blob: 2bcbd8cc14d4b23780a3cfc262857a87f7c5d67e [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
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400189 if (unlikely(q->elevator))
190 return 0;
191
Tejun Heocb98fc82005-10-28 08:29:39 +0200192 INIT_LIST_HEAD(&q->queue_head);
193 q->last_merge = NULL;
194 q->end_sector = 0;
195 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200196
Jens Axboe4eb166d2008-02-01 00:37:27 +0100197 if (name) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800198 e = elevator_get(name, true);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100199 if (!e)
200 return -EINVAL;
201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800203 /*
204 * Use the default elevator specified by config boot param or
205 * config option. Don't try to load modules as we could be running
206 * off async and request_module() isn't allowed from async.
207 */
Jens Axboe4eb166d2008-02-01 00:37:27 +0100208 if (!e && *chosen_elevator) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800209 e = elevator_get(chosen_elevator, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100210 if (!e)
211 printk(KERN_ERR "I/O scheduler %s not found\n",
212 chosen_elevator);
213 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100214
Jens Axboe4eb166d2008-02-01 00:37:27 +0100215 if (!e) {
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800216 e = elevator_get(CONFIG_DEFAULT_IOSCHED, false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100217 if (!e) {
218 printk(KERN_ERR
219 "Default I/O scheduler not found. " \
220 "Using noop.\n");
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800221 e = elevator_get("noop", false);
Jens Axboe4eb166d2008-02-01 00:37:27 +0100222 }
Nate Diller5f003972006-01-24 10:07:58 +0100223 }
224
Jianpeng Mad50235b2013-07-03 13:25:24 +0200225 err = e->ops.elevator_init_fn(q, e);
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400226 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
Jens Axboe2e662b62006-07-13 11:55:04 +0200228EXPORT_SYMBOL(elevator_init);
229
Jens Axboeb374d182008-10-31 10:05:07 +0100230void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
Al Viro3d1ab402006-03-18 18:35:43 -0500232 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100233 if (e->type->ops.elevator_exit_fn)
234 e->type->ops.elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500235 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Al Viro3d1ab402006-03-18 18:35:43 -0500237 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
Jens Axboe2e662b62006-07-13 11:55:04 +0200239EXPORT_SYMBOL(elevator_exit);
240
Jens Axboe98170642006-07-28 09:23:08 +0200241static inline void __elv_rqhash_del(struct request *rq)
242{
Sasha Levin242d98f2012-12-17 10:01:27 -0500243 hash_del(&rq->hash);
Jens Axboe98170642006-07-28 09:23:08 +0200244}
245
Jens Axboe165125e2007-07-24 09:28:11 +0200246static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200247{
248 if (ELV_ON_HASH(rq))
249 __elv_rqhash_del(rq);
250}
251
Jens Axboe165125e2007-07-24 09:28:11 +0200252static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200253{
Jens Axboeb374d182008-10-31 10:05:07 +0100254 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200255
256 BUG_ON(ELV_ON_HASH(rq));
Sasha Levin242d98f2012-12-17 10:01:27 -0500257 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
Jens Axboe98170642006-07-28 09:23:08 +0200258}
259
Jens Axboe165125e2007-07-24 09:28:11 +0200260static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200261{
262 __elv_rqhash_del(rq);
263 elv_rqhash_add(q, rq);
264}
265
Jens Axboe165125e2007-07-24 09:28:11 +0200266static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200267{
Jens Axboeb374d182008-10-31 10:05:07 +0100268 struct elevator_queue *e = q->elevator;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800269 struct hlist_node *next;
Jens Axboe98170642006-07-28 09:23:08 +0200270 struct request *rq;
271
Linus Torvaldsee89f812013-02-28 12:52:24 -0800272 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
Jens Axboe98170642006-07-28 09:23:08 +0200273 BUG_ON(!ELV_ON_HASH(rq));
274
275 if (unlikely(!rq_mergeable(rq))) {
276 __elv_rqhash_del(rq);
277 continue;
278 }
279
280 if (rq_hash_key(rq) == offset)
281 return rq;
282 }
283
284 return NULL;
285}
286
Tejun Heo8922e162005-10-20 16:23:44 +0200287/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200288 * RB-tree support functions for inserting/lookup/removal of requests
289 * in a sorted RB tree.
