blob: 487dd3da8853971d9bcb77cf6ecf51eb39faedf3 [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/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * can we safely merge with this request?
55 */
56inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
57{
58 if (!rq_mergeable(rq))
59 return 0;
60
61 /*
62 * different data direction or already started, don't merge
63 */
64 if (bio_data_dir(bio) != rq_data_dir(rq))
65 return 0;
66
67 /*
68 * same device and no special stuff set, merge is ok
69 */
Jens Axboec00895a2006-09-30 20:29:12 +020070 if (rq->rq_disk == bio->bi_bdev->bd_disk && !rq->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 return 1;
72
73 return 0;
74}
75EXPORT_SYMBOL(elv_rq_merge_ok);
76
Coywolf Qi Hunt769db452005-12-28 10:55:49 +010077static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078{
79 int ret = ELEVATOR_NO_MERGE;
80
81 /*
82 * we can merge and sequence is ok, check if it's possible
83 */
84 if (elv_rq_merge_ok(__rq, bio)) {
85 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
86 ret = ELEVATOR_BACK_MERGE;
87 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
88 ret = ELEVATOR_FRONT_MERGE;
89 }
90
91 return ret;
92}
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Linus Torvalds1da177e2005-04-16 15:20:36 -070094static struct elevator_type *elevator_find(const char *name)
95{
96 struct elevator_type *e = NULL;
97 struct list_head *entry;
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 list_for_each(entry, &elv_list) {
100 struct elevator_type *__e;
101
102 __e = list_entry(entry, struct elevator_type, list);
103
104 if (!strcmp(__e->elevator_name, name)) {
105 e = __e;
106 break;
107 }
108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110 return e;
111}
112
113static void elevator_put(struct elevator_type *e)
114{
115 module_put(e->elevator_owner);
116}
117
118static struct elevator_type *elevator_get(const char *name)
119{
Tejun Heo2824bc932005-10-20 10:56:41 +0200120 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Tejun Heo2824bc932005-10-20 10:56:41 +0200122 spin_lock_irq(&elv_list_lock);
123
124 e = elevator_find(name);
125 if (e && !try_module_get(e->elevator_owner))
126 e = NULL;
127
128 spin_unlock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130 return e;
131}
132
Jens Axboebc1c1162006-06-08 08:49:06 +0200133static void *elevator_init_queue(request_queue_t *q, struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134{
Jens Axboebc1c1162006-06-08 08:49:06 +0200135 return eq->ops->elevator_init_fn(q, eq);
136}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
Jens Axboebc1c1162006-06-08 08:49:06 +0200138static void elevator_attach(request_queue_t *q, struct elevator_queue *eq,
139 void *data)
140{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200142 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
145static char chosen_elevator[16];
146
Nate Diller5f003972006-01-24 10:07:58 +0100147static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100149 /*
150 * Be backwards-compatible with previous kernels, so users
151 * won't get the wrong elevator.
152 */
Nate Diller5f003972006-01-24 10:07:58 +0100153 if (!strcmp(str, "as"))
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100154 strcpy(chosen_elevator, "anticipatory");
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100155 else
Nate Diller5f003972006-01-24 10:07:58 +0100156 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800157 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
160__setup("elevator=", elevator_setup);
161
Al Viro3d1ab402006-03-18 18:35:43 -0500162static struct kobj_type elv_ktype;
163
Jens Axboeb5deef92006-07-19 23:39:40 +0200164static elevator_t *elevator_alloc(request_queue_t *q, struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500165{
Jens Axboe98170642006-07-28 09:23:08 +0200166 elevator_t *eq;
167 int i;
168
Jens Axboeb5deef92006-07-19 23:39:40 +0200169 eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200170 if (unlikely(!eq))
171 goto err;
172
173 memset(eq, 0, sizeof(*eq));
174 eq->ops = &e->ops;
175 eq->elevator_type = e;
176 kobject_init(&eq->kobj);
177 snprintf(eq->kobj.name, KOBJ_NAME_LEN, "%s", "iosched");
178 eq->kobj.ktype = &elv_ktype;
179 mutex_init(&eq->sysfs_lock);
180
Jens Axboeb5deef92006-07-19 23:39:40 +0200181 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
