blob: 536be740ba4e6443f4ecb3501b4d45631385e4e6 [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>
Tejun Heocb98fc82005-10-28 08:29:39 +020034#include <linux/delay.h>
Jens Axboe2056a782006-03-23 20:00:26 +010035#include <linux/blktrace_api.h>
Jens Axboe98170642006-07-28 09:23:08 +020036#include <linux/hash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/uaccess.h>
39
40static DEFINE_SPINLOCK(elv_list_lock);
41static LIST_HEAD(elv_list);
42
43/*
Jens Axboe98170642006-07-28 09:23:08 +020044 * Merge hash stuff.
45 */
46static const int elv_hash_shift = 6;
47#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
48#define ELV_HASH_FN(sec) (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
49#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
50#define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors)
51#define ELV_ON_HASH(rq) (!hlist_unhashed(&(rq)->hash))
52
53/*
Jens Axboeda775262006-12-20 11:04:12 +010054 * Query io scheduler to see if the current process issuing bio may be
55 * merged with rq.
56 */
57static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
58{
59 request_queue_t *q = rq->q;
60 elevator_t *e = q->elevator;
61
62 if (e->ops->elevator_allow_merge_fn)
63 return e->ops->elevator_allow_merge_fn(q, rq, bio);
64
65 return 1;
66}
67
68/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * can we safely merge with this request?
70 */
71inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
72{
73 if (!rq_mergeable(rq))
74 return 0;
75
76 /*
77 * different data direction or already started, don't merge
78 */
79 if (bio_data_dir(bio) != rq_data_dir(rq))
80 return 0;
81
82 /*
Jens Axboeda775262006-12-20 11:04:12 +010083 * must be same device and not a special request
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 */
Jens Axboebb4067e2006-12-21 21:20:01 +010085 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
Jens Axboeda775262006-12-20 11:04:12 +010086 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Jens Axboeda775262006-12-20 11:04:12 +010088 if (!elv_iosched_allow_merge(rq, bio))
89 return 0;
90
91 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93EXPORT_SYMBOL(elv_rq_merge_ok);
94
Coywolf Qi Hunt769db452005-12-28 10:55:49 +010095static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
97 int ret = ELEVATOR_NO_MERGE;
98
99 /*
100 * we can merge and sequence is ok, check if it's possible
101 */
102 if (elv_rq_merge_ok(__rq, bio)) {
103 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
104 ret = ELEVATOR_BACK_MERGE;
105 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
106 ret = ELEVATOR_FRONT_MERGE;
107 }
108
109 return ret;
110}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112static struct elevator_type *elevator_find(const char *name)
113{
Vasily Tarasova22b1692006-10-11 09:24:27 +0200114 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 struct list_head *entry;
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 list_for_each(entry, &elv_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Vasily Tarasova22b1692006-10-11 09:24:27 +0200119 e = list_entry(entry, struct elevator_type, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Vasily Tarasova22b1692006-10-11 09:24:27 +0200121 if (!strcmp(e->elevator_name, name))
122 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Vasily Tarasova22b1692006-10-11 09:24:27 +0200125 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126}
127
128static void elevator_put(struct elevator_type *e)
129{
130 module_put(e->elevator_owner);
131}
132
133static struct elevator_type *elevator_get(const char *name)
134{
Tejun Heo2824bc932005-10-20 10:56:41 +0200135 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Tejun Heo2824bc932005-10-20 10:56:41 +0200137 spin_lock_irq(&elv_list_lock);
138
139 e = elevator_find(name);
140 if (e && !try_module_get(e->elevator_owner))
141 e = NULL;
142
143 spin_unlock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145 return e;
146}
147
Jens Axboebc1c1162006-06-08 08:49:06 +0200148static void *elevator_init_queue(request_queue_t *q, struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149{
Jens Axboebb37b942006-12-01 10:42:33 +0100150 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200151}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Jens Axboebc1c1162006-06-08 08:49:06 +0200153static void elevator_attach(request_queue_t *q, struct elevator_queue *eq,
154 void *data)
155{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200157 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
160static char chosen_elevator[16];
161
Nate Diller5f003972006-01-24 10:07:58 +0100162static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100164 /*
165 * Be backwards-compatible with previous kernels, so users
166 * won't get the wrong elevator.