290 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200291void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200292{
293 struct rb_node **p = &root->rb_node;
294 struct rb_node *parent = NULL;
295 struct request *__rq;
296
297 while (*p) {
298 parent = *p;
299 __rq = rb_entry(parent, struct request, rb_node);
300
Tejun Heo83096eb2009-05-07 22:24:39 +0900301 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200302 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200303 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200304 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200305 }
306
307 rb_link_node(&rq->rb_node, parent, p);
308 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200309}
Jens Axboe2e662b62006-07-13 11:55:04 +0200310EXPORT_SYMBOL(elv_rb_add);
311
312void elv_rb_del(struct rb_root *root, struct request *rq)
313{
314 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
315 rb_erase(&rq->rb_node, root);
316 RB_CLEAR_NODE(&rq->rb_node);
317}
Jens Axboe2e662b62006-07-13 11:55:04 +0200318EXPORT_SYMBOL(elv_rb_del);
319
320struct request *elv_rb_find(struct rb_root *root, sector_t sector)
321{
322 struct rb_node *n = root->rb_node;
323 struct request *rq;
324
325 while (n) {
326 rq = rb_entry(n, struct request, rb_node);
327
Tejun Heo83096eb2009-05-07 22:24:39 +0900328 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200329 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900330 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200331 n = n->rb_right;
332 else
333 return rq;
334 }
335
336 return NULL;
337}
Jens Axboe2e662b62006-07-13 11:55:04 +0200338EXPORT_SYMBOL(elv_rb_find);
339
340/*
Tejun Heo8922e162005-10-20 16:23:44 +0200341 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200342 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200343 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200344 */
Jens Axboe165125e2007-07-24 09:28:11 +0200345void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200346{
347 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200348 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100349 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200350
Tejun Heo06b86242005-10-20 16:46:23 +0200351 if (q->last_merge == rq)
352 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200353
354 elv_rqhash_del(q, rq);
355
Tejun Heo15853af2005-11-10 08:52:05 +0100356 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200357
Jens Axboe1b47f532005-10-20 16:37:00 +0200358 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100359 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200360 list_for_each_prev(entry, &q->queue_head) {
361 struct request *pos = list_entry_rq(entry);
362
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200363 if ((rq->cmd_flags & REQ_DISCARD) !=
364 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100365 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100366 if (rq_data_dir(rq) != rq_data_dir(pos))
367 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100368 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200369 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900370 if (blk_rq_pos(rq) >= boundary) {
371 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200372 continue;
373 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900374 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200375 break;
376 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900377 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200378 break;
379 }
380
381 list_add(&rq->queuelist, entry);
382}
Jens Axboe2e662b62006-07-13 11:55:04 +0200383EXPORT_SYMBOL(elv_dispatch_sort);
384
Jens Axboe98170642006-07-28 09:23:08 +0200385/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200386 * Insert rq into dispatch queue of q. Queue lock must be held on
387 * entry. rq is added to the back of the dispatch queue. To be used by
388 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200389 */
390void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
391{
392 if (q->last_merge == rq)
393 q->last_merge = NULL;
394
395 elv_rqhash_del(q, rq);
396
397 q->nr_sorted--;
398
399 q->end_sector = rq_end_sector(rq);
400 q->boundary_rq = rq;
401 list_add_tail(&rq->queuelist, &q->queue_head);
402}
Jens Axboe2e662b62006-07-13 11:55:04 +0200403EXPORT_SYMBOL(elv_dispatch_add_tail);
404
Jens Axboe165125e2007-07-24 09:28:11 +0200405int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Jens Axboeb374d182008-10-31 10:05:07 +0100407 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200408 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200409 int ret;
410
Jens Axboe98170642006-07-28 09:23:08 +0200411 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100412 * Levels of merges:
413 * nomerges: No merges at all attempted
414 * noxmerges: Only simple one-hit cache try
415 * merges: All merge tries attempted
416 */
417 if (blk_queue_nomerges(q))
418 return ELEVATOR_NO_MERGE;
419
420 /*
Jens Axboe98170642006-07-28 09:23:08 +0200421 * First try one-hit cache.
