blob: ee3a883840f268b1343300ad40368005e2d317dc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Block device elevator/IO-scheduler.
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 *
Jens Axboe0fe23472006-09-04 15:41:16 +02006 * 30042000 Jens Axboe <axboe@kernel.dk> :
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
8 * Split the elevator a bit so that it is possible to choose a different
9 * one or even write a new "plug in". There are three pieces:
10 * - elevator_fn, inserts a new request in the queue list
11 * - elevator_merge_fn, decides whether a new buffer can be merged with
12 * an existing request
13 * - elevator_dequeue_fn, called when a request is taken off the active list
14 *
15 * 20082000 Dave Jones <davej@suse.de> :
16 * Removed tests for max-bomb-segments, which was breaking elvtune
17 * when run without -bN
18 *
19 * Jens:
20 * - Rework again to work with bio instead of buffer_heads
21 * - loose bi_dev comparisons, partition handling is right now
22 * - completely modularize elevator setup and teardown
23 *
24 */
25#include <linux/kernel.h>
26#include <linux/fs.h>
27#include <linux/blkdev.h>
28#include <linux/elevator.h>
29#include <linux/bio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/module.h>
31#include <linux/slab.h>
32#include <linux/init.h>
33#include <linux/compiler.h>
Tejun Heocb98fc82005-10-28 08:29:39 +020034#include <linux/delay.h>
Jens Axboe2056a782006-03-23 20:00:26 +010035#include <linux/blktrace_api.h>
Jens Axboe98170642006-07-28 09:23:08 +020036#include <linux/hash.h>
Jens Axboe0835da62008-08-26 09:15:47 +020037#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Li Zefan55782132009-06-09 13:43:05 +080039#include <trace/events/block.h>
40
Jens Axboe242f9dc2008-09-14 05:55:09 -070041#include "blk.h"
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static DEFINE_SPINLOCK(elv_list_lock);
44static LIST_HEAD(elv_list);
45
46/*
Jens Axboe98170642006-07-28 09:23:08 +020047 * Merge hash stuff.
48 */
49static const int elv_hash_shift = 6;
50#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
Jens Axboe4eb166d2008-02-01 00:37:27 +010051#define ELV_HASH_FN(sec) \
52 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
Jens Axboe98170642006-07-28 09:23:08 +020053#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
Tejun Heo83096eb2009-05-07 22:24:39 +090054#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
Jens Axboe98170642006-07-28 09:23:08 +020055
56/*
Jens Axboeda775262006-12-20 11:04:12 +010057 * Query io scheduler to see if the current process issuing bio may be
58 * merged with rq.
59 */
60static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
61{
Jens Axboe165125e2007-07-24 09:28:11 +020062 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010063 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010064
65 if (e->ops->elevator_allow_merge_fn)
66 return e->ops->elevator_allow_merge_fn(q, rq, bio);
67
68 return 1;
69}
70
71/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * can we safely merge with this request?
73 */
Adrian Bunk72ed0bf2008-04-29 09:49:05 +020074int elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
76 if (!rq_mergeable(rq))
77 return 0;
78
79 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +010080 * Don't merge file system requests and discard requests
81 */
Jens Axboe1f98a132009-09-11 14:32:04 +020082 if (bio_rw_flagged(bio, BIO_RW_DISCARD) !=
83 bio_rw_flagged(rq->bio, BIO_RW_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +010084 return 0;
85
86 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * different data direction or already started, don't merge
88 */
89 if (bio_data_dir(bio) != rq_data_dir(rq))
90 return 0;
91
92 /*
Jens Axboeda775262006-12-20 11:04:12 +010093 * must be same device and not a special request
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Jens Axboebb4067e2006-12-21 21:20:01 +010095 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
Jens Axboeda775262006-12-20 11:04:12 +010096 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020098 /*
99 * only merge integrity protected bio into ditto rq
100 */
101 if (bio_integrity(bio) != blk_integrity_rq(rq))
102 return 0;
103
Jens Axboeda775262006-12-20 11:04:12 +0100104 if (!elv_iosched_allow_merge(rq, bio))
105 return 0;
106
107 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109EXPORT_SYMBOL(elv_rq_merge_ok);
110
Coywolf Qi Hunt769db452005-12-28 10:55:49 +0100111static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
113 int ret = ELEVATOR_NO_MERGE;
114
115 /*
116 * we can merge and sequence is ok, check if it's possible
117 */
118 if (elv_rq_merge_ok(__rq, bio)) {
Tejun Heo83096eb2009-05-07 22:24:39 +0900119 if (blk_rq_pos(__rq) + blk_rq_sectors(__rq) == bio->bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 ret = ELEVATOR_BACK_MERGE;
Tejun Heo83096eb2009-05-07 22:24:39 +0900121 else if (blk_rq_pos(__rq) - bio_sectors(bio) == bio->bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 ret = ELEVATOR_FRONT_MERGE;
