blob: 40f0c04e5ad3d076ee24d34f3424b9e42194d454 [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>
Tejun Heoeea8f412015-05-22 17:13:17 -040038#include <linux/blk-cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Li Zefan55782132009-06-09 13:43:05 +080040#include <trace/events/block.h>
41
Jens Axboe242f9dc2008-09-14 05:55:09 -070042#include "blk.h"
43
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 */
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070056static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio)
Jens Axboeda775262006-12-20 11:04:12 +010057{
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
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070061 if (e->type->ops.elevator_allow_bio_merge_fn)
62 return e->type->ops.elevator_allow_bio_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 */
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070070bool elv_bio_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))
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070073 return false;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020074
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070075 if (!elv_iosched_allow_bio_merge(rq, bio))
76 return false;
Jens Axboeda775262006-12-20 11:04:12 +010077
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070078 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079}
Tahsin Erdogan72ef7992016-07-07 11:48:22 -070080EXPORT_SYMBOL(elv_bio_merge_ok);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
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))
Chao Yu8406a4d2015-04-23 10:47:44 -0600160 return NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200161
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;
168}
Jianpeng Mad50235b2013-07-03 13:25:24 +0200169EXPORT_SYMBOL(elevator_alloc);
Al Viro3d1ab402006-03-18 18:35:43 -0500170
171static void elevator_release(struct kobject *kobj)
172{
Jens Axboeb374d182008-10-31 10:05:07 +0100173 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200174
Jens Axboeb374d182008-10-31 10:05:07 +0100175 e = container_of(kobj, struct elevator_queue, kobj);
Tejun Heo22f746e2011-12-14 00:33:41 +0100176 elevator_put(e->type);
Al Viro3d1ab402006-03-18 18:35:43 -0500177 kfree(e);
178}
179
Jens Axboe165125e2007-07-24 09:28:11 +0200180int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
182 struct elevator_type *e = NULL;
Tejun Heof8fc8772011-12-14 00:33:40 +0100183 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600185 /*
186 * q->sysfs_lock must be held to provide mutual exclusion between
187 * elevator_switch() and here.
188 */
189 lockdep_assert_held(&q->sysfs_lock);
190
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400191 if (unlikely(q->elevator))
192 return 0;
193
Tejun Heocb98fc82005-10-28 08:29:39 +0200194 INIT_LIST_HEAD(&q->queue_head);
195 q->last_merge = NULL;
196 q->end_sector = 0;
197 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200198
Jens Axboe4eb166d2008-02-01 00:37:27 +0100199 if (name) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800200 e = elevator_get(name, true);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100201 if (!e)
202 return -EINVAL;
203 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800205 /*
206 * Use the default elevator specified by config boot param or
207 * config option. Don't try to load modules as we could be running
208 * off async and request_module() isn't allowed from async.