182 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200183 if (!eq->hash)
184 goto err;
185
186 for (i = 0; i < ELV_HASH_ENTRIES; i++)
187 INIT_HLIST_HEAD(&eq->hash[i]);
188
Al Viro3d1ab402006-03-18 18:35:43 -0500189 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200190err:
191 kfree(eq);
192 elevator_put(e);
193 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500194}
195
196static void elevator_release(struct kobject *kobj)
197{
198 elevator_t *e = container_of(kobj, elevator_t, kobj);
Jens Axboe98170642006-07-28 09:23:08 +0200199
Al Viro3d1ab402006-03-18 18:35:43 -0500200 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200201 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500202 kfree(e);
203}
204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205int elevator_init(request_queue_t *q, char *name)
206{
207 struct elevator_type *e = NULL;
208 struct elevator_queue *eq;
209 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200210 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Tejun Heocb98fc82005-10-28 08:29:39 +0200212 INIT_LIST_HEAD(&q->queue_head);
213 q->last_merge = NULL;
214 q->end_sector = 0;
215 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200216
Nate Diller5f003972006-01-24 10:07:58 +0100217 if (name && !(e = elevator_get(name)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return -EINVAL;
219
Nate Diller248d5ca2006-01-24 10:09:14 +0100220 if (!e && *chosen_elevator && !(e = elevator_get(chosen_elevator)))
221 printk("I/O scheduler %s not found\n", chosen_elevator);
222
223 if (!e && !(e = elevator_get(CONFIG_DEFAULT_IOSCHED))) {
224 printk("Default I/O scheduler not found, using no-op\n");
225 e = elevator_get("noop");
Nate Diller5f003972006-01-24 10:07:58 +0100226 }
227
Jens Axboeb5deef92006-07-19 23:39:40 +0200228 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500229 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Jens Axboebc1c1162006-06-08 08:49:06 +0200232 data = elevator_init_queue(q, eq);
233 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500234 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200235 return -ENOMEM;
236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Jens Axboebc1c1162006-06-08 08:49:06 +0200238 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return ret;
240}
241
Jens Axboe2e662b62006-07-13 11:55:04 +0200242EXPORT_SYMBOL(elevator_init);
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244void elevator_exit(elevator_t *e)
245{
Al Viro3d1ab402006-03-18 18:35:43 -0500246 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 if (e->ops->elevator_exit_fn)
248 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500249 e->ops = NULL;
250 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Al Viro3d1ab402006-03-18 18:35:43 -0500252 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
Jens Axboe2e662b62006-07-13 11:55:04 +0200255EXPORT_SYMBOL(elevator_exit);
256
Jens Axboe98170642006-07-28 09:23:08 +0200257static inline void __elv_rqhash_del(struct request *rq)
258{
259 hlist_del_init(&rq->hash);
260}
261
262static void elv_rqhash_del(request_queue_t *q, struct request *rq)
263{
264 if (ELV_ON_HASH(rq))
265 __elv_rqhash_del(rq);
266}
267
268static void elv_rqhash_add(request_queue_t *q, struct request *rq)
269{
270 elevator_t *e = q->elevator;
271
272 BUG_ON(ELV_ON_HASH(rq));
273 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
274}
275
276static void elv_rqhash_reposition(request_queue_t *q, struct request *rq)
277{
278 __elv_rqhash_del(rq);
279 elv_rqhash_add(q, rq);
280}
281
282static struct request *elv_rqhash_find(request_queue_t *q, sector_t offset)
283{
284 elevator_t *e = q->elevator;
285 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
286 struct hlist_node *entry, *next;
287 struct request *rq;
288
289 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
290 BUG_ON(!ELV_ON_HASH(rq));
291
292 if (unlikely(!rq_mergeable(rq))) {
293 __elv_rqhash_del(rq);
294 continue;
295 }
296
297 if (rq_hash_key(rq) == offset)
298 return rq;
299 }
300
301 return NULL;
302}
303
Tejun Heo8922e162005-10-20 16:23:44 +0200304/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200305 * RB-tree support functions for inserting/lookup/removal of requests
306 * in a sorted RB tree.
