blob: 59794d0d38e34604a24b6e7a63bf309570b2f8fb [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>
Lin Mingc8158812013-03-23 11:42:27 +080037#include <linux/pm_runtime.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Li Zefan55782132009-06-09 13:43:05 +080039#include <trace/events/block.h>
40
Jens Axboe242f9dc2008-09-14 05:55:09 -070041#include "blk.h"
Tejun Heo72e06c22012-03-05 13:15:00 -080042#include "blk-cgroup.h"
Jens Axboe242f9dc2008-09-14 05:55:09 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static DEFINE_SPINLOCK(elv_list_lock);
45static LIST_HEAD(elv_list);
46
47/*
Jens Axboe98170642006-07-28 09:23:08 +020048 * Merge hash stuff.
49 */
Tejun Heo83096eb2009-05-07 22:24:39 +090050#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
Jens Axboe98170642006-07-28 09:23:08 +020051
52/*
Jens Axboeda775262006-12-20 11:04:12 +010053 * Query io scheduler to see if the current process issuing bio may be
54 * merged with rq.
55 */
56static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
57{
Jens Axboe165125e2007-07-24 09:28:11 +020058 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010059 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010060
Tejun Heo22f746e2011-12-14 00:33:41 +010061 if (e->type->ops.elevator_allow_merge_fn)
62 return e->type->ops.elevator_allow_merge_fn(q, rq, bio);
Jens Axboeda775262006-12-20 11:04:12 +010063
64 return 1;
65}
66
67/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * can we safely merge with this request?
69 */
Tejun Heo050c8ea2012-02-08 09:19:38 +010070bool elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
Tejun Heo050c8ea2012-02-08 09:19:38 +010072 if (!blk_rq_merge_ok(rq, bio))
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020073 return 0;
74
Jens Axboeda775262006-12-20 11:04:12 +010075 if (!elv_iosched_allow_merge(rq, bio))
76 return 0;
77
78 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079}
80EXPORT_SYMBOL(elv_rq_merge_ok);
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082static struct elevator_type *elevator_find(const char *name)
83{
Vasily Tarasova22b1692006-10-11 09:24:27 +020084 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Matthias Kaehlcke70cee262007-07-10 12:26:24 +020086 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +020087 if (!strcmp(e->elevator_name, name))
88 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Vasily Tarasova22b1692006-10-11 09:24:27 +020091 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93
94static void elevator_put(struct elevator_type *e)
95{
96 module_put(e->elevator_owner);
97}
98
Tejun Heo21c3c5d2013-01-22 16:48:03 -080099static struct elevator_type *elevator_get(const char *name, bool try_loading)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
Tejun Heo2824bc932005-10-20 10:56:41 +0200101 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200103 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200104
105 e = elevator_find(name);
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800106 if (!e && try_loading) {
Jens Axboee1640942008-02-19 10:20:37 +0100107 spin_unlock(&elv_list_lock);
Kees Cook490b94b2011-05-05 18:02:12 -0600108 request_module("%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100109 spin_lock(&elv_list_lock);
110 e = elevator_find(name);
111 }
112
Tejun Heo2824bc932005-10-20 10:56:41 +0200113 if (e && !try_module_get(e->elevator_owner))
114 e = NULL;
115
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200116 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
118 return e;
119}
120
Wang Sheng-Hui484fc252011-09-08 12:32:14 +0200121static char chosen_elevator[ELV_NAME_MAX];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Nate Diller5f003972006-01-24 10:07:58 +0100123static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100125 /*
126 * Be backwards-compatible with previous kernels, so users
127 * won't get the wrong elevator.
