cfq-iosched.c 56.6 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
#define CFQ_SLICE_SCALE		(5)

43 44
#define RQ_CIC(rq)		\
	((struct cfq_io_context *) (rq)->elevator_private)
45
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
Linus Torvalds's avatar
Linus Torvalds committed
46

47 48
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
49

50
static DEFINE_PER_CPU(unsigned long, ioc_count);
51
static struct completion *ioc_gone;
52
static DEFINE_SPINLOCK(ioc_gone_lock);
53

54 55 56 57
#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)

Jens Axboe's avatar
Jens Axboe committed
58 59 60
#define ASYNC			(0)
#define SYNC			(1)

61 62
#define sample_valid(samples)	((samples) > 80)

63 64 65 66 67 68 69 70 71 72 73 74
/*
 * 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, }

75 76 77
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
78
struct cfq_data {
79
	struct request_queue *queue;
80 81 82 83

	/*
	 * rr list of queues with requests and the count of them
	 */
84
	struct cfq_rb_root service_tree;
85 86 87
	unsigned int busy_queues;

	int rq_in_driver;
88
	int sync_flight;
89
	int hw_tag;
Linus Torvalds's avatar
Linus Torvalds committed
90

91 92 93 94 95
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
96

97 98 99
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

100 101 102 103 104
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
105

Jens Axboe's avatar
Jens Axboe committed
106
	sector_t last_position;
107
	unsigned long last_end_request;
Linus Torvalds's avatar
Linus Torvalds committed
108 109 110 111 112

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
113
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
114 115
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
116 117 118
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
119 120

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
121 122
};

123 124 125
/*
 * Per process-grouping structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
126 127 128
struct cfq_queue {
	/* reference count */
	atomic_t ref;
129 130
	/* various state flags, see below */
	unsigned int flags;
Linus Torvalds's avatar
Linus Torvalds committed
131 132
	/* parent cfq_data */
	struct cfq_data *cfqd;
133 134 135 136
	/* service_tree member */
	struct rb_node rb_node;
	/* service_tree key */
	unsigned long rb_key;
Linus Torvalds's avatar
Linus Torvalds committed
137 138 139
	/* sorted list of pending requests */
	struct rb_root sort_list;
	/* if fifo isn't expired, next request to serve */
Jens Axboe's avatar
Jens Axboe committed
140
	struct request *next_rq;
Linus Torvalds's avatar
Linus Torvalds committed
141 142 143 144 145
	/* requests queued in sort_list */
	int queued[2];
	/* currently allocated requests */
	int allocated[2];
	/* fifo list of requests in sort_list */
146
	struct list_head fifo;
Linus Torvalds's avatar
Linus Torvalds committed
147

148
	unsigned long slice_end;
149
	long slice_resid;
Linus Torvalds's avatar
Linus Torvalds committed
150

151 152
	/* pending metadata requests */
	int meta_pending;
Jens Axboe's avatar
Jens Axboe committed
153 154
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
155 156 157 158 159

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

160
	pid_t pid;
Linus Torvalds's avatar
Linus Torvalds committed
161 162
};

Jens Axboe's avatar
Jens Axboe committed
163
enum cfqq_state_flags {
164 165 166 167 168 169 170 171 172
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
	CFQ_CFQQ_FLAG_must_alloc,	/* must be allowed rq alloc */
	CFQ_CFQQ_FLAG_must_alloc_slice,	/* per-slice must_alloc flag */
	CFQ_CFQQ_FLAG_must_dispatch,	/* must dispatch, even if expired */
	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 */
	CFQ_CFQQ_FLAG_queue_new,	/* queue never been serviced */
173
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
174
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
Jens Axboe's avatar
Jens Axboe committed
175 176 177 178 179
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
180
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
181 182 183
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
184
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
185 186 187
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
188
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
189 190 191 192 193 194 195 196 197 198
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
CFQ_CFQQ_FNS(must_alloc);
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(must_dispatch);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
199
CFQ_CFQQ_FNS(queue_new);
200
CFQ_CFQQ_FNS(slice_new);
201
CFQ_CFQQ_FNS(sync);
Jens Axboe's avatar
Jens Axboe committed
202 203
#undef CFQ_CFQQ_FNS