167 */
Nate Diller5f003972006-01-24 10:07:58 +0100168 if (!strcmp(str, "as"))
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100169 strcpy(chosen_elevator, "anticipatory");
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100170 else
Nate Diller5f003972006-01-24 10:07:58 +0100171 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800172 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
175__setup("elevator=", elevator_setup);
176
Al Viro3d1ab402006-03-18 18:35:43 -0500177static struct kobj_type elv_ktype;
178
Jens Axboeb5deef92006-07-19 23:39:40 +0200179static elevator_t *elevator_alloc(request_queue_t *q, struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500180{
Jens Axboe98170642006-07-28 09:23:08 +0200181 elevator_t *eq;
182 int i;
183
Jens Axboeb5deef92006-07-19 23:39:40 +0200184 eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200185 if (unlikely(!eq))
186 goto err;
187
188 memset(eq, 0, sizeof(*eq));
189 eq->ops = &e->ops;
190 eq->elevator_type = e;
191 kobject_init(&eq->kobj);
192 snprintf(eq->kobj.name, KOBJ_NAME_LEN, "%s", "iosched");
193 eq->kobj.ktype = &elv_ktype;
194 mutex_init(&eq->sysfs_lock);
195
Jens Axboeb5deef92006-07-19 23:39:40 +0200196 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
197 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200198 if (!eq->hash)
199 goto err;
200
201 for (i = 0; i < ELV_HASH_ENTRIES; i++)
202 INIT_HLIST_HEAD(&eq->hash[i]);
203
Al Viro3d1ab402006-03-18 18:35:43 -0500204 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200205err:
206 kfree(eq);
207 elevator_put(e);
208 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500209}
210
211static void elevator_release(struct kobject *kobj)
212{
213 elevator_t *e = container_of(kobj, elevator_t, kobj);
Jens Axboe98170642006-07-28 09:23:08 +0200214
Al Viro3d1ab402006-03-18 18:35:43 -0500215 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200216 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500217 kfree(e);
218}
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220int elevator_init(request_queue_t *q, char *name)
221{
222 struct elevator_type *e = NULL;
223 struct elevator_queue *eq;
224 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200225 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Tejun Heocb98fc82005-10-28 08:29:39 +0200227 INIT_LIST_HEAD(&q->queue_head);
228 q->last_merge = NULL;
229 q->end_sector = 0;
230 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200231
Nate Diller5f003972006-01-24 10:07:58 +0100232 if (name && !(e = elevator_get(name)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return -EINVAL;
234
Nate Diller248d5ca2006-01-24 10:09:14 +0100235 if (!e && *chosen_elevator && !(e = elevator_get(chosen_elevator)))
236 printk("I/O scheduler %s not found\n", chosen_elevator);
237
238 if (!e && !(e = elevator_get(CONFIG_DEFAULT_IOSCHED))) {
239 printk("Default I/O scheduler not found, using no-op\n");
240 e = elevator_get("noop");
Nate Diller5f003972006-01-24 10:07:58 +0100241 }
242
Jens Axboeb5deef92006-07-19 23:39:40 +0200243 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500244 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Jens Axboebc1c1162006-06-08 08:49:06 +0200247 data = elevator_init_queue(q, eq);
248 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500249 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200250 return -ENOMEM;
251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Jens Axboebc1c1162006-06-08 08:49:06 +0200253 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 return ret;
255}
256
Jens Axboe2e662b62006-07-13 11:55:04 +0200257EXPORT_SYMBOL(elevator_init);
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259void elevator_exit(elevator_t *e)
260{
Al Viro3d1ab402006-03-18 18:35:43 -0500261 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 if (e->ops->elevator_exit_fn)
263 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500264 e->ops = NULL;
265 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Al Viro3d1ab402006-03-18 18:35:43 -0500267 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268}
269
Jens Axboe2e662b62006-07-13 11:55:04 +0200270EXPORT_SYMBOL(elevator_exit);
271
Jens Axboe98170642006-07-28 09:23:08 +0200272static inline void __elv_rqhash_del(struct request *rq)
273{
274 hlist_del_init(&rq->hash);
275}
276
277static void elv_rqhash_del(request_queue_t *q, struct request *rq)
278{
279 if (ELV_ON_HASH(rq))
280 __elv_rqhash_del(rq);
281}
282
283static void elv_rqhash_add(request_queue_t *q, struct request *rq)
284{
285 elevator_t *e = q->elevator;
286
287 BUG_ON(ELV_ON_HASH(rq));
288 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
289}
290
291static void elv_rqhash_reposition(request_queue_t *q, struct request *rq)
292{
293 __elv_rqhash_del(rq);
294 elv_rqhash_add(q, rq);
295}
296
297static struct request *elv_rqhash_find(request_queue_t *q, sector_t offset)
298{
299 elevator_t *e = q->elevator;
300 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
301 struct hlist_node *entry, *next;
302 struct request *rq;
303
304 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
305 BUG_ON(!ELV_ON_HASH(rq));
306
307 if (unlikely(!rq_mergeable(rq))) {
308 __elv_rqhash_del(rq);
309 continue;
310 }
311
312 if (rq_hash_key(rq) == offset)
313 return rq;
314 }
315
316 return NULL;
317}
318
Tejun Heo8922e162005-10-20 16:23:44 +0200319/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200320 * RB-tree support functions for inserting/lookup/removal of requests
321 * in a sorted RB tree.
