blob: efec4576b67a7c5faa6f4b3fcc4e24af3d84f11d [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>
Jens Axboe2056a782006-03-23 20:00:26 +010034#include <linux/blktrace_api.h>
Jens Axboe98170642006-07-28 09:23:08 +020035#include <linux/hash.h>
Jens Axboe0835da62008-08-26 09:15:47 +020036#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Li Zefan55782132009-06-09 13:43:05 +080038#include <trace/events/block.h>
39
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
45/*
Jens Axboe98170642006-07-28 09:23:08 +020046 * Merge hash stuff.
47 */
48static const int elv_hash_shift = 6;
49#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
Jens Axboe4eb166d2008-02-01 00:37:27 +010050#define ELV_HASH_FN(sec) \
51 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
Jens Axboe98170642006-07-28 09:23:08 +020052#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
Tejun Heo83096eb2009-05-07 22:24:39 +090053#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
Jens Axboe98170642006-07-28 09:23:08 +020054
55/*
Jens Axboeda775262006-12-20 11:04:12 +010056 * Query io scheduler to see if the current process issuing bio may be
57 * merged with rq.
58 */
59static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
60{
Jens Axboe165125e2007-07-24 09:28:11 +020061 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010062 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010063
Tejun Heo22f746e2011-12-14 00:33:41 +010064 if (e->type->ops.elevator_allow_merge_fn)
65 return e->type->ops.elevator_allow_merge_fn(q, rq, bio);
Jens Axboeda775262006-12-20 11:04:12 +010066
67 return 1;
68}
69
70/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * can we safely merge with this request?
72 */
Tejun Heo050c8ea2012-02-08 09:19:38 +010073bool elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 if (!rq_mergeable(rq))
76 return 0;
77
78 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +010079 * Don't merge file system requests and discard requests
80 */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +020081 if ((bio->bi_rw & REQ_DISCARD) != (rq->bio->bi_rw & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +010082 return 0;
83
84 /*
Adrian Hunter8d57a982010-08-11 14:17:49 -070085 * Don't merge discard requests and secure discard requests
86 */
87 if ((bio->bi_rw & REQ_SECURE) != (rq->bio->bi_rw & REQ_SECURE))
88 return 0;
89
90 /*
Maya Erez73937f52012-05-24 23:33:05 +030091 * Don't merge sanitize requests
92 */
93 if ((bio->bi_rw & REQ_SANITIZE) != (rq->bio->bi_rw & REQ_SANITIZE))
94 return 0;
95
96 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * different data direction or already started, don't merge
98 */
99 if (bio_data_dir(bio) != rq_data_dir(rq))
100 return 0;
101
102 /*
Jens Axboeda775262006-12-20 11:04:12 +0100103 * must be same device and not a special request
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 */
Jens Axboebb4067e2006-12-21 21:20:01 +0100105 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
Jens Axboeda775262006-12-20 11:04:12 +0100106 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200108 /*
109 * only merge integrity protected bio into ditto rq
110 */
111 if (bio_integrity(bio) != blk_integrity_rq(rq))
112 return 0;
113
Jens Axboeda775262006-12-20 11:04:12 +0100114 if (!elv_iosched_allow_merge(rq, bio))
115 return 0;
116
117 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
119EXPORT_SYMBOL(elv_rq_merge_ok);
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121static struct elevator_type *elevator_find(const char *name)
122{
Vasily Tarasova22b1692006-10-11 09:24:27 +0200123 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200125 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +0200126 if (!strcmp(e->elevator_name, name))
127 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Vasily Tarasova22b1692006-10-11 09:24:27 +0200130 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
133static void elevator_put(struct elevator_type *e)
134{
135 module_put(e->elevator_owner);
136}
137
138static struct elevator_type *elevator_get(const char *name)
139{
Tejun Heo2824bc932005-10-20 10:56:41 +0200140 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200142 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200143
144 e = elevator_find(name);
Jens Axboee1640942008-02-19 10:20:37 +0100145 if (!e) {
Jens Axboee1640942008-02-19 10:20:37 +0100146 spin_unlock(&elv_list_lock);
Kees Cook490b94b2011-05-05 18:02:12 -0600147 request_module("%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100148 spin_lock(&elv_list_lock);
149 e = elevator_find(name);
150 }
151
Tejun Heo2824bc932005-10-20 10:56:41 +0200152 if (e && !try_module_get(e->elevator_owner))
153 e = NULL;
154
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200155 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 return e;
158}
159
Tejun Heof8fc8772011-12-14 00:33:40 +0100160static int elevator_init_queue(struct request_queue *q,
161 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
Tejun Heo22f746e2011-12-14 00:33:41 +0100163 eq->elevator_data = eq->type->ops.elevator_init_fn(q);
Tejun Heof8fc8772011-12-14 00:33:40 +0100164 if (eq->elevator_data)
165 return 0;
166 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167}
168
Wang Sheng-Hui484fc252011-09-08 12:32:14 +0200169static char chosen_elevator[ELV_NAME_MAX];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Nate Diller5f003972006-01-24 10:07:58 +0100171static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100173 /*
174 * Be backwards-compatible with previous kernels, so users
175 * won't get the wrong elevator.
