blob: 1af5d9f04affe47f187d59df37a287f3f5ad8071 [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>
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010036#include <trace/block.h>
Jens Axboe98170642006-07-28 09:23:08 +020037#include <linux/hash.h>
Jens Axboe0835da62008-08-26 09:15:47 +020038#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Jens Axboe242f9dc2008-09-14 05:55:09 -070040#include "blk.h"
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042static DEFINE_SPINLOCK(elv_list_lock);
43static LIST_HEAD(elv_list);
44
Ingo Molnar0bfc2452008-11-26 11:59:56 +010045DEFINE_TRACE(block_rq_abort);
46
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
Jens Axboe98170642006-07-28 09:23:08 +020048 * Merge hash stuff.
49 */
50static const int elv_hash_shift = 6;
51#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
Jens Axboe4eb166d2008-02-01 00:37:27 +010052#define ELV_HASH_FN(sec) \
53 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
Jens Axboe98170642006-07-28 09:23:08 +020054#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
55#define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors)
Jens Axboe98170642006-07-28 09:23:08 +020056
Ingo Molnar0bfc2452008-11-26 11:59:56 +010057DEFINE_TRACE(block_rq_insert);
58DEFINE_TRACE(block_rq_issue);
59
Jens Axboe98170642006-07-28 09:23:08 +020060/*
Jens Axboeda775262006-12-20 11:04:12 +010061 * Query io scheduler to see if the current process issuing bio may be
62 * merged with rq.
63 */
64static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
65{
Jens Axboe165125e2007-07-24 09:28:11 +020066 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010067 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010068
69 if (e->ops->elevator_allow_merge_fn)
70 return e->ops->elevator_allow_merge_fn(q, rq, bio);
71
72 return 1;
73}
74
75/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * can we safely merge with this request?
77 */
Adrian Bunk72ed0bf2008-04-29 09:49:05 +020078int elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
80 if (!rq_mergeable(rq))
81 return 0;
82
83 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +010084 * Don't merge file system requests and discard requests
85 */
86 if (bio_discard(bio) != bio_discard(rq->bio))
87 return 0;
88
89 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 * different data direction or already started, don't merge
91 */
92 if (bio_data_dir(bio) != rq_data_dir(rq))
93 return 0;
94
95 /*
Jens Axboeda775262006-12-20 11:04:12 +010096 * must be same device and not a special request
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 */
Jens Axboebb4067e2006-12-21 21:20:01 +010098 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
Jens Axboeda775262006-12-20 11:04:12 +010099 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200101 /*
102 * only merge integrity protected bio into ditto rq
103 */
104 if (bio_integrity(bio) != blk_integrity_rq(rq))
105 return 0;
106
Jens Axboeda775262006-12-20 11:04:12 +0100107 if (!elv_iosched_allow_merge(rq, bio))
108 return 0;
109
110 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111}
112EXPORT_SYMBOL(elv_rq_merge_ok);
113
Coywolf Qi Hunt769db452005-12-28 10:55:49 +0100114static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
116 int ret = ELEVATOR_NO_MERGE;
117
118 /*
119 * we can merge and sequence is ok, check if it's possible
120 */
121 if (elv_rq_merge_ok(__rq, bio)) {
122 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
123 ret = ELEVATOR_BACK_MERGE;
124 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
125 ret = ELEVATOR_FRONT_MERGE;
126 }
127
128 return ret;
129}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131static struct elevator_type *elevator_find(const char *name)
132{
Vasily Tarasova22b1692006-10-11 09:24:27 +0200133 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200135 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +0200136 if (!strcmp(e->elevator_name, name))
137 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Vasily Tarasova22b1692006-10-11 09:24:27 +0200140 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
143static void elevator_put(struct elevator_type *e)
144{
145 module_put(e->elevator_owner);
146}
147
148static struct elevator_type *elevator_get(const char *name)
149{
Tejun Heo2824bc932005-10-20 10:56:41 +0200150 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200152 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200153