322 */
323struct request *elv_rb_add(struct rb_root *root, struct request *rq)
324{
325 struct rb_node **p = &root->rb_node;
326 struct rb_node *parent = NULL;
327 struct request *__rq;
328
329 while (*p) {
330 parent = *p;
331 __rq = rb_entry(parent, struct request, rb_node);
332
333 if (rq->sector < __rq->sector)
334 p = &(*p)->rb_left;
335 else if (rq->sector > __rq->sector)
336 p = &(*p)->rb_right;
337 else
338 return __rq;
339 }
340
341 rb_link_node(&rq->rb_node, parent, p);
342 rb_insert_color(&rq->rb_node, root);
343 return NULL;
344}
345
346EXPORT_SYMBOL(elv_rb_add);
347
348void elv_rb_del(struct rb_root *root, struct request *rq)
349{
350 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
351 rb_erase(&rq->rb_node, root);
352 RB_CLEAR_NODE(&rq->rb_node);
353}
354
355EXPORT_SYMBOL(elv_rb_del);
356
357struct request *elv_rb_find(struct rb_root *root, sector_t sector)
358{
359 struct rb_node *n = root->rb_node;
360 struct request *rq;
361
362 while (n) {
363 rq = rb_entry(n, struct request, rb_node);
364
365 if (sector < rq->sector)
366 n = n->rb_left;
367 else if (sector > rq->sector)
368 n = n->rb_right;
369 else
370 return rq;
371 }
372
373 return NULL;
374}
375
376EXPORT_SYMBOL(elv_rb_find);
377
378/*
Tejun Heo8922e162005-10-20 16:23:44 +0200379 * Insert rq into dispatch queue of q. Queue lock must be held on
Jens Axboe2e662b62006-07-13 11:55:04 +0200380 * entry. rq is sort insted into the dispatch queue. To be used by
381 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200382 */
Jens Axboe1b47f532005-10-20 16:37:00 +0200383void elv_dispatch_sort(request_queue_t *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200384{
385 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200386 struct list_head *entry;
387
Tejun Heo06b86242005-10-20 16:46:23 +0200388 if (q->last_merge == rq)
389 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200390
391 elv_rqhash_del(q, rq);
392
Tejun Heo15853af2005-11-10 08:52:05 +0100393 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200394
Jens Axboe1b47f532005-10-20 16:37:00 +0200395 boundary = q->end_sector;
Tejun Heocb198332005-10-24 08:35:58 +0200396
Tejun Heo8922e162005-10-20 16:23:44 +0200397 list_for_each_prev(entry, &q->queue_head) {
398 struct request *pos = list_entry_rq(entry);
399
Jens Axboe4aff5e22006-08-10 08:44:47 +0200400 if (pos->cmd_flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED))
Tejun Heo8922e162005-10-20 16:23:44 +0200401 break;
402 if (rq->sector >= boundary) {
403 if (pos->sector < boundary)
404 continue;
405 } else {
406 if (pos->sector >= boundary)
407 break;
408 }
409 if (rq->sector >= pos->sector)
410 break;
411 }
412
413 list_add(&rq->queuelist, entry);
414}
415
Jens Axboe2e662b62006-07-13 11:55:04 +0200416EXPORT_SYMBOL(elv_dispatch_sort);
417
Jens Axboe98170642006-07-28 09:23:08 +0200418/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200419 * Insert rq into dispatch queue of q. Queue lock must be held on
420 * entry. rq is added to the back of the dispatch queue. To be used by
421 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200422 */
423void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
424{
425 if (q->last_merge == rq)
426 q->last_merge = NULL;
427
428 elv_rqhash_del(q, rq);
429
430 q->nr_sorted--;
431
432 q->end_sector = rq_end_sector(rq);
433 q->boundary_rq = rq;
434 list_add_tail(&rq->queuelist, &q->queue_head);
435}
436
Jens Axboe2e662b62006-07-13 11:55:04 +0200437EXPORT_SYMBOL(elv_dispatch_add_tail);
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439int elv_merge(request_queue_t *q, struct request **req, struct bio *bio)
440{
441 elevator_t *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200442 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200443 int ret;
444
Jens Axboe98170642006-07-28 09:23:08 +0200445 /*
446 * First try one-hit cache.
447 */
Tejun Heo06b86242005-10-20 16:46:23 +0200448 if (q->last_merge) {
449 ret = elv_try_merge(q->last_merge, bio);
450 if (ret != ELEVATOR_NO_MERGE) {
451 *req = q->last_merge;
452 return ret;
453 }
454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
Jens Axboe98170642006-07-28 09:23:08 +0200456 /*
457 * See if our hash lookup can find a potential backmerge.