128 */
Jens Axboe492af632009-10-03 09:37:51 +0200129 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800130 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
133__setup("elevator=", elevator_setup);
134
Tejun Heobb813f42013-01-18 14:05:56 -0800135/* called during boot to load the elevator chosen by the elevator param */
136void __init load_default_elevator_module(void)
137{
138 struct elevator_type *e;
139
140 if (!chosen_elevator[0])
141 return;
142
143 spin_lock(&elv_list_lock);
144 e = elevator_find(chosen_elevator);
145 spin_unlock(&elv_list_lock);
146
147 if (!e)
148 request_module("%s-iosched", chosen_elevator);
149}
150
Al Viro3d1ab402006-03-18 18:35:43 -0500151static struct kobj_type elv_ktype;
152
Jianpeng Mad50235b2013-07-03 13:25:24 +0200153struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200154 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500155{
Jens Axboeb374d182008-10-31 10:05:07 +0100156 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200157
Joe Perchesc1b511e2013-08-29 15:21:42 -0700158 eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200159 if (unlikely(!eq))
160 goto err;
161
Tejun Heo22f746e2011-12-14 00:33:41 +0100162 eq->type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700163 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200164 mutex_init(&eq->sysfs_lock);
Sasha Levin242d98f2012-12-17 10:01:27 -0500165 hash_init(eq->hash);
Jens Axboe98170642006-07-28 09:23:08 +0200166
Al Viro3d1ab402006-03-18 18:35:43 -0500167 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200168err:
169 kfree(eq);
170 elevator_put(e);
171 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500172}
Jianpeng Mad50235b2013-07-03 13:25:24 +0200173EXPORT_SYMBOL(elevator_alloc);
Al Viro3d1ab402006-03-18 18:35:43 -0500174
175static void elevator_release(struct kobject *kobj)
176{
Jens Axboeb374d182008-10-31 10:05:07 +0100177 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200178
Jens Axboeb374d182008-10-31 10:05:07 +0100179 e = container_of(kobj, struct elevator_queue, kobj);
Tejun Heo22f746e2011-12-14 00:33:41 +0100180 elevator_put(e->type);
Al Viro3d1ab402006-03-18 18:35:43 -0500181 kfree(e);
182}
183
Jens Axboe165125e2007-07-24 09:28:11 +0200184int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
186 struct elevator_type *e = NULL;
Tejun Heof8fc8772011-12-14 00:33:40 +0100187 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600189 /*
190 * q->sysfs_lock must be held to provide mutual exclusion between
191 * elevator_switch() and here.
192 */
193 lockdep_assert_held(&q->sysfs_lock);
194
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400195 if (unlikely(q->elevator))
196 return 0;
197
Tejun Heocb98fc82005-10-28 08:29:39 +0200198 INIT_LIST_HEAD(&q->queue_head);
199 q->last_merge = NULL;
200 q->end_sector = 0;
201 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200202
Jens Axboe4eb166d2008-02-01 00:37:27 +0100203 if (name) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800204 e = elevator_get(name, true);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100205 if (!e)
206 return -EINVAL;
207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800209 /*
210 * Use the default elevator specified by config boot param or
211 * config option. Don't try to load modules as we could be running
212 * off async and request_module() isn't allowed from async.
213 */
Jens Axboe4eb166d2008-02-01 00:37:27 +0100214 if (!e && *chosen_elevator) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800215 e = elevator_get(chosen_elevator, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100216 if (!e)
217 printk(KERN_ERR "I/O scheduler %s not found\n",
218 chosen_elevator);
219 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100220
Jens Axboe4eb166d2008-02-01 00:37:27 +0100221 if (!e) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800222 e = elevator_get(CONFIG_DEFAULT_IOSCHED, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100223 if (!e) {
224 printk(KERN_ERR
225 "Default I/O scheduler not found. " \
226 "Using noop.\n");
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800227 e = elevator_get("noop", false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100228 }
Nate Diller5f003972006-01-24 10:07:58 +0100229 }
230
Jianpeng Mad50235b2013-07-03 13:25:24 +0200231 err = e->ops.elevator_init_fn(q, e);
Sudip Mukherjeed32f6b52014-10-23 22:16:48 +0530232 if (err)
233 elevator_put(e);
234 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
Jens Axboe2e662b62006-07-13 11:55:04 +0200236EXPORT_SYMBOL(elevator_init);
237
Jens Axboeb374d182008-10-31 10:05:07 +0100238void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
