cfq-iosched.c 127.2 KB
Newer Older
L
Linus Torvalds 已提交
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>
L
Linus Torvalds 已提交
8 9
 */
#include <linux/module.h>
10
#include <linux/slab.h>
A
Al Viro 已提交
11 12
#include <linux/blkdev.h>
#include <linux/elevator.h>
13
#include <linux/ktime.h>
L
Linus Torvalds 已提交
14
#include <linux/rbtree.h>
15
#include <linux/ioprio.h>
16
#include <linux/blktrace_api.h>
17
#include <linux/blk-cgroup.h>
18
#include "blk.h"
J
Jens Axboe 已提交
19
#include "blk-wbt.h"
L
Linus Torvalds 已提交
20 21 22 23

/*
 * tunables
 */
24
/* max queue in one round of service */
S
Shaohua Li 已提交
25
static const int cfq_quantum = 8;
26
static const u64 cfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 };
27 28 29 30
/* 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;
31 32
static const u64 cfq_slice_sync = NSEC_PER_SEC / 10;
static u64 cfq_slice_async = NSEC_PER_SEC / 25;
33
static const int cfq_slice_async_rq = 2;
34 35 36
static u64 cfq_slice_idle = NSEC_PER_SEC / 125;
static u64 cfq_group_idle = NSEC_PER_SEC / 125;
static const u64 cfq_target_latency = (u64)NSEC_PER_SEC * 3/10; /* 300 ms */
37
static const int cfq_hist_divisor = 4;
38

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

/*
 * below this threshold, we consider thinktime immediate
 */
47
#define CFQ_MIN_TT		(2 * NSEC_PER_SEC / HZ)
48

49
#define CFQ_SLICE_SCALE		(5)
50
#define CFQ_HW_QUEUE_MIN	(5)
51
#define CFQ_SERVICE_SHIFT       12
52

53
#define CFQQ_SEEK_THR		(sector_t)(8 * 100)
54
#define CFQQ_CLOSE_THR		(sector_t)(8 * 1024)
55
#define CFQQ_SECT_THR_NONROT	(sector_t)(2 * 32)
56
#define CFQQ_SEEKY(cfqq)	(hweight32(cfqq->seek_history) > 32/8)
57

58 59 60
#define RQ_CIC(rq)		icq_to_cic((rq)->elv.icq)
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elv.priv[0])
#define RQ_CFQG(rq)		(struct cfq_group *) ((rq)->elv.priv[1])
L
Linus Torvalds 已提交
61

62
static struct kmem_cache *cfq_pool;
L
Linus Torvalds 已提交
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
/* blkio-related constants */
72 73 74
#define CFQ_WEIGHT_LEGACY_MIN	10
#define CFQ_WEIGHT_LEGACY_DFL	500
#define CFQ_WEIGHT_LEGACY_MAX	1000
75

76
struct cfq_ttime {
77
	u64 last_end_request;
78

79 80
	u64 ttime_total;
	u64 ttime_mean;
81 82 83
	unsigned long ttime_samples;
};

84 85 86 87 88 89 90 91 92
/*
 * 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;
93
	unsigned count;
94
	u64 min_vdisktime;
95
	struct cfq_ttime ttime;
96
};
97
#define CFQ_RB_ROOT	(struct cfq_rb_root) { .rb = RB_ROOT, \
98
			.ttime = {.last_end_request = ktime_get_ns(),},}
99

100 101 102 103 104
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
105
	int ref;
106 107 108 109 110 111 112
	/* 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 */
113
	u64 rb_key;
114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
	/* 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;

129
	/* time when queue got scheduled in to dispatch first request. */
130 131 132
	u64 dispatch_start;
	u64 allocated_slice;
	u64 slice_dispatch;
133
	/* time when first request from queue completed and slice started. */
134 135
	u64 slice_start;
	u64 slice_end;
136
	s64 slice_resid;
137

138 139
	/* pending priority requests */
	int prio_pending;
140 141 142 143 144
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;

	/* io prio of this group */
	unsigned short ioprio, org_ioprio;
145
	unsigned short ioprio_class, org_ioprio_class;
146

147 148
	pid_t pid;

149
	u32 seek_history;
150 151
	sector_t last_request_pos;

152
	struct cfq_rb_root *service_tree;
J
Jeff Moyer 已提交
153
	struct cfq_queue *new_cfqq;
154
	struct cfq_group *cfqg;
155 156
	/* Number of sectors dispatched from queue in single dispatch round */
	unsigned long nr_sectors;
157 158
};

159
/*
160
 * First index in the service_trees.
161 162
 * IDLE is handled separately, so it has negative index
 */
163
enum wl_class_t {
164
	BE_WORKLOAD = 0,
165 166
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
167
	CFQ_PRIO_NR,
168 169
};

170 171 172 173 174 175 176 177 178
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
struct cfqg_stats {
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	/* number of ios merged */
	struct blkg_rwstat		merged;
	/* total time spent on device in ns, may not be accurate w/ queueing */
	struct blkg_rwstat		service_time;
	/* total time spent waiting in scheduler queue in ns */
	struct blkg_rwstat		wait_time;
	/* number of IOs queued up */
	struct blkg_rwstat		queued;
	/* total disk time and nr sectors dispatched by this group */
	struct blkg_stat		time;
#ifdef CONFIG_DEBUG_BLK_CGROUP
	/* time not charged to this cgroup */
	struct blkg_stat		unaccounted_time;
	/* sum of number of ios queued across all samples */
	struct blkg_stat		avg_queue_size_sum;
	/* count of samples taken for average */
	struct blkg_stat		avg_queue_size_samples;
	/* how many times this group has been removed from service tree */
	struct blkg_stat		dequeue;
	/* total time spent waiting for it to be assigned a timeslice. */
	struct blkg_stat		group_wait_time;
T
Tejun Heo 已提交
202
	/* time spent idling for this blkcg_gq */
203 204 205 206 207 208 209 210 211 212 213 214
	struct blkg_stat		idle_time;
	/* total time with empty current active q with other requests queued */
	struct blkg_stat		empty_time;
	/* fields after this shouldn't be cleared on stat reset */
	uint64_t			start_group_wait_time;
	uint64_t			start_idle_time;
	uint64_t			start_empty_time;
	uint16_t			flags;
#endif	/* CONFIG_DEBUG_BLK_CGROUP */
#endif	/* CONFIG_CFQ_GROUP_IOSCHED */
};

215 216 217
/* Per-cgroup data */
struct cfq_group_data {
	/* must be the first member */
218
	struct blkcg_policy_data cpd;
219 220 221 222 223

	unsigned int weight;
	unsigned int leaf_weight;
};

224 225
/* This is per cgroup per device grouping structure */
struct cfq_group {
226 227 228
	/* must be the first member */
	struct blkg_policy_data pd;

229 230 231 232 233
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
T
Tejun Heo 已提交
234

235 236 237 238 239 240 241 242 243 244 245 246
	/*
	 * The number of active cfqgs and sum of their weights under this
	 * cfqg.  This covers this cfqg's leaf_weight and all children's
	 * weights, but does not cover weights of further descendants.
	 *
	 * If a cfqg is on the service tree, it's active.  An active cfqg
	 * also activates its parent and contributes to the children_weight
	 * of the parent.
	 */
	int nr_active;
	unsigned int children_weight;

247 248 249 250 251 252 253 254 255 256 257 258
	/*
	 * vfraction is the fraction of vdisktime that the tasks in this
	 * cfqg are entitled to.  This is determined by compounding the
	 * ratios walking up from this cfqg to the root.
	 *
	 * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all
	 * vfractions on a service tree is approximately 1.  The sum may
	 * deviate a bit due to rounding errors and fluctuations caused by
	 * cfqgs entering and leaving the service tree.
	 */
	unsigned int vfraction;

T
Tejun Heo 已提交
259 260 261 262 263 264
	/*
	 * There are two weights - (internal) weight is the weight of this
	 * cfqg against the sibling cfqgs.  leaf_weight is the wight of
	 * this cfqg against the child cfqgs.  For the root cfqg, both
	 * weights are kept in sync for backward compatibility.
	 */
265
	unsigned int weight;
266
	unsigned int new_weight;
267
	unsigned int dev_weight;
268

T
Tejun Heo 已提交
269 270 271 272
	unsigned int leaf_weight;
	unsigned int new_leaf_weight;
	unsigned int dev_leaf_weight;

273 274 275
	/* number of cfqq currently on this group */
	int nr_cfqq;

276
	/*
277
	 * Per group busy queues average. Useful for workload slice calc. We
278 279 280 281 282 283 284 285 286 287 288
	 * 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.
289 290 291 292
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
293

294
	u64 saved_wl_slice;
295 296
	enum wl_type_t saved_wl_type;
	enum wl_class_t saved_wl_class;
297

298 299
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
S
Shaohua Li 已提交
300
	struct cfq_ttime ttime;
301
	struct cfqg_stats stats;	/* stats for this cfqg */
302 303 304 305 306

	/* async queue for each priority case */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;

307
};
308

309 310 311 312
struct cfq_io_cq {
	struct io_cq		icq;		/* must be the first member */
	struct cfq_queue	*cfqq[2];
	struct cfq_ttime	ttime;
T
Tejun Heo 已提交
313 314
	int			ioprio;		/* the current ioprio */
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
315
	uint64_t		blkcg_serial_nr; /* the current blkcg serial */
T
Tejun Heo 已提交
316
#endif
317 318
};

319 320 321
/*
 * Per block device queue structure
 */
L
Linus Torvalds 已提交
322
struct cfq_data {
323
	struct request_queue *queue;
324 325
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
326
	struct cfq_group *root_group;
327

328 329
	/*
	 * The priority currently being served
330
	 */
331 332
	enum wl_class_t serving_wl_class;
	enum wl_type_t serving_wl_type;
333
	u64 workload_expires;
334
	struct cfq_group *serving_group;
335 336 337 338 339 340 341 342

	/*
	 * 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];

