blob: 1975b619c86d4e7f4212116c7fd3dd11b2a090eb [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
157 if (!strcmp(name, "anticipatory"))
158 sprintf(elv, "as-iosched");
159 else
160 sprintf(elv, "%s-iosched", name);
161
maximilian attemse180f592008-07-01 09:42:47 +0200162 request_module("%s", elv);
Jens Axboee1640942008-02-19 10:20:37 +0100163 spin_lock(&elv_list_lock);
164 e = elevator_find(name);
165 }
166
Tejun Heo2824bc932005-10-20 10:56:41 +0200167 if (e && !try_module_get(e->elevator_owner))
168 e = NULL;
169
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200170 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
172 return e;
173}
174
Jens Axboe165125e2007-07-24 09:28:11 +0200175static void *elevator_init_queue(struct request_queue *q,
176 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
Jens Axboebb37b942006-12-01 10:42:33 +0100178 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200179}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Jens Axboe165125e2007-07-24 09:28:11 +0200181static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboebc1c1162006-06-08 08:49:06 +0200182 void *data)
183{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200185 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
188static char chosen_elevator[16];
189
Nate Diller5f003972006-01-24 10:07:58 +0100190static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100192 /*
193 * Be backwards-compatible with previous kernels, so users
194 * won't get the wrong elevator.
195 */
Nate Diller5f003972006-01-24 10:07:58 +0100196 if (!strcmp(str, "as"))
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100197 strcpy(chosen_elevator, "anticipatory");
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100198 else
Nate Diller5f003972006-01-24 10:07:58 +0100199 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800200 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
203__setup("elevator=", elevator_setup);
204
Al Viro3d1ab402006-03-18 18:35:43 -0500205static struct kobj_type elv_ktype;
206
Jens Axboeb374d182008-10-31 10:05:07 +0100207static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200208 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500209{
Jens Axboeb374d182008-10-31 10:05:07 +0100210 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200211 int i;
212
Jens Axboeb374d182008-10-31 10:05:07 +0100213 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200214 if (unlikely(!eq))
215 goto err;
216
Jens Axboe98170642006-07-28 09:23:08 +0200217 eq->ops = &e->ops;
218 eq->elevator_type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700219 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200220 mutex_init(&eq->sysfs_lock);
221
Jens Axboeb5deef92006-07-19 23:39:40 +0200222 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
223 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200224 if (!eq->hash)
225 goto err;
226
227 for (i = 0; i < ELV_HASH_ENTRIES; i++)
228 INIT_HLIST_HEAD(&eq->hash[i]);
229
Al Viro3d1ab402006-03-18 18:35:43 -0500230 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200231err:
232 kfree(eq);
233 elevator_put(e);
234 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500235}
236
237static void elevator_release(struct kobject *kobj)
238{
Jens Axboeb374d182008-10-31 10:05:07 +0100239 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200240
Jens Axboeb374d182008-10-31 10:05:07 +0100241 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500242 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200243 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500244 kfree(e);
245}
246
Jens Axboe165125e2007-07-24 09:28:11 +0200247int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
249 struct elevator_type *e = NULL;
250 struct elevator_queue *eq;
251 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200252 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Tejun Heocb98fc82005-10-28 08:29:39 +0200254 INIT_LIST_HEAD(&q->queue_head);
255 q->last_merge = NULL;
256 q->end_sector = 0;
257 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200258
Jens Axboe4eb166d2008-02-01 00:37:27 +0100259 if (name) {
260 e = elevator_get(name);
261 if (!e)
262 return -EINVAL;
