blob: 6a55d418896f5ceee0042da69c0177c495219cbc [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Li Zefan55782132009-06-09 13:43:05 +080038#include <trace/events/block.h>
39
Jens Axboe242f9dc2008-09-14 05:55:09 -070040#include "blk.h"
Tejun Heo72e06c22012-03-05 13:15:00 -080041#include "blk-cgroup.h"
Jens Axboe242f9dc2008-09-14 05:55:09 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static DEFINE_SPINLOCK(elv_list_lock);
44static LIST_HEAD(elv_list);
45
46/*
Jens Axboe98170642006-07-28 09:23:08 +020047 * Merge hash stuff.
48 */
49static const int elv_hash_shift = 6;
50#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
Jens Axboe4eb166d2008-02-01 00:37:27 +010051#define ELV_HASH_FN(sec) \
52 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
Jens Axboe98170642006-07-28 09:23:08 +020053#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
Tejun Heo83096eb2009-05-07 22:24:39 +090054#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
Jens Axboe98170642006-07-28 09:23:08 +020055
56/*
Jens Axboeda775262006-12-20 11:04:12 +010057 * Query io scheduler to see if the current process issuing bio may be
58 * merged with rq.
59 */
60static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
61{
Jens Axboe165125e2007-07-24 09:28:11 +020062 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010063 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010064
Tejun Heo22f746e2011-12-14 00:33:41 +010065 if (e->type->ops.elevator_allow_merge_fn)
66 return e->type->ops.elevator_allow_merge_fn(q, rq, bio);
Jens Axboeda775262006-12-20 11:04:12 +010067
68 return 1;
69}
70
71/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * can we safely merge with this request?
73 */
Tejun Heo050c8ea2012-02-08 09:19:38 +010074bool elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Tejun Heo050c8ea2012-02-08 09:19:38 +010076 if (!blk_rq_merge_ok(rq, bio))
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +020077 return 0;
78
Jens Axboeda775262006-12-20 11:04:12 +010079 if (!elv_iosched_allow_merge(rq, bio))
80 return 0;
81
82 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
84EXPORT_SYMBOL(elv_rq_merge_ok);
85
Linus Torvalds1da177e2005-04-16 15:20:36 -070086static struct elevator_type *elevator_find(const char *name)
87{
Vasily Tarasova22b1692006-10-11 09:24:27 +020088 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Matthias Kaehlcke70cee262007-07-10 12:26:24 +020090 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +020091 if (!strcmp(e->elevator_name, name))
92 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Vasily Tarasova22b1692006-10-11 09:24:27 +020095 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
97
98static void elevator_put(struct elevator_type *e)
99{
100 module_put(e->elevator_owner);
101}
102
103static struct elevator_type *elevator_get(const char *name)
104{
Tejun Heo2824bc932005-10-20 10:56:41 +0200105 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200107 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200108
109 e = elevator_find(name);
Jens Axboee1640942008-02-19 10:20:37 +0100110 if (!e) {
Jens Axboee1640942008-02-19 10:20:37 +0100111 spin_unlock(&elv_list_lock);
Kees Cook490b94b2011-05-05 18:02:12 -0600112 request_module("%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100113 spin_lock(&elv_list_lock);
114 e = elevator_find(name);
115 }
116
Tejun Heo2824bc932005-10-20 10:56:41 +0200117 if (e && !try_module_get(e->elevator_owner))
118 e = NULL;
119
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200120 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122 return e;
123}
124
Wang Sheng-Hui484fc252011-09-08 12:32:14 +0200125static char chosen_elevator[ELV_NAME_MAX];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Nate Diller5f003972006-01-24 10:07:58 +0100127static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100129 /*
130 * Be backwards-compatible with previous kernels, so users
131 * won't get the wrong elevator.
