cfq-iosched.c 104 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>
10
#include <linux/slab.h>
Al Viro's avatar
Al Viro committed
11 12
#include <linux/blkdev.h>
#include <linux/elevator.h>
Randy Dunlap's avatar
Randy Dunlap committed
13
#include <linux/jiffies.h>
Linus Torvalds's avatar
Linus Torvalds committed
14
#include <linux/rbtree.h>
15
#include <linux/ioprio.h>
16
#include <linux/blktrace_api.h>
17
#include "cfq.h"
Linus Torvalds's avatar
Linus Torvalds committed
18 19 20 21

/*
 * tunables
 */
22
/* max queue in one round of service */
Shaohua Li's avatar
Shaohua Li committed
23
static const int cfq_quantum = 8;
24
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
25 26 27 28
/* 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;
29
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
30
static int cfq_slice_async = HZ / 25;
31
static const int cfq_slice_async_rq = 2;
32
static int cfq_slice_idle = HZ / 125;
33
static int cfq_group_idle = HZ / 125;
34 35
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
36

37
/*
38
 * offset from end of service tree
39
 */
40
#define CFQ_IDLE_DELAY		(HZ / 5)
41 42 43 44 45 46

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

47
#define CFQ_SLICE_SCALE		(5)
48
#define CFQ_HW_QUEUE_MIN	(5)
49
#define CFQ_SERVICE_SHIFT       12
50

51
#define CFQQ_SEEK_THR		(sector_t)(8 * 100)
52
#define CFQQ_CLOSE_THR		(sector_t)(8 * 1024)
53
#define CFQQ_SECT_THR_NONROT	(sector_t)(2 * 32)
54
#define CFQQ_SEEKY(cfqq)	(hweight32(cfqq->seek_history) > 32/8)
55

56
#define RQ_CIC(rq)		\
57 58 59
	((struct cfq_io_context *) (rq)->elevator_private[0])
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private[1])
#define RQ_CFQG(rq)		(struct cfq_group *) ((rq)->elevator_private[2])
Linus Torvalds's avatar
Linus Torvalds committed
60

61 62
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
63

64
static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
65
static struct completion *ioc_gone;
66
static DEFINE_SPINLOCK(ioc_gone_lock);
67

68 69 70
static DEFINE_SPINLOCK(cic_index_lock);
static DEFINE_IDA(cic_index_ida);

71 72 73 74
#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)

75
#define sample_valid(samples)	((samples) > 80)
76
#define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
77

78 79 80 81 82 83 84 85 86
/*
 * 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;
87
	unsigned count;
88
	unsigned total_weight;
89
	u64 min_vdisktime;
90
};
91 92
#define CFQ_RB_ROOT	(struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \
			.count = 0, .min_vdisktime = 0, }
93

94 95 96 97 98
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
99
	int ref;
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122
	/* 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;

123 124
	/* time when queue got scheduled in to dispatch first request. */
	unsigned long dispatch_start;
125
	unsigned int allocated_slice;
126
	unsigned int slice_dispatch;
127 128
	/* time when first request from queue completed and slice started. */
	unsigned long slice_start;
129 130 131 132 133 134 135 136 137 138 139 140
	unsigned long slice_end;
	long slice_resid;

	/* 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;

141 142
	pid_t pid;

143
	u32 seek_history;
144 145
	sector_t last_request_pos;

146
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
147
	struct cfq_queue *new_cfqq;
148
	struct cfq_group *cfqg;
149 150
	/* Number of sectors dispatched from queue in single dispatch round */
	unsigned long nr_sectors;
151 152
};

153
/*
154
 * First index in the service_trees.
155 156 157 158
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
159 160
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
161
	CFQ_PRIO_NR,
162 163
};

164 165 166 167 168 169 170 171 172
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

173 174
/* This is per cgroup per device grouping structure */
struct cfq_group {
175 176 177 178 179
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
180
	unsigned int weight;
181 182 183 184