209 */
Jens Axboe4eb166d2008-02-01 00:37:27 +0100210 if (!e && *chosen_elevator) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800211 e = elevator_get(chosen_elevator, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100212 if (!e)
213 printk(KERN_ERR "I/O scheduler %s not found\n",
214 chosen_elevator);
215 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100216
Jens Axboe4eb166d2008-02-01 00:37:27 +0100217 if (!e) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800218 e = elevator_get(CONFIG_DEFAULT_IOSCHED, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100219 if (!e) {
220 printk(KERN_ERR
221 "Default I/O scheduler not found. " \
222 "Using noop.\n");
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800223 e = elevator_get("noop", false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100224 }
Nate Diller5f003972006-01-24 10:07:58 +0100225 }
226
Jianpeng Mad50235b2013-07-03 13:25:24 +0200227 err = e->ops.elevator_init_fn(q, e);
Sudip Mukherjeed32f6b52014-10-23 22:16:48 +0530228 if (err)
229 elevator_put(e);
230 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
Jens Axboe2e662b62006-07-13 11:55:04 +0200232EXPORT_SYMBOL(elevator_init);
233
Jens Axboeb374d182008-10-31 10:05:07 +0100234void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Al Viro3d1ab402006-03-18 18:35:43 -0500236 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100237 if (e->type->ops.elevator_exit_fn)
238 e->type->ops.elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500239 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Al Viro3d1ab402006-03-18 18:35:43 -0500241 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
Jens Axboe2e662b62006-07-13 11:55:04 +0200243EXPORT_SYMBOL(elevator_exit);
244
Jens Axboe98170642006-07-28 09:23:08 +0200245static inline void __elv_rqhash_del(struct request *rq)
246{
Sasha Levin242d98f2012-12-17 10:01:27 -0500247 hash_del(&rq->hash);
Christoph Hellwige8064022016-10-20 15:12:13 +0200248 rq->rq_flags &= ~RQF_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200249}
250
Jens Axboe70b3ea02016-12-07 08:43:31 -0700251void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200252{
253 if (ELV_ON_HASH(rq))
254 __elv_rqhash_del(rq);
255}
256
Jens Axboe70b3ea02016-12-07 08:43:31 -0700257void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200258{
Jens Axboeb374d182008-10-31 10:05:07 +0100259 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200260
261 BUG_ON(ELV_ON_HASH(rq));
Sasha Levin242d98f2012-12-17 10:01:27 -0500262 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
Christoph Hellwige8064022016-10-20 15:12:13 +0200263 rq->rq_flags |= RQF_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200264}
265
Jens Axboe70b3ea02016-12-07 08:43:31 -0700266void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200267{
268 __elv_rqhash_del(rq);
269 elv_rqhash_add(q, rq);
270}
271
Jens Axboe70b3ea02016-12-07 08:43:31 -0700272struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200273{
Jens Axboeb374d182008-10-31 10:05:07 +0100274 struct elevator_queue *e = q->elevator;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800275 struct hlist_node *next;
Jens Axboe98170642006-07-28 09:23:08 +0200276 struct request *rq;
277
Linus Torvaldsee89f812013-02-28 12:52:24 -0800278 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
Jens Axboe98170642006-07-28 09:23:08 +0200279 BUG_ON(!ELV_ON_HASH(rq));
280
281 if (unlikely(!rq_mergeable(rq))) {
282 __elv_rqhash_del(rq);
283 continue;
284 }
285
286 if (rq_hash_key(rq) == offset)
287 return rq;
288 }
289
290 return NULL;
291}
292
Tejun Heo8922e162005-10-20 16:23:44 +0200293/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200294 * RB-tree support functions for inserting/lookup/removal of requests
295 * in a sorted RB tree.
296 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200297void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200298{
299 struct rb_node **p = &root->rb_node;
300 struct rb_node *parent = NULL;
301 struct request *__rq;
302
303 while (*p) {
304 parent = *p;
305 __rq = rb_entry(parent, struct request, rb_node);
306
Tejun Heo83096eb2009-05-07 22:24:39 +0900307 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200308 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200309 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200310 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200311 }
312
313 rb_link_node(&rq->rb_node, parent, p);
314 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200315}
Jens Axboe2e662b62006-07-13 11:55:04 +0200316EXPORT_SYMBOL(elv_rb_add);
317
318void elv_rb_del(struct rb_root *root, struct request *rq)
319{
320 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
321 rb_erase(&rq->rb_node, root);
322 RB_CLEAR_NODE(&rq->rb_node);
323}
Jens Axboe2e662b62006-07-13 11:55:04 +0200324EXPORT_SYMBOL(elv_rb_del);
325
326struct request *elv_rb_find(struct rb_root *root, sector_t sector)
327{
328 struct rb_node *n = root->rb_node;
329 struct request *rq;
330
331 while (n) {