123 }
124
125 return ret;
126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128static struct elevator_type *elevator_find(const char *name)
129{
Vasily Tarasova22b1692006-10-11 09:24:27 +0200130 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200132 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +0200133 if (!strcmp(e->elevator_name, name))
134 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Vasily Tarasova22b1692006-10-11 09:24:27 +0200137 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138}
139
140static void elevator_put(struct elevator_type *e)
141{
142 module_put(e->elevator_owner);
143}
144
145static struct elevator_type *elevator_get(const char *name)
146{
Tejun Heo2824bc932005-10-20 10:56:41 +0200147 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200149 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200150
151 e = elevator_find(name);
Jens Axboee1640942008-02-19 10:20:37 +0100152 if (!e) {
153 char elv[ELV_NAME_MAX + strlen("-iosched")];
154
155 spin_unlock(&elv_list_lock);
156
Jens Axboe492af632009-10-03 09:37:51 +0200157 sprintf(elv, "%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100158
maximilian attemse180f592008-07-01 09:42:47 +0200159 request_module("%s", elv);
Jens Axboee1640942008-02-19 10:20:37 +0100160 spin_lock(&elv_list_lock);
161 e = elevator_find(name);
162 }
163
Tejun Heo2824bc932005-10-20 10:56:41 +0200164 if (e && !try_module_get(e->elevator_owner))
165 e = NULL;
166
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200167 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
169 return e;
170}
171
Jens Axboe165125e2007-07-24 09:28:11 +0200172static void *elevator_init_queue(struct request_queue *q,
173 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
Jens Axboebb37b942006-12-01 10:42:33 +0100175 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Jens Axboe165125e2007-07-24 09:28:11 +0200178static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboebc1c1162006-06-08 08:49:06 +0200179 void *data)
180{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200182 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185static char chosen_elevator[16];
186
Nate Diller5f003972006-01-24 10:07:58 +0100187static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100189 /*
190 * Be backwards-compatible with previous kernels, so users
191 * won't get the wrong elevator.
192 */
Jens Axboe492af632009-10-03 09:37:51 +0200193 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800194 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195}
196
197__setup("elevator=", elevator_setup);
198
Al Viro3d1ab402006-03-18 18:35:43 -0500199static struct kobj_type elv_ktype;
200
Jens Axboeb374d182008-10-31 10:05:07 +0100201static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200202 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500203{
Jens Axboeb374d182008-10-31 10:05:07 +0100204 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200205 int i;
206
Jens Axboeb374d182008-10-31 10:05:07 +0100207 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200208 if (unlikely(!eq))
209 goto err;
210
Jens Axboe98170642006-07-28 09:23:08 +0200211 eq->ops = &e->ops;
212 eq->elevator_type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700213 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200214 mutex_init(&eq->sysfs_lock);
215
Jens Axboeb5deef92006-07-19 23:39:40 +0200216 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
217 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200218 if (!eq->hash)
219 goto err;
220
221 for (i = 0; i < ELV_HASH_ENTRIES; i++)
222 INIT_HLIST_HEAD(&eq->hash[i]);
223
Al Viro3d1ab402006-03-18 18:35:43 -0500224 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200225err:
226 kfree(eq);
227 elevator_put(e);
228 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500229}
230
231static void elevator_release(struct kobject *kobj)
232{
Jens Axboeb374d182008-10-31 10:05:07 +0100233 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200234
Jens Axboeb374d182008-10-31 10:05:07 +0100235 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500236 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200237 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500238 kfree(e);
239}
240
Jens Axboe165125e2007-07-24 09:28:11 +0200241int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
243 struct elevator_type *e = NULL;
244 struct elevator_queue *eq;
245 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200246 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Tejun Heocb98fc82005-10-28 08:29:39 +0200248 INIT_LIST_HEAD(&q->queue_head);
249 q->last_merge = NULL;
250 q->end_sector = 0;
251 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200252
Jens Axboe4eb166d2008-02-01 00:37:27 +0100253 if (name) {
254 e = elevator_get(name);
255 if (!e)
256 return -EINVAL;
257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Jens Axboe4eb166d2008-02-01 00:37:27 +0100259 if (!e && *chosen_elevator) {
