cfq-iosched.c 58.1 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
#define CFQ_SLICE_SCALE		(5)
42
#define CFQ_HW_QUEUE_MIN	(5)
43

44 45
#define RQ_CIC(rq)		\
	((struct cfq_io_context *) (rq)->elevator_private)
46
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
Linus Torvalds's avatar
Linus Torvalds committed
47

48 49
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
50

51
static DEFINE_PER_CPU(unsigned long, ioc_count);
52
static struct completion *ioc_gone;
53
static DEFINE_SPINLOCK(ioc_gone_lock);
54

55 56 57 58
#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
59 60 61
#define ASYNC			(0)
#define SYNC			(1)

62 63
#define sample_valid(samples)	((samples) > 80)

64 65 66 67 68 69 70 71 72 73 74 75
/*
 * 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, }

76 77 78
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
79
struct cfq_data {
80
	struct request_queue *queue;
81 82 83 84

	/*
	 * rr list of queues with requests and the count of them
	 */
85
	struct cfq_rb_root service_tree;
86
	unsigned int busy_queues;
87 88 89 90 91
	/*
	 * Used to track any pending rt requests so we can pre-empt current
	 * non-RT cfqq in service when this value is non-zero.
	 */
	unsigned int busy_rt_queues;
92 93

	int rq_in_driver;
94
	int sync_flight;
95 96 97 98 99

	/*
	 * queue-depth detection
	 */
	int rq_queued;
100
	int hw_tag;
101 102
	int hw_tag_samples;
	int rq_in_driver_peak;
Linus Torvalds's avatar
Linus Torvalds committed
103

104 105 106 107 108
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
109

110 111 112
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

113 114 115 116 117
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
118

Jens Axboe's avatar
Jens Axboe committed
119
	sector_t last_position;
120
	unsigned long last_end_request;
Linus Torvalds's avatar
Linus Torvalds committed
121 122 123 124 125

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
126
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
127 128
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
129 130 131
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
132 133

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
134 135
};

136 137 138
/*
 * Per process-grouping structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
139 140 141
struct cfq_queue {
	/* reference count */
	atomic_t ref;
142 143
	/* various state flags, see below */
	unsigned int flags;
Linus Torvalds's avatar
Linus Torvalds committed
144 145
	/* parent cfq_data */
	struct cfq_data *cfqd;
146 147 148 149
	/* service_tree member */
	struct rb_node rb_node;
	/* service_tree key */
	unsigned long rb_key;
Linus Torvalds's avatar
Linus Torvalds committed
150 151 152
	/* 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
153
	struct request *next_rq;
Linus Torvalds's avatar
Linus Torvalds committed
154 155 156 157 158
	/* requests queued in sort_list */
	int queued[2];
	/* currently allocated requests */
	int allocated[2];
	/* fifo list of requests in sort_list */
159
	struct list_head fifo;
Linus Torvalds's avatar
Linus Torvalds committed
160

161
	unsigned long slice_end;
162
	long slice_resid;
163
	unsigned int slice_dispatch;
Linus Torvalds's avatar
Linus Torvalds committed
164

165 166
	/* pending metadata requests */
	int meta_pending;
Jens Axboe's avatar
Jens Axboe committed
167 168
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
169 170 171 172 173

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

174
	pid_t pid;
Linus Torvalds's avatar
Linus Torvalds committed
175 176
};

Jens Axboe's avatar
Jens Axboe committed
177
enum cfqq_state_flags {
178 179 180 181 182 183 184
	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_fifo_expire,	/* FIFO checked in this slice */
	CFQ_CFQQ_FLAG_idle_window,	/* slice idling enabled */
	CFQ_CFQQ_FLAG_prio_changed,	/* task priority has changed */
185
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
186
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
Jens Axboe's avatar
Jens Axboe committed
187 188 189 190 191
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
192
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
193 194 195
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
196
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
197 198 199
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
200
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
201 202 203 204 205 206 207 208 209
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
CFQ_CFQQ_FNS(must_alloc);
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
210
CFQ_CFQQ_FNS(slice_new);
211
CFQ_CFQQ_FNS(sync);
Jens Axboe's avatar
Jens Axboe committed
212 213
#undef CFQ_CFQQ_FNS

