We are currently experiencing downtime impacting viewing & cloning the Mesa repo, and some GitLab pages returning 503. Please see #freedesktop on IRC for more updates.

elevator.c 24.6 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5
/*
 *  Block device elevator/IO-scheduler.
 *
 *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
 *
6
 * 30042000 Jens Axboe <axboe@kernel.dk> :
Linus Torvalds's avatar
Linus Torvalds committed
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
 *
 * Split the elevator a bit so that it is possible to choose a different
 * one or even write a new "plug in". There are three pieces:
 * - elevator_fn, inserts a new request in the queue list
 * - elevator_merge_fn, decides whether a new buffer can be merged with
 *   an existing request
 * - elevator_dequeue_fn, called when a request is taken off the active list
 *
 * 20082000 Dave Jones <davej@suse.de> :
 * Removed tests for max-bomb-segments, which was breaking elvtune
 *  when run without -bN
 *
 * Jens:
 * - Rework again to work with bio instead of buffer_heads
 * - loose bi_dev comparisons, partition handling is right now
 * - completely modularize elevator setup and teardown
 *
 */
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/blkdev.h>
#include <linux/elevator.h>
#include <linux/bio.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/init.h>
#include <linux/compiler.h>
34
#include <linux/delay.h>
35
#include <linux/blktrace_api.h>
36
#include <linux/hash.h>
37
#include <linux/uaccess.h>
Linus Torvalds's avatar
Linus Torvalds committed
38

39 40
#include <trace/events/block.h>

41 42
#include "blk.h"

Linus Torvalds's avatar
Linus Torvalds committed
43 44 45
static DEFINE_SPINLOCK(elv_list_lock);
static LIST_HEAD(elv_list);

46 47 48 49 50
/*
 * Merge hash stuff.
 */
static const int elv_hash_shift = 6;
#define ELV_HASH_BLOCK(sec)	((sec) >> 3)
51 52
#define ELV_HASH_FN(sec)	\
		(hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
53
#define ELV_HASH_ENTRIES	(1 << elv_hash_shift)
54
#define rq_hash_key(rq)		(blk_rq_pos(rq) + blk_rq_sectors(rq))
55

56 57 58 59 60 61
/*
 * Query io scheduler to see if the current process issuing bio may be
 * merged with rq.
 */
static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
{
62
	struct request_queue *q = rq->q;
63
	struct elevator_queue *e = q->elevator;
64 65 66 67 68 69 70

	if (e->ops->elevator_allow_merge_fn)
		return e->ops->elevator_allow_merge_fn(q, rq, bio);

	return 1;
}

Linus Torvalds's avatar
Linus Torvalds committed
71 72 73
/*
 * can we safely merge with this request?
 */
74
int elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
75 76 77 78
{
	if (!rq_mergeable(rq))
		return 0;

79 80 81 82 83 84
	/*
	 * Don't merge file system requests and discard requests
	 */
	if (bio_discard(bio) != bio_discard(rq->bio))
		return 0;

Linus Torvalds's avatar
Linus Torvalds committed
85 86 87 88 89 90 91
	/*
	 * different data direction or already started, don't merge
	 */
	if (bio_data_dir(bio) != rq_data_dir(rq))
		return 0;

	/*
92
	 * must be same device and not a special request
Linus Torvalds's avatar
Linus Torvalds committed
93
	 */
94
	if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
95 96
		return 0;

97 98 99 100 101 102
	/*
	 * only merge integrity protected bio into ditto rq
	 */
	if (bio_integrity(bio) != blk_integrity_rq(rq))
		return 0;

103 104 105 106 107 108 109 110
	/*
	 * Don't merge if failfast settings don't match
	 */
	if (bio_failfast_dev(bio)	!= blk_failfast_dev(rq)		||
	    bio_failfast_transport(bio)	!= blk_failfast_transport(rq)	||
	    bio_failfast_driver(bio)	!= blk_failfast_driver(rq))
		return 0;

111 112
	if (!elv_iosched_allow_merge(rq, bio))
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
113

114
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
115 116 117
}
EXPORT_SYMBOL(elv_rq_merge_ok);

118
static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
119 120 121 122 123 124 125
{
	int ret = ELEVATOR_NO_MERGE;

