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 "blk.h"
18
#include "cfq.h"
Linus Torvalds's avatar
Linus Torvalds committed
19 20 21 22

/*
 * tunables
 */
23
/* max queue in one round of service */
Shaohua Li's avatar
Shaohua Li committed
24
static const int cfq_quantum = 8;
25
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
26 27 28 29
/* 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;
30
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
31
static int cfq_slice_async = HZ / 25;
32
static const int cfq_slice_async_rq = 2;
33
static int cfq_slice_idle = HZ / 125;
34
static int cfq_group_idle = HZ / 125;
35 36
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
37

38
/*
39
 * offset from end of service tree
40
 */
41
#define CFQ_IDLE_DELAY		(HZ / 5)
42 43 44 45 46 47

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

48
#define CFQ_SLICE_SCALE		(5)
49
#define CFQ_HW_QUEUE_MIN	(5)
50
#define CFQ_SERVICE_SHIFT       12
51

52
#define CFQQ_SEEK_THR		(sector_t)(8 * 100)
53
#define CFQQ_CLOSE_THR		(sector_t)(8 * 1024)
54
#define CFQQ_SECT_THR_NONROT	(sector_t)(2 * 32)
55
#define CFQQ_SEEKY(cfqq)	(hweight32(cfqq->seek_history) > 32/8)
56

57
#define RQ_CIC(rq)		\
58 59 60
	((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
61

62 63
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
64

65 66 67 68
#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)

69
#define sample_valid(samples)	((samples) > 80)
70
#define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
71

72 73 74 75 76 77 78 79 80
/*
 * 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;
81
	unsigned count;
82
	unsigned total_weight;
83
	u64 min_vdisktime;
84
	struct cfq_ttime ttime;
85
};
86 87
#define CFQ_RB_ROOT	(struct cfq_rb_root) { .rb = RB_ROOT, \
			.ttime = {.last_end_request = jiffies,},}
88

89 90 91 92 93
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
94
	int ref;
95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
	/* 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;

118 119
	/* time when queue got scheduled in to dispatch first request. */
	unsigned long dispatch_start;
120
	unsigned int allocated_slice;
121
	unsigned int slice_dispatch;
122 123
	/* time when first request from queue completed and slice started. */
	unsigned long slice_start;
124 125 126
	unsigned long slice_end;
	long slice_resid;

127 128
	/* pending priority requests */
	int prio_pending;
129 130 131 132 133
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;

	/* io prio of this group */
	unsigned short ioprio, org_ioprio;
134
	unsigned short ioprio_class;
135

136 137
	pid_t pid;

138
	u32 seek_history;
139 140
	sector_t last_request_pos;

141
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
142
	struct cfq_queue *new_cfqq;
143
	struct cfq_group *cfqg;
144 145
	/* Number of sectors dispatched from queue in single dispatch round */
	unsigned long nr_sectors;
146 147
};

148
/*
149
 * First index in the service_trees.
150 151 152 153
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
154 155
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
156
	CFQ_PRIO_NR,
157 158
};

159 160 161 162 163 164 165 166 167
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

168 169
/* This is per cgroup per device grouping structure */
struct cfq_group {
170 171 172 173 174
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
175
	unsigned int weight;
176 177
	unsigned int new_weight;
	bool needs_update;
178 179 180 181

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

182
	/*
183
	 * Per group busy queues average. Useful for workload slice calc. We
184 185 186 187 188 189 190 191 192 193 194
	 * 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.
195 196 197 198
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
199 200 201 202

	unsigned long saved_workload_slice;
	enum wl_type_t saved_workload;
	enum wl_prio_t saved_serving_prio;
203 204 205
	struct blkio_group blkg;
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	struct hlist_node cfqd_node;
206
	int ref;
207
#endif
208 209
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
210
	struct cfq_ttime ttime;
211
};
212

213 214 215
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
216
struct cfq_data {
217
	struct request_queue *queue;
218 219
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
220
	struct cfq_group root_group;
221

222 223
	/*
	 * The priority currently being served
224
	 */
225
	enum wl_prio_t serving_prio;
226 227
	enum wl_type_t serving_type;
	unsigned long workload_expires;
228
	struct cfq_group *serving_group;
229 230 231 232 233 234 235 236

	/*
	 * 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];

237
	unsigned int busy_queues;
238
	unsigned int busy_sync_queues;
239

240 241
	int rq_in_driver;
	int rq_in_flight[2];
242 243 244 245 246

	/*
	 * queue-depth detection
	 */
	int rq_queued;
247
	int hw_tag;
248 249 250 251 252 253 254 255
	/*
	 * 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
256

