]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - block/ll_rw_blk.c
[PATCH] cfq-iosched: use new io context counting mechanism
[mirror_ubuntu-bionic-kernel.git] / block / ll_rw_blk.c
CommitLineData
1da177e4 1/*
1da177e4
LT
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
8 */
9
10/*
11 * This handles all read/write requests to block devices
12 */
1da177e4
LT
13#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/bio.h>
17#include <linux/blkdev.h>
18#include <linux/highmem.h>
19#include <linux/mm.h>
20#include <linux/kernel_stat.h>
21#include <linux/string.h>
22#include <linux/init.h>
23#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
ff856bad
JA
28#include <linux/interrupt.h>
29#include <linux/cpu.h>
2056a782 30#include <linux/blktrace_api.h>
1da177e4
LT
31
32/*
33 * for max sense size
34 */
35#include <scsi/scsi_cmnd.h>
36
37static void blk_unplug_work(void *data);
38static void blk_unplug_timeout(unsigned long data);
93d17d3d 39static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
52d9e675
TH
40static void init_request_from_bio(struct request *req, struct bio *bio);
41static int __make_request(request_queue_t *q, struct bio *bio);
1da177e4
LT
42
43/*
44 * For the allocated request tables
45 */
46static kmem_cache_t *request_cachep;
47
48/*
49 * For queue allocation
50 */
51static kmem_cache_t *requestq_cachep;
52
53/*
54 * For io context allocations
55 */
56static kmem_cache_t *iocontext_cachep;
57
58static wait_queue_head_t congestion_wqh[2] = {
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
60 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
61 };
62
63/*
64 * Controlling structure to kblockd
65 */
ff856bad 66static struct workqueue_struct *kblockd_workqueue;
1da177e4
LT
67
68unsigned long blk_max_low_pfn, blk_max_pfn;
69
70EXPORT_SYMBOL(blk_max_low_pfn);
71EXPORT_SYMBOL(blk_max_pfn);
72
ff856bad
JA
73static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
74
1da177e4
LT
75/* Amount of time in which a process may batch requests */
76#define BLK_BATCH_TIME (HZ/50UL)
77
78/* Number of requests a "batching" process may submit */
79#define BLK_BATCH_REQ 32
80
81/*
82 * Return the threshold (number of used requests) at which the queue is
83 * considered to be congested. It include a little hysteresis to keep the
84 * context switch rate down.
85 */
86static inline int queue_congestion_on_threshold(struct request_queue *q)
87{
88 return q->nr_congestion_on;
89}
90
91/*
92 * The threshold at which a queue is considered to be uncongested
93 */
94static inline int queue_congestion_off_threshold(struct request_queue *q)
95{
96 return q->nr_congestion_off;
97}
98
99static void blk_queue_congestion_threshold(struct request_queue *q)
100{
101 int nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) + 1;
104 if (nr > q->nr_requests)
105 nr = q->nr_requests;
106 q->nr_congestion_on = nr;
107
108 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
109 if (nr < 1)
110 nr = 1;
111 q->nr_congestion_off = nr;
112}
113
114/*
115 * A queue has just exitted congestion. Note this in the global counter of
116 * congested queues, and wake up anyone who was waiting for requests to be
117 * put back.
118 */
119static void clear_queue_congested(request_queue_t *q, int rw)
120{
121 enum bdi_state bit;
122 wait_queue_head_t *wqh = &congestion_wqh[rw];
123
124 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
125 clear_bit(bit, &q->backing_dev_info.state);
126 smp_mb__after_clear_bit();
127 if (waitqueue_active(wqh))
128 wake_up(wqh);
129}
130
131/*
132 * A queue has just entered congestion. Flag that in the queue's VM-visible
133 * state flags and increment the global gounter of congested queues.
134 */
135static void set_queue_congested(request_queue_t *q, int rw)
136{
137 enum bdi_state bit;
138
139 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
140 set_bit(bit, &q->backing_dev_info.state);
141}
142
143/**
144 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
145 * @bdev: device
146 *
147 * Locates the passed device's request queue and returns the address of its
148 * backing_dev_info
149 *
150 * Will return NULL if the request queue cannot be located.
151 */
152struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
153{
154 struct backing_dev_info *ret = NULL;
155 request_queue_t *q = bdev_get_queue(bdev);
156
157 if (q)
158 ret = &q->backing_dev_info;
159 return ret;
160}
161
162EXPORT_SYMBOL(blk_get_backing_dev_info);
163
164void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
165{
166 q->activity_fn = fn;
167 q->activity_data = data;
168}
169
170EXPORT_SYMBOL(blk_queue_activity_fn);
171
172/**
173 * blk_queue_prep_rq - set a prepare_request function for queue
174 * @q: queue
175 * @pfn: prepare_request function
176 *
177 * It's possible for a queue to register a prepare_request callback which
178 * is invoked before the request is handed to the request_fn. The goal of
179 * the function is to prepare a request for I/O, it can be used to build a
180 * cdb from the request data for instance.
181 *
182 */
183void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
184{
185 q->prep_rq_fn = pfn;
186}
187
188EXPORT_SYMBOL(blk_queue_prep_rq);
189
190/**
191 * blk_queue_merge_bvec - set a merge_bvec function for queue
192 * @q: queue
193 * @mbfn: merge_bvec_fn
194 *
195 * Usually queues have static limitations on the max sectors or segments that
196 * we can put in a request. Stacking drivers may have some settings that
197 * are dynamic, and thus we have to query the queue whether it is ok to
198 * add a new bio_vec to a bio at a given offset or not. If the block device
199 * has such limitations, it needs to register a merge_bvec_fn to control
200 * the size of bio's sent to it. Note that a block device *must* allow a
201 * single page to be added to an empty bio. The block device driver may want
202 * to use the bio_split() function to deal with these bio's. By default
203 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
204 * honored.
205 */
206void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
207{
208 q->merge_bvec_fn = mbfn;
209}
210
211EXPORT_SYMBOL(blk_queue_merge_bvec);
212
ff856bad
JA
213void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
214{
215 q->softirq_done_fn = fn;
216}
217
218EXPORT_SYMBOL(blk_queue_softirq_done);
219
1da177e4
LT
220/**
221 * blk_queue_make_request - define an alternate make_request function for a device
222 * @q: the request queue for the device to be affected
223 * @mfn: the alternate make_request function
224 *
225 * Description:
226 * The normal way for &struct bios to be passed to a device
227 * driver is for them to be collected into requests on a request
228 * queue, and then to allow the device driver to select requests
229 * off that queue when it is ready. This works well for many block
230 * devices. However some block devices (typically virtual devices
231 * such as md or lvm) do not benefit from the processing on the
232 * request queue, and are served best by having the requests passed
233 * directly to them. This can be achieved by providing a function
234 * to blk_queue_make_request().
235 *
236 * Caveat:
237 * The driver that does this *must* be able to deal appropriately
238 * with buffers in "highmemory". This can be accomplished by either calling
239 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
240 * blk_queue_bounce() to create a buffer in normal memory.
241 **/
242void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
243{
244 /*
245 * set defaults
246 */
247 q->nr_requests = BLKDEV_MAX_RQ;
309c0a1d
SM
248 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
249 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1da177e4
LT
250 q->make_request_fn = mfn;
251 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
252 q->backing_dev_info.state = 0;
253 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
defd94b7 254 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
1da177e4
LT
255 blk_queue_hardsect_size(q, 512);
256 blk_queue_dma_alignment(q, 511);
257 blk_queue_congestion_threshold(q);
258 q->nr_batching = BLK_BATCH_REQ;
259
260 q->unplug_thresh = 4; /* hmm */
261 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
262 if (q->unplug_delay == 0)
263 q->unplug_delay = 1;
264
265 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
266
267 q->unplug_timer.function = blk_unplug_timeout;
268 q->unplug_timer.data = (unsigned long)q;
269
270 /*
271 * by default assume old behaviour and bounce for any highmem page
272 */
273 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
274
275 blk_queue_activity_fn(q, NULL, NULL);
1da177e4
LT
276}
277
278EXPORT_SYMBOL(blk_queue_make_request);
279
280static inline void rq_init(request_queue_t *q, struct request *rq)
281{
282 INIT_LIST_HEAD(&rq->queuelist);
ff856bad 283 INIT_LIST_HEAD(&rq->donelist);
1da177e4
LT
284
285 rq->errors = 0;
1da177e4 286 rq->bio = rq->biotail = NULL;
2e662b65
JA
287 INIT_HLIST_NODE(&rq->hash);
288 RB_CLEAR_NODE(&rq->rb_node);
22e2c507 289 rq->ioprio = 0;
1da177e4
LT
290 rq->buffer = NULL;
291 rq->ref_count = 1;
292 rq->q = q;
1da177e4
LT
293 rq->special = NULL;
294 rq->data_len = 0;
295 rq->data = NULL;
df46b9a4 296 rq->nr_phys_segments = 0;
1da177e4
LT
297 rq->sense = NULL;
298 rq->end_io = NULL;
299 rq->end_io_data = NULL;
ff856bad 300 rq->completion_data = NULL;
1da177e4
LT
301}
302
303/**
304 * blk_queue_ordered - does this queue support ordered writes
797e7dbb
TH
305 * @q: the request queue
306 * @ordered: one of QUEUE_ORDERED_*
fddfdeaf 307 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
1da177e4
LT
308 *
309 * Description:
310 * For journalled file systems, doing ordered writes on a commit
311 * block instead of explicitly doing wait_on_buffer (which is bad
312 * for performance) can be a big win. Block drivers supporting this
313 * feature should call this function and indicate so.
314 *
315 **/
797e7dbb
TH
316int blk_queue_ordered(request_queue_t *q, unsigned ordered,
317 prepare_flush_fn *prepare_flush_fn)
318{
319 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
320 prepare_flush_fn == NULL) {
321 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
322 return -EINVAL;
323 }
324
325 if (ordered != QUEUE_ORDERED_NONE &&
326 ordered != QUEUE_ORDERED_DRAIN &&
327 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
328 ordered != QUEUE_ORDERED_DRAIN_FUA &&
329 ordered != QUEUE_ORDERED_TAG &&
330 ordered != QUEUE_ORDERED_TAG_FLUSH &&
331 ordered != QUEUE_ORDERED_TAG_FUA) {
332 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
333 return -EINVAL;
1da177e4 334 }
797e7dbb 335
60481b12 336 q->ordered = ordered;
797e7dbb
TH
337 q->next_ordered = ordered;
338 q->prepare_flush_fn = prepare_flush_fn;
339
340 return 0;
1da177e4
LT
341}
342
343EXPORT_SYMBOL(blk_queue_ordered);
344
345/**
346 * blk_queue_issue_flush_fn - set function for issuing a flush
347 * @q: the request queue
348 * @iff: the function to be called issuing the flush
349 *
350 * Description:
351 * If a driver supports issuing a flush command, the support is notified
352 * to the block layer by defining it through this call.
353 *
354 **/
355void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
356{
357 q->issue_flush_fn = iff;
358}
359
360EXPORT_SYMBOL(blk_queue_issue_flush_fn);
361
362/*
363 * Cache flushing for ordered writes handling
364 */
797e7dbb 365inline unsigned blk_ordered_cur_seq(request_queue_t *q)
1da177e4 366{
797e7dbb
TH
367 if (!q->ordseq)
368 return 0;
369 return 1 << ffz(q->ordseq);
1da177e4
LT
370}
371
797e7dbb 372unsigned blk_ordered_req_seq(struct request *rq)
1da177e4 373{
1da177e4
LT
374 request_queue_t *q = rq->q;
375
797e7dbb 376 BUG_ON(q->ordseq == 0);
8922e16c 377
797e7dbb
TH
378 if (rq == &q->pre_flush_rq)
379 return QUEUE_ORDSEQ_PREFLUSH;
380 if (rq == &q->bar_rq)
381 return QUEUE_ORDSEQ_BAR;
382 if (rq == &q->post_flush_rq)
383 return QUEUE_ORDSEQ_POSTFLUSH;
1da177e4 384
4aff5e23
JA
385 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
386 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
797e7dbb
TH
387 return QUEUE_ORDSEQ_DRAIN;
388 else
389 return QUEUE_ORDSEQ_DONE;
1da177e4
LT
390}
391
797e7dbb 392void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
1da177e4 393{
797e7dbb
TH
394 struct request *rq;
395 int uptodate;
1da177e4 396
797e7dbb
TH
397 if (error && !q->orderr)
398 q->orderr = error;
1da177e4 399
797e7dbb
TH
400 BUG_ON(q->ordseq & seq);
401 q->ordseq |= seq;
1da177e4 402
797e7dbb
TH
403 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
404 return;
1da177e4
LT
405
406 /*
797e7dbb 407 * Okay, sequence complete.
1da177e4 408 */
797e7dbb
TH
409 rq = q->orig_bar_rq;
410 uptodate = q->orderr ? q->orderr : 1;
1da177e4 411
797e7dbb 412 q->ordseq = 0;
1da177e4 413
797e7dbb
TH
414 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
415 end_that_request_last(rq, uptodate);
1da177e4
LT
416}
417
797e7dbb 418static void pre_flush_end_io(struct request *rq, int error)
1da177e4 419{
797e7dbb
TH
420 elv_completed_request(rq->q, rq);
421 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
422}
1da177e4 423
797e7dbb
TH
424static void bar_end_io(struct request *rq, int error)
425{
426 elv_completed_request(rq->q, rq);
427 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
428}
1da177e4 429
797e7dbb
TH
430static void post_flush_end_io(struct request *rq, int error)
431{
432 elv_completed_request(rq->q, rq);
433 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
434}
1da177e4 435
797e7dbb
TH
436static void queue_flush(request_queue_t *q, unsigned which)
437{
438 struct request *rq;
439 rq_end_io_fn *end_io;
1da177e4 440
797e7dbb
TH
441 if (which == QUEUE_ORDERED_PREFLUSH) {
442 rq = &q->pre_flush_rq;
443 end_io = pre_flush_end_io;
444 } else {
445 rq = &q->post_flush_rq;
446 end_io = post_flush_end_io;
1da177e4 447 }
797e7dbb 448
4aff5e23 449 rq->cmd_flags = REQ_HARDBARRIER;
797e7dbb 450 rq_init(q, rq);
797e7dbb 451 rq->elevator_private = NULL;
c00895ab 452 rq->elevator_private2 = NULL;
797e7dbb 453 rq->rq_disk = q->bar_rq.rq_disk;
797e7dbb
TH
454 rq->end_io = end_io;
455 q->prepare_flush_fn(q, rq);
456
30e9656c 457 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
1da177e4
LT
458}
459
797e7dbb
TH
460static inline struct request *start_ordered(request_queue_t *q,
461 struct request *rq)
1da177e4 462{
797e7dbb
TH
463 q->bi_size = 0;
464 q->orderr = 0;
465 q->ordered = q->next_ordered;
466 q->ordseq |= QUEUE_ORDSEQ_STARTED;
467
468 /*
469 * Prep proxy barrier request.
470 */
471 blkdev_dequeue_request(rq);
472 q->orig_bar_rq = rq;
473 rq = &q->bar_rq;
4aff5e23 474 rq->cmd_flags = 0;
797e7dbb 475 rq_init(q, rq);
4aff5e23
JA
476 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
477 rq->cmd_flags |= REQ_RW;
478 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
797e7dbb 479 rq->elevator_private = NULL;
c00895ab 480 rq->elevator_private2 = NULL;
797e7dbb
TH
481 init_request_from_bio(rq, q->orig_bar_rq->bio);
482 rq->end_io = bar_end_io;
483
484 /*
485 * Queue ordered sequence. As we stack them at the head, we
486 * need to queue in reverse order. Note that we rely on that
487 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
488 * request gets inbetween ordered sequence.
