blob: f35edddfe9b5421e38e39d8e9e9a43e61e7ca2f4 [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);
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400232 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
Jens Axboe2e662b62006-07-13 11:55:04 +0200234EXPORT_SYMBOL(elevator_init);
235
Jens Axboeb374d182008-10-31 10:05:07 +0100236void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Al Viro3d1ab402006-03-18 18:35:43 -0500238 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100239 if (e->type->ops.elevator_exit_fn)
240 e->type->ops.elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500241 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Al Viro3d1ab402006-03-18 18:35:43 -0500243 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
Jens Axboe2e662b62006-07-13 11:55:04 +0200245EXPORT_SYMBOL(elevator_exit);
246
Jens Axboe98170642006-07-28 09:23:08 +0200247static inline void __elv_rqhash_del(struct request *rq)
248{
Sasha Levin242d98f2012-12-17 10:01:27 -0500249 hash_del(&rq->hash);
Jens Axboe360f92c2014-04-09 20:27:01 -0600250 rq->cmd_flags &= ~REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200251}
252
Jens Axboe165125e2007-07-24 09:28:11 +0200253static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200254{
255 if (ELV_ON_HASH(rq))
256 __elv_rqhash_del(rq);
257}
258
Jens Axboe165125e2007-07-24 09:28:11 +0200259static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200260{
Jens Axboeb374d182008-10-31 10:05:07 +0100261 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200262
263 BUG_ON(ELV_ON_HASH(rq));
Sasha Levin242d98f2012-12-17 10:01:27 -0500264 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
Jens Axboe360f92c2014-04-09 20:27:01 -0600265 rq->cmd_flags |= REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200266}
267
Jens Axboe165125e2007-07-24 09:28:11 +0200268static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200269{
270 __elv_rqhash_del(rq);
271 elv_rqhash_add(q, rq);
272}
273
Jens Axboe165125e2007-07-24 09:28:11 +0200274static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200275{
Jens Axboeb374d182008-10-31 10:05:07 +0100276 struct elevator_queue *e = q->elevator;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800277 struct hlist_node *next;
Jens Axboe98170642006-07-28 09:23:08 +0200278 struct request *rq;
279
Linus Torvaldsee89f812013-02-28 12:52:24 -0800280 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
Jens Axboe98170642006-07-28 09:23:08 +0200281 BUG_ON(!ELV_ON_HASH(rq));
282
283 if (unlikely(!rq_mergeable(rq))) {
284 __elv_rqhash_del(rq);
285 continue;
286 }
287
288 if (rq_hash_key(rq) == offset)
289 return rq;
290 }
291
292 return NULL;
293}
294
Tejun Heo8922e162005-10-20 16:23:44 +0200295/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200296 * RB-tree support functions for inserting/lookup/removal of requests
297 * in a sorted RB tree.
298 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200299void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200300{
301 struct rb_node **p = &root->rb_node;
302 struct rb_node *parent = NULL;
303 struct request *__rq;
304
305 while (*p) {
306 parent = *p;
307 __rq = rb_entry(parent, struct request, rb_node);
308
Tejun Heo83096eb2009-05-07 22:24:39 +0900309 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200310 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200311 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200312 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200313 }
314
315 rb_link_node(&rq->rb_node, parent, p);
316 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200317}
Jens Axboe2e662b62006-07-13 11:55:04 +0200318EXPORT_SYMBOL(elv_rb_add);
319
320void elv_rb_del(struct rb_root *root, struct request *rq)
321{
322 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
323 rb_erase(&rq->rb_node, root);
324 RB_CLEAR_NODE(&rq->rb_node);
325}
Jens Axboe2e662b62006-07-13 11:55:04 +0200326EXPORT_SYMBOL(elv_rb_del);
327
328struct request *elv_rb_find(struct rb_root *root, sector_t sector)
329{
330 struct rb_node *n = root->rb_node;
331 struct request *rq;
332
333 while (n) {
334 rq = rb_entry(n, struct request, rb_node);
335
Tejun Heo83096eb2009-05-07 22:24:39 +0900336 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200337 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900338 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200339 n = n->rb_right;
340 else
341 return rq;
342 }
343
344 return NULL;
345}
Jens Axboe2e662b62006-07-13 11:55:04 +0200346EXPORT_SYMBOL(elv_rb_find);
347
348/*
Tejun Heo8922e162005-10-20 16:23:44 +0200349 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200350 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200351 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200352 */
Jens Axboe165125e2007-07-24 09:28:11 +0200353void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200354{