	/*
	 * we can merge and sequence is ok, check if it's possible
	 */
	if (elv_rq_merge_ok(__rq, bio)) {
126
		if (blk_rq_pos(__rq) + blk_rq_sectors(__rq) == bio->bi_sector)
Linus Torvalds's avatar
Linus Torvalds committed
127
			ret = ELEVATOR_BACK_MERGE;
128
		else if (blk_rq_pos(__rq) - bio_sectors(bio) == bio->bi_sector)
Linus Torvalds's avatar
Linus Torvalds committed
129 130 131 132 133 134 135 136
			ret = ELEVATOR_FRONT_MERGE;
	}

	return ret;
}

static struct elevator_type *elevator_find(const char *name)
{
137
	struct elevator_type *e;
Linus Torvalds's avatar
Linus Torvalds committed
138

139
	list_for_each_entry(e, &elv_list, list) {
140 141
		if (!strcmp(e->elevator_name, name))
			return e;
Linus Torvalds's avatar
Linus Torvalds committed
142 143
	}

144
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
145 146 147 148 149 150 151 152 153
}

static void elevator_put(struct elevator_type *e)
{
	module_put(e->elevator_owner);
}

static struct elevator_type *elevator_get(const char *name)
{
154
	struct elevator_type *e;
Linus Torvalds's avatar
Linus Torvalds committed
155

156
	spin_lock(&elv_list_lock);
157 158

	e = elevator_find(name);
159 160 161 162 163 164 165 166 167 168
	if (!e) {
		char elv[ELV_NAME_MAX + strlen("-iosched")];

		spin_unlock(&elv_list_lock);

		if (!strcmp(name, "anticipatory"))
			sprintf(elv, "as-iosched");
		else
			sprintf(elv, "%s-iosched", name);

169
		request_module("%s", elv);
170 171 172 173
		spin_lock(&elv_list_lock);
		e = elevator_find(name);
	}

174 175 176
	if (e && !try_module_get(e->elevator_owner))
		e = NULL;

177
	spin_unlock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
178 179 180 181

	return e;
}

182 183
static void *elevator_init_queue(struct request_queue *q,
				 struct elevator_queue *eq)
Linus Torvalds's avatar
Linus Torvalds committed
184
{
185
	return eq->ops->elevator_init_fn(q);
Jens Axboe's avatar
Jens Axboe committed
186
}
Linus Torvalds's avatar
Linus Torvalds committed
187

188
static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboe's avatar
Jens Axboe committed
189 190
			   void *data)
{
Linus Torvalds's avatar
Linus Torvalds committed
191
	q->elevator = eq;
Jens Axboe's avatar
Jens Axboe committed
192
	eq->elevator_data = data;
Linus Torvalds's avatar
Linus Torvalds committed
193 194 195 196
}

static char chosen_elevator[16];

197
static int __init elevator_setup(char *str)
Linus Torvalds's avatar
Linus Torvalds committed
198
{
199 200 201 202
	/*
	 * Be backwards-compatible with previous kernels, so users
	 * won't get the wrong elevator.
	 */
203
	if (!strcmp(str, "as"))
204
		strcpy(chosen_elevator, "anticipatory");
205
	else
206
		strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
207
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
208 209 210 211
}

__setup("elevator=", elevator_setup);

212 213
static struct kobj_type elv_ktype;

214
static struct elevator_queue *elevator_alloc(struct request_queue *q,
215
				  struct elevator_type *e)
216
{
217
	struct elevator_queue *eq;
218 219
	int i;

220
	eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
221 222 223 224 225
	if (unlikely(!eq))
		goto err;

	eq->ops = &e->ops;
	eq->elevator_type = e;
226
	kobject_init(&eq->kobj, &elv_ktype);
227 228
	mutex_init(&eq->sysfs_lock);

229 230
	eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
					GFP_KERNEL, q->node);
231 232 233 234 235 236
	if (!eq->hash)
		goto err;

	for (i = 0; i < ELV_HASH_ENTRIES; i++)
		INIT_HLIST_HEAD(&eq->hash[i]);

237
	return eq;
238 239 240 241
err:
	kfree(eq);
	elevator_put(e);
	return NULL;
242 243 244 245
}

static void elevator_release(struct kobject *kobj)
{
246
	struct elevator_queue *e;
247

248
	e = container_of(kobj, struct elevator_queue, kobj);
249
	elevator_put(e->elevator_type);
250
	kfree(e->hash);
251 252 253
	kfree(e);
}

254
int elevator_init(struct request_queue *q, char *name)
Linus Torvalds's avatar
Linus Torvalds committed
255 256 257 258
{
	struct elevator_type *e = NULL;
	struct elevator_queue *eq;
	int ret = 0;
Jens Axboe's avatar
Jens Axboe committed
259
	void *data;
Linus Torvalds's avatar
Linus Torvalds committed
260

