blob: 84d63943f2de2f386ff35e6a395f68ada173b5b2 [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 */
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))
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);
Jens Axboe360f92c2014-04-09 20:27:01 -0600248 rq->cmd_flags &= ~REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200249}
250
Jens Axboe165125e2007-07-24 09:28:11 +0200251static void 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 Axboe165125e2007-07-24 09:28:11 +0200257static void 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));
Jens Axboe360f92c2014-04-09 20:27:01 -0600263 rq->cmd_flags |= REQ_HASHED;
Jens Axboe98170642006-07-28 09:23:08 +0200264}
265
Jens Axboe165125e2007-07-24 09:28:11 +0200266static void 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 Axboe165125e2007-07-24 09:28:11 +0200272static struct 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;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100355 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200356
Tejun Heo06b86242005-10-20 16:46:23 +0200357 if (q->last_merge == rq)
358 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200359
360 elv_rqhash_del(q, rq);
361
Tejun Heo15853af2005-11-10 08:52:05 +0100362 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200363
Jens Axboe1b47f532005-10-20 16:37:00 +0200364 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100365 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200366 list_for_each_prev(entry, &q->queue_head) {
367 struct request *pos = list_entry_rq(entry);
368
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200369 if ((rq->cmd_flags & REQ_DISCARD) !=
370 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100371 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100372 if (rq_data_dir(rq) != rq_data_dir(pos))
373 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100374 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200375 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900376 if (blk_rq_pos(rq) >= boundary) {
377 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200378 continue;
379 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900380 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200381 break;
382 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900383 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200384 break;
385 }
386
387 list_add(&rq->queuelist, entry);
388}
Jens Axboe2e662b62006-07-13 11:55:04 +0200389EXPORT_SYMBOL(elv_dispatch_sort);
390
Jens Axboe98170642006-07-28 09:23:08 +0200391/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200392 * Insert rq into dispatch queue of q. Queue lock must be held on
393 * entry. rq is added to the back of the dispatch queue. To be used by
394 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200395 */
396void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
397{
398 if (q->last_merge == rq)
399 q->last_merge = NULL;
400
401 elv_rqhash_del(q, rq);
402
403 q->nr_sorted--;
404
405 q->end_sector = rq_end_sector(rq);
406 q->boundary_rq = rq;
407 list_add_tail(&rq->queuelist, &q->queue_head);
408}
Jens Axboe2e662b62006-07-13 11:55:04 +0200409EXPORT_SYMBOL(elv_dispatch_add_tail);
410
Jens Axboe165125e2007-07-24 09:28:11 +0200411int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Jens Axboeb374d182008-10-31 10:05:07 +0100413 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200414 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200415 int ret;
416
Jens Axboe98170642006-07-28 09:23:08 +0200417 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100418 * Levels of merges:
419 * nomerges: No merges at all attempted
420 * noxmerges: Only simple one-hit cache try
421 * merges: All merge tries attempted
422 */
423 if (blk_queue_nomerges(q))
424 return ELEVATOR_NO_MERGE;
425
426 /*
Jens Axboe98170642006-07-28 09:23:08 +0200427 * First try one-hit cache.
428 */
Tejun Heo050c8ea2012-02-08 09:19:38 +0100429 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
430 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200431 if (ret != ELEVATOR_NO_MERGE) {
432 *req = q->last_merge;
433 return ret;
434 }
435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100437 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200438 return ELEVATOR_NO_MERGE;
439
Jens Axboe98170642006-07-28 09:23:08 +0200440 /*
441 * See if our hash lookup can find a potential backmerge.
442 */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700443 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
Jens Axboe98170642006-07-28 09:23:08 +0200444 if (__rq && elv_rq_merge_ok(__rq, bio)) {
445 *req = __rq;
446 return ELEVATOR_BACK_MERGE;
447 }
448
Tejun Heo22f746e2011-12-14 00:33:41 +0100449 if (e->type->ops.elevator_merge_fn)
450 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 return ELEVATOR_NO_MERGE;
453}
454
Jens Axboe5e84ea32011-03-21 10:14:27 +0100455/*
456 * Attempt to do an insertion back merge. Only check for the case where
457 * we can append 'rq' to an existing request, so we can throw 'rq' away
458 * afterwards.
459 *
460 * Returns true if we merged, false otherwise
461 */
462static bool elv_attempt_insert_merge(struct request_queue *q,
463 struct request *rq)
464{
465 struct request *__rq;
Shaohua Libee03932012-11-09 08:44:27 +0100466 bool ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100467
468 if (blk_queue_nomerges(q))
469 return false;
470
471 /*
472 * First try one-hit cache.
