cfq-iosched.c 71.3 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
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
32

33
/*
34
 * offset from end of service tree
35
 */
36
#define CFQ_IDLE_DELAY		(HZ / 5)
37 38 39 40 41 42

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

43 44 45 46 47 48
/*
 * Allow merged cfqqs to perform this amount of seeky I/O before
 * deciding to break the queues up again.
 */
#define CFQQ_COOP_TOUT		(HZ)

49
#define CFQ_SLICE_SCALE		(5)
50
#define CFQ_HW_QUEUE_MIN	(5)
51

52 53
#define RQ_CIC(rq)		\
	((struct cfq_io_context *) (rq)->elevator_private)
54
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
Linus Torvalds's avatar
Linus Torvalds committed
55

56 57
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
58

59
static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
60
static struct completion *ioc_gone;
61
static DEFINE_SPINLOCK(ioc_gone_lock);
62

63 64 65 66
#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)

67 68
#define sample_valid(samples)	((samples) > 80)

69 70 71 72 73 74 75 76 77
/*
 * 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;
78
	unsigned count;
79
};
80
#define CFQ_RB_ROOT	(struct cfq_rb_root) { RB_ROOT, NULL, 0, }
81

82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123
/*
 * 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;

124 125 126 127
	unsigned int seek_samples;
	u64 seek_total;
	sector_t seek_mean;
	sector_t last_request_pos;
128
	unsigned long seeky_start;
129

130
	pid_t pid;
Jeff Moyer's avatar
Jeff Moyer committed
131

132
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
133
	struct cfq_queue *new_cfqq;
134 135
};

136 137 138
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
139
struct cfq_data {
140
	struct request_queue *queue;
141 142 143 144

	/*
	 * rr list of queues with requests and the count of them
	 */
145
	struct cfq_rb_root service_tree;
146 147 148 149 150 151 152 153

	/*
	 * 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];

154
	unsigned int busy_queues;
155 156
	unsigned int busy_rt_queues;
	unsigned int busy_queues_avg[2];
157

158
	int rq_in_driver[2];
159
	int sync_flight;
160 161 162 163 164

	/*
	 * queue-depth detection
	 */
	int rq_queued;
165
	int hw_tag;
166 167
	int hw_tag_samples;
	int rq_in_driver_peak;
Linus Torvalds's avatar
Linus Torvalds committed
168

169 170 171 172
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
173
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
174

175 176 177
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

178 179 180 181 182
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
183

Jens Axboe's avatar
Jens Axboe committed
184
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
185 186 187 188 189

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
190
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
191 192
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
193 194 195
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
196
	unsigned int cfq_latency;
197 198

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
199

200 201 202 203
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
204 205

	unsigned long last_end_sync_rq;
Linus Torvalds's avatar
Linus Torvalds committed
206 207
};

Jens Axboe's avatar
Jens Axboe committed
208
enum cfqq_state_flags {
209 210
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
211
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
212 213 214 215
	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 */
216
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
217
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
218
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
Jens Axboe's avatar
Jens Axboe committed
219 220 221 222 223
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
224
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
225 226 227
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
228
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
229 230 231
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
232
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
233 234 235 236
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
237
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
238 239 240 241
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
242
CFQ_CFQQ_FNS(slice_new);
243
CFQ_CFQQ_FNS(sync);
244
CFQ_CFQQ_FNS(coop);
Jens Axboe's avatar
Jens Axboe committed
245 246
#undef CFQ_CFQQ_FNS

247 248 249 250 251
#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)

252
static void cfq_dispatch_insert(struct request_queue *, struct request *);
253
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
254
				       struct io_context *, gfp_t);
255
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
256 257
						struct io_context *);

258 259 260 261 262
static inline int rq_in_driver(struct cfq_data *cfqd)
{
	return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
}

263
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
264
					    bool is_sync)
265
{
266
	return cic->cfqq[is_sync];
267 268 269
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
270
				struct cfq_queue *cfqq, bool is_sync)
271
{
272
	cic->cfqq[is_sync] = cfqq;
273 274 275 276 277 278
}

/*
 * 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).
 */
279
static inline bool cfq_bio_sync(struct bio *bio)
280
{
281
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
282
}
Linus Torvalds's avatar
Linus Torvalds committed
283

Andrew Morton's avatar
Andrew Morton committed
284 285 286 287
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
288
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
289
{
290 291
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
292
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
293
	}
Andrew Morton's avatar
Andrew Morton committed
294 295
}

296
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
297 298 299
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

300
	return !cfqd->busy_queues;
Andrew Morton's avatar
Andrew Morton committed
301 302
}

303 304 305 306 307
/*
 * 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.
 */
308
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
309
				 unsigned short prio)
310
{
311
	const int base_slice = cfqd->cfq_slice[sync];
312