261 262 263 264 265
	INIT_LIST_HEAD(&q->queue_head);
	q->last_merge = NULL;
	q->end_sector = 0;
	q->boundary_rq = NULL;

266 267 268 269 270
	if (name) {
		e = elevator_get(name);
		if (!e)
			return -EINVAL;
	}
Linus Torvalds's avatar
Linus Torvalds committed
271

272 273 274 275 276 277
	if (!e && *chosen_elevator) {
		e = elevator_get(chosen_elevator);
		if (!e)
			printk(KERN_ERR "I/O scheduler %s not found\n",
							chosen_elevator);
	}
278

279 280 281 282 283 284 285 286
	if (!e) {
		e = elevator_get(CONFIG_DEFAULT_IOSCHED);
		if (!e) {
			printk(KERN_ERR
				"Default I/O scheduler not found. " \
				"Using noop.\n");
			e = elevator_get("noop");
		}
287 288
	}

289
	eq = elevator_alloc(q, e);
290
	if (!eq)
Linus Torvalds's avatar
Linus Torvalds committed
291 292
		return -ENOMEM;

Jens Axboe's avatar
Jens Axboe committed
293 294
	data = elevator_init_queue(q, eq);
	if (!data) {
295
		kobject_put(&eq->kobj);
Jens Axboe's avatar
Jens Axboe committed
296 297
		return -ENOMEM;
	}
Linus Torvalds's avatar
Linus Torvalds committed
298

Jens Axboe's avatar
Jens Axboe committed
299
	elevator_attach(q, eq, data);
Linus Torvalds's avatar
Linus Torvalds committed
300 301
	return ret;
}
302 303
EXPORT_SYMBOL(elevator_init);

304
void elevator_exit(struct elevator_queue *e)
Linus Torvalds's avatar
Linus Torvalds committed
305
{
306
	mutex_lock(&e->sysfs_lock);
Linus Torvalds's avatar
Linus Torvalds committed
307 308
	if (e->ops->elevator_exit_fn)
		e->ops->elevator_exit_fn(e);
309 310
	e->ops = NULL;
	mutex_unlock(&e->sysfs_lock);
Linus Torvalds's avatar
Linus Torvalds committed
311

312
	kobject_put(&e->kobj);
Linus Torvalds's avatar
Linus Torvalds committed
313
}
314 315
EXPORT_SYMBOL(elevator_exit);

316 317 318 319 320
static inline void __elv_rqhash_del(struct request *rq)
{
	hlist_del_init(&rq->hash);
}

321
static void elv_rqhash_del(struct request_queue *q, struct request *rq)
322 323 324 325 326
{
	if (ELV_ON_HASH(rq))
		__elv_rqhash_del(rq);
}

327
static void elv_rqhash_add(struct request_queue *q, struct request *rq)
328
{
329
	struct elevator_queue *e = q->elevator;
330 331 332 333 334

	BUG_ON(ELV_ON_HASH(rq));
	hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
}

335
static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
336 337 338 339 340
{
	__elv_rqhash_del(rq);
	elv_rqhash_add(q, rq);
}

341
static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
342
{
343
	struct elevator_queue *e = q->elevator;
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
	struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
	struct hlist_node *entry, *next;
	struct request *rq;

	hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
		BUG_ON(!ELV_ON_HASH(rq));

		if (unlikely(!rq_mergeable(rq))) {
			__elv_rqhash_del(rq);
			continue;
		}

		if (rq_hash_key(rq) == offset)
			return rq;
	}

	return NULL;
}

363 364 365 366 367 368 369 370 371 372 373 374 375 376
/*
 * RB-tree support functions for inserting/lookup/removal of requests
 * in a sorted RB tree.
 */
struct request *elv_rb_add(struct rb_root *root, struct request *rq)
{
	struct rb_node **p = &root->rb_node;
	struct rb_node *parent = NULL;
	struct request *__rq;

	while (*p) {
		parent = *p;
		__rq = rb_entry(parent, struct request, rb_node);

377
		if (blk_rq_pos(rq) < blk_rq_pos(__rq))
378
			p = &(*p)->rb_left;
379
		else if (blk_rq_pos(rq) > blk_rq_pos(__rq))
380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
			p = &(*p)->rb_right;
		else
			return __rq;
	}

	rb_link_node(&rq->rb_node, parent, p);
	rb_insert_color(&rq->rb_node, root);
	return NULL;
}
EXPORT_SYMBOL(elv_rb_add);

void elv_rb_del(struct rb_root *root, struct request *rq)
{
	BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
	rb_erase(&rq->rb_node, root);
	RB_CLEAR_NODE(&rq->rb_node);
}
EXPORT_SYMBOL(elv_rb_del);

struct request *elv_rb_find(struct rb_root *root, sector_t sector)
{
	struct rb_node *n = root->rb_node;
	struct request *rq;

	while (n) {
		rq = rb_entry(n, struct request, rb_node);