Al Viro3d1ab402006-03-18 18:35:43 -0500240 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100241 if (e->type->ops.elevator_exit_fn)
242 e->type->ops.elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500243 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Al Viro3d1ab402006-03-18 18:35:43 -0500245 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
Jens Axboe2e662b62006-07-13 11:55:04 +0200247EXPORT_SYMBOL(elevator_exit);
248
Jens Axboe98170642006-07-28 09:23:08 +0200249static inline void __elv_rqhash_del(struct request *rq)
250{
Sasha Levin242d98f2012-12-17 10:01:27 -0500251 hash_del(&rq->hash);
Jens Axboe360f92c2014-04-09 20:27:01 -0600252 rq->cmd_flags &= ~REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200253}
254
Jens Axboe165125e2007-07-24 09:28:11 +0200255static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200256{
257 if (ELV_ON_HASH(rq))
258 __elv_rqhash_del(rq);
259}
260
Jens Axboe165125e2007-07-24 09:28:11 +0200261static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200262{
Jens Axboeb374d182008-10-31 10:05:07 +0100263 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200264
265 BUG_ON(ELV_ON_HASH(rq));
Sasha Levin242d98f2012-12-17 10:01:27 -0500266 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
Jens Axboe360f92c2014-04-09 20:27:01 -0600267 rq->cmd_flags |= REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200268}
269
Jens Axboe165125e2007-07-24 09:28:11 +0200270static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200271{
272 __elv_rqhash_del(rq);
273 elv_rqhash_add(q, rq);
274}
275
Jens Axboe165125e2007-07-24 09:28:11 +0200276static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200277{
Jens Axboeb374d182008-10-31 10:05:07 +0100278 struct elevator_queue *e = q->elevator;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800279 struct hlist_node *next;
Jens Axboe98170642006-07-28 09:23:08 +0200280 struct request *rq;
281
Linus Torvaldsee89f812013-02-28 12:52:24 -0800282 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
Jens Axboe98170642006-07-28 09:23:08 +0200283 BUG_ON(!ELV_ON_HASH(rq));
284
285 if (unlikely(!rq_mergeable(rq))) {
286 __elv_rqhash_del(rq);
287 continue;
288 }
289
290 if (rq_hash_key(rq) == offset)
291 return rq;
292 }
293
294 return NULL;
295}
296
Tejun Heo8922e162005-10-20 16:23:44 +0200297/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200298 * RB-tree support functions for inserting/lookup/removal of requests
299 * in a sorted RB tree.
300 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200301void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200302{
303 struct rb_node **p = &root->rb_node;
304 struct rb_node *parent = NULL;
305 struct request *__rq;
306
307 while (*p) {
308 parent = *p;
309 __rq = rb_entry(parent, struct request, rb_node);
310
Tejun Heo83096eb2009-05-07 22:24:39 +0900311 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200312 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200313 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200314 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200315 }
316
317 rb_link_node(&rq->rb_node, parent, p);
318 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200319}
Jens Axboe2e662b62006-07-13 11:55:04 +0200320EXPORT_SYMBOL(elv_rb_add);
321
322void elv_rb_del(struct rb_root *root, struct request *rq)
323{
324 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
325 rb_erase(&rq->rb_node, root);
326 RB_CLEAR_NODE(&rq->rb_node);
327}
Jens Axboe2e662b62006-07-13 11:55:04 +0200328EXPORT_SYMBOL(elv_rb_del);
329
330struct request *elv_rb_find(struct rb_root *root, sector_t sector)
331{
332 struct rb_node *n = root->rb_node;
333 struct request *rq;
334
335 while (n) {
336 rq = rb_entry(n, struct request, rb_node);
337
Tejun Heo83096eb2009-05-07 22:24:39 +0900338 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200339 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900340 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200341 n = n->rb_right;
342 else
343 return rq;
344 }
345
346 return NULL;
347}
Jens Axboe2e662b62006-07-13 11:55:04 +0200348EXPORT_SYMBOL(elv_rb_find);
349
350/*
Tejun Heo8922e162005-10-20 16:23:44 +0200351 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200352 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200353 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200354 */
Jens Axboe165125e2007-07-24 09:28:11 +0200355void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200356{
357 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200358 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100359 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200360
Tejun Heo06b86242005-10-20 16:46:23 +0200361 if (q->last_merge == rq)