473 */
474 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
475 return true;
476
477 if (blk_queue_noxmerges(q))
478 return false;
479
Shaohua Libee03932012-11-09 08:44:27 +0100480 ret = false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100481 /*
482 * See if our hash lookup can find a potential backmerge.
483 */
Shaohua Libee03932012-11-09 08:44:27 +0100484 while (1) {
485 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
486 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
487 break;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100488
Shaohua Libee03932012-11-09 08:44:27 +0100489 /* The merged request could be merged with others, try again */
490 ret = true;
491 rq = __rq;
492 }
493
494 return ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100495}
496
Jens Axboe165125e2007-07-24 09:28:11 +0200497void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Jens Axboeb374d182008-10-31 10:05:07 +0100499 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Tejun Heo22f746e2011-12-14 00:33:41 +0100501 if (e->type->ops.elevator_merged_fn)
502 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200503
Jens Axboe2e662b62006-07-13 11:55:04 +0200504 if (type == ELEVATOR_BACK_MERGE)
505 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200506
Tejun Heo06b86242005-10-20 16:46:23 +0200507 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
509
Jens Axboe165125e2007-07-24 09:28:11 +0200510void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 struct request *next)
512{
Jens Axboeb374d182008-10-31 10:05:07 +0100513 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100514 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Tejun Heo22f746e2011-12-14 00:33:41 +0100516 if (next_sorted && e->type->ops.elevator_merge_req_fn)
517 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200518
Jens Axboe98170642006-07-28 09:23:08 +0200519 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200520
Jens Axboe5e84ea32011-03-21 10:14:27 +0100521 if (next_sorted) {
522 elv_rqhash_del(q, next);
523 q->nr_sorted--;
524 }
525
Tejun Heo06b86242005-10-20 16:46:23 +0200526 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527}
528
Divyesh Shah812d4022010-04-08 21:14:23 -0700529void elv_bio_merged(struct request_queue *q, struct request *rq,
530 struct bio *bio)
531{
532 struct elevator_queue *e = q->elevator;
533
Tejun Heo22f746e2011-12-14 00:33:41 +0100534 if (e->type->ops.elevator_bio_merged_fn)
535 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700536}
537
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +0100538#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +0800539static void blk_pm_requeue_request(struct request *rq)
540{
541 if (rq->q->dev && !(rq->cmd_flags & REQ_PM))
542 rq->q->nr_pending--;
543}
544
545static void blk_pm_add_request(struct request_queue *q, struct request *rq)
546{
547 if (q->dev && !(rq->cmd_flags & REQ_PM) && q->nr_pending++ == 0 &&
548 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
549 pm_request_resume(q->dev);
550}
551#else
552static inline void blk_pm_requeue_request(struct request *rq) {}
553static inline void blk_pm_add_request(struct request_queue *q,
554 struct request *rq)
555{
556}
557#endif
558
Jens Axboe165125e2007-07-24 09:28:11 +0200559void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 /*
562 * it already went through dequeue, we need to decrement the
563 * in_flight count again
564 */
Tejun Heo8922e162005-10-20 16:23:44 +0200565 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200566 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200567 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100568 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Jens Axboe4aff5e22006-08-10 08:44:47 +0200571 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Lin Mingc8158812013-03-23 11:42:27 +0800573 blk_pm_requeue_request(rq);
574
Jens Axboeb710a482011-03-30 09:52:30 +0200575 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577
Jerome Marchand26308ea2009-03-27 10:31:51 +0100578void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100579{
580 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200581
582 lockdep_assert_held(q->queue_lock);
583
Tejun Heo22f746e2011-12-14 00:33:41 +0100584 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100585 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200586 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100587 printk(KERN_ERR "%s: forced dispatching is broken "
588 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100589 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100590 }
591}
592
Jens Axboeb710a482011-03-30 09:52:30 +0200593void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100595 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100596
Lin Mingc8158812013-03-23 11:42:27 +0800597 blk_pm_add_request(q, rq);
598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 rq->q = q;
600
Jens Axboeb710a482011-03-30 09:52:30 +0200601 if (rq->cmd_flags & REQ_SOFTBARRIER) {
602 /* barriers are scheduling boundary, update end_sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400603 if (rq->cmd_type == REQ_TYPE_FS) {
Jens Axboeb710a482011-03-30 09:52:30 +0200604 q->end_sector = rq_end_sector(rq);
605 q->boundary_rq = rq;
606 }
607 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200608 (where == ELEVATOR_INSERT_SORT ||
609 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200610 where = ELEVATOR_INSERT_BACK;
611
Tejun Heo8922e162005-10-20 16:23:44 +0200612 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200613 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200614 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200615 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200616 list_add(&rq->queuelist, &q->queue_head);
617 break;
618
619 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200620 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100621 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200622 list_add_tail(&rq->queuelist, &q->queue_head);
623 /*
624 * We kick the queue here for the following reasons.