407
		if (sector < blk_rq_pos(rq))
408
			n = n->rb_left;
409
		else if (sector > blk_rq_pos(rq))
410 411 412 413 414 415 416 417 418
			n = n->rb_right;
		else
			return rq;
	}

	return NULL;
}
EXPORT_SYMBOL(elv_rb_find);

419 420
/*
 * Insert rq into dispatch queue of q.  Queue lock must be held on
421
 * entry.  rq is sort instead into the dispatch queue. To be used by
422
 * specific elevators.
423
 */
424
void elv_dispatch_sort(struct request_queue *q, struct request *rq)
425 426 427
{
	sector_t boundary;
	struct list_head *entry;
428
	int stop_flags;
429

430 431
	if (q->last_merge == rq)
		q->last_merge = NULL;
432 433 434

	elv_rqhash_del(q, rq);

435
	q->nr_sorted--;
436

Jens Axboe's avatar
Jens Axboe committed
437
	boundary = q->end_sector;
438
	stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
439 440 441
	list_for_each_prev(entry, &q->queue_head) {
		struct request *pos = list_entry_rq(entry);

442 443
		if (blk_discard_rq(rq) != blk_discard_rq(pos))
			break;
444 445
		if (rq_data_dir(rq) != rq_data_dir(pos))
			break;
446
		if (pos->cmd_flags & stop_flags)
447
			break;
448 449
		if (blk_rq_pos(rq) >= boundary) {
			if (blk_rq_pos(pos) < boundary)
450 451
				continue;
		} else {
452
			if (blk_rq_pos(pos) >= boundary)
453 454
				break;
		}
455
		if (blk_rq_pos(rq) >= blk_rq_pos(pos))
456 457 458 459 460
			break;
	}

	list_add(&rq->queuelist, entry);
}
461 462
EXPORT_SYMBOL(elv_dispatch_sort);

463
/*
464 465 466
 * Insert rq into dispatch queue of q.  Queue lock must be held on
 * entry.  rq is added to the back of the dispatch queue. To be used by
 * specific elevators.
467 468 469 470 471 472 473 474 475 476 477 478 479 480
 */
void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
{
	if (q->last_merge == rq)
		q->last_merge = NULL;

	elv_rqhash_del(q, rq);

	q->nr_sorted--;

	q->end_sector = rq_end_sector(rq);
	q->boundary_rq = rq;
	list_add_tail(&rq->queuelist, &q->queue_head);
}
481 482
EXPORT_SYMBOL(elv_dispatch_add_tail);

483
int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
484
{
485
	struct elevator_queue *e = q->elevator;
486
	struct request *__rq;
487 488
	int ret;

489 490 491
	/*
	 * First try one-hit cache.
	 */
492 493 494 495 496 497 498
	if (q->last_merge) {
		ret = elv_try_merge(q->last_merge, bio);
		if (ret != ELEVATOR_NO_MERGE) {
			*req = q->last_merge;
			return ret;
		}
	}
Linus Torvalds's avatar
Linus Torvalds committed
499

500 501 502
	if (blk_queue_nomerges(q))
		return ELEVATOR_NO_MERGE;

503 504 505 506 507 508 509 510 511
	/*
	 * See if our hash lookup can find a potential backmerge.
	 */
	__rq = elv_rqhash_find(q, bio->bi_sector);
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
		*req = __rq;
		return ELEVATOR_BACK_MERGE;
	}

Linus Torvalds's avatar
Linus Torvalds committed
512 513 514 515 516 517
	if (e->ops->elevator_merge_fn)
		return e->ops->elevator_merge_fn(q, req, bio);

	return ELEVATOR_NO_MERGE;
}

518
void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds's avatar
Linus Torvalds committed
519
{
520
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
521 522

	if (e->ops->elevator_merged_fn)
523
		e->ops->elevator_merged_fn(q, rq, type);
524

525 526
	if (type == ELEVATOR_BACK_MERGE)
		elv_rqhash_reposition(q, rq);
527

528
	q->last_merge = rq;
Linus Torvalds's avatar
Linus Torvalds committed
529 530
}