307 */
308struct request *elv_rb_add(struct rb_root *root, struct request *rq)
309{
310 struct rb_node **p = &root->rb_node;
311 struct rb_node *parent = NULL;
312 struct request *__rq;
313
314 while (*p) {
315 parent = *p;
316 __rq = rb_entry(parent, struct request, rb_node);
317
318 if (rq->sector < __rq->sector)
319 p = &(*p)->rb_left;
320 else if (rq->sector > __rq->sector)
321 p = &(*p)->rb_right;
322 else
323 return __rq;
324 }
325
326 rb_link_node(&rq->rb_node, parent, p);
327 rb_insert_color(&rq->rb_node, root);
328 return NULL;
329}
330
331EXPORT_SYMBOL(elv_rb_add);
332
333void elv_rb_del(struct rb_root *root, struct request *rq)
334{
335 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
336 rb_erase(&rq->rb_node, root);
337 RB_CLEAR_NODE(&rq->rb_node);
338}
339
340EXPORT_SYMBOL(elv_rb_del);
341
342struct request *elv_rb_find(struct rb_root *root, sector_t sector)
343{
344 struct rb_node *n = root->rb_node;
345 struct request *rq;
346
347 while (n) {
348 rq = rb_entry(n, struct request, rb_node);
349
350 if (sector < rq->sector)
351 n = n->rb_left;
352 else if (sector > rq->sector)
353 n = n->rb_right;
354 else
355 return rq;
356 }
357
358 return NULL;
359}
360
361EXPORT_SYMBOL(elv_rb_find);
362
363/*
Tejun Heo8922e162005-10-20 16:23:44 +0200364 * Insert rq into dispatch queue of q. Queue lock must be held on
Jens Axboe2e662b62006-07-13 11:55:04 +0200365 * entry. rq is sort insted into the dispatch queue. To be used by
366 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200367 */
Jens Axboe1b47f532005-10-20 16:37:00 +0200368void elv_dispatch_sort(request_queue_t *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200369{
370 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200371 struct list_head *entry;
372
Tejun Heo06b86242005-10-20 16:46:23 +0200373 if (q->last_merge == rq)
374 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200375
376 elv_rqhash_del(q, rq);
377
Tejun Heo15853af2005-11-10 08:52:05 +0100378 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200379
Jens Axboe1b47f532005-10-20 16:37:00 +0200380 boundary = q->end_sector;
Tejun Heocb198332005-10-24 08:35:58 +0200381
Tejun Heo8922e162005-10-20 16:23:44 +0200382 list_for_each_prev(entry, &q->queue_head) {
383 struct request *pos = list_entry_rq(entry);
384
Jens Axboe4aff5e22006-08-10 08:44:47 +0200385 if (pos->cmd_flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED))
Tejun Heo8922e162005-10-20 16:23:44 +0200386 break;
387 if (rq->sector >= boundary) {
388 if (pos->sector < boundary)
389 continue;
390 } else {
391 if (pos->sector >= boundary)
392 break;
393 }
394 if (rq->sector >= pos->sector)
395 break;
396 }
397
398 list_add(&rq->queuelist, entry);
399}
400
Jens Axboe2e662b62006-07-13 11:55:04 +0200401EXPORT_SYMBOL(elv_dispatch_sort);
402
Jens Axboe98170642006-07-28 09:23:08 +0200403/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200404 * Insert rq into dispatch queue of q. Queue lock must be held on
405 * entry. rq is added to the back of the dispatch queue. To be used by
406 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200407 */
408void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
409{
410 if (q->last_merge == rq)
411 q->last_merge = NULL;
412
413 elv_rqhash_del(q, rq);
414
415 q->nr_sorted--;
416
417 q->end_sector = rq_end_sector(rq);
418 q->boundary_rq = rq;
419 list_add_tail(&rq->queuelist, &q->queue_head);
420}
421
Jens Axboe2e662b62006-07-13 11:55:04 +0200422EXPORT_SYMBOL(elv_dispatch_add_tail);
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424int elv_merge(request_queue_t *q, struct request **req, struct bio *bio)
425{
426 elevator_t *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200427 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200428 int ret;
429
Jens Axboe98170642006-07-28 09:23:08 +0200430 /*
431 * First try one-hit cache.
432 */
Tejun Heo06b86242005-10-20 16:46:23 +0200433 if (q->last_merge) {
434 ret = elv_try_merge(q->last_merge, bio);
435 if (ret != ELEVATOR_NO_MERGE) {
436 *req = q->last_merge;
437 return ret;
438 }
439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Jens Axboe98170642006-07-28 09:23:08 +0200441 /*
442 * See if our hash lookup can find a potential backmerge.