263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Jens Axboe4eb166d2008-02-01 00:37:27 +0100265 if (!e && *chosen_elevator) {
266 e = elevator_get(chosen_elevator);
267 if (!e)
268 printk(KERN_ERR "I/O scheduler %s not found\n",
269 chosen_elevator);
270 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100271
Jens Axboe4eb166d2008-02-01 00:37:27 +0100272 if (!e) {
273 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
274 if (!e) {
275 printk(KERN_ERR
276 "Default I/O scheduler not found. " \
277 "Using noop.\n");
278 e = elevator_get("noop");
279 }
Nate Diller5f003972006-01-24 10:07:58 +0100280 }
281
Jens Axboeb5deef92006-07-19 23:39:40 +0200282 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500283 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Jens Axboebc1c1162006-06-08 08:49:06 +0200286 data = elevator_init_queue(q, eq);
287 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500288 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200289 return -ENOMEM;
290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
Jens Axboebc1c1162006-06-08 08:49:06 +0200292 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 return ret;
294}
Jens Axboe2e662b62006-07-13 11:55:04 +0200295EXPORT_SYMBOL(elevator_init);
296
Jens Axboeb374d182008-10-31 10:05:07 +0100297void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Al Viro3d1ab402006-03-18 18:35:43 -0500299 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 if (e->ops->elevator_exit_fn)
301 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500302 e->ops = NULL;
303 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Al Viro3d1ab402006-03-18 18:35:43 -0500305 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
Jens Axboe2e662b62006-07-13 11:55:04 +0200307EXPORT_SYMBOL(elevator_exit);
308
Jens Axboe98170642006-07-28 09:23:08 +0200309static inline void __elv_rqhash_del(struct request *rq)
310{
311 hlist_del_init(&rq->hash);
312}
313
Jens Axboe165125e2007-07-24 09:28:11 +0200314static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200315{
316 if (ELV_ON_HASH(rq))
317 __elv_rqhash_del(rq);
318}
319
Jens Axboe165125e2007-07-24 09:28:11 +0200320static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200321{
Jens Axboeb374d182008-10-31 10:05:07 +0100322 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200323
324 BUG_ON(ELV_ON_HASH(rq));
325 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
326}
327
Jens Axboe165125e2007-07-24 09:28:11 +0200328static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200329{
330 __elv_rqhash_del(rq);
331 elv_rqhash_add(q, rq);
332}
333
Jens Axboe165125e2007-07-24 09:28:11 +0200334static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200335{
Jens Axboeb374d182008-10-31 10:05:07 +0100336 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200337 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
338 struct hlist_node *entry, *next;
339 struct request *rq;
340
341 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
342 BUG_ON(!ELV_ON_HASH(rq));
343
344 if (unlikely(!rq_mergeable(rq))) {
345 __elv_rqhash_del(rq);
346 continue;
347 }
348
349 if (rq_hash_key(rq) == offset)
350 return rq;
351 }
352
353 return NULL;
354}
355
Tejun Heo8922e162005-10-20 16:23:44 +0200356/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200357 * RB-tree support functions for inserting/lookup/removal of requests
358 * in a sorted RB tree.
359 */
360struct request *elv_rb_add(struct rb_root *root, struct request *rq)
361{
362 struct rb_node **p = &root->rb_node;
363 struct rb_node *parent = NULL;
364 struct request *__rq;
365
366 while (*p) {
367 parent = *p;
368 __rq = rb_entry(parent, struct request, rb_node);
369
Tejun Heo83096eb2009-05-07 22:24:39 +0900370 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200371 p = &(*p)->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900372 else if (blk_rq_pos(rq) > blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200373 p = &(*p)->rb_right;
374 else
375 return __rq;
376 }
377
378 rb_link_node(&rq->rb_node, parent, p);
379 rb_insert_color(&rq->rb_node, root);
380 return NULL;
381}
Jens Axboe2e662b62006-07-13 11:55:04 +0200382EXPORT_SYMBOL(elv_rb_add);
383
384void elv_rb_del(struct rb_root *root, struct request *rq)
385{