214 215 216 217 218
#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)

219
static void cfq_dispatch_insert(struct request_queue *, struct request *);
220
static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
221
				       struct io_context *, gfp_t);
222
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
						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
248

Andrew Morton's avatar
Andrew Morton committed
249 250 251 252 253 254
/*
 * 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)
{
255 256
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
257
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
258
	}
Andrew Morton's avatar
Andrew Morton committed
259 260
}

261
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
262 263 264
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

265
	return !cfqd->busy_queues;
Andrew Morton's avatar
Andrew Morton committed
266 267
}

268 269 270 271 272
/*
 * 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.
 */
273 274
static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync,
				 unsigned short prio)
275
{
276
	const int base_slice = cfqd->cfq_slice[sync];
277

278 279 280 281
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
282

283 284 285 286
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);
287 288 289 290 291 292
}

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;
293
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310
}

/*
 * 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
311
/*
Jens Axboe's avatar
Jens Axboe committed
312
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
313
 * We choose the request that is closest to the head right now. Distance
314
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
315
 */
Jens Axboe's avatar
Jens Axboe committed
316 317
static struct request *
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
Linus Torvalds's avatar
Linus Torvalds committed
318 319 320
{
	sector_t last, s1, s2, d1 = 0, d2 = 0;
	unsigned long back_max;
321 322 323
#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
324

Jens Axboe's avatar
Jens Axboe committed
325 326 327 328
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
329

Jens Axboe's avatar
Jens Axboe committed
330 331 332 333
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
334 335 336 337
	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
338

Jens Axboe's avatar
Jens Axboe committed
339 340
	s1 = rq1->sector;
	s2 = rq2->sector;
Linus Torvalds's avatar
Linus Torvalds committed
341

Jens Axboe's avatar
Jens Axboe committed
342
	last = cfqd->last_position;
Linus Torvalds's avatar
Linus Torvalds committed
343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358

	/*
	 * 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
359
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
360 361 362 363 364 365

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
366
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
367 368

	/* Found required data */
369 370 371 372 373 374

	/*
	 * 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
375
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
376
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
377
			return rq1;
378
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
379
			return rq2;
380 381
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
382
				return rq1;
383
			else
Jens Axboe's avatar
Jens Axboe committed
384
				return rq2;
385
		}
Linus Torvalds's avatar
Linus Torvalds committed
386

387
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
388
		return rq1;
389
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
390 391
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
392 393 394 395 396 397 398 399
	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
400
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
401
		else
Jens Axboe's avatar
Jens Axboe committed
402
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
403 404 405
	}
}

406 407 408
/*
 * The below is leftmost cache rbtree addon
 */
409
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
410 411 412 413
{
	if (!root->left)
		root->left = rb_first(&root->rb);

414 415 416 417
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
418 419 420 421 422 423 424 425 426 427 428
}

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
429 430 431
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
432 433 434
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
435
{
436 437
	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
438
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
439

440
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
441 442

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
443
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
444

445
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
446
		next = rb_entry_rq(rbnext);
447 448 449
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
450
			next = rb_entry_rq(rbnext);
451
	}
Linus Torvalds's avatar
Linus Torvalds committed
452

453
	return cfq_choose_req(cfqd, next, prev);
Linus Torvalds's avatar
Linus Torvalds committed
454 455
}

456 457
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
458
{
459 460 461
	/*
	 * just an approximation, should be ok.
	 */
462 463
	return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
464 465
}

466 467 468 469 470
/*
 * 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.
 */
471
static void cfq_service_tree_add(struct cfq_data *cfqd,
472
				    struct cfq_queue *cfqq, int add_front)
473
{
474 475
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
476
	unsigned long rb_key;
477
	int left;
478

479 480 481 482 483 484 485 486 487
	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) {
488 489 490 491 492
		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
493

494
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
495
		/*
496
		 * same position, nothing more to do
497
		 */
498 499
		if (rb_key == cfqq->rb_key)
			return;
Linus Torvalds's avatar
Linus Torvalds committed
500

501
		cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
Linus Torvalds's avatar
Linus Torvalds committed
502
	}
503