260 e = elevator_get(chosen_elevator);
261 if (!e)
262 printk(KERN_ERR "I/O scheduler %s not found\n",
263 chosen_elevator);
264 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100265
Jens Axboe4eb166d2008-02-01 00:37:27 +0100266 if (!e) {
267 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
268 if (!e) {
269 printk(KERN_ERR
270 "Default I/O scheduler not found. " \
271 "Using noop.\n");
272 e = elevator_get("noop");
273 }
Nate Diller5f003972006-01-24 10:07:58 +0100274 }
275
Jens Axboeb5deef92006-07-19 23:39:40 +0200276 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500277 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Jens Axboebc1c1162006-06-08 08:49:06 +0200280 data = elevator_init_queue(q, eq);
281 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500282 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200283 return -ENOMEM;
284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Jens Axboebc1c1162006-06-08 08:49:06 +0200286 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return ret;
288}
Jens Axboe2e662b62006-07-13 11:55:04 +0200289EXPORT_SYMBOL(elevator_init);
290
Jens Axboeb374d182008-10-31 10:05:07 +0100291void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
Al Viro3d1ab402006-03-18 18:35:43 -0500293 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 if (e->ops->elevator_exit_fn)
295 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500296 e->ops = NULL;
297 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Al Viro3d1ab402006-03-18 18:35:43 -0500299 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
Jens Axboe2e662b62006-07-13 11:55:04 +0200301EXPORT_SYMBOL(elevator_exit);
302
Jens Axboe98170642006-07-28 09:23:08 +0200303static inline void __elv_rqhash_del(struct request *rq)
304{
305 hlist_del_init(&rq->hash);
306}
307
Jens Axboe165125e2007-07-24 09:28:11 +0200308static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200309{
310 if (ELV_ON_HASH(rq))
311 __elv_rqhash_del(rq);
312}
313
Jens Axboe165125e2007-07-24 09:28:11 +0200314static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200315{
Jens Axboeb374d182008-10-31 10:05:07 +0100316 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200317
318 BUG_ON(ELV_ON_HASH(rq));
319 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
320}
321
Jens Axboe165125e2007-07-24 09:28:11 +0200322static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200323{
324 __elv_rqhash_del(rq);
325 elv_rqhash_add(q, rq);
326}
327
Jens Axboe165125e2007-07-24 09:28:11 +0200328static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200329{
Jens Axboeb374d182008-10-31 10:05:07 +0100330 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200331 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
332 struct hlist_node *entry, *next;
333 struct request *rq;
334
335 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
336 BUG_ON(!ELV_ON_HASH(rq));
337
338 if (unlikely(!rq_mergeable(rq))) {
339 __elv_rqhash_del(rq);
340 continue;
341 }
342
343 if (rq_hash_key(rq) == offset)
344 return rq;
345 }
346
347 return NULL;
348}
349
Tejun Heo8922e162005-10-20 16:23:44 +0200350/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200351 * RB-tree support functions for inserting/lookup/removal of requests
352 * in a sorted RB tree.
353 */
354struct request *elv_rb_add(struct rb_root *root, struct request *rq)
355{
356 struct rb_node **p = &root->rb_node;
357 struct rb_node *parent = NULL;
358 struct request *__rq;
359
360 while (*p) {
361 parent = *p;
362 __rq = rb_entry(parent, struct request, rb_node);
363
Tejun Heo83096eb2009-05-07 22:24:39 +0900364 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200365 p = &(*p)->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900366 else if (blk_rq_pos(rq) > blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200367 p = &(*p)->rb_right;
368 else
369 return __rq;
370 }
371
372 rb_link_node(&rq->rb_node, parent, p);
373 rb_insert_color(&rq->rb_node, root);
374 return NULL;
375}
Jens Axboe2e662b62006-07-13 11:55:04 +0200376EXPORT_SYMBOL(elv_rb_add);
377
378void elv_rb_del(struct rb_root *root, struct request *rq)
379{
380 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
381 rb_erase(&rq->rb_node, root);
382 RB_CLEAR_NODE(&rq->rb_node);
383}
Jens Axboe2e662b62006-07-13 11:55:04 +0200384EXPORT_SYMBOL(elv_rb_del);
385
386struct request *elv_rb_find(struct rb_root *root, sector_t sector)
387{
388 struct rb_node *n = root->rb_node;
389 struct request *rq;
390
391 while (n) {
392 rq = rb_entry(n, struct request, rb_node);
393
Tejun Heo83096eb2009-05-07 22:24:39 +0900394 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200395 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900396 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200397 n = n->rb_right;
398 else
399 return rq;
400 }
401
402 return NULL;
403}
Jens Axboe2e662b62006-07-13 11:55:04 +0200404EXPORT_SYMBOL(elv_rb_find);
405
406/*