362 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200363
364 elv_rqhash_del(q, rq);
365
Tejun Heo15853af2005-11-10 08:52:05 +0100366 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200367
Jens Axboe1b47f532005-10-20 16:37:00 +0200368 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100369 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200370 list_for_each_prev(entry, &q->queue_head) {
371 struct request *pos = list_entry_rq(entry);
372
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200373 if ((rq->cmd_flags & REQ_DISCARD) !=
374 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100375 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100376 if (rq_data_dir(rq) != rq_data_dir(pos))
377 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100378 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200379 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900380 if (blk_rq_pos(rq) >= boundary) {
381 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200382 continue;
383 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900384 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200385 break;
386 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900387 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200388 break;
389 }
390
391 list_add(&rq->queuelist, entry);
392}
Jens Axboe2e662b62006-07-13 11:55:04 +0200393EXPORT_SYMBOL(elv_dispatch_sort);
394
Jens Axboe98170642006-07-28 09:23:08 +0200395/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200396 * Insert rq into dispatch queue of q. Queue lock must be held on
397 * entry. rq is added to the back of the dispatch queue. To be used by
398 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200399 */
400void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
401{
402 if (q->last_merge == rq)
403 q->last_merge = NULL;
404
405 elv_rqhash_del(q, rq);
406
407 q->nr_sorted--;
408
409 q->end_sector = rq_end_sector(rq);
410 q->boundary_rq = rq;
411 list_add_tail(&rq->queuelist, &q->queue_head);
412}
Jens Axboe2e662b62006-07-13 11:55:04 +0200413EXPORT_SYMBOL(elv_dispatch_add_tail);
414
Jens Axboe165125e2007-07-24 09:28:11 +0200415int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
Jens Axboeb374d182008-10-31 10:05:07 +0100417 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200418 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200419 int ret;
420
Jens Axboe98170642006-07-28 09:23:08 +0200421 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100422 * Levels of merges:
423 * nomerges: No merges at all attempted
424 * noxmerges: Only simple one-hit cache try
425 * merges: All merge tries attempted
426 */
427 if (blk_queue_nomerges(q))
428 return ELEVATOR_NO_MERGE;
429
430 /*
Jens Axboe98170642006-07-28 09:23:08 +0200431 * First try one-hit cache.
432 */
Tejun Heo050c8ea2012-02-08 09:19:38 +0100433 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
434 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200435 if (ret != ELEVATOR_NO_MERGE) {
436 *req = q->last_merge;
437 return ret;
438 }
439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100441 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200442 return ELEVATOR_NO_MERGE;
443
Jens Axboe98170642006-07-28 09:23:08 +0200444 /*
445 * See if our hash lookup can find a potential backmerge.
446 */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700447 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
Jens Axboe98170642006-07-28 09:23:08 +0200448 if (__rq && elv_rq_merge_ok(__rq, bio)) {
449 *req = __rq;
450 return ELEVATOR_BACK_MERGE;
451 }
452
Tejun Heo22f746e2011-12-14 00:33:41 +0100453 if (e->type->ops.elevator_merge_fn)
454 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456 return ELEVATOR_NO_MERGE;
457}
458
Jens Axboe5e84ea32011-03-21 10:14:27 +0100459/*
460 * Attempt to do an insertion back merge. Only check for the case where
461 * we can append 'rq' to an existing request, so we can throw 'rq' away
462 * afterwards.
463 *
464 * Returns true if we merged, false otherwise
465 */
466static bool elv_attempt_insert_merge(struct request_queue *q,
467 struct request *rq)
468{
469 struct request *__rq;
Shaohua Libee03932012-11-09 08:44:27 +0100470 bool ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100471
472 if (blk_queue_nomerges(q))
473 return false;
474
475 /*
476 * First try one-hit cache.
477 */
478 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
479 return true;
480
481 if (blk_queue_noxmerges(q))
482 return false;
483
Shaohua Libee03932012-11-09 08:44:27 +0100484 ret = false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100485 /*
486 * See if our hash lookup can find a potential backmerge.