313 314 315 316
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
317

318 319 320 321
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);
322 323
}

324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
/*
 * get averaged number of queues of RT/BE priority.
 * average is updated, with a formula that gives more weight to higher numbers,
 * to quickly follows sudden increases and decrease slowly
 */

static inline unsigned
cfq_get_avg_queues(struct cfq_data *cfqd, bool rt) {
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
	unsigned busy = cfqd->busy_rt_queues;

	if (!rt)
		busy = cfqd->busy_queues - cfqd->busy_rt_queues;

	min_q = min(cfqd->busy_queues_avg[rt], busy);
	max_q = max(cfqd->busy_queues_avg[rt], busy);
	cfqd->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
		cfq_hist_divisor;
	return cfqd->busy_queues_avg[rt];
}

347 348 349
static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
		/* interested queues (we consider only the ones with the same
		 * priority class) */
		unsigned iq = cfq_get_avg_queues(cfqd, cfq_class_rt(cfqq));
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
		if (expect_latency > cfq_target_latency) {
			unsigned base_low_slice = 2 * cfqd->cfq_slice_idle;
			/* scale low_slice according to IO priority
			 * and sync vs async */
			unsigned low_slice =
				min(slice, base_low_slice * slice / sync_slice);
			/* the adapted slice value is scaled to fit all iqs
			 * into the target latency */
			slice = max(slice * cfq_target_latency / expect_latency,
				    low_slice);
		}
	}
	cfqq->slice_end = jiffies + slice;
370
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
371 372 373 374 375 376 377
}

/*
 * 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.
 */
378
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
379 380 381 382 383 384 385 386 387
{
	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
388
/*
Jens Axboe's avatar
Jens Axboe committed
389
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
390
 * We choose the request that is closest to the head right now. Distance
391
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
392
 */
Jens Axboe's avatar
Jens Axboe committed
393 394
static struct request *
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
Linus Torvalds's avatar
Linus Torvalds committed
395 396 397
{
	sector_t last, s1, s2, d1 = 0, d2 = 0;
	unsigned long back_max;
398 399 400
#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
401

Jens Axboe's avatar
Jens Axboe committed
402 403 404 405
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
406

Jens Axboe's avatar
Jens Axboe committed
407 408 409 410
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
411 412 413 414
	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
415

416 417
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
418

Jens Axboe's avatar
Jens Axboe committed
419
	last = cfqd->last_position;
Linus Torvalds's avatar
Linus Torvalds committed
420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435

	/*
	 * 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
436
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
437 438 439 440 441 442

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
443
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
444 445

	/* Found required data */
446 447 448 449 450 451

	/*
	 * 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
452
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
453
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
454
			return rq1;
455
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
456
			return rq2;
457 458
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
459
				return rq1;
460
			else
Jens Axboe's avatar
Jens Axboe committed
461
				return rq2;
462
		}
Linus Torvalds's avatar
Linus Torvalds committed
463

464
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
465
		return rq1;
466
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
467 468
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
469 470 471 472 473 474 475 476
	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
477
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
478
		else
Jens Axboe's avatar
Jens Axboe committed
479
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
480 481 482
	}
}

483 484 485
/*
 * The below is leftmost cache rbtree addon
 */
486
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
487 488 489 490
{
	if (!root->left)
		root->left = rb_first(&root->rb);

491 492 493 494
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
495 496
}

497 498 499 500 501 502
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

503 504 505 506
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
507
	rb_erase_init(n, &root->rb);
508
	--root->count;
509 510
}

Linus Torvalds's avatar
Linus Torvalds committed
511 512 513
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
514 515 516
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
517
{
518 519
	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
520
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
521

522
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
523 524

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
525
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
526

527
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
528
		next = rb_entry_rq(rbnext);
529 530 531
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
532
			next = rb_entry_rq(rbnext);
533
	}
Linus Torvalds's avatar
Linus Torvalds committed
534

535
	return cfq_choose_req(cfqd, next, prev);
Linus Torvalds's avatar
Linus Torvalds committed
536 537
}

538 539
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
540
{
541 542 543
	/*
	 * just an approximation, should be ok.
	 */
544 545
	return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
546 547
}

548 549 550 551 552
/*
 * 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.
 */
553
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
554
				 bool add_front)
555
{
556 557
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
558
	unsigned long rb_key;
559
	struct cfq_rb_root *service_tree = &cfqd->service_tree;
560
	int left;
561

562 563
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
564
		parent = rb_last(&service_tree->rb);
565 566 567 568 569 570
		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) {
571 572 573 574 575 576
		/*
		 * 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.
		 */
577
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
578
		rb_key -= cfqq->slice_resid;
579
		cfqq->slice_resid = 0;
580 581
	} else {
		rb_key = -HZ;
582
		__cfqq = cfq_rb_first(service_tree);
583 584
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
Linus Torvalds's avatar
Linus Torvalds committed
585