531
void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds's avatar
Linus Torvalds committed
532 533
			     struct request *next)
{
534
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
535 536 537

	if (e->ops->elevator_merge_req_fn)
		e->ops->elevator_merge_req_fn(q, rq, next);
538

539 540 541 542
	elv_rqhash_reposition(q, rq);
	elv_rqhash_del(q, next);

	q->nr_sorted--;
543
	q->last_merge = rq;
Linus Torvalds's avatar
Linus Torvalds committed
544 545
}

546
void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
547 548 549 550 551
{
	/*
	 * it already went through dequeue, we need to decrement the
	 * in_flight count again
	 */
552
	if (blk_account_rq(rq)) {
553
		q->in_flight[rq_is_sync(rq)]--;
554 555
		if (blk_sorted_rq(rq))
			elv_deactivate_rq(q, rq);
556
	}
Linus Torvalds's avatar
Linus Torvalds committed
557

558
	rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds's avatar
Linus Torvalds committed
559

560
	elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds's avatar
Linus Torvalds committed
561 562
}

563
void elv_drain_elevator(struct request_queue *q)
564 565 566 567 568 569 570 571 572 573 574 575 576
{
	static int printed;
	while (q->elevator->ops->elevator_dispatch_fn(q, 1))
		;
	if (q->nr_sorted == 0)
		return;
	if (printed++ < 10) {
		printk(KERN_ERR "%s: forced dispatching is broken "
		       "(nr_sorted=%u), please report this\n",
		       q->elevator->elevator_type->elevator_name, q->nr_sorted);
	}
}

Jens Axboe's avatar
Jens Axboe committed
577 578 579
/*
 * Call with queue lock held, interrupts disabled
 */
580
void elv_quiesce_start(struct request_queue *q)
Jens Axboe's avatar
Jens Axboe committed
581
{
582 583 584
	if (!q->elevator)
		return;

Jens Axboe's avatar
Jens Axboe committed
585 586 587 588 589 590 591
	queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);

	/*
	 * make sure we don't have any requests in flight
	 */
	elv_drain_elevator(q);
	while (q->rq.elvpriv) {
Tejun Heo's avatar
Tejun Heo committed
592
		__blk_run_queue(q);
Jens Axboe's avatar
Jens Axboe committed
593 594 595 596 597 598 599
		spin_unlock_irq(q->queue_lock);
		msleep(10);
		spin_lock_irq(q->queue_lock);
		elv_drain_elevator(q);
	}
}

600
void elv_quiesce_end(struct request_queue *q)
Jens Axboe's avatar
Jens Axboe committed
601 602 603 604
{
	queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
}

605
void elv_insert(struct request_queue *q, struct request *rq, int where)
Linus Torvalds's avatar
Linus Torvalds committed
606
{
607 608
	struct list_head *pos;
	unsigned ordseq;
609
	int unplug_it = 1;
610

611
	trace_block_rq_insert(q, rq);
612

Linus Torvalds's avatar
Linus Torvalds committed
613 614
	rq->q = q;

615 616
	switch (where) {
	case ELEVATOR_INSERT_FRONT:
617
		rq->cmd_flags |= REQ_SOFTBARRIER;
618 619 620 621 622

		list_add(&rq->queuelist, &q->queue_head);
		break;

	case ELEVATOR_INSERT_BACK:
623
		rq->cmd_flags |= REQ_SOFTBARRIER;
624
		elv_drain_elevator(q);
625 626 627 628 629 630 631 632 633 634 635
		list_add_tail(&rq->queuelist, &q->queue_head);
		/*
		 * We kick the queue here for the following reasons.
		 * - The elevator might have returned NULL previously
		 *   to delay requests and returned them now.  As the
		 *   queue wasn't empty before this request, ll_rw_blk
		 *   won't run the queue on return, resulting in hang.
		 * - Usually, back inserted requests won't be merged
		 *   with anything.  There's no point in delaying queue
		 *   processing.
		 */
Tejun Heo's avatar
Tejun Heo committed
636
		__blk_run_queue(q);
637 638 639
		break;

	case ELEVATOR_INSERT_SORT:
640
		BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq));
641
		rq->cmd_flags |= REQ_SORTED;
642
		q->nr_sorted++;
643 644 645 646 647 648
		if (rq_mergeable(rq)) {
			elv_rqhash_add(q, rq);
			if (!q->last_merge)
				q->last_merge = rq;
		}

649 650 651 652 653 654
		/*
		 * Some ioscheds (cfq) run q->request_fn directly, so
		 * rq cannot be accessed after calling
		 * elevator_add_req_fn.
		 */
		q->elevator->ops->elevator_add_req_fn(q, rq);
655 656
		break;