487 */
Shaohua Libee03932012-11-09 08:44:27 +0100488 while (1) {
489 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
490 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
491 break;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100492
Shaohua Libee03932012-11-09 08:44:27 +0100493 /* The merged request could be merged with others, try again */
494 ret = true;
495 rq = __rq;
496 }
497
498 return ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100499}
500
Jens Axboe165125e2007-07-24 09:28:11 +0200501void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Jens Axboeb374d182008-10-31 10:05:07 +0100503 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Tejun Heo22f746e2011-12-14 00:33:41 +0100505 if (e->type->ops.elevator_merged_fn)
506 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200507
Jens Axboe2e662b62006-07-13 11:55:04 +0200508 if (type == ELEVATOR_BACK_MERGE)
509 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200510
Tejun Heo06b86242005-10-20 16:46:23 +0200511 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512}
513
Jens Axboe165125e2007-07-24 09:28:11 +0200514void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 struct request *next)
516{
Jens Axboeb374d182008-10-31 10:05:07 +0100517 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100518 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Tejun Heo22f746e2011-12-14 00:33:41 +0100520 if (next_sorted && e->type->ops.elevator_merge_req_fn)
521 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200522
Jens Axboe98170642006-07-28 09:23:08 +0200523 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200524
Jens Axboe5e84ea32011-03-21 10:14:27 +0100525 if (next_sorted) {
526 elv_rqhash_del(q, next);
527 q->nr_sorted--;
528 }
529
Tejun Heo06b86242005-10-20 16:46:23 +0200530 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
Divyesh Shah812d4022010-04-08 21:14:23 -0700533void elv_bio_merged(struct request_queue *q, struct request *rq,
534 struct bio *bio)
535{
536 struct elevator_queue *e = q->elevator;
537
Tejun Heo22f746e2011-12-14 00:33:41 +0100538 if (e->type->ops.elevator_bio_merged_fn)
539 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700540}
541
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +0100542#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +0800543static void blk_pm_requeue_request(struct request *rq)
544{
545 if (rq->q->dev && !(rq->cmd_flags & REQ_PM))
546 rq->q->nr_pending--;
547}
548
549static void blk_pm_add_request(struct request_queue *q, struct request *rq)
550{
551 if (q->dev && !(rq->cmd_flags & REQ_PM) && q->nr_pending++ == 0 &&
552 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
553 pm_request_resume(q->dev);
554}
555#else
556static inline void blk_pm_requeue_request(struct request *rq) {}
557static inline void blk_pm_add_request(struct request_queue *q,
558 struct request *rq)
559{
560}
561#endif
562
Jens Axboe165125e2007-07-24 09:28:11 +0200563void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 /*
566 * it already went through dequeue, we need to decrement the
567 * in_flight count again
568 */
Tejun Heo8922e162005-10-20 16:23:44 +0200569 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200570 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200571 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100572 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Jens Axboe4aff5e22006-08-10 08:44:47 +0200575 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Lin Mingc8158812013-03-23 11:42:27 +0800577 blk_pm_requeue_request(rq);
578
Jens Axboeb710a482011-03-30 09:52:30 +0200579 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580}
581
Jerome Marchand26308ea2009-03-27 10:31:51 +0100582void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100583{
584 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200585
586 lockdep_assert_held(q->queue_lock);
587
Tejun Heo22f746e2011-12-14 00:33:41 +0100588 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100589 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200590 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100591 printk(KERN_ERR "%s: forced dispatching is broken "
592 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100593 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100594 }
595}
596
Jens Axboeb710a482011-03-30 09:52:30 +0200597void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100599 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100600
Lin Mingc8158812013-03-23 11:42:27 +0800601 blk_pm_add_request(q, rq);
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 rq->q = q;
604
Jens Axboeb710a482011-03-30 09:52:30 +0200605 if (rq->cmd_flags & REQ_SOFTBARRIER) {
606 /* barriers are scheduling boundary, update end_sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400607 if (rq->cmd_type == REQ_TYPE_FS) {
Jens Axboeb710a482011-03-30 09:52:30 +0200608 q->end_sector = rq_end_sector(rq);
609 q->boundary_rq = rq;
610 }
611 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200612 (where == ELEVATOR_INSERT_SORT ||
613 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200614 where = ELEVATOR_INSERT_BACK;
615
Tejun Heo8922e162005-10-20 16:23:44 +0200616 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200617 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200618 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200619 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200620 list_add(&rq->queuelist, &q->queue_head);
621 break;
622
623 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200624 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100625 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200626 list_add_tail(&rq->queuelist, &q->queue_head);
627 /*
628 * We kick the queue here for the following reasons.