422 */
Tejun Heo050c8ea2012-02-08 09:19:38 +0100423 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
424 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200425 if (ret != ELEVATOR_NO_MERGE) {
426 *req = q->last_merge;
427 return ret;
428 }
429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100431 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200432 return ELEVATOR_NO_MERGE;
433
Jens Axboe98170642006-07-28 09:23:08 +0200434 /*
435 * See if our hash lookup can find a potential backmerge.
436 */
437 __rq = elv_rqhash_find(q, bio->bi_sector);
438 if (__rq && elv_rq_merge_ok(__rq, bio)) {
439 *req = __rq;
440 return ELEVATOR_BACK_MERGE;
441 }
442
Tejun Heo22f746e2011-12-14 00:33:41 +0100443 if (e->type->ops.elevator_merge_fn)
444 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 return ELEVATOR_NO_MERGE;
447}
448
Jens Axboe5e84ea32011-03-21 10:14:27 +0100449/*
450 * Attempt to do an insertion back merge. Only check for the case where
451 * we can append 'rq' to an existing request, so we can throw 'rq' away
452 * afterwards.
453 *
454 * Returns true if we merged, false otherwise
455 */
456static bool elv_attempt_insert_merge(struct request_queue *q,
457 struct request *rq)
458{
459 struct request *__rq;
Shaohua Libee03932012-11-09 08:44:27 +0100460 bool ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100461
462 if (blk_queue_nomerges(q))
463 return false;
464
465 /*
466 * First try one-hit cache.
467 */
468 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
469 return true;
470
471 if (blk_queue_noxmerges(q))
472 return false;
473
Shaohua Libee03932012-11-09 08:44:27 +0100474 ret = false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100475 /*
476 * See if our hash lookup can find a potential backmerge.
477 */
Shaohua Libee03932012-11-09 08:44:27 +0100478 while (1) {
479 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
480 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
481 break;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100482
Shaohua Libee03932012-11-09 08:44:27 +0100483 /* The merged request could be merged with others, try again */
484 ret = true;
485 rq = __rq;
486 }
487
488 return ret;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100489}
490
Jens Axboe165125e2007-07-24 09:28:11 +0200491void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Jens Axboeb374d182008-10-31 10:05:07 +0100493 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Tejun Heo22f746e2011-12-14 00:33:41 +0100495 if (e->type->ops.elevator_merged_fn)
496 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200497
Jens Axboe2e662b62006-07-13 11:55:04 +0200498 if (type == ELEVATOR_BACK_MERGE)
499 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200500
Tejun Heo06b86242005-10-20 16:46:23 +0200501 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
503
Jens Axboe165125e2007-07-24 09:28:11 +0200504void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 struct request *next)
506{
Jens Axboeb374d182008-10-31 10:05:07 +0100507 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100508 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Tejun Heo22f746e2011-12-14 00:33:41 +0100510 if (next_sorted && e->type->ops.elevator_merge_req_fn)
511 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200512
Jens Axboe98170642006-07-28 09:23:08 +0200513 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200514
Jens Axboe5e84ea32011-03-21 10:14:27 +0100515 if (next_sorted) {
516 elv_rqhash_del(q, next);
517 q->nr_sorted--;
518 }
519
Tejun Heo06b86242005-10-20 16:46:23 +0200520 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521}
522
Divyesh Shah812d4022010-04-08 21:14:23 -0700523void elv_bio_merged(struct request_queue *q, struct request *rq,
524 struct bio *bio)
525{
526 struct elevator_queue *e = q->elevator;
527
Tejun Heo22f746e2011-12-14 00:33:41 +0100528 if (e->type->ops.elevator_bio_merged_fn)
529 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700530}
531
Lin Mingc8158812013-03-23 11:42:27 +0800532#ifdef CONFIG_PM_RUNTIME
533static void blk_pm_requeue_request(struct request *rq)
534{
535 if (rq->q->dev && !(rq->cmd_flags & REQ_PM))
536 rq->q->nr_pending--;
537}
538
539static void blk_pm_add_request(struct request_queue *q, struct request *rq)
540{
541 if (q->dev && !(rq->cmd_flags & REQ_PM) && q->nr_pending++ == 0 &&
542 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
543 pm_request_resume(q->dev);
544}
545#else