657 658 659 660 661 662
	case ELEVATOR_INSERT_REQUEUE:
		/*
		 * If ordered flush isn't in progress, we do front
		 * insertion; otherwise, requests should be requeued
		 * in ordseq order.
		 */
663
		rq->cmd_flags |= REQ_SOFTBARRIER;
664

665 666 667 668 669 670
		/*
		 * Most requeues happen because of a busy condition,
		 * don't force unplug of the queue for that case.
		 */
		unplug_it = 0;

671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686
		if (q->ordseq == 0) {
			list_add(&rq->queuelist, &q->queue_head);
			break;
		}

		ordseq = blk_ordered_req_seq(rq);

		list_for_each(pos, &q->queue_head) {
			struct request *pos_rq = list_entry_rq(pos);
			if (ordseq <= blk_ordered_req_seq(pos_rq))
				break;
		}

		list_add_tail(&rq->queuelist, pos);
		break;

687 688
	default:
		printk(KERN_ERR "%s: bad insertion point %d\n",
689
		       __func__, where);
690 691 692
		BUG();
	}

693
	if (unplug_it && blk_queue_plugged(q)) {
694
		int nrq = q->rq.count[BLK_RW_SYNC] + q->rq.count[BLK_RW_ASYNC]
695
				- queue_in_flight(q);
696 697 698 699

		if (nrq >= q->unplug_thresh)
			__generic_unplug_device(q);
	}
Linus Torvalds's avatar
Linus Torvalds committed
700 701
}

702
void __elv_add_request(struct request_queue *q, struct request *rq, int where,
703 704 705
		       int plug)
{
	if (q->ordcolor)
706
		rq->cmd_flags |= REQ_ORDERED_COLOR;
707

708
	if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724
		/*
		 * toggle ordered color
		 */
		if (blk_barrier_rq(rq))
			q->ordcolor ^= 1;

		/*
		 * barriers implicitly indicate back insertion
		 */
		if (where == ELEVATOR_INSERT_SORT)
			where = ELEVATOR_INSERT_BACK;

		/*
		 * this request is scheduling boundary, update
		 * end_sector
		 */
725
		if (blk_fs_request(rq) || blk_discard_rq(rq)) {
726 727 728
			q->end_sector = rq_end_sector(rq);
			q->boundary_rq = rq;
		}
729 730
	} else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
		    where == ELEVATOR_INSERT_SORT)
731 732 733 734 735 736 737
		where = ELEVATOR_INSERT_BACK;

	if (plug)
		blk_plug_device(q);

	elv_insert(q, rq, where);
}
738 739
EXPORT_SYMBOL(__elv_add_request);

740
void elv_add_request(struct request_queue *q, struct request *rq, int where,
Linus Torvalds's avatar
Linus Torvalds committed
741 742 743 744 745 746 747 748
		     int plug)
{
	unsigned long flags;

	spin_lock_irqsave(q->queue_lock, flags);
	__elv_add_request(q, rq, where, plug);
	spin_unlock_irqrestore(q->queue_lock, flags);
}
749 750
EXPORT_SYMBOL(elv_add_request);

751
int elv_queue_empty(struct request_queue *q)
Linus Torvalds's avatar
Linus Torvalds committed
752
{
753
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
754

755 756 757
	if (!list_empty(&q->queue_head))
		return 0;

Linus Torvalds's avatar
Linus Torvalds committed
758 759 760
	if (e->ops->elevator_queue_empty_fn)
		return e->ops->elevator_queue_empty_fn(q);

761
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
762
}
763 764
EXPORT_SYMBOL(elv_queue_empty);

765
struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
766
{
767
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
768 769 770 771 772 773

	if (e->ops->elevator_latter_req_fn)
		return e->ops->elevator_latter_req_fn(q, rq);
	return NULL;
}

774
struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
775
{
776
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
777 778 779 780 781 782

	if (e->ops->elevator_former_req_fn)
		return e->ops->elevator_former_req_fn(q, rq);
	return NULL;
}

783
int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds's avatar
Linus Torvalds committed
784
{
785
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
786 787

	if (e->ops->elevator_set_req_fn)
788
		return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds's avatar
Linus Torvalds committed
789 790 791 792 793

	rq->elevator_private = NULL;
	return 0;
}

794
void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
795
{
796
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
797 798

	if (e->ops->elevator_put_req_fn)
799
		e->ops->elevator_put_req_fn(rq);
Linus Torvalds's avatar
Linus Torvalds committed
800 801
}