629 * - The elevator might have returned NULL previously
630 * to delay requests and returned them now. As the
631 * queue wasn't empty before this request, ll_rw_blk
632 * won't run the queue on return, resulting in hang.
633 * - Usually, back inserted requests won't be merged
634 * with anything. There's no point in delaying queue
635 * processing.
636 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200637 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200638 break;
639
Jens Axboe5e84ea32011-03-21 10:14:27 +0100640 case ELEVATOR_INSERT_SORT_MERGE:
641 /*
642 * If we succeed in merging this request with one in the
643 * queue already, we are done - rq has now been freed,
644 * so no need to do anything further.
645 */
646 if (elv_attempt_insert_merge(q, rq))
647 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200648 case ELEVATOR_INSERT_SORT:
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400649 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200650 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100651 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200652 if (rq_mergeable(rq)) {
653 elv_rqhash_add(q, rq);
654 if (!q->last_merge)
655 q->last_merge = rq;
656 }
657
Tejun Heoca235092005-11-01 17:23:49 +0900658 /*
659 * Some ioscheds (cfq) run q->request_fn directly, so
660 * rq cannot be accessed after calling
661 * elevator_add_req_fn.
662 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100663 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200664 break;
665
Tejun Heoae1b1532011-01-25 12:43:54 +0100666 case ELEVATOR_INSERT_FLUSH:
667 rq->cmd_flags |= REQ_SOFTBARRIER;
668 blk_insert_flush(rq);
669 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200670 default:
671 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700672 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200673 BUG();
674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
Jens Axboe2e662b62006-07-13 11:55:04 +0200676EXPORT_SYMBOL(__elv_add_request);
677
Jens Axboe7eaceac2011-03-10 08:52:07 +0100678void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680 unsigned long flags;
681
682 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100683 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 spin_unlock_irqrestore(q->queue_lock, flags);
685}
Jens Axboe2e662b62006-07-13 11:55:04 +0200686EXPORT_SYMBOL(elv_add_request);
687
Jens Axboe165125e2007-07-24 09:28:11 +0200688struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
Jens Axboeb374d182008-10-31 10:05:07 +0100690 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Tejun Heo22f746e2011-12-14 00:33:41 +0100692 if (e->type->ops.elevator_latter_req_fn)
693 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 return NULL;
695}
696
Jens Axboe165125e2007-07-24 09:28:11 +0200697struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Jens Axboeb374d182008-10-31 10:05:07 +0100699 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Tejun Heo22f746e2011-12-14 00:33:41 +0100701 if (e->type->ops.elevator_former_req_fn)
702 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return NULL;
704}
705
Tejun Heo852c7882012-03-05 13:15:27 -0800706int elv_set_request(struct request_queue *q, struct request *rq,
707 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
Jens Axboeb374d182008-10-31 10:05:07 +0100709 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Tejun Heo22f746e2011-12-14 00:33:41 +0100711 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800712 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 return 0;
714}
715
Jens Axboe165125e2007-07-24 09:28:11 +0200716void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
Jens Axboeb374d182008-10-31 10:05:07 +0100718 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Tejun Heo22f746e2011-12-14 00:33:41 +0100720 if (e->type->ops.elevator_put_req_fn)
721 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
Jens Axboe165125e2007-07-24 09:28:11 +0200724int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Jens Axboeb374d182008-10-31 10:05:07 +0100726 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Tejun Heo22f746e2011-12-14 00:33:41 +0100728 if (e->type->ops.elevator_may_queue_fn)
729 return e->type->ops.elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 return ELV_MQUEUE_MAY;
732}
733
Jens Axboe165125e2007-07-24 09:28:11 +0200734void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Jens Axboeb374d182008-10-31 10:05:07 +0100736 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
738 /*
739 * request is released from the driver, io must be done
740 */
Tejun Heo8922e162005-10-20 16:23:44 +0200741 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200742 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200743 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100744 e->type->ops.elevator_completed_req_fn)
745 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100746 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
Al Viro3d1ab402006-03-18 18:35:43 -0500749#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
750
751static ssize_t
752elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
753{
Al Viro3d1ab402006-03-18 18:35:43 -0500754 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100755 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500756 ssize_t error;
757
758 if (!entry->show)
759 return -EIO;
760
Jens Axboeb374d182008-10-31 10:05:07 +0100761 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500762 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100763 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500764 mutex_unlock(&e->sysfs_lock);