546static inline void blk_pm_requeue_request(struct request *rq) {}
547static inline void blk_pm_add_request(struct request_queue *q,
548 struct request *rq)
549{
550}
551#endif
552
Jens Axboe165125e2007-07-24 09:28:11 +0200553void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 /*
556 * it already went through dequeue, we need to decrement the
557 * in_flight count again
558 */
Tejun Heo8922e162005-10-20 16:23:44 +0200559 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200560 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200561 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100562 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Jens Axboe4aff5e22006-08-10 08:44:47 +0200565 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Lin Mingc8158812013-03-23 11:42:27 +0800567 blk_pm_requeue_request(rq);
568
Jens Axboeb710a482011-03-30 09:52:30 +0200569 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570}
571
Jerome Marchand26308ea2009-03-27 10:31:51 +0100572void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100573{
574 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200575
576 lockdep_assert_held(q->queue_lock);
577
Tejun Heo22f746e2011-12-14 00:33:41 +0100578 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100579 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200580 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100581 printk(KERN_ERR "%s: forced dispatching is broken "
582 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100583 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100584 }
585}
586
Jens Axboeb710a482011-03-30 09:52:30 +0200587void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100589 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100590
Lin Mingc8158812013-03-23 11:42:27 +0800591 blk_pm_add_request(q, rq);
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 rq->q = q;
594
Jens Axboeb710a482011-03-30 09:52:30 +0200595 if (rq->cmd_flags & REQ_SOFTBARRIER) {
596 /* barriers are scheduling boundary, update end_sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400597 if (rq->cmd_type == REQ_TYPE_FS) {
Jens Axboeb710a482011-03-30 09:52:30 +0200598 q->end_sector = rq_end_sector(rq);
599 q->boundary_rq = rq;
600 }
601 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200602 (where == ELEVATOR_INSERT_SORT ||
603 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200604 where = ELEVATOR_INSERT_BACK;
605
Tejun Heo8922e162005-10-20 16:23:44 +0200606 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200607 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200608 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200609 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200610 list_add(&rq->queuelist, &q->queue_head);
611 break;
612
613 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200614 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100615 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200616 list_add_tail(&rq->queuelist, &q->queue_head);
617 /*
618 * We kick the queue here for the following reasons.
619 * - The elevator might have returned NULL previously
620 * to delay requests and returned them now. As the
621 * queue wasn't empty before this request, ll_rw_blk
622 * won't run the queue on return, resulting in hang.
623 * - Usually, back inserted requests won't be merged
624 * with anything. There's no point in delaying queue
625 * processing.
626 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200627 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200628 break;
629
Jens Axboe5e84ea32011-03-21 10:14:27 +0100630 case ELEVATOR_INSERT_SORT_MERGE:
631 /*
632 * If we succeed in merging this request with one in the
633 * queue already, we are done - rq has now been freed,
634 * so no need to do anything further.
635 */
636 if (elv_attempt_insert_merge(q, rq))
637 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200638 case ELEVATOR_INSERT_SORT:
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400639 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200640 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100641 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200642 if (rq_mergeable(rq)) {
643 elv_rqhash_add(q, rq);
644 if (!q->last_merge)
645 q->last_merge = rq;
646 }
647
Tejun Heoca235092005-11-01 17:23:49 +0900648 /*
649 * Some ioscheds (cfq) run q->request_fn directly, so
650 * rq cannot be accessed after calling
651 * elevator_add_req_fn.