332 rq = rb_entry(n, struct request, rb_node);
333
Tejun Heo83096eb2009-05-07 22:24:39 +0900334 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200335 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900336 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200337 n = n->rb_right;
338 else
339 return rq;
340 }
341
342 return NULL;
343}
Jens Axboe2e662b62006-07-13 11:55:04 +0200344EXPORT_SYMBOL(elv_rb_find);
345
346/*
Tejun Heo8922e162005-10-20 16:23:44 +0200347 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200348 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200349 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200350 */
Jens Axboe165125e2007-07-24 09:28:11 +0200351void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200352{
353 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200354 struct list_head *entry;
355
Tejun Heo06b86242005-10-20 16:46:23 +0200356 if (q->last_merge == rq)
357 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200358
359 elv_rqhash_del(q, rq);
360
Tejun Heo15853af2005-11-10 08:52:05 +0100361 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200362
Jens Axboe1b47f532005-10-20 16:37:00 +0200363 boundary = q->end_sector;
Tejun Heo8922e162005-10-20 16:23:44 +0200364 list_for_each_prev(entry, &q->queue_head) {
365 struct request *pos = list_entry_rq(entry);
366
Adrian Hunter7afafc82016-08-16 10:59:35 +0300367 if (req_op(rq) != req_op(pos))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100368 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100369 if (rq_data_dir(rq) != rq_data_dir(pos))
370 break;
Christoph Hellwige8064022016-10-20 15:12:13 +0200371 if (pos->rq_flags & (RQF_STARTED | RQF_SOFTBARRIER))
Tejun Heo8922e162005-10-20 16:23:44 +0200372 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900373 if (blk_rq_pos(rq) >= boundary) {
374 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200375 continue;
376 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900377 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200378 break;
379 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900380 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200381 break;
382 }
383
384 list_add(&rq->queuelist, entry);
385}
Jens Axboe2e662b62006-07-13 11:55:04 +0200386EXPORT_SYMBOL(elv_dispatch_sort);
387
Jens Axboe98170642006-07-28 09:23:08 +0200388/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200389 * Insert rq into dispatch queue of q. Queue lock must be held on
390 * entry. rq is added to the back of the dispatch queue. To be used by
391 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200392 */
393void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
394{
395 if (q->last_merge == rq)
396 q->last_merge = NULL;
397
398 elv_rqhash_del(q, rq);
399
400 q->nr_sorted--;
401
402 q->end_sector = rq_end_sector(rq);
403 q->boundary_rq = rq;
404 list_add_tail(&rq->queuelist, &q->queue_head);
405}
Jens Axboe2e662b62006-07-13 11:55:04 +0200406EXPORT_SYMBOL(elv_dispatch_add_tail);
407
Jens Axboe165125e2007-07-24 09:28:11 +0200408int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
Jens Axboeb374d182008-10-31 10:05:07 +0100410 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200411 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200412 int ret;
413
Jens Axboe98170642006-07-28 09:23:08 +0200414 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100415 * Levels of merges:
416 * nomerges: No merges at all attempted
417 * noxmerges: Only simple one-hit cache try
418 * merges: All merge tries attempted
419 */
Ming Lei7460d382015-10-20 23:13:55 +0800420 if (blk_queue_nomerges(q) || !bio_mergeable(bio))
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100421 return ELEVATOR_NO_MERGE;
422
423 /*
Jens Axboe98170642006-07-28 09:23:08 +0200424 * First try one-hit cache.
425 */
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700426 if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) {
Tejun Heo050c8ea2012-02-08 09:19:38 +0100427 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200428 if (ret != ELEVATOR_NO_MERGE) {
429 *req = q->last_merge;
430 return ret;
431 }
432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100434 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200435 return ELEVATOR_NO_MERGE;
436
Jens Axboe98170642006-07-28 09:23:08 +0200437 /*
438 * See if our hash lookup can find a potential backmerge.
439 */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700440 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700441 if (__rq && elv_bio_merge_ok(__rq, bio)) {
Jens Axboe98170642006-07-28 09:23:08 +0200442 *req = __rq;
443 return ELEVATOR_BACK_MERGE;
444 }
445
Tejun Heo22f746e2011-12-14 00:33:41 +0100446 if (e->type->ops.elevator_merge_fn)
447 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 return ELEVATOR_NO_MERGE;
450}
451
Jens Axboe5e84ea32011-03-21 10:14:27 +0100452/*
453 * Attempt to do an insertion back merge. Only check for the case where
454 * we can append 'rq' to an existing request, so we can throw 'rq' away
455 * afterwards.
