q: the request queue for the device
272 * @size: the hardware sector size, in bytes
273 *
274 * Description:
275 * This should typically be set to the lowest possible sector size
276 * that the hardware can operate on (possible without reverting to
277 * even internal read-modify-write operations). Usually the default
278 * of 512 covers most hardware.
279 **/
280void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
281{
282 q->hardsect_size = size;
283}
Jens Axboe86db1e22008-01-29 14:53:40 +0100284EXPORT_SYMBOL(blk_queue_hardsect_size);
285
286/*
287 * Returns the minimum that is _not_ zero, unless both are zero.
288 */
289#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
290
291/**
292 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
293 * @t: the stacking driver (top)
294 * @b: the underlying device (bottom)
295 **/
296void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
297{
298 /* zero is "infinity" */
Jens Axboe6728cb02008-01-31 13:03:55 +0100299 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
300 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100301
Jens Axboe6728cb02008-01-31 13:03:55 +0100302 t->max_phys_segments = min(t->max_phys_segments, b->max_phys_segments);
303 t->max_hw_segments = min(t->max_hw_segments, b->max_hw_segments);
304 t->max_segment_size = min(t->max_segment_size, b->max_segment_size);
305 t->hardsect_size = max(t->hardsect_size, b->hardsect_size);
Neil Browne7e72bf2008-05-14 16:05:54 -0700306 if (!t->queue_lock)
307 WARN_ON_ONCE(1);
308 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
309 unsigned long flags;
310 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200311 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700312 spin_unlock_irqrestore(t->queue_lock, flags);
313 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100314}
Jens Axboe86db1e22008-01-29 14:53:40 +0100315EXPORT_SYMBOL(blk_queue_stack_limits);
316
317/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100318 * blk_queue_dma_pad - set pad mask
319 * @q: the request queue for the device
320 * @mask: pad mask
321 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200322 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100323 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200324 * Appending pad buffer to a request modifies the last entry of a
325 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100326 **/
327void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
328{
329 q->dma_pad_mask = mask;
330}
331EXPORT_SYMBOL(blk_queue_dma_pad);
332
333/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200334 * blk_queue_update_dma_pad - update pad mask
335 * @q: the request queue for the device
336 * @mask: pad mask
337 *
338 * Update dma pad mask.
339 *
340 * Appending pad buffer to a request modifies the last entry of a
341 * scatter list such that it includes the pad buffer.
342 **/
343void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
344{
345 if (mask > q->dma_pad_mask)
346 q->dma_pad_mask = mask;
347}
348EXPORT_SYMBOL(blk_queue_update_dma_pad);
349
350/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100351 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100352 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100353 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100354 * @buf: physically contiguous buffer
355 * @size: size of the buffer in bytes
356 *
357 * Some devices have excess DMA problems and can't simply discard (or
358 * zero fill) the unwanted piece of the transfer. They have to have a
359 * real area of memory to transfer it into. The use case for this is
360 * ATAPI devices in DMA mode. If the packet command causes a transfer
361 * bigger than the transfer size some HBAs will lock up if there
362 * aren't DMA elements to contain the excess transfer. What this API
363 * does is adjust the queue so that the buf is always appended
364 * silently to the scatterlist.
365 *
366 * Note: This routine adjusts max_hw_segments to make room for
367 * appending the drain buffer. If you call
368 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
369 * calling this routine, you must set the limit to one fewer than your
370 * device can support otherwise there won't be room for the drain
371 * buffer.
372 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100373int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100374 dma_drain_needed_fn *dma_drain_needed,
375 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100376{
377 if (q->max_hw_segments < 2 || q->max_phys_segments < 2)
378 return -EINVAL;
379 /* make room for appending the drain */
380 --q->max_hw_segments;
381 --q->max_phys_segments;
Tejun Heo2fb98e82008-02-19 11:36:53 +0100382 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100383 q->dma_drain_buffer = buf;
384 q->dma_drain_size = size;
385
386 return 0;
387}
Jens Axboe86db1e22008-01-29 14:53:40 +0100388EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
389
390/**
391 * blk_queue_segment_boundary - set boundary rules for segment merging
392 * @q: the request queue for the device
393 * @mask: the memory boundary mask
394 **/
395void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
396{
397 if (mask < PAGE_CACHE_SIZE - 1) {
398 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700399 printk(KERN_INFO "%s: set to minimum %lx\n",
400 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100401 }
402
403 q->seg_boundary_mask = mask;
404}
Jens Axboe86db1e22008-01-29 14:53:40 +0100405EXPORT_SYMBOL(blk_queue_segment_boundary);
406
407/**
408 * blk_queue_dma_alignment - set dma length and memory alignment
409 * @q: the request queue for the device
410 * @mask: alignment mask
411 *
412 * description:
413 * set required memory and length aligment for direct dma transactions.
414 * this is used when buiding direct io requests for the queue.
415 *
416 **/
417void blk_queue_dma_alignment(struct request_queue *q, int mask)
418{
419 q->dma_alignment = mask;
420}
Jens Axboe86db1e22008-01-29 14:53:40 +0100421EXPORT_SYMBOL(blk_queue_dma_alignment);
422
423/**
424 * blk_queue_update_dma_alignment - update dma length and memory alignment
425 * @q: the request queue for the device
426 * @mask: alignment mask
427 *
428 * description:
429 * update required memory and length aligment for direct dma transactions.
430 * If the requested alignment is larger than the current alignment, then
431 * the current queue alignment is updated to the new value, otherwise it
432 * is left alone. The design of this is to allow multiple objects
433 * (driver, device, transport etc) to set their respective
434 * alignments without having them interfere.
435 *
436 **/
437void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
438{
439 BUG_ON(mask > PAGE_SIZE);
440
441 if (mask > q->dma_alignment)
442 q->dma_alignment = mask;
443}
Jens Axboe86db1e22008-01-29 14:53:40 +0100444EXPORT_SYMBOL(blk_queue_update_dma_alignment);
445
Adrian Bunk52ff4ca2008-02-18 13:45:55 +0100446static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100447{
448 blk_max_low_pfn = max_low_pfn - 1;
449 blk_max_pfn = max_pfn - 1;
450 return 0;
451}
452subsys_initcall(blk_settings_init);