132 */
Jens Axboe492af632009-10-03 09:37:51 +0200133 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800134 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
137__setup("elevator=", elevator_setup);
138
Al Viro3d1ab402006-03-18 18:35:43 -0500139static struct kobj_type elv_ktype;
140
Jens Axboeb374d182008-10-31 10:05:07 +0100141static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200142 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500143{
Jens Axboeb374d182008-10-31 10:05:07 +0100144 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200145 int i;
146
Jens Axboeb374d182008-10-31 10:05:07 +0100147 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200148 if (unlikely(!eq))
149 goto err;
150
Tejun Heo22f746e2011-12-14 00:33:41 +0100151 eq->type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700152 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200153 mutex_init(&eq->sysfs_lock);
154
Jens Axboeb5deef92006-07-19 23:39:40 +0200155 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
156 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200157 if (!eq->hash)
158 goto err;
159
160 for (i = 0; i < ELV_HASH_ENTRIES; i++)
161 INIT_HLIST_HEAD(&eq->hash[i]);
162
Al Viro3d1ab402006-03-18 18:35:43 -0500163 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200164err:
165 kfree(eq);
166 elevator_put(e);
167 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500168}
169
170static void elevator_release(struct kobject *kobj)
171{
Jens Axboeb374d182008-10-31 10:05:07 +0100172 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200173
Jens Axboeb374d182008-10-31 10:05:07 +0100174 e = container_of(kobj, struct elevator_queue, kobj);
Tejun Heo22f746e2011-12-14 00:33:41 +0100175 elevator_put(e->type);
Jens Axboe98170642006-07-28 09:23:08 +0200176 kfree(e->hash);
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
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400185 if (unlikely(q->elevator))
186 return 0;
187
Tejun Heocb98fc82005-10-28 08:29:39 +0200188 INIT_LIST_HEAD(&q->queue_head);
189 q->last_merge = NULL;
190 q->end_sector = 0;
191 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200192
Jens Axboe4eb166d2008-02-01 00:37:27 +0100193 if (name) {
194 e = elevator_get(name);
195 if (!e)
196 return -EINVAL;
197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Jens Axboe4eb166d2008-02-01 00:37:27 +0100199 if (!e && *chosen_elevator) {
200 e = elevator_get(chosen_elevator);
201 if (!e)
202 printk(KERN_ERR "I/O scheduler %s not found\n",
203 chosen_elevator);
204 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100205
Jens Axboe4eb166d2008-02-01 00:37:27 +0100206 if (!e) {
207 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
208 if (!e) {
209 printk(KERN_ERR
210 "Default I/O scheduler not found. " \
211 "Using noop.\n");
212 e = elevator_get("noop");
213 }
Nate Diller5f003972006-01-24 10:07:58 +0100214 }
215
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800216 q->elevator = elevator_alloc(q, e);
217 if (!q->elevator)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Tejun Heob2fab5a2012-03-05 13:14:57 -0800220 err = e->ops.elevator_init_fn(q);
Tejun Heof8fc8772011-12-14 00:33:40 +0100221 if (err) {
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800222 kobject_put(&q->elevator->kobj);
Tejun Heof8fc8772011-12-14 00:33:40 +0100223 return err;
Jens Axboebc1c1162006-06-08 08:49:06 +0200224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
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{
243 hlist_del_init(&rq->hash);
244}
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));
257 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
258}
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;
Jens Axboe98170642006-07-28 09:23:08 +0200269 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
270 struct hlist_node *entry, *next;
271 struct request *rq;
272
273 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
274 BUG_ON(!ELV_ON_HASH(rq));
275
276 if (unlikely(!rq_mergeable(rq))) {
277 __elv_rqhash_del(rq);
278 continue;
279 }
280
281 if (rq_hash_key(rq) == offset)
282 return rq;
283 }
284
285 return NULL;
286}
287
Tejun Heo8922e162005-10-20 16:23:44 +0200288/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200289 * RB-tree support functions for inserting/lookup/removal of requests
290 * in a sorted RB tree.