456 *
457 * Returns true if we merged, false otherwise
458 */
459static bool elv_attempt_insert_merge(struct request_queue *q,
460 struct request *rq)
461{
462 struct request *__rq;
Shaohua Libee03932012-11-09 08:44:27 +0100463 bool ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100464
465 if (blk_queue_nomerges(q))
466 return false;
467
468 /*
469 * First try one-hit cache.
470 */
471 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
472 return true;
473
474 if (blk_queue_noxmerges(q))
475 return false;
476
Shaohua Libee03932012-11-09 08:44:27 +0100477 ret = false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100478 /*
479 * See if our hash lookup can find a potential backmerge.
480 */
Shaohua Libee03932012-11-09 08:44:27 +0100481 while (1) {
482 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
483 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
484 break;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100485
Shaohua Libee03932012-11-09 08:44:27 +0100486 /* The merged request could be merged with others, try again */
487 ret = true;
488 rq = __rq;
489 }
490
491 return ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100492}
493
Jens Axboe165125e2007-07-24 09:28:11 +0200494void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Jens Axboeb374d182008-10-31 10:05:07 +0100496 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Tejun Heo22f746e2011-12-14 00:33:41 +0100498 if (e->type->ops.elevator_merged_fn)
499 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200500
Jens Axboe2e662b62006-07-13 11:55:04 +0200501 if (type == ELEVATOR_BACK_MERGE)
502 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200503
Tejun Heo06b86242005-10-20 16:46:23 +0200504 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
Jens Axboe165125e2007-07-24 09:28:11 +0200507void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct request *next)
509{
Jens Axboeb374d182008-10-31 10:05:07 +0100510 struct elevator_queue *e = q->elevator;
Christoph Hellwige8064022016-10-20 15:12:13 +0200511 const int next_sorted = next->rq_flags & RQF_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Tejun Heo22f746e2011-12-14 00:33:41 +0100513 if (next_sorted && e->type->ops.elevator_merge_req_fn)
514 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200515
Jens Axboe98170642006-07-28 09:23:08 +0200516 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200517
Jens Axboe5e84ea32011-03-21 10:14:27 +0100518 if (next_sorted) {
519 elv_rqhash_del(q, next);
520 q->nr_sorted--;
521 }
522
Tejun Heo06b86242005-10-20 16:46:23 +0200523 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
Divyesh Shah812d4022010-04-08 21:14:23 -0700526void elv_bio_merged(struct request_queue *q, struct request *rq,
527 struct bio *bio)
528{
529 struct elevator_queue *e = q->elevator;
530
Tejun Heo22f746e2011-12-14 00:33:41 +0100531 if (e->type->ops.elevator_bio_merged_fn)
532 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700533}
534
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +0100535#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +0800536static void blk_pm_requeue_request(struct request *rq)
537{
Christoph Hellwige8064022016-10-20 15:12:13 +0200538 if (rq->q->dev && !(rq->rq_flags & RQF_PM))
Lin Mingc8158812013-03-23 11:42:27 +0800539 rq->q->nr_pending--;
540}
541
542static void blk_pm_add_request(struct request_queue *q, struct request *rq)
543{
Christoph Hellwige8064022016-10-20 15:12:13 +0200544 if (q->dev && !(rq->rq_flags & RQF_PM) && q->nr_pending++ == 0 &&
Lin Mingc8158812013-03-23 11:42:27 +0800545 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
546 pm_request_resume(q->dev);
547}
548#else
549static inline void blk_pm_requeue_request(struct request *rq) {}
550static inline void blk_pm_add_request(struct request_queue *q,
551 struct request *rq)
552{
553}
554#endif
555
Jens Axboe165125e2007-07-24 09:28:11 +0200556void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 /*
559 * it already went through dequeue, we need to decrement the
560 * in_flight count again
561 */