343
	unsigned int busy_queues;
344
	unsigned int busy_sync_queues;
345

346 347
	int rq_in_driver;
	int rq_in_flight[2];
348 349 350 351 352

	/*
	 * queue-depth detection
	 */
	int rq_queued;
353
	int hw_tag;
354 355 356 357 358 359 360 361
	/*
	 * 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;
L
Linus Torvalds 已提交
362

363 364 365
	/*
	 * idle window management
	 */
366
	struct hrtimer idle_slice_timer;
367
	struct work_struct unplug_work;
L
Linus Torvalds 已提交
368

369
	struct cfq_queue *active_queue;
370
	struct cfq_io_cq *active_cic;
371

J
Jens Axboe 已提交
372
	sector_t last_position;
L
Linus Torvalds 已提交
373 374 375 376 377 378 379

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
380
	unsigned int cfq_slice_async_rq;
381
	unsigned int cfq_latency;
382 383 384 385 386
	u64 cfq_fifo_expire[2];
	u64 cfq_slice[2];
	u64 cfq_slice_idle;
	u64 cfq_group_idle;
	u64 cfq_target_latency;
387

388 389 390 391
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
392

393
	u64 last_delayed_sync;
L
Linus Torvalds 已提交
394 395
};

396
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
397
static void cfq_put_queue(struct cfq_queue *cfqq);
398

399
static struct cfq_rb_root *st_for(struct cfq_group *cfqg,
400
					    enum wl_class_t class,
401
					    enum wl_type_t type)
402
{
403 404 405
	if (!cfqg)
		return NULL;

406
	if (class == IDLE_WORKLOAD)
407
		return &cfqg->service_tree_idle;
408

409
	return &cfqg->service_trees[class][type];
410 411
}

J
Jens Axboe 已提交
412
enum cfqq_state_flags {
413 414
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
415
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
416 417 418 419
	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 */
420
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
421
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
422
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
423
	CFQ_CFQQ_FLAG_split_coop,	/* shared cfqq will be splitted */
424
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
425
	CFQ_CFQQ_FLAG_wait_busy,	/* Waiting for next request */
J
Jens Axboe 已提交
426 427 428 429 430
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
431
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
J
Jens Axboe 已提交
432 433 434
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
435
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
J
Jens Axboe 已提交
436 437 438
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
439
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
J
Jens Axboe 已提交
440 441 442 443
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
444
CFQ_CFQQ_FNS(must_dispatch);
J
Jens Axboe 已提交
445 446 447 448
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
449
CFQ_CFQQ_FNS(slice_new);
450
CFQ_CFQQ_FNS(sync);
451
CFQ_CFQQ_FNS(coop);
452
CFQ_CFQQ_FNS(split_coop);
453
CFQ_CFQQ_FNS(deep);
454
CFQ_CFQQ_FNS(wait_busy);
J
Jens Axboe 已提交
455 456
#undef CFQ_CFQQ_FNS

457
#if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP)
458

459 460 461 462 463
/* cfqg stats flags */
enum cfqg_stats_flags {
	CFQG_stats_waiting = 0,
	CFQG_stats_idling,
	CFQG_stats_empty,
464 465
};

466 467
#define CFQG_FLAG_FNS(name)						\
static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats)	\
468
{									\
469
	stats->flags |= (1 << CFQG_stats_##name);			\
470
}									\
471
static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats)	\
472
{									\
473
	stats->flags &= ~(1 << CFQG_stats_##name);			\
474
}									\
475
static inline int cfqg_stats_##name(struct cfqg_stats *stats)		\
476
{									\
477
	return (stats->flags & (1 << CFQG_stats_##name)) != 0;		\
478 479
}									\

480 481 482 483
CFQG_FLAG_FNS(waiting)
CFQG_FLAG_FNS(idling)
CFQG_FLAG_FNS(empty)
#undef CFQG_FLAG_FNS
484 485

/* This should be called with the queue_lock held. */
486
static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats)
487 488 489
{
	unsigned long long now;

490
	if (!cfqg_stats_waiting(stats))
491 492 493 494 495 496
		return;

	now = sched_clock();
	if (time_after64(now, stats->start_group_wait_time))
		blkg_stat_add(&stats->group_wait_time,
			      now - stats->start_group_wait_time);
497
	cfqg_stats_clear_waiting(stats);
498 499 500
}

/* This should be called with the queue_lock held. */
501 502
static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg,
						 struct cfq_group *curr_cfqg)
503
{
504
	struct cfqg_stats *stats = &cfqg->stats;
505

506
	if (cfqg_stats_waiting(stats))
507
		return;
508
	if (cfqg == curr_cfqg)
509
		return;
510 511
	stats->start_group_wait_time = sched_clock();
	cfqg_stats_mark_waiting(stats);
512 513 514
}

/* This should be called with the queue_lock held. */
515
static void cfqg_stats_end_empty_time(struct cfqg_stats *stats)
516 517 518
{
	unsigned long long now;

519
	if (!cfqg_stats_empty(stats))
520 521 522 523 524 525
		return;

	now = sched_clock();
	if (time_after64(now, stats->start_empty_time))
		blkg_stat_add(&stats->empty_time,
			      now - stats->start_empty_time);
526
	cfqg_stats_clear_empty(stats);
527 528
}

529
static void cfqg_stats_update_dequeue(struct cfq_group *cfqg)
530
{
531
	blkg_stat_add(&cfqg->stats.dequeue, 1);
532 533
}

534
static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg)
535
{
536
	struct cfqg_stats *stats = &cfqg->stats;
537

538
	if (blkg_rwstat_total(&stats->queued))
539 540 541 542 543 544 545
		return;

	/*
	 * group is already marked empty. This can happen if cfqq got new
	 * request in parent group and moved to this group while being added
	 * to service tree. Just ignore the event and move on.
	 */
546
	if (cfqg_stats_empty(stats))
547 548 549
		return;

	stats->start_empty_time = sched_clock();
550
	cfqg_stats_mark_empty(stats);
551 552
}

553
static void cfqg_stats_update_idle_time(struct cfq_group *cfqg)
554
{
555
	struct cfqg_stats *stats = &cfqg->stats;
556

557
	if (cfqg_stats_idling(stats)) {
558 559 560 561 562
		unsigned long long now = sched_clock();

		if (time_after64(now, stats->start_idle_time))
			blkg_stat_add(&stats->idle_time,
				      now - stats->start_idle_time);
563
		cfqg_stats_clear_idling(stats);
564 565 566
	}
}

567
static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg)
568
{
569
	struct cfqg_stats *stats = &cfqg->stats;
570

571
	BUG_ON(cfqg_stats_idling(stats));
572 573

	stats->start_idle_time = sched_clock();
574
	cfqg_stats_mark_idling(stats);
575 576
}

577
static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg)
578
{
579
	struct cfqg_stats *stats = &cfqg->stats;
580 581

	blkg_stat_add(&stats->avg_queue_size_sum,
582
		      blkg_rwstat_total(&stats->queued));
583
	blkg_stat_add(&stats->avg_queue_size_samples, 1);
584
	cfqg_stats_update_group_wait_time(stats);
585 586 587 588
}

#else	/* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */

T
Tejun Heo 已提交
589 590 591 592 593 594 595
static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { }
static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { }
static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { }
static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { }
static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { }
static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { }
static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { }
596 597 598 599

#endif	/* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */

#ifdef CONFIG_CFQ_GROUP_IOSCHED
600

601 602 603 604 605 606 607 608
static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd)
{
	return pd ? container_of(pd, struct cfq_group, pd) : NULL;
}

static struct cfq_group_data
*cpd_to_cfqgd(struct blkcg_policy_data *cpd)
{
609
	return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL;
610 611 612 613 614 615 616
}

static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg)
{
	return pd_to_blkg(&cfqg->pd);
}

617 618 619 620 621 622 623
static struct blkcg_policy blkcg_policy_cfq;

static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg)
{
	return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq));
}

624 625 626 627 628
static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg)
{
	return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq));
}

629
static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg)
630
{
631
	struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent;
632

633
	return pblkg ? blkg_to_cfqg(pblkg) : NULL;
634 635
}

636 637 638 639 640 641 642
static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
				      struct cfq_group *ancestor)
{
	return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup,
				    cfqg_to_blkg(ancestor)->blkcg->css.cgroup);
}

643 644 645 646 647 648 649 650 651 652
static inline void cfqg_get(struct cfq_group *cfqg)
{
	return blkg_get(cfqg_to_blkg(cfqg));
}

static inline void cfqg_put(struct cfq_group *cfqg)
{
	return blkg_put(cfqg_to_blkg(cfqg));
}

T
Tejun Heo 已提交
653 654 655 656
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	do {			\
	char __pbuf[128];						\
									\
	blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf));	\
657 658 659
	blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \
			cfq_cfqq_sync((cfqq)) ? 'S' : 'A',		\
			cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
T
Tejun Heo 已提交
660 661 662 663 664 665 666 667 668
			  __pbuf, ##args);				\
} while (0)

#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)	do {			\
	char __pbuf[128];						\
									\
	blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf));		\
	blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args);	\
} while (0)
V
Vivek Goyal 已提交
669

670
static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
671 672
					    struct cfq_group *curr_cfqg,
					    unsigned int op)
673
{
674
	blkg_rwstat_add(&cfqg->stats.queued, op, 1);
675 676
	cfqg_stats_end_empty_time(&cfqg->stats);
	cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg);
677 678
}

679
static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
680
			uint64_t time, unsigned long unaccounted_time)
681
{
682
	blkg_stat_add(&cfqg->stats.time, time);
683
#ifdef CONFIG_DEBUG_BLK_CGROUP
684
	blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time);
685
#endif
686 687
}

688 689
static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg,
					       unsigned int op)
690
{
691
	blkg_rwstat_add(&cfqg->stats.queued, op, -1);
692 693
}

694 695
static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg,
					       unsigned int op)
696
{
697
	blkg_rwstat_add(&cfqg->stats.merged, op, 1);
698 699
}