257 258 259 260
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
261
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
262

263 264 265
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

266 267 268 269 270
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
271

Jens Axboe's avatar
Jens Axboe committed
272
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
273 274 275 276 277

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
278
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
279 280
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
281 282 283
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
284
	unsigned int cfq_group_idle;
285
	unsigned int cfq_latency;
286 287

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
288

289 290 291 292
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
293

294
	unsigned long last_delayed_sync;
295 296 297

	/* List of cfq groups being managed on this device*/
	struct hlist_head cfqg_list;
298 299 300

	/* Number of groups which are on blkcg->blkg_list */
	unsigned int nr_blkcg_linked_grps;
Linus Torvalds's avatar
Linus Torvalds committed
301 302
};

303 304
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);

305 306
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
307
					    enum wl_type_t type)
308
{
309 310 311
	if (!cfqg)
		return NULL;

312
	if (prio == IDLE_WORKLOAD)
313
		return &cfqg->service_tree_idle;
314

315
	return &cfqg->service_trees[prio][type];
316 317
}

Jens Axboe's avatar
Jens Axboe committed
318
enum cfqq_state_flags {
319 320
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
321
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
322 323 324 325
	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 */
326
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
327
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
328
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
329
	CFQ_CFQQ_FLAG_split_coop,	/* shared cfqq will be splitted */
330
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
331
	CFQ_CFQQ_FLAG_wait_busy,	/* Waiting for next request */
Jens Axboe's avatar
Jens Axboe committed
332 333 334 335 336
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
337
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
338 339 340
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
341
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
342 343 344
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
345
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
346 347 348 349
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
350
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
351 352 353 354
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
355
CFQ_CFQQ_FNS(slice_new);
356
CFQ_CFQQ_FNS(sync);
357
CFQ_CFQQ_FNS(coop);
358
CFQ_CFQQ_FNS(split_coop);
359
CFQ_CFQQ_FNS(deep);
360
CFQ_CFQQ_FNS(wait_busy);
Jens Axboe's avatar
Jens Axboe committed
361 362
#undef CFQ_CFQQ_FNS

363
#ifdef CONFIG_CFQ_GROUP_IOSCHED
364 365 366
#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', \
367
			blkg_path(&(cfqq)->cfqg->blkg), ##args)
368 369 370

#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)				\
	blk_add_trace_msg((cfqd)->queue, "%s " fmt,			\
371
				blkg_path(&(cfqg)->blkg), ##args)       \
372 373

#else
374 375
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
376
#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0)
377
#endif
378 379 380
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

381 382 383 384 385 386 387 388 389 390
/* 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) \

391 392 393 394 395 396 397 398 399 400 401 402
static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd,
	struct cfq_ttime *ttime, bool group_idle)
{
	unsigned long slice;
	if (!sample_valid(ttime->ttime_samples))
		return false;
	if (group_idle)
		slice = cfqd->cfq_group_idle;
	else
		slice = cfqd->cfq_slice_idle;
	return ttime->ttime_mean > slice;
}
403

404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
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;
}

419 420 421 422 423 424 425 426 427
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;
}

428 429 430 431 432 433 434 435 436 437

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;
}

438 439 440
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
441 442
{
	if (wl == IDLE_WORKLOAD)
443
		return cfqg->service_tree_idle.count;
444

445 446 447
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
448 449
}

450 451 452 453 454 455 456
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;
}

457
static void cfq_dispatch_insert(struct request_queue *, struct request *);
458
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
459
				       struct io_context *, gfp_t);
460
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
461 462 463
						struct io_context *);

static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
464
					    bool is_sync)
465
{
466
	return cic->cfqq[is_sync];
467 468 469
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
470
				struct cfq_queue *cfqq, bool is_sync)
471
{
472
	cic->cfqq[is_sync] = cfqq;
473 474
}

475 476
static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic)
{
Tejun Heo's avatar
Tejun Heo committed
477
	return cic->q->elevator->elevator_data;
478 479
}

480 481 482 483
/*
 * 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).
 */
484
static inline bool cfq_bio_sync(struct bio *bio)
485
{
486
	return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC);
487
}
Linus Torvalds's avatar
Linus Torvalds committed
488