204 205 206 207 208
#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)

209
static void cfq_dispatch_insert(struct request_queue *, struct request *);
210
static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
211
				       struct io_context *, gfp_t);
212
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
						struct io_context *);

static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
					    int is_sync)
{
	return cic->cfqq[!!is_sync];
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
				struct cfq_queue *cfqq, int is_sync)
{
	cic->cfqq[!!is_sync] = cfqq;
}

/*
 * 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).
 */
static inline int cfq_bio_sync(struct bio *bio)
{
	if (bio_data_dir(bio) == READ || bio_sync(bio))
		return 1;

	return 0;
}
Linus Torvalds's avatar
Linus Torvalds committed
238

Andrew Morton's avatar
Andrew Morton committed
239 240 241 242 243 244
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
{
245 246
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
247
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
248
	}
Andrew Morton's avatar
Andrew Morton committed
249 250
}

251
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
252 253 254
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

255
	return !cfqd->busy_queues;
Andrew Morton's avatar
Andrew Morton committed
256 257
}

258 259 260 261 262
/*
 * 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.
 */
263 264
static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync,
				 unsigned short prio)
265
{
266
	const int base_slice = cfqd->cfq_slice[sync];
267

268 269 270 271
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
272

273 274 275 276
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);
277 278 279 280 281 282
}

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;
283
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
}

/*
 * 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.
 */
static inline int cfq_slice_used(struct cfq_queue *cfqq)
{
	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
301
/*
Jens Axboe's avatar
Jens Axboe committed
302
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
303
 * We choose the request that is closest to the head right now. Distance
304
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
305
 */
Jens Axboe's avatar
Jens Axboe committed
306 307
static struct request *
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
Linus Torvalds's avatar
Linus Torvalds committed
308 309 310
{
	sector_t last, s1, s2, d1 = 0, d2 = 0;
	unsigned long back_max;
311 312 313
#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
314

Jens Axboe's avatar
Jens Axboe committed
315 316 317 318
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
319

Jens Axboe's avatar
Jens Axboe committed
320 321 322 323
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
324 325 326 327
	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
328

Jens Axboe's avatar
Jens Axboe committed
329 330
	s1 = rq1->sector;
	s2 = rq2->sector;
Linus Torvalds's avatar
Linus Torvalds committed
331

Jens Axboe's avatar
Jens Axboe committed
332
	last = cfqd->last_position;
Linus Torvalds's avatar
Linus Torvalds committed
333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348

	/*
	 * 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
349
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
350 351 352 353 354 355

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
356
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
357 358

	/* Found required data */
359 360 361 362 363 364

	/*
	 * 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
365
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
366
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
367
			return rq1;
368
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
369
			return rq2;
370 371
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
372
				return rq1;
373
			else
Jens Axboe's avatar
Jens Axboe committed
374
				return rq2;
375
		}
Linus Torvalds's avatar
Linus Torvalds committed
376

377
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
378
		return rq1;
379
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
380 381
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
382 383 384 385 386 387 388 389
	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
390
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
391
		else
Jens Axboe's avatar
Jens Axboe committed
392
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
393 394 395
	}
}

396 397 398
/*
 * The below is leftmost cache rbtree addon
 */
399
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
400 401 402 403
{
	if (!root->left)
		root->left = rb_first(&root->rb);

404 405 406 407
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
408 409 410 411 412 413 414 415 416 417 418
}

static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;

	rb_erase(n, &root->rb);
	RB_CLEAR_NODE(n);
}

Linus Torvalds's avatar
Linus Torvalds committed
419 420 421
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
422 423 424
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
425
{
426 427
	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
428
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
429

430
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
431 432

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
433
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
434