652 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100653 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200654 break;
655
Tejun Heoae1b1532011-01-25 12:43:54 +0100656 case ELEVATOR_INSERT_FLUSH:
657 rq->cmd_flags |= REQ_SOFTBARRIER;
658 blk_insert_flush(rq);
659 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200660 default:
661 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700662 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200663 BUG();
664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
Jens Axboe2e662b62006-07-13 11:55:04 +0200666EXPORT_SYMBOL(__elv_add_request);
667
Jens Axboe7eaceac2011-03-10 08:52:07 +0100668void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 unsigned long flags;
671
672 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100673 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_unlock_irqrestore(q->queue_lock, flags);
675}
Jens Axboe2e662b62006-07-13 11:55:04 +0200676EXPORT_SYMBOL(elv_add_request);
677
Jens Axboe165125e2007-07-24 09:28:11 +0200678struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Jens Axboeb374d182008-10-31 10:05:07 +0100680 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Tejun Heo22f746e2011-12-14 00:33:41 +0100682 if (e->type->ops.elevator_latter_req_fn)
683 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 return NULL;
685}
686
Jens Axboe165125e2007-07-24 09:28:11 +0200687struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
Jens Axboeb374d182008-10-31 10:05:07 +0100689 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Tejun Heo22f746e2011-12-14 00:33:41 +0100691 if (e->type->ops.elevator_former_req_fn)
692 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return NULL;
694}
695
Tejun Heo852c7882012-03-05 13:15:27 -0800696int elv_set_request(struct request_queue *q, struct request *rq,
697 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Jens Axboeb374d182008-10-31 10:05:07 +0100699 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Tejun Heo22f746e2011-12-14 00:33:41 +0100701 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800702 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return 0;
704}
705
Jens Axboe165125e2007-07-24 09:28:11 +0200706void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
Jens Axboeb374d182008-10-31 10:05:07 +0100708 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Tejun Heo22f746e2011-12-14 00:33:41 +0100710 if (e->type->ops.elevator_put_req_fn)
711 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712}
713
Jens Axboe165125e2007-07-24 09:28:11 +0200714int elv_may_queue(struct request_queue *q, int rw)
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_may_queue_fn)
719 return e->type->ops.elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721 return ELV_MQUEUE_MAY;
722}
723
Mike Anderson11914a52008-09-13 20:31:27 +0200724void elv_abort_queue(struct request_queue *q)
725{
726 struct request *rq;
727
Tejun Heoae1b1532011-01-25 12:43:54 +0100728 blk_abort_flushes(q);
729
Mike Anderson11914a52008-09-13 20:31:27 +0200730 while (!list_empty(&q->queue_head)) {
731 rq = list_entry_rq(q->queue_head.next);
732 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100733 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200734 /*
735 * Mark this request as started so we don't trigger
736 * any debug logic in the end I/O path.
737 */
738 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900739 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200740 }
741}
742EXPORT_SYMBOL(elv_abort_queue);
743
Jens Axboe165125e2007-07-24 09:28:11 +0200744void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Jens Axboeb374d182008-10-31 10:05:07 +0100746 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 /*
749 * request is released from the driver, io must be done
750 */
Tejun Heo8922e162005-10-20 16:23:44 +0200751 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200752 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200753 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100754 e->type->ops.elevator_completed_req_fn)
755 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
Al Viro3d1ab402006-03-18 18:35:43 -0500759#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
760
761static ssize_t
762elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
763{
Al Viro3d1ab402006-03-18 18:35:43 -0500764 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100765 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500766 ssize_t error;
767
768 if (!entry->show)
769 return -EIO;
770
Jens Axboeb374d182008-10-31 10:05:07 +0100771 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500772 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100773 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500774 mutex_unlock(&e->sysfs_lock);
775 return error;
776}
777
778static ssize_t
779elv_attr_store(struct kobject *kobj, struct attribute *attr,
780 const char *page, size_t length)
781{
Al Viro3d1ab402006-03-18 18:35:43 -0500782 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100783 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500784 ssize_t error;
785
786 if (!entry->store)
787 return -EIO;
788
Jens Axboeb374d182008-10-31 10:05:07 +0100789 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500790 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100791 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500792 mutex_unlock(&e->sysfs_lock);
793 return error;
794}
795
Emese Revfy52cf25d2010-01-19 02:58:23 +0100796static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500797 .show = elv_attr_show,
798 .store = elv_attr_store,
799};
800
801static struct kobj_type elv_ktype = {
802 .sysfs_ops = &elv_sysfs_ops,
803 .release = elevator_release,
804};
805
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800806int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800808 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500809 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700811 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500812 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100813 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500814 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500815 while (attr->attr.name) {
816 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500817 break;
Al Viroe572ec72006-03-18 22:27:18 -0500818 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500819 }
820 }
821 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200822 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500823 }