Andrew Morton's avatar
Andrew Morton committed
489 490 491 492
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
493
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
494
{
495 496
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
497
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
498
	}
Andrew Morton's avatar
Andrew Morton committed
499 500
}

501 502 503 504 505
/*
 * 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.
 */
506
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
507
				 unsigned short prio)
508
{
509
	const int base_slice = cfqd->cfq_slice[sync];
510

511 512 513 514
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
515

516 517 518 519
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);
520 521
}

522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554
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)
{
	struct cfq_group *cfqg;

	if (st->left) {
		cfqg = rb_entry_cfqg(st->left);
555 556
		st->min_vdisktime = max_vdisktime(st->min_vdisktime,
						  cfqg->vdisktime);
557 558 559
	}
}

560 561 562 563 564 565
/*
 * 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
 */

566 567
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
568
{
569 570 571
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
572
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
573

574 575 576
	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) /
577
		cfq_hist_divisor;
578 579 580 581 582 583 584 585 586
	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;
587 588
}

589
static inline unsigned
590
cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
591
{
592 593
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
594 595 596 597 598 599
		/*
		 * 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));
600 601
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
602 603 604
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
605 606 607 608 609 610 611
			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 */
612
			slice = max(slice * group_slice / expect_latency,
613 614 615
				    low_slice);
		}
	}
616 617 618 619 620 621
	return slice;
}

static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
622
	unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
623

624
	cfqq->slice_start = jiffies;
625
	cfqq->slice_end = jiffies + slice;
626
	cfqq->allocated_slice = slice;
627
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
628 629 630 631 632 633 634
}

/*
 * 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.
 */
635
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
636 637
{
	if (cfq_cfqq_slice_new(cfqq))
Shaohua Li's avatar
Shaohua Li committed
638
		return false;
639
	if (time_before(jiffies, cfqq->slice_end))
Shaohua Li's avatar
Shaohua Li committed
640
		return false;
641

Shaohua Li's avatar
Shaohua Li committed
642
	return true;
643 644
}

Linus Torvalds's avatar
Linus Torvalds committed
645
/*
Jens Axboe's avatar
Jens Axboe committed
646
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
647
 * We choose the request that is closest to the head right now. Distance
648
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
649
 */
Jens Axboe's avatar
Jens Axboe committed
650
static struct request *
651
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
652
{
653
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
654
	unsigned long back_max;
655 656 657
#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
658

Jens Axboe's avatar
Jens Axboe committed
659 660 661 662
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
663

664 665 666
	if (rq_is_sync(rq1) != rq_is_sync(rq2))
		return rq_is_sync(rq1) ? rq1 : rq2;

667 668
	if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
		return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
669

670 671
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687

	/*
	 * 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
688
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
689 690 691 692 693 694

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
695
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
696 697

	/* Found required data */
698 699 700 701 702 703

	/*
	 * 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
704
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
705
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
706
			return rq1;
707
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
708
			return rq2;
709 710
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
711
				return rq1;
712
			else
Jens Axboe's avatar
Jens Axboe committed
713
				return rq2;
714
		}
Linus Torvalds's avatar
Linus Torvalds committed
715

716
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
717
		return rq1;
718
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
719 720
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
721 722 723 724 725 726 727 728
	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
729
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
730
		else
Jens Axboe's avatar
Jens Axboe committed
731
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
732 733 734
	}
}

735 736 737
/*
 * The below is leftmost cache rbtree addon
 */
738
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
739
{
740 741 742 743
	/* Service tree is empty */
	if (!root->count)
		return NULL;

744 745 746
	if (!root->left)
		root->left = rb_first(&root->rb);

747 748 749 750
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
751 752
}

753 754 755 756 757 758 759 760 761 762 763
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;
}

764 765 766 767 768 769
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

770 771 772 773
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
774
	rb_erase_init(n, &root->rb);
775
	--root->count;
776 777
}

Linus Torvalds's avatar
Linus Torvalds committed
778 779 780
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
781 782 783
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
784
{
785 786
	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
787
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
788

789
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
790 791

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
792
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
793

794
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
795
		next = rb_entry_rq(rbnext);
796 797 798
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
799
			next = rb_entry_rq(rbnext);
800
	}
Linus Torvalds's avatar
Linus Torvalds committed
801

802
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
803 804
}

805 806
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
807
{
808 809 810
	/*
	 * just an approximation, should be ok.
	 */
811
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
812
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
813 814
}