176 */
Jens Axboe492af632009-10-03 09:37:51 +0200177 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800178 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
181__setup("elevator=", elevator_setup);
182
Al Viro3d1ab402006-03-18 18:35:43 -0500183static struct kobj_type elv_ktype;
184
Jens Axboeb374d182008-10-31 10:05:07 +0100185static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200186 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500187{
Jens Axboeb374d182008-10-31 10:05:07 +0100188 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200189 int i;
190
Jens Axboeb374d182008-10-31 10:05:07 +0100191 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200192 if (unlikely(!eq))
193 goto err;
194
Tejun Heo22f746e2011-12-14 00:33:41 +0100195 eq->type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700196 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200197 mutex_init(&eq->sysfs_lock);
198
Jens Axboeb5deef92006-07-19 23:39:40 +0200199 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
200 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200201 if (!eq->hash)
202 goto err;
203
204 for (i = 0; i < ELV_HASH_ENTRIES; i++)
205 INIT_HLIST_HEAD(&eq->hash[i]);
206
Al Viro3d1ab402006-03-18 18:35:43 -0500207 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200208err:
209 kfree(eq);
210 elevator_put(e);
211 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500212}
213
214static void elevator_release(struct kobject *kobj)
215{
Jens Axboeb374d182008-10-31 10:05:07 +0100216 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200217
Jens Axboeb374d182008-10-31 10:05:07 +0100218 e = container_of(kobj, struct elevator_queue, kobj);
Tejun Heo22f746e2011-12-14 00:33:41 +0100219 elevator_put(e->type);
Jens Axboe98170642006-07-28 09:23:08 +0200220 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500221 kfree(e);
222}
223
Jens Axboe165125e2007-07-24 09:28:11 +0200224int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 struct elevator_type *e = NULL;
227 struct elevator_queue *eq;
Tejun Heof8fc8772011-12-14 00:33:40 +0100228 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400230 if (unlikely(q->elevator))
231 return 0;
232
Tejun Heocb98fc82005-10-28 08:29:39 +0200233 INIT_LIST_HEAD(&q->queue_head);
234 q->last_merge = NULL;
235 q->end_sector = 0;
236 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200237
Jens Axboe4eb166d2008-02-01 00:37:27 +0100238 if (name) {
239 e = elevator_get(name);
240 if (!e)
241 return -EINVAL;
242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Jens Axboe4eb166d2008-02-01 00:37:27 +0100244 if (!e && *chosen_elevator) {
245 e = elevator_get(chosen_elevator);
246 if (!e)
247 printk(KERN_ERR "I/O scheduler %s not found\n",
248 chosen_elevator);
249 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100250
Jens Axboe4eb166d2008-02-01 00:37:27 +0100251 if (!e) {
252 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
253 if (!e) {
254 printk(KERN_ERR
255 "Default I/O scheduler not found. " \
256 "Using noop.\n");
257 e = elevator_get("noop");
258 }
Nate Diller5f003972006-01-24 10:07:58 +0100259 }
260
Jens Axboeb5deef92006-07-19 23:39:40 +0200261 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500262 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Tejun Heof8fc8772011-12-14 00:33:40 +0100265 err = elevator_init_queue(q, eq);
266 if (err) {
Al Viro3d1ab402006-03-18 18:35:43 -0500267 kobject_put(&eq->kobj);
Tejun Heof8fc8772011-12-14 00:33:40 +0100268 return err;
Jens Axboebc1c1162006-06-08 08:49:06 +0200269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Tejun Heof8fc8772011-12-14 00:33:40 +0100271 q->elevator = eq;
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400272 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}
Jens Axboe2e662b62006-07-13 11:55:04 +0200274EXPORT_SYMBOL(elevator_init);
275
Jens Axboeb374d182008-10-31 10:05:07 +0100276void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
Al Viro3d1ab402006-03-18 18:35:43 -0500278 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100279 if (e->type->ops.elevator_exit_fn)
280 e->type->ops.elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500281 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
Al Viro3d1ab402006-03-18 18:35:43 -0500283 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
Jens Axboe2e662b62006-07-13 11:55:04 +0200285EXPORT_SYMBOL(elevator_exit);
286
Jens Axboe98170642006-07-28 09:23:08 +0200287static inline void __elv_rqhash_del(struct request *rq)
288{
289 hlist_del_init(&rq->hash);
290}
291
Jens Axboe165125e2007-07-24 09:28:11 +0200292static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200293{
294 if (ELV_ON_HASH(rq))
295 __elv_rqhash_del(rq);
296}
297
Jens Axboe165125e2007-07-24 09:28:11 +0200298static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200299{
Jens Axboeb374d182008-10-31 10:05:07 +0100300 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200301
302 BUG_ON(ELV_ON_HASH(rq));
303 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
304}
305
Jens Axboe165125e2007-07-24 09:28:11 +0200306static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200307{
308 __elv_rqhash_del(rq);
309 elv_rqhash_add(q, rq);
310}
311
Jens Axboe165125e2007-07-24 09:28:11 +0200312static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200313{
Jens Axboeb374d182008-10-31 10:05:07 +0100314 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200315 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
316 struct hlist_node *entry, *next;
317 struct request *rq;