458 */
459 __rq = elv_rqhash_find(q, bio->bi_sector);
460 if (__rq && elv_rq_merge_ok(__rq, bio)) {
461 *req = __rq;
462 return ELEVATOR_BACK_MERGE;
463 }
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 if (e->ops->elevator_merge_fn)
466 return e->ops->elevator_merge_fn(q, req, bio);
467
468 return ELEVATOR_NO_MERGE;
469}
470
Jens Axboe2e662b62006-07-13 11:55:04 +0200471void elv_merged_request(request_queue_t *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
473 elevator_t *e = q->elevator;
474
475 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200476 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200477
Jens Axboe2e662b62006-07-13 11:55:04 +0200478 if (type == ELEVATOR_BACK_MERGE)
479 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200480
Tejun Heo06b86242005-10-20 16:46:23 +0200481 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
484void elv_merge_requests(request_queue_t *q, struct request *rq,
485 struct request *next)
486{
487 elevator_t *e = q->elevator;
488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (e->ops->elevator_merge_req_fn)
490 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200491
Jens Axboe98170642006-07-28 09:23:08 +0200492 elv_rqhash_reposition(q, rq);
493 elv_rqhash_del(q, next);
494
495 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200496 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
Tejun Heo8922e162005-10-20 16:23:44 +0200499void elv_requeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 elevator_t *e = q->elevator;
502
503 /*
504 * it already went through dequeue, we need to decrement the
505 * in_flight count again
506 */
Tejun Heo8922e162005-10-20 16:23:44 +0200507 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 q->in_flight--;
Tejun Heo8922e162005-10-20 16:23:44 +0200509 if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn)
510 e->ops->elevator_deactivate_req_fn(q, rq);
511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Jens Axboe4aff5e22006-08-10 08:44:47 +0200513 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Tejun Heo30e96562006-02-08 01:01:31 -0800515 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
Tejun Heo15853af2005-11-10 08:52:05 +0100518static void elv_drain_elevator(request_queue_t *q)
519{
520 static int printed;
521 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
522 ;
523 if (q->nr_sorted == 0)
524 return;
525 if (printed++ < 10) {
526 printk(KERN_ERR "%s: forced dispatching is broken "
527 "(nr_sorted=%u), please report this\n",
528 q->elevator->elevator_type->elevator_name, q->nr_sorted);
529 }
530}
531
Tejun Heo30e96562006-02-08 01:01:31 -0800532void elv_insert(request_queue_t *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
Tejun Heo797e7db2006-01-06 09:51:03 +0100534 struct list_head *pos;
535 unsigned ordseq;
Jens Axboedac07ec2006-05-11 08:20:16 +0200536 int unplug_it = 1;
Tejun Heo797e7db2006-01-06 09:51:03 +0100537
Jens Axboe2056a782006-03-23 20:00:26 +0100538 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 rq->q = q;
541
Tejun Heo8922e162005-10-20 16:23:44 +0200542 switch (where) {
543 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200544 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200545
546 list_add(&rq->queuelist, &q->queue_head);
547 break;
548
549 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200550 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100551 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200552 list_add_tail(&rq->queuelist, &q->queue_head);
553 /*
554 * We kick the queue here for the following reasons.
555 * - The elevator might have returned NULL previously
556 * to delay requests and returned them now. As the
557 * queue wasn't empty before this request, ll_rw_blk
558 * won't run the queue on return, resulting in hang.
559 * - Usually, back inserted requests won't be merged
560 * with anything. There's no point in delaying queue
561 * processing.
562 */
563 blk_remove_plug(q);
564 q->request_fn(q);
565 break;
566
567 case ELEVATOR_INSERT_SORT:
568 BUG_ON(!blk_fs_request(rq));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200569 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100570 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200571 if (rq_mergeable(rq)) {
572 elv_rqhash_add(q, rq);
573 if (!q->last_merge)
574 q->last_merge = rq;
575 }
576
Tejun Heoca235092005-11-01 17:23:49 +0900577 /*
578 * Some ioscheds (cfq) run q->request_fn directly, so
579 * rq cannot be accessed after calling
580 * elevator_add_req_fn.
581 */
582 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200583 break;
584
Tejun Heo797e7db2006-01-06 09:51:03 +0100585 case ELEVATOR_INSERT_REQUEUE:
586 /*
587 * If ordered flush isn't in progress, we do front
588 * insertion; otherwise, requests should be requeued
589 * in ordseq order.