802
int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds's avatar
Linus Torvalds committed
803
{
804
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
805 806

	if (e->ops->elevator_may_queue_fn)
807
		return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds's avatar
Linus Torvalds committed
808 809 810 811

	return ELV_MQUEUE_MAY;
}

812 813 814 815 816 817 818
void elv_abort_queue(struct request_queue *q)
{
	struct request *rq;

	while (!list_empty(&q->queue_head)) {
		rq = list_entry_rq(q->queue_head.next);
		rq->cmd_flags |= REQ_QUIET;
819
		trace_block_rq_abort(q, rq);
820 821 822 823 824
		/*
		 * Mark this request as started so we don't trigger
		 * any debug logic in the end I/O path.
		 */
		blk_start_request(rq);
825
		__blk_end_request_all(rq, -EIO);
826 827 828 829
	}
}
EXPORT_SYMBOL(elv_abort_queue);

830
void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
831
{
832
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
833 834 835 836

	/*
	 * request is released from the driver, io must be done
	 */
837
	if (blk_account_rq(rq)) {
838
		q->in_flight[rq_is_sync(rq)]--;
839 840 841
		if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
			e->ops->elevator_completed_req_fn(q, rq);
	}
842

843 844 845 846 847
	/*
	 * Check if the queue is waiting for fs requests to be
	 * drained for flush sequence.
	 */
	if (unlikely(q->ordseq)) {
848 849 850 851 852
		struct request *next = NULL;

		if (!list_empty(&q->queue_head))
			next = list_entry_rq(q->queue_head.next);

853
		if (!queue_in_flight(q) &&
854
		    blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
855
		    (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
856
			blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
Tejun Heo's avatar
Tejun Heo committed
857
			__blk_run_queue(q);
858
		}
859
	}
Linus Torvalds's avatar
Linus Torvalds committed
860 861
}

862 863 864 865
#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)

static ssize_t
elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
Linus Torvalds's avatar
Linus Torvalds committed
866
{
867
	struct elv_fs_entry *entry = to_elv(attr);
868
	struct elevator_queue *e;
869 870 871 872 873
	ssize_t error;

	if (!entry->show)
		return -EIO;

874
	e = container_of(kobj, struct elevator_queue, kobj);
875 876 877 878 879
	mutex_lock(&e->sysfs_lock);
	error = e->ops ? entry->show(e, page) : -ENOENT;
	mutex_unlock(&e->sysfs_lock);
	return error;
}
Linus Torvalds's avatar
Linus Torvalds committed
880

881 882 883 884 885
static ssize_t
elv_attr_store(struct kobject *kobj, struct attribute *attr,
	       const char *page, size_t length)
{
	struct elv_fs_entry *entry = to_elv(attr);
886
	struct elevator_queue *e;
887
	ssize_t error;
Linus Torvalds's avatar
Linus Torvalds committed
888

889 890
	if (!entry->store)
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
891

892
	e = container_of(kobj, struct elevator_queue, kobj);
893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910
	mutex_lock(&e->sysfs_lock);
	error = e->ops ? entry->store(e, page, length) : -ENOENT;
	mutex_unlock(&e->sysfs_lock);
	return error;
}

static struct sysfs_ops elv_sysfs_ops = {
	.show	= elv_attr_show,
	.store	= elv_attr_store,
};

static struct kobj_type elv_ktype = {
	.sysfs_ops	= &elv_sysfs_ops,
	.release	= elevator_release,
};

int elv_register_queue(struct request_queue *q)
{
911
	struct elevator_queue *e = q->elevator;
912 913
	int error;

914
	error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
915
	if (!error) {
916
		struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
917
		if (attr) {
918 919
			while (attr->attr.name) {
				if (sysfs_create_file(&e->kobj, &attr->attr))
920
					break;
921
				attr++;
922 923 924 925 926
			}
		}
		kobject_uevent(&e->kobj, KOBJ_ADD);
	}
	return error;
Linus Torvalds's avatar
Linus Torvalds committed
927 928
}

929
static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboe's avatar
Jens Axboe committed
930 931 932 933 934
{
	kobject_uevent(&e->kobj, KOBJ_REMOVE);
	kobject_del(&e->kobj);
}

Linus Torvalds's avatar
Linus Torvalds committed
935 936
void elv_unregister_queue(struct request_queue *q)
{
Jens Axboe's avatar
Jens Axboe committed
937 938
	if (q)
		__elv_unregister_queue(q->elevator);
Linus Torvalds's avatar
Linus Torvalds committed
939 940
}