318
319 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
320 BUG_ON(!ELV_ON_HASH(rq));
321
322 if (unlikely(!rq_mergeable(rq))) {
323 __elv_rqhash_del(rq);
324 continue;
325 }
326
327 if (rq_hash_key(rq) == offset)
328 return rq;
329 }
330
331 return NULL;
332}
333
Tejun Heo8922e162005-10-20 16:23:44 +0200334/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200335 * RB-tree support functions for inserting/lookup/removal of requests
336 * in a sorted RB tree.
337 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200338void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200339{
340 struct rb_node **p = &root->rb_node;
341 struct rb_node *parent = NULL;
342 struct request *__rq;
343
344 while (*p) {
345 parent = *p;
346 __rq = rb_entry(parent, struct request, rb_node);
347
Tejun Heo83096eb2009-05-07 22:24:39 +0900348 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200349 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200350 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200351 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200352 }
353
354 rb_link_node(&rq->rb_node, parent, p);
355 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200356}
Jens Axboe2e662b62006-07-13 11:55:04 +0200357EXPORT_SYMBOL(elv_rb_add);
358
359void elv_rb_del(struct rb_root *root, struct request *rq)
360{
361 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
362 rb_erase(&rq->rb_node, root);
363 RB_CLEAR_NODE(&rq->rb_node);
364}
Jens Axboe2e662b62006-07-13 11:55:04 +0200365EXPORT_SYMBOL(elv_rb_del);
366
367struct request *elv_rb_find(struct rb_root *root, sector_t sector)
368{
369 struct rb_node *n = root->rb_node;
370 struct request *rq;
371
372 while (n) {
373 rq = rb_entry(n, struct request, rb_node);
374
Tejun Heo83096eb2009-05-07 22:24:39 +0900375 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200376 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900377 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200378 n = n->rb_right;
379 else
380 return rq;
381 }
382
383 return NULL;
384}
Jens Axboe2e662b62006-07-13 11:55:04 +0200385EXPORT_SYMBOL(elv_rb_find);
386
387/*
Tejun Heo8922e162005-10-20 16:23:44 +0200388 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200389 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200390 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200391 */
Jens Axboe165125e2007-07-24 09:28:11 +0200392void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200393{
394 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200395 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100396 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200397
Tejun Heo06b86242005-10-20 16:46:23 +0200398 if (q->last_merge == rq)
399 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200400
401 elv_rqhash_del(q, rq);
402
Tejun Heo15853af2005-11-10 08:52:05 +0100403 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200404
Jens Axboe1b47f532005-10-20 16:37:00 +0200405 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100406 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200407 list_for_each_prev(entry, &q->queue_head) {
408 struct request *pos = list_entry_rq(entry);
409
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200410 if ((rq->cmd_flags & REQ_DISCARD) !=
411 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100412 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100413 if (rq_data_dir(rq) != rq_data_dir(pos))
414 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100415 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200416 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900417 if (blk_rq_pos(rq) >= boundary) {
418 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200419 continue;
420 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900421 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200422 break;
423 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900424 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200425 break;
426 }
427
428 list_add(&rq->queuelist, entry);
429}
Jens Axboe2e662b62006-07-13 11:55:04 +0200430EXPORT_SYMBOL(elv_dispatch_sort);
431
Jens Axboe98170642006-07-28 09:23:08 +0200432/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200433 * Insert rq into dispatch queue of q. Queue lock must be held on
434 * entry. rq is added to the back of the dispatch queue. To be used by
435 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200436 */
437void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
438{
439 if (q->last_merge == rq)
440 q->last_merge = NULL;
441
442 elv_rqhash_del(q, rq);
443
444 q->nr_sorted--;
445
446 q->end_sector = rq_end_sector(rq);
447 q->boundary_rq = rq;
448 list_add_tail(&rq->queuelist, &q->queue_head);
449}
Jens Axboe2e662b62006-07-13 11:55:04 +0200450EXPORT_SYMBOL(elv_dispatch_add_tail);
451
Jens Axboe165125e2007-07-24 09:28:11 +0200452int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Jens Axboeb374d182008-10-31 10:05:07 +0100454 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200455 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200456 int ret;
457
Jens Axboe98170642006-07-28 09:23:08 +0200458 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100459 * Levels of merges:
460 * nomerges: No merges at all attempted
461 * noxmerges: Only simple one-hit cache try
462 * merges: All merge tries attempted
463 */
464 if (blk_queue_nomerges(q))
465 return ELEVATOR_NO_MERGE;
466
467 /*
Jens Axboe98170642006-07-28 09:23:08 +0200468 * First try one-hit cache.