586
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
587
		/*
588
		 * same position, nothing more to do
589
		 */
590 591
		if (rb_key == cfqq->rb_key)
			return;
Linus Torvalds's avatar
Linus Torvalds committed
592

593 594
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
595
	}
596

597
	left = 1;
598
	parent = NULL;
599 600
	cfqq->service_tree = service_tree;
	p = &service_tree->rb.rb_node;
601
	while (*p) {
602
		struct rb_node **n;
603

604 605 606
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

607 608
		/*
		 * sort RT queues first, we always want to give
609 610
		 * preference to them. IDLE queues goes to the back.
		 * after that, sort on the next service time.
611 612
		 */
		if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
613
			n = &(*p)->rb_left;
614
		else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
615 616 617 618 619
			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;
620
		else if (time_before(rb_key, __cfqq->rb_key))
621 622 623 624 625
			n = &(*p)->rb_left;
		else
			n = &(*p)->rb_right;

		if (n == &(*p)->rb_right)
626
			left = 0;
627 628

		p = n;
629 630
	}

631
	if (left)
632
		service_tree->left = &cfqq->rb_node;
633

634 635
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
636 637
	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
	service_tree->count++;
Linus Torvalds's avatar
Linus Torvalds committed
638 639
}

640
static struct cfq_queue *
641 642 643
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)
644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659
{
	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.
		 */
660
		if (sector > blk_rq_pos(cfqq->next_rq))
661
			n = &(*p)->rb_right;
662
		else if (sector < blk_rq_pos(cfqq->next_rq))
663 664 665 666
			n = &(*p)->rb_left;
		else
			break;
		p = n;
667
		cfqq = NULL;
668 669 670 671 672
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
673
	return cfqq;
674 675 676 677 678 679 680
}

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

681 682 683 684
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
685 686 687 688 689 690

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

691
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
692 693
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
694 695
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
696 697 698
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
699 700
}

701 702 703
/*
 * Update cfqq's position in the service tree.
 */
704
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
705 706 707 708
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
709
	if (cfq_cfqq_on_rr(cfqq)) {
710
		cfq_service_tree_add(cfqd, cfqq, 0);
711 712
		cfq_prio_tree_add(cfqd, cfqq);
	}
Jens Axboe's avatar
Jens Axboe committed
713 714
}

Linus Torvalds's avatar
Linus Torvalds committed
715 716
/*
 * add to busy list of queues for service, trying to be fair in ordering
717
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
718
 */
719
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
720
{
721
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
722 723
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
724
	cfqd->busy_queues++;
725 726
	if (cfq_class_rt(cfqq))
		cfqd->busy_rt_queues++;
727
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
728 729
}

730 731 732 733
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
734
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
735
{
736
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
737 738
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
739

740 741 742 743
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
	}
744 745 746 747
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
748

Linus Torvalds's avatar
Linus Torvalds committed
749 750
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
751 752
	if (cfq_class_rt(cfqq))
		cfqd->busy_rt_queues--;
Linus Torvalds's avatar
Linus Torvalds committed
753 754 755 756 757
}

/*
 * rb tree support functions
 */
758
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
759
{
Jens Axboe's avatar
Jens Axboe committed
760
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
761
	struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe's avatar
Jens Axboe committed
762
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
763

764 765
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
766

Jens Axboe's avatar
Jens Axboe committed
767
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
768

769
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
770
		cfq_del_cfqq_rr(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
771 772
}

Jens Axboe's avatar
Jens Axboe committed
773
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
774
{
Jens Axboe's avatar
Jens Axboe committed
775
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds's avatar
Linus Torvalds committed
776
	struct cfq_data *cfqd = cfqq->cfqd;
777
	struct request *__alias, *prev;
Linus Torvalds's avatar
Linus Torvalds committed
778

779
	cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds's avatar
Linus Torvalds committed
780 781 782 783 784

	/*
	 * looks a little odd, but the first insert might return an alias.
	 * if that happens, put the alias on the dispatch list
	 */
785
	while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe's avatar
Jens Axboe committed
786
		cfq_dispatch_insert(cfqd->queue, __alias);
787 788 789

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
790 791 792 793

	/*
	 * check if this request is a better next-serve candidate
	 */
794
	prev = cfqq->next_rq;
795
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
796 797 798 799 800 801 802

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

803
	BUG_ON(!cfqq->next_rq);
Linus Torvalds's avatar
Linus Torvalds committed
804 805
}

806
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
807
{
808 809
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
Jens Axboe's avatar
Jens Axboe committed
810
	cfq_add_rq_rb(rq);
Linus Torvalds's avatar
Linus Torvalds committed
811 812
}