700
static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
701 702
			uint64_t start_time, uint64_t io_start_time,
			unsigned int op)
703
{
704
	struct cfqg_stats *stats = &cfqg->stats;
705 706 707
	unsigned long long now = sched_clock();

	if (time_after64(now, io_start_time))
708
		blkg_rwstat_add(&stats->service_time, op, now - io_start_time);
709
	if (time_after64(io_start_time, start_time))
710
		blkg_rwstat_add(&stats->wait_time, op,
711
				io_start_time - start_time);
712 713
}

714 715
/* @stats = 0 */
static void cfqg_stats_reset(struct cfqg_stats *stats)
716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
{
	/* queued stats shouldn't be cleared */
	blkg_rwstat_reset(&stats->merged);
	blkg_rwstat_reset(&stats->service_time);
	blkg_rwstat_reset(&stats->wait_time);
	blkg_stat_reset(&stats->time);
#ifdef CONFIG_DEBUG_BLK_CGROUP
	blkg_stat_reset(&stats->unaccounted_time);
	blkg_stat_reset(&stats->avg_queue_size_sum);
	blkg_stat_reset(&stats->avg_queue_size_samples);
	blkg_stat_reset(&stats->dequeue);
	blkg_stat_reset(&stats->group_wait_time);
	blkg_stat_reset(&stats->idle_time);
	blkg_stat_reset(&stats->empty_time);
#endif
}

733
/* @to += @from */
734
static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from)
735 736
{
	/* queued stats shouldn't be cleared */
737 738 739 740
	blkg_rwstat_add_aux(&to->merged, &from->merged);
	blkg_rwstat_add_aux(&to->service_time, &from->service_time);
	blkg_rwstat_add_aux(&to->wait_time, &from->wait_time);
	blkg_stat_add_aux(&from->time, &from->time);
741
#ifdef CONFIG_DEBUG_BLK_CGROUP
742 743 744 745 746 747 748
	blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time);
	blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
	blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples);
	blkg_stat_add_aux(&to->dequeue, &from->dequeue);
	blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time);
	blkg_stat_add_aux(&to->idle_time, &from->idle_time);
	blkg_stat_add_aux(&to->empty_time, &from->empty_time);
749 750 751 752
#endif
}

/*
753
 * Transfer @cfqg's stats to its parent's aux counts so that the ancestors'
754 755 756 757 758 759 760 761 762 763 764 765
 * recursive stats can still account for the amount used by this cfqg after
 * it's gone.
 */
static void cfqg_stats_xfer_dead(struct cfq_group *cfqg)
{
	struct cfq_group *parent = cfqg_parent(cfqg);

	lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock);

	if (unlikely(!parent))
		return;

766
	cfqg_stats_add_aux(&parent->stats, &cfqg->stats);
767 768 769
	cfqg_stats_reset(&cfqg->stats);
}

770 771
#else	/* CONFIG_CFQ_GROUP_IOSCHED */

772
static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; }
773 774 775 776 777
static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
				      struct cfq_group *ancestor)
{
	return true;
}
778 779 780
static inline void cfqg_get(struct cfq_group *cfqg) { }
static inline void cfqg_put(struct cfq_group *cfqg) { }

781
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
782 783 784 785
	blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid,	\
			cfq_cfqq_sync((cfqq)) ? 'S' : 'A',		\
			cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
				##args)
786
#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0)
787

788
static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
789
			struct cfq_group *curr_cfqg, unsigned int op) { }
790
static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
791
			uint64_t time, unsigned long unaccounted_time) { }
792 793 794 795
static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg,
			unsigned int op) { }
static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg,
			unsigned int op) { }
796
static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
797 798
			uint64_t start_time, uint64_t io_start_time,
			unsigned int op) { }
799

800 801
#endif	/* CONFIG_CFQ_GROUP_IOSCHED */

802 803 804
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

805 806 807 808 809 810 811 812 813 814
/* 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) \

815 816 817
static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd,
	struct cfq_ttime *ttime, bool group_idle)
{
818
	u64 slice;
819 820 821 822 823 824 825 826
	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;
}
827

828 829 830 831 832 833 834 835 836 837 838 839 840 841 842
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;
}

843
static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq)
844 845 846 847 848 849 850 851
{
	if (cfq_class_idle(cfqq))
		return IDLE_WORKLOAD;
	if (cfq_class_rt(cfqq))
		return RT_WORKLOAD;
	return BE_WORKLOAD;
}

852 853 854 855 856 857 858 859 860 861

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;
}

862
static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class,
863 864
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
865
{
866
	if (wl_class == IDLE_WORKLOAD)
867
		return cfqg->service_tree_idle.count;
868

869 870 871
	return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count +
		cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count +
		cfqg->service_trees[wl_class][SYNC_WORKLOAD].count;
872 873
}

874 875 876
static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg)
{
877 878
	return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count +
		cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
879 880
}

881
static void cfq_dispatch_insert(struct request_queue *, struct request *);
882
static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync,
883
				       struct cfq_io_cq *cic, struct bio *bio);
884

885 886 887 888 889 890
static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq)
{
	/* cic->icq is the first member, %NULL will convert to %NULL */
	return container_of(icq, struct cfq_io_cq, icq);
}

891 892 893 894 895 896 897 898
static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd,
					       struct io_context *ioc)
{
	if (ioc)
		return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue));
	return NULL;
}

899
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync)
900
{
901
	return cic->cfqq[is_sync];
902 903
}

904 905
static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq,
				bool is_sync)
906
{
907
	cic->cfqq[is_sync] = cfqq;
908 909
}

910
static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic)
911
{
912
	return cic->icq.q->elevator->elevator_data;
913 914
}

A
Andrew Morton 已提交
915 916 917 918
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
919
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
A
Andrew Morton 已提交
920
{
921 922
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
923
		kblockd_schedule_work(&cfqd->unplug_work);
924
	}
A
Andrew Morton 已提交
925 926
}

927 928 929 930 931
/*
 * 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.
 */
932
static inline u64 cfq_prio_slice(struct cfq_data *cfqd, bool sync,
933
				 unsigned short prio)
934
{
935 936
	u64 base_slice = cfqd->cfq_slice[sync];
	u64 slice = div_u64(base_slice, CFQ_SLICE_SCALE);
937

938 939
	WARN_ON(prio >= IOPRIO_BE_NR);

940
	return base_slice + (slice * (4 - prio));
941
}
942

943
static inline u64
944 945 946
cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
947 948
}

949 950 951 952 953 954 955 956 957 958 959 960
/**
 * cfqg_scale_charge - scale disk time charge according to cfqg weight
 * @charge: disk time being charged
 * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT
 *
 * Scale @charge according to @vfraction, which is in range (0, 1].  The
 * scaling is inversely proportional.
 *
 * scaled = charge / vfraction
 *
 * The result is also in fixed point w/ CFQ_SERVICE_SHIFT.
 */
961
static inline u64 cfqg_scale_charge(u64 charge,
962
				    unsigned int vfraction)
963
{
964
	u64 c = charge << CFQ_SERVICE_SHIFT;	/* make it fixed point */
965

966 967
	/* charge / vfraction */
	c <<= CFQ_SERVICE_SHIFT;
968
	return div_u64(c, vfraction);
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994
}

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);
995 996
		st->min_vdisktime = max_vdisktime(st->min_vdisktime,
						  cfqg->vdisktime);
997 998 999
	}
}

1000 1001 1002 1003 1004 1005
/*
 * 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
 */

1006 1007
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
1008
{
1009 1010 1011
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
1012
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
1013

1014 1015 1016
	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) /
1017
		cfq_hist_divisor;
1018 1019 1020
	return cfqg->busy_queues_avg[rt];
}

1021
static inline u64
1022 1023
cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
1024
	return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT;
1025 1026
}

1027
static inline u64
1028
cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1029
{
1030
	u64 slice = cfq_prio_to_slice(cfqd, cfqq);
1031
	if (cfqd->cfq_latency) {
1032 1033 1034 1035 1036 1037
		/*
		 * 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));
1038 1039 1040
		u64 sync_slice = cfqd->cfq_slice[1];
		u64 expect_latency = sync_slice * iq;
		u64 group_slice = cfq_group_slice(cfqd, cfqq->cfqg);
1041 1042

		if (expect_latency > group_slice) {
1043 1044 1045
			u64 base_low_slice = 2 * cfqd->cfq_slice_idle;
			u64 low_slice;

1046 1047
			/* scale low_slice according to IO priority
			 * and sync vs async */
1048 1049
			low_slice = div64_u64(base_low_slice*slice, sync_slice);
			low_slice = min(slice, low_slice);
1050 1051
			/* the adapted slice value is scaled to fit all iqs
			 * into the target latency */
1052 1053
			slice = div64_u64(slice*group_slice, expect_latency);
			slice = max(slice, low_slice);
1054 1055
		}
	}
1056 1057 1058 1059 1060 1061
	return slice;
}

static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
1062 1063
	u64 slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
	u64 now = ktime_get_ns();
1064

1065 1066
	cfqq->slice_start = now;
	cfqq->slice_end = now + slice;
1067
	cfqq->allocated_slice = slice;
1068
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%llu", cfqq->slice_end - now);
1069 1070 1071 1072 1073 1074 1075
}

/*
 * 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.
 */
1076
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
1077 1078
{
	if (cfq_cfqq_slice_new(cfqq))
S
Shaohua Li 已提交
1079
		return false;
1080
	if (ktime_get_ns() < cfqq->slice_end)
S
Shaohua Li 已提交
1081
		return false;
1082

S
Shaohua Li 已提交
1083
	return true;
1084 1085
}

L
Linus Torvalds 已提交
1086
/*
J
Jens Axboe 已提交
1087
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
L
Linus Torvalds 已提交
1088
 * We choose the request that is closest to the head right now. Distance
1089
 * behind the head is penalized and only allowed to a certain extent.
L
Linus Torvalds 已提交
1090
 */
J
Jens Axboe 已提交
1091
static struct request *
1092
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
L
Linus Torvalds 已提交
1093
{
1094
	sector_t s1, s2, d1 = 0, d2 = 0;
L
Linus Torvalds 已提交
1095
	unsigned long back_max;
1096 1097 1098
#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? */
L
Linus Torvalds 已提交
1099