386 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
387 rb_erase(&rq->rb_node, root);
388 RB_CLEAR_NODE(&rq->rb_node);
389}
Jens Axboe2e662b62006-07-13 11:55:04 +0200390EXPORT_SYMBOL(elv_rb_del);
391
392struct request *elv_rb_find(struct rb_root *root, sector_t sector)
393{
394 struct rb_node *n = root->rb_node;
395 struct request *rq;
396
397 while (n) {
398 rq = rb_entry(n, struct request, rb_node);
399
Tejun Heo83096eb2009-05-07 22:24:39 +0900400 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200401 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900402 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200403 n = n->rb_right;
404 else
405 return rq;
406 }
407
408 return NULL;
409}
Jens Axboe2e662b62006-07-13 11:55:04 +0200410EXPORT_SYMBOL(elv_rb_find);
411
412/*
Tejun Heo8922e162005-10-20 16:23:44 +0200413 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200414 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200415 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200416 */
Jens Axboe165125e2007-07-24 09:28:11 +0200417void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200418{
419 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200420 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100421 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200422
Tejun Heo06b86242005-10-20 16:46:23 +0200423 if (q->last_merge == rq)
424 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200425
426 elv_rqhash_del(q, rq);
427
Tejun Heo15853af2005-11-10 08:52:05 +0100428 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200429
Jens Axboe1b47f532005-10-20 16:37:00 +0200430 boundary = q->end_sector;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100431 stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200432 list_for_each_prev(entry, &q->queue_head) {
433 struct request *pos = list_entry_rq(entry);
434
David Woodhousee17fc0a2008-08-09 16:42:20 +0100435 if (blk_discard_rq(rq) != blk_discard_rq(pos))
436 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100437 if (rq_data_dir(rq) != rq_data_dir(pos))
438 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100439 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200440 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900441 if (blk_rq_pos(rq) >= boundary) {
442 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200443 continue;
444 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900445 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200446 break;
447 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900448 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200449 break;
450 }
451
452 list_add(&rq->queuelist, entry);
453}
Jens Axboe2e662b62006-07-13 11:55:04 +0200454EXPORT_SYMBOL(elv_dispatch_sort);
455
Jens Axboe98170642006-07-28 09:23:08 +0200456/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200457 * Insert rq into dispatch queue of q. Queue lock must be held on
458 * entry. rq is added to the back of the dispatch queue. To be used by
459 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200460 */
461void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
462{
463 if (q->last_merge == rq)
464 q->last_merge = NULL;
465
466 elv_rqhash_del(q, rq);
467
468 q->nr_sorted--;
469
470 q->end_sector = rq_end_sector(rq);
471 q->boundary_rq = rq;
472 list_add_tail(&rq->queuelist, &q->queue_head);
473}
Jens Axboe2e662b62006-07-13 11:55:04 +0200474EXPORT_SYMBOL(elv_dispatch_add_tail);
475
Jens Axboe165125e2007-07-24 09:28:11 +0200476int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Jens Axboeb374d182008-10-31 10:05:07 +0100478 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200479 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200480 int ret;
481
Jens Axboe98170642006-07-28 09:23:08 +0200482 /*
483 * First try one-hit cache.
484 */
Tejun Heo06b86242005-10-20 16:46:23 +0200485 if (q->last_merge) {
486 ret = elv_try_merge(q->last_merge, bio);
487 if (ret != ELEVATOR_NO_MERGE) {
488 *req = q->last_merge;
489 return ret;
490 }
491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200493 if (blk_queue_nomerges(q))
494 return ELEVATOR_NO_MERGE;
495
Jens Axboe98170642006-07-28 09:23:08 +0200496 /*
497 * See if our hash lookup can find a potential backmerge.