154 e = elevator_find(name);
Jens Axboee1640942008-02-19 10:20:37 +0100155 if (!e) {
156 char elv[ELV_NAME_MAX + strlen("-iosched")];
157
158 spin_unlock(&elv_list_lock);
159
160 if (!strcmp(name, "anticipatory"))
161 sprintf(elv, "as-iosched");
162 else
163 sprintf(elv, "%s-iosched", name);
164
maximilian attemse180f592008-07-01 09:42:47 +0200165 request_module("%s", elv);
Jens Axboee1640942008-02-19 10:20:37 +0100166 spin_lock(&elv_list_lock);
167 e = elevator_find(name);
168 }
169
Tejun Heo2824bc932005-10-20 10:56:41 +0200170 if (e && !try_module_get(e->elevator_owner))
171 e = NULL;
172
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200173 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175 return e;
176}
177
Jens Axboe165125e2007-07-24 09:28:11 +0200178static void *elevator_init_queue(struct request_queue *q,
179 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Jens Axboebb37b942006-12-01 10:42:33 +0100181 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200182}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Jens Axboe165125e2007-07-24 09:28:11 +0200184static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboebc1c1162006-06-08 08:49:06 +0200185 void *data)
186{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200188 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190
191static char chosen_elevator[16];
192
Nate Diller5f003972006-01-24 10:07:58 +0100193static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100195 /*
196 * Be backwards-compatible with previous kernels, so users
197 * won't get the wrong elevator.
198 */
Nate Diller5f003972006-01-24 10:07:58 +0100199 if (!strcmp(str, "as"))
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100200 strcpy(chosen_elevator, "anticipatory");
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100201 else
Nate Diller5f003972006-01-24 10:07:58 +0100202 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800203 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
205
206__setup("elevator=", elevator_setup);
207
Al Viro3d1ab402006-03-18 18:35:43 -0500208static struct kobj_type elv_ktype;
209
Jens Axboeb374d182008-10-31 10:05:07 +0100210static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200211 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500212{
Jens Axboeb374d182008-10-31 10:05:07 +0100213 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200214 int i;
215
Jens Axboeb374d182008-10-31 10:05:07 +0100216 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200217 if (unlikely(!eq))
218 goto err;
219
Jens Axboe98170642006-07-28 09:23:08 +0200220 eq->ops = &e->ops;
221 eq->elevator_type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700222 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200223 mutex_init(&eq->sysfs_lock);
224
Jens Axboeb5deef92006-07-19 23:39:40 +0200225 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
226 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200227 if (!eq->hash)
228 goto err;
229
230 for (i = 0; i < ELV_HASH_ENTRIES; i++)
231 INIT_HLIST_HEAD(&eq->hash[i]);
232
Al Viro3d1ab402006-03-18 18:35:43 -0500233 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200234err:
235 kfree(eq);
236 elevator_put(e);
237 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500238}
239
240static void elevator_release(struct kobject *kobj)
241{
Jens Axboeb374d182008-10-31 10:05:07 +0100242 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200243
Jens Axboeb374d182008-10-31 10:05:07 +0100244 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500245 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200246 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500247 kfree(e);
248}
249
Jens Axboe165125e2007-07-24 09:28:11 +0200250int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
252 struct elevator_type *e = NULL;
253 struct elevator_queue *eq;
254 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200255 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Tejun Heocb98fc82005-10-28 08:29:39 +0200257 INIT_LIST_HEAD(&q->queue_head);
258 q->last_merge = NULL;
259 q->end_sector = 0;
260 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200261
Jens Axboe4eb166d2008-02-01 00:37:27 +0100262 if (name) {
263 e = elevator_get(name);
264 if (!e)
265 return -EINVAL;
266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Jens Axboe4eb166d2008-02-01 00:37:27 +0100268 if (!e && *chosen_elevator) {