813 814
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
815
{
816
	struct task_struct *tsk = current;
817
	struct cfq_io_context *cic;
818
	struct cfq_queue *cfqq;
Linus Torvalds's avatar
Linus Torvalds committed
819

820
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
821 822 823 824
	if (!cic)
		return NULL;

	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
825 826 827
	if (cfqq) {
		sector_t sector = bio->bi_sector + bio_sectors(bio);

828
		return elv_rb_find(&cfqq->sort_list, sector);
829
	}
Linus Torvalds's avatar
Linus Torvalds committed
830 831 832 833

	return NULL;
}

834
static void cfq_activate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
835
{
836
	struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe's avatar
Jens Axboe committed
837

838
	cfqd->rq_in_driver[rq_is_sync(rq)]++;
839
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
840
						rq_in_driver(cfqd));
841

842
	cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
Linus Torvalds's avatar
Linus Torvalds committed
843 844
}

845
static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
846
{
847
	struct cfq_data *cfqd = q->elevator->elevator_data;
848
	const int sync = rq_is_sync(rq);
849

850 851
	WARN_ON(!cfqd->rq_in_driver[sync]);
	cfqd->rq_in_driver[sync]--;
852
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
853
						rq_in_driver(cfqd));
Linus Torvalds's avatar
Linus Torvalds committed
854 855
}

856
static void cfq_remove_request(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
857
{
Jens Axboe's avatar
Jens Axboe committed
858
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
859

Jens Axboe's avatar
Jens Axboe committed
860 861
	if (cfqq->next_rq == rq)
		cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds's avatar
Linus Torvalds committed
862

863
	list_del_init(&rq->queuelist);
Jens Axboe's avatar
Jens Axboe committed
864
	cfq_del_rq_rb(rq);
865

866
	cfqq->cfqd->rq_queued--;
867 868 869 870
	if (rq_is_meta(rq)) {
		WARN_ON(!cfqq->meta_pending);
		cfqq->meta_pending--;
	}
Linus Torvalds's avatar
Linus Torvalds committed
871 872
}

873 874
static int cfq_merge(struct request_queue *q, struct request **req,
		     struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
875 876 877 878
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request *__rq;

879
	__rq = cfq_find_rq_fmerge(cfqd, bio);
880
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
881 882
		*req = __rq;
		return ELEVATOR_FRONT_MERGE;
Linus Torvalds's avatar
Linus Torvalds committed
883 884 885 886 887
	}

	return ELEVATOR_NO_MERGE;
}

888
static void cfq_merged_request(struct request_queue *q, struct request *req,
889
			       int type)
Linus Torvalds's avatar
Linus Torvalds committed
890
{
891
	if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe's avatar
Jens Axboe committed
892
		struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds's avatar
Linus Torvalds committed
893

Jens Axboe's avatar
Jens Axboe committed
894
		cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds's avatar
Linus Torvalds committed
895 896 897 898
	}
}

static void
899
cfq_merged_requests(struct request_queue *q, struct request *rq,
Linus Torvalds's avatar
Linus Torvalds committed
900 901
		    struct request *next)
{
902 903 904 905
	/*
	 * reposition in fifo if next is older than rq
	 */
	if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
906
	    time_before(rq_fifo_time(next), rq_fifo_time(rq))) {
907
		list_move(&rq->queuelist, &next->queuelist);
908 909
		rq_set_fifo_time(rq, rq_fifo_time(next));
	}
910

911
	cfq_remove_request(next);
912 913
}

914
static int cfq_allow_merge(struct request_queue *q, struct request *rq,
915 916 917
			   struct bio *bio)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
918
	struct cfq_io_context *cic;
919 920 921
	struct cfq_queue *cfqq;

	/*
922
	 * Disallow merge of a sync bio into an async request.
923
	 */
924
	if (cfq_bio_sync(bio) && !rq_is_sync(rq))
925
		return false;
926 927

	/*
928 929
	 * Lookup the cfqq that this bio will be queued with. Allow
	 * merge only if rq is queued there.
930
	 */
931
	cic = cfq_cic_lookup(cfqd, current->io_context);
932
	if (!cic)
933
		return false;
934

935
	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
936
	return cfqq == RQ_CFQQ(rq);
937 938
}

939 940
static void __cfq_set_active_queue(struct cfq_data *cfqd,
				   struct cfq_queue *cfqq)
941 942
{
	if (cfqq) {
943
		cfq_log_cfqq(cfqd, cfqq, "set_active");
944
		cfqq->slice_end = 0;
945 946 947
		cfqq->slice_dispatch = 0;

		cfq_clear_cfqq_wait_request(cfqq);