765 return error;
766}
767
768static ssize_t
769elv_attr_store(struct kobject *kobj, struct attribute *attr,
770 const char *page, size_t length)
771{
Al Viro3d1ab402006-03-18 18:35:43 -0500772 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100773 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500774 ssize_t error;
775
776 if (!entry->store)
777 return -EIO;
778
Jens Axboeb374d182008-10-31 10:05:07 +0100779 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500780 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100781 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500782 mutex_unlock(&e->sysfs_lock);
783 return error;
784}
785
Emese Revfy52cf25d2010-01-19 02:58:23 +0100786static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500787 .show = elv_attr_show,
788 .store = elv_attr_store,
789};
790
791static struct kobj_type elv_ktype = {
792 .sysfs_ops = &elv_sysfs_ops,
793 .release = elevator_release,
794};
795
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800796int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800798 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500799 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700801 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500802 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100803 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500804 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500805 while (attr->attr.name) {
806 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500807 break;
Al Viroe572ec72006-03-18 22:27:18 -0500808 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500809 }
810 }
811 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200812 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500813 }
814 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815}
Tejun Heof8fc8772011-12-14 00:33:40 +0100816EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818void elv_unregister_queue(struct request_queue *q)
819{
Tejun Heof8fc8772011-12-14 00:33:40 +0100820 if (q) {
821 struct elevator_queue *e = q->elevator;
822
823 kobject_uevent(&e->kobj, KOBJ_REMOVE);
824 kobject_del(&e->kobj);
825 e->registered = 0;
826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
Mike Snitzer01effb02010-05-11 08:57:42 +0200828EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
Jens Axboee567bf72014-06-22 16:32:48 -0600830int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100832 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200833
Tejun Heo3d3c2372011-12-14 00:33:42 +0100834 /* create icq_cache if requested */
835 if (e->icq_size) {
836 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
837 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
838 return -EINVAL;
839
840 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
841 "%s_io_cq", e->elevator_name);
842 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
843 e->icq_align, 0, NULL);
844 if (!e->icq_cache)
845 return -ENOMEM;
846 }
847
848 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200849 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100850 if (elevator_find(e->elevator_name)) {
851 spin_unlock(&elv_list_lock);
852 if (e->icq_cache)
853 kmem_cache_destroy(e->icq_cache);
854 return -EBUSY;
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200857 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Tejun Heo3d3c2372011-12-14 00:33:42 +0100859 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100860 if (!strcmp(e->elevator_name, chosen_elevator) ||
861 (!*chosen_elevator &&
862 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100863 def = " (default)";
864
Jens Axboe4eb166d2008-02-01 00:37:27 +0100865 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
866 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100867 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868}
869EXPORT_SYMBOL_GPL(elv_register);
870
871void elv_unregister(struct elevator_type *e)
872{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100873 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200874 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200876 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100877
878 /*
879 * Destroy icq_cache if it exists. icq's are RCU managed. Make
880 * sure all RCU operations are complete before proceeding.
881 */
882 if (e->icq_cache) {
883 rcu_barrier();
884 kmem_cache_destroy(e->icq_cache);
885 e->icq_cache = NULL;
886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888EXPORT_SYMBOL_GPL(elv_unregister);
889
890/*
891 * switch to new_e io scheduler. be careful not to introduce deadlocks -
892 * we don't free the old io scheduler, before we have allocated what we
893 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200894 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 */
Jens Axboe165125e2007-07-24 09:28:11 +0200896static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800898 struct elevator_queue *old = q->elevator;
899 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800900 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800902 /*
903 * Turn on BYPASS and drain all requests w/ elevator private data.
904 * Block layer doesn't call into a quiesced elevator - all requests
905 * are directly put on the dispatch list without elevator data
906 * using INSERT_BACK. All requests have SOFTBARRIER set and no
907 * merge happens either.