469 */
Tejun Heo050c8ea2012-02-08 09:19:38 +0100470 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
471 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200472 if (ret != ELEVATOR_NO_MERGE) {
473 *req = q->last_merge;
474 return ret;
475 }
476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100478 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200479 return ELEVATOR_NO_MERGE;
480
Jens Axboe98170642006-07-28 09:23:08 +0200481 /*
482 * See if our hash lookup can find a potential backmerge.
483 */
484 __rq = elv_rqhash_find(q, bio->bi_sector);
485 if (__rq && elv_rq_merge_ok(__rq, bio)) {
486 *req = __rq;
487 return ELEVATOR_BACK_MERGE;
488 }
489
Tejun Heo22f746e2011-12-14 00:33:41 +0100490 if (e->type->ops.elevator_merge_fn)
491 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 return ELEVATOR_NO_MERGE;
494}
495
Jens Axboe5e84ea32011-03-21 10:14:27 +0100496/*
497 * Attempt to do an insertion back merge. Only check for the case where
498 * we can append 'rq' to an existing request, so we can throw 'rq' away
499 * afterwards.
500 *
501 * Returns true if we merged, false otherwise
502 */
503static bool elv_attempt_insert_merge(struct request_queue *q,
504 struct request *rq)
505{
506 struct request *__rq;
507
508 if (blk_queue_nomerges(q))
509 return false;
510
511 /*
512 * First try one-hit cache.
513 */
514 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
515 return true;
516
517 if (blk_queue_noxmerges(q))
518 return false;
519
520 /*
521 * See if our hash lookup can find a potential backmerge.
522 */
Jens Axboe5d381ef2012-01-15 10:29:48 +0100523 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
524 if (__rq && blk_attempt_req_merge(q, __rq, rq))
525 return true;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100526
Jens Axboe5d381ef2012-01-15 10:29:48 +0100527 return false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100528}
529
Jens Axboe165125e2007-07-24 09:28:11 +0200530void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Jens Axboeb374d182008-10-31 10:05:07 +0100532 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Tejun Heo22f746e2011-12-14 00:33:41 +0100534 if (e->type->ops.elevator_merged_fn)
535 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200536
Jens Axboe2e662b62006-07-13 11:55:04 +0200537 if (type == ELEVATOR_BACK_MERGE)
538 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200539
Tejun Heo06b86242005-10-20 16:46:23 +0200540 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541}
542
Jens Axboe165125e2007-07-24 09:28:11 +0200543void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 struct request *next)
545{
Jens Axboeb374d182008-10-31 10:05:07 +0100546 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100547 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Tejun Heo22f746e2011-12-14 00:33:41 +0100549 if (next_sorted && e->type->ops.elevator_merge_req_fn)
550 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200551
Jens Axboe98170642006-07-28 09:23:08 +0200552 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200553
Jens Axboe5e84ea32011-03-21 10:14:27 +0100554 if (next_sorted) {
555 elv_rqhash_del(q, next);
556 q->nr_sorted--;
557 }
558
Tejun Heo06b86242005-10-20 16:46:23 +0200559 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
Divyesh Shah812d4022010-04-08 21:14:23 -0700562void elv_bio_merged(struct request_queue *q, struct request *rq,
563 struct bio *bio)
564{
565 struct elevator_queue *e = q->elevator;
566
Tejun Heo22f746e2011-12-14 00:33:41 +0100567 if (e->type->ops.elevator_bio_merged_fn)
568 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700569}
570
Jens Axboe165125e2007-07-24 09:28:11 +0200571void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 /*
574 * it already went through dequeue, we need to decrement the
575 * in_flight count again
576 */
Tejun Heo8922e162005-10-20 16:23:44 +0200577 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200578 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200579 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100580 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Jens Axboe4aff5e22006-08-10 08:44:47 +0200583 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Jens Axboeb710a482011-03-30 09:52:30 +0200585 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
Tatyana Brokhman57d80192012-12-04 15:54:43 +0200588/**
589 * elv_reinsert_request() - Insert a request back to the scheduler
590 * @q: request queue where request should be inserted
591 * @rq: request to be inserted
592 *
593 * This function returns the request back to the scheduler to be