489 */
490 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
491 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
492 else
493 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
494
30e9656c 495 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
797e7dbb
TH
496
497 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
498 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
499 rq = &q->pre_flush_rq;
500 } else
501 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
1da177e4 502
797e7dbb
TH
503 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
504 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
505 else
506 rq = NULL;
507
508 return rq;
1da177e4
LT
509}
510
797e7dbb 511int blk_do_ordered(request_queue_t *q, struct request **rqp)
1da177e4 512{
9a7a67af 513 struct request *rq = *rqp;
797e7dbb 514 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
1da177e4 515
797e7dbb
TH
516 if (!q->ordseq) {
517 if (!is_barrier)
518 return 1;
1da177e4 519
797e7dbb
TH
520 if (q->next_ordered != QUEUE_ORDERED_NONE) {
521 *rqp = start_ordered(q, rq);
522 return 1;
523 } else {
524 /*
525 * This can happen when the queue switches to
526 * ORDERED_NONE while this request is on it.
527 */
528 blkdev_dequeue_request(rq);
529 end_that_request_first(rq, -EOPNOTSUPP,
530 rq->hard_nr_sectors);
531 end_that_request_last(rq, -EOPNOTSUPP);
532 *rqp = NULL;
533 return 0;
534 }
535 }
1da177e4 536
9a7a67af
JA
537 /*
538 * Ordered sequence in progress
539 */
540
541 /* Special requests are not subject to ordering rules. */
542 if (!blk_fs_request(rq) &&
543 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
544 return 1;
545
797e7dbb 546 if (q->ordered & QUEUE_ORDERED_TAG) {
9a7a67af 547 /* Ordered by tag. Blocking the next barrier is enough. */
797e7dbb
TH
548 if (is_barrier && rq != &q->bar_rq)
549 *rqp = NULL;
9a7a67af
JA
550 } else {
551 /* Ordered by draining. Wait for turn. */
552 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
553 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
554 *rqp = NULL;
1da177e4
LT
555 }
556
557 return 1;
558}
559
797e7dbb 560static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
1da177e4 561{
797e7dbb
TH
562 request_queue_t *q = bio->bi_private;
563 struct bio_vec *bvec;
564 int i;
565
566 /*
567 * This is dry run, restore bio_sector and size. We'll finish
568 * this request again with the original bi_end_io after an
569 * error occurs or post flush is complete.
570 */
571 q->bi_size += bytes;
572
573 if (bio->bi_size)
574 return 1;
575
576 /* Rewind bvec's */
577 bio->bi_idx = 0;
578 bio_for_each_segment(bvec, bio, i) {
579 bvec->bv_len += bvec->bv_offset;
580 bvec->bv_offset = 0;
581 }
582
583 /* Reset bio */
584 set_bit(BIO_UPTODATE, &bio->bi_flags);
585 bio->bi_size = q->bi_size;
586 bio->bi_sector -= (q->bi_size >> 9);
587 q->bi_size = 0;
588
589 return 0;
1da177e4 590}
1da177e4 591
797e7dbb
TH
592static inline int ordered_bio_endio(struct request *rq, struct bio *bio,
593 unsigned int nbytes, int error)
1da177e4 594{
797e7dbb
TH
595 request_queue_t *q = rq->q;
596 bio_end_io_t *endio;
597 void *private;
598
599 if (&q->bar_rq != rq)
600 return 0;
601
602 /*
603 * Okay, this is the barrier request in progress, dry finish it.
604 */
605 if (error && !q->orderr)
606 q->orderr = error;
607
608 endio = bio->bi_end_io;
609 private = bio->bi_private;
610 bio->bi_end_io = flush_dry_bio_endio;
611 bio->bi_private = q;
612
613 bio_endio(bio, nbytes, error);
614
615 bio->bi_end_io = endio;
616 bio->bi_private = private;
617
618 return 1;
1da177e4 619}
1da177e4
LT
620
621/**
622 * blk_queue_bounce_limit - set bounce buffer limit for queue
623 * @q: the request queue for the device
624 * @dma_addr: bus address limit
625 *
626 * Description:
627 * Different hardware can have different requirements as to what pages
628 * it can do I/O directly to. A low level driver can call
629 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
5ee1af9f 630 * buffers for doing I/O to pages residing above @page.
1da177e4
LT
631 **/
632void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
633{
634 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
5ee1af9f
AK
635 int dma = 0;
636
637 q->bounce_gfp = GFP_NOIO;
638#if BITS_PER_LONG == 64
639 /* Assume anything <= 4GB can be handled by IOMMU.
640 Actually some IOMMUs can handle everything, but I don't
641 know of a way to test this here. */
8269730b 642 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
5ee1af9f
AK
643 dma = 1;
644 q->bounce_pfn = max_low_pfn;
645#else
646 if (bounce_pfn < blk_max_low_pfn)
647 dma = 1;
648 q->bounce_pfn = bounce_pfn;
649#endif
650 if (dma) {
1da177e4
LT
651 init_emergency_isa_pool();
652 q->bounce_gfp = GFP_NOIO | GFP_DMA;
5ee1af9f
AK
653 q->bounce_pfn = bounce_pfn;
654 }
1da177e4
LT
655}
656
657EXPORT_SYMBOL(blk_queue_bounce_limit);
658
659/**
660 * blk_queue_max_sectors - set max sectors for a request for this queue
661 * @q: the request queue for the device
662 * @max_sectors: max sectors in the usual 512b unit
663 *
664 * Description:
665 * Enables a low level driver to set an upper limit on the size of
666 * received requests.
667 **/
2cb2e147 668void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
1da177e4
LT
669{
670 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
671 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
672 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
673 }
674
defd94b7
MC
675 if (BLK_DEF_MAX_SECTORS > max_sectors)
676 q->max_hw_sectors = q->max_sectors = max_sectors;
677 else {
678 q->max_sectors = BLK_DEF_MAX_SECTORS;
679 q->max_hw_sectors = max_sectors;
680 }
1da177e4
LT
681}
682
683EXPORT_SYMBOL(blk_queue_max_sectors);
684
685/**
686 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
687 * @q: the request queue for the device
688 * @max_segments: max number of segments
689 *
690 * Description:
691 * Enables a low level driver to set an upper limit on the number of
692 * physical data segments in a request. This would be the largest sized
693 * scatter list the driver could handle.
694 **/
695void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
696{
697 if (!max_segments) {
698 max_segments = 1;
699 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
700 }
701
702 q->max_phys_segments = max_segments;
703}
704
705EXPORT_SYMBOL(blk_queue_max_phys_segments);
706
707/**
708 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
709 * @q: the request queue for the device
710 * @max_segments: max number of segments
711 *
712 * Description:
713 * Enables a low level driver to set an upper limit on the number of
714 * hw data segments in a request. This would be the largest number of
715 * address/length pairs the host adapter can actually give as once
716 * to the device.
717 **/
718void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
719{
720 if (!max_segments) {
721 max_segments = 1;
722 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
723 }
724
725 q->max_hw_segments = max_segments;
726}
727
728EXPORT_SYMBOL(blk_queue_max_hw_segments);
729
730/**
731 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
732 * @q: the request queue for the device
733 * @max_size: max size of segment in bytes
734 *
735 * Description:
736 * Enables a low level driver to set an upper limit on the size of a
737 * coalesced segment
738 **/
739void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
740{
741 if (max_size < PAGE_CACHE_SIZE) {
742 max_size = PAGE_CACHE_SIZE;
743 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
744 }
745
746 q->max_segment_size = max_size;
747}
748
749EXPORT_SYMBOL(blk_queue_max_segment_size);
750
751/**
752 * blk_queue_hardsect_size - set hardware sector size for the queue
753 * @q: the request queue for the device
754 * @size: the hardware sector size, in bytes
755 *
756 * Description:
757 * This should typically be set to the lowest possible sector size
758 * that the hardware can operate on (possible without reverting to
759 * even internal read-modify-write operations). Usually the default
760 * of 512 covers most hardware.
761 **/
762void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
763{
764 q->hardsect_size = size;
765}
766
767EXPORT_SYMBOL(blk_queue_hardsect_size);
768
769/*
770 * Returns the minimum that is _not_ zero, unless both are zero.
771 */
772#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
773
774/**
775 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
776 * @t: the stacking driver (top)
777 * @b: the underlying device (bottom)
778 **/
779void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
780{
781 /* zero is "infinity" */
defd94b7
MC
782 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
783 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
1da177e4
LT
784
785 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
786 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
787 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
788 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
89e5c8b5
N
789 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
790 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
1da177e4
LT
791}
792
793EXPORT_SYMBOL(blk_queue_stack_limits);
794
795/**
796 * blk_queue_segment_boundary - set boundary rules for segment merging
797 * @q: the request queue for the device
798 * @mask: the memory boundary mask
799 **/
800void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
801{
802 if (mask < PAGE_CACHE_SIZE - 1) {
803 mask = PAGE_CACHE_SIZE - 1;
804 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
805 }
806
807 q->seg_boundary_mask = mask;
808}
809
810EXPORT_SYMBOL(blk_queue_segment_boundary);
811
812/**
813 * blk_queue_dma_alignment - set dma length and memory alignment
814 * @q: the request queue for the device
815 * @mask: alignment mask
816 *
817 * description:
818 * set required memory and length aligment for direct dma transactions.
819 * this is used when buiding direct io requests for the queue.
820 *
821 **/
822void blk_queue_dma_alignment(request_queue_t *q, int mask)
823{
824 q->dma_alignment = mask;
825}
826
827EXPORT_SYMBOL(blk_queue_dma_alignment);
828
829/**
830 * blk_queue_find_tag - find a request by its tag and queue
1da177e4
LT
831 * @q: The request queue for the device
832 * @tag: The tag of the request
833 *
834 * Notes:
835 * Should be used when a device returns a tag and you want to match
836 * it with a request.
837 *
838 * no locks need be held.
839 **/
840struct request *blk_queue_find_tag(request_queue_t *q, int tag)
841{
842 struct blk_queue_tag *bqt = q->queue_tags;
843
ba025082 844 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
1da177e4
LT
845 return NULL;
846
847 return bqt->tag_index[tag];
848}
849
850EXPORT_SYMBOL(blk_queue_find_tag);
851
852/**
492dfb48
JB
853 * __blk_free_tags - release a given set of tag maintenance info
854 * @bqt: the tag map to free
1da177e4 855 *
492dfb48
JB
856 * Tries to free the specified @bqt@. Returns true if it was
857 * actually freed and false if there are still references using it
858 */
859static int __blk_free_tags(struct blk_queue_tag *bqt)
1da177e4 860{
492dfb48 861 int retval;
1da177e4 862
492dfb48
JB
863 retval = atomic_dec_and_test(&bqt->refcnt);
864 if (retval) {
1da177e4
LT
865 BUG_ON(bqt->busy);
866 BUG_ON(!list_empty(&bqt->busy_list));
867
868 kfree(bqt->tag_index);
869 bqt->tag_index = NULL;
870
871 kfree(bqt->tag_map);
872 bqt->tag_map = NULL;
873
874 kfree(bqt);
492dfb48 875
1da177e4
LT
876 }
877
492dfb48
JB
878 return retval;
879}
880
881/**
882 * __blk_queue_free_tags - release tag maintenance info
883 * @q: the request queue for the device
884 *
885 * Notes:
886 * blk_cleanup_queue() will take care of calling this function, if tagging
887 * has been used. So there's no need to call this directly.
888 **/
889static void __blk_queue_free_tags(request_queue_t *q)
890{
891 struct blk_queue_tag *bqt = q->queue_tags;
892
893 if (!bqt)
894 return;
895
896 __blk_free_tags(bqt);
897
1da177e4
LT
898 q->queue_tags = NULL;
899 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
900}
901
492dfb48
JB
902
903/**
904 * blk_free_tags - release a given set of tag maintenance info
905 * @bqt: the tag map to free
906 *
907 * For externally managed @bqt@ frees the map. Callers of this
908 * function must guarantee to have released all the queues that
909 * might have been using this tag map.
910 */
911void blk_free_tags(struct blk_queue_tag *bqt)
912{
913 if (unlikely(!__blk_free_tags(bqt)))
914 BUG();
915}
916EXPORT_SYMBOL(blk_free_tags);
917
1da177e4
LT
918/**
919 * blk_queue_free_tags - release tag maintenance info
920 * @q: the request queue for the device
921 *
922 * Notes:
923 * This is used to disabled tagged queuing to a device, yet leave
924 * queue in function.
925 **/
926void blk_queue_free_tags(request_queue_t *q)
927{
928 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
929}
930
931EXPORT_SYMBOL(blk_queue_free_tags);
932
933static int
934init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
935{
1da177e4
LT
936 struct request **tag_index;
937 unsigned long *tag_map;
fa72b903 938 int nr_ulongs;
1da177e4 939
492dfb48 940 if (q && depth > q->nr_requests * 2) {
1da177e4
LT
941 depth = q->nr_requests * 2;
942 printk(KERN_ERR "%s: adjusted depth to %d\n",
943 __FUNCTION__, depth);
944 }
945
f68110fc 946 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
1da177e4
LT
947 if (!tag_index)
948 goto fail;
949
f7d37d02 950 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
f68110fc 951 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
1da177e4
LT
952 if (!tag_map)
953 goto fail;
954
ba025082 955 tags->real_max_depth = depth;
1da177e4 956 tags->max_depth = depth;
1da177e4
LT
957 tags->tag_index = tag_index;
958 tags->tag_map = tag_map;
959
1da177e4
LT
960 return 0;
961fail:
962 kfree(tag_index);
963 return -ENOMEM;
964}
965
492dfb48
JB
966static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
967 int depth)
968{
969 struct blk_queue_tag *tags;
970
971 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
972 if (!tags)
973 goto fail;
974
975 if (init_tag_map(q, tags, depth))
976 goto fail;
977
978 INIT_LIST_HEAD(&tags->busy_list);
979 tags->busy = 0;
980 atomic_set(&tags->refcnt, 1);
981 return tags;
982fail:
983 kfree(tags);
984 return NULL;
985}
986
987/**
988 * blk_init_tags - initialize the tag info for an external tag map
989 * @depth: the maximum queue depth supported
990 * @tags: the tag to use
991 **/
992struct blk_queue_tag *blk_init_tags(int depth)
993{
994 return __blk_queue_init_tags(NULL, depth);
995}
996EXPORT_SYMBOL(blk_init_tags);
997
1da177e4
LT
998/**
999 * blk_queue_init_tags - initialize the queue tag info
1000 * @q: the request queue for the device
1001 * @depth: the maximum queue depth supported
1002 * @tags: the tag to use
1003 **/
1004int blk_queue_init_tags(request_queue_t *q, int depth,
1005 struct blk_queue_tag *tags)
1006{
1007 int rc;
1008
1009 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
1010
1011 if (!tags && !q->queue_tags) {
492dfb48 1012 tags = __blk_queue_init_tags(q, depth);
1da177e4 1013
492dfb48 1014 if (!tags)
1da177e4 1015 goto fail;
1da177e4
LT
1016 } else if (q->queue_tags) {
1017 if ((rc = blk_queue_resize_tags(q, depth)))
1018 return rc;
1019 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
1020 return 0;
1021 } else
1022 atomic_inc(&tags->refcnt);
1023
1024 /*
1025 * assign it, all done
1026 */
1027 q->queue_tags = tags;
1028 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
1029 return 0;
1030fail:
1031 kfree(tags);
1032 return -ENOMEM;
1033}
1034
1035EXPORT_SYMBOL(blk_queue_init_tags);
1036
1037/**
1038 * blk_queue_resize_tags - change the queueing depth
1039 * @q: the request queue for the device
1040 * @new_depth: the new max command queueing depth
1041 *
1042 * Notes:
1043 * Must be called with the queue lock held.
1044 **/
1045int blk_queue_resize_tags(request_queue_t *q, int new_depth)
1046{
1047 struct blk_queue_tag *bqt = q->queue_tags;
1048 struct request **tag_index;
1049 unsigned long *tag_map;
fa72b903 1050 int max_depth, nr_ulongs;
1da177e4
LT
1051
1052 if (!bqt)
1053 return -ENXIO;
1054
ba025082
TH
1055 /*
1056 * if we already have large enough real_max_depth. just
1057 * adjust max_depth. *NOTE* as requests with tag value
1058 * between new_depth and real_max_depth can be in-flight, tag
1059 * map can not be shrunk blindly here.