291 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200292void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200293{
294 struct rb_node **p = &root->rb_node;
295 struct rb_node *parent = NULL;
296 struct request *__rq;
297
298 while (*p) {
299 parent = *p;
300 __rq = rb_entry(parent, struct request, rb_node);
301
Tejun Heo83096eb2009-05-07 22:24:39 +0900302 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200303 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200304 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200305 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200306 }
307
308 rb_link_node(&rq->rb_node, parent, p);
309 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200310}
Jens Axboe2e662b62006-07-13 11:55:04 +0200311EXPORT_SYMBOL(elv_rb_add);
312
313void elv_rb_del(struct rb_root *root, struct request *rq)
314{
315 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
316 rb_erase(&rq->rb_node, root);
317 RB_CLEAR_NODE(&rq->rb_node);
318}
Jens Axboe2e662b62006-07-13 11:55:04 +0200319EXPORT_SYMBOL(elv_rb_del);
320
321struct request *elv_rb_find(struct rb_root *root, sector_t sector)
322{
323 struct rb_node *n = root->rb_node;
324 struct request *rq;
325
326 while (n) {
327 rq = rb_entry(n, struct request, rb_node);
328
Tejun Heo83096eb2009-05-07 22:24:39 +0900329 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200330 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900331 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200332 n = n->rb_right;
333 else
334 return rq;
335 }
336
337 return NULL;
338}
Jens Axboe2e662b62006-07-13 11:55:04 +0200339EXPORT_SYMBOL(elv_rb_find);
340
341/*
Tejun Heo8922e162005-10-20 16:23:44 +0200342 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200343 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200344 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200345 */
Jens Axboe165125e2007-07-24 09:28:11 +0200346void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200347{
348 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200349 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100350 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200351
Tejun Heo06b86242005-10-20 16:46:23 +0200352 if (q->last_merge == rq)
353 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200354
355 elv_rqhash_del(q, rq);
356
Tejun Heo15853af2005-11-10 08:52:05 +0100357 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200358
Jens Axboe1b47f532005-10-20 16:37:00 +0200359 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100360 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200361 list_for_each_prev(entry, &q->queue_head) {
362 struct request *pos = list_entry_rq(entry);
363
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200364 if ((rq->cmd_flags & REQ_DISCARD) !=
365 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100366 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100367 if (rq_data_dir(rq) != rq_data_dir(pos))
368 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100369 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200370 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900371 if (blk_rq_pos(rq) >= boundary) {
372 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200373 continue;
374 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900375 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200376 break;
377 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900378 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200379 break;
380 }
381
382 list_add(&rq->queuelist, entry);
383}
Jens Axboe2e662b62006-07-13 11:55:04 +0200384EXPORT_SYMBOL(elv_dispatch_sort);
385
Jens Axboe98170642006-07-28 09:23:08 +0200386/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200387 * Insert rq into dispatch queue of q. Queue lock must be held on
388 * entry. rq is added to the back of the dispatch queue. To be used by
389 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200390 */
391void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
392{
393 if (q->last_merge == rq)
394 q->last_merge = NULL;
395
396 elv_rqhash_del(q, rq);
397
398 q->nr_sorted--;
399
400 q->end_sector = rq_end_sector(rq);
401 q->boundary_rq = rq;
402 list_add_tail(&rq->queuelist, &q->queue_head);
403}
Jens Axboe2e662b62006-07-13 11:55:04 +0200404EXPORT_SYMBOL(elv_dispatch_add_tail);
405
Jens Axboe165125e2007-07-24 09:28:11 +0200406int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Jens Axboeb374d182008-10-31 10:05:07 +0100408 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200409 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200410 int ret;
411
Jens Axboe98170642006-07-28 09:23:08 +0200412 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100413 * Levels of merges:
414 * nomerges: No merges at all attempted
415 * noxmerges: Only simple one-hit cache try
416 * merges: All merge tries attempted
417 */
418 if (blk_queue_nomerges(q))
419 return ELEVATOR_NO_MERGE;
420
421 /*
Jens Axboe98170642006-07-28 09:23:08 +0200422 * First try one-hit cache.
423 */
Tejun Heo050c8ea2012-02-08 09:19:38 +0100424 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
425 ret = blk_try_merge(q->last_merge, bio);
Tejun Heo06b86242005-10-20 16:46:23 +0200426 if (ret != ELEVATOR_NO_MERGE) {
427 *req = q->last_merge;
428 return ret;
429 }
430 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100432 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200433 return ELEVATOR_NO_MERGE;
434
Jens Axboe98170642006-07-28 09:23:08 +0200435 /*
436 * See if our hash lookup can find a potential backmerge.