J
Jens Axboe 已提交
1100 1101 1102 1103
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
J
Jens Axboe 已提交
1104

1105 1106 1107
	if (rq_is_sync(rq1) != rq_is_sync(rq2))
		return rq_is_sync(rq1) ? rq1 : rq2;

1108 1109
	if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
		return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
1110

1111 1112
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
L
Linus Torvalds 已提交
1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128

	/*
	 * 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
1129
		wrap |= CFQ_RQ1_WRAP;
L
Linus Torvalds 已提交
1130 1131 1132 1133 1134 1135

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
1136
		wrap |= CFQ_RQ2_WRAP;
L
Linus Torvalds 已提交
1137 1138

	/* Found required data */
1139 1140 1141 1142 1143 1144

	/*
	 * By doing switch() on the bit mask "wrap" we avoid having to
	 * check two variables for all permutations: --> faster!
	 */
	switch (wrap) {
J
Jens Axboe 已提交
1145
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
1146
		if (d1 < d2)
J
Jens Axboe 已提交
1147
			return rq1;
1148
		else if (d2 < d1)
J
Jens Axboe 已提交
1149
			return rq2;
1150 1151
		else {
			if (s1 >= s2)
J
Jens Axboe 已提交
1152
				return rq1;
1153
			else
J
Jens Axboe 已提交
1154
				return rq2;
1155
		}
L
Linus Torvalds 已提交
1156

1157
	case CFQ_RQ2_WRAP:
J
Jens Axboe 已提交
1158
		return rq1;
1159
	case CFQ_RQ1_WRAP:
J
Jens Axboe 已提交
1160 1161
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
1162 1163 1164 1165 1166 1167 1168 1169
	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)
J
Jens Axboe 已提交
1170
			return rq1;
L
Linus Torvalds 已提交
1171
		else
J
Jens Axboe 已提交
1172
			return rq2;
L
Linus Torvalds 已提交
1173 1174 1175
	}
}

1176 1177 1178
/*
 * The below is leftmost cache rbtree addon
 */
1179
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
1180
{
1181 1182 1183 1184
	/* Service tree is empty */
	if (!root->count)
		return NULL;

1185 1186 1187
	if (!root->left)
		root->left = rb_first(&root->rb);

1188 1189 1190 1191
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
1192 1193
}

1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
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;
}

1205 1206 1207 1208 1209 1210
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

1211 1212 1213 1214
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
1215
	rb_erase_init(n, &root->rb);
1216
	--root->count;
1217 1218
}

L
Linus Torvalds 已提交
1219 1220 1221
/*
 * would be nice to take fifo expire time into account as well
 */
J
Jens Axboe 已提交
1222 1223 1224
static struct request *
cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		  struct request *last)
L
Linus Torvalds 已提交
1225
{
1226 1227
	struct rb_node *rbnext = rb_next(&last->rb_node);
	struct rb_node *rbprev = rb_prev(&last->rb_node);
J
Jens Axboe 已提交
1228
	struct request *next = NULL, *prev = NULL;
L
Linus Torvalds 已提交
1229

1230
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
L
Linus Torvalds 已提交
1231 1232

	if (rbprev)
J
Jens Axboe 已提交
1233
		prev = rb_entry_rq(rbprev);
L
Linus Torvalds 已提交
1234

1235
	if (rbnext)
J
Jens Axboe 已提交
1236
		next = rb_entry_rq(rbnext);
1237 1238 1239
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
J
Jens Axboe 已提交
1240
			next = rb_entry_rq(rbnext);
1241
	}
L
Linus Torvalds 已提交
1242

1243
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
L
Linus Torvalds 已提交
1244 1245
}

1246 1247
static u64 cfq_slice_offset(struct cfq_data *cfqd,
			    struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
1248
{
1249 1250 1251
	/*
	 * just an approximation, should be ok.
	 */
1252
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
1253
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
1254 1255
}

1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289
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);
}

1290 1291 1292
/*
 * This has to be called only on activation of cfqg
 */
1293
static void
1294 1295
cfq_update_group_weight(struct cfq_group *cfqg)
{
1296
	if (cfqg->new_weight) {
1297
		cfqg->weight = cfqg->new_weight;
1298
		cfqg->new_weight = 0;
1299
	}
1300 1301 1302 1303 1304 1305
}

static void
cfq_update_group_leaf_weight(struct cfq_group *cfqg)
{
	BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
T
Tejun Heo 已提交
1306 1307 1308 1309 1310

	if (cfqg->new_leaf_weight) {
		cfqg->leaf_weight = cfqg->new_leaf_weight;
		cfqg->new_leaf_weight = 0;
	}
1311 1312 1313 1314 1315
}

static void
cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
1316
	unsigned int vfr = 1 << CFQ_SERVICE_SHIFT;	/* start with 1 */
1317
	struct cfq_group *pos = cfqg;
1318
	struct cfq_group *parent;
1319 1320 1321
	bool propagate;

	/* add to the service tree */
1322 1323
	BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));

1324 1325 1326 1327 1328
	/*
	 * Update leaf_weight.  We cannot update weight at this point
	 * because cfqg might already have been activated and is
	 * contributing its current weight to the parent's child_weight.
	 */
1329
	cfq_update_group_leaf_weight(cfqg);
1330
	__cfq_group_service_tree_add(st, cfqg);
1331 1332

	/*
1333 1334 1335 1336 1337 1338 1339
	 * Activate @cfqg and calculate the portion of vfraction @cfqg is
	 * entitled to.  vfraction is calculated by walking the tree
	 * towards the root calculating the fraction it has at each level.
	 * The compounded ratio is how much vfraction @cfqg owns.
	 *
	 * Start with the proportion tasks in this cfqg has against active
	 * children cfqgs - its leaf_weight against children_weight.
1340 1341 1342
	 */
	propagate = !pos->nr_active++;
	pos->children_weight += pos->leaf_weight;
1343
	vfr = vfr * pos->leaf_weight / pos->children_weight;
1344

1345 1346 1347 1348 1349 1350
	/*
	 * Compound ->weight walking up the tree.  Both activation and
	 * vfraction calculation are done in the same loop.  Propagation
	 * stops once an already activated node is met.  vfraction
	 * calculation should always continue to the root.
	 */
1351
	while ((parent = cfqg_parent(pos))) {
1352
		if (propagate) {
1353
			cfq_update_group_weight(pos);
1354 1355 1356 1357
			propagate = !parent->nr_active++;
			parent->children_weight += pos->weight;
		}
		vfr = vfr * pos->weight / parent->children_weight;
1358 1359
		pos = parent;
	}
1360 1361

	cfqg->vfraction = max_t(unsigned, vfr, 1);
1362 1363 1364 1365
}

static void
cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
1366 1367 1368 1369 1370 1371
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
	struct cfq_group *__cfqg;
	struct rb_node *n;

	cfqg->nr_cfqq++;
G
Gui Jianfeng 已提交
1372
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
1373 1374 1375 1376 1377
		return;

	/*
	 * Currently put the group at the end. Later implement something
	 * so that groups get lesser vtime based on their weights, so that
L
Lucas De Marchi 已提交
1378
	 * if group does not loose all if it was not continuously backlogged.
1379 1380 1381 1382 1383 1384 1385
	 */
	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;
1386 1387
	cfq_group_service_tree_add(st, cfqg);
}
1388

1389 1390 1391
static void
cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
	struct cfq_group *pos = cfqg;
	bool propagate;

	/*
	 * Undo activation from cfq_group_service_tree_add().  Deactivate
	 * @cfqg and propagate deactivation upwards.
	 */
	propagate = !--pos->nr_active;
	pos->children_weight -= pos->leaf_weight;

	while (propagate) {
1403
		struct cfq_group *parent = cfqg_parent(pos);
1404 1405 1406

		/* @pos has 0 nr_active at this point */
		WARN_ON_ONCE(pos->children_weight);
1407
		pos->vfraction = 0;
1408 1409 1410 1411 1412 1413 1414 1415 1416 1417

		if (!parent)
			break;

		propagate = !--parent->nr_active;
		parent->children_weight -= pos->weight;
		pos = parent;
	}

	/* remove from the service tree */
1418 1419
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
1420 1421 1422
}

static void
1423
cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
1424 1425 1426 1427 1428
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
1429

1430 1431 1432 1433
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

V
Vivek Goyal 已提交
1434
	cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
1435
	cfq_group_service_tree_del(st, cfqg);
1436
	cfqg->saved_wl_slice = 0;
1437
	cfqg_stats_update_dequeue(cfqg);
1438 1439
}

1440 1441
static inline u64 cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
				       u64 *unaccounted_time)
1442
{
1443 1444
	u64 slice_used;
	u64 now = ktime_get_ns();
1445 1446 1447 1448 1449

	/*
	 * Queue got expired before even a single request completed or
	 * got expired immediately after first request completion.
	 */
1450
	if (!cfqq->slice_start || cfqq->slice_start == now) {
1451 1452 1453 1454 1455 1456
		/*
		 * 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.
		 */
1457 1458
		slice_used = max_t(u64, (now - cfqq->dispatch_start),
					jiffies_to_nsecs(1));
1459
	} else {
1460
		slice_used = now - cfqq->slice_start;
1461 1462
		if (slice_used > cfqq->allocated_slice) {
			*unaccounted_time = slice_used - cfqq->allocated_slice;
1463
			slice_used = cfqq->allocated_slice;
1464
		}
1465
		if (cfqq->slice_start > cfqq->dispatch_start)
1466 1467
			*unaccounted_time += cfqq->slice_start -
					cfqq->dispatch_start;
1468 1469 1470 1471 1472 1473
	}

	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
1474
				struct cfq_queue *cfqq)
1475 1476
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
1477
	u64 used_sl, charge, unaccounted_sl = 0;
1478 1479
	int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
			- cfqg->service_tree_idle.count;
1480
	unsigned int vfr;
1481
	u64 now = ktime_get_ns();
1482 1483

	BUG_ON(nr_sync < 0);
1484
	used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
1485