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

185
	/*
186 187 188 189 190 191 192 193 194 195 196 197
	 * Per group busy queus average. Useful for workload slice calc. We
	 * create the array for each prio class but at run time it is used
	 * only for RT and BE class and slot for IDLE class remains unused.
	 * This is primarily done to avoid confusion and a gcc warning.
	 */
	unsigned int busy_queues_avg[CFQ_PRIO_NR];
	/*
	 * rr lists of queues with requests. We maintain service trees for
	 * RT and BE classes. These trees are subdivided in subclasses
	 * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE
	 * class there is no subclassification and all the cfq queues go on
	 * a single tree service_tree_idle.
198 199 200 201
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
202 203 204 205

	unsigned long saved_workload_slice;
	enum wl_type_t saved_workload;
	enum wl_prio_t saved_serving_prio;
206 207 208
	struct blkio_group blkg;
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	struct hlist_node cfqd_node;
209
	int ref;
210
#endif
211 212
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
213
};
214

215 216 217
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
218
struct cfq_data {
219
	struct request_queue *queue;
220 221
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
222
	struct cfq_group root_group;
223

224 225
	/*
	 * The priority currently being served
226
	 */
227
	enum wl_prio_t serving_prio;
228 229
	enum wl_type_t serving_type;
	unsigned long workload_expires;
230
	struct cfq_group *serving_group;
231 232 233 234 235 236 237 238

	/*
	 * 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];

239
	unsigned int busy_queues;
240
	unsigned int busy_sync_queues;
241

242 243
	int rq_in_driver;
	int rq_in_flight[2];
244 245 246 247 248

	/*
	 * queue-depth detection
	 */
	int rq_queued;
249
	int hw_tag;
250 251 252 253 254 255 256 257
	/*
	 * 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
258

259 260 261 262
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
263
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
264

265 266 267
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

268 269 270 271 272
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
273

Jens Axboe's avatar
Jens Axboe committed
274
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
275 276 277 278 279

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
280
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
281 282
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
283 284 285
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
286
	unsigned int cfq_group_idle;
287
	unsigned int cfq_latency;
288

289
	unsigned int cic_index;
290
	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
291

292 293 294 295
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
296

297
	unsigned long last_delayed_sync;
298 299 300

	/* List of cfq groups being managed on this device*/
	struct hlist_head cfqg_list;
301
	struct rcu_head rcu;
Linus Torvalds's avatar
Linus Torvalds committed
302 303
};

304 305
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);

306 307
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
308
					    enum wl_type_t type)
309
{
310 311 312
	if (!cfqg)
		return NULL;

313
	if (prio == IDLE_WORKLOAD)
314
		return &cfqg->service_tree_idle;
315

316
	return &cfqg->service_trees[prio][type];
317 318
}

Jens Axboe's avatar
Jens Axboe committed
319
enum cfqq_state_flags {
320 321
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
322
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
323 324 325 326
	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 */
327
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
328
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
329
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
330
	CFQ_CFQQ_FLAG_split_coop,	/* shared cfqq will be splitted */
331
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
332
	CFQ_CFQQ_FLAG_wait_busy,	/* Waiting for next request */
Jens Axboe's avatar
Jens Axboe committed
333 334 335 336 337
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
338
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
339 340 341
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
342
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
343 344 345
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
346
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
347 348 349 350
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
351
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
352 353 354 355
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
356
CFQ_CFQQ_FNS(slice_new);
357
CFQ_CFQQ_FNS(sync);
358
CFQ_CFQQ_FNS(coop);
359
CFQ_CFQQ_FNS(split_coop);
360
CFQ_CFQQ_FNS(deep);
361
CFQ_CFQQ_FNS(wait_busy);
Jens Axboe's avatar
Jens Axboe committed
362 363
#undef CFQ_CFQQ_FNS

364
#ifdef CONFIG_CFQ_GROUP_IOSCHED
365 366 367 368 369 370 371 372 373 374
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \
			cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
			blkg_path(&(cfqq)->cfqg->blkg), ##args);

#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)				\
	blk_add_trace_msg((cfqd)->queue, "%s " fmt,			\
				blkg_path(&(cfqg)->blkg), ##args);      \

#else
375 376
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
377 378
#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0);
#endif
379 380 381
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

382 383 384 385 386 387 388 389 390 391 392
/* 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) \