824 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
Tejun Heof8fc8772011-12-14 00:33:40 +0100826EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828void elv_unregister_queue(struct request_queue *q)
829{
Tejun Heof8fc8772011-12-14 00:33:40 +0100830 if (q) {
831 struct elevator_queue *e = q->elevator;
832
833 kobject_uevent(&e->kobj, KOBJ_REMOVE);
834 kobject_del(&e->kobj);
835 e->registered = 0;
836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
Mike Snitzer01effb02010-05-11 08:57:42 +0200838EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Tejun Heo3d3c2372011-12-14 00:33:42 +0100840int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100842 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200843
Tejun Heo3d3c2372011-12-14 00:33:42 +0100844 /* create icq_cache if requested */
845 if (e->icq_size) {
846 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
847 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
848 return -EINVAL;
849
850 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
851 "%s_io_cq", e->elevator_name);
852 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
853 e->icq_align, 0, NULL);
854 if (!e->icq_cache)
855 return -ENOMEM;
856 }
857
858 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200859 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100860 if (elevator_find(e->elevator_name)) {
861 spin_unlock(&elv_list_lock);
862 if (e->icq_cache)
863 kmem_cache_destroy(e->icq_cache);
864 return -EBUSY;
865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200867 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Tejun Heo3d3c2372011-12-14 00:33:42 +0100869 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100870 if (!strcmp(e->elevator_name, chosen_elevator) ||
871 (!*chosen_elevator &&
872 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100873 def = " (default)";
874
Jens Axboe4eb166d2008-02-01 00:37:27 +0100875 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
876 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100877 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879EXPORT_SYMBOL_GPL(elv_register);
880
881void elv_unregister(struct elevator_type *e)
882{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100883 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200884 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200886 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100887
888 /*
889 * Destroy icq_cache if it exists. icq's are RCU managed. Make
890 * sure all RCU operations are complete before proceeding.
891 */
892 if (e->icq_cache) {
893 rcu_barrier();
894 kmem_cache_destroy(e->icq_cache);
895 e->icq_cache = NULL;
896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898EXPORT_SYMBOL_GPL(elv_unregister);
899
900/*
901 * switch to new_e io scheduler. be careful not to introduce deadlocks -
902 * we don't free the old io scheduler, before we have allocated what we
903 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200904 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 */
Jens Axboe165125e2007-07-24 09:28:11 +0200906static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800908 struct elevator_queue *old = q->elevator;
909 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800910 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800912 /*
913 * Turn on BYPASS and drain all requests w/ elevator private data.
914 * Block layer doesn't call into a quiesced elevator - all requests
915 * are directly put on the dispatch list without elevator data
916 * using INSERT_BACK. All requests have SOFTBARRIER set and no
917 * merge happens either.
918 */
Tejun Heod7325802012-03-05 13:14:58 -0800919 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200920
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800921 /* unregister and clear all auxiliary data of the old elevator */
922 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100923 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800924
925 spin_lock_irq(q->queue_lock);
926 ioc_clear_queue(q);
927 spin_unlock_irq(q->queue_lock);
928
929 /* allocate, init and register new elevator */
Jianpeng Mad50235b2013-07-03 13:25:24 +0200930 err = new_e->ops.elevator_init_fn(q, new_e);
931 if (err)
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800932 goto fail_init;
933
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800934 if (registered) {
935 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200936 if (err)
937 goto fail_register;
938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800940 /* done, kill the old one and finish */
941 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800942 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200943
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800944 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200945
Jens Axboe5dd531a2010-08-23 13:52:19 +0200946 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800949 elevator_exit(q->elevator);
950fail_init:
951 /* switch failed, restore and re-register old elevator */
952 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800954 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200955
Jens Axboe5dd531a2010-08-23 13:52:19 +0200956 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
Jens Axboe5dd531a2010-08-23 13:52:19 +0200959/*
960 * Switch this queue to the given IO scheduler.
961 */
962int elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
964 char elevator_name[ELV_NAME_MAX];
965 struct elevator_type *e;
966
Martin K. Petersencd43e262009-05-22 17:17:52 -0400967 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200968 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400969
Li Zefanee2e9922008-10-14 08:49:56 +0200970 strlcpy(elevator_name, name, sizeof(elevator_name));
Tejun Heo21c3c5d2013-01-22 16:48:03 -0800971 e = elevator_get(strstrip(elevator_name), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (!e) {
973 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
974 return -EINVAL;
975 }
976
Tejun Heo22f746e2011-12-14 00:33:41 +0100977 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800978 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200979 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800980 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Jens Axboe5dd531a2010-08-23 13:52:19 +0200982 return elevator_switch(q, e);
983}
984EXPORT_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
994 ret = elevator_change(q, name);
995 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);