1060 */
1061 if (new_depth <= bqt->real_max_depth) {
1062 bqt->max_depth = new_depth;
1063 return 0;
1064 }
1065
492dfb48
JB
1066 /*
1067 * Currently cannot replace a shared tag map with a new
1068 * one, so error out if this is the case
1069 */
1070 if (atomic_read(&bqt->refcnt) != 1)
1071 return -EBUSY;
1072
1da177e4
LT
1073 /*
1074 * save the old state info, so we can copy it back
1075 */
1076 tag_index = bqt->tag_index;
1077 tag_map = bqt->tag_map;
ba025082 1078 max_depth = bqt->real_max_depth;
1da177e4
LT
1079
1080 if (init_tag_map(q, bqt, new_depth))
1081 return -ENOMEM;
1082
1083 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
f7d37d02 1084 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
fa72b903 1085 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
1da177e4
LT
1086
1087 kfree(tag_index);
1088 kfree(tag_map);
1089 return 0;
1090}
1091
1092EXPORT_SYMBOL(blk_queue_resize_tags);
1093
1094/**
1095 * blk_queue_end_tag - end tag operations for a request
1096 * @q: the request queue for the device
1097 * @rq: the request that has completed
1098 *
1099 * Description:
1100 * Typically called when end_that_request_first() returns 0, meaning
1101 * all transfers have been done for a request. It's important to call
1102 * this function before end_that_request_last(), as that will put the
1103 * request back on the free list thus corrupting the internal tag list.
1104 *
1105 * Notes:
1106 * queue lock must be held.
1107 **/
1108void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1109{
1110 struct blk_queue_tag *bqt = q->queue_tags;
1111 int tag = rq->tag;
1112
1113 BUG_ON(tag == -1);
1114
ba025082 1115 if (unlikely(tag >= bqt->real_max_depth))
040c928c
TH
1116 /*
1117 * This can happen after tag depth has been reduced.
1118 * FIXME: how about a warning or info message here?
1119 */
1da177e4
LT
1120 return;
1121
1122 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
040c928c
TH
1123 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1124 __FUNCTION__, tag);
1da177e4
LT
1125 return;
1126 }
1127
1128 list_del_init(&rq->queuelist);
4aff5e23 1129 rq->cmd_flags &= ~REQ_QUEUED;
1da177e4
LT
1130 rq->tag = -1;
1131
1132 if (unlikely(bqt->tag_index[tag] == NULL))
040c928c
TH
1133 printk(KERN_ERR "%s: tag %d is missing\n",
1134 __FUNCTION__, tag);
1da177e4
LT
1135
1136 bqt->tag_index[tag] = NULL;
1137 bqt->busy--;
1138}
1139
1140EXPORT_SYMBOL(blk_queue_end_tag);
1141
1142/**
1143 * blk_queue_start_tag - find a free tag and assign it
1144 * @q: the request queue for the device
1145 * @rq: the block request that needs tagging
1146 *
1147 * Description:
1148 * This can either be used as a stand-alone helper, or possibly be
1149 * assigned as the queue &prep_rq_fn (in which case &struct request
1150 * automagically gets a tag assigned). Note that this function
1151 * assumes that any type of request can be queued! if this is not
1152 * true for your device, you must check the request type before
1153 * calling this function. The request will also be removed from
1154 * the request queue, so it's the drivers responsibility to readd
1155 * it if it should need to be restarted for some reason.
1156 *
1157 * Notes:
1158 * queue lock must be held.
1159 **/
1160int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1161{
1162 struct blk_queue_tag *bqt = q->queue_tags;
2bf0fdad 1163 int tag;
1da177e4 1164
4aff5e23 1165 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
1da177e4 1166 printk(KERN_ERR
040c928c
TH
1167 "%s: request %p for device [%s] already tagged %d",
1168 __FUNCTION__, rq,
1169 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
1da177e4
LT
1170 BUG();
1171 }
1172
2bf0fdad
TH
1173 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1174 if (tag >= bqt->max_depth)
1175 return 1;
1da177e4 1176
1da177e4
LT
1177 __set_bit(tag, bqt->tag_map);
1178
4aff5e23 1179 rq->cmd_flags |= REQ_QUEUED;
1da177e4
LT
1180 rq->tag = tag;
1181 bqt->tag_index[tag] = rq;
1182 blkdev_dequeue_request(rq);
1183 list_add(&rq->queuelist, &bqt->busy_list);
1184 bqt->busy++;
1185 return 0;
1186}
1187
1188EXPORT_SYMBOL(blk_queue_start_tag);
1189
1190/**
1191 * blk_queue_invalidate_tags - invalidate all pending tags
1192 * @q: the request queue for the device
1193 *
1194 * Description:
1195 * Hardware conditions may dictate a need to stop all pending requests.
1196 * In this case, we will safely clear the block side of the tag queue and
1197 * readd all requests to the request queue in the right order.
1198 *
1199 * Notes:
1200 * queue lock must be held.
1201 **/
1202void blk_queue_invalidate_tags(request_queue_t *q)
1203{
1204 struct blk_queue_tag *bqt = q->queue_tags;
1205 struct list_head *tmp, *n;
1206 struct request *rq;
1207
1208 list_for_each_safe(tmp, n, &bqt->busy_list) {
1209 rq = list_entry_rq(tmp);
1210
1211 if (rq->tag == -1) {
040c928c
TH
1212 printk(KERN_ERR
1213 "%s: bad tag found on list\n", __FUNCTION__);
1da177e4 1214 list_del_init(&rq->queuelist);
4aff5e23 1215 rq->cmd_flags &= ~REQ_QUEUED;
1da177e4
LT
1216 } else
1217 blk_queue_end_tag(q, rq);
1218
4aff5e23 1219 rq->cmd_flags &= ~REQ_STARTED;
1da177e4
LT
1220 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1221 }
1222}
1223
1224EXPORT_SYMBOL(blk_queue_invalidate_tags);
1225
1da177e4
LT
1226void blk_dump_rq_flags(struct request *rq, char *msg)
1227{
1228 int bit;
1229
4aff5e23
JA
1230 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1231 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1232 rq->cmd_flags);
1da177e4
LT
1233
1234 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1235 rq->nr_sectors,
1236 rq->current_nr_sectors);
1237 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1238
4aff5e23 1239 if (blk_pc_request(rq)) {
1da177e4
LT
1240 printk("cdb: ");
1241 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1242 printk("%02x ", rq->cmd[bit]);
1243 printk("\n");
1244 }
1245}
1246
1247EXPORT_SYMBOL(blk_dump_rq_flags);
1248
1249void blk_recount_segments(request_queue_t *q, struct bio *bio)
1250{
1251 struct bio_vec *bv, *bvprv = NULL;
1252 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1253 int high, highprv = 1;
1254
1255 if (unlikely(!bio->bi_io_vec))
1256 return;
1257
1258 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1259 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1260 bio_for_each_segment(bv, bio, i) {
1261 /*
1262 * the trick here is making sure that a high page is never
1263 * considered part of another segment, since that might
1264 * change with the bounce page.
1265 */
1266 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1267 if (high || highprv)
1268 goto new_hw_segment;
1269 if (cluster) {
1270 if (seg_size + bv->bv_len > q->max_segment_size)
1271 goto new_segment;
1272 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1273 goto new_segment;
1274 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1275 goto new_segment;
1276 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1277 goto new_hw_segment;
1278
1279 seg_size += bv->bv_len;
1280 hw_seg_size += bv->bv_len;
1281 bvprv = bv;
1282 continue;
1283 }
1284new_segment:
1285 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1286 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1287 hw_seg_size += bv->bv_len;
1288 } else {
1289new_hw_segment:
1290 if (hw_seg_size > bio->bi_hw_front_size)
1291 bio->bi_hw_front_size = hw_seg_size;
1292 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1293 nr_hw_segs++;
1294 }
1295
1296 nr_phys_segs++;
1297 bvprv = bv;
1298 seg_size = bv->bv_len;
1299 highprv = high;
1300 }
1301 if (hw_seg_size > bio->bi_hw_back_size)
1302 bio->bi_hw_back_size = hw_seg_size;
1303 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1304 bio->bi_hw_front_size = hw_seg_size;
1305 bio->bi_phys_segments = nr_phys_segs;
1306 bio->bi_hw_segments = nr_hw_segs;
1307 bio->bi_flags |= (1 << BIO_SEG_VALID);
1308}
1309
1310
93d17d3d 1311static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
1da177e4
LT
1312 struct bio *nxt)
1313{
1314 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1315 return 0;
1316
1317 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1318 return 0;
1319 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1320 return 0;
1321
1322 /*
1323 * bio and nxt are contigous in memory, check if the queue allows
1324 * these two to be merged into one
1325 */
1326 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1327 return 1;
1328
1329 return 0;
1330}
1331
93d17d3d 1332static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1da177e4
LT
1333 struct bio *nxt)
1334{
1335 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1336 blk_recount_segments(q, bio);
1337 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1338 blk_recount_segments(q, nxt);
1339 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1340 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1341 return 0;
1342 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1343 return 0;
1344
1345 return 1;
1346}
1347
1da177e4
LT
1348/*
1349 * map a request to scatterlist, return number of sg entries setup. Caller
1350 * must make sure sg can hold rq->nr_phys_segments entries
1351 */
1352int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1353{
1354 struct bio_vec *bvec, *bvprv;
1355 struct bio *bio;
1356 int nsegs, i, cluster;
1357
1358 nsegs = 0;
1359 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1360
1361 /*
1362 * for each bio in rq
1363 */
1364 bvprv = NULL;
1365 rq_for_each_bio(bio, rq) {
1366 /*
1367 * for each segment in bio
1368 */
1369 bio_for_each_segment(bvec, bio, i) {
1370 int nbytes = bvec->bv_len;
1371
1372 if (bvprv && cluster) {
1373 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1374 goto new_segment;
1375
1376 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1377 goto new_segment;
1378 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1379 goto new_segment;
1380
1381 sg[nsegs - 1].length += nbytes;
1382 } else {
1383new_segment:
1384 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1385 sg[nsegs].page = bvec->bv_page;
1386 sg[nsegs].length = nbytes;
1387 sg[nsegs].offset = bvec->bv_offset;
1388
1389 nsegs++;
1390 }
1391 bvprv = bvec;
1392 } /* segments in bio */
1393 } /* bios in rq */
1394
1395 return nsegs;
1396}
1397
1398EXPORT_SYMBOL(blk_rq_map_sg);
1399
1400/*
1401 * the standard queue merge functions, can be overridden with device
1402 * specific ones if so desired
1403 */
1404
1405static inline int ll_new_mergeable(request_queue_t *q,
1406 struct request *req,
1407 struct bio *bio)
1408{
1409 int nr_phys_segs = bio_phys_segments(q, bio);
1410
1411 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
4aff5e23 1412 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1413 if (req == q->last_merge)
1414 q->last_merge = NULL;
1415 return 0;
1416 }
1417
1418 /*
1419 * A hw segment is just getting larger, bump just the phys
1420 * counter.
1421 */
1422 req->nr_phys_segments += nr_phys_segs;
1423 return 1;
1424}
1425
1426static inline int ll_new_hw_segment(request_queue_t *q,
1427 struct request *req,
1428 struct bio *bio)
1429{
1430 int nr_hw_segs = bio_hw_segments(q, bio);
1431 int nr_phys_segs = bio_phys_segments(q, bio);
1432
1433 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1434 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
4aff5e23 1435 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1436 if (req == q->last_merge)
1437 q->last_merge = NULL;
1438 return 0;
1439 }
1440
1441 /*
1442 * This will form the start of a new hw segment. Bump both
1443 * counters.
1444 */
1445 req->nr_hw_segments += nr_hw_segs;
1446 req->nr_phys_segments += nr_phys_segs;
1447 return 1;
1448}
1449
1450static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1451 struct bio *bio)
1452{
defd94b7 1453 unsigned short max_sectors;
1da177e4
LT
1454 int len;
1455
defd94b7
MC
1456 if (unlikely(blk_pc_request(req)))
1457 max_sectors = q->max_hw_sectors;
1458 else
1459 max_sectors = q->max_sectors;
1460
1461 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
4aff5e23 1462 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1463 if (req == q->last_merge)
1464 q->last_merge = NULL;
1465 return 0;
1466 }
1467 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1468 blk_recount_segments(q, req->biotail);
1469 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1470 blk_recount_segments(q, bio);
1471 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1472 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1473 !BIOVEC_VIRT_OVERSIZE(len)) {
1474 int mergeable = ll_new_mergeable(q, req, bio);
1475
1476 if (mergeable) {
1477 if (req->nr_hw_segments == 1)
1478 req->bio->bi_hw_front_size = len;
1479 if (bio->bi_hw_segments == 1)
1480 bio->bi_hw_back_size = len;
1481 }
1482 return mergeable;
1483 }
1484
1485 return ll_new_hw_segment(q, req, bio);
1486}
1487
1488static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1489 struct bio *bio)
1490{
defd94b7 1491 unsigned short max_sectors;
1da177e4
LT
1492 int len;
1493
defd94b7
MC
1494 if (unlikely(blk_pc_request(req)))
1495 max_sectors = q->max_hw_sectors;
1496 else
1497 max_sectors = q->max_sectors;
1498
1499
1500 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
4aff5e23 1501 req->cmd_flags |= REQ_NOMERGE;
1da177e4
LT
1502 if (req == q->last_merge)
1503 q->last_merge = NULL;
1504 return 0;
1505 }
1506 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1507 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1508 blk_recount_segments(q, bio);
1509 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1510 blk_recount_segments(q, req->bio);
1511 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1512 !BIOVEC_VIRT_OVERSIZE(len)) {
1513 int mergeable = ll_new_mergeable(q, req, bio);
1514
1515 if (mergeable) {
1516 if (bio->bi_hw_segments == 1)
1517 bio->bi_hw_front_size = len;
1518 if (req->nr_hw_segments == 1)
1519 req->biotail->bi_hw_back_size = len;
1520 }
1521 return mergeable;
1522 }
1523
1524 return ll_new_hw_segment(q, req, bio);
1525}
1526
1527static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1528 struct request *next)
1529{
dfa1a553
ND
1530 int total_phys_segments;
1531 int total_hw_segments;
1da177e4
LT
1532
1533 /*
1534 * First check if the either of the requests are re-queued
1535 * requests. Can't merge them if they are.
1536 */
1537 if (req->special || next->special)
1538 return 0;
1539
1540 /*
dfa1a553 1541 * Will it become too large?
1da177e4
LT
1542 */
1543 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1544 return 0;
1545
1546 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1547 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1548 total_phys_segments--;
1549
1550 if (total_phys_segments > q->max_phys_segments)
1551 return 0;
1552
1553 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1554 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1555 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1556 /*
1557 * propagate the combined length to the end of the requests
1558 */
1559 if (req->nr_hw_segments == 1)
1560 req->bio->bi_hw_front_size = len;
1561 if (next->nr_hw_segments == 1)
1562 next->biotail->bi_hw_back_size = len;
1563 total_hw_segments--;
1564 }
1565
1566 if (total_hw_segments > q->max_hw_segments)
1567 return 0;
1568
1569 /* Merge is OK... */
1570 req->nr_phys_segments = total_phys_segments;
1571 req->nr_hw_segments = total_hw_segments;
1572 return 1;
1573}
1574
1575/*
1576 * "plug" the device if there are no outstanding requests: this will
1577 * force the transfer to start only after we have put all the requests
1578 * on the list.
1579 *
1580 * This is called with interrupts off and no requests on the queue and
1581 * with the queue lock held.
1582 */
1583void blk_plug_device(request_queue_t *q)
1584{
1585 WARN_ON(!irqs_disabled());
1586
1587 /*
1588 * don't plug a stopped queue, it must be paired with blk_start_queue()
1589 * which will restart the queueing
1590 */
7daac490 1591 if (blk_queue_stopped(q))
1da177e4
LT
1592 return;
1593
2056a782 1594 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
1da177e4 1595 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
2056a782
JA
1596 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1597 }
1da177e4
LT
1598}
1599
1600EXPORT_SYMBOL(blk_plug_device);
1601
1602/*
1603 * remove the queue from the plugged list, if present. called with
1604 * queue lock held and interrupts disabled.