594 * inserted as if it was never dispatched
595 *
596 * Return: 0 on success, error code on failure
597 */
598int elv_reinsert_request(struct request_queue *q, struct request *rq)
599{
600 int res;
601
602 if (!q->elevator->type->ops.elevator_reinsert_req_fn)
603 return -EPERM;
604
605 res = q->elevator->type->ops.elevator_reinsert_req_fn(q, rq);
606 if (!res) {
607 /*
608 * it already went through dequeue, we need to decrement the
609 * in_flight count again
610 */
611 if (blk_account_rq(rq)) {
612 q->in_flight[rq_is_sync(rq)]--;
613 if (rq->cmd_flags & REQ_SORTED)
614 elv_deactivate_rq(q, rq);
615 }
616 rq->cmd_flags &= ~REQ_STARTED;
617 q->nr_sorted++;
618 }
619
620 return res;
621}
622
Jerome Marchand26308ea2009-03-27 10:31:51 +0100623void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100624{
625 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200626
627 lockdep_assert_held(q->queue_lock);
628
Tejun Heo22f746e2011-12-14 00:33:41 +0100629 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100630 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200631 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100632 printk(KERN_ERR "%s: forced dispatching is broken "
633 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100634 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100635 }
636}
637
Jens Axboef600abe2009-04-08 14:22:01 +0200638void elv_quiesce_start(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100639{
Martin K. Petersencd43e262009-05-22 17:17:52 -0400640 if (!q->elevator)
641 return;
642
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200643 spin_lock_irq(q->queue_lock);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100644 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200645 spin_unlock_irq(q->queue_lock);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100646
Tejun Heoc9a929d2011-10-19 14:42:16 +0200647 blk_drain_queue(q, false);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100648}
649
Jens Axboef600abe2009-04-08 14:22:01 +0200650void elv_quiesce_end(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100651{
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200652 spin_lock_irq(q->queue_lock);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100653 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200654 spin_unlock_irq(q->queue_lock);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100655}
656
Jens Axboeb710a482011-03-30 09:52:30 +0200657void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100659 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 rq->q = q;
662
Jens Axboeb710a482011-03-30 09:52:30 +0200663 if (rq->cmd_flags & REQ_SOFTBARRIER) {
664 /* barriers are scheduling boundary, update end_sector */
665 if (rq->cmd_type == REQ_TYPE_FS ||
Maya Erez73937f52012-05-24 23:33:05 +0300666 (rq->cmd_flags & (REQ_DISCARD | REQ_SANITIZE))) {
Jens Axboeb710a482011-03-30 09:52:30 +0200667 q->end_sector = rq_end_sector(rq);
668 q->boundary_rq = rq;
669 }
670 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200671 (where == ELEVATOR_INSERT_SORT ||
672 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200673 where = ELEVATOR_INSERT_BACK;
674
Tejun Heo8922e162005-10-20 16:23:44 +0200675 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200676 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200677 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200678 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200679 list_add(&rq->queuelist, &q->queue_head);
680 break;
681
682 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200683 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100684 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200685 list_add_tail(&rq->queuelist, &q->queue_head);
686 /*
687 * We kick the queue here for the following reasons.
688 * - The elevator might have returned NULL previously
689 * to delay requests and returned them now. As the
690 * queue wasn't empty before this request, ll_rw_blk
691 * won't run the queue on return, resulting in hang.
692 * - Usually, back inserted requests won't be merged
693 * with anything. There's no point in delaying queue
694 * processing.
695 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200696 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200697 break;
698
Jens Axboe5e84ea32011-03-21 10:14:27 +0100699 case ELEVATOR_INSERT_SORT_MERGE:
700 /*
701 * If we succeed in merging this request with one in the
702 * queue already, we are done - rq has now been freed,
703 * so no need to do anything further.