Tejun Heo8922e162005-10-20 16:23:44 +0200407 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200408 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200409 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200410 */
Jens Axboe165125e2007-07-24 09:28:11 +0200411void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200412{
413 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200414 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100415 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200416
Tejun Heo06b86242005-10-20 16:46:23 +0200417 if (q->last_merge == rq)
418 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200419
420 elv_rqhash_del(q, rq);
421
Tejun Heo15853af2005-11-10 08:52:05 +0100422 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200423
Jens Axboe1b47f532005-10-20 16:37:00 +0200424 boundary = q->end_sector;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100425 stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200426 list_for_each_prev(entry, &q->queue_head) {
427 struct request *pos = list_entry_rq(entry);
428
David Woodhousee17fc0a2008-08-09 16:42:20 +0100429 if (blk_discard_rq(rq) != blk_discard_rq(pos))
430 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100431 if (rq_data_dir(rq) != rq_data_dir(pos))
432 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100433 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200434 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900435 if (blk_rq_pos(rq) >= boundary) {
436 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200437 continue;
438 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900439 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200440 break;
441 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900442 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200443 break;
444 }
445
446 list_add(&rq->queuelist, entry);
447}
Jens Axboe2e662b62006-07-13 11:55:04 +0200448EXPORT_SYMBOL(elv_dispatch_sort);
449
Jens Axboe98170642006-07-28 09:23:08 +0200450/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200451 * Insert rq into dispatch queue of q. Queue lock must be held on
452 * entry. rq is added to the back of the dispatch queue. To be used by
453 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200454 */
455void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
456{
457 if (q->last_merge == rq)
458 q->last_merge = NULL;
459
460 elv_rqhash_del(q, rq);
461
462 q->nr_sorted--;
463
464 q->end_sector = rq_end_sector(rq);
465 q->boundary_rq = rq;
466 list_add_tail(&rq->queuelist, &q->queue_head);
467}
Jens Axboe2e662b62006-07-13 11:55:04 +0200468EXPORT_SYMBOL(elv_dispatch_add_tail);
469
Jens Axboe165125e2007-07-24 09:28:11 +0200470int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Jens Axboeb374d182008-10-31 10:05:07 +0100472 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200473 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200474 int ret;
475
Jens Axboe98170642006-07-28 09:23:08 +0200476 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100477 * Levels of merges:
478 * nomerges: No merges at all attempted
479 * noxmerges: Only simple one-hit cache try
480 * merges: All merge tries attempted
481 */
482 if (blk_queue_nomerges(q))
483 return ELEVATOR_NO_MERGE;
484
485 /*
Jens Axboe98170642006-07-28 09:23:08 +0200486 * First try one-hit cache.
487 */
Tejun Heo06b86242005-10-20 16:46:23 +0200488 if (q->last_merge) {
489 ret = elv_try_merge(q->last_merge, bio);
490 if (ret != ELEVATOR_NO_MERGE) {
491 *req = q->last_merge;
492 return ret;
493 }
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100496 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200497 return ELEVATOR_NO_MERGE;
498
Jens Axboe98170642006-07-28 09:23:08 +0200499 /*
500 * See if our hash lookup can find a potential backmerge.
501 */
502 __rq = elv_rqhash_find(q, bio->bi_sector);
503 if (__rq && elv_rq_merge_ok(__rq, bio)) {
504 *req = __rq;
505 return ELEVATOR_BACK_MERGE;
506 }
507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (e->ops->elevator_merge_fn)
509 return e->ops->elevator_merge_fn(q, req, bio);
510
511 return ELEVATOR_NO_MERGE;
512}
513
Jens Axboe165125e2007-07-24 09:28:11 +0200514void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515{
Jens Axboeb374d182008-10-31 10:05:07 +0100516 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
518 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200519 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200520
Jens Axboe2e662b62006-07-13 11:55:04 +0200521 if (type == ELEVATOR_BACK_MERGE)
522 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200523
Tejun Heo06b86242005-10-20 16:46:23 +0200524 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
526
Jens Axboe165125e2007-07-24 09:28:11 +0200527void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 struct request *next)
529{
Jens Axboeb374d182008-10-31 10:05:07 +0100530 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (e->ops->elevator_merge_req_fn)