498 */
499 __rq = elv_rqhash_find(q, bio->bi_sector);
500 if (__rq && elv_rq_merge_ok(__rq, bio)) {
501 *req = __rq;
502 return ELEVATOR_BACK_MERGE;
503 }
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 if (e->ops->elevator_merge_fn)
506 return e->ops->elevator_merge_fn(q, req, bio);
507
508 return ELEVATOR_NO_MERGE;
509}
510
Jens Axboe165125e2007-07-24 09:28:11 +0200511void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512{
Jens Axboeb374d182008-10-31 10:05:07 +0100513 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
515 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200516 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200517
Jens Axboe2e662b62006-07-13 11:55:04 +0200518 if (type == ELEVATOR_BACK_MERGE)
519 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200520
Tejun Heo06b86242005-10-20 16:46:23 +0200521 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
Jens Axboe165125e2007-07-24 09:28:11 +0200524void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 struct request *next)
526{
Jens Axboeb374d182008-10-31 10:05:07 +0100527 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 if (e->ops->elevator_merge_req_fn)
530 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200531
Jens Axboe98170642006-07-28 09:23:08 +0200532 elv_rqhash_reposition(q, rq);
533 elv_rqhash_del(q, next);
534
535 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200536 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
Jens Axboe165125e2007-07-24 09:28:11 +0200539void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 /*
542 * it already went through dequeue, we need to decrement the
543 * in_flight count again
544 */
Tejun Heo8922e162005-10-20 16:23:44 +0200545 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200546 q->in_flight[rq_is_sync(rq)]--;
Jens Axboecad97512007-01-14 22:26:09 +1100547 if (blk_sorted_rq(rq))
548 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Jens Axboe4aff5e22006-08-10 08:44:47 +0200551 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Tejun Heo30e96562006-02-08 01:01:31 -0800553 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
555
Jerome Marchand26308ea2009-03-27 10:31:51 +0100556void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100557{
558 static int printed;
559 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
560 ;
561 if (q->nr_sorted == 0)
562 return;
563 if (printed++ < 10) {
564 printk(KERN_ERR "%s: forced dispatching is broken "
565 "(nr_sorted=%u), please report this\n",
566 q->elevator->elevator_type->elevator_name, q->nr_sorted);
567 }
568}
569
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100570/*
571 * Call with queue lock held, interrupts disabled
572 */
Jens Axboef600abe2009-04-08 14:22:01 +0200573void elv_quiesce_start(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100574{
Martin K. Petersencd43e262009-05-22 17:17:52 -0400575 if (!q->elevator)
576 return;
577
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100578 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
579
580 /*
581 * make sure we don't have any requests in flight
582 */
583 elv_drain_elevator(q);
584 while (q->rq.elvpriv) {
Tejun Heoa7f55792009-04-23 11:05:17 +0900585 __blk_run_queue(q);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100586 spin_unlock_irq(q->queue_lock);
587 msleep(10);
588 spin_lock_irq(q->queue_lock);
589 elv_drain_elevator(q);
590 }
591}
592
Jens Axboef600abe2009-04-08 14:22:01 +0200593void elv_quiesce_end(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100594{
595 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
596}
597
Jens Axboe165125e2007-07-24 09:28:11 +0200598void elv_insert(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
Tejun Heo797e7db2006-01-06 09:51:03 +0100600 struct list_head *pos;
601 unsigned ordseq;
Jens Axboedac07ec2006-05-11 08:20:16 +0200602 int unplug_it = 1;
Tejun Heo797e7db2006-01-06 09:51:03 +0100603
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100604 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100605
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 rq->q = q;
607
Tejun Heo8922e162005-10-20 16:23:44 +0200608 switch (where) {
609 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200610 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200611
612 list_add(&rq->queuelist, &q->queue_head);
613 break;
614
615 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200616 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100617 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200618 list_add_tail(&rq->queuelist, &q->queue_head);
619 /*
620 * We kick the queue here for the following reasons.
621 * - The elevator might have returned NULL previously
622 * to delay requests and returned them now. As the
623 * queue wasn't empty before this request, ll_rw_blk
624 * won't run the queue on return, resulting in hang.
625 * - Usually, back inserted requests won't be merged
626 * with anything. There's no point in delaying queue
627 * processing.