504
	left = 1;
505 506
	parent = NULL;
	p = &cfqd->service_tree.rb.rb_node;
507
	while (*p) {
508
		struct rb_node **n;
509

510 511 512
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

513 514
		/*
		 * sort RT queues first, we always want to give
515 516
		 * preference to them. IDLE queues goes to the back.
		 * after that, sort on the next service time.
517 518
		 */
		if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
519
			n = &(*p)->rb_left;
520
		else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
521 522 523 524 525
			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;
526
		else if (rb_key < __cfqq->rb_key)
527 528 529 530 531
			n = &(*p)->rb_left;
		else
			n = &(*p)->rb_right;

		if (n == &(*p)->rb_right)
532
			left = 0;
533 534

		p = n;
535 536
	}

537 538 539
	if (left)
		cfqd->service_tree.left = &cfqq->rb_node;

540 541
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
542
	rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
Linus Torvalds's avatar
Linus Torvalds committed
543 544
}

545 546 547
/*
 * Update cfqq's position in the service tree.
 */
548
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
549 550 551 552
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
553
	if (cfq_cfqq_on_rr(cfqq))
554
		cfq_service_tree_add(cfqd, cfqq, 0);
Jens Axboe's avatar
Jens Axboe committed
555 556
}

Linus Torvalds's avatar
Linus Torvalds committed
557 558
/*
 * add to busy list of queues for service, trying to be fair in ordering
559
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
560
 */
561
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
562
{
563
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
564 565
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
566
	cfqd->busy_queues++;
567 568
	if (cfq_class_rt(cfqq))
		cfqd->busy_rt_queues++;
Linus Torvalds's avatar
Linus Torvalds committed
569

570
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
571 572
}

573 574 575 576
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
577
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
578
{
579
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
580 581
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
582

583 584
	if (!RB_EMPTY_NODE(&cfqq->rb_node))
		cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
585

Linus Torvalds's avatar
Linus Torvalds committed
586 587
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
588 589
	if (cfq_class_rt(cfqq))
		cfqd->busy_rt_queues--;
Linus Torvalds's avatar
Linus Torvalds committed
590 591 592 593 594
}

/*
 * rb tree support functions
 */
595
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
596
{
Jens Axboe's avatar
Jens Axboe committed
597
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
598
	struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe's avatar
Jens Axboe committed
599
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
600

601 602
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
603

Jens Axboe's avatar
Jens Axboe committed
604
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
605

606
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
607
		cfq_del_cfqq_rr(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
608 609
}

Jens Axboe's avatar
Jens Axboe committed
610
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
611
{
Jens Axboe's avatar
Jens Axboe committed
612
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds's avatar
Linus Torvalds committed
613
	struct cfq_data *cfqd = cfqq->cfqd;
614
	struct request *__alias;
Linus Torvalds's avatar
Linus Torvalds committed
615

616
	cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds's avatar
Linus Torvalds committed
617 618 619 620 621

	/*
	 * looks a little odd, but the first insert might return an alias.
	 * if that happens, put the alias on the dispatch list
	 */
622
	while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe's avatar
Jens Axboe committed
623
		cfq_dispatch_insert(cfqd->queue, __alias);
624 625 626

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
627 628 629 630 631 632

	/*
	 * 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
633 634
}

635
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
636
{
637 638
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
Jens Axboe's avatar
Jens Axboe committed
639
	cfq_add_rq_rb(rq);
Linus Torvalds's avatar
Linus Torvalds committed
640 641
}

642 643
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
644
{
645
	struct task_struct *tsk = current;
646
	struct cfq_io_context *cic;
647
	struct cfq_queue *cfqq;
Linus Torvalds's avatar
Linus Torvalds committed
648

649
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
650 651 652 653
	if (!cic)
		return NULL;

	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
654 655 656
	if (cfqq) {
		sector_t sector = bio->bi_sector + bio_sectors(bio);

657
		return elv_rb_find(&cfqq->sort_list, sector);
658
	}
Linus Torvalds's avatar
Linus Torvalds committed
659 660 661 662

	return NULL;
}

663
static void cfq_activate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
664
{
665
	struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe's avatar
Jens Axboe committed
666