590 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200591 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100592
593 if (q->ordseq == 0) {
594 list_add(&rq->queuelist, &q->queue_head);
595 break;
596 }
597
598 ordseq = blk_ordered_req_seq(rq);
599
600 list_for_each(pos, &q->queue_head) {
601 struct request *pos_rq = list_entry_rq(pos);
602 if (ordseq <= blk_ordered_req_seq(pos_rq))
603 break;
604 }
605
606 list_add_tail(&rq->queuelist, pos);
Jens Axboedac07ec2006-05-11 08:20:16 +0200607 /*
608 * most requeues happen because of a busy condition, don't
609 * force unplug of the queue for that case.
610 */
611 unplug_it = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100612 break;
613
Tejun Heo8922e162005-10-20 16:23:44 +0200614 default:
615 printk(KERN_ERR "%s: bad insertion point %d\n",
616 __FUNCTION__, where);
617 BUG();
618 }
619
Jens Axboedac07ec2006-05-11 08:20:16 +0200620 if (unplug_it && blk_queue_plugged(q)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200621 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
622 - q->in_flight;
623
624 if (nrq >= q->unplug_thresh)
625 __generic_unplug_device(q);
626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
628
Tejun Heo30e96562006-02-08 01:01:31 -0800629void __elv_add_request(request_queue_t *q, struct request *rq, int where,
630 int plug)
631{
632 if (q->ordcolor)
Jens Axboe4aff5e22006-08-10 08:44:47 +0200633 rq->cmd_flags |= REQ_ORDERED_COLOR;
Tejun Heo30e96562006-02-08 01:01:31 -0800634
Jens Axboe4aff5e22006-08-10 08:44:47 +0200635 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800636 /*
637 * toggle ordered color
638 */
639 if (blk_barrier_rq(rq))
640 q->ordcolor ^= 1;
641
642 /*
643 * barriers implicitly indicate back insertion
644 */
645 if (where == ELEVATOR_INSERT_SORT)
646 where = ELEVATOR_INSERT_BACK;
647
648 /*
649 * this request is scheduling boundary, update
650 * end_sector
651 */
652 if (blk_fs_request(rq)) {
653 q->end_sector = rq_end_sector(rq);
654 q->boundary_rq = rq;
655 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200656 } else if (!(rq->cmd_flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
Tejun Heo30e96562006-02-08 01:01:31 -0800657 where = ELEVATOR_INSERT_BACK;
658
659 if (plug)
660 blk_plug_device(q);
661
662 elv_insert(q, rq, where);
663}
664
Jens Axboe2e662b62006-07-13 11:55:04 +0200665EXPORT_SYMBOL(__elv_add_request);
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667void elv_add_request(request_queue_t *q, struct request *rq, int where,
668 int plug)
669{
670 unsigned long flags;
671
672 spin_lock_irqsave(q->queue_lock, flags);
673 __elv_add_request(q, rq, where, plug);
674 spin_unlock_irqrestore(q->queue_lock, flags);
675}
676
Jens Axboe2e662b62006-07-13 11:55:04 +0200677EXPORT_SYMBOL(elv_add_request);
678
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679static inline struct request *__elv_next_request(request_queue_t *q)
680{
Tejun Heo8922e162005-10-20 16:23:44 +0200681 struct request *rq;
682
Tejun Heo797e7db2006-01-06 09:51:03 +0100683 while (1) {
684 while (!list_empty(&q->queue_head)) {
685 rq = list_entry_rq(q->queue_head.next);
686 if (blk_do_ordered(q, &rq))
687 return rq;
688 }
Tejun Heo8922e162005-10-20 16:23:44 +0200689
Tejun Heo797e7db2006-01-06 09:51:03 +0100690 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
691 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
695struct request *elv_next_request(request_queue_t *q)
696{
697 struct request *rq;
698 int ret;
699
700 while ((rq = __elv_next_request(q)) != NULL) {
Jens Axboe4aff5e22006-08-10 08:44:47 +0200701 if (!(rq->cmd_flags & REQ_STARTED)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200702 elevator_t *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Tejun Heo8922e162005-10-20 16:23:44 +0200704 /*
705 * This is the first time the device driver
706 * sees this request (possibly after
707 * requeueing). Notify IO scheduler.