628 */
Tejun Heoa7f55792009-04-23 11:05:17 +0900629 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200630 break;
631
632 case ELEVATOR_INSERT_SORT:
David Woodhousee17fc0a2008-08-09 16:42:20 +0100633 BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200634 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100635 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200636 if (rq_mergeable(rq)) {
637 elv_rqhash_add(q, rq);
638 if (!q->last_merge)
639 q->last_merge = rq;
640 }
641
Tejun Heoca235092005-11-01 17:23:49 +0900642 /*
643 * Some ioscheds (cfq) run q->request_fn directly, so
644 * rq cannot be accessed after calling
645 * elevator_add_req_fn.
646 */
647 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200648 break;
649
Tejun Heo797e7db2006-01-06 09:51:03 +0100650 case ELEVATOR_INSERT_REQUEUE:
651 /*
652 * If ordered flush isn't in progress, we do front
653 * insertion; otherwise, requests should be requeued
654 * in ordseq order.
655 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200656 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100657
Linas Vepstas95543172007-01-23 19:40:54 +0100658 /*
659 * Most requeues happen because of a busy condition,
660 * don't force unplug of the queue for that case.
661 */
662 unplug_it = 0;
663
Tejun Heo797e7db2006-01-06 09:51:03 +0100664 if (q->ordseq == 0) {
665 list_add(&rq->queuelist, &q->queue_head);
666 break;
667 }
668
669 ordseq = blk_ordered_req_seq(rq);
670
671 list_for_each(pos, &q->queue_head) {
672 struct request *pos_rq = list_entry_rq(pos);
673 if (ordseq <= blk_ordered_req_seq(pos_rq))
674 break;
675 }
676
677 list_add_tail(&rq->queuelist, pos);
678 break;
679
Tejun Heo8922e162005-10-20 16:23:44 +0200680 default:
681 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700682 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200683 BUG();
684 }
685
Jens Axboedac07ec2006-05-11 08:20:16 +0200686 if (unplug_it && blk_queue_plugged(q)) {
Jens Axboe1faa16d2009-04-06 14:48:01 +0200687 int nrq = q->rq.count[BLK_RW_SYNC] + q->rq.count[BLK_RW_ASYNC]
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200688 - queue_in_flight(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200689
690 if (nrq >= q->unplug_thresh)
691 __generic_unplug_device(q);
692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Jens Axboe165125e2007-07-24 09:28:11 +0200695void __elv_add_request(struct request_queue *q, struct request *rq, int where,
Tejun Heo30e96562006-02-08 01:01:31 -0800696 int plug)
697{
698 if (q->ordcolor)
Jens Axboe4aff5e22006-08-10 08:44:47 +0200699 rq->cmd_flags |= REQ_ORDERED_COLOR;
Tejun Heo30e96562006-02-08 01:01:31 -0800700
Jens Axboe4aff5e22006-08-10 08:44:47 +0200701 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800702 /*
703 * toggle ordered color
704 */
705 if (blk_barrier_rq(rq))
706 q->ordcolor ^= 1;
707
708 /*
709 * barriers implicitly indicate back insertion
710 */
711 if (where == ELEVATOR_INSERT_SORT)
712 where = ELEVATOR_INSERT_BACK;
713
714 /*
715 * this request is scheduling boundary, update
716 * end_sector
717 */
David Woodhousee17fc0a2008-08-09 16:42:20 +0100718 if (blk_fs_request(rq) || blk_discard_rq(rq)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800719 q->end_sector = rq_end_sector(rq);
720 q->boundary_rq = rq;
721 }
Jens Axboe4eb166d2008-02-01 00:37:27 +0100722 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
723 where == ELEVATOR_INSERT_SORT)
Tejun Heo30e96562006-02-08 01:01:31 -0800724 where = ELEVATOR_INSERT_BACK;
725
726 if (plug)
727 blk_plug_device(q);
728
729 elv_insert(q, rq, where);
730}
Jens Axboe2e662b62006-07-13 11:55:04 +0200731EXPORT_SYMBOL(__elv_add_request);
732
Jens Axboe165125e2007-07-24 09:28:11 +0200733void elv_add_request(struct request_queue *q, struct request *rq, int where,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 int plug)
735{
736 unsigned long flags;
737
738 spin_lock_irqsave(q->queue_lock, flags);