437 */
438 __rq = elv_rqhash_find(q, bio->bi_sector);
439 if (__rq && elv_rq_merge_ok(__rq, bio)) {
440 *req = __rq;
441 return ELEVATOR_BACK_MERGE;
442 }
443
Tejun Heo22f746e2011-12-14 00:33:41 +0100444 if (e->type->ops.elevator_merge_fn)
445 return e->type->ops.elevator_merge_fn(q, req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
447 return ELEVATOR_NO_MERGE;
448}
449
Jens Axboe5e84ea32011-03-21 10:14:27 +0100450/*
451 * Attempt to do an insertion back merge. Only check for the case where
452 * we can append 'rq' to an existing request, so we can throw 'rq' away
453 * afterwards.
454 *
455 * Returns true if we merged, false otherwise
456 */
457static bool elv_attempt_insert_merge(struct request_queue *q,
458 struct request *rq)
459{
460 struct request *__rq;
461
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
474 /*
475 * See if our hash lookup can find a potential backmerge.
476 */
Jens Axboe5d381ef2012-01-15 10:29:48 +0100477 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
478 if (__rq && blk_attempt_req_merge(q, __rq, rq))
479 return true;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100480
Jens Axboe5d381ef2012-01-15 10:29:48 +0100481 return false;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100482}
483
Jens Axboe165125e2007-07-24 09:28:11 +0200484void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
Jens Axboeb374d182008-10-31 10:05:07 +0100486 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Tejun Heo22f746e2011-12-14 00:33:41 +0100488 if (e->type->ops.elevator_merged_fn)
489 e->type->ops.elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200490
Jens Axboe2e662b62006-07-13 11:55:04 +0200491 if (type == ELEVATOR_BACK_MERGE)
492 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200493
Tejun Heo06b86242005-10-20 16:46:23 +0200494 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
Jens Axboe165125e2007-07-24 09:28:11 +0200497void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 struct request *next)
499{
Jens Axboeb374d182008-10-31 10:05:07 +0100500 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100501 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Tejun Heo22f746e2011-12-14 00:33:41 +0100503 if (next_sorted && e->type->ops.elevator_merge_req_fn)
504 e->type->ops.elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200505
Jens Axboe98170642006-07-28 09:23:08 +0200506 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200507
Jens Axboe5e84ea32011-03-21 10:14:27 +0100508 if (next_sorted) {
509 elv_rqhash_del(q, next);
510 q->nr_sorted--;
511 }
512
Tejun Heo06b86242005-10-20 16:46:23 +0200513 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
Divyesh Shah812d4022010-04-08 21:14:23 -0700516void elv_bio_merged(struct request_queue *q, struct request *rq,
517 struct bio *bio)
518{
519 struct elevator_queue *e = q->elevator;
520
Tejun Heo22f746e2011-12-14 00:33:41 +0100521 if (e->type->ops.elevator_bio_merged_fn)
522 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
Divyesh Shah812d4022010-04-08 21:14:23 -0700523}
524
Jens Axboe165125e2007-07-24 09:28:11 +0200525void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 /*
528 * it already went through dequeue, we need to decrement the
529 * in_flight count again
530 */
Tejun Heo8922e162005-10-20 16:23:44 +0200531 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200532 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200533 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100534 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
Jens Axboe4aff5e22006-08-10 08:44:47 +0200537 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
Jens Axboeb710a482011-03-30 09:52:30 +0200539 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
Jerome Marchand26308ea2009-03-27 10:31:51 +0100542void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100543{
544 static int printed;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200545
546 lockdep_assert_held(q->queue_lock);
547
Tejun Heo22f746e2011-12-14 00:33:41 +0100548 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
Tejun Heo15853af2005-11-10 08:52:05 +0100549 ;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200550 if (q->nr_sorted && printed++ < 10) {
Tejun Heo15853af2005-11-10 08:52:05 +0100551 printk(KERN_ERR "%s: forced dispatching is broken "
552 "(nr_sorted=%u), please report this\n",
Tejun Heo22f746e2011-12-14 00:33:41 +0100553 q->elevator->type->elevator_name, q->nr_sorted);
Tejun Heo15853af2005-11-10 08:52:05 +0100554 }
555}
556
Jens Axboeb710a482011-03-30 09:52:30 +0200557void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100559 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 rq->q = q;
562
Jens Axboeb710a482011-03-30 09:52:30 +0200563 if (rq->cmd_flags & REQ_SOFTBARRIER) {
564 /* barriers are scheduling boundary, update end_sector */
565 if (rq->cmd_type == REQ_TYPE_FS ||
566 (rq->cmd_flags & REQ_DISCARD)) {
567 q->end_sector = rq_end_sector(rq);
568 q->boundary_rq = rq;
569 }
570 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200571 (where == ELEVATOR_INSERT_SORT ||
572 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200573 where = ELEVATOR_INSERT_BACK;
574
Tejun Heo8922e162005-10-20 16:23:44 +0200575 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200576 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200577 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200578 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200579 list_add(&rq->queuelist, &q->queue_head);
580 break;
581
582 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200583 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100584 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200585 list_add_tail(&rq->queuelist, &q->queue_head);
586 /*
587 * We kick the queue here for the following reasons.