533 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200534
Jens Axboe98170642006-07-28 09:23:08 +0200535 elv_rqhash_reposition(q, rq);
536 elv_rqhash_del(q, next);
537
538 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200539 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
Jens Axboe165125e2007-07-24 09:28:11 +0200542void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 /*
545 * it already went through dequeue, we need to decrement the
546 * in_flight count again
547 */
Tejun Heo8922e162005-10-20 16:23:44 +0200548 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200549 q->in_flight[rq_is_sync(rq)]--;
Jens Axboecad97512007-01-14 22:26:09 +1100550 if (blk_sorted_rq(rq))
551 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Jens Axboe4aff5e22006-08-10 08:44:47 +0200554 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Tejun Heo30e96562006-02-08 01:01:31 -0800556 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
Jerome Marchand26308ea2009-03-27 10:31:51 +0100559void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100560{
561 static int printed;
562 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
563 ;
564 if (q->nr_sorted == 0)
565 return;
566 if (printed++ < 10) {
567 printk(KERN_ERR "%s: forced dispatching is broken "
568 "(nr_sorted=%u), please report this\n",
569 q->elevator->elevator_type->elevator_name, q->nr_sorted);
570 }
571}
572
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100573/*
574 * Call with queue lock held, interrupts disabled
575 */
Jens Axboef600abe2009-04-08 14:22:01 +0200576void elv_quiesce_start(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100577{
Martin K. Petersencd43e262009-05-22 17:17:52 -0400578 if (!q->elevator)
579 return;
580
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100581 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
582
583 /*
584 * make sure we don't have any requests in flight
585 */
586 elv_drain_elevator(q);
587 while (q->rq.elvpriv) {
Tejun Heoa7f55792009-04-23 11:05:17 +0900588 __blk_run_queue(q);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100589 spin_unlock_irq(q->queue_lock);
590 msleep(10);
591 spin_lock_irq(q->queue_lock);
592 elv_drain_elevator(q);
593 }
594}
595
Jens Axboef600abe2009-04-08 14:22:01 +0200596void elv_quiesce_end(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100597{
598 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
599}
600
Jens Axboe165125e2007-07-24 09:28:11 +0200601void elv_insert(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Tejun Heo797e7db2006-01-06 09:51:03 +0100603 struct list_head *pos;
604 unsigned ordseq;
Jens Axboedac07ec2006-05-11 08:20:16 +0200605 int unplug_it = 1;
Tejun Heo797e7db2006-01-06 09:51:03 +0100606
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100607 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 rq->q = q;
610
Tejun Heo8922e162005-10-20 16:23:44 +0200611 switch (where) {
612 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200613 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200614
615 list_add(&rq->queuelist, &q->queue_head);
616 break;
617
618 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200619 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100620 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200621 list_add_tail(&rq->queuelist, &q->queue_head);
622 /*
623 * We kick the queue here for the following reasons.
624 * - The elevator might have returned NULL previously
625 * to delay requests and returned them now. As the
626 * queue wasn't empty before this request, ll_rw_blk
627 * won't run the queue on return, resulting in hang.
628 * - Usually, back inserted requests won't be merged
629 * with anything. There's no point in delaying queue
630 * processing.
631 */
Tejun Heoa7f55792009-04-23 11:05:17 +0900632 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200633 break;
634
635 case ELEVATOR_INSERT_SORT:
David Woodhousee17fc0a2008-08-09 16:42:20 +0100636 BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200637 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100638 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200639 if (rq_mergeable(rq)) {
640 elv_rqhash_add(q, rq);
641 if (!q->last_merge)
642 q->last_merge = rq;
643 }
644
Tejun Heoca235092005-11-01 17:23:49 +0900645 /*
646 * Some ioscheds (cfq) run q->request_fn directly, so
647 * rq cannot be accessed after calling
648 * elevator_add_req_fn.
649 */
650 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200651 break;
652
Tejun Heo797e7db2006-01-06 09:51:03 +0100653 case ELEVATOR_INSERT_REQUEUE:
654 /*
655 * If ordered flush isn't in progress, we do front
656 * insertion; otherwise, requests should be requeued
657 * in ordseq order.
658 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200659 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100660
Linas Vepstas95543172007-01-23 19:40:54 +0100661 /*
662 * Most requeues happen because of a busy condition,
663 * don't force unplug of the queue for that case.