739 __elv_add_request(q, rq, where, plug);
740 spin_unlock_irqrestore(q->queue_lock, flags);
741}
Jens Axboe2e662b62006-07-13 11:55:04 +0200742EXPORT_SYMBOL(elv_add_request);
743
Jens Axboe165125e2007-07-24 09:28:11 +0200744int elv_queue_empty(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Jens Axboeb374d182008-10-31 10:05:07 +0100746 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Tejun Heo8922e162005-10-20 16:23:44 +0200748 if (!list_empty(&q->queue_head))
749 return 0;
750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 if (e->ops->elevator_queue_empty_fn)
752 return e->ops->elevator_queue_empty_fn(q);
753
Tejun Heo8922e162005-10-20 16:23:44 +0200754 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755}
Jens Axboe2e662b62006-07-13 11:55:04 +0200756EXPORT_SYMBOL(elv_queue_empty);
757
Jens Axboe165125e2007-07-24 09:28:11 +0200758struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Jens Axboeb374d182008-10-31 10:05:07 +0100760 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
762 if (e->ops->elevator_latter_req_fn)
763 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return NULL;
765}
766
Jens Axboe165125e2007-07-24 09:28:11 +0200767struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Jens Axboeb374d182008-10-31 10:05:07 +0100769 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 if (e->ops->elevator_former_req_fn)
772 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 return NULL;
774}
775
Jens Axboe165125e2007-07-24 09:28:11 +0200776int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Jens Axboeb374d182008-10-31 10:05:07 +0100778 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 if (e->ops->elevator_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200781 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 rq->elevator_private = NULL;
784 return 0;
785}
786
Jens Axboe165125e2007-07-24 09:28:11 +0200787void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Jens Axboeb374d182008-10-31 10:05:07 +0100789 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 if (e->ops->elevator_put_req_fn)
Jens Axboebb37b942006-12-01 10:42:33 +0100792 e->ops->elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
Jens Axboe165125e2007-07-24 09:28:11 +0200795int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
Jens Axboeb374d182008-10-31 10:05:07 +0100797 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
799 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200800 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 return ELV_MQUEUE_MAY;
803}
804
Mike Anderson11914a52008-09-13 20:31:27 +0200805void elv_abort_queue(struct request_queue *q)
806{
807 struct request *rq;
808
809 while (!list_empty(&q->queue_head)) {
810 rq = list_entry_rq(q->queue_head.next);
811 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100812 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200813 /*
814 * Mark this request as started so we don't trigger
815 * any debug logic in the end I/O path.
816 */
817 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900818 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200819 }
820}
821EXPORT_SYMBOL(elv_abort_queue);
822
Jens Axboe165125e2007-07-24 09:28:11 +0200823void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824{
Jens Axboeb374d182008-10-31 10:05:07 +0100825 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827 /*
828 * request is released from the driver, io must be done
829 */
Tejun Heo8922e162005-10-20 16:23:44 +0200830 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200831 q->in_flight[rq_is_sync(rq)]--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100832 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
833 e->ops->elevator_completed_req_fn(q, rq);
834 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100835
Tejun Heo1bc691d2006-01-12 15:39:26 +0100836 /*
837 * Check if the queue is waiting for fs requests to be
838 * drained for flush sequence.