435
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
436
		next = rb_entry_rq(rbnext);
437 438 439
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
440
			next = rb_entry_rq(rbnext);
441
	}
Linus Torvalds's avatar
Linus Torvalds committed
442

443
	return cfq_choose_req(cfqd, next, prev);
Linus Torvalds's avatar
Linus Torvalds committed
444 445
}

446 447
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
448
{
449 450 451
	/*
	 * just an approximation, should be ok.
	 */
452 453
	return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
454 455
}

456 457 458 459 460
/*
 * 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.
 */
461
static void cfq_service_tree_add(struct cfq_data *cfqd,
462
				    struct cfq_queue *cfqq, int add_front)
463
{
464 465
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
466
	unsigned long rb_key;
467
	int left;
468

469 470 471 472 473 474 475 476 477
	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) {
478 479 480 481 482
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
		rb_key += cfqq->slice_resid;
		cfqq->slice_resid = 0;
	} else
		rb_key = 0;
Linus Torvalds's avatar
Linus Torvalds committed
483

484
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
485
		/*
486
		 * same position, nothing more to do
487
		 */
488 489
		if (rb_key == cfqq->rb_key)
			return;
Linus Torvalds's avatar
Linus Torvalds committed
490

491
		cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
Linus Torvalds's avatar
Linus Torvalds committed
492
	}
493

494
	left = 1;
495 496
	parent = NULL;
	p = &cfqd->service_tree.rb.rb_node;
497
	while (*p) {
498
		struct rb_node **n;
499

500 501 502
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

503 504
		/*
		 * sort RT queues first, we always want to give
505 506
		 * preference to them. IDLE queues goes to the back.
		 * after that, sort on the next service time.
507 508
		 */
		if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
509
			n = &(*p)->rb_left;
510
		else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
511 512 513 514 515
			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;
516
		else if (rb_key < __cfqq->rb_key)
517 518 519 520 521
			n = &(*p)->rb_left;
		else
			n = &(*p)->rb_right;

		if (n == &(*p)->rb_right)
522
			left = 0;
523 524

		p = n;
525 526
	}

527 528 529
	if (left)
		cfqd->service_tree.left = &cfqq->rb_node;

530 531
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
532
	rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
Linus Torvalds's avatar
Linus Torvalds committed
533 534
}

535 536 537
/*
 * Update cfqq's position in the service tree.
 */
538
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
539 540 541 542
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
543
	if (cfq_cfqq_on_rr(cfqq))
544
		cfq_service_tree_add(cfqd, cfqq, 0);
Jens Axboe's avatar
Jens Axboe committed
545 546
}

Linus Torvalds's avatar
Linus Torvalds committed
547 548
/*
 * add to busy list of queues for service, trying to be fair in ordering
549
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
550
 */
551
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
552
{
553
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
554 555
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
556 557
	cfqd->busy_queues++;

558
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
559 560
}

561 562 563 564
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
565
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
566
{
567
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
568 569
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
570

571 572
	if (!RB_EMPTY_NODE(&cfqq->rb_node))
		cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
573

Linus Torvalds's avatar
Linus Torvalds committed
574 575 576 577 578 579 580
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
}

/*
 * rb tree support functions
 */
581
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
582
{
Jens Axboe's avatar
Jens Axboe committed
583
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
584
	struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe's avatar
Jens Axboe committed
585
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
586

587 588
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
589

Jens Axboe's avatar
Jens Axboe committed
590
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
591

592
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
593
		cfq_del_cfqq_rr(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
594 595
}

Jens Axboe's avatar
Jens Axboe committed
596
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
597
{
Jens Axboe's avatar
Jens Axboe committed
598
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds's avatar
Linus Torvalds committed
599
	struct cfq_data *cfqd = cfqq->cfqd;
600
	struct request *__alias;
Linus Torvalds's avatar
Linus Torvalds committed
601

602
	cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds's avatar
Linus Torvalds committed
603 604 605 606 607