355 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200356 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100357 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200358
Tejun Heo06b86242005-10-20 16:46:23 +0200359 if (q->last_merge == rq)
360 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200361
362 elv_rqhash_del(q, rq);
363
Tejun Heo15853af2005-11-10 08:52:05 +0100364 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200365
Jens Axboe1b47f532005-10-20 16:37:00 +0200366 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100367 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200368 list_for_each_prev(entry, &q->queue_head) {
369 struct request *pos = list_entry_rq(entry);
370
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200371 if ((rq->cmd_flags & REQ_DISCARD) !=
372 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100373 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100374 if (rq_data_dir(rq) != rq_data_dir(pos))
375 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100376 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200377 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900378 if (blk_rq_pos(rq) >= boundary) {
379 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200380 continue;
381 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900382 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200383 break;
384 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900385 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200386 break;
387 }
388
389 list_add(&rq->queuelist, entry);
390}
Jens Axboe2e662b62006-07-13 11:55:04 +0200391EXPORT_SYMBOL(elv_dispatch_sort);
392
Jens Axboe98170642006-07-28 09:23:08 +0200393/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200394 * Insert rq into dispatch queue of q. Queue lock must be held on
395 * entry. rq is added to the back of the dispatch queue. To be used by
396 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200397 */
398void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
399{
400 if (q->last_merge == rq)
401 q->last_merge = NULL;
402
403 elv_rqhash_del(q, rq);
404
405 q->nr_sorted--;
406
407 q->end_sector = rq_end_sector(rq);
408 q->boundary_rq = rq;
409 list_add_tail(&rq->queuelist, &q->queue_head);
410}
Jens Axboe2e662b62006-07-13 11:55:04 +0200411EXPORT_SYMBOL(elv_dispatch_add_tail);
412
Jens Axboe165125e2007-07-24 09:28:11 +0200413int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414{
Jens Axboeb374d182008-10-31 10:05:07 +0100415 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200416 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200417 int ret;
418
Jens Axboe98170642006-07-28 09:23:08 +0200419 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100420 * Levels of merges:
421 * nomerges: No merges at all attempted
422 * noxmerges: Only simple one-hit cache try
423 * merges: All merge tries attempted
424 */
425 if (blk_queue_nomerges(q))
426 return ELEVATOR_NO_MERGE;
427
428 /*
Jens Axboe98170642006-07-28 09:23:08 +0200429 * First try one-hit cache.
430 */
Tejun Heo050c8ea2012-02-08 09:19:38 +0100431 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
432 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200433 if (ret != ELEVATOR_NO_MERGE) {
434 *req = q->last_merge;
435 return ret;
436 }
437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100439 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200440 return ELEVATOR_NO_MERGE;
441
Jens Axboe98170642006-07-28 09:23:08 +0200442 /*
443 * See if our hash lookup can find a potential backmerge.
444 */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700445 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
Jens Axboe98170642006-07-28 09:23:08 +0200446 if (__rq && elv_rq_merge_ok(__rq, bio)) {
447 *req = __rq;
448 return ELEVATOR_BACK_MERGE;
449 }
450
Tejun Heo22f746e2011-12-14 00:33:41 +0100451 if (e->type->ops.elevator_merge_fn)
452 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 return ELEVATOR_NO_MERGE;
455}
456
Jens Axboe5e84ea32011-03-21 10:14:27 +0100457/*
458 * Attempt to do an insertion back merge. Only check for the case where
459 * we can append 'rq' to an existing request, so we can throw 'rq' away
460 * afterwards.
461 *
462 * Returns true if we merged, false otherwise
463 */
464static bool elv_attempt_insert_merge(struct request_queue *q,
465 struct request *rq)
466{
467 struct request *__rq;
Shaohua Libee03932012-11-09 08:44:27 +0100468 bool ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100469
470 if (blk_queue_nomerges(q))
471 return false;
472
473 /*
474 * First try one-hit cache.
475 */
476 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
477 return true;
478
479 if (blk_queue_noxmerges(q))
480 return false;
481
Shaohua Libee03932012-11-09 08:44:27 +0100482 ret = false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100483 /*
484 * See if our hash lookup can find a potential backmerge.