1605 */
1606int blk_remove_plug(request_queue_t *q)
1607{
1608 WARN_ON(!irqs_disabled());
1609
1610 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1611 return 0;
1612
1613 del_timer(&q->unplug_timer);
1614 return 1;
1615}
1616
1617EXPORT_SYMBOL(blk_remove_plug);
1618
1619/*
1620 * remove the plug and let it rip..
1621 */
1622void __generic_unplug_device(request_queue_t *q)
1623{
7daac490 1624 if (unlikely(blk_queue_stopped(q)))
1da177e4
LT
1625 return;
1626
1627 if (!blk_remove_plug(q))
1628 return;
1629
22e2c507 1630 q->request_fn(q);
1da177e4
LT
1631}
1632EXPORT_SYMBOL(__generic_unplug_device);
1633
1634/**
1635 * generic_unplug_device - fire a request queue
1636 * @q: The &request_queue_t in question
1637 *
1638 * Description:
1639 * Linux uses plugging to build bigger requests queues before letting
1640 * the device have at them. If a queue is plugged, the I/O scheduler
1641 * is still adding and merging requests on the queue. Once the queue
1642 * gets unplugged, the request_fn defined for the queue is invoked and
1643 * transfers started.
1644 **/
1645void generic_unplug_device(request_queue_t *q)
1646{
1647 spin_lock_irq(q->queue_lock);
1648 __generic_unplug_device(q);
1649 spin_unlock_irq(q->queue_lock);
1650}
1651EXPORT_SYMBOL(generic_unplug_device);
1652
1653static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1654 struct page *page)
1655{
1656 request_queue_t *q = bdi->unplug_io_data;
1657
1658 /*
1659 * devices don't necessarily have an ->unplug_fn defined
1660 */
2056a782
JA
1661 if (q->unplug_fn) {
1662 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1663 q->rq.count[READ] + q->rq.count[WRITE]);
1664
1da177e4 1665 q->unplug_fn(q);
2056a782 1666 }
1da177e4
LT
1667}
1668
1669static void blk_unplug_work(void *data)
1670{
1671 request_queue_t *q = data;
1672
2056a782
JA
1673 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1674 q->rq.count[READ] + q->rq.count[WRITE]);
1675
1da177e4
LT
1676 q->unplug_fn(q);
1677}
1678
1679static void blk_unplug_timeout(unsigned long data)
1680{
1681 request_queue_t *q = (request_queue_t *)data;
1682
2056a782
JA
1683 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1684 q->rq.count[READ] + q->rq.count[WRITE]);
1685
1da177e4
LT
1686 kblockd_schedule_work(&q->unplug_work);
1687}
1688
1689/**
1690 * blk_start_queue - restart a previously stopped queue
1691 * @q: The &request_queue_t in question
1692 *
1693 * Description:
1694 * blk_start_queue() will clear the stop flag on the queue, and call
1695 * the request_fn for the queue if it was in a stopped state when
1696 * entered. Also see blk_stop_queue(). Queue lock must be held.
1697 **/
1698void blk_start_queue(request_queue_t *q)
1699{
a038e253
PBG
1700 WARN_ON(!irqs_disabled());
1701
1da177e4
LT
1702 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1703
1704 /*
1705 * one level of recursion is ok and is much faster than kicking
1706 * the unplug handling
1707 */
1708 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1709 q->request_fn(q);
1710 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1711 } else {
1712 blk_plug_device(q);
1713 kblockd_schedule_work(&q->unplug_work);
1714 }
1715}
1716
1717EXPORT_SYMBOL(blk_start_queue);
1718
1719/**
1720 * blk_stop_queue - stop a queue
1721 * @q: The &request_queue_t in question
1722 *
1723 * Description:
1724 * The Linux block layer assumes that a block driver will consume all
1725 * entries on the request queue when the request_fn strategy is called.
1726 * Often this will not happen, because of hardware limitations (queue
1727 * depth settings). If a device driver gets a 'queue full' response,
1728 * or if it simply chooses not to queue more I/O at one point, it can
1729 * call this function to prevent the request_fn from being called until
1730 * the driver has signalled it's ready to go again. This happens by calling
1731 * blk_start_queue() to restart queue operations. Queue lock must be held.
1732 **/
1733void blk_stop_queue(request_queue_t *q)
1734{
1735 blk_remove_plug(q);
1736 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1737}
1738EXPORT_SYMBOL(blk_stop_queue);
1739
1740/**
1741 * blk_sync_queue - cancel any pending callbacks on a queue
1742 * @q: the queue
1743 *
1744 * Description:
1745 * The block layer may perform asynchronous callback activity
1746 * on a queue, such as calling the unplug function after a timeout.
1747 * A block device may call blk_sync_queue to ensure that any
1748 * such activity is cancelled, thus allowing it to release resources
1749 * the the callbacks might use. The caller must already have made sure
1750 * that its ->make_request_fn will not re-add plugging prior to calling
1751 * this function.
1752 *
1753 */
1754void blk_sync_queue(struct request_queue *q)
1755{
1756 del_timer_sync(&q->unplug_timer);
1757 kblockd_flush();
1758}
1759EXPORT_SYMBOL(blk_sync_queue);
1760
1761/**
1762 * blk_run_queue - run a single device queue
1763 * @q: The queue to run
1764 */
1765void blk_run_queue(struct request_queue *q)
1766{
1767 unsigned long flags;
1768
1769 spin_lock_irqsave(q->queue_lock, flags);
1770 blk_remove_plug(q);
dac07ec1
JA
1771
1772 /*
1773 * Only recurse once to avoid overrunning the stack, let the unplug
1774 * handling reinvoke the handler shortly if we already got there.
1775 */
1776 if (!elv_queue_empty(q)) {
1777 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1778 q->request_fn(q);
1779 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1780 } else {
1781 blk_plug_device(q);
1782 kblockd_schedule_work(&q->unplug_work);
1783 }
1784 }
1785
1da177e4
LT
1786 spin_unlock_irqrestore(q->queue_lock, flags);
1787}
1788EXPORT_SYMBOL(blk_run_queue);
1789
1790/**
1791 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
a580290c 1792 * @kobj: the kobj belonging of the request queue to be released
1da177e4
LT
1793 *
1794 * Description:
1795 * blk_cleanup_queue is the pair to blk_init_queue() or
1796 * blk_queue_make_request(). It should be called when a request queue is
1797 * being released; typically when a block device is being de-registered.
1798 * Currently, its primary task it to free all the &struct request
1799 * structures that were allocated to the queue and the queue itself.
1800 *
1801 * Caveat:
1802 * Hopefully the low level driver will have finished any
1803 * outstanding requests first...
1804 **/
483f4afc 1805static void blk_release_queue(struct kobject *kobj)
1da177e4 1806{
483f4afc 1807 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
1da177e4
LT
1808 struct request_list *rl = &q->rq;
1809
1da177e4
LT
1810 blk_sync_queue(q);
1811
1812 if (rl->rq_pool)
1813 mempool_destroy(rl->rq_pool);
1814
1815 if (q->queue_tags)
1816 __blk_queue_free_tags(q);
1817
6c5c9341 1818 blk_trace_shutdown(q);
2056a782 1819
1da177e4
LT
1820 kmem_cache_free(requestq_cachep, q);
1821}
1822
483f4afc
AV
1823void blk_put_queue(request_queue_t *q)
1824{
1825 kobject_put(&q->kobj);
1826}
1827EXPORT_SYMBOL(blk_put_queue);
1828
1829void blk_cleanup_queue(request_queue_t * q)
1830{
1831 mutex_lock(&q->sysfs_lock);
1832 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1833 mutex_unlock(&q->sysfs_lock);
1834
1835 if (q->elevator)
1836 elevator_exit(q->elevator);
1837
1838 blk_put_queue(q);
1839}
1840
1da177e4
LT
1841EXPORT_SYMBOL(blk_cleanup_queue);
1842
1843static int blk_init_free_list(request_queue_t *q)
1844{
1845 struct request_list *rl = &q->rq;
1846
1847 rl->count[READ] = rl->count[WRITE] = 0;
1848 rl->starved[READ] = rl->starved[WRITE] = 0;
cb98fc8b 1849 rl->elvpriv = 0;
1da177e4
LT
1850 init_waitqueue_head(&rl->wait[READ]);
1851 init_waitqueue_head(&rl->wait[WRITE]);
1da177e4 1852
1946089a
CL
1853 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1854 mempool_free_slab, request_cachep, q->node);
1da177e4
LT
1855
1856 if (!rl->rq_pool)
1857 return -ENOMEM;
1858
1859 return 0;
1860}
1861
8267e268 1862request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
1da177e4 1863{
1946089a
CL
1864 return blk_alloc_queue_node(gfp_mask, -1);
1865}
1866EXPORT_SYMBOL(blk_alloc_queue);
1da177e4 1867
483f4afc
AV
1868static struct kobj_type queue_ktype;
1869
8267e268 1870request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
1946089a
CL
1871{
1872 request_queue_t *q;
1873
1874 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
1da177e4
LT
1875 if (!q)
1876 return NULL;
1877
1878 memset(q, 0, sizeof(*q));
1879 init_timer(&q->unplug_timer);
483f4afc
AV
1880
1881 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1882 q->kobj.ktype = &queue_ktype;
1883 kobject_init(&q->kobj);
1da177e4
LT
1884
1885 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1886 q->backing_dev_info.unplug_io_data = q;
1887
483f4afc
AV
1888 mutex_init(&q->sysfs_lock);
1889
1da177e4
LT
1890 return q;
1891}
1946089a 1892EXPORT_SYMBOL(blk_alloc_queue_node);
1da177e4
LT
1893
1894/**
1895 * blk_init_queue - prepare a request queue for use with a block device
1896 * @rfn: The function to be called to process requests that have been
1897 * placed on the queue.
1898 * @lock: Request queue spin lock
1899 *
1900 * Description:
1901 * If a block device wishes to use the standard request handling procedures,
1902 * which sorts requests and coalesces adjacent requests, then it must
1903 * call blk_init_queue(). The function @rfn will be called when there
1904 * are requests on the queue that need to be processed. If the device
1905 * supports plugging, then @rfn may not be called immediately when requests
1906 * are available on the queue, but may be called at some time later instead.
1907 * Plugged queues are generally unplugged when a buffer belonging to one
1908 * of the requests on the queue is needed, or due to memory pressure.
1909 *
1910 * @rfn is not required, or even expected, to remove all requests off the
1911 * queue, but only as many as it can handle at a time. If it does leave
1912 * requests on the queue, it is responsible for arranging that the requests
1913 * get dealt with eventually.
1914 *
1915 * The queue spin lock must be held while manipulating the requests on the
a038e253
PBG
1916 * request queue; this lock will be taken also from interrupt context, so irq
1917 * disabling is needed for it.
1da177e4
LT
1918 *
1919 * Function returns a pointer to the initialized request queue, or NULL if
1920 * it didn't succeed.
1921 *
1922 * Note:
1923 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1924 * when the block device is deactivated (such as at module unload).
1925 **/
1946089a 1926
1da177e4
LT
1927request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1928{
1946089a
CL
1929 return blk_init_queue_node(rfn, lock, -1);
1930}
1931EXPORT_SYMBOL(blk_init_queue);
1932
1933request_queue_t *
1934blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1935{
1936 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1da177e4
LT
1937
1938 if (!q)
1939 return NULL;
1940
1946089a 1941 q->node = node_id;
8669aafd
AV
1942 if (blk_init_free_list(q)) {
1943 kmem_cache_free(requestq_cachep, q);
1944 return NULL;
1945 }
1da177e4 1946
152587de
JA
1947 /*
1948 * if caller didn't supply a lock, they get per-queue locking with
1949 * our embedded lock
1950 */
1951 if (!lock) {
1952 spin_lock_init(&q->__queue_lock);
1953 lock = &q->__queue_lock;
1954 }
1955
1da177e4
LT
1956 q->request_fn = rfn;
1957 q->back_merge_fn = ll_back_merge_fn;
1958 q->front_merge_fn = ll_front_merge_fn;
1959 q->merge_requests_fn = ll_merge_requests_fn;
1960 q->prep_rq_fn = NULL;
1961 q->unplug_fn = generic_unplug_device;
1962 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1963 q->queue_lock = lock;
1964
1965 blk_queue_segment_boundary(q, 0xffffffff);
1966
1967 blk_queue_make_request(q, __make_request);
1968 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1969
1970 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1971 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1972
1973 /*
1974 * all done
1975 */
1976 if (!elevator_init(q, NULL)) {
1977 blk_queue_congestion_threshold(q);
1978 return q;
1979 }
1980
8669aafd 1981 blk_put_queue(q);
1da177e4
LT
1982 return NULL;
1983}
1946089a 1984EXPORT_SYMBOL(blk_init_queue_node);
1da177e4
LT
1985
1986int blk_get_queue(request_queue_t *q)
1987{
fde6ad22 1988 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
483f4afc 1989 kobject_get(&q->kobj);
1da177e4
LT
1990 return 0;
1991 }
1992
1993 return 1;
1994}
1995
1996EXPORT_SYMBOL(blk_get_queue);
1997
1998static inline void blk_free_request(request_queue_t *q, struct request *rq)
1999{
4aff5e23 2000 if (rq->cmd_flags & REQ_ELVPRIV)
cb98fc8b 2001 elv_put_request(q, rq);
1da177e4
LT
2002 mempool_free(rq, q->rq.rq_pool);
2003}
2004
22e2c507 2005static inline struct request *
cb78b285 2006blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask)
1da177e4
LT
2007{
2008 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
2009
2010 if (!rq)
2011 return NULL;
2012
2013 /*
4aff5e23 2014 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1da177e4
LT
2015 * see bio.h and blkdev.h
2016 */
49171e5c 2017 rq->cmd_flags = rw | REQ_ALLOCED;
1da177e4 2018
cb98fc8b 2019 if (priv) {
cb78b285 2020 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
cb98fc8b
TH
2021 mempool_free(rq, q->rq.rq_pool);
2022 return NULL;
2023 }
4aff5e23 2024 rq->cmd_flags |= REQ_ELVPRIV;
cb98fc8b 2025 }
1da177e4 2026
cb98fc8b 2027 return rq;
1da177e4
LT
2028}
2029
2030/*
2031 * ioc_batching returns true if the ioc is a valid batching request and
2032 * should be given priority access to a request.
2033 */
2034static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
2035{
2036 if (!ioc)
2037 return 0;
2038
2039 /*
2040 * Make sure the process is able to allocate at least 1 request
2041 * even if the batch times out, otherwise we could theoretically
2042 * lose wakeups.
2043 */
2044 return ioc->nr_batch_requests == q->nr_batching ||
2045 (ioc->nr_batch_requests > 0
2046 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2047}
2048
2049/*
2050 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2051 * will cause the process to be a "batcher" on all queues in the system. This
2052 * is the behaviour we want though - once it gets a wakeup it should be given
2053 * a nice run.
2054 */
93d17d3d 2055static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
1da177e4
LT
2056{
2057 if (!ioc || ioc_batching(q, ioc))
2058 return;
2059
2060 ioc->nr_batch_requests = q->nr_batching;
2061 ioc->last_waited = jiffies;
2062}
2063
2064static void __freed_request(request_queue_t *q, int rw)
2065{
2066 struct request_list *rl = &q->rq;
2067
2068 if (rl->count[rw] < queue_congestion_off_threshold(q))
2069 clear_queue_congested(q, rw);
2070
2071 if (rl->count[rw] + 1 <= q->nr_requests) {
1da177e4
LT
2072 if (waitqueue_active(&rl->wait[rw]))
2073 wake_up(&rl->wait[rw]);
2074
2075 blk_clear_queue_full(q, rw);
2076 }
2077}
2078
2079/*
2080 * A request has just been released. Account for it, update the full and
2081 * congestion status, wake up any waiters. Called under q->queue_lock.
2082 */
cb98fc8b 2083static void freed_request(request_queue_t *q, int rw, int priv)
1da177e4
LT
2084{
2085 struct request_list *rl = &q->rq;
2086
2087 rl->count[rw]--;
cb98fc8b
TH
2088 if (priv)
2089 rl->elvpriv--;
1da177e4
LT
2090
2091 __freed_request(q, rw);
2092
2093 if (unlikely(rl->starved[rw ^ 1]))
2094 __freed_request(q, rw ^ 1);
1da177e4
LT
2095}
2096
2097#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2098/*
d6344532
NP
2099 * Get a free request, queue_lock must be held.