Tejun Heo8922e162005-10-20 16:23:44 +0200562 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200563 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwige8064022016-10-20 15:12:13 +0200564 if (rq->rq_flags & RQF_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100565 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Christoph Hellwige8064022016-10-20 15:12:13 +0200568 rq->rq_flags &= ~RQF_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Lin Mingc8158812013-03-23 11:42:27 +0800570 blk_pm_requeue_request(rq);
571
Jens Axboeb710a482011-03-30 09:52:30 +0200572 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573}
574
Jerome Marchand26308ea2009-03-27 10:31:51 +0100575void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100576{
577 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200578
579 lockdep_assert_held(q->queue_lock);
580
Tejun Heo22f746e2011-12-14 00:33:41 +0100581 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100582 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200583 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100584 printk(KERN_ERR "%s: forced dispatching is broken "
585 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100586 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100587 }
588}
589
Jens Axboeb710a482011-03-30 09:52:30 +0200590void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100592 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100593
Lin Mingc8158812013-03-23 11:42:27 +0800594 blk_pm_add_request(q, rq);
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 rq->q = q;
597
Christoph Hellwige8064022016-10-20 15:12:13 +0200598 if (rq->rq_flags & RQF_SOFTBARRIER) {
Jens Axboeb710a482011-03-30 09:52:30 +0200599 /* barriers are scheduling boundary, update end_sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400600 if (rq->cmd_type == REQ_TYPE_FS) {
Jens Axboeb710a482011-03-30 09:52:30 +0200601 q->end_sector = rq_end_sector(rq);
602 q->boundary_rq = rq;
603 }
Christoph Hellwige8064022016-10-20 15:12:13 +0200604 } else if (!(rq->rq_flags & RQF_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200605 (where == ELEVATOR_INSERT_SORT ||
606 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200607 where = ELEVATOR_INSERT_BACK;
608
Tejun Heo8922e162005-10-20 16:23:44 +0200609 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200610 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200611 case ELEVATOR_INSERT_FRONT:
Christoph Hellwige8064022016-10-20 15:12:13 +0200612 rq->rq_flags |= RQF_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200613 list_add(&rq->queuelist, &q->queue_head);
614 break;
615
616 case ELEVATOR_INSERT_BACK:
Christoph Hellwige8064022016-10-20 15:12:13 +0200617 rq->rq_flags |= RQF_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100618 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200619 list_add_tail(&rq->queuelist, &q->queue_head);
620 /*
621 * We kick the queue here for the following reasons.
622 * - The elevator might have returned NULL previously
623 * to delay requests and returned them now. As the
624 * queue wasn't empty before this request, ll_rw_blk
625 * won't run the queue on return, resulting in hang.
626 * - Usually, back inserted requests won't be merged
627 * with anything. There's no point in delaying queue
628 * processing.
629 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200630 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200631 break;
632
Jens Axboe5e84ea32011-03-21 10:14:27 +0100633 case ELEVATOR_INSERT_SORT_MERGE:
634 /*
635 * If we succeed in merging this request with one in the
636 * queue already, we are done - rq has now been freed,
637 * so no need to do anything further.
638 */
639 if (elv_attempt_insert_merge(q, rq))
640 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200641 case ELEVATOR_INSERT_SORT:
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400642 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
Christoph Hellwige8064022016-10-20 15:12:13 +0200643 rq->rq_flags |= RQF_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100644 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200645 if (rq_mergeable(rq)) {
646 elv_rqhash_add(q, rq);
647 if (!q->last_merge)
648 q->last_merge = rq;
649 }
650
Tejun Heoca235092005-11-01 17:23:49 +0900651 /*
652 * Some ioscheds (cfq) run q->request_fn directly, so
653 * rq cannot be accessed after calling
654 * elevator_add_req_fn.