393 394 395 396 397 398 399 400 401 402 403 404 405 406 407
static inline bool iops_mode(struct cfq_data *cfqd)
{
	/*
	 * If we are not idling on queues and it is a NCQ drive, parallel
	 * execution of requests is on and measuring time is not possible
	 * in most of the cases until and unless we drive shallower queue
	 * depths and that becomes a performance bottleneck. In such cases
	 * switch to start providing fairness in terms of number of IOs.
	 */
	if (!cfqd->cfq_slice_idle && cfqd->hw_tag)
		return true;
	else
		return false;
}

408 409 410 411 412 413 414 415 416
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;
}

417 418 419 420 421 422 423 424 425 426

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;
}

427 428 429
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
430 431
{
	if (wl == IDLE_WORKLOAD)
432
		return cfqg->service_tree_idle.count;
433

434 435 436
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
437 438
}

439 440 441 442 443 444 445
static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg)
{
	return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
}

446
static void cfq_dispatch_insert(struct request_queue *, struct request *);
447
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
448
				       struct io_context *, gfp_t);
449
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
450 451 452
						struct io_context *);

static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
453
					    bool is_sync)
454
{
455
	return cic->cfqq[is_sync];
456 457 458
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
459
				struct cfq_queue *cfqq, bool is_sync)
460
{
461
	cic->cfqq[is_sync] = cfqq;
462 463
}

464
#define CIC_DEAD_KEY	1ul
465
#define CIC_DEAD_INDEX_SHIFT	1
466 467 468

static inline void *cfqd_dead_key(struct cfq_data *cfqd)
{
469
	return (void *)(cfqd->cic_index << CIC_DEAD_INDEX_SHIFT | CIC_DEAD_KEY);
470 471 472 473 474 475 476 477 478 479 480 481
}

static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic)
{
	struct cfq_data *cfqd = cic->key;

	if (unlikely((unsigned long) cfqd & CIC_DEAD_KEY))
		return NULL;

	return cfqd;
}

482 483 484 485
/*
 * 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).
 */
486
static inline bool cfq_bio_sync(struct bio *bio)
487
{
488
	return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC);
489
}
Linus Torvalds's avatar
Linus Torvalds committed
490

Andrew Morton's avatar
Andrew Morton committed
491 492 493 494
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
495
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
496
{
497 498
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
499
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
500
	}
Andrew Morton's avatar
Andrew Morton committed
501 502
}

503
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
504 505 506
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

507
	return !cfqd->rq_queued;
Andrew Morton's avatar
Andrew Morton committed
508 509
}

510 511 512 513 514
/*
 * 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.
 */
515
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
516
				 unsigned short prio)
517
{
518
	const int base_slice = cfqd->cfq_slice[sync];
519

520 521 522 523
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
524

525 526 527 528
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);
529 530
}

531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570
static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg)
{
	u64 d = delta << CFQ_SERVICE_SHIFT;

	d = d * BLKIO_WEIGHT_DEFAULT;
	do_div(d, cfqg->weight);
	return d;
}

static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime)
{
	s64 delta = (s64)(vdisktime - min_vdisktime);
	if (delta > 0)
		min_vdisktime = vdisktime;

	return min_vdisktime;
}

static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime)
{
	s64 delta = (s64)(vdisktime - min_vdisktime);
	if (delta < 0)
		min_vdisktime = vdisktime;

	return min_vdisktime;
}

static void update_min_vdisktime(struct cfq_rb_root *st)
{
	u64 vdisktime = st->min_vdisktime;
	struct cfq_group *cfqg;

	if (st->left) {
		cfqg = rb_entry_cfqg(st->left);
		vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime);
	}

	st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime);
}

571 572 573 574 575 576
/*
 * 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
 */

577 578
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
579
{
580 581 582
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
583
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
584

585 586 587
	min_q = min(cfqg->busy_queues_avg[rt], busy);
	max_q = max(cfqg->busy_queues_avg[rt], busy);
	cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
588
		cfq_hist_divisor;
589 590 591 592 593 594 595 596 597
	return cfqg->busy_queues_avg[rt];
}

static inline unsigned
cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

	return cfq_target_latency * cfqg->weight / st->total_weight;
598 599
}

600
static inline unsigned
601
cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
602
{
603 604
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
605 606 607 608 609 610
		/*
		 * interested queues (we consider only the ones with the same
		 * priority class in the cfq group)
		 */
		unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg,
						cfq_class_rt(cfqq));
611 612
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
613 614 615
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
616 617 618 619 620 621 622
			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 */
623
			slice = max(slice * group_slice / expect_latency,
624 625 626
				    low_slice);
		}
	}
627 628 629 630 631 632
	return slice;
}

static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
633
	unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
634