1486 1487 1488 1489
	if (iops_mode(cfqd))
		charge = cfqq->slice_dispatch;
	else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
		charge = cfqq->allocated_slice;
1490

1491 1492 1493 1494 1495 1496 1497
	/*
	 * Can't update vdisktime while on service tree and cfqg->vfraction
	 * is valid only while on it.  Cache vfr, leave the service tree,
	 * update vdisktime and go back on.  The re-addition to the tree
	 * will also update the weights as necessary.
	 */
	vfr = cfqg->vfraction;
1498
	cfq_group_service_tree_del(st, cfqg);
1499
	cfqg->vdisktime += cfqg_scale_charge(charge, vfr);
1500
	cfq_group_service_tree_add(st, cfqg);
1501 1502

	/* This group is being expired. Save the context */
1503 1504
	if (cfqd->workload_expires > now) {
		cfqg->saved_wl_slice = cfqd->workload_expires - now;
1505 1506
		cfqg->saved_wl_type = cfqd->serving_wl_type;
		cfqg->saved_wl_class = cfqd->serving_wl_class;
1507
	} else
1508
		cfqg->saved_wl_slice = 0;
V
Vivek Goyal 已提交
1509 1510 1511

	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
					st->min_vdisktime);
1512
	cfq_log_cfqq(cfqq->cfqd, cfqq,
1513
		     "sl_used=%llu disp=%llu charge=%llu iops=%u sect=%lu",
1514 1515
		     used_sl, cfqq->slice_dispatch, charge,
		     iops_mode(cfqd), cfqq->nr_sectors);
1516 1517
	cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl);
	cfqg_stats_set_start_empty_time(cfqg);
1518 1519
}

1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535
/**
 * cfq_init_cfqg_base - initialize base part of a cfq_group
 * @cfqg: cfq_group to initialize
 *
 * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED
 * is enabled or not.
 */
static void cfq_init_cfqg_base(struct cfq_group *cfqg)
{
	struct cfq_rb_root *st;
	int i, j;

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

1536
	cfqg->ttime.last_end_request = ktime_get_ns();
1537 1538
}

1539
#ifdef CONFIG_CFQ_GROUP_IOSCHED
1540 1541 1542
static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
			    bool on_dfl, bool reset_dev, bool is_leaf_weight);

T
Tejun Heo 已提交
1543
static void cfqg_stats_exit(struct cfqg_stats *stats)
1544
{
T
Tejun Heo 已提交
1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562
	blkg_rwstat_exit(&stats->merged);
	blkg_rwstat_exit(&stats->service_time);
	blkg_rwstat_exit(&stats->wait_time);
	blkg_rwstat_exit(&stats->queued);
	blkg_stat_exit(&stats->time);
#ifdef CONFIG_DEBUG_BLK_CGROUP
	blkg_stat_exit(&stats->unaccounted_time);
	blkg_stat_exit(&stats->avg_queue_size_sum);
	blkg_stat_exit(&stats->avg_queue_size_samples);
	blkg_stat_exit(&stats->dequeue);
	blkg_stat_exit(&stats->group_wait_time);
	blkg_stat_exit(&stats->idle_time);
	blkg_stat_exit(&stats->empty_time);
#endif
}

static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp)
{
1563
	if (blkg_rwstat_init(&stats->merged, gfp) ||
T
Tejun Heo 已提交
1564 1565 1566 1567 1568
	    blkg_rwstat_init(&stats->service_time, gfp) ||
	    blkg_rwstat_init(&stats->wait_time, gfp) ||
	    blkg_rwstat_init(&stats->queued, gfp) ||
	    blkg_stat_init(&stats->time, gfp))
		goto err;
1569 1570

#ifdef CONFIG_DEBUG_BLK_CGROUP
T
Tejun Heo 已提交
1571 1572 1573 1574 1575 1576 1577 1578
	if (blkg_stat_init(&stats->unaccounted_time, gfp) ||
	    blkg_stat_init(&stats->avg_queue_size_sum, gfp) ||
	    blkg_stat_init(&stats->avg_queue_size_samples, gfp) ||
	    blkg_stat_init(&stats->dequeue, gfp) ||
	    blkg_stat_init(&stats->group_wait_time, gfp) ||
	    blkg_stat_init(&stats->idle_time, gfp) ||
	    blkg_stat_init(&stats->empty_time, gfp))
		goto err;
1579
#endif
T
Tejun Heo 已提交
1580 1581 1582 1583
	return 0;
err:
	cfqg_stats_exit(stats);
	return -ENOMEM;
1584 1585
}

1586 1587 1588 1589
static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp)
{
	struct cfq_group_data *cgd;

T
Tejun Heo 已提交
1590
	cgd = kzalloc(sizeof(*cgd), gfp);
1591 1592 1593 1594 1595
	if (!cgd)
		return NULL;
	return &cgd->cpd;
}

1596
static void cfq_cpd_init(struct blkcg_policy_data *cpd)
1597
{
1598
	struct cfq_group_data *cgd = cpd_to_cfqgd(cpd);
1599
	unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ?
1600
			      CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
1601

1602 1603 1604 1605 1606
	if (cpd_to_blkcg(cpd) == &blkcg_root)
		weight *= 2;

	cgd->weight = weight;
	cgd->leaf_weight = weight;
1607 1608
}

1609 1610 1611 1612 1613
static void cfq_cpd_free(struct blkcg_policy_data *cpd)
{
	kfree(cpd_to_cfqgd(cpd));
}

1614 1615 1616
static void cfq_cpd_bind(struct blkcg_policy_data *cpd)
{
	struct blkcg *blkcg = cpd_to_blkcg(cpd);
1617
	bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys);
1618 1619 1620 1621 1622 1623 1624 1625 1626
	unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;

	if (blkcg == &blkcg_root)
		weight *= 2;

	WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false));
	WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true));
}

1627 1628
static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node)
{
1629 1630 1631 1632 1633 1634 1635
	struct cfq_group *cfqg;

	cfqg = kzalloc_node(sizeof(*cfqg), gfp, node);
	if (!cfqg)
		return NULL;

	cfq_init_cfqg_base(cfqg);
T
Tejun Heo 已提交
1636 1637 1638 1639
	if (cfqg_stats_init(&cfqg->stats, gfp)) {
		kfree(cfqg);
		return NULL;
	}
1640 1641

	return &cfqg->pd;
1642 1643
}

1644
static void cfq_pd_init(struct blkg_policy_data *pd)
1645
{
1646 1647
	struct cfq_group *cfqg = pd_to_cfqg(pd);
	struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg);
1648

1649 1650
	cfqg->weight = cgd->weight;
	cfqg->leaf_weight = cgd->leaf_weight;
1651 1652
}

1653
static void cfq_pd_offline(struct blkg_policy_data *pd)
1654
{
1655
	struct cfq_group *cfqg = pd_to_cfqg(pd);
1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667
	int i;

	for (i = 0; i < IOPRIO_BE_NR; i++) {
		if (cfqg->async_cfqq[0][i])
			cfq_put_queue(cfqg->async_cfqq[0][i]);
		if (cfqg->async_cfqq[1][i])
			cfq_put_queue(cfqg->async_cfqq[1][i]);
	}

	if (cfqg->async_idle_cfqq)
		cfq_put_queue(cfqg->async_idle_cfqq);

1668 1669 1670 1671 1672 1673
	/*
	 * @blkg is going offline and will be ignored by
	 * blkg_[rw]stat_recursive_sum().  Transfer stats to the parent so
	 * that they don't get lost.  If IOs complete after this point, the
	 * stats for them will be lost.  Oh well...
	 */
1674
	cfqg_stats_xfer_dead(cfqg);
1675 1676
}

1677 1678
static void cfq_pd_free(struct blkg_policy_data *pd)
{
T
Tejun Heo 已提交
1679 1680 1681 1682
	struct cfq_group *cfqg = pd_to_cfqg(pd);

	cfqg_stats_exit(&cfqg->stats);
	return kfree(cfqg);
1683 1684
}

1685
static void cfq_pd_reset_stats(struct blkg_policy_data *pd)
1686
{
1687
	struct cfq_group *cfqg = pd_to_cfqg(pd);
1688 1689

	cfqg_stats_reset(&cfqg->stats);
1690 1691
}

1692 1693
static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
					 struct blkcg *blkcg)
1694
{
1695
	struct blkcg_gq *blkg;
1696

1697 1698 1699 1700
	blkg = blkg_lookup(blkcg, cfqd->queue);
	if (likely(blkg))
		return blkg_to_cfqg(blkg);
	return NULL;
1701 1702 1703 1704 1705
}

static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg)
{
	cfqq->cfqg = cfqg;
1706
	/* cfqq reference on cfqg */
1707
	cfqg_get(cfqg);
1708 1709
}

1710 1711
static u64 cfqg_prfill_weight_device(struct seq_file *sf,
				     struct blkg_policy_data *pd, int off)
1712
{
1713
	struct cfq_group *cfqg = pd_to_cfqg(pd);
1714 1715

	if (!cfqg->dev_weight)
1716
		return 0;
1717
	return __blkg_prfill_u64(sf, pd, cfqg->dev_weight);
1718 1719
}

1720
static int cfqg_print_weight_device(struct seq_file *sf, void *v)
1721
{
1722 1723 1724
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_weight_device, &blkcg_policy_cfq,
			  0, false);
1725 1726 1727
	return 0;
}

T
Tejun Heo 已提交
1728 1729 1730 1731 1732 1733 1734 1735 1736 1737
static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf,
					  struct blkg_policy_data *pd, int off)
{
	struct cfq_group *cfqg = pd_to_cfqg(pd);

	if (!cfqg->dev_leaf_weight)
		return 0;
	return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight);
}

1738
static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v)
T
Tejun Heo 已提交
1739
{
1740 1741 1742
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq,
			  0, false);
T
Tejun Heo 已提交
1743 1744 1745
	return 0;
}

1746
static int cfq_print_weight(struct seq_file *sf, void *v)
1747
{
1748
	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1749 1750
	struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
	unsigned int val = 0;
1751