485 */
Shaohua Libee03932012-11-09 08:44:27 +0100486 while (1) {
487 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
488 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
489 break;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100490
Shaohua Libee03932012-11-09 08:44:27 +0100491 /* The merged request could be merged with others, try again */
492 ret = true;
493 rq = __rq;
494 }
495
496 return ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100497}
498
Jens Axboe165125e2007-07-24 09:28:11 +0200499void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Jens Axboeb374d182008-10-31 10:05:07 +0100501 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Tejun Heo22f746e2011-12-14 00:33:41 +0100503 if (e->type->ops.elevator_merged_fn)
504 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200505
Jens Axboe2e662b62006-07-13 11:55:04 +0200506 if (type == ELEVATOR_BACK_MERGE)
507 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200508
Tejun Heo06b86242005-10-20 16:46:23 +0200509 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
Jens Axboe165125e2007-07-24 09:28:11 +0200512void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 struct request *next)
514{
Jens Axboeb374d182008-10-31 10:05:07 +0100515 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100516 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Tejun Heo22f746e2011-12-14 00:33:41 +0100518 if (next_sorted && e->type->ops.elevator_merge_req_fn)
519 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200520
Jens Axboe98170642006-07-28 09:23:08 +0200521 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200522
Jens Axboe5e84ea32011-03-21 10:14:27 +0100523 if (next_sorted) {
524 elv_rqhash_del(q, next);
525 q->nr_sorted--;
526 }
527
Tejun Heo06b86242005-10-20 16:46:23 +0200528 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529}
530
Divyesh Shah812d4022010-04-08 21:14:23 -0700531void elv_bio_merged(struct request_queue *q, struct request *rq,
532 struct bio *bio)
533{
534 struct elevator_queue *e = q->elevator;
535
Tejun Heo22f746e2011-12-14 00:33:41 +0100536 if (e->type->ops.elevator_bio_merged_fn)
537 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700538}
539
Lin Mingc8158812013-03-23 11:42:27 +0800540#ifdef CONFIG_PM_RUNTIME
541static void blk_pm_requeue_request(struct request *rq)
542{
543 if (rq->q->dev && !(rq->cmd_flags & REQ_PM))
544 rq->q->nr_pending--;
545}
546
547static void blk_pm_add_request(struct request_queue *q, struct request *rq)
548{
549 if (q->dev && !(rq->cmd_flags & REQ_PM) && q->nr_pending++ == 0 &&
550 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
551 pm_request_resume(q->dev);
552}
553#else
554static inline void blk_pm_requeue_request(struct request *rq) {}
555static inline void blk_pm_add_request(struct request_queue *q,
556 struct request *rq)
557{
558}
559#endif
560
Jens Axboe165125e2007-07-24 09:28:11 +0200561void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 /*
564 * it already went through dequeue, we need to decrement the
565 * in_flight count again
566 */
Tejun Heo8922e162005-10-20 16:23:44 +0200567 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200568 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200569 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100570 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Jens Axboe4aff5e22006-08-10 08:44:47 +0200573 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Lin Mingc8158812013-03-23 11:42:27 +0800575 blk_pm_requeue_request(rq);
576
Jens Axboeb710a482011-03-30 09:52:30 +0200577 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Jerome Marchand26308ea2009-03-27 10:31:51 +0100580void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100581{
582 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200583
584 lockdep_assert_held(q->queue_lock);
585
Tejun Heo22f746e2011-12-14 00:33:41 +0100586 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100587 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200588 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100589 printk(KERN_ERR "%s: forced dispatching is broken "
590 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100591 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100592 }
593}
594
Jens Axboeb710a482011-03-30 09:52:30 +0200595void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100597 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100598
Lin Mingc8158812013-03-23 11:42:27 +0800599 blk_pm_add_request(q, rq);
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 rq->q = q;
602
Jens Axboeb710a482011-03-30 09:52:30 +0200603 if (rq->cmd_flags & REQ_SOFTBARRIER) {
604 /* barriers are scheduling boundary, update end_sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400605 if (rq->cmd_type == REQ_TYPE_FS) {
Jens Axboeb710a482011-03-30 09:52:30 +0200606 q->end_sector = rq_end_sector(rq);
607 q->boundary_rq = rq;
608 }
609 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200610 (where == ELEVATOR_INSERT_SORT ||
611 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200612 where = ELEVATOR_INSERT_BACK;
613
Tejun Heo8922e162005-10-20 16:23:44 +0200614 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200615 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200616 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200617 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200618 list_add(&rq->queuelist, &q->queue_head);
619 break;
620
621 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200622 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100623 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200624 list_add_tail(&rq->queuelist, &q->queue_head);
625 /*
626 * We kick the queue here for the following reasons.