269 e = elevator_get(chosen_elevator);
270 if (!e)
271 printk(KERN_ERR "I/O scheduler %s not found\n",
272 chosen_elevator);
273 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100274
Jens Axboe4eb166d2008-02-01 00:37:27 +0100275 if (!e) {
276 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
277 if (!e) {
278 printk(KERN_ERR
279 "Default I/O scheduler not found. " \
280 "Using noop.\n");
281 e = elevator_get("noop");
282 }
Nate Diller5f003972006-01-24 10:07:58 +0100283 }
284
Jens Axboeb5deef92006-07-19 23:39:40 +0200285 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500286 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Jens Axboebc1c1162006-06-08 08:49:06 +0200289 data = elevator_init_queue(q, eq);
290 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500291 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200292 return -ENOMEM;
293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Jens Axboebc1c1162006-06-08 08:49:06 +0200295 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 return ret;
297}
Jens Axboe2e662b62006-07-13 11:55:04 +0200298EXPORT_SYMBOL(elevator_init);
299
Jens Axboeb374d182008-10-31 10:05:07 +0100300void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Al Viro3d1ab402006-03-18 18:35:43 -0500302 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 if (e->ops->elevator_exit_fn)
304 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500305 e->ops = NULL;
306 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Al Viro3d1ab402006-03-18 18:35:43 -0500308 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
Jens Axboe2e662b62006-07-13 11:55:04 +0200310EXPORT_SYMBOL(elevator_exit);
311
Jens Axboe98170642006-07-28 09:23:08 +0200312static inline void __elv_rqhash_del(struct request *rq)
313{
314 hlist_del_init(&rq->hash);
315}
316
Jens Axboe165125e2007-07-24 09:28:11 +0200317static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200318{
319 if (ELV_ON_HASH(rq))
320 __elv_rqhash_del(rq);
321}
322
Jens Axboe165125e2007-07-24 09:28:11 +0200323static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200324{
Jens Axboeb374d182008-10-31 10:05:07 +0100325 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200326
327 BUG_ON(ELV_ON_HASH(rq));
328 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
329}
330
Jens Axboe165125e2007-07-24 09:28:11 +0200331static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200332{
333 __elv_rqhash_del(rq);
334 elv_rqhash_add(q, rq);
335}
336
Jens Axboe165125e2007-07-24 09:28:11 +0200337static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200338{
Jens Axboeb374d182008-10-31 10:05:07 +0100339 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200340 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
341 struct hlist_node *entry, *next;
342 struct request *rq;
343
344 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
345 BUG_ON(!ELV_ON_HASH(rq));
346
347 if (unlikely(!rq_mergeable(rq))) {
348 __elv_rqhash_del(rq);
349 continue;
350 }
351
352 if (rq_hash_key(rq) == offset)
353 return rq;
354 }
355
356 return NULL;
357}
358
Tejun Heo8922e162005-10-20 16:23:44 +0200359/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200360 * RB-tree support functions for inserting/lookup/removal of requests
361 * in a sorted RB tree.
362 */
363struct request *elv_rb_add(struct rb_root *root, struct request *rq)
364{
365 struct rb_node **p = &root->rb_node;
366 struct rb_node *parent = NULL;
367 struct request *__rq;
368
369 while (*p) {
370 parent = *p;
371 __rq = rb_entry(parent, struct request, rb_node);
372
373 if (rq->sector < __rq->sector)
374 p = &(*p)->rb_left;
375 else if (rq->sector > __rq->sector)
376 p = &(*p)->rb_right;
377 else
378 return __rq;
379 }
380
381 rb_link_node(&rq->rb_node, parent, p);
382 rb_insert_color(&rq->rb_node, root);
383 return NULL;
384}
Jens Axboe2e662b62006-07-13 11:55:04 +0200385EXPORT_SYMBOL(elv_rb_add);
386
387void elv_rb_del(struct rb_root *root, struct request *rq)
388{
389 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
390 rb_erase(&rq->rb_node, root);
391 RB_CLEAR_NODE(&rq->rb_node);
392}
Jens Axboe2e662b62006-07-13 11:55:04 +0200393EXPORT_SYMBOL(elv_rb_del);
394
395struct request *elv_rb_find(struct rb_root *root, sector_t sector)
396{
397 struct rb_node *n = root->rb_node;
398 struct request *rq;
399
400 while (n) {
401 rq = rb_entry(n, struct request, rb_node);
402