443 */
444 __rq = elv_rqhash_find(q, bio->bi_sector);
445 if (__rq && elv_rq_merge_ok(__rq, bio)) {
446 *req = __rq;
447 return ELEVATOR_BACK_MERGE;
448 }
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 if (e->ops->elevator_merge_fn)
451 return e->ops->elevator_merge_fn(q, req, bio);
452
453 return ELEVATOR_NO_MERGE;
454}
455
Jens Axboe2e662b62006-07-13 11:55:04 +0200456void elv_merged_request(request_queue_t *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 elevator_t *e = q->elevator;
459
460 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200461 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200462
Jens Axboe2e662b62006-07-13 11:55:04 +0200463 if (type == ELEVATOR_BACK_MERGE)
464 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200465
Tejun Heo06b86242005-10-20 16:46:23 +0200466 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469void elv_merge_requests(request_queue_t *q, struct request *rq,
470 struct request *next)
471{
472 elevator_t *e = q->elevator;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 if (e->ops->elevator_merge_req_fn)
475 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200476
Jens Axboe98170642006-07-28 09:23:08 +0200477 elv_rqhash_reposition(q, rq);
478 elv_rqhash_del(q, next);
479
480 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200481 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Tejun Heo8922e162005-10-20 16:23:44 +0200484void elv_requeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
486 elevator_t *e = q->elevator;
487
488 /*
489 * it already went through dequeue, we need to decrement the
490 * in_flight count again
491 */
Tejun Heo8922e162005-10-20 16:23:44 +0200492 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 q->in_flight--;
Tejun Heo8922e162005-10-20 16:23:44 +0200494 if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn)
495 e->ops->elevator_deactivate_req_fn(q, rq);
496 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Jens Axboe4aff5e22006-08-10 08:44:47 +0200498 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Tejun Heo30e96562006-02-08 01:01:31 -0800500 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501}
502
Tejun Heo15853af2005-11-10 08:52:05 +0100503static void elv_drain_elevator(request_queue_t *q)
504{
505 static int printed;
506 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
507 ;
508 if (q->nr_sorted == 0)
509 return;
510 if (printed++ < 10) {
511 printk(KERN_ERR "%s: forced dispatching is broken "
512 "(nr_sorted=%u), please report this\n",
513 q->elevator->elevator_type->elevator_name, q->nr_sorted);
514 }
515}
516
Tejun Heo30e96562006-02-08 01:01:31 -0800517void elv_insert(request_queue_t *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
Tejun Heo797e7db2006-01-06 09:51:03 +0100519 struct list_head *pos;
520 unsigned ordseq;
Jens Axboedac07ec2006-05-11 08:20:16 +0200521 int unplug_it = 1;
Tejun Heo797e7db2006-01-06 09:51:03 +0100522
Jens Axboe2056a782006-03-23 20:00:26 +0100523 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 rq->q = q;
526
Tejun Heo8922e162005-10-20 16:23:44 +0200527 switch (where) {
528 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200529 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200530
531 list_add(&rq->queuelist, &q->queue_head);
532 break;
533
534 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200535 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100536 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200537 list_add_tail(&rq->queuelist, &q->queue_head);
538 /*
539 * We kick the queue here for the following reasons.
540 * - The elevator might have returned NULL previously
541 * to delay requests and returned them now. As the
542 * queue wasn't empty before this request, ll_rw_blk
543 * won't run the queue on return, resulting in hang.
544 * - Usually, back inserted requests won't be merged
545 * with anything. There's no point in delaying queue
546 * processing.
547 */
548 blk_remove_plug(q);
549 q->request_fn(q);
550 break;
551
552 case ELEVATOR_INSERT_SORT:
553 BUG_ON(!blk_fs_request(rq));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200554 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100555 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200556 if (rq_mergeable(rq)) {
557 elv_rqhash_add(q, rq);
558 if (!q->last_merge)
559 q->last_merge = rq;
560 }
561
Tejun Heoca235092005-11-01 17:23:49 +0900562 /*
563 * Some ioscheds (cfq) run q->request_fn directly, so
564 * rq cannot be accessed after calling
565 * elevator_add_req_fn.
566 */
567 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200568 break;
569
Tejun Heo797e7db2006-01-06 09:51:03 +0100570 case ELEVATOR_INSERT_REQUEUE:
571 /*
572 * If ordered flush isn't in progress, we do front
573 * insertion; otherwise, requests should be requeued
574 * in ordseq order.
575 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200576 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100577
578 if (q->ordseq == 0) {
579 list_add(&rq->queuelist, &q->queue_head);
580 break;
581 }
582
583 ordseq = blk_ordered_req_seq(rq);
584
585 list_for_each(pos, &q->queue_head) {
586 struct request *pos_rq = list_entry_rq(pos);
587 if (ordseq <= blk_ordered_req_seq(pos_rq))
588 break;
589 }
590
591 list_add_tail(&rq->queuelist, pos);
Jens Axboedac07ec2006-05-11 08:20:16 +0200592 /*
593 * most requeues happen because of a busy condition, don't
594 * force unplug of the queue for that case.