	/*
	 * looks a little odd, but the first insert might return an alias.
	 * if that happens, put the alias on the dispatch list
	 */
608
	while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe's avatar
Jens Axboe committed
609
		cfq_dispatch_insert(cfqd->queue, __alias);
610 611 612

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
613 614 615 616 617 618

	/*
	 * check if this request is a better next-serve candidate
	 */
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
	BUG_ON(!cfqq->next_rq);
Linus Torvalds's avatar
Linus Torvalds committed
619 620
}

621
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
622
{
623 624
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
Jens Axboe's avatar
Jens Axboe committed
625
	cfq_add_rq_rb(rq);
Linus Torvalds's avatar
Linus Torvalds committed
626 627
}

628 629
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
630
{
631
	struct task_struct *tsk = current;
632
	struct cfq_io_context *cic;
633
	struct cfq_queue *cfqq;
Linus Torvalds's avatar
Linus Torvalds committed
634

635
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
636 637 638 639
	if (!cic)
		return NULL;

	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
640 641 642
	if (cfqq) {
		sector_t sector = bio->bi_sector + bio_sectors(bio);

643
		return elv_rb_find(&cfqq->sort_list, sector);
644
	}
Linus Torvalds's avatar
Linus Torvalds committed
645 646 647 648

	return NULL;
}

649
static void cfq_activate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
650
{
651
	struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe's avatar
Jens Axboe committed
652

653
	cfqd->rq_in_driver++;
654 655
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
						cfqd->rq_in_driver);
656 657 658 659 660 661 662 663 664

	/*
	 * If the depth is larger 1, it really could be queueing. But lets
	 * make the mark a little higher - idling could still be good for
	 * low queueing, and a low queueing number could also just indicate
	 * a SCSI mid layer like behaviour where limit+1 is often seen.
	 */
	if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
		cfqd->hw_tag = 1;
Jens Axboe's avatar
Jens Axboe committed
665 666

	cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors;
Linus Torvalds's avatar
Linus Torvalds committed
667 668
}

669
static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
670
{
671 672 673 674
	struct cfq_data *cfqd = q->elevator->elevator_data;

	WARN_ON(!cfqd->rq_in_driver);
	cfqd->rq_in_driver--;
675 676
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
						cfqd->rq_in_driver);
Linus Torvalds's avatar
Linus Torvalds committed
677 678
}

679
static void cfq_remove_request(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
680
{
Jens Axboe's avatar
Jens Axboe committed
681
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
682

Jens Axboe's avatar
Jens Axboe committed
683 684
	if (cfqq->next_rq == rq)
		cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds's avatar
Linus Torvalds committed
685

686
	list_del_init(&rq->queuelist);
Jens Axboe's avatar
Jens Axboe committed
687
	cfq_del_rq_rb(rq);
688 689 690 691 692

	if (rq_is_meta(rq)) {
		WARN_ON(!cfqq->meta_pending);
		cfqq->meta_pending--;
	}
Linus Torvalds's avatar
Linus Torvalds committed
693 694
}

695 696
static int cfq_merge(struct request_queue *q, struct request **req,
		     struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
697 698 699 700
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request *__rq;

701
	__rq = cfq_find_rq_fmerge(cfqd, bio);
702
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
703 704
		*req = __rq;
		return ELEVATOR_FRONT_MERGE;
Linus Torvalds's avatar
Linus Torvalds committed
705 706 707 708 709
	}

	return ELEVATOR_NO_MERGE;
}

710
static void cfq_merged_request(struct request_queue *q, struct request *req,
711
			       int type)
Linus Torvalds's avatar
Linus Torvalds committed
712
{
713
	if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe's avatar
Jens Axboe committed
714
		struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds's avatar
Linus Torvalds committed
715

Jens Axboe's avatar
Jens Axboe committed
716
		cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds's avatar
Linus Torvalds committed
717 718 719 720
	}
}

static void
721
cfq_merged_requests(struct request_queue *q, struct request *rq,
Linus Torvalds's avatar
Linus Torvalds committed
722 723
		    struct request *next)
{
724 725 726 727 728 729 730
	/*
	 * reposition in fifo if next is older than rq
	 */
	if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
	    time_before(next->start_time, rq->start_time))
		list_move(&rq->queuelist, &next->queuelist);