403 if (sector < rq->sector)
404 n = n->rb_left;
405 else if (sector > rq->sector)
406 n = n->rb_right;
407 else
408 return rq;
409 }
410
411 return NULL;
412}
Jens Axboe2e662b62006-07-13 11:55:04 +0200413EXPORT_SYMBOL(elv_rb_find);
414
415/*
Tejun Heo8922e162005-10-20 16:23:44 +0200416 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200417 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200418 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200419 */
Jens Axboe165125e2007-07-24 09:28:11 +0200420void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200421{
422 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200423 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100424 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200425
Tejun Heo06b86242005-10-20 16:46:23 +0200426 if (q->last_merge == rq)
427 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200428
429 elv_rqhash_del(q, rq);
430
Tejun Heo15853af2005-11-10 08:52:05 +0100431 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200432
Jens Axboe1b47f532005-10-20 16:37:00 +0200433 boundary = q->end_sector;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100434 stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200435 list_for_each_prev(entry, &q->queue_head) {
436 struct request *pos = list_entry_rq(entry);
437
David Woodhousee17fc0a2008-08-09 16:42:20 +0100438 if (blk_discard_rq(rq) != blk_discard_rq(pos))
439 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100440 if (rq_data_dir(rq) != rq_data_dir(pos))
441 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100442 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200443 break;
444 if (rq->sector >= boundary) {
445 if (pos->sector < boundary)
446 continue;
447 } else {
448 if (pos->sector >= boundary)
449 break;
450 }
451 if (rq->sector >= pos->sector)
452 break;
453 }
454
455 list_add(&rq->queuelist, entry);
456}
Jens Axboe2e662b62006-07-13 11:55:04 +0200457EXPORT_SYMBOL(elv_dispatch_sort);
458
Jens Axboe98170642006-07-28 09:23:08 +0200459/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200460 * Insert rq into dispatch queue of q. Queue lock must be held on
461 * entry. rq is added to the back of the dispatch queue. To be used by
462 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200463 */
464void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
465{
466 if (q->last_merge == rq)
467 q->last_merge = NULL;
468
469 elv_rqhash_del(q, rq);
470
471 q->nr_sorted--;
472
473 q->end_sector = rq_end_sector(rq);
474 q->boundary_rq = rq;
475 list_add_tail(&rq->queuelist, &q->queue_head);
476}
Jens Axboe2e662b62006-07-13 11:55:04 +0200477EXPORT_SYMBOL(elv_dispatch_add_tail);
478
Jens Axboe165125e2007-07-24 09:28:11 +0200479int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
Jens Axboeb374d182008-10-31 10:05:07 +0100481 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200482 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200483 int ret;
484
Jens Axboe98170642006-07-28 09:23:08 +0200485 /*
486 * 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. Brunelleac9fafa2008-04-29 14:44:19 +0200496 if (blk_queue_nomerges(q))
497 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)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 q->in_flight--;
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{
578 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]
Tejun Heo8922e162005-10-20 16:23:44 +0200688 - q->in_flight;
689
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);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900813 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200814 }
815}
816EXPORT_SYMBOL(elv_abort_queue);
817
Jens Axboe165125e2007-07-24 09:28:11 +0200818void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
Jens Axboeb374d182008-10-31 10:05:07 +0100820 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 /*
823 * request is released from the driver, io must be done
824 */
Tejun Heo8922e162005-10-20 16:23:44 +0200825 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 q->in_flight--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100827 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
828 e->ops->elevator_completed_req_fn(q, rq);
829 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100830
Tejun Heo1bc691d2006-01-12 15:39:26 +0100831 /*
832 * Check if the queue is waiting for fs requests to be
833 * drained for flush sequence.