704 */
705 if (elv_attempt_insert_merge(q, rq))
706 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200707 case ELEVATOR_INSERT_SORT:
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200708 BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
709 !(rq->cmd_flags & REQ_DISCARD));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200710 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100711 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200712 if (rq_mergeable(rq)) {
713 elv_rqhash_add(q, rq);
714 if (!q->last_merge)
715 q->last_merge = rq;
716 }
717
Tejun Heoca235092005-11-01 17:23:49 +0900718 /*
719 * Some ioscheds (cfq) run q->request_fn directly, so
720 * rq cannot be accessed after calling
721 * elevator_add_req_fn.
722 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100723 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200724 break;
725
Tejun Heoae1b1532011-01-25 12:43:54 +0100726 case ELEVATOR_INSERT_FLUSH:
727 rq->cmd_flags |= REQ_SOFTBARRIER;
728 blk_insert_flush(rq);
729 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200730 default:
731 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700732 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200733 BUG();
734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
Jens Axboe2e662b62006-07-13 11:55:04 +0200736EXPORT_SYMBOL(__elv_add_request);
737
Jens Axboe7eaceac2011-03-10 08:52:07 +0100738void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
740 unsigned long flags;
741
742 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100743 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 spin_unlock_irqrestore(q->queue_lock, flags);
745}
Jens Axboe2e662b62006-07-13 11:55:04 +0200746EXPORT_SYMBOL(elv_add_request);
747
Jens Axboe165125e2007-07-24 09:28:11 +0200748struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Jens Axboeb374d182008-10-31 10:05:07 +0100750 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Tejun Heo22f746e2011-12-14 00:33:41 +0100752 if (e->type->ops.elevator_latter_req_fn)
753 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return NULL;
755}
756
Jens Axboe165125e2007-07-24 09:28:11 +0200757struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
Jens Axboeb374d182008-10-31 10:05:07 +0100759 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Tejun Heo22f746e2011-12-14 00:33:41 +0100761 if (e->type->ops.elevator_former_req_fn)
762 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return NULL;
764}
765
Jens Axboe165125e2007-07-24 09:28:11 +0200766int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
Jens Axboeb374d182008-10-31 10:05:07 +0100768 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Tejun Heo22f746e2011-12-14 00:33:41 +0100770 if (e->type->ops.elevator_set_req_fn)
771 return e->type->ops.elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 return 0;
773}
774
Jens Axboe165125e2007-07-24 09:28:11 +0200775void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
Jens Axboeb374d182008-10-31 10:05:07 +0100777 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Tejun Heo22f746e2011-12-14 00:33:41 +0100779 if (e->type->ops.elevator_put_req_fn)
780 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
Jens Axboe165125e2007-07-24 09:28:11 +0200783int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
Jens Axboeb374d182008-10-31 10:05:07 +0100785 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Tejun Heo22f746e2011-12-14 00:33:41 +0100787 if (e->type->ops.elevator_may_queue_fn)
788 return e->type->ops.elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 return ELV_MQUEUE_MAY;
791}
792
Mike Anderson11914a52008-09-13 20:31:27 +0200793void elv_abort_queue(struct request_queue *q)
794{
795 struct request *rq;
796
Tejun Heoae1b1532011-01-25 12:43:54 +0100797 blk_abort_flushes(q);
798
Mike Anderson11914a52008-09-13 20:31:27 +0200799 while (!list_empty(&q->queue_head)) {
800 rq = list_entry_rq(q->queue_head.next);
801 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100802 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200803 /*
804 * Mark this request as started so we don't trigger
805 * any debug logic in the end I/O path.