908 */
Tejun Heod7325802012-03-05 13:14:58 -0800909 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200910
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800911 /* unregister and clear all auxiliary data of the old elevator */
912 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100913 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800914
915 spin_lock_irq(q->queue_lock);
916 ioc_clear_queue(q);
917 spin_unlock_irq(q->queue_lock);
918
919 /* allocate, init and register new elevator */
Jianpeng Mad50235b2013-07-03 13:25:24 +0200920 err = new_e->ops.elevator_init_fn(q, new_e);
921 if (err)
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800922 goto fail_init;
923
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800924 if (registered) {
925 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200926 if (err)
927 goto fail_register;
928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800930 /* done, kill the old one and finish */
931 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800932 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200933
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800934 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200935
Jens Axboe5dd531a2010-08-23 13:52:19 +0200936 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800939 elevator_exit(q->elevator);
940fail_init:
941 /* switch failed, restore and re-register old elevator */
942 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800944 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200945
Jens Axboe5dd531a2010-08-23 13:52:19 +0200946 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947}
948
Jens Axboe5dd531a2010-08-23 13:52:19 +0200949/*
950 * Switch this queue to the given IO scheduler.
951 */
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600952static int __elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
954 char elevator_name[ELV_NAME_MAX];
955 struct elevator_type *e;
956
Martin K. Petersencd43e262009-05-22 17:17:52 -0400957 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200958 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400959
Li Zefanee2e9922008-10-14 08:49:56 +0200960 strlcpy(elevator_name, name, sizeof(elevator_name));
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800961 e = elevator_get(strstrip(elevator_name), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 if (!e) {
963 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
964 return -EINVAL;
965 }
966
Tejun Heo22f746e2011-12-14 00:33:41 +0100967 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800968 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200969 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Jens Axboe5dd531a2010-08-23 13:52:19 +0200972 return elevator_switch(q, e);
973}
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600974
975int elevator_change(struct request_queue *q, const char *name)
976{
977 int ret;
978
979 /* Protect q->elevator from elevator_init() */
980 mutex_lock(&q->sysfs_lock);
981 ret = __elevator_change(q, name);
982 mutex_unlock(&q->sysfs_lock);
983
984 return ret;
985}
Jens Axboe5dd531a2010-08-23 13:52:19 +0200986EXPORT_SYMBOL(elevator_change);
987
988ssize_t elv_iosched_store(struct request_queue *q, const char *name,
989 size_t count)
990{
991 int ret;
992
993 if (!q->elevator)
994 return count;
995
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600996 ret = __elevator_change(q, name);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200997 if (!ret)
998 return count;
999
1000 printk(KERN_ERR "elevator: switch to %s failed\n", name);
1001 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002}
1003
Jens Axboe165125e2007-07-24 09:28:11 +02001004ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005{
Jens Axboeb374d182008-10-31 10:05:07 +01001006 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001007 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001008 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 int len = 0;
1010
Mike Snitzere36f7242010-05-24 09:07:32 +02001011 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001012 return sprintf(name, "none\n");
1013
Tejun Heo22f746e2011-12-14 00:33:41 +01001014 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001015
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001016 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001017 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (!strcmp(elv->elevator_name, __e->elevator_name))
1019 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1020 else
1021 len += sprintf(name+len, "%s ", __e->elevator_name);
1022 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001023 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 len += sprintf(len+name, "\n");
1026 return len;
1027}
1028
Jens Axboe165125e2007-07-24 09:28:11 +02001029struct request *elv_rb_former_request(struct request_queue *q,
1030 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001031{
1032 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1033
1034 if (rbprev)
1035 return rb_entry_rq(rbprev);
1036
1037 return NULL;
1038}
Jens Axboe2e662b62006-07-13 11:55:04 +02001039EXPORT_SYMBOL(elv_rb_former_request);
1040
Jens Axboe165125e2007-07-24 09:28:11 +02001041struct request *elv_rb_latter_request(struct request_queue *q,
1042 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001043{
1044 struct rb_node *rbnext = rb_next(&rq->rb_node);
1045
1046 if (rbnext)
1047 return rb_entry_rq(rbnext);
1048
1049 return NULL;
1050}
Jens Axboe2e662b62006-07-13 11:55:04 +02001051EXPORT_SYMBOL(elv_rb_latter_request);