cfq-iosched.c 69.5 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6
/*
 *  CFQ, or complete fairness queueing, disk scheduler.
 *
 *  Based on ideas from a previously unfinished io
 *  scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
 *
7
 *  Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
Linus Torvalds's avatar
Linus Torvalds committed
8 9
 */
#include <linux/module.h>
Al Viro's avatar
Al Viro committed
10 11
#include <linux/blkdev.h>
#include <linux/elevator.h>
Linus Torvalds's avatar
Linus Torvalds committed
12
#include <linux/rbtree.h>
13
#include <linux/ioprio.h>
14
#include <linux/blktrace_api.h>
Linus Torvalds's avatar
Linus Torvalds committed
15 16 17 18

/*
 * tunables
 */
19 20
/* max queue in one round of service */
static const int cfq_quantum = 4;
21
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
22 23 24 25
/* maximum backwards seek, in KiB */
static const int cfq_back_max = 16 * 1024;
/* penalty of a backwards seek */
static const int cfq_back_penalty = 2;
26
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
27
static int cfq_slice_async = HZ / 25;
28
static const int cfq_slice_async_rq = 2;
29
static int cfq_slice_idle = HZ / 125;
30

31
/*
32
 * offset from end of service tree
33
 */
34
#define CFQ_IDLE_DELAY		(HZ / 5)
35 36 37 38 39 40

/*
 * below this threshold, we consider thinktime immediate
 */
#define CFQ_MIN_TT		(2)

41 42 43 44 45 46
/*
 * Allow merged cfqqs to perform this amount of seeky I/O before
 * deciding to break the queues up again.
 */
#define CFQQ_COOP_TOUT		(HZ)

47
#define CFQ_SLICE_SCALE		(5)
48
#define CFQ_HW_QUEUE_MIN	(5)
49

50 51
#define RQ_CIC(rq)		\
	((struct cfq_io_context *) (rq)->elevator_private)
52
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
Linus Torvalds's avatar
Linus Torvalds committed
53

54 55
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
56

57
static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
58
static struct completion *ioc_gone;
59
static DEFINE_SPINLOCK(ioc_gone_lock);
60

61 62 63 64
#define CFQ_PRIO_LISTS		IOPRIO_BE_NR
#define cfq_class_idle(cfqq)	((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
#define cfq_class_rt(cfqq)	((cfqq)->ioprio_class == IOPRIO_CLASS_RT)

65 66
#define sample_valid(samples)	((samples) > 80)

67 68 69 70 71 72 73 74 75 76 77 78
/*
 * Most of our rbtree usage is for sorting with min extraction, so
 * if we cache the leftmost node we don't have to walk down the tree
 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
 * move this into the elevator for the rq sorting as well.
 */
struct cfq_rb_root {
	struct rb_root rb;
	struct rb_node *left;
};
#define CFQ_RB_ROOT	(struct cfq_rb_root) { RB_ROOT, NULL, }

79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
	atomic_t ref;
	/* various state flags, see below */
	unsigned int flags;
	/* parent cfq_data */
	struct cfq_data *cfqd;
	/* service_tree member */
	struct rb_node rb_node;
	/* service_tree key */
	unsigned long rb_key;
	/* prio tree member */
	struct rb_node p_node;
	/* prio tree root we belong to, if any */
	struct rb_root *p_root;
	/* sorted list of pending requests */
	struct rb_root sort_list;
	/* if fifo isn't expired, next request to serve */
	struct request *next_rq;
	/* requests queued in sort_list */
	int queued[2];
	/* currently allocated requests */
	int allocated[2];
	/* fifo list of requests in sort_list */
	struct list_head fifo;

	unsigned long slice_end;
	long slice_resid;
	unsigned int slice_dispatch;

	/* pending metadata requests */
	int meta_pending;
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;

	/* io prio of this group */
	unsigned short ioprio, org_ioprio;
	unsigned short ioprio_class, org_ioprio_class;

121 122 123 124
	unsigned int seek_samples;
	u64 seek_total;
	sector_t seek_mean;
	sector_t last_request_pos;
125
	unsigned long seeky_start;
126

127
	pid_t pid;
Jeff Moyer's avatar
Jeff Moyer committed
128 129

	struct cfq_queue *new_cfqq;
130 131
};

132 133 134
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
135
struct cfq_data {
136
	struct request_queue *queue;
137 138 139 140