595 */
596 unplug_it = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100597 break;
598
Tejun Heo8922e162005-10-20 16:23:44 +0200599 default:
600 printk(KERN_ERR "%s: bad insertion point %d\n",
601 __FUNCTION__, where);
602 BUG();
603 }
604
Jens Axboedac07ec2006-05-11 08:20:16 +0200605 if (unplug_it && blk_queue_plugged(q)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200606 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
607 - q->in_flight;
608
609 if (nrq >= q->unplug_thresh)
610 __generic_unplug_device(q);
611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
Tejun Heo30e96562006-02-08 01:01:31 -0800614void __elv_add_request(request_queue_t *q, struct request *rq, int where,
615 int plug)
616{
617 if (q->ordcolor)
Jens Axboe4aff5e22006-08-10 08:44:47 +0200618 rq->cmd_flags |= REQ_ORDERED_COLOR;
Tejun Heo30e96562006-02-08 01:01:31 -0800619
Jens Axboe4aff5e22006-08-10 08:44:47 +0200620 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800621 /*
622 * toggle ordered color
623 */
624 if (blk_barrier_rq(rq))
625 q->ordcolor ^= 1;
626
627 /*
628 * barriers implicitly indicate back insertion
629 */
630 if (where == ELEVATOR_INSERT_SORT)
631 where = ELEVATOR_INSERT_BACK;
632
633 /*
634 * this request is scheduling boundary, update
635 * end_sector
636 */
637 if (blk_fs_request(rq)) {
638 q->end_sector = rq_end_sector(rq);
639 q->boundary_rq = rq;
640 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200641 } else if (!(rq->cmd_flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
Tejun Heo30e96562006-02-08 01:01:31 -0800642 where = ELEVATOR_INSERT_BACK;
643
644 if (plug)
645 blk_plug_device(q);
646
647 elv_insert(q, rq, where);
648}
649
Jens Axboe2e662b62006-07-13 11:55:04 +0200650EXPORT_SYMBOL(__elv_add_request);
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652void elv_add_request(request_queue_t *q, struct request *rq, int where,
653 int plug)
654{
655 unsigned long flags;
656
657 spin_lock_irqsave(q->queue_lock, flags);
658 __elv_add_request(q, rq, where, plug);
659 spin_unlock_irqrestore(q->queue_lock, flags);
660}
661
Jens Axboe2e662b62006-07-13 11:55:04 +0200662EXPORT_SYMBOL(elv_add_request);
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664static inline struct request *__elv_next_request(request_queue_t *q)
665{
Tejun Heo8922e162005-10-20 16:23:44 +0200666 struct request *rq;
667
Tejun Heo797e7db2006-01-06 09:51:03 +0100668 while (1) {
669 while (!list_empty(&q->queue_head)) {
670 rq = list_entry_rq(q->queue_head.next);
671 if (blk_do_ordered(q, &rq))
672 return rq;
673 }
Tejun Heo8922e162005-10-20 16:23:44 +0200674
Tejun Heo797e7db2006-01-06 09:51:03 +0100675 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
676 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
679
680struct request *elv_next_request(request_queue_t *q)
681{
682 struct request *rq;
683 int ret;
684
685 while ((rq = __elv_next_request(q)) != NULL) {
Jens Axboe4aff5e22006-08-10 08:44:47 +0200686 if (!(rq->cmd_flags & REQ_STARTED)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200687 elevator_t *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Tejun Heo8922e162005-10-20 16:23:44 +0200689 /*
690 * This is the first time the device driver
691 * sees this request (possibly after
692 * requeueing). Notify IO scheduler.
693 */
694 if (blk_sorted_rq(rq) &&
695 e->ops->elevator_activate_req_fn)
696 e->ops->elevator_activate_req_fn(q, rq);
697
698 /*
699 * just mark as started even if we don't start
700 * it, a request that has been delayed should
701 * not be passed by new incoming requests
702 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200703 rq->cmd_flags |= REQ_STARTED;
Jens Axboe2056a782006-03-23 20:00:26 +0100704 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
Tejun Heo8922e162005-10-20 16:23:44 +0200705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Tejun Heo8922e162005-10-20 16:23:44 +0200707 if (!q->boundary_rq || q->boundary_rq == rq) {
Jens Axboe1b47f532005-10-20 16:37:00 +0200708 q->end_sector = rq_end_sector(rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200709 q->boundary_rq = NULL;
710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Jens Axboe4aff5e22006-08-10 08:44:47 +0200712 if ((rq->cmd_flags & REQ_DONTPREP) || !q->prep_rq_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 break;
714
715 ret = q->prep_rq_fn(q, rq);
716 if (ret == BLKPREP_OK) {
717 break;
718 } else if (ret == BLKPREP_DEFER) {
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500719 /*
720 * the request may have been (partially) prepped.