664 */
665 unplug_it = 0;
666
Tejun Heo797e7db2006-01-06 09:51:03 +0100667 if (q->ordseq == 0) {
668 list_add(&rq->queuelist, &q->queue_head);
669 break;
670 }
671
672 ordseq = blk_ordered_req_seq(rq);
673
674 list_for_each(pos, &q->queue_head) {
675 struct request *pos_rq = list_entry_rq(pos);
676 if (ordseq <= blk_ordered_req_seq(pos_rq))
677 break;
678 }
679
680 list_add_tail(&rq->queuelist, pos);
681 break;
682
Tejun Heo8922e162005-10-20 16:23:44 +0200683 default:
684 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700685 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200686 BUG();
687 }
688
Jens Axboedac07ec2006-05-11 08:20:16 +0200689 if (unplug_it && blk_queue_plugged(q)) {
Jens Axboe1faa16d2009-04-06 14:48:01 +0200690 int nrq = q->rq.count[BLK_RW_SYNC] + q->rq.count[BLK_RW_ASYNC]
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200691 - queue_in_flight(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200692
693 if (nrq >= q->unplug_thresh)
694 __generic_unplug_device(q);
695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
Jens Axboe165125e2007-07-24 09:28:11 +0200698void __elv_add_request(struct request_queue *q, struct request *rq, int where,
Tejun Heo30e96562006-02-08 01:01:31 -0800699 int plug)
700{
701 if (q->ordcolor)
Jens Axboe4aff5e22006-08-10 08:44:47 +0200702 rq->cmd_flags |= REQ_ORDERED_COLOR;
Tejun Heo30e96562006-02-08 01:01:31 -0800703
Jens Axboe4aff5e22006-08-10 08:44:47 +0200704 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800705 /*
706 * toggle ordered color
707 */
708 if (blk_barrier_rq(rq))
709 q->ordcolor ^= 1;
710
711 /*
712 * barriers implicitly indicate back insertion
713 */
714 if (where == ELEVATOR_INSERT_SORT)
715 where = ELEVATOR_INSERT_BACK;
716
717 /*
718 * this request is scheduling boundary, update
719 * end_sector
720 */
David Woodhousee17fc0a2008-08-09 16:42:20 +0100721 if (blk_fs_request(rq) || blk_discard_rq(rq)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800722 q->end_sector = rq_end_sector(rq);
723 q->boundary_rq = rq;
724 }
Jens Axboe4eb166d2008-02-01 00:37:27 +0100725 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
726 where == ELEVATOR_INSERT_SORT)
Tejun Heo30e96562006-02-08 01:01:31 -0800727 where = ELEVATOR_INSERT_BACK;
728
729 if (plug)
730 blk_plug_device(q);
731
732 elv_insert(q, rq, where);
733}
Jens Axboe2e662b62006-07-13 11:55:04 +0200734EXPORT_SYMBOL(__elv_add_request);
735
Jens Axboe165125e2007-07-24 09:28:11 +0200736void elv_add_request(struct request_queue *q, struct request *rq, int where,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int plug)
738{
739 unsigned long flags;
740
741 spin_lock_irqsave(q->queue_lock, flags);
742 __elv_add_request(q, rq, where, plug);
743 spin_unlock_irqrestore(q->queue_lock, flags);
744}
Jens Axboe2e662b62006-07-13 11:55:04 +0200745EXPORT_SYMBOL(elv_add_request);
746
Jens Axboe165125e2007-07-24 09:28:11 +0200747int elv_queue_empty(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Jens Axboeb374d182008-10-31 10:05:07 +0100749 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Tejun Heo8922e162005-10-20 16:23:44 +0200751 if (!list_empty(&q->queue_head))
752 return 0;
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 if (e->ops->elevator_queue_empty_fn)
755 return e->ops->elevator_queue_empty_fn(q);
756
Tejun Heo8922e162005-10-20 16:23:44 +0200757 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
Jens Axboe2e662b62006-07-13 11:55:04 +0200759EXPORT_SYMBOL(elv_queue_empty);
760
Jens Axboe165125e2007-07-24 09:28:11 +0200761struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
Jens Axboeb374d182008-10-31 10:05:07 +0100763 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 if (e->ops->elevator_latter_req_fn)
766 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 return NULL;
768}
769
Jens Axboe165125e2007-07-24 09:28:11 +0200770struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
Jens Axboeb374d182008-10-31 10:05:07 +0100772 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774 if (e->ops->elevator_former_req_fn)
775 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 return NULL;
777}
778
Jens Axboe165125e2007-07-24 09:28:11 +0200779int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Jens Axboeb374d182008-10-31 10:05:07 +0100781 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 if (e->ops->elevator_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200784 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
786 rq->elevator_private = NULL;
787 return 0;
788}
789
Jens Axboe165125e2007-07-24 09:28:11 +0200790void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
Jens Axboeb374d182008-10-31 10:05:07 +0100792 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794 if (e->ops->elevator_put_req_fn)
Jens Axboebb37b942006-12-01 10:42:33 +0100795 e->ops->elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
Jens Axboe165125e2007-07-24 09:28:11 +0200798int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Jens Axboeb374d182008-10-31 10:05:07 +0100800 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200803 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805 return ELV_MQUEUE_MAY;
806}
807
Mike Anderson11914a52008-09-13 20:31:27 +0200808void elv_abort_queue(struct request_queue *q)
809{
810 struct request *rq;
811
812 while (!list_empty(&q->queue_head)) {
813 rq = list_entry_rq(q->queue_head.next);
814 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100815 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200816 /*
817 * Mark this request as started so we don't trigger
818 * any debug logic in the end I/O path.