	/*
	 * rr list of queues with requests and the count of them
	 */
141
	struct cfq_rb_root service_tree;
142 143 144 145 146 147 148 149

	/*
	 * Each priority tree is sorted by next_request position.  These
	 * trees are used when determining if two or more queues are
	 * interleaving requests (see cfq_close_cooperator).
	 */
	struct rb_root prio_trees[CFQ_PRIO_LISTS];

150 151
	unsigned int busy_queues;

152
	int rq_in_driver[2];
153
	int sync_flight;
154 155 156 157 158

	/*
	 * queue-depth detection
	 */
	int rq_queued;
159
	int hw_tag;
160 161
	int hw_tag_samples;
	int rq_in_driver_peak;
Linus Torvalds's avatar
Linus Torvalds committed
162

163 164 165 166
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
167
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
168

169 170 171
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

172 173 174 175 176
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
177

Jens Axboe's avatar
Jens Axboe committed
178
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
179 180 181 182 183

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
184
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
185 186
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
187 188 189
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
190
	unsigned int cfq_latency;
191 192

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
193

194 195 196 197
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
198 199

	unsigned long last_end_sync_rq;
Linus Torvalds's avatar
Linus Torvalds committed
200 201
};

Jens Axboe's avatar
Jens Axboe committed
202
enum cfqq_state_flags {
203 204
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
205
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
206 207 208 209
	CFQ_CFQQ_FLAG_must_alloc_slice,	/* per-slice must_alloc flag */
	CFQ_CFQQ_FLAG_fifo_expire,	/* FIFO checked in this slice */
	CFQ_CFQQ_FLAG_idle_window,	/* slice idling enabled */
	CFQ_CFQQ_FLAG_prio_changed,	/* task priority has changed */
210
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
211
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
212
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
Jens Axboe's avatar
Jens Axboe committed
213 214 215 216 217
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
218
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
219 220 221
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
222
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
223 224 225
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
226
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
227 228 229 230
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
231
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
232 233 234 235
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
236
CFQ_CFQQ_FNS(slice_new);
237
CFQ_CFQQ_FNS(sync);
238
CFQ_CFQQ_FNS(coop);
Jens Axboe's avatar
Jens Axboe committed
239 240
#undef CFQ_CFQQ_FNS

241 242 243 244 245
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

246
static void cfq_dispatch_insert(struct request_queue *, struct request *);
247
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
248
				       struct io_context *, gfp_t);
249
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
250 251
						struct io_context *);

252 253 254 255 256
static inline int rq_in_driver(struct cfq_data *cfqd)
{
	return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
}

257
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
258
					    bool is_sync)
259
{
260
	return cic->cfqq[is_sync];
261 262 263
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
264
				struct cfq_queue *cfqq, bool is_sync)
265
{
266
	cic->cfqq[is_sync] = cfqq;
267 268 269 270 271 272
}

/*
 * We regard a request as SYNC, if it's either a read or has the SYNC bit
 * set (in which case it could also be direct WRITE).
 */
273
static inline bool cfq_bio_sync(struct bio *bio)
274
{
275
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
276
}
Linus Torvalds's avatar
Linus Torvalds committed
277

Andrew Morton's avatar
Andrew Morton committed
278 279 280 281
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
282
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
283
{
284 285
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
286
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
287
	}
Andrew Morton's avatar
Andrew Morton committed
288 289
}

290
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
291 292 293
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

294
	return !cfqd->busy_queues;
Andrew Morton's avatar
Andrew Morton committed
295 296
}

297 298 299 300 301
/*
 * Scale schedule slice based on io priority. Use the sync time slice only
 * if a queue is marked sync and has sync io queued. A sync queue with async
 * io only, should not get full sync slice length.
 */
302
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
303
				 unsigned short prio)
304
{
305
	const int base_slice = cfqd->cfq_slice[sync];
306

307 308 309 310
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
311

312 313 314 315
static inline int
cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
316 317 318 319 320 321
}

static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
322
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
323 324 325 326 327 328 329
}