667
	cfqd->rq_in_driver++;
668 669
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
						cfqd->rq_in_driver);
670

Jens Axboe's avatar
Jens Axboe committed
671
	cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors;
Linus Torvalds's avatar
Linus Torvalds committed
672 673
}

674
static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
675
{
676 677 678 679
	struct cfq_data *cfqd = q->elevator->elevator_data;

	WARN_ON(!cfqd->rq_in_driver);
	cfqd->rq_in_driver--;
680 681
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
						cfqd->rq_in_driver);
Linus Torvalds's avatar
Linus Torvalds committed
682 683
}

684
static void cfq_remove_request(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
685
{
Jens Axboe's avatar
Jens Axboe committed
686
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
687

Jens Axboe's avatar
Jens Axboe committed
688 689
	if (cfqq->next_rq == rq)
		cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds's avatar
Linus Torvalds committed
690

691
	list_del_init(&rq->queuelist);
Jens Axboe's avatar
Jens Axboe committed
692
	cfq_del_rq_rb(rq);
693

694
	cfqq->cfqd->rq_queued--;
695 696 697 698
	if (rq_is_meta(rq)) {
		WARN_ON(!cfqq->meta_pending);
		cfqq->meta_pending--;
	}
Linus Torvalds's avatar
Linus Torvalds committed
699 700
}

701 702
static int cfq_merge(struct request_queue *q, struct request **req,
		     struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
703 704 705 706
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request *__rq;

707
	__rq = cfq_find_rq_fmerge(cfqd, bio);
708
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
709 710
		*req = __rq;
		return ELEVATOR_FRONT_MERGE;
Linus Torvalds's avatar
Linus Torvalds committed
711 712 713 714 715
	}

	return ELEVATOR_NO_MERGE;
}

716
static void cfq_merged_request(struct request_queue *q, struct request *req,
717
			       int type)
Linus Torvalds's avatar
Linus Torvalds committed
718
{
719
	if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe's avatar
Jens Axboe committed
720
		struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds's avatar
Linus Torvalds committed
721

Jens Axboe's avatar
Jens Axboe committed
722
		cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds's avatar
Linus Torvalds committed
723 724 725 726
	}
}

static void
727
cfq_merged_requests(struct request_queue *q, struct request *rq,
Linus Torvalds's avatar
Linus Torvalds committed
728 729
		    struct request *next)
{
730 731 732 733 734 735 736
	/*
	 * 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);

737
	cfq_remove_request(next);
738 739
}

740
static int cfq_allow_merge(struct request_queue *q, struct request *rq,
741 742 743
			   struct bio *bio)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
744
	struct cfq_io_context *cic;
745 746 747
	struct cfq_queue *cfqq;

	/*
748
	 * Disallow merge of a sync bio into an async request.
749
	 */
750
	if (cfq_bio_sync(bio) && !rq_is_sync(rq))
751 752 753
		return 0;

	/*
754 755
	 * Lookup the cfqq that this bio will be queued with. Allow
	 * merge only if rq is queued there.
756
	 */
757
	cic = cfq_cic_lookup(cfqd, current->io_context);
758 759
	if (!cic)
		return 0;
760

761
	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
762 763
	if (cfqq == RQ_CFQQ(rq))
		return 1;
764

765
	return 0;
766 767
}

768 769
static void __cfq_set_active_queue(struct cfq_data *cfqd,
				   struct cfq_queue *cfqq)
770 771
{
	if (cfqq) {
772
		cfq_log_cfqq(cfqd, cfqq, "set_active");
773
		cfqq->slice_end = 0;
774 775 776
		cfqq->slice_dispatch = 0;

		cfq_clear_cfqq_wait_request(cfqq);
Jens Axboe's avatar
Jens Axboe committed
777 778
		cfq_clear_cfqq_must_alloc_slice(cfqq);
		cfq_clear_cfqq_fifo_expire(cfqq);
779
		cfq_mark_cfqq_slice_new(cfqq);
780 781

		del_timer(&cfqd->idle_slice_timer);
782 783 784 785 786
	}

	cfqd->active_queue = cfqq;
}

787 788 789 790 791
/*
 * 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,
792
		    int timed_out)
793
{
794 795
	cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);

