cfq-iosched.c 81.7 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>
Randy Dunlap's avatar
Randy Dunlap committed
12
#include <linux/jiffies.h>
Linus Torvalds's avatar
Linus Torvalds committed
13
#include <linux/rbtree.h>
14
#include <linux/ioprio.h>
15
#include <linux/blktrace_api.h>
Linus Torvalds's avatar
Linus Torvalds committed
16 17 18 19

/*
 * tunables
 */
20 21
/* max queue in one round of service */
static const int cfq_quantum = 4;
22
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
23 24 25 26
/* 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;
27
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
28
static int cfq_slice_async = HZ / 25;
29
static const int cfq_slice_async_rq = 2;
30
static int cfq_slice_idle = HZ / 125;
31 32
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
33

34
/*
35
 * offset from end of service tree
36
 */
37
#define CFQ_IDLE_DELAY		(HZ / 5)
38 39 40 41 42 43

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

44 45 46 47 48 49
/*
 * Allow merged cfqqs to perform this amount of seeky I/O before
 * deciding to break the queues up again.
 */
#define CFQQ_COOP_TOUT		(HZ)

50
#define CFQ_SLICE_SCALE		(5)
51
#define CFQ_HW_QUEUE_MIN	(5)
52

53 54
#define RQ_CIC(rq)		\
	((struct cfq_io_context *) (rq)->elevator_private)
55
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
Linus Torvalds's avatar
Linus Torvalds committed
56

57 58
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
59

60
static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
61
static struct completion *ioc_gone;
62
static DEFINE_SPINLOCK(ioc_gone_lock);
63

64 65 66 67
#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)

68
#define sample_valid(samples)	((samples) > 80)
69
#define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
70

71 72 73 74 75 76 77 78 79
/*
 * 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;
80
	unsigned count;
81
	u64 min_vdisktime;
82
};
83
#define CFQ_RB_ROOT	(struct cfq_rb_root) { RB_ROOT, NULL, 0, 0, }
84

85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126
/*
 * 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;

127 128 129 130
	unsigned int seek_samples;
	u64 seek_total;
	sector_t seek_mean;
	sector_t last_request_pos;
131
	unsigned long seeky_start;
132

133
	pid_t pid;
Jeff Moyer's avatar
Jeff Moyer committed
134

135
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
136
	struct cfq_queue *new_cfqq;
137
	struct cfq_group *cfqg;
138 139
};

140
/*
141
 * First index in the service_trees.
142 143 144 145
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
146 147
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
148 149
};

150 151 152 153 154 155 156 157 158
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

159 160
/* This is per cgroup per device grouping structure */
struct cfq_group {
161 162 163 164 165 166 167 168 169 170
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
	bool on_st;

	/* number of cfqq currently on this group */
	int nr_cfqq;

171 172 173 174 175 176 177
	/*
	 * rr lists of queues with requests, onle rr for each priority class.
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
};
178

179 180 181
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
182
struct cfq_data {
183
	struct request_queue *queue;
184 185
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
186
	struct cfq_group root_group;
187

188 189
	/*
	 * The priority currently being served
190
	 */
191
	enum wl_prio_t serving_prio;
192 193
	enum wl_type_t serving_type;
	unsigned long workload_expires;
194
	struct cfq_group *serving_group;
195
	bool noidle_tree_requires_idle;
196 197 198 199 200 201 202 203

	/*
	 * 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];

204
	unsigned int busy_queues;
205
	unsigned int busy_queues_avg[2];
206

207
	int rq_in_driver[2];
208
	int sync_flight;
209 210 211 212 213

	/*
	 * queue-depth detection
	 */
	int rq_queued;
214
	int hw_tag;
215 216 217 218 219 220 221 222
	/*
	 * hw_tag can be
	 * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection)
	 *  1 => NCQ is present (hw_tag_est_depth is the estimated max depth)
	 *  0 => no NCQ
	 */
	int hw_tag_est_depth;
	unsigned int hw_tag_samples;
Linus Torvalds's avatar
Linus Torvalds committed
223

224 225 226 227
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
228
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
229

230 231 232
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

233 234 235 236 237
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
238

Jens Axboe's avatar
Jens Axboe committed
239
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
240 241 242 243 244

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
245
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
246 247
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
248 249 250
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
251
	unsigned int cfq_latency;
252 253

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
254

255 256 257 258
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
259 260

	unsigned long last_end_sync_rq;
Linus Torvalds's avatar
Linus Torvalds committed
261 262
};

263 264
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
265
					    enum wl_type_t type,
266 267
					    struct cfq_data *cfqd)
{
268 269 270
	if (!cfqg)
		return NULL;