/*
 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
 * isn't valid until the first request from the dispatch is activated
 * and the slice time set.
 */
330
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
331 332 333 334 335 336 337 338 339
{
	if (cfq_cfqq_slice_new(cfqq))
		return 0;
	if (time_before(jiffies, cfqq->slice_end))
		return 0;

	return 1;
}

Linus Torvalds's avatar
Linus Torvalds committed
340
/*
Jens Axboe's avatar
Jens Axboe committed
341
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
342
 * We choose the request that is closest to the head right now. Distance
343
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
344
 */
Jens Axboe's avatar
Jens Axboe committed
345 346
static struct request *
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
Linus Torvalds's avatar
Linus Torvalds committed
347 348 349
{
	sector_t last, s1, s2, d1 = 0, d2 = 0;
	unsigned long back_max;
350 351 352
#define CFQ_RQ1_WRAP	0x01 /* request 1 wraps */
#define CFQ_RQ2_WRAP	0x02 /* request 2 wraps */
	unsigned wrap = 0; /* bit mask: requests behind the disk head? */
Linus Torvalds's avatar
Linus Torvalds committed
353

Jens Axboe's avatar
Jens Axboe committed
354 355 356 357
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
358

Jens Axboe's avatar
Jens Axboe committed
359 360 361 362
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
363 364 365 366
	if (rq_is_meta(rq1) && !rq_is_meta(rq2))
		return rq1;
	else if (rq_is_meta(rq2) && !rq_is_meta(rq1))
		return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
367

368 369
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
370

Jens Axboe's avatar
Jens Axboe committed
371
	last = cfqd->last_position;
Linus Torvalds's avatar
Linus Torvalds committed
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387

	/*
	 * by definition, 1KiB is 2 sectors
	 */
	back_max = cfqd->cfq_back_max * 2;

	/*
	 * Strict one way elevator _except_ in the case where we allow
	 * short backward seeks which are biased as twice the cost of a
	 * similar forward seek.
	 */
	if (s1 >= last)
		d1 = s1 - last;
	else if (s1 + back_max >= last)
		d1 = (last - s1) * cfqd->cfq_back_penalty;
	else
388
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
389 390 391 392 393 394

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
395
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
396 397

	/* Found required data */
398 399 400 401 402 403

	/*
	 * By doing switch() on the bit mask "wrap" we avoid having to
	 * check two variables for all permutations: --> faster!
	 */
	switch (wrap) {
Jens Axboe's avatar
Jens Axboe committed
404
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
405
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
406
			return rq1;
407
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
408
			return rq2;
409 410
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
411
				return rq1;
412
			else
Jens Axboe's avatar
Jens Axboe committed
413
				return rq2;
414
		}
Linus Torvalds's avatar
Linus Torvalds committed
415

416
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
417
		return rq1;
418
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
419 420
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
421 422 423 424 425 426 427 428
	default:
		/*
		 * Since both rqs are wrapped,
		 * start with the one that's further behind head
		 * (--> only *one* back seek required),
		 * since back seek takes more time than forward.
		 */
		if (s1 <= s2)
Jens Axboe's avatar
Jens Axboe committed
429
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
430
		else
Jens Axboe's avatar
Jens Axboe committed
431
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
432 433 434
	}
}

435 436 437
/*
 * The below is leftmost cache rbtree addon
 */
438
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
439 440 441 442
{
	if (!root->left)
		root->left = rb_first(&root->rb);

443 444 445 446
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
447 448
}

449 450 451 452 453 454
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

455 456 457 458
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
459
	rb_erase_init(n, &root->rb);
460 461
}

Linus Torvalds's avatar
Linus Torvalds committed
462 463 464
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
465 466 467
static struct request *
cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		  struct request *last)
Linus Torvalds's avatar
Linus Torvalds committed
468
{
469 470
	struct rb_node *rbnext = rb_next(&last->rb_node);
	struct rb_node *rbprev = rb_prev(&last->rb_node);
Jens Axboe's avatar
Jens Axboe committed
471
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
472

473
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
474 475

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
476
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
477

478
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
479
		next = rb_entry_rq(rbnext);
480 481 482
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
483
			next = rb_entry_rq(rbnext);
484
	}
Linus Torvalds's avatar
Linus Torvalds committed
485

486
	return cfq_choose_req(cfqd, next, prev);
Linus Torvalds's avatar
Linus Torvalds committed
487 488
}

489 490
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
491
{
492 493 494
	/*
	 * just an approximation, should be ok.
	 */
495 496
	return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
497 498
}