806 */
807 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900808 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200809 }
810}
811EXPORT_SYMBOL(elv_abort_queue);
812
Jens Axboe165125e2007-07-24 09:28:11 +0200813void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
Jens Axboeb374d182008-10-31 10:05:07 +0100815 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Tatyana Brokhman88fd1b42012-12-04 16:04:15 +0200817 if (test_bit(REQ_ATOM_URGENT, &rq->atomic_flags)) {
818 q->notified_urgent = false;
819 q->dispatched_urgent = false;
820 blk_clear_rq_urgent(rq);
821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 /*
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)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200826 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200827 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100828 e->type->ops.elevator_completed_req_fn)
829 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
832
Al Viro3d1ab402006-03-18 18:35:43 -0500833#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
834
835static ssize_t
836elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
837{
Al Viro3d1ab402006-03-18 18:35:43 -0500838 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100839 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500840 ssize_t error;
841
842 if (!entry->show)
843 return -EIO;
844
Jens Axboeb374d182008-10-31 10:05:07 +0100845 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500846 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100847 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500848 mutex_unlock(&e->sysfs_lock);
849 return error;
850}
851
852static ssize_t
853elv_attr_store(struct kobject *kobj, struct attribute *attr,
854 const char *page, size_t length)
855{
Al Viro3d1ab402006-03-18 18:35:43 -0500856 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100857 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500858 ssize_t error;
859
860 if (!entry->store)
861 return -EIO;
862
Jens Axboeb374d182008-10-31 10:05:07 +0100863 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500864 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100865 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500866 mutex_unlock(&e->sysfs_lock);
867 return error;
868}
869
Emese Revfy52cf25d2010-01-19 02:58:23 +0100870static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500871 .show = elv_attr_show,
872 .store = elv_attr_store,
873};
874
875static struct kobj_type elv_ktype = {
876 .sysfs_ops = &elv_sysfs_ops,
877 .release = elevator_release,
878};
879
Tejun Heof8fc8772011-12-14 00:33:40 +0100880int __elv_register_queue(struct request_queue *q, struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881{
Al Viro3d1ab402006-03-18 18:35:43 -0500882 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700884 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500885 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100886 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500887 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500888 while (attr->attr.name) {
889 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500890 break;
Al Viroe572ec72006-03-18 22:27:18 -0500891 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500892 }
893 }
894 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200895 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500896 }
897 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
Tejun Heof8fc8772011-12-14 00:33:40 +0100900int elv_register_queue(struct request_queue *q)
Jens Axboebc1c1162006-06-08 08:49:06 +0200901{
Tejun Heof8fc8772011-12-14 00:33:40 +0100902 return __elv_register_queue(q, q->elevator);
Jens Axboebc1c1162006-06-08 08:49:06 +0200903}
Tejun Heof8fc8772011-12-14 00:33:40 +0100904EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906void elv_unregister_queue(struct request_queue *q)
907{
Tejun Heof8fc8772011-12-14 00:33:40 +0100908 if (q) {
909 struct elevator_queue *e = q->elevator;
910
911 kobject_uevent(&e->kobj, KOBJ_REMOVE);
912 kobject_del(&e->kobj);
913 e->registered = 0;
914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
Mike Snitzer01effb02010-05-11 08:57:42 +0200916EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Tejun Heo3d3c2372011-12-14 00:33:42 +0100918int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100920 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200921
Tejun Heo3d3c2372011-12-14 00:33:42 +0100922 /* create icq_cache if requested */
923 if (e->icq_size) {
924 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
925 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
926 return -EINVAL;
927
928 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
929 "%s_io_cq", e->elevator_name);
930 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
931 e->icq_align, 0, NULL);
932 if (!e->icq_cache)
933 return -ENOMEM;
934 }
935
936 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200937 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100938 if (elevator_find(e->elevator_name)) {
939 spin_unlock(&elv_list_lock);
940 if (e->icq_cache)
941 kmem_cache_destroy(e->icq_cache);
942 return -EBUSY;
943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200945 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Tejun Heo3d3c2372011-12-14 00:33:42 +0100947 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100948 if (!strcmp(e->elevator_name, chosen_elevator) ||
949 (!*chosen_elevator &&
950 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100951 def = " (default)";
952
Jens Axboe4eb166d2008-02-01 00:37:27 +0100953 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
954 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100955 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956}
957EXPORT_SYMBOL_GPL(elv_register);
958
959void elv_unregister(struct elevator_type *e)
960{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100961 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200962 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200964 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100965
966 /*
967 * Destroy icq_cache if it exists. icq's are RCU managed. Make
968 * sure all RCU operations are complete before proceeding.