588 * - The elevator might have returned NULL previously
589 * to delay requests and returned them now. As the
590 * queue wasn't empty before this request, ll_rw_blk
591 * won't run the queue on return, resulting in hang.
592 * - Usually, back inserted requests won't be merged
593 * with anything. There's no point in delaying queue
594 * processing.
595 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200596 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200597 break;
598
Jens Axboe5e84ea32011-03-21 10:14:27 +0100599 case ELEVATOR_INSERT_SORT_MERGE:
600 /*
601 * If we succeed in merging this request with one in the
602 * queue already, we are done - rq has now been freed,
603 * so no need to do anything further.
604 */
605 if (elv_attempt_insert_merge(q, rq))
606 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200607 case ELEVATOR_INSERT_SORT:
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200608 BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
609 !(rq->cmd_flags & REQ_DISCARD));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200610 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100611 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200612 if (rq_mergeable(rq)) {
613 elv_rqhash_add(q, rq);
614 if (!q->last_merge)
615 q->last_merge = rq;
616 }
617
Tejun Heoca235092005-11-01 17:23:49 +0900618 /*
619 * Some ioscheds (cfq) run q->request_fn directly, so
620 * rq cannot be accessed after calling
621 * elevator_add_req_fn.
622 */
Tejun Heo22f746e2011-12-14 00:33:41 +0100623 q->elevator->type->ops.elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200624 break;
625
Tejun Heoae1b1532011-01-25 12:43:54 +0100626 case ELEVATOR_INSERT_FLUSH:
627 rq->cmd_flags |= REQ_SOFTBARRIER;
628 blk_insert_flush(rq);
629 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200630 default:
631 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700632 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200633 BUG();
634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
Jens Axboe2e662b62006-07-13 11:55:04 +0200636EXPORT_SYMBOL(__elv_add_request);
637
Jens Axboe7eaceac2011-03-10 08:52:07 +0100638void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
640 unsigned long flags;
641
642 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100643 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 spin_unlock_irqrestore(q->queue_lock, flags);
645}
Jens Axboe2e662b62006-07-13 11:55:04 +0200646EXPORT_SYMBOL(elv_add_request);
647
Jens Axboe165125e2007-07-24 09:28:11 +0200648struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Jens Axboeb374d182008-10-31 10:05:07 +0100650 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Tejun Heo22f746e2011-12-14 00:33:41 +0100652 if (e->type->ops.elevator_latter_req_fn)
653 return e->type->ops.elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 return NULL;
655}
656
Jens Axboe165125e2007-07-24 09:28:11 +0200657struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
Jens Axboeb374d182008-10-31 10:05:07 +0100659 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Tejun Heo22f746e2011-12-14 00:33:41 +0100661 if (e->type->ops.elevator_former_req_fn)
662 return e->type->ops.elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 return NULL;
664}
665
Tejun Heo852c7882012-03-05 13:15:27 -0800666int elv_set_request(struct request_queue *q, struct request *rq,
667 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
Jens Axboeb374d182008-10-31 10:05:07 +0100669 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Tejun Heo22f746e2011-12-14 00:33:41 +0100671 if (e->type->ops.elevator_set_req_fn)
Tejun Heo852c7882012-03-05 13:15:27 -0800672 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 return 0;
674}
675
Jens Axboe165125e2007-07-24 09:28:11 +0200676void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677{
Jens Axboeb374d182008-10-31 10:05:07 +0100678 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Tejun Heo22f746e2011-12-14 00:33:41 +0100680 if (e->type->ops.elevator_put_req_fn)
681 e->type->ops.elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