721 * we need to keep this request in the front to
Tejun Heo8922e162005-10-20 16:23:44 +0200722 * avoid resource deadlock. REQ_STARTED will
723 * prevent other fs requests from passing this one.
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500724 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 rq = NULL;
726 break;
727 } else if (ret == BLKPREP_KILL) {
728 int nr_bytes = rq->hard_nr_sectors << 9;
729
730 if (!nr_bytes)
731 nr_bytes = rq->data_len;
732
733 blkdev_dequeue_request(rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200734 rq->cmd_flags |= REQ_QUIET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 end_that_request_chunk(rq, 0, nr_bytes);
Tejun Heo8ffdc652006-01-06 09:49:03 +0100736 end_that_request_last(rq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 } else {
738 printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
739 ret);
740 break;
741 }
742 }
743
744 return rq;
745}
746
Jens Axboe2e662b62006-07-13 11:55:04 +0200747EXPORT_SYMBOL(elv_next_request);
748
Tejun Heo8922e162005-10-20 16:23:44 +0200749void elv_dequeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
Tejun Heo8922e162005-10-20 16:23:44 +0200751 BUG_ON(list_empty(&rq->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +0200752 BUG_ON(ELV_ON_HASH(rq));
Tejun Heo8922e162005-10-20 16:23:44 +0200753
754 list_del_init(&rq->queuelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756 /*
757 * the time frame between a request being removed from the lists
758 * and to it is freed is accounted as io that is in progress at
Tejun Heo8922e162005-10-20 16:23:44 +0200759 * the driver side.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 */
761 if (blk_account_rq(rq))
762 q->in_flight++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
Jens Axboe2e662b62006-07-13 11:55:04 +0200765EXPORT_SYMBOL(elv_dequeue_request);
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767int elv_queue_empty(request_queue_t *q)
768{
769 elevator_t *e = q->elevator;
770
Tejun Heo8922e162005-10-20 16:23:44 +0200771 if (!list_empty(&q->queue_head))
772 return 0;
773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 if (e->ops->elevator_queue_empty_fn)
775 return e->ops->elevator_queue_empty_fn(q);
776
Tejun Heo8922e162005-10-20 16:23:44 +0200777 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Jens Axboe2e662b62006-07-13 11:55:04 +0200780EXPORT_SYMBOL(elv_queue_empty);
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782struct request *elv_latter_request(request_queue_t *q, struct request *rq)
783{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 elevator_t *e = q->elevator;
785
786 if (e->ops->elevator_latter_req_fn)
787 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 return NULL;
789}
790
791struct request *elv_former_request(request_queue_t *q, struct request *rq)
792{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 elevator_t *e = q->elevator;
794
795 if (e->ops->elevator_former_req_fn)
796 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 return NULL;
798}
799
Jens Axboecb78b282006-07-28 09:32:57 +0200800int elv_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
802 elevator_t *e = q->elevator;
803
804 if (e->ops->elevator_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200805 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
807 rq->elevator_private = NULL;
808 return 0;
809}
810
811void elv_put_request(request_queue_t *q, struct request *rq)
812{
813 elevator_t *e = q->elevator;
814
815 if (e->ops->elevator_put_req_fn)
816 e->ops->elevator_put_req_fn(q, rq);
817}
818
Jens Axboecb78b282006-07-28 09:32:57 +0200819int elv_may_queue(request_queue_t *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
821 elevator_t *e = q->elevator;
822
823 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200824 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 return ELV_MQUEUE_MAY;
827}
828
829void elv_completed_request(request_queue_t *q, struct request *rq)
830{
831 elevator_t *e = q->elevator;
832
833 /*
834 * request is released from the driver, io must be done
835 */
Tejun Heo8922e162005-10-20 16:23:44 +0200836 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 q->in_flight--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100838 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
839 e->ops->elevator_completed_req_fn(q, rq);
840 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100841
Tejun Heo1bc691d2006-01-12 15:39:26 +0100842 /*
843 * Check if the queue is waiting for fs requests to be
844 * drained for flush sequence.