796 797 798 799 800 801
	if (cfq_cfqq_wait_request(cfqq))
		del_timer(&cfqd->idle_slice_timer);

	cfq_clear_cfqq_wait_request(cfqq);

	/*
802
	 * store what was left of this slice, if the queue idled/timed out
803
	 */
804
	if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
805
		cfqq->slice_resid = cfqq->slice_end - jiffies;
806 807
		cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
	}
808

809
	cfq_resort_rr_list(cfqd, cfqq);
810 811 812 813 814 815 816 817 818 819

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

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

820
static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out)
821 822 823 824
{
	struct cfq_queue *cfqq = cfqd->active_queue;

	if (cfqq)
825
		__cfq_slice_expired(cfqd, cfqq, timed_out);
826 827
}

828 829 830 831
/*
 * 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
832
static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
833
{
834 835
	if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
		return NULL;
836

837
	return cfq_rb_first(&cfqd->service_tree);
Jens Axboe's avatar
Jens Axboe committed
838 839
}

840 841 842
/*
 * Get and set a new active queue for service.
 */
Jens Axboe's avatar
Jens Axboe committed
843 844 845 846
static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
{
	struct cfq_queue *cfqq;

847
	cfqq = cfq_get_next_queue(cfqd);
848
	__cfq_set_active_queue(cfqd, cfqq);
Jens Axboe's avatar
Jens Axboe committed
849
	return cfqq;
850 851
}

852 853 854 855 856 857 858 859 860
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
861 862 863 864 865 866 867 868 869 870
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;
}

871 872
static int cfq_close_cooperator(struct cfq_data *cfq_data,
				struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
873 874
{
	/*
875 876 877
	 * 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
878
	 */
879
	return 0;
Jens Axboe's avatar
Jens Axboe committed
880 881 882
}

#define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024))
883

Jens Axboe's avatar
Jens Axboe committed
884
static void cfq_arm_slice_timer(struct cfq_data *cfqd)
885
{
886
	struct cfq_queue *cfqq = cfqd->active_queue;
887
	struct cfq_io_context *cic;
888 889
	unsigned long sl;

890
	/*
891 892 893
	 * SSD device without seek penalty, disable idling. But only do so
	 * for devices that support queuing, otherwise we still have a problem
	 * with sync vs async workloads.
894
	 */
895
	if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
896 897
		return;

898
	WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
Jens Axboe's avatar
Jens Axboe committed
899
	WARN_ON(cfq_cfqq_slice_new(cfqq));
900 901 902 903

	/*
	 * idle is disabled, either manually or by past process history
	 */
Jens Axboe's avatar
Jens Axboe committed
904 905 906
	if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq))
		return;

907 908 909 910 911 912
	/*
	 * still requests with the driver, don't idle
	 */
	if (cfqd->rq_in_driver)
		return;

913 914 915
	/*
	 * task has exited, don't wait
	 */
916
	cic = cfqd->active_cic;
917
	if (!cic || !atomic_read(&cic->ioc->nr_tasks))
Jens Axboe's avatar
Jens Axboe committed
918 919 920 921 922
		return;

	/*
	 * See if this prio level has a good candidate
	 */
Jens Axboe's avatar
Jens Axboe committed
923 924
	if (cfq_close_cooperator(cfqd, cfqq) &&
	    (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2))
Jens Axboe's avatar
Jens Axboe committed
925
		return;
926

Jens Axboe's avatar
Jens Axboe committed
927
	cfq_mark_cfqq_wait_request(cfqq);
928

929 930 931 932 933
	/*
	 * 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
934
	sl = cfqd->cfq_slice_idle;
935
	if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
936
		sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
937

938
	mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
939
	cfq_log(cfqd, "arm_idle: %lu", sl);
Linus Torvalds's avatar
Linus Torvalds committed
940 941
}

942 943 944
/*
 * Move request from internal lists to the request queue dispatch list.
 */
945
static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
946
{
947
	struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe's avatar
Jens Axboe committed
948
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
949

950 951
	cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");

952
	cfq_remove_request(rq);
Jens Axboe's avatar
Jens Axboe committed
953
	cfqq->dispatched++;