834 */
835 if (unlikely(q->ordseq)) {
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900836 struct request *next = NULL;
837
838 if (!list_empty(&q->queue_head))
839 next = list_entry_rq(q->queue_head.next);
840
841 if (!q->in_flight &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100842 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900843 (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
Tejun Heo797e7db2006-01-06 09:51:03 +0100844 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +0900845 __blk_run_queue(q);
Tejun Heo797e7db2006-01-06 09:51:03 +0100846 }
Tejun Heo8922e162005-10-20 16:23:44 +0200847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848}
849
Al Viro3d1ab402006-03-18 18:35:43 -0500850#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
851
852static ssize_t
853elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
854{
Al Viro3d1ab402006-03-18 18:35:43 -0500855 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100856 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500857 ssize_t error;
858
859 if (!entry->show)
860 return -EIO;
861
Jens Axboeb374d182008-10-31 10:05:07 +0100862 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500863 mutex_lock(&e->sysfs_lock);
864 error = e->ops ? entry->show(e, page) : -ENOENT;
865 mutex_unlock(&e->sysfs_lock);
866 return error;
867}
868
869static ssize_t
870elv_attr_store(struct kobject *kobj, struct attribute *attr,
871 const char *page, size_t length)
872{
Al Viro3d1ab402006-03-18 18:35:43 -0500873 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100874 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500875 ssize_t error;
876
877 if (!entry->store)
878 return -EIO;
879
Jens Axboeb374d182008-10-31 10:05:07 +0100880 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500881 mutex_lock(&e->sysfs_lock);
882 error = e->ops ? entry->store(e, page, length) : -ENOENT;
883 mutex_unlock(&e->sysfs_lock);
884 return error;
885}
886
887static struct sysfs_ops elv_sysfs_ops = {
888 .show = elv_attr_show,
889 .store = elv_attr_store,
890};
891
892static struct kobj_type elv_ktype = {
893 .sysfs_ops = &elv_sysfs_ops,
894 .release = elevator_release,
895};
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897int elv_register_queue(struct request_queue *q)
898{
Jens Axboeb374d182008-10-31 10:05:07 +0100899 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500900 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700902 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500903 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500904 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500905 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500906 while (attr->attr.name) {
907 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500908 break;
Al Viroe572ec72006-03-18 22:27:18 -0500909 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500910 }
911 }
912 kobject_uevent(&e->kobj, KOBJ_ADD);
913 }
914 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
916
Jens Axboeb374d182008-10-31 10:05:07 +0100917static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboebc1c1162006-06-08 08:49:06 +0200918{
919 kobject_uevent(&e->kobj, KOBJ_REMOVE);
920 kobject_del(&e->kobj);
921}
922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923void elv_unregister_queue(struct request_queue *q)
924{
Jens Axboebc1c1162006-06-08 08:49:06 +0200925 if (q)
926 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927}
928
Adrian Bunk2fdd82b2007-12-12 18:51:56 +0100929void elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100931 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200932
933 spin_lock(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100934 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200936 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
Nate Diller5f003972006-01-24 10:07:58 +0100938 if (!strcmp(e->elevator_name, chosen_elevator) ||
939 (!*chosen_elevator &&
940 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100941 def = " (default)";
942
Jens Axboe4eb166d2008-02-01 00:37:27 +0100943 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
944 def);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
946EXPORT_SYMBOL_GPL(elv_register);
947
948void elv_unregister(struct elevator_type *e)
949{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800950 struct task_struct *g, *p;
951
952 /*
953 * Iterate every thread in the process to remove the io contexts.