731
	cfq_remove_request(next);
732 733
}

734
static int cfq_allow_merge(struct request_queue *q, struct request *rq,
735 736 737
			   struct bio *bio)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
738
	struct cfq_io_context *cic;
739 740 741
	struct cfq_queue *cfqq;

	/*
742
	 * Disallow merge of a sync bio into an async request.
743
	 */
744
	if (cfq_bio_sync(bio) && !rq_is_sync(rq))
745 746 747
		return 0;

	/*
748 749
	 * Lookup the cfqq that this bio will be queued with. Allow
	 * merge only if rq is queued there.
750
	 */
751
	cic = cfq_cic_lookup(cfqd, current->io_context);
752 753
	if (!cic)
		return 0;
754

755
	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
756 757
	if (cfqq == RQ_CFQQ(rq))
		return 1;
758

759
	return 0;
760 761
}

762 763
static void __cfq_set_active_queue(struct cfq_data *cfqd,
				   struct cfq_queue *cfqq)
764 765
{
	if (cfqq) {
766
		cfq_log_cfqq(cfqd, cfqq, "set_active");
767
		cfqq->slice_end = 0;
Jens Axboe's avatar
Jens Axboe committed
768 769
		cfq_clear_cfqq_must_alloc_slice(cfqq);
		cfq_clear_cfqq_fifo_expire(cfqq);
770
		cfq_mark_cfqq_slice_new(cfqq);
Jens Axboe's avatar
Jens Axboe committed
771
		cfq_clear_cfqq_queue_new(cfqq);
772 773 774 775 776
	}

	cfqd->active_queue = cfqq;
}

777 778 779 780 781
/*
 * 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,
782
		    int timed_out)
783
{
784 785
	cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);

786 787 788 789 790 791 792
	if (cfq_cfqq_wait_request(cfqq))
		del_timer(&cfqd->idle_slice_timer);

	cfq_clear_cfqq_must_dispatch(cfqq);
	cfq_clear_cfqq_wait_request(cfqq);

	/*
793
	 * store what was left of this slice, if the queue idled/timed out
794
	 */
795
	if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
796
		cfqq->slice_resid = cfqq->slice_end - jiffies;
797 798
		cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
	}
799

800
	cfq_resort_rr_list(cfqd, cfqq);
801 802 803 804 805 806 807 808 809 810

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

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

811
static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out)
812 813 814 815
{
	struct cfq_queue *cfqq = cfqd->active_queue;

	if (cfqq)
816
		__cfq_slice_expired(cfqd, cfqq, timed_out);
817 818
}

819 820 821 822
/*
 * 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
823
static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
824
{
825 826
	if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
		return NULL;
827

828
	return cfq_rb_first(&cfqd->service_tree);
Jens Axboe's avatar
Jens Axboe committed
829 830
}

831 832 833
/*
 * Get and set a new active queue for service.
 */
Jens Axboe's avatar
Jens Axboe committed
834 835 836 837
static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
{
	struct cfq_queue *cfqq;

838
	cfqq = cfq_get_next_queue(cfqd);
839
	__cfq_set_active_queue(cfqd, cfqq);
Jens Axboe's avatar
Jens Axboe committed
840
	return cfqq;
841 842
}

843 844 845 846 847 848 849 850 851
static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
					  struct request *rq)
{
	if (rq->sector >= cfqd->last_position)
		return rq->sector - cfqd->last_position;
	else
		return cfqd->last_position - rq->sector;
}