839 */
840 if (unlikely(q->ordseq)) {
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900841 struct request *next = NULL;
842
843 if (!list_empty(&q->queue_head))
844 next = list_entry_rq(q->queue_head.next);
845
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200846 if (!queue_in_flight(q) &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100847 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900848 (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
Tejun Heo797e7db2006-01-06 09:51:03 +0100849 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +0900850 __blk_run_queue(q);
Tejun Heo797e7db2006-01-06 09:51:03 +0100851 }
Tejun Heo8922e162005-10-20 16:23:44 +0200852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
854
Al Viro3d1ab402006-03-18 18:35:43 -0500855#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
856
857static ssize_t
858elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
859{
Al Viro3d1ab402006-03-18 18:35:43 -0500860 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100861 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500862 ssize_t error;
863
864 if (!entry->show)
865 return -EIO;
866
Jens Axboeb374d182008-10-31 10:05:07 +0100867 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500868 mutex_lock(&e->sysfs_lock);
869 error = e->ops ? entry->show(e, page) : -ENOENT;
870 mutex_unlock(&e->sysfs_lock);
871 return error;
872}
873
874static ssize_t
875elv_attr_store(struct kobject *kobj, struct attribute *attr,
876 const char *page, size_t length)
877{
Al Viro3d1ab402006-03-18 18:35:43 -0500878 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100879 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500880 ssize_t error;
881
882 if (!entry->store)
883 return -EIO;
884
Jens Axboeb374d182008-10-31 10:05:07 +0100885 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500886 mutex_lock(&e->sysfs_lock);
887 error = e->ops ? entry->store(e, page, length) : -ENOENT;
888 mutex_unlock(&e->sysfs_lock);
889 return error;
890}
891
892static struct sysfs_ops elv_sysfs_ops = {
893 .show = elv_attr_show,
894 .store = elv_attr_store,
895};
896
897static struct kobj_type elv_ktype = {
898 .sysfs_ops = &elv_sysfs_ops,
899 .release = elevator_release,
900};
901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902int elv_register_queue(struct request_queue *q)
903{
Jens Axboeb374d182008-10-31 10:05:07 +0100904 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500905 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700907 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500908 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500909 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500910 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500911 while (attr->attr.name) {
912 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500913 break;
Al Viroe572ec72006-03-18 22:27:18 -0500914 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500915 }
916 }
917 kobject_uevent(&e->kobj, KOBJ_ADD);
918 }
919 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920}
921
Jens Axboeb374d182008-10-31 10:05:07 +0100922static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboebc1c1162006-06-08 08:49:06 +0200923{
924 kobject_uevent(&e->kobj, KOBJ_REMOVE);
925 kobject_del(&e->kobj);
926}
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928void elv_unregister_queue(struct request_queue *q)
929{
Jens Axboebc1c1162006-06-08 08:49:06 +0200930 if (q)
931 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932}
933
Adrian Bunk2fdd82b2007-12-12 18:51:56 +0100934void elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100936 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200937
938 spin_lock(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100939 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200941 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Nate Diller5f003972006-01-24 10:07:58 +0100943 if (!strcmp(e->elevator_name, chosen_elevator) ||
944 (!*chosen_elevator &&
945 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100946 def = " (default)";
947
Jens Axboe4eb166d2008-02-01 00:37:27 +0100948 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
949 def);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950}
951EXPORT_SYMBOL_GPL(elv_register);
952
953void elv_unregister(struct elevator_type *e)
954{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800955 struct task_struct *g, *p;
956
957 /*
958 * Iterate every thread in the process to remove the io contexts.