2100 * Returns NULL on failure, with queue_lock held.
2101 * Returns !NULL on success, with queue_lock *not held*.
1da177e4 2102 */
22e2c507 2103static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
8267e268 2104 gfp_t gfp_mask)
1da177e4
LT
2105{
2106 struct request *rq = NULL;
2107 struct request_list *rl = &q->rq;
88ee5ef1
JA
2108 struct io_context *ioc = NULL;
2109 int may_queue, priv;
2110
cb78b285 2111 may_queue = elv_may_queue(q, rw);
88ee5ef1
JA
2112 if (may_queue == ELV_MQUEUE_NO)
2113 goto rq_starved;
2114
2115 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2116 if (rl->count[rw]+1 >= q->nr_requests) {
2117 ioc = current_io_context(GFP_ATOMIC);
2118 /*
2119 * The queue will fill after this allocation, so set
2120 * it as full, and mark this process as "batching".
2121 * This process will be allowed to complete a batch of
2122 * requests, others will be blocked.
2123 */
2124 if (!blk_queue_full(q, rw)) {
2125 ioc_set_batching(q, ioc);
2126 blk_set_queue_full(q, rw);
2127 } else {
2128 if (may_queue != ELV_MQUEUE_MUST
2129 && !ioc_batching(q, ioc)) {
2130 /*
2131 * The queue is full and the allocating
2132 * process is not a "batcher", and not
2133 * exempted by the IO scheduler
2134 */
2135 goto out;
2136 }
2137 }
1da177e4 2138 }
88ee5ef1 2139 set_queue_congested(q, rw);
1da177e4
LT
2140 }
2141
082cf69e
JA
2142 /*
2143 * Only allow batching queuers to allocate up to 50% over the defined
2144 * limit of requests, otherwise we could have thousands of requests
2145 * allocated with any setting of ->nr_requests
2146 */
fd782a4a 2147 if (rl->count[rw] >= (3 * q->nr_requests / 2))
082cf69e 2148 goto out;
fd782a4a 2149
1da177e4
LT
2150 rl->count[rw]++;
2151 rl->starved[rw] = 0;
cb98fc8b 2152
64521d1a 2153 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
cb98fc8b
TH
2154 if (priv)
2155 rl->elvpriv++;
2156
1da177e4
LT
2157 spin_unlock_irq(q->queue_lock);
2158
cb78b285 2159 rq = blk_alloc_request(q, rw, priv, gfp_mask);
88ee5ef1 2160 if (unlikely(!rq)) {
1da177e4
LT
2161 /*
2162 * Allocation failed presumably due to memory. Undo anything
2163 * we might have messed up.
2164 *
2165 * Allocating task should really be put onto the front of the
2166 * wait queue, but this is pretty rare.
2167 */
2168 spin_lock_irq(q->queue_lock);
cb98fc8b 2169 freed_request(q, rw, priv);
1da177e4
LT
2170
2171 /*
2172 * in the very unlikely event that allocation failed and no
2173 * requests for this direction was pending, mark us starved
2174 * so that freeing of a request in the other direction will
2175 * notice us. another possible fix would be to split the
2176 * rq mempool into READ and WRITE
2177 */
2178rq_starved:
2179 if (unlikely(rl->count[rw] == 0))
2180 rl->starved[rw] = 1;
2181
1da177e4
LT
2182 goto out;
2183 }
2184
88ee5ef1
JA
2185 /*
2186 * ioc may be NULL here, and ioc_batching will be false. That's
2187 * OK, if the queue is under the request limit then requests need
2188 * not count toward the nr_batch_requests limit. There will always
2189 * be some limit enforced by BLK_BATCH_TIME.
2190 */
1da177e4
LT
2191 if (ioc_batching(q, ioc))
2192 ioc->nr_batch_requests--;
2193
2194 rq_init(q, rq);
2056a782
JA
2195
2196 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
1da177e4 2197out:
1da177e4
LT
2198 return rq;
2199}
2200
2201/*
2202 * No available requests for this queue, unplug the device and wait for some
2203 * requests to become available.
d6344532
NP
2204 *
2205 * Called with q->queue_lock held, and returns with it unlocked.
1da177e4 2206 */
22e2c507
JA
2207static struct request *get_request_wait(request_queue_t *q, int rw,
2208 struct bio *bio)
1da177e4 2209{
1da177e4
LT
2210 struct request *rq;
2211
450991bc
NP
2212 rq = get_request(q, rw, bio, GFP_NOIO);
2213 while (!rq) {
2214 DEFINE_WAIT(wait);
1da177e4
LT
2215 struct request_list *rl = &q->rq;
2216
2217 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2218 TASK_UNINTERRUPTIBLE);
2219
22e2c507 2220 rq = get_request(q, rw, bio, GFP_NOIO);
1da177e4
LT
2221
2222 if (!rq) {
2223 struct io_context *ioc;
2224
2056a782
JA
2225 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2226
d6344532
NP
2227 __generic_unplug_device(q);
2228 spin_unlock_irq(q->queue_lock);
1da177e4
LT
2229 io_schedule();
2230
2231 /*
2232 * After sleeping, we become a "batching" process and
2233 * will be able to allocate at least one request, and
2234 * up to a big batch of them for a small period time.
2235 * See ioc_batching, ioc_set_batching
2236 */
fb3cc432 2237 ioc = current_io_context(GFP_NOIO);
1da177e4 2238 ioc_set_batching(q, ioc);
d6344532
NP
2239
2240 spin_lock_irq(q->queue_lock);
1da177e4
LT
2241 }
2242 finish_wait(&rl->wait[rw], &wait);
450991bc 2243 }
1da177e4
LT
2244
2245 return rq;
2246}
2247
8267e268 2248struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
1da177e4
LT
2249{
2250 struct request *rq;
2251
2252 BUG_ON(rw != READ && rw != WRITE);
2253
d6344532
NP
2254 spin_lock_irq(q->queue_lock);
2255 if (gfp_mask & __GFP_WAIT) {
22e2c507 2256 rq = get_request_wait(q, rw, NULL);
d6344532 2257 } else {
22e2c507 2258 rq = get_request(q, rw, NULL, gfp_mask);
d6344532
NP
2259 if (!rq)
2260 spin_unlock_irq(q->queue_lock);
2261 }
2262 /* q->queue_lock is unlocked at this point */
1da177e4
LT
2263
2264 return rq;
2265}
1da177e4
LT
2266EXPORT_SYMBOL(blk_get_request);
2267
2268/**
2269 * blk_requeue_request - put a request back on queue
2270 * @q: request queue where request should be inserted
2271 * @rq: request to be inserted
2272 *
2273 * Description:
2274 * Drivers often keep queueing requests until the hardware cannot accept
2275 * more, when that condition happens we need to put the request back
2276 * on the queue. Must be called with queue lock held.
2277 */
2278void blk_requeue_request(request_queue_t *q, struct request *rq)
2279{
2056a782
JA
2280 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2281
1da177e4
LT
2282 if (blk_rq_tagged(rq))
2283 blk_queue_end_tag(q, rq);
2284
2285 elv_requeue_request(q, rq);
2286}
2287
2288EXPORT_SYMBOL(blk_requeue_request);
2289
2290/**
2291 * blk_insert_request - insert a special request in to a request queue
2292 * @q: request queue where request should be inserted
2293 * @rq: request to be inserted
2294 * @at_head: insert request at head or tail of queue
2295 * @data: private data
1da177e4
LT
2296 *
2297 * Description:
2298 * Many block devices need to execute commands asynchronously, so they don't
2299 * block the whole kernel from preemption during request execution. This is
2300 * accomplished normally by inserting aritficial requests tagged as
2301 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2302 * scheduled for actual execution by the request queue.
2303 *
2304 * We have the option of inserting the head or the tail of the queue.
2305 * Typically we use the tail for new ioctls and so forth. We use the head
2306 * of the queue for things like a QUEUE_FULL message from a device, or a
2307 * host that is unable to accept a particular command.
2308 */
2309void blk_insert_request(request_queue_t *q, struct request *rq,
867d1191 2310 int at_head, void *data)
1da177e4 2311{
867d1191 2312 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1da177e4
LT
2313 unsigned long flags;
2314
2315 /*
2316 * tell I/O scheduler that this isn't a regular read/write (ie it
2317 * must not attempt merges on this) and that it acts as a soft
2318 * barrier
2319 */
4aff5e23
JA
2320 rq->cmd_type = REQ_TYPE_SPECIAL;
2321 rq->cmd_flags |= REQ_SOFTBARRIER;
1da177e4
LT
2322
2323 rq->special = data;
2324
2325 spin_lock_irqsave(q->queue_lock, flags);
2326
2327 /*
2328 * If command is tagged, release the tag
2329 */
867d1191
TH
2330 if (blk_rq_tagged(rq))
2331 blk_queue_end_tag(q, rq);
1da177e4 2332
867d1191
TH
2333 drive_stat_acct(rq, rq->nr_sectors, 1);
2334 __elv_add_request(q, rq, where, 0);
1da177e4 2335
1da177e4
LT
2336 if (blk_queue_plugged(q))
2337 __generic_unplug_device(q);
2338 else
2339 q->request_fn(q);
2340 spin_unlock_irqrestore(q->queue_lock, flags);
2341}
2342
2343EXPORT_SYMBOL(blk_insert_request);
2344
2345/**
2346 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2347 * @q: request queue where request should be inserted
73747aed 2348 * @rq: request structure to fill
1da177e4
LT
2349 * @ubuf: the user buffer
2350 * @len: length of user data
2351 *
2352 * Description:
2353 * Data will be mapped directly for zero copy io, if possible. Otherwise
2354 * a kernel bounce buffer is used.
2355 *
2356 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2357 * still in process context.
2358 *
2359 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2360 * before being submitted to the device, as pages mapped may be out of
2361 * reach. It's the callers responsibility to make sure this happens. The
2362 * original bio must be passed back in to blk_rq_unmap_user() for proper
2363 * unmapping.
2364 */
dd1cab95
JA
2365int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2366 unsigned int len)
1da177e4
LT
2367{
2368 unsigned long uaddr;
1da177e4 2369 struct bio *bio;
dd1cab95 2370 int reading;
1da177e4 2371
defd94b7 2372 if (len > (q->max_hw_sectors << 9))
dd1cab95
JA
2373 return -EINVAL;
2374 if (!len || !ubuf)
2375 return -EINVAL;
1da177e4 2376
dd1cab95 2377 reading = rq_data_dir(rq) == READ;
1da177e4
LT
2378
2379 /*
2380 * if alignment requirement is satisfied, map in user pages for
2381 * direct dma. else, set up kernel bounce buffers
2382 */
2383 uaddr = (unsigned long) ubuf;
2384 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
dd1cab95 2385 bio = bio_map_user(q, NULL, uaddr, len, reading);
1da177e4 2386 else
dd1cab95 2387 bio = bio_copy_user(q, uaddr, len, reading);
1da177e4
LT
2388
2389 if (!IS_ERR(bio)) {
2390 rq->bio = rq->biotail = bio;
2391 blk_rq_bio_prep(q, rq, bio);
2392
2393 rq->buffer = rq->data = NULL;
2394 rq->data_len = len;
dd1cab95 2395 return 0;
1da177e4
LT
2396 }
2397
2398 /*
2399 * bio is the err-ptr
2400 */
dd1cab95 2401 return PTR_ERR(bio);
1da177e4
LT
2402}
2403
2404EXPORT_SYMBOL(blk_rq_map_user);
2405
f1970baf
JB
2406/**
2407 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2408 * @q: request queue where request should be inserted
2409 * @rq: request to map data to
2410 * @iov: pointer to the iovec
2411 * @iov_count: number of elements in the iovec
2412 *
2413 * Description:
2414 * Data will be mapped directly for zero copy io, if possible. Otherwise
2415 * a kernel bounce buffer is used.
2416 *
2417 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2418 * still in process context.
2419 *
2420 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2421 * before being submitted to the device, as pages mapped may be out of
2422 * reach. It's the callers responsibility to make sure this happens. The
2423 * original bio must be passed back in to blk_rq_unmap_user() for proper
2424 * unmapping.
2425 */
2426int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2427 struct sg_iovec *iov, int iov_count)
2428{
2429 struct bio *bio;
2430
2431 if (!iov || iov_count <= 0)
2432 return -EINVAL;
2433
2434 /* we don't allow misaligned data like bio_map_user() does. If the
2435 * user is using sg, they're expected to know the alignment constraints
2436 * and respect them accordingly */
2437 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2438 if (IS_ERR(bio))
2439 return PTR_ERR(bio);
2440
2441 rq->bio = rq->biotail = bio;
2442 blk_rq_bio_prep(q, rq, bio);
2443 rq->buffer = rq->data = NULL;
2444 rq->data_len = bio->bi_size;
2445 return 0;
2446}
2447
2448EXPORT_SYMBOL(blk_rq_map_user_iov);
2449
1da177e4
LT
2450/**
2451 * blk_rq_unmap_user - unmap a request with user data
73747aed 2452 * @bio: bio to be unmapped
1da177e4
LT
2453 * @ulen: length of user buffer
2454 *
2455 * Description:
73747aed 2456 * Unmap a bio previously mapped by blk_rq_map_user().
1da177e4 2457 */
dd1cab95 2458int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
1da177e4
LT
2459{
2460 int ret = 0;
2461
2462 if (bio) {
2463 if (bio_flagged(bio, BIO_USER_MAPPED))
2464 bio_unmap_user(bio);
2465 else
2466 ret = bio_uncopy_user(bio);
2467 }
2468
dd1cab95 2469 return 0;
1da177e4
LT
2470}
2471
2472EXPORT_SYMBOL(blk_rq_unmap_user);
2473
df46b9a4
MC
2474/**
2475 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2476 * @q: request queue where request should be inserted
73747aed 2477 * @rq: request to fill
df46b9a4
MC
2478 * @kbuf: the kernel buffer
2479 * @len: length of user data
73747aed 2480 * @gfp_mask: memory allocation flags
df46b9a4 2481 */
dd1cab95 2482int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
8267e268 2483 unsigned int len, gfp_t gfp_mask)
df46b9a4 2484{
df46b9a4
MC
2485 struct bio *bio;
2486
defd94b7 2487 if (len > (q->max_hw_sectors << 9))
dd1cab95
JA
2488 return -EINVAL;
2489 if (!len || !kbuf)
2490 return -EINVAL;
df46b9a4
MC
2491
2492 bio = bio_map_kern(q, kbuf, len, gfp_mask);
dd1cab95
JA
2493 if (IS_ERR(bio))
2494 return PTR_ERR(bio);
df46b9a4 2495
dd1cab95
JA
2496 if (rq_data_dir(rq) == WRITE)
2497 bio->bi_rw |= (1 << BIO_RW);
df46b9a4 2498
dd1cab95
JA
2499 rq->bio = rq->biotail = bio;
2500 blk_rq_bio_prep(q, rq, bio);
df46b9a4 2501
dd1cab95
JA
2502 rq->buffer = rq->data = NULL;
2503 rq->data_len = len;
2504 return 0;
df46b9a4
MC
2505}
2506
2507EXPORT_SYMBOL(blk_rq_map_kern);
2508
73747aed
CH
2509/**
2510 * blk_execute_rq_nowait - insert a request into queue for execution
2511 * @q: queue to insert the request in
2512 * @bd_disk: matching gendisk
2513 * @rq: request to insert
2514 * @at_head: insert request at head or tail of queue
2515 * @done: I/O completion handler
2516 *
2517 * Description:
2518 * Insert a fully prepared request at the back of the io scheduler queue
2519 * for execution. Don't wait for completion.