954 */
Al Viroe17a9482006-03-18 13:21:20 -0500955 if (e->ops.trim) {
956 read_lock(&tasklist_lock);
957 do_each_thread(g, p) {
958 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400959 if (p->io_context)
960 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500961 task_unlock(p);
962 } while_each_thread(g, p);
963 read_unlock(&tasklist_lock);
964 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800965
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200966 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200968 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969}
970EXPORT_SYMBOL_GPL(elv_unregister);
971
972/*
973 * switch to new_e io scheduler. be careful not to introduce deadlocks -
974 * we don't free the old io scheduler, before we have allocated what we
975 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200976 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 */
Jens Axboe165125e2007-07-24 09:28:11 +0200978static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Jens Axboeb374d182008-10-31 10:05:07 +0100980 struct elevator_queue *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200981 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Tejun Heocb98fc82005-10-28 08:29:39 +0200983 /*
984 * Allocate new elevator
985 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200986 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -0500988 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Jens Axboebc1c1162006-06-08 08:49:06 +0200990 data = elevator_init_queue(q, e);
991 if (!data) {
992 kobject_put(&e->kobj);
993 return 0;
994 }
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200997 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 */
Tejun Heocb98fc82005-10-28 08:29:39 +0200999 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001000 elv_quiesce_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 /*
Jens Axboebc1c1162006-06-08 08:49:06 +02001003 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 old_elevator = q->elevator;
1006
1007 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 * attach and start new elevator
1009 */
Jens Axboebc1c1162006-06-08 08:49:06 +02001010 elevator_attach(q, e, data);
1011
1012 spin_unlock_irq(q->queue_lock);
1013
1014 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 if (elv_register_queue(q))
1017 goto fail_register;
1018
1019 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001020 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 */
1022 elevator_exit(old_elevator);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001023 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001024 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001025 spin_unlock_irq(q->queue_lock);
1026
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001027 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1028
Al Viro3d1ab402006-03-18 18:35:43 -05001029 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
1031fail_register:
1032 /*
1033 * switch failed, exit the new io scheduler and reattach the old
1034 * one again (along with re-adding the sysfs dir)
1035 */
1036 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 q->elevator = old_elevator;
1038 elv_register_queue(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001039
1040 spin_lock_irq(q->queue_lock);
1041 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1042 spin_unlock_irq(q->queue_lock);
1043
Al Viro3d1ab402006-03-18 18:35:43 -05001044 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045}
1046
Jens Axboe165125e2007-07-24 09:28:11 +02001047ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1048 size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049{
1050 char elevator_name[ELV_NAME_MAX];
1051 struct elevator_type *e;
1052
Li Zefanee2e9922008-10-14 08:49:56 +02001053 strlcpy(elevator_name, name, sizeof(elevator_name));
1054 strstrip(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 e = elevator_get(elevator_name);
1057 if (!e) {
1058 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1059 return -EINVAL;
1060 }
1061
Nate Diller2ca7d932005-10-30 15:02:24 -08001062 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1063 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -08001065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Al Viro3d1ab402006-03-18 18:35:43 -05001067 if (!elevator_switch(q, e))
Jens Axboe4eb166d2008-02-01 00:37:27 +01001068 printk(KERN_ERR "elevator: switch to %s failed\n",
1069 elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 return count;
1071}
1072
Jens Axboe165125e2007-07-24 09:28:11 +02001073ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Jens Axboeb374d182008-10-31 10:05:07 +01001075 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 struct elevator_type *elv = e->elevator_type;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001077 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 int len = 0;
1079
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001080 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001081 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 if (!strcmp(elv->elevator_name, __e->elevator_name))
1083 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1084 else
1085 len += sprintf(name+len, "%s ", __e->elevator_name);
1086 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001087 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089 len += sprintf(len+name, "\n");
1090 return len;
1091}
1092
Jens Axboe165125e2007-07-24 09:28:11 +02001093struct request *elv_rb_former_request(struct request_queue *q,
1094 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001095{
1096 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1097
1098 if (rbprev)
1099 return rb_entry_rq(rbprev);
1100
1101 return NULL;
1102}
Jens Axboe2e662b62006-07-13 11:55:04 +02001103EXPORT_SYMBOL(elv_rb_former_request);
1104
Jens Axboe165125e2007-07-24 09:28:11 +02001105struct request *elv_rb_latter_request(struct request_queue *q,
1106 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001107{
1108 struct rb_node *rbnext = rb_next(&rq->rb_node);
1109
1110 if (rbnext)
1111 return rb_entry_rq(rbnext);
1112
1113 return NULL;
1114}
Jens Axboe2e662b62006-07-13 11:55:04 +02001115EXPORT_SYMBOL(elv_rb_latter_request);