625 * - The elevator might have returned NULL previously
626 * to delay requests and returned them now. As the
627 * queue wasn't empty before this request, ll_rw_blk
628 * won't run the queue on return, resulting in hang.
629 * - Usually, back inserted requests won't be merged
630 * with anything. There's no point in delaying queue
631 * processing.
632 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200633 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200634 break;
635
Jens Axboe5e84ea32011-03-21 10:14:27 +0100636 case ELEVATOR_INSERT_SORT_MERGE:
637 /*
638 * If we succeed in merging this request with one in the
639 * queue already, we are done - rq has now been freed,
640 * so no need to do anything further.
641 */
642 if (elv_attempt_insert_merge(q, rq))
643 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200644 case ELEVATOR_INSERT_SORT:
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400645 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200646 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100647 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200648 if (rq_mergeable(rq)) {
649 elv_rqhash_add(q, rq);
650 if (!q->last_merge)
651 q->last_merge = rq;
652 }
653
Tejun Heoca235092005-11-01 17:23:49 +0900654 /*
655 * Some ioscheds (cfq) run q->request_fn directly, so
656 * rq cannot be accessed after calling
657 * elevator_add_req_fn.
658 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100659 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200660 break;
661
Tejun Heoae1b1532011-01-25 12:43:54 +0100662 case ELEVATOR_INSERT_FLUSH:
663 rq->cmd_flags |= REQ_SOFTBARRIER;
664 blk_insert_flush(rq);
665 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200666 default:
667 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700668 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200669 BUG();
670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671}
Jens Axboe2e662b62006-07-13 11:55:04 +0200672EXPORT_SYMBOL(__elv_add_request);
673
Jens Axboe7eaceac2011-03-10 08:52:07 +0100674void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
676 unsigned long flags;
677
678 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100679 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 spin_unlock_irqrestore(q->queue_lock, flags);
681}
Jens Axboe2e662b62006-07-13 11:55:04 +0200682EXPORT_SYMBOL(elv_add_request);
683
Jens Axboe165125e2007-07-24 09:28:11 +0200684struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Jens Axboeb374d182008-10-31 10:05:07 +0100686 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Tejun Heo22f746e2011-12-14 00:33:41 +0100688 if (e->type->ops.elevator_latter_req_fn)
689 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 return NULL;
691}
692
Jens Axboe165125e2007-07-24 09:28:11 +0200693struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
Jens Axboeb374d182008-10-31 10:05:07 +0100695 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Tejun Heo22f746e2011-12-14 00:33:41 +0100697 if (e->type->ops.elevator_former_req_fn)
698 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 return NULL;
700}
701
Tejun Heo852c7882012-03-05 13:15:27 -0800702int elv_set_request(struct request_queue *q, struct request *rq,
703 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
Jens Axboeb374d182008-10-31 10:05:07 +0100705 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Tejun Heo22f746e2011-12-14 00:33:41 +0100707 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800708 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 return 0;
710}
711
Jens Axboe165125e2007-07-24 09:28:11 +0200712void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
Jens Axboeb374d182008-10-31 10:05:07 +0100714 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Tejun Heo22f746e2011-12-14 00:33:41 +0100716 if (e->type->ops.elevator_put_req_fn)
717 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Jens Axboe165125e2007-07-24 09:28:11 +0200720int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Jens Axboeb374d182008-10-31 10:05:07 +0100722 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Tejun Heo22f746e2011-12-14 00:33:41 +0100724 if (e->type->ops.elevator_may_queue_fn)
725 return e->type->ops.elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727 return ELV_MQUEUE_MAY;
728}
729
Jens Axboe165125e2007-07-24 09:28:11 +0200730void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
Jens Axboeb374d182008-10-31 10:05:07 +0100732 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 /*
735 * request is released from the driver, io must be done
736 */
Tejun Heo8922e162005-10-20 16:23:44 +0200737 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200738 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200739 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100740 e->type->ops.elevator_completed_req_fn)
741 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
Al Viro3d1ab402006-03-18 18:35:43 -0500745#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
746
747static ssize_t
748elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
749{
Al Viro3d1ab402006-03-18 18:35:43 -0500750 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100751 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500752 ssize_t error;
753
754 if (!entry->show)
755 return -EIO;
756
Jens Axboeb374d182008-10-31 10:05:07 +0100757 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500758 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100759 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500760 mutex_unlock(&e->sysfs_lock);