627 * - The elevator might have returned NULL previously
628 * to delay requests and returned them now. As the
629 * queue wasn't empty before this request, ll_rw_blk
630 * won't run the queue on return, resulting in hang.
631 * - Usually, back inserted requests won't be merged
632 * with anything. There's no point in delaying queue
633 * processing.
634 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200635 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200636 break;
637
Jens Axboe5e84ea32011-03-21 10:14:27 +0100638 case ELEVATOR_INSERT_SORT_MERGE:
639 /*
640 * If we succeed in merging this request with one in the
641 * queue already, we are done - rq has now been freed,
642 * so no need to do anything further.
643 */
644 if (elv_attempt_insert_merge(q, rq))
645 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200646 case ELEVATOR_INSERT_SORT:
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400647 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200648 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100649 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200650 if (rq_mergeable(rq)) {
651 elv_rqhash_add(q, rq);
652 if (!q->last_merge)
653 q->last_merge = rq;
654 }
655
Tejun Heoca235092005-11-01 17:23:49 +0900656 /*
657 * Some ioscheds (cfq) run q->request_fn directly, so
658 * rq cannot be accessed after calling
659 * elevator_add_req_fn.
660 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100661 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200662 break;
663
Tejun Heoae1b1532011-01-25 12:43:54 +0100664 case ELEVATOR_INSERT_FLUSH:
665 rq->cmd_flags |= REQ_SOFTBARRIER;
666 blk_insert_flush(rq);
667 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200668 default:
669 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700670 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200671 BUG();
672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
Jens Axboe2e662b62006-07-13 11:55:04 +0200674EXPORT_SYMBOL(__elv_add_request);
675
Jens Axboe7eaceac2011-03-10 08:52:07 +0100676void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677{
678 unsigned long flags;
679
680 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100681 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 spin_unlock_irqrestore(q->queue_lock, flags);
683}
Jens Axboe2e662b62006-07-13 11:55:04 +0200684EXPORT_SYMBOL(elv_add_request);
685
Jens Axboe165125e2007-07-24 09:28:11 +0200686struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
Jens Axboeb374d182008-10-31 10:05:07 +0100688 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Tejun Heo22f746e2011-12-14 00:33:41 +0100690 if (e->type->ops.elevator_latter_req_fn)
691 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 return NULL;
693}
694
Jens Axboe165125e2007-07-24 09:28:11 +0200695struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Jens Axboeb374d182008-10-31 10:05:07 +0100697 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Tejun Heo22f746e2011-12-14 00:33:41 +0100699 if (e->type->ops.elevator_former_req_fn)
700 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return NULL;
702}
703
Tejun Heo852c7882012-03-05 13:15:27 -0800704int elv_set_request(struct request_queue *q, struct request *rq,
705 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
Jens Axboeb374d182008-10-31 10:05:07 +0100707 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Tejun Heo22f746e2011-12-14 00:33:41 +0100709 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800710 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return 0;
712}
713
Jens Axboe165125e2007-07-24 09:28:11 +0200714void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Jens Axboeb374d182008-10-31 10:05:07 +0100716 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Tejun Heo22f746e2011-12-14 00:33:41 +0100718 if (e->type->ops.elevator_put_req_fn)
719 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
721
Jens Axboe165125e2007-07-24 09:28:11 +0200722int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723{
Jens Axboeb374d182008-10-31 10:05:07 +0100724 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Tejun Heo22f746e2011-12-14 00:33:41 +0100726 if (e->type->ops.elevator_may_queue_fn)
727 return e->type->ops.elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
729 return ELV_MQUEUE_MAY;
730}
731
Mike Anderson11914a52008-09-13 20:31:27 +0200732void elv_abort_queue(struct request_queue *q)
733{
734 struct request *rq;
735
Tejun Heoae1b1532011-01-25 12:43:54 +0100736 blk_abort_flushes(q);
737
Mike Anderson11914a52008-09-13 20:31:27 +0200738 while (!list_empty(&q->queue_head)) {
739 rq = list_entry_rq(q->queue_head.next);
740 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100741 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200742 /*
743 * Mark this request as started so we don't trigger
744 * any debug logic in the end I/O path.