635
	cfqq->slice_start = jiffies;
636
	cfqq->slice_end = jiffies + slice;
637
	cfqq->allocated_slice = slice;
638
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
639 640 641 642 643 644 645
}

/*
 * 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.
 */
646
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
647 648
{
	if (cfq_cfqq_slice_new(cfqq))
Shaohua Li's avatar
Shaohua Li committed
649
		return false;
650
	if (time_before(jiffies, cfqq->slice_end))
Shaohua Li's avatar
Shaohua Li committed
651
		return false;
652

Shaohua Li's avatar
Shaohua Li committed
653
	return true;
654 655
}

Linus Torvalds's avatar
Linus Torvalds committed
656
/*
Jens Axboe's avatar
Jens Axboe committed
657
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
658
 * We choose the request that is closest to the head right now. Distance
659
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
660
 */
Jens Axboe's avatar
Jens Axboe committed
661
static struct request *
662
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
663
{
664
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
665
	unsigned long back_max;
666 667 668
#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
669

Jens Axboe's avatar
Jens Axboe committed
670 671 672 673
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
674

Jens Axboe's avatar
Jens Axboe committed
675 676 677 678
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
679
	if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
680
		return rq1;
681 682
	else if ((rq2->cmd_flags & REQ_META) &&
		 !(rq1->cmd_flags & REQ_META))
683
		return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
684

685 686
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702

	/*
	 * 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
703
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
704 705 706 707 708 709

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
710
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
711 712

	/* Found required data */
713 714 715 716 717 718

	/*
	 * 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
719
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
720
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
721
			return rq1;
722
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
723
			return rq2;
724 725
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
726
				return rq1;
727
			else
Jens Axboe's avatar
Jens Axboe committed
728
				return rq2;
729
		}
Linus Torvalds's avatar
Linus Torvalds committed
730

731
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
732
		return rq1;
733
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
734 735
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
736 737 738 739 740 741 742 743
	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
744
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
745
		else
Jens Axboe's avatar
Jens Axboe committed
746
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
747 748 749
	}
}

750 751 752
/*
 * The below is leftmost cache rbtree addon
 */
753
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
754
{
755 756 757 758
	/* Service tree is empty */
	if (!root->count)
		return NULL;

759 760 761
	if (!root->left)
		root->left = rb_first(&root->rb);

762 763 764 765
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
766 767
}

768 769 770 771 772 773 774 775 776 777 778
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;
}

779 780 781 782 783 784
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

785 786 787 788
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
789
	rb_erase_init(n, &root->rb);
790
	--root->count;
791 792
}

Linus Torvalds's avatar
Linus Torvalds committed
793 794 795
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
796 797 798
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
799
{
800 801
	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
802
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
803

804
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
805 806

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
807
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
808

809
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
810
		next = rb_entry_rq(rbnext);
811 812 813
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
814
			next = rb_entry_rq(rbnext);
815
	}
Linus Torvalds's avatar
Linus Torvalds committed
816

817
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
818 819
}

820 821
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
822
{
823 824 825
	/*
	 * just an approximation, should be ok.
	 */
826
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
827
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
828 829
}

830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871
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++;
872
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
873 874 875 876 877 878 879 880 881 882 883 884 885 886 887
		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);
888
	st->total_weight += cfqg->weight;
889 890 891 892 893 894 895 896 897
}

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--;
898

899 900 901 902
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

903
	cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
904
	st->total_weight -= cfqg->weight;
905 906
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
907
	cfqg->saved_workload_slice = 0;
908
	cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1);
909 910 911 912
}

static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq)
{
913
	unsigned int slice_used;
914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929

	/*
	 * Queue got expired before even a single request completed or
	 * got expired immediately after first request completion.
	 */
	if (!cfqq->slice_start || cfqq->slice_start == jiffies) {
		/*
		 * Also charge the seek time incurred to the group, otherwise
		 * if there are mutiple queues in the group, each can dispatch
		 * a single request on seeky media and cause lots of seek time
		 * and group will never know it.
		 */
		slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start),
					1);
	} else {
		slice_used = jiffies - cfqq->slice_start;
930 931
		if (slice_used > cfqq->allocated_slice)
			slice_used = cfqq->allocated_slice;
932 933 934 935 936 937
	}