845 */
846 if (unlikely(q->ordseq)) {
847 struct request *first_rq = list_entry_rq(q->queue_head.next);
848 if (q->in_flight == 0 &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100849 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
850 blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
851 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
852 q->request_fn(q);
853 }
Tejun Heo8922e162005-10-20 16:23:44 +0200854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855}
856
Al Viro3d1ab402006-03-18 18:35:43 -0500857#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
858
859static ssize_t
860elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
861{
862 elevator_t *e = container_of(kobj, elevator_t, kobj);
863 struct elv_fs_entry *entry = to_elv(attr);
864 ssize_t error;
865
866 if (!entry->show)
867 return -EIO;
868
869 mutex_lock(&e->sysfs_lock);
870 error = e->ops ? entry->show(e, page) : -ENOENT;
871 mutex_unlock(&e->sysfs_lock);
872 return error;
873}
874
875static ssize_t
876elv_attr_store(struct kobject *kobj, struct attribute *attr,
877 const char *page, size_t length)
878{
879 elevator_t *e = container_of(kobj, elevator_t, kobj);
880 struct elv_fs_entry *entry = to_elv(attr);
881 ssize_t error;
882
883 if (!entry->store)
884 return -EIO;
885
886 mutex_lock(&e->sysfs_lock);
887 error = e->ops ? entry->store(e, page, length) : -ENOENT;
888 mutex_unlock(&e->sysfs_lock);
889 return error;
890}
891
892static struct sysfs_ops elv_sysfs_ops = {
893 .show = elv_attr_show,
894 .store = elv_attr_store,
895};
896
897static struct kobj_type elv_ktype = {
898 .sysfs_ops = &elv_sysfs_ops,
899 .release = elevator_release,
900};
901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902int elv_register_queue(struct request_queue *q)
903{
904 elevator_t *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500905 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Al Viro3d1ab402006-03-18 18:35:43 -0500907 e->kobj.parent = &q->kobj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Al Viro3d1ab402006-03-18 18:35:43 -0500909 error = kobject_add(&e->kobj);
910 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500911 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500912 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500913 while (attr->attr.name) {
914 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500915 break;
Al Viroe572ec72006-03-18 22:27:18 -0500916 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500917 }
918 }
919 kobject_uevent(&e->kobj, KOBJ_ADD);
920 }
921 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922}
923
Jens Axboebc1c1162006-06-08 08:49:06 +0200924static void __elv_unregister_queue(elevator_t *e)
925{
926 kobject_uevent(&e->kobj, KOBJ_REMOVE);
927 kobject_del(&e->kobj);
928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930void elv_unregister_queue(struct request_queue *q)
931{
Jens Axboebc1c1162006-06-08 08:49:06 +0200932 if (q)
933 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
936int elv_register(struct elevator_type *e)
937{
Tejun Heo2824bc932005-10-20 10:56:41 +0200938 spin_lock_irq(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100939 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 list_add_tail(&e->list, &elv_list);
941 spin_unlock_irq(&elv_list_lock);
942
943 printk(KERN_INFO "io scheduler %s registered", e->elevator_name);
Nate Diller5f003972006-01-24 10:07:58 +0100944 if (!strcmp(e->elevator_name, chosen_elevator) ||
945 (!*chosen_elevator &&
946 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
947 printk(" (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 printk("\n");
949 return 0;
950}
951EXPORT_SYMBOL_GPL(elv_register);
952
953void elv_unregister(struct elevator_type *e)
954{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800955 struct task_struct *g, *p;
956
957 /*
958 * Iterate every thread in the process to remove the io contexts.