499 500 501 502 503
/*
 * The cfqd->service_tree holds all pending cfq_queue's that have
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
504
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
505
				 bool add_front)
506
{
507 508
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
509
	unsigned long rb_key;
510
	int left;
511

512 513 514 515 516 517 518 519 520
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
		parent = rb_last(&cfqd->service_tree.rb);
		if (parent && parent != &cfqq->rb_node) {
			__cfqq = rb_entry(parent, struct cfq_queue, rb_node);
			rb_key += __cfqq->rb_key;
		} else
			rb_key += jiffies;
	} else if (!add_front) {
521 522 523 524 525 526
		/*
		 * Get our rb key offset. Subtract any residual slice
		 * value carried from last service. A negative resid
		 * count indicates slice overrun, and this should position
		 * the next service time further away in the tree.
		 */
527
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
528
		rb_key -= cfqq->slice_resid;
529
		cfqq->slice_resid = 0;
530 531 532 533 534
	} else {
		rb_key = -HZ;
		__cfqq = cfq_rb_first(&cfqd->service_tree);
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
Linus Torvalds's avatar
Linus Torvalds committed
535

536
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
537
		/*
538
		 * same position, nothing more to do
539
		 */
540 541
		if (rb_key == cfqq->rb_key)
			return;
Linus Torvalds's avatar
Linus Torvalds committed
542

543
		cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
Linus Torvalds's avatar
Linus Torvalds committed
544
	}
545

546
	left = 1;
547 548
	parent = NULL;
	p = &cfqd->service_tree.rb.rb_node;
549
	while (*p) {
550
		struct rb_node **n;
551

552 553 554
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

555 556
		/*
		 * sort RT queues first, we always want to give
557 558
		 * preference to them. IDLE queues goes to the back.
		 * after that, sort on the next service time.
559 560
		 */
		if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
561
			n = &(*p)->rb_left;
562
		else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
563 564 565 566 567
			n = &(*p)->rb_right;
		else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq))
			n = &(*p)->rb_left;
		else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq))
			n = &(*p)->rb_right;
568
		else if (time_before(rb_key, __cfqq->rb_key))
569 570 571 572 573
			n = &(*p)->rb_left;
		else
			n = &(*p)->rb_right;

		if (n == &(*p)->rb_right)
574
			left = 0;
575 576

		p = n;
577 578
	}

579 580 581
	if (left)
		cfqd->service_tree.left = &cfqq->rb_node;

582 583
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
584
	rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
Linus Torvalds's avatar
Linus Torvalds committed
585 586
}

587
static struct cfq_queue *
588 589 590
cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
		     sector_t sector, struct rb_node **ret_parent,
		     struct rb_node ***rb_link)
591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606
{
	struct rb_node **p, *parent;
	struct cfq_queue *cfqq = NULL;

	parent = NULL;
	p = &root->rb_node;
	while (*p) {
		struct rb_node **n;

		parent = *p;
		cfqq = rb_entry(parent, struct cfq_queue, p_node);

		/*
		 * Sort strictly based on sector.  Smallest to the left,
		 * largest to the right.
		 */
607
		if (sector > blk_rq_pos(cfqq->next_rq))
608
			n = &(*p)->rb_right;
609
		else if (sector < blk_rq_pos(cfqq->next_rq))
610 611 612 613
			n = &(*p)->rb_left;
		else
			break;
		p = n;
614
		cfqq = NULL;
615 616 617 618 619
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
620
	return cfqq;
621 622 623 624 625 626 627
}

static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;

628 629 630 631
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
632 633 634 635 636 637

	if (cfq_class_idle(cfqq))
		return;
	if (!cfqq->next_rq)
		return;

638
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
639 640
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
641 642
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
643 644 645
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
646 647
}

648 649 650
/*
 * Update cfqq's position in the service tree.
 */
651
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
652 653 654 655
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
656
	if (cfq_cfqq_on_rr(cfqq)) {
657
		cfq_service_tree_add(cfqd, cfqq, 0);
658 659
		cfq_prio_tree_add(cfqd, cfqq);
	}
Jens Axboe's avatar
Jens Axboe committed
660 661
}

Linus Torvalds's avatar
Linus Torvalds committed
662 663
/*
 * add to busy list of queues for service, trying to be fair in ordering
664
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
665
 */
666
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
667
{
668
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
669 670
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
671 672
	cfqd->busy_queues++;