271
	if (prio == IDLE_WORKLOAD)
272
		return &cfqg->service_tree_idle;
273

274
	return &cfqg->service_trees[prio][type];
275 276
}

Jens Axboe's avatar
Jens Axboe committed
277
enum cfqq_state_flags {
278 279
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
280
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
281 282 283 284
	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 */
285
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
286
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
287
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
288
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
Jens Axboe's avatar
Jens Axboe committed
289 290 291 292 293
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
294
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
295 296 297
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
298
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
299 300 301
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
302
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
303 304 305 306
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
307
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
308 309 310 311
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
312
CFQ_CFQQ_FNS(slice_new);
313
CFQ_CFQQ_FNS(sync);
314
CFQ_CFQQ_FNS(coop);
315
CFQ_CFQQ_FNS(deep);
Jens Axboe's avatar
Jens Axboe committed
316 317
#undef CFQ_CFQQ_FNS

318 319 320 321 322
#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)

323 324 325 326 327 328 329 330 331 332 333
/* Traverses through cfq group service trees */
#define for_each_cfqg_st(cfqg, i, j, st) \
	for (i = 0; i <= IDLE_WORKLOAD; i++) \
		for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\
			: &cfqg->service_tree_idle; \
			(i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \
			(i == IDLE_WORKLOAD && j == 0); \
			j++, st = i < IDLE_WORKLOAD ? \
			&cfqg->service_trees[i][j]: NULL) \


334 335 336 337 338 339 340 341 342
static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq)
{
	if (cfq_class_idle(cfqq))
		return IDLE_WORKLOAD;
	if (cfq_class_rt(cfqq))
		return RT_WORKLOAD;
	return BE_WORKLOAD;
}

343 344 345 346 347 348 349 350 351 352

static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
{
	if (!cfq_cfqq_sync(cfqq))
		return ASYNC_WORKLOAD;
	if (!cfq_cfqq_idle_window(cfqq))
		return SYNC_NOIDLE_WORKLOAD;
	return SYNC_WORKLOAD;
}

353 354
static inline int cfq_busy_queues_wl(enum wl_prio_t wl, struct cfq_data *cfqd)
{
355 356
	struct cfq_group *cfqg = &cfqd->root_group;

357
	if (wl == IDLE_WORKLOAD)
358
		return cfqg->service_tree_idle.count;
359

360 361 362
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
363 364
}

365
static void cfq_dispatch_insert(struct request_queue *, struct request *);
366
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
367
				       struct io_context *, gfp_t);
368
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
369 370
						struct io_context *);

371 372 373 374 375
static inline int rq_in_driver(struct cfq_data *cfqd)
{
	return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
}

376
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
377
					    bool is_sync)
378
{
379
	return cic->cfqq[is_sync];
380 381 382
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
383
				struct cfq_queue *cfqq, bool is_sync)
384
{
385
	cic->cfqq[is_sync] = cfqq;
386 387 388 389 390 391
}

/*
 * 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).
 */
392
static inline bool cfq_bio_sync(struct bio *bio)
393
{
394
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
395
}
Linus Torvalds's avatar
Linus Torvalds committed
396

Andrew Morton's avatar
Andrew Morton committed
397 398 399 400
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
401
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
402
{
403 404
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
405
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
406
	}
Andrew Morton's avatar
Andrew Morton committed
407 408
}

409
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
410 411 412
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

413
	return !cfqd->rq_queued;
Andrew Morton's avatar
Andrew Morton committed
414 415
}

416 417 418 419 420
/*
 * 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.
 */
421
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
422
				 unsigned short prio)
423
{
424
	const int base_slice = cfqd->cfq_slice[sync];
425

426 427 428 429
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
430

431 432 433 434
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);
435 436
}

437 438 439 440 441 442
/*
 * 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
 */

443 444
static inline unsigned cfq_get_avg_queues(struct cfq_data *cfqd, bool rt)
{
445 446 447
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
448
	unsigned busy = cfq_busy_queues_wl(rt, cfqd);
449 450 451 452 453 454 455 456

	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];
}

457 458 459
static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
	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;
480
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
481 482 483 484 485 486 487
}

/*
 * 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.
 */
488
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
489 490 491 492 493 494 495 496 497
{
	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
498
/*
Jens Axboe's avatar
Jens Axboe committed
499
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
500
 * We choose the request that is closest to the head right now. Distance
501
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
502
 */
Jens Axboe's avatar
Jens Axboe committed
503
static struct request *
504
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
505
{
506
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
507
	unsigned long back_max;
508 509 510
#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
511

Jens Axboe's avatar
Jens Axboe committed
512 513 514 515
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
516

Jens Axboe's avatar
Jens Axboe committed
517 518 519 520
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
521 522 523 524
	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
525