761 return error;
762}
763
764static ssize_t
765elv_attr_store(struct kobject *kobj, struct attribute *attr,
766 const char *page, size_t length)
767{
Al Viro3d1ab402006-03-18 18:35:43 -0500768 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100769 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500770 ssize_t error;
771
772 if (!entry->store)
773 return -EIO;
774
Jens Axboeb374d182008-10-31 10:05:07 +0100775 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500776 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100777 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500778 mutex_unlock(&e->sysfs_lock);
779 return error;
780}
781
Emese Revfy52cf25d2010-01-19 02:58:23 +0100782static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500783 .show = elv_attr_show,
784 .store = elv_attr_store,
785};
786
787static struct kobj_type elv_ktype = {
788 .sysfs_ops = &elv_sysfs_ops,
789 .release = elevator_release,
790};
791
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800792int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800794 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500795 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700797 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500798 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100799 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500800 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500801 while (attr->attr.name) {
802 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500803 break;
Al Viroe572ec72006-03-18 22:27:18 -0500804 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500805 }
806 }
807 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200808 e->registered = 1;
Jens Axboe0bb97942015-06-10 08:01:20 -0600809 if (e->type->ops.elevator_registered_fn)
810 e->type->ops.elevator_registered_fn(q);
Al Viro3d1ab402006-03-18 18:35:43 -0500811 }
812 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
Tejun Heof8fc8772011-12-14 00:33:40 +0100814EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200815
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816void elv_unregister_queue(struct request_queue *q)
817{
Tejun Heof8fc8772011-12-14 00:33:40 +0100818 if (q) {
819 struct elevator_queue *e = q->elevator;
820
821 kobject_uevent(&e->kobj, KOBJ_REMOVE);
822 kobject_del(&e->kobj);
823 e->registered = 0;
824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
Mike Snitzer01effb02010-05-11 08:57:42 +0200826EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Jens Axboee567bf72014-06-22 16:32:48 -0600828int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100830 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200831
Tejun Heo3d3c2372011-12-14 00:33:42 +0100832 /* create icq_cache if requested */
833 if (e->icq_size) {
834 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
835 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
836 return -EINVAL;
837
838 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
839 "%s_io_cq", e->elevator_name);
840 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
841 e->icq_align, 0, NULL);
842 if (!e->icq_cache)
843 return -ENOMEM;
844 }
845
846 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200847 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100848 if (elevator_find(e->elevator_name)) {
849 spin_unlock(&elv_list_lock);
850 if (e->icq_cache)
851 kmem_cache_destroy(e->icq_cache);
852 return -EBUSY;
853 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200855 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
Tejun Heo3d3c2372011-12-14 00:33:42 +0100857 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100858 if (!strcmp(e->elevator_name, chosen_elevator) ||
859 (!*chosen_elevator &&
860 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100861 def = " (default)";
862
Jens Axboe4eb166d2008-02-01 00:37:27 +0100863 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
864 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100865 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866}
867EXPORT_SYMBOL_GPL(elv_register);
868
869void elv_unregister(struct elevator_type *e)
870{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100871 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200872 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200874 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100875
876 /*
877 * Destroy icq_cache if it exists. icq's are RCU managed. Make
878 * sure all RCU operations are complete before proceeding.
879 */
880 if (e->icq_cache) {
881 rcu_barrier();
882 kmem_cache_destroy(e->icq_cache);
883 e->icq_cache = NULL;
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886EXPORT_SYMBOL_GPL(elv_unregister);
887
888/*
889 * switch to new_e io scheduler. be careful not to introduce deadlocks -
890 * we don't free the old io scheduler, before we have allocated what we
891 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200892 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 */
Jens Axboe165125e2007-07-24 09:28:11 +0200894static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800896 struct elevator_queue *old = q->elevator;
897 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800898 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800900 /*
901 * Turn on BYPASS and drain all requests w/ elevator private data.