1752 1753 1754 1755
	if (cgd)
		val = cgd->weight;

	seq_printf(sf, "%u\n", val);
1756 1757 1758
	return 0;
}

1759
static int cfq_print_leaf_weight(struct seq_file *sf, void *v)
T
Tejun Heo 已提交
1760
{
1761
	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1762 1763 1764 1765 1766
	struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
	unsigned int val = 0;

	if (cgd)
		val = cgd->leaf_weight;
1767

1768
	seq_printf(sf, "%u\n", val);
T
Tejun Heo 已提交
1769 1770 1771
	return 0;
}

1772 1773
static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of,
					char *buf, size_t nbytes, loff_t off,
1774
					bool on_dfl, bool is_leaf_weight)
1775
{
1776 1777
	unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
	unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
1778
	struct blkcg *blkcg = css_to_blkcg(of_css(of));
1779
	struct blkg_conf_ctx ctx;
1780
	struct cfq_group *cfqg;
1781
	struct cfq_group_data *cfqgd;
1782
	int ret;
1783
	u64 v;
1784

T
Tejun Heo 已提交
1785
	ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx);
1786 1787 1788
	if (ret)
		return ret;

1789 1790 1791 1792 1793 1794 1795 1796 1797
	if (sscanf(ctx.body, "%llu", &v) == 1) {
		/* require "default" on dfl */
		ret = -ERANGE;
		if (!v && on_dfl)
			goto out_finish;
	} else if (!strcmp(strim(ctx.body), "default")) {
		v = 0;
	} else {
		ret = -EINVAL;
1798
		goto out_finish;
1799
	}
1800

1801
	cfqg = blkg_to_cfqg(ctx.blkg);
1802
	cfqgd = blkcg_to_cfqgd(blkcg);
1803

1804
	ret = -ERANGE;
1805
	if (!v || (v >= min && v <= max)) {
T
Tejun Heo 已提交
1806
		if (!is_leaf_weight) {
1807 1808
			cfqg->dev_weight = v;
			cfqg->new_weight = v ?: cfqgd->weight;
T
Tejun Heo 已提交
1809
		} else {
1810 1811
			cfqg->dev_leaf_weight = v;
			cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight;
T
Tejun Heo 已提交
1812
		}
1813 1814
		ret = 0;
	}
1815
out_finish:
1816
	blkg_conf_finish(&ctx);
1817
	return ret ?: nbytes;
1818 1819
}

1820 1821
static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of,
				      char *buf, size_t nbytes, loff_t off)
T
Tejun Heo 已提交
1822
{
1823
	return __cfqg_set_weight_device(of, buf, nbytes, off, false, false);
T
Tejun Heo 已提交
1824 1825
}

1826 1827
static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of,
					   char *buf, size_t nbytes, loff_t off)
T
Tejun Heo 已提交
1828
{
1829
	return __cfqg_set_weight_device(of, buf, nbytes, off, false, true);
T
Tejun Heo 已提交
1830 1831
}

1832
static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
1833
			    bool on_dfl, bool reset_dev, bool is_leaf_weight)
1834
{
1835 1836
	unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
	unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
1837
	struct blkcg *blkcg = css_to_blkcg(css);
T
Tejun Heo 已提交
1838
	struct blkcg_gq *blkg;
1839
	struct cfq_group_data *cfqgd;
1840
	int ret = 0;
1841

1842 1843
	if (val < min || val > max)
		return -ERANGE;
1844 1845

	spin_lock_irq(&blkcg->lock);
1846
	cfqgd = blkcg_to_cfqgd(blkcg);
1847 1848 1849 1850
	if (!cfqgd) {
		ret = -EINVAL;
		goto out;
	}
T
Tejun Heo 已提交
1851 1852

	if (!is_leaf_weight)
1853
		cfqgd->weight = val;
T
Tejun Heo 已提交
1854
	else
1855
		cfqgd->leaf_weight = val;
1856

1857
	hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
1858
		struct cfq_group *cfqg = blkg_to_cfqg(blkg);
1859

T
Tejun Heo 已提交
1860 1861 1862 1863
		if (!cfqg)
			continue;

		if (!is_leaf_weight) {
1864 1865
			if (reset_dev)
				cfqg->dev_weight = 0;
T
Tejun Heo 已提交
1866
			if (!cfqg->dev_weight)
1867
				cfqg->new_weight = cfqgd->weight;
T
Tejun Heo 已提交
1868
		} else {
1869 1870
			if (reset_dev)
				cfqg->dev_leaf_weight = 0;
T
Tejun Heo 已提交
1871
			if (!cfqg->dev_leaf_weight)
1872
				cfqg->new_leaf_weight = cfqgd->leaf_weight;
T
Tejun Heo 已提交
1873
		}
1874 1875
	}

1876
out:
1877
	spin_unlock_irq(&blkcg->lock);
1878
	return ret;
1879 1880
}

1881 1882
static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft,
			  u64 val)
T
Tejun Heo 已提交
1883
{
1884
	return __cfq_set_weight(css, val, false, false, false);
T
Tejun Heo 已提交
1885 1886
}

1887 1888
static int cfq_set_leaf_weight(struct cgroup_subsys_state *css,
			       struct cftype *cft, u64 val)
T
Tejun Heo 已提交
1889
{
1890
	return __cfq_set_weight(css, val, false, false, true);
T
Tejun Heo 已提交
1891 1892
}

1893
static int cfqg_print_stat(struct seq_file *sf, void *v)
1894
{
1895 1896
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat,
			  &blkcg_policy_cfq, seq_cft(sf)->private, false);
1897 1898 1899
	return 0;
}

1900
static int cfqg_print_rwstat(struct seq_file *sf, void *v)
1901
{
1902 1903
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat,
			  &blkcg_policy_cfq, seq_cft(sf)->private, true);
1904 1905 1906
	return 0;
}

1907 1908 1909
static u64 cfqg_prfill_stat_recursive(struct seq_file *sf,
				      struct blkg_policy_data *pd, int off)
{
1910 1911
	u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd),
					  &blkcg_policy_cfq, off);
1912 1913 1914 1915 1916 1917
	return __blkg_prfill_u64(sf, pd, sum);
}

static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf,
					struct blkg_policy_data *pd, int off)
{
1918 1919
	struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd),
							&blkcg_policy_cfq, off);
1920 1921 1922
	return __blkg_prfill_rwstat(sf, pd, &sum);
}

1923
static int cfqg_print_stat_recursive(struct seq_file *sf, void *v)
1924
{
1925 1926 1927
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_stat_recursive, &blkcg_policy_cfq,
			  seq_cft(sf)->private, false);
1928 1929 1930
	return 0;
}

1931
static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v)
1932
{
1933 1934 1935
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq,
			  seq_cft(sf)->private, true);
1936 1937 1938
	return 0;
}

T
Tejun Heo 已提交
1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972
static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd,
			       int off)
{
	u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes);

	return __blkg_prfill_u64(sf, pd, sum >> 9);
}

static int cfqg_print_stat_sectors(struct seq_file *sf, void *v)
{
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false);
	return 0;
}

static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf,
					 struct blkg_policy_data *pd, int off)
{
	struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL,
					offsetof(struct blkcg_gq, stat_bytes));
	u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
		atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);

	return __blkg_prfill_u64(sf, pd, sum >> 9);
}

static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v)
{
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0,
			  false);
	return 0;
}

1973
#ifdef CONFIG_DEBUG_BLK_CGROUP
1974 1975
static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf,
				      struct blkg_policy_data *pd, int off)
1976
{
1977
	struct cfq_group *cfqg = pd_to_cfqg(pd);
1978
	u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples);
1979 1980 1981
	u64 v = 0;

	if (samples) {
1982
		v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum);
1983
		v = div64_u64(v, samples);
1984
	}
1985
	__blkg_prfill_u64(sf, pd, v);
1986 1987 1988 1989
	return 0;
}

/* print avg_queue_size */
1990
static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v)
1991
{
1992 1993 1994
	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
			  cfqg_prfill_avg_queue_size, &blkcg_policy_cfq,
			  0, false);
1995 1996 1997 1998
	return 0;
}
#endif	/* CONFIG_DEBUG_BLK_CGROUP */

1999
static struct cftype cfq_blkcg_legacy_files[] = {
2000
	/* on root, weight is mapped to leaf_weight */
2001 2002
	{
		.name = "weight_device",
2003
		.flags = CFTYPE_ONLY_ON_ROOT,
2004
		.seq_show = cfqg_print_leaf_weight_device,
2005
		.write = cfqg_set_leaf_weight_device,
2006 2007 2008
	},
	{
		.name = "weight",
2009
		.flags = CFTYPE_ONLY_ON_ROOT,
2010
		.seq_show = cfq_print_leaf_weight,
2011
		.write_u64 = cfq_set_leaf_weight,
2012
	},
T
Tejun Heo 已提交
2013

2014
	/* no such mapping necessary for !roots */
2015 2016
	{
		.name = "weight_device",
2017
		.flags = CFTYPE_NOT_ON_ROOT,
2018
		.seq_show = cfqg_print_weight_device,
2019
		.write = cfqg_set_weight_device,
2020 2021 2022
	},
	{
		.name = "weight",
2023
		.flags = CFTYPE_NOT_ON_ROOT,
2024
		.seq_show = cfq_print_weight,
2025
		.write_u64 = cfq_set_weight,
2026
	},
T
Tejun Heo 已提交
2027 2028 2029

	{
		.name = "leaf_weight_device",
2030
		.seq_show = cfqg_print_leaf_weight_device,
2031
		.write = cfqg_set_leaf_weight_device,
T
Tejun Heo 已提交
2032 2033 2034
	},
	{
		.name = "leaf_weight",
2035
		.seq_show = cfq_print_leaf_weight,
T
Tejun Heo 已提交
2036 2037 2038
		.write_u64 = cfq_set_leaf_weight,
	},