708 */
709 if (blk_sorted_rq(rq) &&
710 e->ops->elevator_activate_req_fn)
711 e->ops->elevator_activate_req_fn(q, rq);
712
713 /*
714 * just mark as started even if we don't start
715 * it, a request that has been delayed should
716 * not be passed by new incoming requests
717 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200718 rq->cmd_flags |= REQ_STARTED;
Jens Axboe2056a782006-03-23 20:00:26 +0100719 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
Tejun Heo8922e162005-10-20 16:23:44 +0200720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Tejun Heo8922e162005-10-20 16:23:44 +0200722 if (!q->boundary_rq || q->boundary_rq == rq) {
Jens Axboe1b47f532005-10-20 16:37:00 +0200723 q->end_sector = rq_end_sector(rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200724 q->boundary_rq = NULL;
725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Jens Axboe4aff5e22006-08-10 08:44:47 +0200727 if ((rq->cmd_flags & REQ_DONTPREP) || !q->prep_rq_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 break;
729
730 ret = q->prep_rq_fn(q, rq);
731 if (ret == BLKPREP_OK) {
732 break;
733 } else if (ret == BLKPREP_DEFER) {
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500734 /*
735 * the request may have been (partially) prepped.
736 * we need to keep this request in the front to
Tejun Heo8922e162005-10-20 16:23:44 +0200737 * avoid resource deadlock. REQ_STARTED will
738 * prevent other fs requests from passing this one.
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500739 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 rq = NULL;
741 break;
742 } else if (ret == BLKPREP_KILL) {
743 int nr_bytes = rq->hard_nr_sectors << 9;
744
745 if (!nr_bytes)
746 nr_bytes = rq->data_len;
747
748 blkdev_dequeue_request(rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200749 rq->cmd_flags |= REQ_QUIET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 end_that_request_chunk(rq, 0, nr_bytes);
Tejun Heo8ffdc652006-01-06 09:49:03 +0100751 end_that_request_last(rq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 } else {
753 printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
754 ret);
755 break;
756 }
757 }
758
759 return rq;
760}
761
Jens Axboe2e662b62006-07-13 11:55:04 +0200762EXPORT_SYMBOL(elv_next_request);
763
Tejun Heo8922e162005-10-20 16:23:44 +0200764void elv_dequeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
Tejun Heo8922e162005-10-20 16:23:44 +0200766 BUG_ON(list_empty(&rq->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +0200767 BUG_ON(ELV_ON_HASH(rq));
Tejun Heo8922e162005-10-20 16:23:44 +0200768
769 list_del_init(&rq->queuelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 /*
772 * the time frame between a request being removed from the lists
773 * and to it is freed is accounted as io that is in progress at
Tejun Heo8922e162005-10-20 16:23:44 +0200774 * the driver side.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 */
776 if (blk_account_rq(rq))
777 q->in_flight++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Jens Axboe2e662b62006-07-13 11:55:04 +0200780EXPORT_SYMBOL(elv_dequeue_request);
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782int elv_queue_empty(request_queue_t *q)
783{
784 elevator_t *e = q->elevator;
785
Tejun Heo8922e162005-10-20 16:23:44 +0200786 if (!list_empty(&q->queue_head))
787 return 0;
788
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if (e->ops->elevator_queue_empty_fn)
790 return e->ops->elevator_queue_empty_fn(q);
791
Tejun Heo8922e162005-10-20 16:23:44 +0200792 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
Jens Axboe2e662b62006-07-13 11:55:04 +0200795EXPORT_SYMBOL(elv_queue_empty);
796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797struct request *elv_latter_request(request_queue_t *q, struct request *rq)
798{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 elevator_t *e = q->elevator;
800
801 if (e->ops->elevator_latter_req_fn)
802 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return NULL;
804}
805
806struct request *elv_former_request(request_queue_t *q, struct request *rq)
807{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 elevator_t *e = q->elevator;
809
810 if (e->ops->elevator_former_req_fn)
811 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return NULL;
813}
814
Jens Axboecb78b282006-07-28 09:32:57 +0200815int elv_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
817 elevator_t *e = q->elevator;
818
819 if (e->ops->elevator_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200820 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 rq->elevator_private = NULL;
823 return 0;
824}
825
826void elv_put_request(request_queue_t *q, struct request *rq)
827{
828 elevator_t *e = q->elevator;
829
830 if (e->ops->elevator_put_req_fn)
Jens Axboebb37b942006-12-01 10:42:33 +0100831 e->ops->elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832}
833
Jens Axboecb78b282006-07-28 09:32:57 +0200834int elv_may_queue(request_queue_t *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
836 elevator_t *e = q->elevator;
837
838 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200839 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 return ELV_MQUEUE_MAY;
842}
843
844void elv_completed_request(request_queue_t *q, struct request *rq)
845{
846 elevator_t *e = q->elevator;
847
848 /*
849 * request is released from the driver, io must be done
850 */
Tejun Heo8922e162005-10-20 16:23:44 +0200851 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 q->in_flight--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100853 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
854 e->ops->elevator_completed_req_fn(q, rq);
855 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100856
Tejun Heo1bc691d2006-01-12 15:39:26 +0100857 /*
858 * Check if the queue is waiting for fs requests to be
859 * drained for flush sequence.