655 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100656 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200657 break;
658
Tejun Heoae1b1532011-01-25 12:43:54 +0100659 case ELEVATOR_INSERT_FLUSH:
Christoph Hellwige8064022016-10-20 15:12:13 +0200660 rq->rq_flags |= RQF_SOFTBARRIER;
Tejun Heoae1b1532011-01-25 12:43:54 +0100661 blk_insert_flush(rq);
662 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200663 default:
664 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700665 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200666 BUG();
667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
Jens Axboe2e662b62006-07-13 11:55:04 +0200669EXPORT_SYMBOL(__elv_add_request);
670
Jens Axboe7eaceac2011-03-10 08:52:07 +0100671void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 unsigned long flags;
674
675 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100676 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 spin_unlock_irqrestore(q->queue_lock, flags);
678}
Jens Axboe2e662b62006-07-13 11:55:04 +0200679EXPORT_SYMBOL(elv_add_request);
680
Jens Axboe165125e2007-07-24 09:28:11 +0200681struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
Jens Axboeb374d182008-10-31 10:05:07 +0100683 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Tejun Heo22f746e2011-12-14 00:33:41 +0100685 if (e->type->ops.elevator_latter_req_fn)
686 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return NULL;
688}
689
Jens Axboe165125e2007-07-24 09:28:11 +0200690struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Jens Axboeb374d182008-10-31 10:05:07 +0100692 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Tejun Heo22f746e2011-12-14 00:33:41 +0100694 if (e->type->ops.elevator_former_req_fn)
695 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 return NULL;
697}
698
Tejun Heo852c7882012-03-05 13:15:27 -0800699int elv_set_request(struct request_queue *q, struct request *rq,
700 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
Jens Axboeb374d182008-10-31 10:05:07 +0100702 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Tejun Heo22f746e2011-12-14 00:33:41 +0100704 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800705 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return 0;
707}
708
Jens Axboe165125e2007-07-24 09:28:11 +0200709void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
Jens Axboeb374d182008-10-31 10:05:07 +0100711 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Tejun Heo22f746e2011-12-14 00:33:41 +0100713 if (e->type->ops.elevator_put_req_fn)
714 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Christoph Hellwigef295ec2016-10-28 08:48:16 -0600717int elv_may_queue(struct request_queue *q, unsigned int op)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Jens Axboeb374d182008-10-31 10:05:07 +0100719 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Tejun Heo22f746e2011-12-14 00:33:41 +0100721 if (e->type->ops.elevator_may_queue_fn)
Christoph Hellwigef295ec2016-10-28 08:48:16 -0600722 return e->type->ops.elevator_may_queue_fn(q, op);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
724 return ELV_MQUEUE_MAY;
725}
726
Jens Axboe165125e2007-07-24 09:28:11 +0200727void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
Jens Axboeb374d182008-10-31 10:05:07 +0100729 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 /*
732 * request is released from the driver, io must be done
733 */
Tejun Heo8922e162005-10-20 16:23:44 +0200734 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200735 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwige8064022016-10-20 15:12:13 +0200736 if ((rq->rq_flags & RQF_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100737 e->type->ops.elevator_completed_req_fn)
738 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740}
741
Al Viro3d1ab402006-03-18 18:35:43 -0500742#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
743
744static ssize_t
745elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
746{
Al Viro3d1ab402006-03-18 18:35:43 -0500747 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100748 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500749 ssize_t error;
750
751 if (!entry->show)
752 return -EIO;
753
Jens Axboeb374d182008-10-31 10:05:07 +0100754 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500755 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100756 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500757 mutex_unlock(&e->sysfs_lock);
758 return error;
759}
760
761static ssize_t
762elv_attr_store(struct kobject *kobj, struct attribute *attr,
763 const char *page, size_t length)
764{
Al Viro3d1ab402006-03-18 18:35:43 -0500765 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100766 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500767 ssize_t error;
768
769 if (!entry->store)
770 return -EIO;
771
Jens Axboeb374d182008-10-31 10:05:07 +0100772 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500773 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100774 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500775 mutex_unlock(&e->sysfs_lock);