959 */
Al Viroe17a9482006-03-18 13:21:20 -0500960 if (e->ops.trim) {
961 read_lock(&tasklist_lock);
962 do_each_thread(g, p) {
963 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400964 if (p->io_context)
965 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500966 task_unlock(p);
967 } while_each_thread(g, p);
968 read_unlock(&tasklist_lock);
969 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 spin_lock_irq(&elv_list_lock);
972 list_del_init(&e->list);
973 spin_unlock_irq(&elv_list_lock);
974}
975EXPORT_SYMBOL_GPL(elv_unregister);
976
977/*
978 * switch to new_e io scheduler. be careful not to introduce deadlocks -
979 * we don't free the old io scheduler, before we have allocated what we
980 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200981 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 */
Al Viro3d1ab402006-03-18 18:35:43 -0500983static int elevator_switch(request_queue_t *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Tejun Heocb98fc82005-10-28 08:29:39 +0200985 elevator_t *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200986 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Tejun Heocb98fc82005-10-28 08:29:39 +0200988 /*
989 * Allocate new elevator
990 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200991 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -0500993 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Jens Axboebc1c1162006-06-08 08:49:06 +0200995 data = elevator_init_queue(q, e);
996 if (!data) {
997 kobject_put(&e->kobj);
998 return 0;
999 }
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001002 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001004 spin_lock_irq(q->queue_lock);
1005
Jens Axboe64521d12005-10-28 08:30:39 +02001006 set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02001007
Tejun Heo15853af2005-11-10 08:52:05 +01001008 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001009
1010 while (q->rq.elvpriv) {
Tejun Heo407df2a2005-11-10 08:48:21 +01001011 blk_remove_plug(q);
1012 q->request_fn(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001013 spin_unlock_irq(q->queue_lock);
Jens Axboe64521d12005-10-28 08:30:39 +02001014 msleep(10);
Tejun Heocb98fc82005-10-28 08:29:39 +02001015 spin_lock_irq(q->queue_lock);
Tejun Heo15853af2005-11-10 08:52:05 +01001016 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001017 }
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /*
Jens Axboebc1c1162006-06-08 08:49:06 +02001020 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 old_elevator = q->elevator;
1023
1024 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 * attach and start new elevator
1026 */
Jens Axboebc1c1162006-06-08 08:49:06 +02001027 elevator_attach(q, e, data);
1028
1029 spin_unlock_irq(q->queue_lock);
1030
1031 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033 if (elv_register_queue(q))
1034 goto fail_register;
1035
1036 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001037 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 */
1039 elevator_exit(old_elevator);
Jens Axboe64521d12005-10-28 08:30:39 +02001040 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Al Viro3d1ab402006-03-18 18:35:43 -05001041 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043fail_register:
1044 /*
1045 * switch failed, exit the new io scheduler and reattach the old
1046 * one again (along with re-adding the sysfs dir)
1047 */
1048 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 q->elevator = old_elevator;
1050 elv_register_queue(q);
Jens Axboe64521d12005-10-28 08:30:39 +02001051 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Al Viro3d1ab402006-03-18 18:35:43 -05001052 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053}
1054
1055ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count)
1056{
1057 char elevator_name[ELV_NAME_MAX];
Tejun Heobe561232005-11-10 08:55:01 +01001058 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 struct elevator_type *e;
1060
Tejun Heobe561232005-11-10 08:55:01 +01001061 elevator_name[sizeof(elevator_name) - 1] = '\0';
1062 strncpy(elevator_name, name, sizeof(elevator_name) - 1);
1063 len = strlen(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Tejun Heobe561232005-11-10 08:55:01 +01001065 if (len && elevator_name[len - 1] == '\n')
1066 elevator_name[len - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 e = elevator_get(elevator_name);
1069 if (!e) {
1070 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1071 return -EINVAL;
1072 }
1073
Nate Diller2ca7d932005-10-30 15:02:24 -08001074 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1075 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -08001077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Al Viro3d1ab402006-03-18 18:35:43 -05001079 if (!elevator_switch(q, e))
1080 printk(KERN_ERR "elevator: switch to %s failed\n",elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return count;
1082}
1083
1084ssize_t elv_iosched_show(request_queue_t *q, char *name)
1085{
1086 elevator_t *e = q->elevator;
1087 struct elevator_type *elv = e->elevator_type;
1088 struct list_head *entry;
1089 int len = 0;
1090
1091 spin_lock_irq(q->queue_lock);
1092 list_for_each(entry, &elv_list) {
1093 struct elevator_type *__e;
1094
1095 __e = list_entry(entry, struct elevator_type, list);
1096 if (!strcmp(elv->elevator_name, __e->elevator_name))
1097 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1098 else
1099 len += sprintf(name+len, "%s ", __e->elevator_name);
1100 }
1101 spin_unlock_irq(q->queue_lock);
1102
1103 len += sprintf(len+name, "\n");
1104 return len;
1105}
1106
Jens Axboe2e662b62006-07-13 11:55:04 +02001107struct request *elv_rb_former_request(request_queue_t *q, struct request *rq)
1108{
1109 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1110
1111 if (rbprev)
1112 return rb_entry_rq(rbprev);
1113
1114 return NULL;
1115}
1116
1117EXPORT_SYMBOL(elv_rb_former_request);
1118
1119struct request *elv_rb_latter_request(request_queue_t *q, struct request *rq)
1120{
1121 struct rb_node *rbnext = rb_next(&rq->rb_node);
1122
1123 if (rbnext)
1124 return rb_entry_rq(rbnext);
1125
1126 return NULL;
1127}
1128
1129EXPORT_SYMBOL(elv_rb_latter_request);