2520 */
f1970baf
JB
2521void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2522 struct request *rq, int at_head,
8ffdc655 2523 rq_end_io_fn *done)
f1970baf
JB
2524{
2525 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2526
2527 rq->rq_disk = bd_disk;
4aff5e23 2528 rq->cmd_flags |= REQ_NOMERGE;
f1970baf 2529 rq->end_io = done;
4c5d0bbd
AM
2530 WARN_ON(irqs_disabled());
2531 spin_lock_irq(q->queue_lock);
2532 __elv_add_request(q, rq, where, 1);
2533 __generic_unplug_device(q);
2534 spin_unlock_irq(q->queue_lock);
f1970baf 2535}
6e39b69e
MC
2536EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2537
1da177e4
LT
2538/**
2539 * blk_execute_rq - insert a request into queue for execution
2540 * @q: queue to insert the request in
2541 * @bd_disk: matching gendisk
2542 * @rq: request to insert
994ca9a1 2543 * @at_head: insert request at head or tail of queue
1da177e4
LT
2544 *
2545 * Description:
2546 * Insert a fully prepared request at the back of the io scheduler queue
73747aed 2547 * for execution and wait for completion.
1da177e4
LT
2548 */
2549int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
994ca9a1 2550 struct request *rq, int at_head)
1da177e4 2551{
60be6b9a 2552 DECLARE_COMPLETION_ONSTACK(wait);
1da177e4
LT
2553 char sense[SCSI_SENSE_BUFFERSIZE];
2554 int err = 0;
2555
1da177e4
LT
2556 /*
2557 * we need an extra reference to the request, so we can look at
2558 * it after io completion
2559 */
2560 rq->ref_count++;
2561
2562 if (!rq->sense) {
2563 memset(sense, 0, sizeof(sense));
2564 rq->sense = sense;
2565 rq->sense_len = 0;
2566 }
2567
c00895ab 2568 rq->end_io_data = &wait;
994ca9a1 2569 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
1da177e4 2570 wait_for_completion(&wait);
1da177e4
LT
2571
2572 if (rq->errors)
2573 err = -EIO;
2574
2575 return err;
2576}
2577
2578EXPORT_SYMBOL(blk_execute_rq);
2579
2580/**
2581 * blkdev_issue_flush - queue a flush
2582 * @bdev: blockdev to issue flush for
2583 * @error_sector: error sector
2584 *
2585 * Description:
2586 * Issue a flush for the block device in question. Caller can supply
2587 * room for storing the error offset in case of a flush error, if they
2588 * wish to. Caller must run wait_for_completion() on its own.
2589 */
2590int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2591{
2592 request_queue_t *q;
2593
2594 if (bdev->bd_disk == NULL)
2595 return -ENXIO;
2596
2597 q = bdev_get_queue(bdev);
2598 if (!q)
2599 return -ENXIO;
2600 if (!q->issue_flush_fn)
2601 return -EOPNOTSUPP;
2602
2603 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2604}
2605
2606EXPORT_SYMBOL(blkdev_issue_flush);
2607
93d17d3d 2608static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
1da177e4
LT
2609{
2610 int rw = rq_data_dir(rq);
2611
2612 if (!blk_fs_request(rq) || !rq->rq_disk)
2613 return;
2614
d72d904a 2615 if (!new_io) {
a362357b 2616 __disk_stat_inc(rq->rq_disk, merges[rw]);
d72d904a 2617 } else {
1da177e4
LT
2618 disk_round_stats(rq->rq_disk);
2619 rq->rq_disk->in_flight++;
2620 }
2621}
2622
2623/*
2624 * add-request adds a request to the linked list.
2625 * queue lock is held and interrupts disabled, as we muck with the
2626 * request queue list.
2627 */
2628static inline void add_request(request_queue_t * q, struct request * req)
2629{
2630 drive_stat_acct(req, req->nr_sectors, 1);
2631
2632 if (q->activity_fn)
2633 q->activity_fn(q->activity_data, rq_data_dir(req));
2634
2635 /*
2636 * elevator indicated where it wants this request to be
2637 * inserted at elevator_merge time
2638 */
2639 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2640}
2641
2642/*
2643 * disk_round_stats() - Round off the performance stats on a struct
2644 * disk_stats.
2645 *
2646 * The average IO queue length and utilisation statistics are maintained
2647 * by observing the current state of the queue length and the amount of
2648 * time it has been in this state for.
2649 *
2650 * Normally, that accounting is done on IO completion, but that can result
2651 * in more than a second's worth of IO being accounted for within any one
2652 * second, leading to >100% utilisation. To deal with that, we call this
2653 * function to do a round-off before returning the results when reading
2654 * /proc/diskstats. This accounts immediately for all queue usage up to
2655 * the current jiffies and restarts the counters again.
2656 */
2657void disk_round_stats(struct gendisk *disk)
2658{
2659 unsigned long now = jiffies;
2660
b2982649
KC
2661 if (now == disk->stamp)
2662 return;
1da177e4 2663
20e5c81f
KC
2664 if (disk->in_flight) {
2665 __disk_stat_add(disk, time_in_queue,
2666 disk->in_flight * (now - disk->stamp));
2667 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2668 }
1da177e4 2669 disk->stamp = now;
1da177e4
LT
2670}
2671
3eaf840e
JNN
2672EXPORT_SYMBOL_GPL(disk_round_stats);
2673
1da177e4
LT
2674/*
2675 * queue lock must be held
2676 */
6e39b69e 2677void __blk_put_request(request_queue_t *q, struct request *req)
1da177e4 2678{
1da177e4
LT
2679 if (unlikely(!q))
2680 return;
2681 if (unlikely(--req->ref_count))
2682 return;
2683
8922e16c
TH
2684 elv_completed_request(q, req);
2685
1da177e4
LT
2686 /*
2687 * Request may not have originated from ll_rw_blk. if not,
2688 * it didn't come out of our reserved rq pools
2689 */
49171e5c 2690 if (req->cmd_flags & REQ_ALLOCED) {
1da177e4 2691 int rw = rq_data_dir(req);
4aff5e23 2692 int priv = req->cmd_flags & REQ_ELVPRIV;
1da177e4 2693
1da177e4 2694 BUG_ON(!list_empty(&req->queuelist));
9817064b 2695 BUG_ON(!hlist_unhashed(&req->hash));
1da177e4
LT
2696
2697 blk_free_request(q, req);
cb98fc8b 2698 freed_request(q, rw, priv);
1da177e4
LT
2699 }
2700}
2701
6e39b69e
MC
2702EXPORT_SYMBOL_GPL(__blk_put_request);
2703
1da177e4
LT
2704void blk_put_request(struct request *req)
2705{
8922e16c
TH
2706 unsigned long flags;
2707 request_queue_t *q = req->q;
2708
1da177e4 2709 /*
8922e16c
TH
2710 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2711 * following if (q) test.
1da177e4 2712 */
8922e16c 2713 if (q) {
1da177e4
LT
2714 spin_lock_irqsave(q->queue_lock, flags);
2715 __blk_put_request(q, req);
2716 spin_unlock_irqrestore(q->queue_lock, flags);
2717 }
2718}
2719
2720EXPORT_SYMBOL(blk_put_request);
2721
2722/**
2723 * blk_end_sync_rq - executes a completion event on a request
2724 * @rq: request to complete
fddfdeaf 2725 * @error: end io status of the request
1da177e4 2726 */
8ffdc655 2727void blk_end_sync_rq(struct request *rq, int error)
1da177e4 2728{
c00895ab 2729 struct completion *waiting = rq->end_io_data;
1da177e4 2730
c00895ab 2731 rq->end_io_data = NULL;
1da177e4
LT
2732 __blk_put_request(rq->q, rq);
2733
2734 /*
2735 * complete last, if this is a stack request the process (and thus
2736 * the rq pointer) could be invalid right after this complete()
2737 */
2738 complete(waiting);
2739}
2740EXPORT_SYMBOL(blk_end_sync_rq);
2741
2742/**
2743 * blk_congestion_wait - wait for a queue to become uncongested
2744 * @rw: READ or WRITE
2745 * @timeout: timeout in jiffies
2746 *
2747 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2748 * If no queues are congested then just wait for the next request to be
2749 * returned.
2750 */
2751long blk_congestion_wait(int rw, long timeout)
2752{
2753 long ret;
2754 DEFINE_WAIT(wait);
2755 wait_queue_head_t *wqh = &congestion_wqh[rw];
2756
2757 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2758 ret = io_schedule_timeout(timeout);
2759 finish_wait(wqh, &wait);
2760 return ret;
2761}
2762
2763EXPORT_SYMBOL(blk_congestion_wait);
2764
275a082f
TM
2765/**
2766 * blk_congestion_end - wake up sleepers on a congestion queue
2767 * @rw: READ or WRITE
2768 */
2769void blk_congestion_end(int rw)
2770{
2771 wait_queue_head_t *wqh = &congestion_wqh[rw];
2772
2773 if (waitqueue_active(wqh))
2774 wake_up(wqh);
2775}
2776
1da177e4
LT
2777/*
2778 * Has to be called with the request spinlock acquired
2779 */
2780static int attempt_merge(request_queue_t *q, struct request *req,
2781 struct request *next)
2782{
2783 if (!rq_mergeable(req) || !rq_mergeable(next))
2784 return 0;
2785
2786 /*
d6e05edc 2787 * not contiguous
1da177e4
LT
2788 */
2789 if (req->sector + req->nr_sectors != next->sector)
2790 return 0;
2791
2792 if (rq_data_dir(req) != rq_data_dir(next)
2793 || req->rq_disk != next->rq_disk
c00895ab 2794 || next->special)
1da177e4
LT
2795 return 0;
2796
2797 /*
2798 * If we are allowed to merge, then append bio list
2799 * from next to rq and release next. merge_requests_fn
2800 * will have updated segment counts, update sector
2801 * counts here.
2802 */
2803 if (!q->merge_requests_fn(q, req, next))
2804 return 0;
2805
2806 /*
2807 * At this point we have either done a back merge
2808 * or front merge. We need the smaller start_time of
2809 * the merged requests to be the current request
2810 * for accounting purposes.
2811 */
2812 if (time_after(req->start_time, next->start_time))
2813 req->start_time = next->start_time;
2814
2815 req->biotail->bi_next = next->bio;
2816 req->biotail = next->biotail;
2817
2818 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2819
2820 elv_merge_requests(q, req, next);
2821
2822 if (req->rq_disk) {
2823 disk_round_stats(req->rq_disk);
2824 req->rq_disk->in_flight--;
2825 }
2826
22e2c507
JA
2827 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2828
1da177e4
LT
2829 __blk_put_request(q, next);
2830 return 1;
2831}
2832
2833static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2834{
2835 struct request *next = elv_latter_request(q, rq);
2836
2837 if (next)
2838 return attempt_merge(q, rq, next);
2839
2840 return 0;
2841}
2842
2843static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2844{
2845 struct request *prev = elv_former_request(q, rq);
2846
2847 if (prev)
2848 return attempt_merge(q, prev, rq);
2849
2850 return 0;
2851}
2852
52d9e675
TH
2853static void init_request_from_bio(struct request *req, struct bio *bio)
2854{
4aff5e23 2855 req->cmd_type = REQ_TYPE_FS;
52d9e675
TH
2856
2857 /*
2858 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2859 */
2860 if (bio_rw_ahead(bio) || bio_failfast(bio))
4aff5e23 2861 req->cmd_flags |= REQ_FAILFAST;
52d9e675
TH
2862
2863 /*
2864 * REQ_BARRIER implies no merging, but lets make it explicit
2865 */
2866 if (unlikely(bio_barrier(bio)))
4aff5e23 2867 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
52d9e675 2868
b31dc66a 2869 if (bio_sync(bio))
4aff5e23 2870 req->cmd_flags |= REQ_RW_SYNC;
b31dc66a 2871
52d9e675
TH
2872 req->errors = 0;
2873 req->hard_sector = req->sector = bio->bi_sector;
2874 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2875 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2876 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2877 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2878 req->buffer = bio_data(bio); /* see ->buffer comment above */
52d9e675
TH
2879 req->bio = req->biotail = bio;
2880 req->ioprio = bio_prio(bio);
2881 req->rq_disk = bio->bi_bdev->bd_disk;
2882 req->start_time = jiffies;
2883}
2884
1da177e4
LT
2885static int __make_request(request_queue_t *q, struct bio *bio)
2886{
450991bc 2887 struct request *req;
51da90fc
JA
2888 int el_ret, nr_sectors, barrier, err;
2889 const unsigned short prio = bio_prio(bio);
2890 const int sync = bio_sync(bio);
1da177e4 2891
1da177e4 2892 nr_sectors = bio_sectors(bio);
1da177e4
LT
2893
2894 /*
2895 * low level driver can indicate that it wants pages above a
2896 * certain limit bounced to low memory (ie for highmem, or even
2897 * ISA dma in theory)
2898 */
2899 blk_queue_bounce(q, &bio);
2900
1da177e4 2901 barrier = bio_barrier(bio);
797e7dbb 2902 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
1da177e4
LT
2903 err = -EOPNOTSUPP;
2904 goto end_io;
2905 }
2906
1da177e4
LT
2907 spin_lock_irq(q->queue_lock);
2908
450991bc 2909 if (unlikely(barrier) || elv_queue_empty(q))
1da177e4
LT
2910 goto get_rq;
2911
2912 el_ret = elv_merge(q, &req, bio);
2913 switch (el_ret) {
2914 case ELEVATOR_BACK_MERGE:
2915 BUG_ON(!rq_mergeable(req));
2916
2917 if (!q->back_merge_fn(q, req, bio))
2918 break;
2919
2056a782
JA
2920 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2921
1da177e4
LT
2922 req->biotail->bi_next = bio;
2923 req->biotail = bio;
2924 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
22e2c507 2925 req->ioprio = ioprio_best(req->ioprio, prio);
1da177e4
LT
2926 drive_stat_acct(req, nr_sectors, 0);
2927 if (!attempt_back_merge(q, req))
2e662b65 2928 elv_merged_request(q, req, el_ret);
1da177e4
LT
2929 goto out;
2930
2931 case ELEVATOR_FRONT_MERGE:
2932 BUG_ON(!rq_mergeable(req));
2933
2934 if (!q->front_merge_fn(q, req, bio))
2935 break;
2936
2056a782
JA
2937 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2938
1da177e4
LT
2939 bio->bi_next = req->bio;
2940 req->bio = bio;
2941
2942 /*
2943 * may not be valid. if the low level driver said
2944 * it didn't need a bounce buffer then it better
2945 * not touch req->buffer either...
2946 */
2947 req->buffer = bio_data(bio);
51da90fc
JA
2948 req->current_nr_sectors = bio_cur_sectors(bio);
2949 req->hard_cur_sectors = req->current_nr_sectors;
2950 req->sector = req->hard_sector = bio->bi_sector;
1da177e4 2951 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
22e2c507 2952 req->ioprio = ioprio_best(req->ioprio, prio);
1da177e4
LT
2953 drive_stat_acct(req, nr_sectors, 0);
2954 if (!attempt_front_merge(q, req))
2e662b65 2955 elv_merged_request(q, req, el_ret);
1da177e4
LT
2956 goto out;
2957
450991bc 2958 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1da177e4 2959 default:
450991bc 2960 ;
1da177e4
LT
2961 }
2962
450991bc 2963get_rq:
1da177e4 2964 /*
450991bc 2965 * Grab a free request. This is might sleep but can not fail.
d6344532 2966 * Returns with the queue unlocked.
450991bc 2967 */
51da90fc 2968 req = get_request_wait(q, bio_data_dir(bio), bio);
d6344532 2969
450991bc
NP
2970 /*
2971 * After dropping the lock and possibly sleeping here, our request
2972 * may now be mergeable after it had proven unmergeable (above).
2973 * We don't worry about that case for efficiency. It won't happen
2974 * often, and the elevators are able to handle it.