819 */
820 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900821 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200822 }
823}
824EXPORT_SYMBOL(elv_abort_queue);
825
Jens Axboe165125e2007-07-24 09:28:11 +0200826void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
Jens Axboeb374d182008-10-31 10:05:07 +0100828 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 /*
831 * request is released from the driver, io must be done
832 */
Tejun Heo8922e162005-10-20 16:23:44 +0200833 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200834 q->in_flight[rq_is_sync(rq)]--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100835 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
836 e->ops->elevator_completed_req_fn(q, rq);
837 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100838
Tejun Heo1bc691d2006-01-12 15:39:26 +0100839 /*
840 * Check if the queue is waiting for fs requests to be
841 * drained for flush sequence.
842 */
843 if (unlikely(q->ordseq)) {
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900844 struct request *next = NULL;
845
846 if (!list_empty(&q->queue_head))
847 next = list_entry_rq(q->queue_head.next);
848
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200849 if (!queue_in_flight(q) &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100850 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900851 (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
Tejun Heo797e7db2006-01-06 09:51:03 +0100852 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +0900853 __blk_run_queue(q);
Tejun Heo797e7db2006-01-06 09:51:03 +0100854 }
Tejun Heo8922e162005-10-20 16:23:44 +0200855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
Al Viro3d1ab402006-03-18 18:35:43 -0500858#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
859
860static ssize_t
861elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
862{
Al Viro3d1ab402006-03-18 18:35:43 -0500863 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100864 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500865 ssize_t error;
866
867 if (!entry->show)
868 return -EIO;
869
Jens Axboeb374d182008-10-31 10:05:07 +0100870 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500871 mutex_lock(&e->sysfs_lock);
872 error = e->ops ? entry->show(e, page) : -ENOENT;
873 mutex_unlock(&e->sysfs_lock);
874 return error;
875}
876
877static ssize_t
878elv_attr_store(struct kobject *kobj, struct attribute *attr,
879 const char *page, size_t length)
880{
Al Viro3d1ab402006-03-18 18:35:43 -0500881 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100882 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500883 ssize_t error;
884
885 if (!entry->store)
886 return -EIO;
887
Jens Axboeb374d182008-10-31 10:05:07 +0100888 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500889 mutex_lock(&e->sysfs_lock);
890 error = e->ops ? entry->store(e, page, length) : -ENOENT;
891 mutex_unlock(&e->sysfs_lock);
892 return error;
893}
894
895static struct sysfs_ops elv_sysfs_ops = {
896 .show = elv_attr_show,
897 .store = elv_attr_store,
898};
899
900static struct kobj_type elv_ktype = {
901 .sysfs_ops = &elv_sysfs_ops,
902 .release = elevator_release,
903};
904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905int elv_register_queue(struct request_queue *q)
906{
Jens Axboeb374d182008-10-31 10:05:07 +0100907 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500908 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700910 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500911 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500912 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500913 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500914 while (attr->attr.name) {
915 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500916 break;
Al Viroe572ec72006-03-18 22:27:18 -0500917 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500918 }
919 }
920 kobject_uevent(&e->kobj, KOBJ_ADD);
921 }
922 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
Jens Axboeb374d182008-10-31 10:05:07 +0100925static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboebc1c1162006-06-08 08:49:06 +0200926{
927 kobject_uevent(&e->kobj, KOBJ_REMOVE);
928 kobject_del(&e->kobj);
929}
930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931void elv_unregister_queue(struct request_queue *q)
932{
Jens Axboebc1c1162006-06-08 08:49:06 +0200933 if (q)
934 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935}
936
Adrian Bunk2fdd82b2007-12-12 18:51:56 +0100937void elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100939 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200940
941 spin_lock(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100942 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200944 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Nate Diller5f003972006-01-24 10:07:58 +0100946 if (!strcmp(e->elevator_name, chosen_elevator) ||
947 (!*chosen_elevator &&
948 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100949 def = " (default)";
950
Jens Axboe4eb166d2008-02-01 00:37:27 +0100951 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
952 def);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954EXPORT_SYMBOL_GPL(elv_register);
955
956void elv_unregister(struct elevator_type *e)
957{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800958 struct task_struct *g, *p;
959
960 /*
961 * Iterate every thread in the process to remove the io contexts.