Jens Axboe165125e2007-07-24 09:28:11 +0200684int elv_may_queue(struct request_queue *q, int rw)
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_may_queue_fn)
689 return e->type->ops.elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691 return ELV_MQUEUE_MAY;
692}
693
Mike Anderson11914a52008-09-13 20:31:27 +0200694void elv_abort_queue(struct request_queue *q)
695{
696 struct request *rq;
697
Tejun Heoae1b1532011-01-25 12:43:54 +0100698 blk_abort_flushes(q);
699
Mike Anderson11914a52008-09-13 20:31:27 +0200700 while (!list_empty(&q->queue_head)) {
701 rq = list_entry_rq(q->queue_head.next);
702 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100703 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200704 /*
705 * Mark this request as started so we don't trigger
706 * any debug logic in the end I/O path.
707 */
708 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900709 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200710 }
711}
712EXPORT_SYMBOL(elv_abort_queue);
713
Jens Axboe165125e2007-07-24 09:28:11 +0200714void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Jens Axboeb374d182008-10-31 10:05:07 +0100716 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 /*
719 * request is released from the driver, io must be done
720 */
Tejun Heo8922e162005-10-20 16:23:44 +0200721 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200722 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200723 if ((rq->cmd_flags & REQ_SORTED) &&
Tejun Heo22f746e2011-12-14 00:33:41 +0100724 e->type->ops.elevator_completed_req_fn)
725 e->type->ops.elevator_completed_req_fn(q, rq);
Tejun Heo1bc691d2006-01-12 15:39:26 +0100726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727}
728
Al Viro3d1ab402006-03-18 18:35:43 -0500729#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
730
731static ssize_t
732elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
733{
Al Viro3d1ab402006-03-18 18:35:43 -0500734 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100735 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500736 ssize_t error;
737
738 if (!entry->show)
739 return -EIO;
740
Jens Axboeb374d182008-10-31 10:05:07 +0100741 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500742 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100743 error = e->type ? entry->show(e, page) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500744 mutex_unlock(&e->sysfs_lock);
745 return error;
746}
747
748static ssize_t
749elv_attr_store(struct kobject *kobj, struct attribute *attr,
750 const char *page, size_t length)
751{
Al Viro3d1ab402006-03-18 18:35:43 -0500752 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100753 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500754 ssize_t error;
755
756 if (!entry->store)
757 return -EIO;
758
Jens Axboeb374d182008-10-31 10:05:07 +0100759 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500760 mutex_lock(&e->sysfs_lock);
Tejun Heo22f746e2011-12-14 00:33:41 +0100761 error = e->type ? entry->store(e, page, length) : -ENOENT;
Al Viro3d1ab402006-03-18 18:35:43 -0500762 mutex_unlock(&e->sysfs_lock);
763 return error;
764}
765
Emese Revfy52cf25d2010-01-19 02:58:23 +0100766static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500767 .show = elv_attr_show,
768 .store = elv_attr_store,
769};
770
771static struct kobj_type elv_ktype = {
772 .sysfs_ops = &elv_sysfs_ops,
773 .release = elevator_release,
774};
775
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800776int elv_register_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800778 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500779 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700781 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500782 if (!error) {
Tejun Heo22f746e2011-12-14 00:33:41 +0100783 struct elv_fs_entry *attr = e->type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500784 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500785 while (attr->attr.name) {
786 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500787 break;