969 */
970 if (e->icq_cache) {
971 rcu_barrier();
972 kmem_cache_destroy(e->icq_cache);
973 e->icq_cache = NULL;
974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976EXPORT_SYMBOL_GPL(elv_unregister);
977
978/*
979 * switch to new_e io scheduler. be careful not to introduce deadlocks -
980 * we don't free the old io scheduler, before we have allocated what we
981 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200982 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 */
Jens Axboe165125e2007-07-24 09:28:11 +0200984static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985{
Jens Axboeb374d182008-10-31 10:05:07 +0100986 struct elevator_queue *old_elevator, *e;
Jens Axboe5dd531a2010-08-23 13:52:19 +0200987 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Tejun Heof8fc8772011-12-14 00:33:40 +0100989 /* allocate new elevator */
Jens Axboeb5deef92006-07-19 23:39:40 +0200990 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (!e)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200992 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Tejun Heof8fc8772011-12-14 00:33:40 +0100994 err = elevator_init_queue(q, e);
995 if (err) {
Jens Axboebc1c1162006-06-08 08:49:06 +0200996 kobject_put(&e->kobj);
Tejun Heof8fc8772011-12-14 00:33:40 +0100997 return err;
Jens Axboebc1c1162006-06-08 08:49:06 +0200998 }
999
Tejun Heof8fc8772011-12-14 00:33:40 +01001000 /* turn on BYPASS and drain all requests w/ elevator private data */
Jens Axboef600abe2009-04-08 14:22:01 +02001001 elv_quiesce_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001002
Tejun Heof8fc8772011-12-14 00:33:40 +01001003 /* unregister old queue, register new one and kill old elevator */
1004 if (q->elevator->registered) {
1005 elv_unregister_queue(q);
1006 err = __elv_register_queue(q, e);
Jens Axboe430c62f2010-10-07 09:35:16 +02001007 if (err)
1008 goto fail_register;
1009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
Tejun Heo7e5a8792011-12-14 00:33:42 +01001011 /* done, clear io_cq's, switch elevators and turn off BYPASS */
Tejun Heof8fc8772011-12-14 00:33:40 +01001012 spin_lock_irq(q->queue_lock);
Tejun Heo7e5a8792011-12-14 00:33:42 +01001013 ioc_clear_queue(q);
Tejun Heof8fc8772011-12-14 00:33:40 +01001014 old_elevator = q->elevator;
1015 q->elevator = e;
1016 spin_unlock_irq(q->queue_lock);
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 elevator_exit(old_elevator);
Jens Axboef600abe2009-04-08 14:22:01 +02001019 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001020
Tejun Heo22f746e2011-12-14 00:33:41 +01001021 blk_add_trace_msg(q, "elv switch: %s", e->type->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001022
Jens Axboe5dd531a2010-08-23 13:52:19 +02001023 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025fail_register:
1026 /*
1027 * switch failed, exit the new io scheduler and reattach the old
1028 * one again (along with re-adding the sysfs dir)
1029 */
1030 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 elv_register_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +02001032 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001033
Jens Axboe5dd531a2010-08-23 13:52:19 +02001034 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035}
1036
Jens Axboe5dd531a2010-08-23 13:52:19 +02001037/*
1038 * Switch this queue to the given IO scheduler.
1039 */
1040int elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
1042 char elevator_name[ELV_NAME_MAX];
1043 struct elevator_type *e;
1044
Martin K. Petersencd43e262009-05-22 17:17:52 -04001045 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +02001046 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001047
Li Zefanee2e9922008-10-14 08:49:56 +02001048 strlcpy(elevator_name, name, sizeof(elevator_name));
KOSAKI Motohiro8c279592009-10-09 08:48:08 +02001049 e = elevator_get(strstrip(elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (!e) {
1051 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1052 return -EINVAL;
1053 }
1054
Tejun Heo22f746e2011-12-14 00:33:41 +01001055 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -08001056 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +02001057 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -08001058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Jens Axboe5dd531a2010-08-23 13:52:19 +02001060 return elevator_switch(q, e);
1061}
1062EXPORT_SYMBOL(elevator_change);
1063
1064ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1065 size_t count)
1066{
1067 int ret;
1068
1069 if (!q->elevator)
1070 return count;
1071
1072 ret = elevator_change(q, name);
1073 if (!ret)
1074 return count;
1075
1076 printk(KERN_ERR "elevator: switch to %s failed\n", name);
1077 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
Jens Axboe165125e2007-07-24 09:28:11 +02001080ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
Jens Axboeb374d182008-10-31 10:05:07 +01001082 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001083 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001084 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 int len = 0;
1086
Mike Snitzere36f7242010-05-24 09:07:32 +02001087 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001088 return sprintf(name, "none\n");
1089
Tejun Heo22f746e2011-12-14 00:33:41 +01001090 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001091
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001092 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001093 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (!strcmp(elv->elevator_name, __e->elevator_name))
1095 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1096 else
1097 len += sprintf(name+len, "%s ", __e->elevator_name);
1098 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001099 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 len += sprintf(len+name, "\n");
1102 return len;
1103}
1104
Jens Axboe165125e2007-07-24 09:28:11 +02001105struct request *elv_rb_former_request(struct request_queue *q,
1106 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001107{
1108 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1109
1110 if (rbprev)
1111 return rb_entry_rq(rbprev);
1112
1113 return NULL;
1114}
Jens Axboe2e662b62006-07-13 11:55:04 +02001115EXPORT_SYMBOL(elv_rb_former_request);
1116
Jens Axboe165125e2007-07-24 09:28:11 +02001117struct request *elv_rb_latter_request(struct request_queue *q,
1118 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001119{
1120 struct rb_node *rbnext = rb_next(&rq->rb_node);
1121
1122 if (rbnext)
1123 return rb_entry_rq(rbnext);
1124
1125 return NULL;
1126}
Jens Axboe2e662b62006-07-13 11:55:04 +02001127EXPORT_SYMBOL(elv_rb_latter_request);