673
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
674 675
}

676 677 678 679
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
680
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
681
{
682
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
683 684
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
685

686 687
	if (!RB_EMPTY_NODE(&cfqq->rb_node))
		cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
688 689 690 691
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
692

Linus Torvalds's avatar
Linus Torvalds committed
693 694 695 696 697 698 699
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
}

/*
 * rb tree support functions
 */
700
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
701
{
Jens Axboe's avatar
Jens Axboe committed
702
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
703
	struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe's avatar
Jens Axboe committed
704
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
705

706 707
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
708

Jens Axboe's avatar
Jens Axboe committed
709
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
710

711
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
712
		cfq_del_cfqq_rr(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
713 714
}

Jens Axboe's avatar
Jens Axboe committed
715
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
716
{
Jens Axboe's avatar
Jens Axboe committed
717
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds's avatar
Linus Torvalds committed
718
	struct cfq_data *cfqd = cfqq->cfqd;
719
	struct request *__alias, *prev;
Linus Torvalds's avatar
Linus Torvalds committed
720

721
	cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds's avatar
Linus Torvalds committed
722 723 724 725 726

	/*
	 * looks a little odd, but the first insert might return an alias.
	 * if that happens, put the alias on the dispatch list
	 */
727
	while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe's avatar
Jens Axboe committed
728
		cfq_dispatch_insert(cfqd->queue, __alias);
729 730 731

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
732 733 734 735

	/*
	 * check if this request is a better next-serve candidate
	 */
736
	prev = cfqq->next_rq;
737
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
738 739 740 741 742 743 744

	/*
	 * adjust priority tree position, if ->next_rq changes
	 */
	if (prev != cfqq->next_rq)
		cfq_prio_tree_add(cfqd, cfqq);

745
	BUG_ON(!cfqq->next_rq);
Linus Torvalds's avatar
Linus Torvalds committed
746 747
}

748
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
749
{
750 751
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
Jens Axboe's avatar
Jens Axboe committed
752
	cfq_add_rq_rb(rq);
Linus Torvalds's avatar
Linus Torvalds committed
753 754
}

755 756
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
757
{
758
	struct task_struct *tsk = current;
759
	struct cfq_io_context *cic;
760
	struct cfq_queue *cfqq;
Linus Torvalds's avatar
Linus Torvalds committed
761

762
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
763 764 765 766
	if (!cic)
		return NULL;

	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
767 768 769
	if (cfqq) {
		sector_t sector = bio->bi_sector + bio_sectors(bio);

770
		return elv_rb_find(&cfqq->sort_list, sector);
771
	}
Linus Torvalds's avatar
Linus Torvalds committed
772 773 774 775

	return NULL;
}

776
static void cfq_activate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
777
{
778
	struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe's avatar
Jens Axboe committed
779

780
	cfqd->rq_in_driver[rq_is_sync(rq)]++;
781
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
782
						rq_in_driver(cfqd));
783

784
	cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
Linus Torvalds's avatar
Linus Torvalds committed
785 786
}

787
static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
788
{
789
	struct cfq_data *cfqd = q->elevator->elevator_data;
790
	const int sync = rq_is_sync(rq);
791

792 793
	WARN_ON(!cfqd->rq_in_driver[sync]);
	cfqd->rq_in_driver[sync]--;
794
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
795
						rq_in_driver(cfqd));
Linus Torvalds's avatar
Linus Torvalds committed
796 797
}

798
static void cfq_remove_request(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
799
{
Jens Axboe's avatar
Jens Axboe committed
800
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
801

Jens Axboe's avatar
Jens Axboe committed
802 803
	if (cfqq->next_rq == rq)
		cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds's avatar
Linus Torvalds committed
804

805
	list_del_init(&rq->queuelist);
Jens Axboe's avatar
Jens Axboe committed
806
	cfq_del_rq_rb(rq);
807

808
	cfqq->cfqd->rq_queued--;
809 810 811 812
	if (rq_is_meta(rq)) {
		WARN_ON(!cfqq->meta_pending);
		cfqq->meta_pending--;
	}
Linus Torvalds's avatar
Linus Torvalds committed
813 814
}

815 816
static int cfq_merge(struct request_queue *q, struct request **req,
		     struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
817 818 819 820
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request *__rq;