860 */
861 if (unlikely(q->ordseq)) {
862 struct request *first_rq = list_entry_rq(q->queue_head.next);
863 if (q->in_flight == 0 &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100864 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
865 blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
866 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
867 q->request_fn(q);
868 }
Tejun Heo8922e162005-10-20 16:23:44 +0200869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870}
871
Al Viro3d1ab402006-03-18 18:35:43 -0500872#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
873
874static ssize_t
875elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
876{
877 elevator_t *e = container_of(kobj, elevator_t, kobj);
878 struct elv_fs_entry *entry = to_elv(attr);
879 ssize_t error;
880
881 if (!entry->show)
882 return -EIO;
883
884 mutex_lock(&e->sysfs_lock);
885 error = e->ops ? entry->show(e, page) : -ENOENT;
886 mutex_unlock(&e->sysfs_lock);
887 return error;
888}
889
890static ssize_t
891elv_attr_store(struct kobject *kobj, struct attribute *attr,
892 const char *page, size_t length)
893{
894 elevator_t *e = container_of(kobj, elevator_t, kobj);
895 struct elv_fs_entry *entry = to_elv(attr);
896 ssize_t error;
897
898 if (!entry->store)
899 return -EIO;
900
901 mutex_lock(&e->sysfs_lock);
902 error = e->ops ? entry->store(e, page, length) : -ENOENT;
903 mutex_unlock(&e->sysfs_lock);
904 return error;
905}
906
907static struct sysfs_ops elv_sysfs_ops = {
908 .show = elv_attr_show,
909 .store = elv_attr_store,
910};
911
912static struct kobj_type elv_ktype = {
913 .sysfs_ops = &elv_sysfs_ops,
914 .release = elevator_release,
915};
916
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917int elv_register_queue(struct request_queue *q)
918{
919 elevator_t *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500920 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Al Viro3d1ab402006-03-18 18:35:43 -0500922 e->kobj.parent = &q->kobj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Al Viro3d1ab402006-03-18 18:35:43 -0500924 error = kobject_add(&e->kobj);
925 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500926 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500927 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500928 while (attr->attr.name) {
929 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500930 break;
Al Viroe572ec72006-03-18 22:27:18 -0500931 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500932 }
933 }
934 kobject_uevent(&e->kobj, KOBJ_ADD);
935 }
936 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
938
Jens Axboebc1c1162006-06-08 08:49:06 +0200939static void __elv_unregister_queue(elevator_t *e)
940{
941 kobject_uevent(&e->kobj, KOBJ_REMOVE);
942 kobject_del(&e->kobj);
943}
944
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945void elv_unregister_queue(struct request_queue *q)
946{
Jens Axboebc1c1162006-06-08 08:49:06 +0200947 if (q)
948 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
951int elv_register(struct elevator_type *e)
952{
Tejun Heo2824bc932005-10-20 10:56:41 +0200953 spin_lock_irq(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100954 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 list_add_tail(&e->list, &elv_list);
956 spin_unlock_irq(&elv_list_lock);
957
958 printk(KERN_INFO "io scheduler %s registered", e->elevator_name);
Nate Diller5f003972006-01-24 10:07:58 +0100959 if (!strcmp(e->elevator_name, chosen_elevator) ||
960 (!*chosen_elevator &&
961 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
962 printk(" (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 printk("\n");
964 return 0;
965}
966EXPORT_SYMBOL_GPL(elv_register);
967
968void elv_unregister(struct elevator_type *e)
969{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800970 struct task_struct *g, *p;
971
972 /*
973 * Iterate every thread in the process to remove the io contexts.