2039
	/* statistics, covers only the tasks in the cfqg */
2040 2041
	{
		.name = "time",
2042
		.private = offsetof(struct cfq_group, stats.time),
2043
		.seq_show = cfqg_print_stat,
2044 2045 2046
	},
	{
		.name = "sectors",
T
Tejun Heo 已提交
2047
		.seq_show = cfqg_print_stat_sectors,
2048 2049 2050
	},
	{
		.name = "io_service_bytes",
2051 2052
		.private = (unsigned long)&blkcg_policy_cfq,
		.seq_show = blkg_print_stat_bytes,
2053 2054 2055
	},
	{
		.name = "io_serviced",
2056 2057
		.private = (unsigned long)&blkcg_policy_cfq,
		.seq_show = blkg_print_stat_ios,
2058 2059 2060
	},
	{
		.name = "io_service_time",
2061
		.private = offsetof(struct cfq_group, stats.service_time),
2062
		.seq_show = cfqg_print_rwstat,
2063 2064 2065
	},
	{
		.name = "io_wait_time",
2066
		.private = offsetof(struct cfq_group, stats.wait_time),
2067
		.seq_show = cfqg_print_rwstat,
2068 2069 2070
	},
	{
		.name = "io_merged",
2071
		.private = offsetof(struct cfq_group, stats.merged),
2072
		.seq_show = cfqg_print_rwstat,
2073 2074 2075
	},
	{
		.name = "io_queued",
2076
		.private = offsetof(struct cfq_group, stats.queued),
2077
		.seq_show = cfqg_print_rwstat,
2078
	},
2079 2080 2081 2082 2083

	/* the same statictics which cover the cfqg and its descendants */
	{
		.name = "time_recursive",
		.private = offsetof(struct cfq_group, stats.time),
2084
		.seq_show = cfqg_print_stat_recursive,
2085 2086 2087
	},
	{
		.name = "sectors_recursive",
T
Tejun Heo 已提交
2088
		.seq_show = cfqg_print_stat_sectors_recursive,
2089 2090 2091
	},
	{
		.name = "io_service_bytes_recursive",
2092 2093
		.private = (unsigned long)&blkcg_policy_cfq,
		.seq_show = blkg_print_stat_bytes_recursive,
2094 2095 2096
	},
	{
		.name = "io_serviced_recursive",
2097 2098
		.private = (unsigned long)&blkcg_policy_cfq,
		.seq_show = blkg_print_stat_ios_recursive,
2099 2100 2101 2102
	},
	{
		.name = "io_service_time_recursive",
		.private = offsetof(struct cfq_group, stats.service_time),
2103
		.seq_show = cfqg_print_rwstat_recursive,
2104 2105 2106 2107
	},
	{
		.name = "io_wait_time_recursive",
		.private = offsetof(struct cfq_group, stats.wait_time),
2108
		.seq_show = cfqg_print_rwstat_recursive,
2109 2110 2111 2112
	},
	{
		.name = "io_merged_recursive",
		.private = offsetof(struct cfq_group, stats.merged),
2113
		.seq_show = cfqg_print_rwstat_recursive,
2114 2115 2116 2117
	},
	{
		.name = "io_queued_recursive",
		.private = offsetof(struct cfq_group, stats.queued),
2118
		.seq_show = cfqg_print_rwstat_recursive,
2119
	},
2120 2121 2122
#ifdef CONFIG_DEBUG_BLK_CGROUP
	{
		.name = "avg_queue_size",
2123
		.seq_show = cfqg_print_avg_queue_size,
2124 2125 2126
	},
	{
		.name = "group_wait_time",
2127
		.private = offsetof(struct cfq_group, stats.group_wait_time),
2128
		.seq_show = cfqg_print_stat,
2129 2130 2131
	},
	{
		.name = "idle_time",
2132
		.private = offsetof(struct cfq_group, stats.idle_time),
2133
		.seq_show = cfqg_print_stat,
2134 2135 2136
	},
	{
		.name = "empty_time",
2137
		.private = offsetof(struct cfq_group, stats.empty_time),
2138
		.seq_show = cfqg_print_stat,
2139 2140 2141
	},
	{
		.name = "dequeue",
2142
		.private = offsetof(struct cfq_group, stats.dequeue),
2143
		.seq_show = cfqg_print_stat,
2144 2145 2146
	},
	{
		.name = "unaccounted_time",
2147
		.private = offsetof(struct cfq_group, stats.unaccounted_time),
2148
		.seq_show = cfqg_print_stat,
2149 2150 2151 2152
	},
#endif	/* CONFIG_DEBUG_BLK_CGROUP */
	{ }	/* terminate */
};
2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176

static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v)
{
	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
	struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);

	seq_printf(sf, "default %u\n", cgd->weight);
	blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device,
			  &blkcg_policy_cfq, 0, false);
	return 0;
}

static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of,
				     char *buf, size_t nbytes, loff_t off)
{
	char *endp;
	int ret;
	u64 v;

	buf = strim(buf);

	/* "WEIGHT" or "default WEIGHT" sets the default weight */
	v = simple_strtoull(buf, &endp, 0);
	if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) {
2177
		ret = __cfq_set_weight(of_css(of), v, true, false, false);
2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194
		return ret ?: nbytes;
	}

	/* "MAJ:MIN WEIGHT" */
	return __cfqg_set_weight_device(of, buf, nbytes, off, true, false);
}

static struct cftype cfq_blkcg_files[] = {
	{
		.name = "weight",
		.flags = CFTYPE_NOT_ON_ROOT,
		.seq_show = cfq_print_weight_on_dfl,
		.write = cfq_set_weight_on_dfl,
	},
	{ }	/* terminate */
};

2195
#else /* GROUP_IOSCHED */
2196 2197
static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
					 struct blkcg *blkcg)
2198
{
2199
	return cfqd->root_group;
2200
}
2201

2202 2203 2204 2205 2206 2207 2208
static inline void
cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
	cfqq->cfqg = cfqg;
}

#endif /* GROUP_IOSCHED */

2209
/*
2210
 * The cfqd->service_trees holds all pending cfq_queue's that have
2211 2212 2213
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
2214
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2215
				 bool add_front)
2216
{
2217 2218
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
2219
	u64 rb_key;
2220
	struct cfq_rb_root *st;
2221
	int left;
2222
	int new_cfqq = 1;
2223
	u64 now = ktime_get_ns();
2224

2225
	st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq));
2226 2227
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
2228
		parent = rb_last(&st->rb);
2229 2230 2231 2232
		if (parent && parent != &cfqq->rb_node) {
			__cfqq = rb_entry(parent, struct cfq_queue, rb_node);
			rb_key += __cfqq->rb_key;
		} else
2233
			rb_key += now;
2234
	} else if (!add_front) {
2235 2236 2237 2238 2239 2240
		/*
		 * 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.
		 */
2241
		rb_key = cfq_slice_offset(cfqd, cfqq) + now;
2242
		rb_key -= cfqq->slice_resid;
2243
		cfqq->slice_resid = 0;
2244
	} else {
2245
		rb_key = -NSEC_PER_SEC;
2246
		__cfqq = cfq_rb_first(st);
2247
		rb_key += __cfqq ? __cfqq->rb_key : now;
2248
	}
L
Linus Torvalds 已提交
2249

2250
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
2251
		new_cfqq = 0;
2252
		/*
2253
		 * same position, nothing more to do
2254
		 */
2255
		if (rb_key == cfqq->rb_key && cfqq->service_tree == st)
2256
			return;
L
Linus Torvalds 已提交
2257

2258 2259
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
L
Linus Torvalds 已提交
2260
	}
2261

2262
	left = 1;
2263
	parent = NULL;
2264 2265
	cfqq->service_tree = st;
	p = &st->rb.rb_node;
2266 2267 2268 2269
	while (*p) {
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

2270
		/*
2271
		 * sort by key, that represents service time.
2272
		 */
2273
		if (rb_key < __cfqq->rb_key)
2274
			p = &parent->rb_left;
2275
		else {
2276
			p = &parent->rb_right;
2277
			left = 0;
2278
		}
2279 2280
	}

2281
	if (left)
2282
		st->left = &cfqq->rb_node;
2283

2284 2285
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
2286 2287
	rb_insert_color(&cfqq->rb_node, &st->rb);
	st->count++;
2288
	if (add_front || !new_cfqq)
2289
		return;
2290
	cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
L
Linus Torvalds 已提交
2291 2292
}

2293
static struct cfq_queue *
2294 2295 2296
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)
2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312
{
	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.
		 */
2313
		if (sector > blk_rq_pos(cfqq->next_rq))
2314
			n = &(*p)->rb_right;
2315
		else if (sector < blk_rq_pos(cfqq->next_rq))
2316 2317 2318 2319
			n = &(*p)->rb_left;
		else
			break;
		p = n;
2320
		cfqq = NULL;
2321 2322 2323 2324 2325
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
2326
	return cfqq;
2327 2328 2329 2330 2331 2332 2333
}

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

2334 2335 2336 2337
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
2338 2339 2340 2341 2342 2343

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

2344
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
2345 2346
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
2347 2348
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
2349 2350 2351
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
2352 2353
}

2354 2355 2356
/*
 * Update cfqq's position in the service tree.
 */
2357
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
J
Jens Axboe 已提交
2358 2359 2360 2361
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
2362
	if (cfq_cfqq_on_rr(cfqq)) {
2363
		cfq_service_tree_add(cfqd, cfqq, 0);
2364 2365
		cfq_prio_tree_add(cfqd, cfqq);
	}
J
Jens Axboe 已提交
2366 2367
}

L
Linus Torvalds 已提交
2368 2369
/*
 * add to busy list of queues for service, trying to be fair in ordering
2370
 * the pending list according to last request service
L
Linus Torvalds 已提交
2371
 */
J
Jens Axboe 已提交
2372
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
2373
{
2374
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
J
Jens Axboe 已提交
2375 2376
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
L
Linus Torvalds 已提交
2377
	cfqd->busy_queues++;
2378 2379
	if (cfq_cfqq_sync(cfqq))
		cfqd->busy_sync_queues++;
L
Linus Torvalds 已提交
2380

2381
	cfq_resort_rr_list(cfqd, cfqq);
L
Linus Torvalds 已提交
2382 2383
}

2384 2385 2386 2387
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
J
Jens Axboe 已提交
2388
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
2389
{
2390
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
J
Jens Axboe 已提交
2391 2392
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
L
Linus Torvalds 已提交
2393