745 */
746 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900747 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200748 }
749}
750EXPORT_SYMBOL(elv_abort_queue);
751
Jens Axboe165125e2007-07-24 09:28:11 +0200752void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
Jens Axboeb374d182008-10-31 10:05:07 +0100754 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756 /*
757 * request is released from the driver, io must be done
758 */
Tejun Heo8922e162005-10-20 16:23:44 +0200759 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200760 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200761 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100762 e->type->ops.elevator_completed_req_fn)
763 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
766
Al Viro3d1ab402006-03-18 18:35:43 -0500767#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
768
769static ssize_t
770elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
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->show)
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->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500782 mutex_unlock(&e->sysfs_lock);
783 return error;
784}
785
786static ssize_t
787elv_attr_store(struct kobject *kobj, struct attribute *attr,
788 const char *page, size_t length)
789{
Al Viro3d1ab402006-03-18 18:35:43 -0500790 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100791 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500792 ssize_t error;
793
794 if (!entry->store)
795 return -EIO;
796
Jens Axboeb374d182008-10-31 10:05:07 +0100797 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500798 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100799 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500800 mutex_unlock(&e->sysfs_lock);
801 return error;
802}
803
Emese Revfy52cf25d2010-01-19 02:58:23 +0100804static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500805 .show = elv_attr_show,
806 .store = elv_attr_store,
807};
808
809static struct kobj_type elv_ktype = {
810 .sysfs_ops = &elv_sysfs_ops,
811 .release = elevator_release,
812};
813
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800814int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800816 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500817 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700819 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500820 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100821 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500822 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500823 while (attr->attr.name) {
824 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500825 break;
Al Viroe572ec72006-03-18 22:27:18 -0500826 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500827 }
828 }
829 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200830 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500831 }
832 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
Tejun Heof8fc8772011-12-14 00:33:40 +0100834EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836void elv_unregister_queue(struct request_queue *q)
837{
Tejun Heof8fc8772011-12-14 00:33:40 +0100838 if (q) {
839 struct elevator_queue *e = q->elevator;
840
841 kobject_uevent(&e->kobj, KOBJ_REMOVE);
842 kobject_del(&e->kobj);
843 e->registered = 0;
844 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845}
Mike Snitzer01effb02010-05-11 08:57:42 +0200846EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Fabian Frederickb5097e92014-06-10 20:04:50 +0200848int __init elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100850 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200851
Tejun Heo3d3c2372011-12-14 00:33:42 +0100852 /* create icq_cache if requested */
853 if (e->icq_size) {
854 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
855 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
856 return -EINVAL;
857
858 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
859 "%s_io_cq", e->elevator_name);
860 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
861 e->icq_align, 0, NULL);
862 if (!e->icq_cache)
863 return -ENOMEM;
864 }
865
866 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200867 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100868 if (elevator_find(e->elevator_name)) {
869 spin_unlock(&elv_list_lock);
870 if (e->icq_cache)
871 kmem_cache_destroy(e->icq_cache);
872 return -EBUSY;
873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200875 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Tejun Heo3d3c2372011-12-14 00:33:42 +0100877 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100878 if (!strcmp(e->elevator_name, chosen_elevator) ||
879 (!*chosen_elevator &&
880 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100881 def = " (default)";
882
Jens Axboe4eb166d2008-02-01 00:37:27 +0100883 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
884 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100885 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887EXPORT_SYMBOL_GPL(elv_register);
888
889void elv_unregister(struct elevator_type *e)
890{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100891 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200892 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200894 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100895
896 /*
897 * Destroy icq_cache if it exists. icq's are RCU managed. Make
898 * sure all RCU operations are complete before proceeding.
899 */
900 if (e->icq_cache) {
901 rcu_barrier();
902 kmem_cache_destroy(e->icq_cache);
903 e->icq_cache = NULL;
904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905}
906EXPORT_SYMBOL_GPL(elv_unregister);
907
908/*
909 * switch to new_e io scheduler. be careful not to introduce deadlocks -
910 * we don't free the old io scheduler, before we have allocated what we
911 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200912 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 */
Jens Axboe165125e2007-07-24 09:28:11 +0200914static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800916 struct elevator_queue *old = q->elevator;
917 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800918 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800920 /*
921 * Turn on BYPASS and drain all requests w/ elevator private data.
922 * Block layer doesn't call into a quiesced elevator - all requests
923 * are directly put on the dispatch list without elevator data
924 * using INSERT_BACK. All requests have SOFTBARRIER set and no
925 * merge happens either.