526 527
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543

	/*
	 * 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
544
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
545 546 547 548 549 550

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
551
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
552 553

	/* Found required data */
554 555 556 557 558 559

	/*
	 * 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
560
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
561
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
562
			return rq1;
563
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
564
			return rq2;
565 566
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
567
				return rq1;
568
			else
Jens Axboe's avatar
Jens Axboe committed
569
				return rq2;
570
		}
Linus Torvalds's avatar
Linus Torvalds committed
571

572
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
573
		return rq1;
574
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
575 576
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
577 578 579 580 581 582 583 584
	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
585
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
586
		else
Jens Axboe's avatar
Jens Axboe committed
587
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
588 589 590
	}
}

591 592 593
/*
 * The below is leftmost cache rbtree addon
 */
594
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
595
{
596 597 598 599
	/* Service tree is empty */
	if (!root->count)
		return NULL;

600 601 602
	if (!root->left)
		root->left = rb_first(&root->rb);

603 604 605 606
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
607 608
}

609 610 611 612 613 614 615 616 617 618 619
static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root)
{
	if (!root->left)
		root->left = rb_first(&root->rb);

	if (root->left)
		return rb_entry_cfqg(root->left);

	return NULL;
}

620 621 622 623 624 625
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

626 627 628 629
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
630
	rb_erase_init(n, &root->rb);
631
	--root->count;
632 633
}

Linus Torvalds's avatar
Linus Torvalds committed
634 635 636
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
637 638 639
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
640
{
641 642
	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
643
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
644

645
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
646 647

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
648
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
649

650
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
651
		next = rb_entry_rq(rbnext);
652 653 654
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
655
			next = rb_entry_rq(rbnext);
656
	}
Linus Torvalds's avatar
Linus Torvalds committed
657

658
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
659 660
}

661 662
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
663
{
664 665 666
	/*
	 * just an approximation, should be ok.
	 */
667
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
668
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
669 670
}

671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747
static inline s64
cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	return cfqg->vdisktime - st->min_vdisktime;
}

static void
__cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	struct rb_node **node = &st->rb.rb_node;
	struct rb_node *parent = NULL;
	struct cfq_group *__cfqg;
	s64 key = cfqg_key(st, cfqg);
	int left = 1;

	while (*node != NULL) {
		parent = *node;
		__cfqg = rb_entry_cfqg(parent);

		if (key < cfqg_key(st, __cfqg))
			node = &parent->rb_left;
		else {
			node = &parent->rb_right;
			left = 0;
		}
	}

	if (left)
		st->left = &cfqg->rb_node;

	rb_link_node(&cfqg->rb_node, parent, node);
	rb_insert_color(&cfqg->rb_node, &st->rb);
}

static void
cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
	struct cfq_group *__cfqg;
	struct rb_node *n;

	cfqg->nr_cfqq++;
	if (cfqg->on_st)
		return;

	/*
	 * Currently put the group at the end. Later implement something
	 * so that groups get lesser vtime based on their weights, so that
	 * if group does not loose all if it was not continously backlogged.
	 */
	n = rb_last(&st->rb);
	if (n) {
		__cfqg = rb_entry_cfqg(n);
		cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY;
	} else
		cfqg->vdisktime = st->min_vdisktime;

	__cfq_group_service_tree_add(st, cfqg);
	cfqg->on_st = true;
}

static void
cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

	cfqg->on_st = false;
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
}

748
/*
749
 * The cfqd->service_trees holds all pending cfq_queue's that have
750 751 752
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
753
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
754
				 bool add_front)
755
{
756 757
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
758
	unsigned long rb_key;
759
	struct cfq_rb_root *service_tree;
760
	int left;
761

762 763
	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
						cfqq_type(cfqq), cfqd);
764 765
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
766
		parent = rb_last(&service_tree->rb);
767 768 769 770 771 772
		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) {
773 774 775 776 777 778
		/*
		 * 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.
		 */
779
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
780
		rb_key -= cfqq->slice_resid;
781
		cfqq->slice_resid = 0;
782 783
	} else {
		rb_key = -HZ;
784
		__cfqq = cfq_rb_first(service_tree);
785 786
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
Linus Torvalds's avatar
Linus Torvalds committed
787

788
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
789
		/*
790
		 * same position, nothing more to do
791
		 */
792 793
		if (rb_key == cfqq->rb_key &&
		    cfqq->service_tree == service_tree)
794
			return;
Linus Torvalds's avatar
Linus Torvalds committed
795

796 797
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
798
	}
799

800
	left = 1;
801
	parent = NULL;
802 803
	cfqq->service_tree = service_tree;
	p = &service_tree->rb.rb_node;
804
	while (*p) {
805
		struct rb_node **n;
806