902 * Block layer doesn't call into a quiesced elevator - all requests
903 * are directly put on the dispatch list without elevator data
904 * using INSERT_BACK. All requests have SOFTBARRIER set and no
905 * merge happens either.
906 */
Tejun Heod7325802012-03-05 13:14:58 -0800907 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200908
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800909 /* unregister and clear all auxiliary data of the old elevator */
910 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100911 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800912
913 spin_lock_irq(q->queue_lock);
914 ioc_clear_queue(q);
915 spin_unlock_irq(q->queue_lock);
916
917 /* allocate, init and register new elevator */
Jianpeng Mad50235b2013-07-03 13:25:24 +0200918 err = new_e->ops.elevator_init_fn(q, new_e);
919 if (err)
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800920 goto fail_init;
921
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800922 if (registered) {
923 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200924 if (err)
925 goto fail_register;
926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800928 /* done, kill the old one and finish */
929 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800930 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200931
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800932 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200933
Jens Axboe5dd531a2010-08-23 13:52:19 +0200934 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800937 elevator_exit(q->elevator);
938fail_init:
939 /* switch failed, restore and re-register old elevator */
940 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800942 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200943
Jens Axboe5dd531a2010-08-23 13:52:19 +0200944 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
946
Jens Axboe5dd531a2010-08-23 13:52:19 +0200947/*
948 * Switch this queue to the given IO scheduler.
949 */
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600950static int __elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951{
952 char elevator_name[ELV_NAME_MAX];
953 struct elevator_type *e;
954
Martin K. Petersencd43e262009-05-22 17:17:52 -0400955 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200956 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400957
Li Zefanee2e9922008-10-14 08:49:56 +0200958 strlcpy(elevator_name, name, sizeof(elevator_name));
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800959 e = elevator_get(strstrip(elevator_name), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 if (!e) {
961 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
962 return -EINVAL;
963 }
964
Tejun Heo22f746e2011-12-14 00:33:41 +0100965 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800966 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200967 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Jens Axboe5dd531a2010-08-23 13:52:19 +0200970 return elevator_switch(q, e);
971}
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600972
973int elevator_change(struct request_queue *q, const char *name)
974{
975 int ret;
976
977 /* Protect q->elevator from elevator_init() */
978 mutex_lock(&q->sysfs_lock);
979 ret = __elevator_change(q, name);
980 mutex_unlock(&q->sysfs_lock);
981
982 return ret;
983}
Jens Axboe5dd531a2010-08-23 13:52:19 +0200984EXPORT_SYMBOL(elevator_change);
985
986ssize_t elv_iosched_store(struct request_queue *q, const char *name,
987 size_t count)
988{
989 int ret;
990
991 if (!q->elevator)
992 return count;
993
Tomoki Sekiyama7c8a3672013-10-15 16:42:19 -0600994 ret = __elevator_change(q, name);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200995 if (!ret)
996 return count;
997
998 printk(KERN_ERR "elevator: switch to %s failed\n", name);
999 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
Jens Axboe165125e2007-07-24 09:28:11 +02001002ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003{
Jens Axboeb374d182008-10-31 10:05:07 +01001004 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001005 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001006 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 int len = 0;
1008
Mike Snitzere36f7242010-05-24 09:07:32 +02001009 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001010 return sprintf(name, "none\n");
1011
Tejun Heo22f746e2011-12-14 00:33:41 +01001012 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001013
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001014 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001015 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 if (!strcmp(elv->elevator_name, __e->elevator_name))
1017 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1018 else
1019 len += sprintf(name+len, "%s ", __e->elevator_name);
1020 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001021 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023 len += sprintf(len+name, "\n");
1024 return len;
1025}
1026
Jens Axboe165125e2007-07-24 09:28:11 +02001027struct request *elv_rb_former_request(struct request_queue *q,
1028 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001029{
1030 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1031
1032 if (rbprev)
1033 return rb_entry_rq(rbprev);
1034
1035 return NULL;
1036}
Jens Axboe2e662b62006-07-13 11:55:04 +02001037EXPORT_SYMBOL(elv_rb_former_request);
1038
Jens Axboe165125e2007-07-24 09:28:11 +02001039struct request *elv_rb_latter_request(struct request_queue *q,
1040 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001041{
1042 struct rb_node *rbnext = rb_next(&rq->rb_node);
1043
1044 if (rbnext)
1045 return rb_entry_rq(rbnext);
1046
1047 return NULL;
1048}
Jens Axboe2e662b62006-07-13 11:55:04 +02001049EXPORT_SYMBOL(elv_rb_latter_request);