974 */
Al Viroe17a9482006-03-18 13:21:20 -0500975 if (e->ops.trim) {
976 read_lock(&tasklist_lock);
977 do_each_thread(g, p) {
978 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400979 if (p->io_context)
980 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500981 task_unlock(p);
982 } while_each_thread(g, p);
983 read_unlock(&tasklist_lock);
984 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 spin_lock_irq(&elv_list_lock);
987 list_del_init(&e->list);
988 spin_unlock_irq(&elv_list_lock);
989}
990EXPORT_SYMBOL_GPL(elv_unregister);
991
992/*
993 * switch to new_e io scheduler. be careful not to introduce deadlocks -
994 * we don't free the old io scheduler, before we have allocated what we
995 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200996 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Al Viro3d1ab402006-03-18 18:35:43 -0500998static int elevator_switch(request_queue_t *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Tejun Heocb98fc82005-10-28 08:29:39 +02001000 elevator_t *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +02001001 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Tejun Heocb98fc82005-10-28 08:29:39 +02001003 /*
1004 * Allocate new elevator
1005 */
Jens Axboeb5deef92006-07-19 23:39:40 +02001006 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -05001008 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Jens Axboebc1c1162006-06-08 08:49:06 +02001010 data = elevator_init_queue(q, e);
1011 if (!data) {
1012 kobject_put(&e->kobj);
1013 return 0;
1014 }
1015
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001017 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001019 spin_lock_irq(q->queue_lock);
1020
Jens Axboe64521d12005-10-28 08:30:39 +02001021 set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02001022
Tejun Heo15853af2005-11-10 08:52:05 +01001023 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001024
1025 while (q->rq.elvpriv) {
Tejun Heo407df2a2005-11-10 08:48:21 +01001026 blk_remove_plug(q);
1027 q->request_fn(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001028 spin_unlock_irq(q->queue_lock);
Jens Axboe64521d12005-10-28 08:30:39 +02001029 msleep(10);
Tejun Heocb98fc82005-10-28 08:29:39 +02001030 spin_lock_irq(q->queue_lock);
Tejun Heo15853af2005-11-10 08:52:05 +01001031 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001032 }
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /*
Jens Axboebc1c1162006-06-08 08:49:06 +02001035 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 old_elevator = q->elevator;
1038
1039 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 * attach and start new elevator
1041 */
Jens Axboebc1c1162006-06-08 08:49:06 +02001042 elevator_attach(q, e, data);
1043
1044 spin_unlock_irq(q->queue_lock);
1045
1046 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
1048 if (elv_register_queue(q))
1049 goto fail_register;
1050
1051 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001052 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 */
1054 elevator_exit(old_elevator);
Jens Axboe64521d12005-10-28 08:30:39 +02001055 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Al Viro3d1ab402006-03-18 18:35:43 -05001056 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058fail_register:
1059 /*
1060 * switch failed, exit the new io scheduler and reattach the old
1061 * one again (along with re-adding the sysfs dir)
1062 */
1063 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 q->elevator = old_elevator;
1065 elv_register_queue(q);
Jens Axboe64521d12005-10-28 08:30:39 +02001066 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Al Viro3d1ab402006-03-18 18:35:43 -05001067 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
1070ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count)
1071{
1072 char elevator_name[ELV_NAME_MAX];
Tejun Heobe561232005-11-10 08:55:01 +01001073 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 struct elevator_type *e;
1075
Tejun Heobe561232005-11-10 08:55:01 +01001076 elevator_name[sizeof(elevator_name) - 1] = '\0';
1077 strncpy(elevator_name, name, sizeof(elevator_name) - 1);
1078 len = strlen(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Tejun Heobe561232005-11-10 08:55:01 +01001080 if (len && elevator_name[len - 1] == '\n')
1081 elevator_name[len - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
1083 e = elevator_get(elevator_name);
1084 if (!e) {
1085 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1086 return -EINVAL;
1087 }
1088
Nate Diller2ca7d932005-10-30 15:02:24 -08001089 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1090 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -08001092 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Al Viro3d1ab402006-03-18 18:35:43 -05001094 if (!elevator_switch(q, e))
1095 printk(KERN_ERR "elevator: switch to %s failed\n",elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return count;
1097}
1098
1099ssize_t elv_iosched_show(request_queue_t *q, char *name)
1100{
1101 elevator_t *e = q->elevator;
1102 struct elevator_type *elv = e->elevator_type;
1103 struct list_head *entry;
1104 int len = 0;
1105
Vasily Tarasovc5841642006-10-11 13:26:30 +02001106 spin_lock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 list_for_each(entry, &elv_list) {
1108 struct elevator_type *__e;
1109
1110 __e = list_entry(entry, struct elevator_type, list);
1111 if (!strcmp(elv->elevator_name, __e->elevator_name))
1112 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1113 else
1114 len += sprintf(name+len, "%s ", __e->elevator_name);
1115 }
Vasily Tarasovc5841642006-10-11 13:26:30 +02001116 spin_unlock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
1118 len += sprintf(len+name, "\n");
1119 return len;
1120}
1121
Jens Axboe2e662b62006-07-13 11:55:04 +02001122struct request *elv_rb_former_request(request_queue_t *q, struct request *rq)
1123{
1124 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1125
1126 if (rbprev)
1127 return rb_entry_rq(rbprev);
1128
1129 return NULL;
1130}
1131
1132EXPORT_SYMBOL(elv_rb_former_request);
1133
1134struct request *elv_rb_latter_request(request_queue_t *q, struct request *rq)
1135{
1136 struct rb_node *rbnext = rb_next(&rq->rb_node);
1137
1138 if (rbnext)
1139 return rb_entry_rq(rbnext);
1140
1141 return NULL;
1142}
1143
1144EXPORT_SYMBOL(elv_rb_latter_request);