Al Viroe572ec72006-03-18 22:27:18 -0500788 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500789 }
790 }
791 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200792 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500793 }
794 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
Tejun Heof8fc8772011-12-14 00:33:40 +0100796EXPORT_SYMBOL(elv_register_queue);
Jens Axboebc1c1162006-06-08 08:49:06 +0200797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798void elv_unregister_queue(struct request_queue *q)
799{
Tejun Heof8fc8772011-12-14 00:33:40 +0100800 if (q) {
801 struct elevator_queue *e = q->elevator;
802
803 kobject_uevent(&e->kobj, KOBJ_REMOVE);
804 kobject_del(&e->kobj);
805 e->registered = 0;
806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
Mike Snitzer01effb02010-05-11 08:57:42 +0200808EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Tejun Heo3d3c2372011-12-14 00:33:42 +0100810int elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100812 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200813
Tejun Heo3d3c2372011-12-14 00:33:42 +0100814 /* create icq_cache if requested */
815 if (e->icq_size) {
816 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
817 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
818 return -EINVAL;
819
820 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
821 "%s_io_cq", e->elevator_name);
822 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
823 e->icq_align, 0, NULL);
824 if (!e->icq_cache)
825 return -ENOMEM;
826 }
827
828 /* register, don't allow duplicate names */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200829 spin_lock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100830 if (elevator_find(e->elevator_name)) {
831 spin_unlock(&elv_list_lock);
832 if (e->icq_cache)
833 kmem_cache_destroy(e->icq_cache);
834 return -EBUSY;
835 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200837 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Tejun Heo3d3c2372011-12-14 00:33:42 +0100839 /* print pretty message */
Nate Diller5f003972006-01-24 10:07:58 +0100840 if (!strcmp(e->elevator_name, chosen_elevator) ||
841 (!*chosen_elevator &&
842 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100843 def = " (default)";
844
Jens Axboe4eb166d2008-02-01 00:37:27 +0100845 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
846 def);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100847 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848}
849EXPORT_SYMBOL_GPL(elv_register);
850
851void elv_unregister(struct elevator_type *e)
852{
Tejun Heo3d3c2372011-12-14 00:33:42 +0100853 /* unregister */
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200854 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200856 spin_unlock(&elv_list_lock);
Tejun Heo3d3c2372011-12-14 00:33:42 +0100857
858 /*
859 * Destroy icq_cache if it exists. icq's are RCU managed. Make
860 * sure all RCU operations are complete before proceeding.
861 */
862 if (e->icq_cache) {
863 rcu_barrier();
864 kmem_cache_destroy(e->icq_cache);
865 e->icq_cache = NULL;
866 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867}
868EXPORT_SYMBOL_GPL(elv_unregister);
869
870/*
871 * switch to new_e io scheduler. be careful not to introduce deadlocks -
872 * we don't free the old io scheduler, before we have allocated what we
873 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200874 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 */
Jens Axboe165125e2007-07-24 09:28:11 +0200876static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800878 struct elevator_queue *old = q->elevator;
879 bool registered = old->registered;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800880 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800882 /*
883 * Turn on BYPASS and drain all requests w/ elevator private data.
884 * Block layer doesn't call into a quiesced elevator - all requests
885 * are directly put on the dispatch list without elevator data
886 * using INSERT_BACK. All requests have SOFTBARRIER set and no
887 * merge happens either.