1da177e4 2975 */
52d9e675 2976 init_request_from_bio(req, bio);
1da177e4 2977
450991bc
NP
2978 spin_lock_irq(q->queue_lock);
2979 if (elv_queue_empty(q))
2980 blk_plug_device(q);
1da177e4
LT
2981 add_request(q, req);
2982out:
4a534f93 2983 if (sync)
1da177e4
LT
2984 __generic_unplug_device(q);
2985
2986 spin_unlock_irq(q->queue_lock);
2987 return 0;
2988
2989end_io:
2990 bio_endio(bio, nr_sectors << 9, err);
2991 return 0;
2992}
2993
2994/*
2995 * If bio->bi_dev is a partition, remap the location
2996 */
2997static inline void blk_partition_remap(struct bio *bio)
2998{
2999 struct block_device *bdev = bio->bi_bdev;
3000
3001 if (bdev != bdev->bd_contains) {
3002 struct hd_struct *p = bdev->bd_part;
a362357b
JA
3003 const int rw = bio_data_dir(bio);
3004
3005 p->sectors[rw] += bio_sectors(bio);
3006 p->ios[rw]++;
1da177e4 3007
1da177e4
LT
3008 bio->bi_sector += p->start_sect;
3009 bio->bi_bdev = bdev->bd_contains;
3010 }
3011}
3012
1da177e4
LT
3013static void handle_bad_sector(struct bio *bio)
3014{
3015 char b[BDEVNAME_SIZE];
3016
3017 printk(KERN_INFO "attempt to access beyond end of device\n");
3018 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3019 bdevname(bio->bi_bdev, b),
3020 bio->bi_rw,
3021 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3022 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3023
3024 set_bit(BIO_EOF, &bio->bi_flags);
3025}
3026
3027/**
3028 * generic_make_request: hand a buffer to its device driver for I/O
3029 * @bio: The bio describing the location in memory and on the device.
3030 *
3031 * generic_make_request() is used to make I/O requests of block
3032 * devices. It is passed a &struct bio, which describes the I/O that needs
3033 * to be done.
3034 *
3035 * generic_make_request() does not return any status. The
3036 * success/failure status of the request, along with notification of
3037 * completion, is delivered asynchronously through the bio->bi_end_io
3038 * function described (one day) else where.
3039 *
3040 * The caller of generic_make_request must make sure that bi_io_vec
3041 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3042 * set to describe the device address, and the
3043 * bi_end_io and optionally bi_private are set to describe how
3044 * completion notification should be signaled.
3045 *
3046 * generic_make_request and the drivers it calls may use bi_next if this
3047 * bio happens to be merged with someone else, and may change bi_dev and
3048 * bi_sector for remaps as it sees fit. So the values of these fields
3049 * should NOT be depended on after the call to generic_make_request.
3050 */
3051void generic_make_request(struct bio *bio)
3052{
3053 request_queue_t *q;
3054 sector_t maxsector;
3055 int ret, nr_sectors = bio_sectors(bio);
2056a782 3056 dev_t old_dev;
1da177e4
LT
3057
3058 might_sleep();
3059 /* Test device or partition size, when known. */
3060 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3061 if (maxsector) {
3062 sector_t sector = bio->bi_sector;
3063
3064 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3065 /*
3066 * This may well happen - the kernel calls bread()
3067 * without checking the size of the device, e.g., when
3068 * mounting a device.
3069 */
3070 handle_bad_sector(bio);
3071 goto end_io;
3072 }
3073 }
3074
3075 /*
3076 * Resolve the mapping until finished. (drivers are
3077 * still free to implement/resolve their own stacking
3078 * by explicitly returning 0)
3079 *
3080 * NOTE: we don't repeat the blk_size check for each new device.
3081 * Stacking drivers are expected to know what they are doing.
3082 */
2056a782
JA
3083 maxsector = -1;
3084 old_dev = 0;
1da177e4
LT
3085 do {
3086 char b[BDEVNAME_SIZE];
3087
3088 q = bdev_get_queue(bio->bi_bdev);
3089 if (!q) {
3090 printk(KERN_ERR
3091 "generic_make_request: Trying to access "
3092 "nonexistent block-device %s (%Lu)\n",
3093 bdevname(bio->bi_bdev, b),
3094 (long long) bio->bi_sector);
3095end_io:
3096 bio_endio(bio, bio->bi_size, -EIO);
3097 break;
3098 }
3099
3100 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3101 printk("bio too big device %s (%u > %u)\n",
3102 bdevname(bio->bi_bdev, b),
3103 bio_sectors(bio),
3104 q->max_hw_sectors);
3105 goto end_io;
3106 }
3107
fde6ad22 3108 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1da177e4
LT
3109 goto end_io;
3110
1da177e4
LT
3111 /*
3112 * If this device has partitions, remap block n
3113 * of partition p to block n+start(p) of the disk.
3114 */
3115 blk_partition_remap(bio);
3116
2056a782
JA
3117 if (maxsector != -1)
3118 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
3119 maxsector);
3120
3121 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3122
3123 maxsector = bio->bi_sector;
3124 old_dev = bio->bi_bdev->bd_dev;
3125
1da177e4
LT
3126 ret = q->make_request_fn(q, bio);
3127 } while (ret);
3128}
3129
3130EXPORT_SYMBOL(generic_make_request);
3131
3132/**
3133 * submit_bio: submit a bio to the block device layer for I/O
3134 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3135 * @bio: The &struct bio which describes the I/O
3136 *
3137 * submit_bio() is very similar in purpose to generic_make_request(), and
3138 * uses that function to do most of the work. Both are fairly rough
3139 * interfaces, @bio must be presetup and ready for I/O.
3140 *
3141 */
3142void submit_bio(int rw, struct bio *bio)
3143{
3144 int count = bio_sectors(bio);
3145
3146 BIO_BUG_ON(!bio->bi_size);
3147 BIO_BUG_ON(!bio->bi_io_vec);
22e2c507 3148 bio->bi_rw |= rw;
1da177e4 3149 if (rw & WRITE)
f8891e5e 3150 count_vm_events(PGPGOUT, count);
1da177e4 3151 else
f8891e5e 3152 count_vm_events(PGPGIN, count);
1da177e4
LT
3153
3154 if (unlikely(block_dump)) {
3155 char b[BDEVNAME_SIZE];
3156 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3157 current->comm, current->pid,
3158 (rw & WRITE) ? "WRITE" : "READ",
3159 (unsigned long long)bio->bi_sector,
3160 bdevname(bio->bi_bdev,b));
3161 }
3162
3163 generic_make_request(bio);
3164}
3165
3166EXPORT_SYMBOL(submit_bio);
3167
93d17d3d 3168static void blk_recalc_rq_segments(struct request *rq)
1da177e4
LT
3169{
3170 struct bio *bio, *prevbio = NULL;
3171 int nr_phys_segs, nr_hw_segs;
3172 unsigned int phys_size, hw_size;
3173 request_queue_t *q = rq->q;
3174
3175 if (!rq->bio)
3176 return;
3177
3178 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3179 rq_for_each_bio(bio, rq) {
3180 /* Force bio hw/phys segs to be recalculated. */
3181 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3182
3183 nr_phys_segs += bio_phys_segments(q, bio);
3184 nr_hw_segs += bio_hw_segments(q, bio);
3185 if (prevbio) {
3186 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3187 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3188
3189 if (blk_phys_contig_segment(q, prevbio, bio) &&
3190 pseg <= q->max_segment_size) {
3191 nr_phys_segs--;
3192 phys_size += prevbio->bi_size + bio->bi_size;
3193 } else
3194 phys_size = 0;
3195
3196 if (blk_hw_contig_segment(q, prevbio, bio) &&
3197 hseg <= q->max_segment_size) {
3198 nr_hw_segs--;
3199 hw_size += prevbio->bi_size + bio->bi_size;
3200 } else
3201 hw_size = 0;
3202 }
3203 prevbio = bio;
3204 }
3205
3206 rq->nr_phys_segments = nr_phys_segs;
3207 rq->nr_hw_segments = nr_hw_segs;
3208}
3209
93d17d3d 3210static void blk_recalc_rq_sectors(struct request *rq, int nsect)
1da177e4
LT
3211{
3212 if (blk_fs_request(rq)) {
3213 rq->hard_sector += nsect;
3214 rq->hard_nr_sectors -= nsect;
3215
3216 /*
3217 * Move the I/O submission pointers ahead if required.
3218 */
3219 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3220 (rq->sector <= rq->hard_sector)) {
3221 rq->sector = rq->hard_sector;
3222 rq->nr_sectors = rq->hard_nr_sectors;
3223 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3224 rq->current_nr_sectors = rq->hard_cur_sectors;
3225 rq->buffer = bio_data(rq->bio);
3226 }
3227
3228 /*
3229 * if total number of sectors is less than the first segment
3230 * size, something has gone terribly wrong
3231 */
3232 if (rq->nr_sectors < rq->current_nr_sectors) {
3233 printk("blk: request botched\n");
3234 rq->nr_sectors = rq->current_nr_sectors;
3235 }
3236 }
3237}
3238
3239static int __end_that_request_first(struct request *req, int uptodate,
3240 int nr_bytes)
3241{
3242 int total_bytes, bio_nbytes, error, next_idx = 0;
3243 struct bio *bio;
3244
2056a782
JA
3245 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3246
1da177e4
LT
3247 /*
3248 * extend uptodate bool to allow < 0 value to be direct io error
3249 */
3250 error = 0;
3251 if (end_io_error(uptodate))
3252 error = !uptodate ? -EIO : uptodate;
3253
3254 /*
3255 * for a REQ_BLOCK_PC request, we want to carry any eventual
3256 * sense key with us all the way through
3257 */
3258 if (!blk_pc_request(req))
3259 req->errors = 0;
3260
3261 if (!uptodate) {
4aff5e23 3262 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
1da177e4
LT
3263 printk("end_request: I/O error, dev %s, sector %llu\n",
3264 req->rq_disk ? req->rq_disk->disk_name : "?",
3265 (unsigned long long)req->sector);
3266 }
3267
d72d904a 3268 if (blk_fs_request(req) && req->rq_disk) {
a362357b
JA
3269 const int rw = rq_data_dir(req);
3270
53e86061 3271 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
d72d904a
JA
3272 }
3273
1da177e4
LT
3274 total_bytes = bio_nbytes = 0;
3275 while ((bio = req->bio) != NULL) {
3276 int nbytes;
3277
3278 if (nr_bytes >= bio->bi_size) {
3279 req->bio = bio->bi_next;
3280 nbytes = bio->bi_size;
797e7dbb
TH
3281 if (!ordered_bio_endio(req, bio, nbytes, error))
3282 bio_endio(bio, nbytes, error);
1da177e4
LT
3283 next_idx = 0;
3284 bio_nbytes = 0;
3285 } else {
3286 int idx = bio->bi_idx + next_idx;
3287
3288 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3289 blk_dump_rq_flags(req, "__end_that");
3290 printk("%s: bio idx %d >= vcnt %d\n",
3291 __FUNCTION__,
3292 bio->bi_idx, bio->bi_vcnt);
3293 break;
3294 }
3295
3296 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3297 BIO_BUG_ON(nbytes > bio->bi_size);
3298
3299 /*
3300 * not a complete bvec done
3301 */
3302 if (unlikely(nbytes > nr_bytes)) {
3303 bio_nbytes += nr_bytes;
3304 total_bytes += nr_bytes;
3305 break;
3306 }
3307
3308 /*
3309 * advance to the next vector
3310 */
3311 next_idx++;
3312 bio_nbytes += nbytes;
3313 }
3314
3315 total_bytes += nbytes;
3316 nr_bytes -= nbytes;
3317
3318 if ((bio = req->bio)) {
3319 /*
3320 * end more in this run, or just return 'not-done'
3321 */
3322 if (unlikely(nr_bytes <= 0))
3323 break;
3324 }
3325 }
3326
3327 /*
3328 * completely done
3329 */
3330 if (!req->bio)
3331 return 0;
3332
3333 /*
3334 * if the request wasn't completed, update state
3335 */
3336 if (bio_nbytes) {
797e7dbb
TH
3337 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3338 bio_endio(bio, bio_nbytes, error);
1da177e4
LT
3339 bio->bi_idx += next_idx;
3340 bio_iovec(bio)->bv_offset += nr_bytes;
3341 bio_iovec(bio)->bv_len -= nr_bytes;
3342 }
3343
3344 blk_recalc_rq_sectors(req, total_bytes >> 9);
3345 blk_recalc_rq_segments(req);
3346 return 1;
3347}
3348
3349/**
3350 * end_that_request_first - end I/O on a request
3351 * @req: the request being processed
3352 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3353 * @nr_sectors: number of sectors to end I/O on
3354 *
3355 * Description:
3356 * Ends I/O on a number of sectors attached to @req, and sets it up
3357 * for the next range of segments (if any) in the cluster.
3358 *
3359 * Return:
3360 * 0 - we are done with this request, call end_that_request_last()
3361 * 1 - still buffers pending for this request
3362 **/
3363int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3364{
3365 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3366}
3367
3368EXPORT_SYMBOL(end_that_request_first);
3369
3370/**
3371 * end_that_request_chunk - end I/O on a request
3372 * @req: the request being processed
3373 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3374 * @nr_bytes: number of bytes to complete
3375 *
3376 * Description:
3377 * Ends I/O on a number of bytes attached to @req, and sets it up
3378 * for the next range of segments (if any). Like end_that_request_first(),
3379 * but deals with bytes instead of sectors.
3380 *
3381 * Return:
3382 * 0 - we are done with this request, call end_that_request_last()
3383 * 1 - still buffers pending for this request
3384 **/
3385int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3386{
3387 return __end_that_request_first(req, uptodate, nr_bytes);
3388}
3389
3390EXPORT_SYMBOL(end_that_request_chunk);
3391
ff856bad
JA
3392/*
3393 * splice the completion data to a local structure and hand off to
3394 * process_completion_queue() to complete the requests
3395 */
3396static void blk_done_softirq(struct softirq_action *h)
3397{
626ab0e6 3398 struct list_head *cpu_list, local_list;
ff856bad
JA
3399
3400 local_irq_disable();
3401 cpu_list = &__get_cpu_var(blk_cpu_done);
626ab0e6 3402 list_replace_init(cpu_list, &local_list);
ff856bad
JA
3403 local_irq_enable();
3404
3405 while (!list_empty(&local_list)) {
3406 struct request *rq = list_entry(local_list.next, struct request, donelist);
3407
3408 list_del_init(&rq->donelist);
3409 rq->q->softirq_done_fn(rq);
3410 }
3411}
3412
3413#ifdef CONFIG_HOTPLUG_CPU
3414
3415static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3416 void *hcpu)
3417{
3418 /*
3419 * If a CPU goes away, splice its entries to the current CPU
3420 * and trigger a run of the softirq
3421 */
3422 if (action == CPU_DEAD) {
3423 int cpu = (unsigned long) hcpu;
3424
3425 local_irq_disable();
3426 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3427 &__get_cpu_var(blk_cpu_done));
3428 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3429 local_irq_enable();
3430 }
3431
3432 return NOTIFY_OK;
3433}
3434
3435
054cc8a2 3436static struct notifier_block __devinitdata blk_cpu_notifier = {
ff856bad
JA
3437 .notifier_call = blk_cpu_notify,
3438};
3439
3440#endif /* CONFIG_HOTPLUG_CPU */
3441
3442/**
3443 * blk_complete_request - end I/O on a request
3444 * @req: the request being processed
3445 *
3446 * Description:
3447 * Ends all I/O on a request. It does not handle partial completions,
d6e05edc 3448 * unless the driver actually implements this in its completion callback
ff856bad
JA
3449 * through requeueing. Theh actual completion happens out-of-order,
3450 * through a softirq handler. The user must have registered a completion
3451 * callback through blk_queue_softirq_done().
3452 **/
3453
3454void blk_complete_request(struct request *req)
3455{
3456 struct list_head *cpu_list;
3457 unsigned long flags;
3458
3459 BUG_ON(!req->q->softirq_done_fn);
3460
3461 local_irq_save(flags);
3462
3463 cpu_list = &__get_cpu_var(blk_cpu_done);
3464 list_add_tail(&req->donelist, cpu_list);
3465 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3466
3467 local_irq_restore(flags);
3468}
3469
3470EXPORT_SYMBOL(blk_complete_request);
3471
1da177e4
LT
3472/*
3473 * queue lock must be held
3474 */
8ffdc655 3475void end_that_request_last(struct request *req, int uptodate)
1da177e4
LT
3476{
3477 struct gendisk *disk = req->rq_disk;
8ffdc655
TH
3478 int error;
3479
3480 /*
3481 * extend uptodate bool to allow < 0 value to be direct io error
3482 */
3483 error = 0;
3484 if (end_io_error(uptodate))
3485 error = !uptodate ? -EIO : uptodate;
1da177e4
LT
3486
3487 if (unlikely(laptop_mode) && blk_fs_request(req))
3488 laptop_io_completion();
3489
fd0ff8aa
JA
3490 /*
3491 * Account IO completion. bar_rq isn't accounted as a normal
3492 * IO on queueing nor completion. Accounting the containing
3493 * request is enough.