962 */
Al Viroe17a9482006-03-18 13:21:20 -0500963 if (e->ops.trim) {
964 read_lock(&tasklist_lock);
965 do_each_thread(g, p) {
966 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400967 if (p->io_context)
968 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500969 task_unlock(p);
970 } while_each_thread(g, p);
971 read_unlock(&tasklist_lock);
972 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800973
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200974 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200976 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978EXPORT_SYMBOL_GPL(elv_unregister);
979
980/*
981 * switch to new_e io scheduler. be careful not to introduce deadlocks -
982 * we don't free the old io scheduler, before we have allocated what we
983 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200984 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 */
Jens Axboe165125e2007-07-24 09:28:11 +0200986static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987{
Jens Axboeb374d182008-10-31 10:05:07 +0100988 struct elevator_queue *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200989 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Tejun Heocb98fc82005-10-28 08:29:39 +0200991 /*
992 * Allocate new elevator
993 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200994 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -0500996 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Jens Axboebc1c1162006-06-08 08:49:06 +0200998 data = elevator_init_queue(q, e);
999 if (!data) {
1000 kobject_put(&e->kobj);
1001 return 0;
1002 }
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001005 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001007 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001008 elv_quiesce_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /*
Jens Axboebc1c1162006-06-08 08:49:06 +02001011 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 old_elevator = q->elevator;
1014
1015 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 * attach and start new elevator
1017 */
Jens Axboebc1c1162006-06-08 08:49:06 +02001018 elevator_attach(q, e, data);
1019
1020 spin_unlock_irq(q->queue_lock);
1021
1022 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 if (elv_register_queue(q))
1025 goto fail_register;
1026
1027 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001028 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 */
1030 elevator_exit(old_elevator);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001031 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001032 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001033 spin_unlock_irq(q->queue_lock);
1034
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001035 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1036
Al Viro3d1ab402006-03-18 18:35:43 -05001037 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
1039fail_register:
1040 /*
1041 * switch failed, exit the new io scheduler and reattach the old
1042 * one again (along with re-adding the sysfs dir)
1043 */
1044 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 q->elevator = old_elevator;
1046 elv_register_queue(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001047
1048 spin_lock_irq(q->queue_lock);
1049 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1050 spin_unlock_irq(q->queue_lock);
1051
Al Viro3d1ab402006-03-18 18:35:43 -05001052 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053}
1054
Jens Axboe165125e2007-07-24 09:28:11 +02001055ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1056 size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
1058 char elevator_name[ELV_NAME_MAX];
1059 struct elevator_type *e;
1060
Martin K. Petersencd43e262009-05-22 17:17:52 -04001061 if (!q->elevator)
1062 return count;
1063
Li Zefanee2e9922008-10-14 08:49:56 +02001064 strlcpy(elevator_name, name, sizeof(elevator_name));
KOSAKI Motohiro8c279592009-10-09 08:48:08 +02001065 e = elevator_get(strstrip(elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (!e) {
1067 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1068 return -EINVAL;
1069 }
1070
Nate Diller2ca7d932005-10-30 15:02:24 -08001071 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1072 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -08001074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
Al Viro3d1ab402006-03-18 18:35:43 -05001076 if (!elevator_switch(q, e))
Jens Axboe4eb166d2008-02-01 00:37:27 +01001077 printk(KERN_ERR "elevator: switch to %s failed\n",
1078 elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return count;
1080}
1081
Jens Axboe165125e2007-07-24 09:28:11 +02001082ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
Jens Axboeb374d182008-10-31 10:05:07 +01001084 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001085 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001086 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 int len = 0;
1088
Martin K. Petersencd43e262009-05-22 17:17:52 -04001089 if (!q->elevator)
1090 return sprintf(name, "none\n");
1091
1092 elv = e->elevator_type;
1093
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001094 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001095 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (!strcmp(elv->elevator_name, __e->elevator_name))
1097 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1098 else
1099 len += sprintf(name+len, "%s ", __e->elevator_name);
1100 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001101 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 len += sprintf(len+name, "\n");
1104 return len;
1105}
1106
Jens Axboe165125e2007-07-24 09:28:11 +02001107struct request *elv_rb_former_request(struct request_queue *q,
1108 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001109{
1110 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1111
1112 if (rbprev)
1113 return rb_entry_rq(rbprev);
1114
1115 return NULL;
1116}
Jens Axboe2e662b62006-07-13 11:55:04 +02001117EXPORT_SYMBOL(elv_rb_former_request);
1118
Jens Axboe165125e2007-07-24 09:28:11 +02001119struct request *elv_rb_latter_request(struct request_queue *q,
1120 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001121{
1122 struct rb_node *rbnext = rb_next(&rq->rb_node);
1123
1124 if (rbnext)
1125 return rb_entry_rq(rbnext);
1126
1127 return NULL;
1128}
Jens Axboe2e662b62006-07-13 11:55:04 +02001129EXPORT_SYMBOL(elv_rb_latter_request);