807 808 809
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

810
		/*
811
		 * sort by key, that represents service time.
812
		 */
813
		if (time_before(rb_key, __cfqq->rb_key))
814
			n = &(*p)->rb_left;
815
		else {
816
			n = &(*p)->rb_right;
817
			left = 0;
818
		}
819 820

		p = n;
821 822
	}

823
	if (left)
824
		service_tree->left = &cfqq->rb_node;
825

826 827
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
828 829
	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
	service_tree->count++;
830
	cfq_group_service_tree_add(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
831 832
}

833
static struct cfq_queue *
834 835 836
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)
837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
{
	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.
		 */
853
		if (sector > blk_rq_pos(cfqq->next_rq))
854
			n = &(*p)->rb_right;
855
		else if (sector < blk_rq_pos(cfqq->next_rq))
856 857 858 859
			n = &(*p)->rb_left;
		else
			break;
		p = n;
860
		cfqq = NULL;
861 862 863 864 865
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
866
	return cfqq;
867 868 869 870 871 872 873
}

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

874 875 876 877
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
878 879 880 881 882 883

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

884
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
885 886
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
887 888
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
889 890 891
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
892 893
}

894 895 896
/*
 * Update cfqq's position in the service tree.
 */
897
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
898 899 900 901
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
902
	if (cfq_cfqq_on_rr(cfqq)) {
903
		cfq_service_tree_add(cfqd, cfqq, 0);
904 905
		cfq_prio_tree_add(cfqd, cfqq);
	}
Jens Axboe's avatar
Jens Axboe committed
906 907
}

Linus Torvalds's avatar
Linus Torvalds committed
908 909
/*
 * add to busy list of queues for service, trying to be fair in ordering
910
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
911
 */
912
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
913
{
914
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
915 916
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
917 918
	cfqd->busy_queues++;

919
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
920 921
}

922 923 924 925
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
926
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
927
{
928
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
929 930
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
931

932 933 934 935
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
	}
936 937 938 939
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
940

941
	cfq_group_service_tree_del(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
942 943 944 945 946 947 948
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
}

/*
 * rb tree support functions
 */
949
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
950
{
Jens Axboe's avatar
Jens Axboe committed
951 952
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
953

954 955
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
956

Jens Axboe's avatar
Jens Axboe committed
957
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
958

959 960 961 962 963 964 965 966 967 968 969
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) {
		/*
		 * Queue will be deleted from service tree when we actually
		 * expire it later. Right now just remove it from prio tree
		 * as it is empty.
		 */
		if (cfqq->p_root) {
			rb_erase(&cfqq->p_node, cfqq->p_root);
			cfqq->p_root = NULL;
		}
	}
Linus Torvalds's avatar
Linus Torvalds committed
970 971
}

Jens Axboe's avatar
Jens Axboe committed
972
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
973
{
Jens Axboe's avatar
Jens Axboe committed
974
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds's avatar
Linus Torvalds committed
975
	struct cfq_data *cfqd = cfqq->cfqd;
976
	struct request *__alias, *prev;
Linus Torvalds's avatar
Linus Torvalds committed
977

978
	cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds's avatar
Linus Torvalds committed
979 980 981 982 983

	/*
	 * looks a little odd, but the first insert might return an alias.
	 * if that happens, put the alias on the dispatch list
	 */
984
	while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe's avatar
Jens Axboe committed
985
		cfq_dispatch_insert(cfqd->queue, __alias);
986 987 988

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
989 990 991 992

	/*
	 * check if this request is a better next-serve candidate
	 */
993
	prev = cfqq->next_rq;
994
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
995 996 997 998 999 1000 1001

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

1002
	BUG_ON(!cfqq->next_rq);
Linus Torvalds's avatar
Linus Torvalds committed
1003 1004
}

1005
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
1006
{
1007 1008
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
Jens Axboe's avatar
Jens Axboe committed
1009
	cfq_add_rq_rb(rq);
Linus Torvalds's avatar
Linus Torvalds committed
1010 1011
}

1012 1013
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
1014
{
1015
	struct task_struct *tsk = current;
1016
	struct cfq_io_context *cic;
1017
	struct cfq_queue *cfqq;
Linus Torvalds's avatar
Linus Torvalds committed
1018

1019
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
1020 1021 1022 1023
	if (!cic)
		return NULL;

	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
1024 1025 1026
	if (cfqq) {
		sector_t sector = bio->bi_sector + bio_sectors(bio);

1027
		return elv_rb_find(&cfqq->sort_list, sector);
1028
	}
Linus Torvalds's avatar
Linus Torvalds committed
1029 1030 1031 1032

	return NULL;
}

Jens Axboe's avatar