776 return error;
777}
778
Emese Revfy52cf25d2010-01-19 02:58:23 +0100779static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500780 .show = elv_attr_show,
781 .store = elv_attr_store,
782};
783
784static struct kobj_type elv_ktype = {
785 .sysfs_ops = &elv_sysfs_ops,
786 .release = elevator_release,
787};
788
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800789int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800791 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500792 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700794 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500795 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100796 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500797 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500798 while (attr->attr.name) {
799 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500800 break;
Al Viroe572ec72006-03-18 22:27:18 -0500801 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500802 }
803 }
804 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200805 e->registered = 1;
Jens Axboe0bb97942015-06-10 08:01:20 -0600806 if (e->type->ops.elevator_registered_fn)
807 e->type->ops.elevator_registered_fn(q);
Al Viro3d1ab402006-03-18 18:35:43 -0500808 }
809 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
Tejun Heof8fc8772011-12-14 00:33:40 +0100811EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813void elv_unregister_queue(struct request_queue *q)
814{
Tejun Heof8fc8772011-12-14 00:33:40 +0100815 if (q) {
816 struct elevator_queue *e = q->elevator;
817
818 kobject_uevent(&e->kobj, KOBJ_REMOVE);
819 kobject_del(&e->kobj);
820 e->registered = 0;
821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}
Mike Snitzer01effb02010-05-11 08:57:42 +0200823EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Jens Axboee567bf72014-06-22 16:32:48 -0600825int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100827 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200828
Tejun Heo3d3c2372011-12-14 00:33:42 +0100829 /* create icq_cache if requested */
830 if (e->icq_size) {
831 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
832 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
833 return -EINVAL;
834
835 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
836 "%s_io_cq", e->elevator_name);
837 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
838 e->icq_align, 0, NULL);
839 if (!e->icq_cache)
840 return -ENOMEM;
841 }
842
843 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200844 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100845 if (elevator_find(e->elevator_name)) {
846 spin_unlock(&elv_list_lock);
847 if (e->icq_cache)
848 kmem_cache_destroy(e->icq_cache);
849 return -EBUSY;
850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200852 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
Tejun Heo3d3c2372011-12-14 00:33:42 +0100854 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100855 if (!strcmp(e->elevator_name, chosen_elevator) ||
856 (!*chosen_elevator &&
857 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100858 def = " (default)";
859
Jens Axboe4eb166d2008-02-01 00:37:27 +0100860 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
861 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100862 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
864EXPORT_SYMBOL_GPL(elv_register);
865
866void elv_unregister(struct elevator_type *e)
867{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100868 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200869 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200871 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100872
873 /*
874 * Destroy icq_cache if it exists. icq's are RCU managed. Make
875 * sure all RCU operations are complete before proceeding.
876 */
877 if (e->icq_cache) {
878 rcu_barrier();
879 kmem_cache_destroy(e->icq_cache);
880 e->icq_cache = NULL;
881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883EXPORT_SYMBOL_GPL(elv_unregister);
884
885/*
886 * switch to new_e io scheduler. be careful not to introduce deadlocks -
887 * we don't free the old io scheduler, before we have allocated what we
888 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200889 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 */
Jens Axboe165125e2007-07-24 09:28:11 +0200891static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800893 struct elevator_queue *old = q->elevator;
894 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800895 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800897 /*
898 * Turn on BYPASS and drain all requests w/ elevator private data.
899 * Block layer doesn't call into a quiesced elevator - all requests
900 * are directly put on the dispatch list without elevator data
901 * using INSERT_BACK. All requests have SOFTBARRIER set and no
902 * merge happens either.