	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
938
				struct cfq_queue *cfqq)
939 940
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
941
	unsigned int used_sl, charge;
942 943 944 945
	int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
			- cfqg->service_tree_idle.count;

	BUG_ON(nr_sync < 0);
946
	used_sl = charge = cfq_cfqq_slice_usage(cfqq);
947

948 949 950 951
	if (iops_mode(cfqd))
		charge = cfqq->slice_dispatch;
	else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
		charge = cfqq->allocated_slice;
952 953 954

	/* Can't update vdisktime while group is on service tree */
	cfq_rb_erase(&cfqg->rb_node, st);
955
	cfqg->vdisktime += cfq_scale_slice(charge, cfqg);
956 957 958 959 960 961 962 963 964 965
	__cfq_group_service_tree_add(st, cfqg);

	/* This group is being expired. Save the context */
	if (time_after(cfqd->workload_expires, jiffies)) {
		cfqg->saved_workload_slice = cfqd->workload_expires
						- jiffies;
		cfqg->saved_workload = cfqd->serving_type;
		cfqg->saved_serving_prio = cfqd->serving_prio;
	} else
		cfqg->saved_workload_slice = 0;
966 967 968

	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
					st->min_vdisktime);
969 970 971
	cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u disp=%u charge=%u iops=%u"
			" sect=%u", used_sl, cfqq->slice_dispatch, charge,
			iops_mode(cfqd), cfqq->nr_sectors);
972 973
	cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl);
	cfq_blkiocg_set_start_empty_time(&cfqg->blkg);
974 975
}

976 977 978 979 980 981 982 983
#ifdef CONFIG_CFQ_GROUP_IOSCHED
static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg)
{
	if (blkg)
		return container_of(blkg, struct cfq_group, blkg);
	return NULL;
}

984 985
void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg,
					unsigned int weight)
986 987 988 989
{
	cfqg_of_blkg(blkg)->weight = weight;
}

990 991 992 993 994 995 996 997
static struct cfq_group *
cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create)
{
	struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
	struct cfq_group *cfqg = NULL;
	void *key = cfqd;
	int i, j;
	struct cfq_rb_root *st;
998 999
	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
	unsigned int major, minor;
1000 1001

	cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
1002 1003 1004 1005 1006
	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
		cfqg->blkg.dev = MKDEV(major, minor);
		goto done;
	}
1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017
	if (cfqg || !create)
		goto done;

	cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node);
	if (!cfqg)
		goto done;

	for_each_cfqg_st(cfqg, i, j, st)
		*st = CFQ_RB_ROOT;
	RB_CLEAR_NODE(&cfqg->rb_node);

1018 1019 1020 1021 1022 1023
	/*
	 * Take the initial reference that will be released on destroy
	 * This can be thought of a joint reference by cgroup and
	 * elevator which will be dropped by either elevator exit
	 * or cgroup deletion path depending on who is exiting first.
	 */
1024
	cfqg->ref = 1;
1025

1026 1027
	/*
	 * Add group onto cgroup list. It might happen that bdi->dev is
1028
	 * not initialized yet. Initialize this new group without major
1029 1030 1031 1032 1033 1034
	 * and minor info and this info will be filled in once a new thread
	 * comes for IO. See code above.
	 */
	if (bdi->dev) {
		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
1035
					MKDEV(major, minor));
1036 1037 1038 1039
	} else
		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
					0);

1040
	cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066

	/* Add group on cfqd list */
	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);

done:
	return cfqg;
}

/*
 * Search for the cfq group current task belongs to. If create = 1, then also
 * create the cfq group if it does not exist. request_queue lock must be held.
 */
static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
{
	struct cgroup *cgroup;
	struct cfq_group *cfqg = NULL;

	rcu_read_lock();
	cgroup = task_cgroup(current, blkio_subsys_id);
	cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create);
	if (!cfqg && create)
		cfqg = &cfqd->root_group;
	rcu_read_unlock();
	return cfqg;
}

1067 1068
static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
{
1069
	cfqg->ref++;
1070 1071 1072
	return cfqg;
}