821
	__rq = cfq_find_rq_fmerge(cfqd, bio);
822
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
823 824
		*req = __rq;
		return ELEVATOR_FRONT_MERGE;
Linus Torvalds's avatar
Linus Torvalds committed
825 826 827 828 829
	}

	return ELEVATOR_NO_MERGE;
}

830
static void cfq_merged_request(struct request_queue *q, struct request *req,
831
			       int type)
Linus Torvalds's avatar
Linus Torvalds committed
832
{
833
	if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe's avatar
Jens Axboe committed
834
		struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds's avatar
Linus Torvalds committed
835

Jens Axboe's avatar
Jens Axboe committed
836
		cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds's avatar
Linus Torvalds committed
837 838 839 840
	}
}

static void
841
cfq_merged_requests(struct request_queue *q, struct request *rq,
Linus Torvalds's avatar
Linus Torvalds committed
842 843
		    struct request *next)
{
844 845 846 847
	/*
	 * reposition in fifo if next is older than rq
	 */
	if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
848
	    time_before(rq_fifo_time(next), rq_fifo_time(rq))) {
849
		list_move(&rq->queuelist, &next->queuelist);
850 851
		rq_set_fifo_time(rq, rq_fifo_time(next));
	}
852

853
	cfq_remove_request(next);
854 855
}

856
static int cfq_allow_merge(struct request_queue *q, struct request *rq,
857 858 859
			   struct bio *bio)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
860
	struct cfq_io_context *cic;
861 862 863
	struct cfq_queue *cfqq;

	/*
864
	 * Disallow merge of a sync bio into an async request.
865
	 */
866
	if (cfq_bio_sync(bio) && !rq_is_sync(rq))
867
		return false;
868 869

	/*
870 871
	 * Lookup the cfqq that this bio will be queued with. Allow
	 * merge only if rq is queued there.
872
	 */
873
	cic = cfq_cic_lookup(cfqd, current->io_context);
874
	if (!cic)
875
		return false;
876

877
	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
878
	return cfqq == RQ_CFQQ(rq);
879 880
}

881 882
static void __cfq_set_active_queue(struct cfq_data *cfqd,
				   struct cfq_queue *cfqq)
883 884
{
	if (cfqq) {
885
		cfq_log_cfqq(cfqd, cfqq, "set_active");
886
		cfqq->slice_end = 0;
887 888 889
		cfqq->slice_dispatch = 0;

		cfq_clear_cfqq_wait_request(cfqq);
890
		cfq_clear_cfqq_must_dispatch(cfqq);
Jens Axboe's avatar
Jens Axboe committed
891 892
		cfq_clear_cfqq_must_alloc_slice(cfqq);
		cfq_clear_cfqq_fifo_expire(cfqq);
893
		cfq_mark_cfqq_slice_new(cfqq);
894 895

		del_timer(&cfqd->idle_slice_timer);
896 897 898 899 900
	}

	cfqd->active_queue = cfqq;
}

901 902 903 904 905
/*
 * current cfqq expired its slice (or was too idle), select new one
 */
static void
__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
906
		    bool timed_out)
907
{
908 909
	cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);

910 911 912 913 914 915
	if (cfq_cfqq_wait_request(cfqq))
		del_timer(&cfqd->idle_slice_timer);

	cfq_clear_cfqq_wait_request(cfqq);

	/*
916
	 * store what was left of this slice, if the queue idled/timed out
917
	 */
918
	if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
919
		cfqq->slice_resid = cfqq->slice_end - jiffies;
920 921
		cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
	}
922

923
	cfq_resort_rr_list(cfqd, cfqq);
924 925 926 927 928 929 930 931 932 933

	if (cfqq == cfqd->active_queue)
		cfqd->active_queue = NULL;

	if (cfqd->active_cic) {
		put_io_context(cfqd->active_cic->ioc);
		cfqd->active_cic = NULL;
	}
}

934
static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
935 936 937 938
{
	struct cfq_queue *cfqq = cfqd->active_queue;

	if (cfqq)
939
		__cfq_slice_expired(cfqd, cfqq, timed_out);
940 941
}

942 943 944 945
/*
 * Get next queue for service. Unless we have a queue preemption,
 * we'll simply select the first cfqq in the service tree.
 */
Jens Axboe's avatar
Jens Axboe committed
946
static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
947
{
948 949
	if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
		return NULL;