926 */
Tejun Heod7325802012-03-05 13:14:58 -0800927 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200928
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800929 /* unregister and clear all auxiliary data of the old elevator */
930 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100931 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800932
933 spin_lock_irq(q->queue_lock);
934 ioc_clear_queue(q);
935 spin_unlock_irq(q->queue_lock);
936
937 /* allocate, init and register new elevator */
Jianpeng Mad50235b2013-07-03 13:25:24 +0200938 err = new_e->ops.elevator_init_fn(q, new_e);
939 if (err)
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800940 goto fail_init;
941
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800942 if (registered) {
943 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200944 if (err)
945 goto fail_register;
946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800948 /* done, kill the old one and finish */
949 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800950 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200951
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800952 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200953
Jens Axboe5dd531a2010-08-23 13:52:19 +0200954 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800957 elevator_exit(q->elevator);
958fail_init:
959 /* switch failed, restore and re-register old elevator */
960 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800962 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200963
Jens Axboe5dd531a2010-08-23 13:52:19 +0200964 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
Jens Axboe5dd531a2010-08-23 13:52:19 +0200967/*
968 * Switch this queue to the given IO scheduler.
969 */
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600970static int __elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
972 char elevator_name[ELV_NAME_MAX];
973 struct elevator_type *e;
974
Martin K. Petersencd43e262009-05-22 17:17:52 -0400975 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200976 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400977
Li Zefanee2e9922008-10-14 08:49:56 +0200978 strlcpy(elevator_name, name, sizeof(elevator_name));
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800979 e = elevator_get(strstrip(elevator_name), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 if (!e) {
981 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
982 return -EINVAL;
983 }
984
Tejun Heo22f746e2011-12-14 00:33:41 +0100985 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800986 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200987 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Jens Axboe5dd531a2010-08-23 13:52:19 +0200990 return elevator_switch(q, e);
991}
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600992
993int elevator_change(struct request_queue *q, const char *name)
994{
995 int ret;
996
997 /* Protect q->elevator from elevator_init() */
998 mutex_lock(&q->sysfs_lock);
999 ret = __elevator_change(q, name);
1000 mutex_unlock(&q->sysfs_lock);
1001
1002 return ret;
1003}
Jens Axboe5dd531a2010-08-23 13:52:19 +02001004EXPORT_SYMBOL(elevator_change);
1005
1006ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1007 size_t count)
1008{
1009 int ret;
1010
1011 if (!q->elevator)
1012 return count;
1013
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -06001014 ret = __elevator_change(q, name);
Jens Axboe5dd531a2010-08-23 13:52:19 +02001015 if (!ret)
1016 return count;
1017
1018 printk(KERN_ERR "elevator: switch to %s failed\n", name);
1019 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020}
1021
Jens Axboe165125e2007-07-24 09:28:11 +02001022ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Jens Axboeb374d182008-10-31 10:05:07 +01001024 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001025 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001026 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 int len = 0;
1028
Mike Snitzere36f7242010-05-24 09:07:32 +02001029 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001030 return sprintf(name, "none\n");
1031
Tejun Heo22f746e2011-12-14 00:33:41 +01001032 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001033
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001034 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001035 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (!strcmp(elv->elevator_name, __e->elevator_name))
1037 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1038 else
1039 len += sprintf(name+len, "%s ", __e->elevator_name);
1040 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001041 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 len += sprintf(len+name, "\n");
1044 return len;
1045}
1046
Jens Axboe165125e2007-07-24 09:28:11 +02001047struct request *elv_rb_former_request(struct request_queue *q,
1048 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001049{
1050 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1051
1052 if (rbprev)
1053 return rb_entry_rq(rbprev);
1054
1055 return NULL;
1056}
Jens Axboe2e662b62006-07-13 11:55:04 +02001057EXPORT_SYMBOL(elv_rb_former_request);
1058
Jens Axboe165125e2007-07-24 09:28:11 +02001059struct request *elv_rb_latter_request(struct request_queue *q,
1060 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001061{
1062 struct rb_node *rbnext = rb_next(&rq->rb_node);
1063
1064 if (rbnext)
1065 return rb_entry_rq(rbnext);
1066
1067 return NULL;
1068}
Jens Axboe2e662b62006-07-13 11:55:04 +02001069EXPORT_SYMBOL(elv_rb_latter_request);