959 */
Al Viroe17a9482006-03-18 13:21:20 -0500960 if (e->ops.trim) {
961 read_lock(&tasklist_lock);
962 do_each_thread(g, p) {
963 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400964 if (p->io_context)
965 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500966 task_unlock(p);
967 } while_each_thread(g, p);
968 read_unlock(&tasklist_lock);
969 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800970
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200971 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200973 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
975EXPORT_SYMBOL_GPL(elv_unregister);
976
977/*
978 * switch to new_e io scheduler. be careful not to introduce deadlocks -
979 * we don't free the old io scheduler, before we have allocated what we
980 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200981 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 */
Jens Axboe165125e2007-07-24 09:28:11 +0200983static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Jens Axboeb374d182008-10-31 10:05:07 +0100985 struct elevator_queue *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200986 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Tejun Heocb98fc82005-10-28 08:29:39 +0200988 /*
989 * Allocate new elevator
990 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200991 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -0500993 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Jens Axboebc1c1162006-06-08 08:49:06 +0200995 data = elevator_init_queue(q, e);
996 if (!data) {
997 kobject_put(&e->kobj);
998 return 0;
999 }
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001002 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001004 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001005 elv_quiesce_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 /*
Jens Axboebc1c1162006-06-08 08:49:06 +02001008 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 old_elevator = q->elevator;
1011
1012 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 * attach and start new elevator
1014 */
Jens Axboebc1c1162006-06-08 08:49:06 +02001015 elevator_attach(q, e, data);
1016
1017 spin_unlock_irq(q->queue_lock);
1018
1019 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 if (elv_register_queue(q))
1022 goto fail_register;
1023
1024 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001025 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 */
1027 elevator_exit(old_elevator);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001028 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001029 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001030 spin_unlock_irq(q->queue_lock);
1031
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001032 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1033
Al Viro3d1ab402006-03-18 18:35:43 -05001034 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
1036fail_register:
1037 /*
1038 * switch failed, exit the new io scheduler and reattach the old
1039 * one again (along with re-adding the sysfs dir)
1040 */
1041 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 q->elevator = old_elevator;
1043 elv_register_queue(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001044
1045 spin_lock_irq(q->queue_lock);
1046 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1047 spin_unlock_irq(q->queue_lock);
1048
Al Viro3d1ab402006-03-18 18:35:43 -05001049 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
Jens Axboe165125e2007-07-24 09:28:11 +02001052ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1053 size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
1055 char elevator_name[ELV_NAME_MAX];
1056 struct elevator_type *e;
1057
Martin K. Petersencd43e262009-05-22 17:17:52 -04001058 if (!q->elevator)
1059 return count;
1060
Li Zefanee2e9922008-10-14 08:49:56 +02001061 strlcpy(elevator_name, name, sizeof(elevator_name));
1062 strstrip(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
1064 e = elevator_get(elevator_name);
1065 if (!e) {
1066 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1067 return -EINVAL;
1068 }
1069
Nate Diller2ca7d932005-10-30 15:02:24 -08001070 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1071 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -08001073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Al Viro3d1ab402006-03-18 18:35:43 -05001075 if (!elevator_switch(q, e))
Jens Axboe4eb166d2008-02-01 00:37:27 +01001076 printk(KERN_ERR "elevator: switch to %s failed\n",
1077 elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 return count;
1079}
1080
Jens Axboe165125e2007-07-24 09:28:11 +02001081ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082{
Jens Axboeb374d182008-10-31 10:05:07 +01001083 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001084 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001085 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 int len = 0;
1087
Martin K. Petersencd43e262009-05-22 17:17:52 -04001088 if (!q->elevator)
1089 return sprintf(name, "none\n");
1090
1091 elv = e->elevator_type;
1092
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001093 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001094 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 if (!strcmp(elv->elevator_name, __e->elevator_name))
1096 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1097 else
1098 len += sprintf(name+len, "%s ", __e->elevator_name);
1099 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001100 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 len += sprintf(len+name, "\n");
1103 return len;
1104}
1105
Jens Axboe165125e2007-07-24 09:28:11 +02001106struct request *elv_rb_former_request(struct request_queue *q,
1107 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001108{
1109 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1110
1111 if (rbprev)
1112 return rb_entry_rq(rbprev);
1113
1114 return NULL;
1115}
Jens Axboe2e662b62006-07-13 11:55:04 +02001116EXPORT_SYMBOL(elv_rb_former_request);
1117
Jens Axboe165125e2007-07-24 09:28:11 +02001118struct request *elv_rb_latter_request(struct request_queue *q,
1119 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001120{
1121 struct rb_node *rbnext = rb_next(&rq->rb_node);
1122
1123 if (rbnext)
1124 return rb_entry_rq(rbnext);
1125
1126 return NULL;
1127}
Jens Axboe2e662b62006-07-13 11:55:04 +02001128EXPORT_SYMBOL(elv_rb_latter_request);