Jens Axboe's avatar
Jens Axboe committed
852 853 854 855 856 857 858 859 860 861
static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq)
{
	struct cfq_io_context *cic = cfqd->active_cic;

	if (!sample_valid(cic->seek_samples))
		return 0;

	return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean;
}

862 863
static int cfq_close_cooperator(struct cfq_data *cfq_data,
				struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
864 865
{
	/*
866 867 868
	 * We should notice if some of the queues are cooperating, eg
	 * working closely on the same area of the disk. In that case,
	 * we can group them together and don't waste time idling.
Jens Axboe's avatar
Jens Axboe committed
869
	 */
870
	return 0;
Jens Axboe's avatar
Jens Axboe committed
871 872 873
}

#define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024))
874

Jens Axboe's avatar
Jens Axboe committed
875
static void cfq_arm_slice_timer(struct cfq_data *cfqd)
876
{
877
	struct cfq_queue *cfqq = cfqd->active_queue;
878
	struct cfq_io_context *cic;
879 880
	unsigned long sl;

881
	WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
Jens Axboe's avatar
Jens Axboe committed
882
	WARN_ON(cfq_cfqq_slice_new(cfqq));
883 884 885 886

	/*
	 * idle is disabled, either manually or by past process history
	 */
Jens Axboe's avatar
Jens Axboe committed
887 888 889
	if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq))
		return;

890 891 892 893 894 895
	/*
	 * still requests with the driver, don't idle
	 */
	if (cfqd->rq_in_driver)
		return;

896 897 898
	/*
	 * task has exited, don't wait
	 */
899
	cic = cfqd->active_cic;
900
	if (!cic || !atomic_read(&cic->ioc->nr_tasks))
Jens Axboe's avatar
Jens Axboe committed
901 902 903 904 905
		return;

	/*
	 * See if this prio level has a good candidate
	 */
Jens Axboe's avatar
Jens Axboe committed
906 907
	if (cfq_close_cooperator(cfqd, cfqq) &&
	    (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2))
Jens Axboe's avatar
Jens Axboe committed
908
		return;
909

Jens Axboe's avatar
Jens Axboe committed
910 911
	cfq_mark_cfqq_must_dispatch(cfqq);
	cfq_mark_cfqq_wait_request(cfqq);
912

913 914 915 916 917
	/*
	 * we don't want to idle for seeks, but we do want to allow
	 * fair distribution of slice time for a process doing back-to-back
	 * seeks. so allow a little bit of time for him to submit a new rq
	 */
Jens Axboe's avatar
Jens Axboe committed
918
	sl = cfqd->cfq_slice_idle;
919
	if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
920
		sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
921

922
	mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
923
	cfq_log(cfqd, "arm_idle: %lu", sl);
Linus Torvalds's avatar
Linus Torvalds committed
924 925
}

926 927 928
/*
 * Move request from internal lists to the request queue dispatch list.
 */
929
static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
930
{
931
	struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe's avatar
Jens Axboe committed
932
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
933

934 935
	cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");

936
	cfq_remove_request(rq);
Jens Axboe's avatar
Jens Axboe committed
937
	cfqq->dispatched++;
938
	elv_dispatch_sort(q, rq);
939 940 941

	if (cfq_cfqq_sync(cfqq))
		cfqd->sync_flight++;
Linus Torvalds's avatar
Linus Torvalds committed
942 943 944 945 946
}

/*
 * return expired entry, or NULL to just start from scratch in rbtree
 */
947
static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
948 949
{
	struct cfq_data *cfqd = cfqq->cfqd;
950
	struct request *rq;
951
	int fifo;
Linus Torvalds's avatar
Linus Torvalds committed
952

Jens Axboe's avatar
Jens Axboe committed
953
	if (cfq_cfqq_fifo_expire(cfqq))
Linus Torvalds's avatar
Linus Torvalds committed
954
		return NULL;
955 956 957

	cfq_mark_cfqq_fifo_expire(cfqq);

958 959
	if (list_empty(&cfqq->fifo))
		return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
960

Jens Axboe's avatar
Jens Axboe committed
961
	fifo = cfq_cfqq_sync(cfqq);
962
	rq = rq_entry_fifo(cfqq->fifo.next);
Linus Torvalds's avatar
Linus Torvalds committed
963