903 */
Tejun Heod7325802012-03-05 13:14:58 -0800904 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200905
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800906 /* unregister and clear all auxiliary data of the old elevator */
907 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100908 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800909
910 spin_lock_irq(q->queue_lock);
911 ioc_clear_queue(q);
912 spin_unlock_irq(q->queue_lock);
913
914 /* allocate, init and register new elevator */
Jianpeng Mad50235b2013-07-03 13:25:24 +0200915 err = new_e->ops.elevator_init_fn(q, new_e);
916 if (err)
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800917 goto fail_init;
918
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800919 if (registered) {
920 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200921 if (err)
922 goto fail_register;
923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800925 /* done, kill the old one and finish */
926 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800927 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200928
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800929 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200930
Jens Axboe5dd531a2010-08-23 13:52:19 +0200931 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800934 elevator_exit(q->elevator);
935fail_init:
936 /* switch failed, restore and re-register old elevator */
937 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800939 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200940
Jens Axboe5dd531a2010-08-23 13:52:19 +0200941 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942}
943
Jens Axboe5dd531a2010-08-23 13:52:19 +0200944/*
945 * Switch this queue to the given IO scheduler.
946 */
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600947static int __elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
949 char elevator_name[ELV_NAME_MAX];
950 struct elevator_type *e;
951
Martin K. Petersencd43e262009-05-22 17:17:52 -0400952 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200953 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400954
Li Zefanee2e9922008-10-14 08:49:56 +0200955 strlcpy(elevator_name, name, sizeof(elevator_name));
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800956 e = elevator_get(strstrip(elevator_name), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 if (!e) {
958 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
959 return -EINVAL;
960 }
961
Tejun Heo22f746e2011-12-14 00:33:41 +0100962 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800963 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200964 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800965 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
Jens Axboe5dd531a2010-08-23 13:52:19 +0200967 return elevator_switch(q, e);
968}
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600969
970int elevator_change(struct request_queue *q, const char *name)
971{
972 int ret;
973
974 /* Protect q->elevator from elevator_init() */
975 mutex_lock(&q->sysfs_lock);
976 ret = __elevator_change(q, name);
977 mutex_unlock(&q->sysfs_lock);
978
979 return ret;
980}
Jens Axboe5dd531a2010-08-23 13:52:19 +0200981EXPORT_SYMBOL(elevator_change);
982
983ssize_t elv_iosched_store(struct request_queue *q, const char *name,
984 size_t count)
985{
986 int ret;
987
988 if (!q->elevator)
989 return count;
990
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600991 ret = __elevator_change(q, name);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200992 if (!ret)
993 return count;
994
995 printk(KERN_ERR "elevator: switch to %s failed\n", name);
996 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997}
998
Jens Axboe165125e2007-07-24 09:28:11 +0200999ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
Jens Axboeb374d182008-10-31 10:05:07 +01001001 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001002 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001003 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 int len = 0;
1005
Mike Snitzere36f7242010-05-24 09:07:32 +02001006 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001007 return sprintf(name, "none\n");
1008
Tejun Heo22f746e2011-12-14 00:33:41 +01001009 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001010
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001011 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001012 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (!strcmp(elv->elevator_name, __e->elevator_name))
1014 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1015 else
1016 len += sprintf(name+len, "%s ", __e->elevator_name);
1017 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001018 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 len += sprintf(len+name, "\n");
1021 return len;
1022}
1023
Jens Axboe165125e2007-07-24 09:28:11 +02001024struct request *elv_rb_former_request(struct request_queue *q,
1025 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001026{
1027 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1028
1029 if (rbprev)
1030 return rb_entry_rq(rbprev);
1031
1032 return NULL;
1033}
Jens Axboe2e662b62006-07-13 11:55:04 +02001034EXPORT_SYMBOL(elv_rb_former_request);
1035
Jens Axboe165125e2007-07-24 09:28:11 +02001036struct request *elv_rb_latter_request(struct request_queue *q,
1037 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001038{
1039 struct rb_node *rbnext = rb_next(&rq->rb_node);
1040
1041 if (rbnext)
1042 return rb_entry_rq(rbnext);
1043
1044 return NULL;
1045}
Jens Axboe2e662b62006-07-13 11:55:04 +02001046EXPORT_SYMBOL(elv_rb_latter_request);