888 */
Tejun Heod7325802012-03-05 13:14:58 -0800889 blk_queue_bypass_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200890
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800891 /* unregister and clear all auxiliary data of the old elevator */
892 if (registered)
Tejun Heof8fc8772011-12-14 00:33:40 +0100893 elv_unregister_queue(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800894
895 spin_lock_irq(q->queue_lock);
896 ioc_clear_queue(q);
897 spin_unlock_irq(q->queue_lock);
898
899 /* allocate, init and register new elevator */
900 err = -ENOMEM;
901 q->elevator = elevator_alloc(q, new_e);
902 if (!q->elevator)
903 goto fail_init;
904
Tejun Heob2fab5a2012-03-05 13:14:57 -0800905 err = new_e->ops.elevator_init_fn(q);
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800906 if (err) {
907 kobject_put(&q->elevator->kobj);
908 goto fail_init;
909 }
910
911 if (registered) {
912 err = elv_register_queue(q);
Jens Axboe430c62f2010-10-07 09:35:16 +0200913 if (err)
914 goto fail_register;
915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800917 /* done, kill the old one and finish */
918 elevator_exit(old);
Tejun Heod7325802012-03-05 13:14:58 -0800919 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200920
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800921 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
Alan D. Brunelle4722dc52008-05-27 14:55:00 +0200922
Jens Axboe5dd531a2010-08-23 13:52:19 +0200923 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925fail_register:
Tejun Heo5a5bafd2012-03-05 13:14:56 -0800926 elevator_exit(q->elevator);
927fail_init:
928 /* switch failed, restore and re-register old elevator */
929 q->elevator = old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 elv_register_queue(q);
Tejun Heod7325802012-03-05 13:14:58 -0800931 blk_queue_bypass_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200932
Jens Axboe5dd531a2010-08-23 13:52:19 +0200933 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
Jens Axboe5dd531a2010-08-23 13:52:19 +0200936/*
937 * Switch this queue to the given IO scheduler.
938 */
939int elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 char elevator_name[ELV_NAME_MAX];
942 struct elevator_type *e;
943
Martin K. Petersencd43e262009-05-22 17:17:52 -0400944 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200945 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400946
Li Zefanee2e9922008-10-14 08:49:56 +0200947 strlcpy(elevator_name, name, sizeof(elevator_name));
KOSAKI Motohiro8c279592009-10-09 08:48:08 +0200948 e = elevator_get(strstrip(elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (!e) {
950 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
951 return -EINVAL;
952 }
953
Tejun Heo22f746e2011-12-14 00:33:41 +0100954 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
Nate Diller2ca7d932005-10-30 15:02:24 -0800955 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200956 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -0800957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Jens Axboe5dd531a2010-08-23 13:52:19 +0200959 return elevator_switch(q, e);
960}
961EXPORT_SYMBOL(elevator_change);
962
963ssize_t elv_iosched_store(struct request_queue *q, const char *name,
964 size_t count)
965{
966 int ret;
967
968 if (!q->elevator)
969 return count;
970
971 ret = elevator_change(q, name);
972 if (!ret)
973 return count;
974
975 printk(KERN_ERR "elevator: switch to %s failed\n", name);
976 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978
Jens Axboe165125e2007-07-24 09:28:11 +0200979ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980{
Jens Axboeb374d182008-10-31 10:05:07 +0100981 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400982 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200983 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 int len = 0;
985
Mike Snitzere36f7242010-05-24 09:07:32 +0200986 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -0400987 return sprintf(name, "none\n");
988
Tejun Heo22f746e2011-12-14 00:33:41 +0100989 elv = e->type;
Martin K. Petersencd43e262009-05-22 17:17:52 -0400990
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200991 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200992 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 if (!strcmp(elv->elevator_name, __e->elevator_name))
994 len += sprintf(name+len, "[%s] ", elv->elevator_name);
995 else
996 len += sprintf(name+len, "%s ", __e->elevator_name);
997 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200998 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
1000 len += sprintf(len+name, "\n");
1001 return len;
1002}
1003
Jens Axboe165125e2007-07-24 09:28:11 +02001004struct request *elv_rb_former_request(struct request_queue *q,
1005 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001006{
1007 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1008
1009 if (rbprev)
1010 return rb_entry_rq(rbprev);
1011
1012 return NULL;
1013}
Jens Axboe2e662b62006-07-13 11:55:04 +02001014EXPORT_SYMBOL(elv_rb_former_request);
1015
Jens Axboe165125e2007-07-24 09:28:11 +02001016struct request *elv_rb_latter_request(struct request_queue *q,
1017 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001018{
1019 struct rb_node *rbnext = rb_next(&rq->rb_node);
1020
1021 if (rbnext)
1022 return rb_entry_rq(rbnext);
1023
1024 return NULL;
1025}
Jens Axboe2e662b62006-07-13 11:55:04 +02001026EXPORT_SYMBOL(elv_rb_latter_request);