941
void elv_register(struct elevator_type *e)
Linus Torvalds's avatar
Linus Torvalds committed
942
{
943
	char *def = "";
944 945

	spin_lock(&elv_list_lock);
946
	BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds's avatar
Linus Torvalds committed
947
	list_add_tail(&e->list, &elv_list);
948
	spin_unlock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
949

950 951 952
	if (!strcmp(e->elevator_name, chosen_elevator) ||
			(!*chosen_elevator &&
			 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
953 954
				def = " (default)";

955 956
	printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
								def);
Linus Torvalds's avatar
Linus Torvalds committed
957 958 959 960 961
}
EXPORT_SYMBOL_GPL(elv_register);

void elv_unregister(struct elevator_type *e)
{
962 963 964 965 966
	struct task_struct *g, *p;

	/*
	 * Iterate every thread in the process to remove the io contexts.
	 */
967 968 969 970
	if (e->ops.trim) {
		read_lock(&tasklist_lock);
		do_each_thread(g, p) {
			task_lock(p);
971 972
			if (p->io_context)
				e->ops.trim(p->io_context);
973 974 975 976
			task_unlock(p);
		} while_each_thread(g, p);
		read_unlock(&tasklist_lock);
	}
977

978
	spin_lock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
979
	list_del_init(&e->list);
980
	spin_unlock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
981 982 983 984 985 986 987
}
EXPORT_SYMBOL_GPL(elv_unregister);

/*
 * switch to new_e io scheduler. be careful not to introduce deadlocks -
 * we don't free the old io scheduler, before we have allocated what we
 * need for the new one. this way we have a chance of going back to the old
988
 * one, if the new one fails init for some reason.
Linus Torvalds's avatar
Linus Torvalds committed
989
 */
990
static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds's avatar
Linus Torvalds committed
991
{
992
	struct elevator_queue *old_elevator, *e;
Jens Axboe's avatar
Jens Axboe committed
993
	void *data;
Linus Torvalds's avatar
Linus Torvalds committed
994

995 996 997
	/*
	 * Allocate new elevator
	 */
998
	e = elevator_alloc(q, new_e);
Linus Torvalds's avatar
Linus Torvalds committed
999
	if (!e)
1000
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
1001

Jens Axboe's avatar
Jens Axboe committed
1002 1003 1004 1005 1006 1007
	data = elevator_init_queue(q, e);
	if (!data) {
		kobject_put(&e->kobj);
		return 0;
	}

Linus Torvalds's avatar
Linus Torvalds committed
1008
	/*
1009
	 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds's avatar
Linus Torvalds committed
1010
	 */
1011
	spin_lock_irq(q->queue_lock);
1012
	elv_quiesce_start(q);
1013

Linus Torvalds's avatar
Linus Torvalds committed
1014
	/*
Jens Axboe's avatar
Jens Axboe committed
1015
	 * Remember old elevator.
Linus Torvalds's avatar
Linus Torvalds committed
1016 1017 1018 1019 1020 1021
	 */
	old_elevator = q->elevator;

	/*
	 * attach and start new elevator
	 */
Jens Axboe's avatar
Jens Axboe committed
1022 1023 1024 1025 1026
	elevator_attach(q, e, data);

	spin_unlock_irq(q->queue_lock);

	__elv_unregister_queue(old_elevator);
Linus Torvalds's avatar
Linus Torvalds committed
1027 1028 1029 1030 1031

	if (elv_register_queue(q))
		goto fail_register;

	/*
1032
	 * finally exit old elevator and turn off BYPASS.
Linus Torvalds's avatar
Linus Torvalds committed
1033 1034
	 */
	elevator_exit(old_elevator);
1035
	spin_lock_irq(q->queue_lock);
1036
	elv_quiesce_end(q);
1037 1038
	spin_unlock_irq(q->queue_lock);

1039 1040
	blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);

1041
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
1042 1043 1044 1045 1046 1047 1048 1049 1050

fail_register:
	/*
	 * switch failed, exit the new io scheduler and reattach the old
	 * one again (along with re-adding the sysfs dir)
	 */
	elevator_exit(e);
	q->elevator = old_elevator;
	elv_register_queue(q);
1051 1052 1053 1054 1055

	spin_lock_irq(q->queue_lock);
	queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
	spin_unlock_irq(q->queue_lock);

1056
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
1057 1058
}

1059 1060
ssize_t elv_iosched_store(struct request_queue *q