815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849
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
850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
cfq_update_group_weight(struct cfq_group *cfqg)
{
	BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
	if (cfqg->needs_update) {
		cfqg->weight = cfqg->new_weight;
		cfqg->needs_update = false;
	}
}

static void
cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));

	cfq_update_group_weight(cfqg);
	__cfq_group_service_tree_add(st, cfqg);
	st->total_weight += cfqg->weight;
}

static void
cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
871 872 873 874 875 876
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
	struct cfq_group *__cfqg;
	struct rb_node *n;

	cfqg->nr_cfqq++;
877
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
878 879 880 881 882
		return;

	/*
	 * Currently put the group at the end. Later implement something
	 * so that groups get lesser vtime based on their weights, so that
Lucas De Marchi's avatar
Lucas De Marchi committed
883
	 * if group does not loose all if it was not continuously backlogged.
884 885 886 887 888 889 890
	 */
	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;
891 892
	cfq_group_service_tree_add(st, cfqg);
}
893

894 895 896 897 898 899
static void
cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	st->total_weight -= cfqg->weight;
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
900 901 902
}

static void
903
cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
904 905 906 907 908
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
909

910 911 912 913
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

914
	cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
915
	cfq_group_service_tree_del(st, cfqg);
916
	cfqg->saved_workload_slice = 0;
917
	cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1);
918 919
}

920 921
static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
						unsigned int *unaccounted_time)
922
{
923
	unsigned int slice_used;
924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939

	/*
	 * 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;
940 941
		if (slice_used > cfqq->allocated_slice) {
			*unaccounted_time = slice_used - cfqq->allocated_slice;
942
			slice_used = cfqq->allocated_slice;
943 944 945 946
		}
		if (time_after(cfqq->slice_start, cfqq->dispatch_start))
			*unaccounted_time += cfqq->slice_start -
					cfqq->dispatch_start;
947 948 949 950 951 952
	}

	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
953
				struct cfq_queue *cfqq)
954 955
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
956
	unsigned int used_sl, charge, unaccounted_sl = 0;
957 958 959 960
	int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
			- cfqg->service_tree_idle.count;

	BUG_ON(nr_sync < 0);
961
	used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
962

963 964 965 966
	if (iops_mode(cfqd))
		charge = cfqq->slice_dispatch;
	else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
		charge = cfqq->allocated_slice;
967 968

	/* Can't update vdisktime while group is on service tree */
969
	cfq_group_service_tree_del(st, cfqg);
970
	cfqg->vdisktime += cfq_scale_slice(charge, cfqg);
971 972
	/* If a new weight was requested, update now, off tree */
	cfq_group_service_tree_add(st, cfqg);
973 974 975 976 977 978 979 980 981

	/* 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;
982 983 984

	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
					st->min_vdisktime);
985 986 987 988
	cfq_log_cfqq(cfqq->cfqd, cfqq,
		     "sl_used=%u disp=%u charge=%u iops=%u sect=%lu",
		     used_sl, cfqq->slice_dispatch, charge,
		     iops_mode(cfqd), cfqq->nr_sectors);
989 990
	cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl,
					  unaccounted_sl);
991
	cfq_blkiocg_set_start_empty_time(&cfqg->blkg);
992 993
}

994 995 996 997 998 999 1000 1001
#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;
}

Paul Bolle's avatar
Paul Bolle committed
1002 1003
static void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg,
					  unsigned int weight)
1004
{
1005 1006 1007
	struct cfq_group *cfqg = cfqg_of_blkg(blkg);
	cfqg->new_weight = weight;
	cfqg->needs_update = true;
1008 1009
}

1010 1011
static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd,
			struct cfq_group *cfqg, struct blkio_cgroup *blkcg)
1012
{
1013 1014
	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
	unsigned int major, minor;
1015

1016 1017 1018 1019 1020 1021 1022
	/*
	 * Add group onto cgroup list. It might happen that bdi->dev is
	 * not initialized yet. Initialize this new group without major
	 * and minor info and this info will be filled in once a new thread
	 * comes for IO.
	 */
	if (bdi->dev) {
1023
		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
					(void *)cfqd, MKDEV(major, minor));
	} else
		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
					(void *)cfqd, 0);

	cfqd->nr_blkcg_linked_grps++;
	cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);

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

/*
 * Should be called from sleepable context. No request queue lock as per
 * cpu stats are allocated dynamically and alloc_percpu needs to be called
 * from sleepable context.
 */
static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd)
{
	struct cfq_group *cfqg = NULL;
1045
	int i, j, ret