3494 */
3495 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1da177e4 3496 unsigned long duration = jiffies - req->start_time;
a362357b
JA
3497 const int rw = rq_data_dir(req);
3498
3499 __disk_stat_inc(disk, ios[rw]);
3500 __disk_stat_add(disk, ticks[rw], duration);
1da177e4
LT
3501 disk_round_stats(disk);
3502 disk->in_flight--;
3503 }
3504 if (req->end_io)
8ffdc655 3505 req->end_io(req, error);
1da177e4
LT
3506 else
3507 __blk_put_request(req->q, req);
3508}
3509
3510EXPORT_SYMBOL(end_that_request_last);
3511
3512void end_request(struct request *req, int uptodate)
3513{
3514 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3515 add_disk_randomness(req->rq_disk);
3516 blkdev_dequeue_request(req);
8ffdc655 3517 end_that_request_last(req, uptodate);
1da177e4
LT
3518 }
3519}
3520
3521EXPORT_SYMBOL(end_request);
3522
3523void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3524{
4aff5e23
JA
3525 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3526 rq->cmd_flags |= (bio->bi_rw & 3);
1da177e4
LT
3527
3528 rq->nr_phys_segments = bio_phys_segments(q, bio);
3529 rq->nr_hw_segments = bio_hw_segments(q, bio);
3530 rq->current_nr_sectors = bio_cur_sectors(bio);
3531 rq->hard_cur_sectors = rq->current_nr_sectors;
3532 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3533 rq->buffer = bio_data(bio);
3534
3535 rq->bio = rq->biotail = bio;
3536}
3537
3538EXPORT_SYMBOL(blk_rq_bio_prep);
3539
3540int kblockd_schedule_work(struct work_struct *work)
3541{
3542 return queue_work(kblockd_workqueue, work);
3543}
3544
3545EXPORT_SYMBOL(kblockd_schedule_work);
3546
3547void kblockd_flush(void)
3548{
3549 flush_workqueue(kblockd_workqueue);
3550}
3551EXPORT_SYMBOL(kblockd_flush);
3552
3553int __init blk_dev_init(void)
3554{
ff856bad
JA
3555 int i;
3556
1da177e4
LT
3557 kblockd_workqueue = create_workqueue("kblockd");
3558 if (!kblockd_workqueue)
3559 panic("Failed to create kblockd\n");
3560
3561 request_cachep = kmem_cache_create("blkdev_requests",
3562 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3563
3564 requestq_cachep = kmem_cache_create("blkdev_queue",
3565 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3566
3567 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3568 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3569
0a945022 3570 for_each_possible_cpu(i)
ff856bad
JA
3571 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3572
3573 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
5a67e4c5 3574 register_hotcpu_notifier(&blk_cpu_notifier);
ff856bad 3575
1da177e4
LT
3576 blk_max_low_pfn = max_low_pfn;
3577 blk_max_pfn = max_pfn;
3578
3579 return 0;
3580}
3581
3582/*
3583 * IO Context helper functions
3584 */
3585void put_io_context(struct io_context *ioc)
3586{
3587 if (ioc == NULL)
3588 return;
3589
3590 BUG_ON(atomic_read(&ioc->refcount) == 0);
3591
3592 if (atomic_dec_and_test(&ioc->refcount)) {
e2d74ac0
JA
3593 struct cfq_io_context *cic;
3594
334e94de 3595 rcu_read_lock();
1da177e4
LT
3596 if (ioc->aic && ioc->aic->dtor)
3597 ioc->aic->dtor(ioc->aic);
e2d74ac0 3598 if (ioc->cic_root.rb_node != NULL) {
7143dd4b
JA
3599 struct rb_node *n = rb_first(&ioc->cic_root);
3600
3601 cic = rb_entry(n, struct cfq_io_context, rb_node);
e2d74ac0
JA
3602 cic->dtor(ioc);
3603 }
334e94de 3604 rcu_read_unlock();
1da177e4
LT
3605
3606 kmem_cache_free(iocontext_cachep, ioc);
3607 }
3608}
3609EXPORT_SYMBOL(put_io_context);
3610
3611/* Called by the exitting task */
3612void exit_io_context(void)
3613{
3614 unsigned long flags;
3615 struct io_context *ioc;
e2d74ac0 3616 struct cfq_io_context *cic;
1da177e4
LT
3617
3618 local_irq_save(flags);
22e2c507 3619 task_lock(current);
1da177e4
LT
3620 ioc = current->io_context;
3621 current->io_context = NULL;
22e2c507
JA
3622 ioc->task = NULL;
3623 task_unlock(current);
1da177e4
LT
3624 local_irq_restore(flags);
3625
3626 if (ioc->aic && ioc->aic->exit)
3627 ioc->aic->exit(ioc->aic);
e2d74ac0
JA
3628 if (ioc->cic_root.rb_node != NULL) {
3629 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3630 cic->exit(ioc);
3631 }
3632
1da177e4
LT
3633 put_io_context(ioc);
3634}
3635
3636/*
3637 * If the current task has no IO context then create one and initialise it.
fb3cc432 3638 * Otherwise, return its existing IO context.
1da177e4 3639 *
fb3cc432
NP
3640 * This returned IO context doesn't have a specifically elevated refcount,
3641 * but since the current task itself holds a reference, the context can be
3642 * used in general code, so long as it stays within `current` context.
1da177e4 3643 */
8267e268 3644struct io_context *current_io_context(gfp_t gfp_flags)
1da177e4
LT
3645{
3646 struct task_struct *tsk = current;
1da177e4
LT
3647 struct io_context *ret;
3648
1da177e4 3649 ret = tsk->io_context;
fb3cc432
NP
3650 if (likely(ret))
3651 return ret;
1da177e4
LT
3652
3653 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3654 if (ret) {
3655 atomic_set(&ret->refcount, 1);
22e2c507 3656 ret->task = current;
fc46379d 3657 ret->ioprio_changed = 0;
1da177e4
LT
3658 ret->last_waited = jiffies; /* doesn't matter... */
3659 ret->nr_batch_requests = 0; /* because this is 0 */
3660 ret->aic = NULL;
e2d74ac0 3661 ret->cic_root.rb_node = NULL;
9f83e45e
ON
3662 /* make sure set_task_ioprio() sees the settings above */
3663 smp_wmb();
fb3cc432
NP
3664 tsk->io_context = ret;
3665 }
1da177e4 3666
fb3cc432
NP
3667 return ret;
3668}
3669EXPORT_SYMBOL(current_io_context);
1da177e4 3670
fb3cc432
NP
3671/*
3672 * If the current task has no IO context then create one and initialise it.
3673 * If it does have a context, take a ref on it.
3674 *
3675 * This is always called in the context of the task which submitted the I/O.
3676 */
8267e268 3677struct io_context *get_io_context(gfp_t gfp_flags)
fb3cc432
NP
3678{
3679 struct io_context *ret;
3680 ret = current_io_context(gfp_flags);
3681 if (likely(ret))
1da177e4 3682 atomic_inc(&ret->refcount);
1da177e4
LT
3683 return ret;
3684}
3685EXPORT_SYMBOL(get_io_context);
3686
3687void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3688{
3689 struct io_context *src = *psrc;
3690 struct io_context *dst = *pdst;
3691
3692 if (src) {
3693 BUG_ON(atomic_read(&src->refcount) == 0);
3694 atomic_inc(&src->refcount);
3695 put_io_context(dst);
3696 *pdst = src;
3697 }
3698}
3699EXPORT_SYMBOL(copy_io_context);
3700
3701void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3702{
3703 struct io_context *temp;
3704 temp = *ioc1;
3705 *ioc1 = *ioc2;
3706 *ioc2 = temp;
3707}
3708EXPORT_SYMBOL(swap_io_context);
3709
3710/*
3711 * sysfs parts below
3712 */
3713struct queue_sysfs_entry {
3714 struct attribute attr;
3715 ssize_t (*show)(struct request_queue *, char *);
3716 ssize_t (*store)(struct request_queue *, const char *, size_t);
3717};
3718
3719static ssize_t
3720queue_var_show(unsigned int var, char *page)
3721{
3722 return sprintf(page, "%d\n", var);
3723}
3724
3725static ssize_t
3726queue_var_store(unsigned long *var, const char *page, size_t count)
3727{
3728 char *p = (char *) page;
3729
3730 *var = simple_strtoul(p, &p, 10);
3731 return count;
3732}
3733
3734static ssize_t queue_requests_show(struct request_queue *q, char *page)
3735{
3736 return queue_var_show(q->nr_requests, (page));
3737}
3738
3739static ssize_t
3740queue_requests_store(struct request_queue *q, const char *page, size_t count)
3741{
3742 struct request_list *rl = &q->rq;
c981ff9f
AV
3743 unsigned long nr;
3744 int ret = queue_var_store(&nr, page, count);
3745 if (nr < BLKDEV_MIN_RQ)
3746 nr = BLKDEV_MIN_RQ;
1da177e4 3747
c981ff9f
AV
3748 spin_lock_irq(q->queue_lock);
3749 q->nr_requests = nr;
1da177e4
LT
3750 blk_queue_congestion_threshold(q);
3751
3752 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3753 set_queue_congested(q, READ);
3754 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3755 clear_queue_congested(q, READ);
3756
3757 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3758 set_queue_congested(q, WRITE);
3759 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3760 clear_queue_congested(q, WRITE);
3761
3762 if (rl->count[READ] >= q->nr_requests) {
3763 blk_set_queue_full(q, READ);
3764 } else if (rl->count[READ]+1 <= q->nr_requests) {
3765 blk_clear_queue_full(q, READ);
3766 wake_up(&rl->wait[READ]);
3767 }
3768
3769 if (rl->count[WRITE] >= q->nr_requests) {
3770 blk_set_queue_full(q, WRITE);
3771 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3772 blk_clear_queue_full(q, WRITE);
3773 wake_up(&rl->wait[WRITE]);
3774 }
c981ff9f 3775 spin_unlock_irq(q->queue_lock);
1da177e4
LT
3776 return ret;
3777}
3778
3779static ssize_t queue_ra_show(struct request_queue *q, char *page)
3780{
3781 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3782
3783 return queue_var_show(ra_kb, (page));
3784}
3785
3786static ssize_t
3787queue_ra_store(struct request_queue *q, const char *page, size_t count)
3788{
3789 unsigned long ra_kb;
3790 ssize_t ret = queue_var_store(&ra_kb, page, count);
3791
3792 spin_lock_irq(q->queue_lock);
3793 if (ra_kb > (q->max_sectors >> 1))
3794 ra_kb = (q->max_sectors >> 1);
3795
3796 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3797 spin_unlock_irq(q->queue_lock);
3798
3799 return ret;
3800}
3801
3802static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3803{
3804 int max_sectors_kb = q->max_sectors >> 1;
3805
3806 return queue_var_show(max_sectors_kb, (page));
3807}
3808
3809static ssize_t
3810queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3811{
3812 unsigned long max_sectors_kb,
3813 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3814 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3815 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3816 int ra_kb;
3817
3818 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3819 return -EINVAL;
3820 /*
3821 * Take the queue lock to update the readahead and max_sectors
3822 * values synchronously:
3823 */
3824 spin_lock_irq(q->queue_lock);
3825 /*
3826 * Trim readahead window as well, if necessary:
3827 */
3828 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3829 if (ra_kb > max_sectors_kb)
3830 q->backing_dev_info.ra_pages =
3831 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3832
3833 q->max_sectors = max_sectors_kb << 1;
3834 spin_unlock_irq(q->queue_lock);
3835
3836 return ret;
3837}
3838
3839static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3840{
3841 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3842
3843 return queue_var_show(max_hw_sectors_kb, (page));
3844}
3845
3846
3847static struct queue_sysfs_entry queue_requests_entry = {
3848 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3849 .show = queue_requests_show,
3850 .store = queue_requests_store,
3851};
3852
3853static struct queue_sysfs_entry queue_ra_entry = {
3854 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3855 .show = queue_ra_show,
3856 .store = queue_ra_store,
3857};
3858
3859static struct queue_sysfs_entry queue_max_sectors_entry = {
3860 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3861 .show = queue_max_sectors_show,
3862 .store = queue_max_sectors_store,
3863};
3864
3865static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3866 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3867 .show = queue_max_hw_sectors_show,
3868};
3869
3870static struct queue_sysfs_entry queue_iosched_entry = {
3871 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3872 .show = elv_iosched_show,
3873 .store = elv_iosched_store,
3874};
3875
3876static struct attribute *default_attrs[] = {
3877 &queue_requests_entry.attr,
3878 &queue_ra_entry.attr,
3879 &queue_max_hw_sectors_entry.attr,
3880 &queue_max_sectors_entry.attr,
3881 &queue_iosched_entry.attr,
3882 NULL,
3883};
3884
3885#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3886
3887static ssize_t
3888queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3889{
3890 struct queue_sysfs_entry *entry = to_queue(attr);
483f4afc
AV
3891 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3892 ssize_t res;
1da177e4 3893
1da177e4 3894 if (!entry->show)
6c1852a0 3895 return -EIO;
483f4afc
AV
3896 mutex_lock(&q->sysfs_lock);
3897 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3898 mutex_unlock(&q->sysfs_lock);
3899 return -ENOENT;
3900 }
3901 res = entry->show(q, page);
3902 mutex_unlock(&q->sysfs_lock);
3903 return res;
1da177e4
LT
3904}
3905
3906static ssize_t
3907queue_attr_store(struct kobject *kobj, struct attribute *attr,
3908 const char *page, size_t length)
3909{
3910 struct queue_sysfs_entry *entry = to_queue(attr);
483f4afc
AV
3911 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3912
3913 ssize_t res;
1da177e4 3914
1da177e4 3915 if (!entry->store)
6c1852a0 3916 return -EIO;
483f4afc
AV
3917 mutex_lock(&q->sysfs_lock);
3918 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3919 mutex_unlock(&q->sysfs_lock);
3920 return -ENOENT;
3921 }
3922 res = entry->store(q, page, length);
3923 mutex_unlock(&q->sysfs_lock);
3924 return res;
1da177e4
LT
3925}
3926
3927static struct sysfs_ops queue_sysfs_ops = {
3928 .show = queue_attr_show,
3929 .store = queue_attr_store,
3930};
3931
93d17d3d 3932static struct kobj_type queue_ktype = {
1da177e4
LT
3933 .sysfs_ops = &queue_sysfs_ops,
3934 .default_attrs = default_attrs,
483f4afc 3935 .release = blk_release_queue,
1da177e4
LT
3936};
3937
3938int blk_register_queue(struct gendisk *disk)
3939{
3940 int ret;
3941
3942 request_queue_t *q = disk->queue;
3943
3944 if (!q || !q->request_fn)
3945 return -ENXIO;
3946
3947 q->kobj.parent = kobject_get(&disk->kobj);
1da177e4 3948
483f4afc 3949 ret = kobject_add(&q->kobj);
1da177e4
LT
3950 if (ret < 0)
3951 return ret;
3952
483f4afc
AV
3953 kobject_uevent(&q->kobj, KOBJ_ADD);
3954
1da177e4
LT
3955 ret = elv_register_queue(q);
3956 if (ret) {
483f4afc
AV
3957 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3958 kobject_del(&q->kobj);
1da177e4
LT
3959 return ret;
3960 }
3961
3962 return 0;
3963}
3964
3965void blk_unregister_queue(struct gendisk *disk)
3966{
3967 request_queue_t *q = disk->queue;
3968
3969 if (q && q->request_fn) {
3970 elv_unregister_queue(q);
3971
483f4afc
AV
3972 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3973 kobject_del(&q->kobj);
1da177e4
LT
3974 kobject_put(&disk->kobj);
3975 }
3976}