2394 2395 2396 2397
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
	}
2398 2399 2400 2401
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
2402

2403
	cfq_group_notify_queue_del(cfqd, cfqq->cfqg);
L
Linus Torvalds 已提交
2404 2405
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
2406 2407
	if (cfq_cfqq_sync(cfqq))
		cfqd->busy_sync_queues--;
L
Linus Torvalds 已提交
2408 2409 2410 2411 2412
}

/*
 * rb tree support functions
 */
J
Jens Axboe 已提交
2413
static void cfq_del_rq_rb(struct request *rq)
L
Linus Torvalds 已提交
2414
{
J
Jens Axboe 已提交
2415 2416
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
	const int sync = rq_is_sync(rq);
L
Linus Torvalds 已提交
2417

2418 2419
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
L
Linus Torvalds 已提交
2420

J
Jens Axboe 已提交
2421
	elv_rb_del(&cfqq->sort_list, rq);
L
Linus Torvalds 已提交
2422

2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433
	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;
		}
	}
L
Linus Torvalds 已提交
2434 2435
}

J
Jens Axboe 已提交
2436
static void cfq_add_rq_rb(struct request *rq)
L
Linus Torvalds 已提交
2437
{
J
Jens Axboe 已提交
2438
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
L
Linus Torvalds 已提交
2439
	struct cfq_data *cfqd = cfqq->cfqd;
2440
	struct request *prev;
L
Linus Torvalds 已提交
2441

2442
	cfqq->queued[rq_is_sync(rq)]++;
L
Linus Torvalds 已提交
2443

2444
	elv_rb_add(&cfqq->sort_list, rq);
2445 2446 2447

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
2448 2449 2450 2451

	/*
	 * check if this request is a better next-serve candidate
	 */
2452
	prev = cfqq->next_rq;
2453
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
2454 2455 2456 2457 2458 2459 2460

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

2461
	BUG_ON(!cfqq->next_rq);
L
Linus Torvalds 已提交
2462 2463
}

J
Jens Axboe 已提交
2464
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
L
Linus Torvalds 已提交
2465
{
2466 2467
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
2468
	cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags);
J
Jens Axboe 已提交
2469
	cfq_add_rq_rb(rq);
2470
	cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group,
2471
				 rq->cmd_flags);
L
Linus Torvalds 已提交
2472 2473
}

2474 2475
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
L
Linus Torvalds 已提交
2476
{
2477
	struct task_struct *tsk = current;
2478
	struct cfq_io_cq *cic;
2479
	struct cfq_queue *cfqq;
L
Linus Torvalds 已提交
2480

2481
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
2482 2483 2484
	if (!cic)
		return NULL;

2485
	cfqq = cic_to_cfqq(cic, op_is_sync(bio->bi_opf));
K
Kent Overstreet 已提交
2486 2487
	if (cfqq)
		return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio));
L
Linus Torvalds 已提交
2488 2489 2490 2491

	return NULL;
}

2492
static void cfq_activate_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
2493
{
2494
	struct cfq_data *cfqd = q->elevator->elevator_data;
J
Jens Axboe 已提交
2495

2496
	cfqd->rq_in_driver++;
2497
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
2498
						cfqd->rq_in_driver);
2499

2500
	cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
L
Linus Torvalds 已提交
2501 2502
}

2503
static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
2504
{
2505 2506
	struct cfq_data *cfqd = q->elevator->elevator_data;

2507 2508
	WARN_ON(!cfqd->rq_in_driver);
	cfqd->rq_in_driver--;
2509
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
2510
						cfqd->rq_in_driver);
L
Linus Torvalds 已提交
2511 2512
}

2513
static void cfq_remove_request(struct request *rq)
L
Linus Torvalds 已提交
2514
{
J
Jens Axboe 已提交
2515
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
2516

J
Jens Axboe 已提交
2517 2518
	if (cfqq->next_rq == rq)
		cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
L
Linus Torvalds 已提交
2519

2520
	list_del_init(&rq->queuelist);
J
Jens Axboe 已提交
2521
	cfq_del_rq_rb(rq);
2522

2523
	cfqq->cfqd->rq_queued--;
2524
	cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags);
2525 2526 2527
	if (rq->cmd_flags & REQ_PRIO) {
		WARN_ON(!cfqq->prio_pending);
		cfqq->prio_pending--;
2528
	}
L
Linus Torvalds 已提交
2529 2530
}

2531 2532
static int cfq_merge(struct request_queue *q, struct request **req,
		     struct bio *bio)
L
Linus Torvalds 已提交
2533 2534 2535 2536
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request *__rq;

2537
	__rq = cfq_find_rq_fmerge(cfqd, bio);
2538
	if (__rq && elv_bio_merge_ok(__rq, bio)) {
2539 2540
		*req = __rq;
		return ELEVATOR_FRONT_MERGE;
L
Linus Torvalds 已提交
2541 2542 2543 2544 2545
	}

	return ELEVATOR_NO_MERGE;
}

2546
static void cfq_merged_request(struct request_queue *q, struct request *req,
2547
			       int type)
L
Linus Torvalds 已提交
2548
{
2549
	if (type == ELEVATOR_FRONT_MERGE) {
J
Jens Axboe 已提交
2550
		struct cfq_queue *cfqq = RQ_CFQQ(req);
L
Linus Torvalds 已提交
2551

J
Jens Axboe 已提交
2552
		cfq_reposition_rq_rb(cfqq, req);
L
Linus Torvalds 已提交
2553 2554 2555
	}
}

D
Divyesh Shah 已提交
2556 2557 2558
static void cfq_bio_merged(struct request_queue *q, struct request *req,
				struct bio *bio)
{
2559
	cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_opf);
D
Divyesh Shah 已提交
2560 2561
}

L
Linus Torvalds 已提交
2562
static void
2563
cfq_merged_requests(struct request_queue *q, struct request *rq,
L
Linus Torvalds 已提交
2564 2565
		    struct request *next)
{
2566
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
2567 2568
	struct cfq_data *cfqd = q->elevator->elevator_data;

2569 2570 2571 2572
	/*
	 * reposition in fifo if next is older than rq
	 */
	if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
2573
	    next->fifo_time < rq->fifo_time &&
2574
	    cfqq == RQ_CFQQ(next)) {
2575
		list_move(&rq->queuelist, &next->queuelist);
2576
		rq->fifo_time = next->fifo_time;
2577
	}
2578

2579 2580
	if (cfqq->next_rq == next)
		cfqq->next_rq = rq;
2581
	cfq_remove_request(next);
2582
	cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags);
2583 2584 2585 2586 2587 2588 2589 2590 2591 2592

	cfqq = RQ_CFQQ(next);
	/*
	 * all requests of this queue are merged to other queues, delete it
	 * from the service tree. If it's the active_queue,
	 * cfq_dispatch_requests() will choose to expire it or do idle
	 */
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) &&
	    cfqq != cfqd->active_queue)
		cfq_del_cfqq_rr(cfqd, cfqq);
2593 2594
}

2595 2596
static int cfq_allow_bio_merge(struct request_queue *q, struct request *rq,
			       struct bio *bio)
2597 2598
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
2599
	bool is_sync = op_is_sync(bio->bi_opf);
2600
	struct cfq_io_cq *cic;
2601 2602 2603
	struct cfq_queue *cfqq;

	/*
2604
	 * Disallow merge of a sync bio into an async request.
2605
	 */
2606
	if (is_sync && !rq_is_sync(rq))
2607
		return false;
2608 2609

	/*
T
Tejun Heo 已提交
2610
	 * Lookup the cfqq that this bio will be queued with and allow
2611
	 * merge only if rq is queued there.
T
Tejun Heo 已提交
2612
	 */
2613 2614 2615
	cic = cfq_cic_lookup(cfqd, current->io_context);
	if (!cic)
		return false;
2616

2617
	cfqq = cic_to_cfqq(cic, is_sync);
2618
	return cfqq == RQ_CFQQ(rq);
2619 2620
}

2621 2622 2623 2624 2625 2626
static int cfq_allow_rq_merge(struct request_queue *q, struct request *rq,
			      struct request *next)
{
	return RQ_CFQQ(rq) == RQ_CFQQ(next);
}

2627 2628
static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
2629
	hrtimer_try_to_cancel(&cfqd->idle_slice_timer);
2630
	cfqg_stats_update_idle_time(cfqq->cfqg);
2631 2632
}

J
Jens Axboe 已提交
2633 2634
static void __cfq_set_active_queue(struct cfq_data *cfqd,
				   struct cfq_queue *cfqq)
2635 2636
{
	if (cfqq) {
2637
		cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d",
2638
				cfqd->serving_wl_class, cfqd->serving_wl_type);
2639
		cfqg_stats_update_avg_queue_size(cfqq->cfqg);
2640
		cfqq->slice_start = 0;
2641
		cfqq->dispatch_start = ktime_get_ns();
2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653
		cfqq->allocated_slice = 0;
		cfqq->slice_end = 0;
		cfqq->slice_dispatch = 0;
		cfqq->nr_sectors = 0;

		cfq_clear_cfqq_wait_request(cfqq);
		cfq_clear_cfqq_must_dispatch(cfqq);
		cfq_clear_cfqq_must_alloc_slice(cfqq);
		cfq_clear_cfqq_fifo_expire(cfqq);
		cfq_mark_cfqq_slice_new(cfqq);

		cfq_del_timer(cfqd, cfqq);
2654 2655 2656 2657 2658
	}

	cfqd->active_queue = cfqq;
}

2659 2660 2661 2662 2663
/*
 * current cfqq expired its slice (or was too idle), select new one
 */
static void
__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2664
		    bool timed_out)
2665
{
2666 2667
	cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);

2668
	if (cfq_cfqq_wait_request(cfqq))
2669
		cfq_del_timer(cfqd, cfqq);
2670 2671

	cfq_clear_cfqq_wait_request(cfqq);
2672
	cfq_clear_cfqq_wait_busy(cfqq);
2673

2674 2675 2676 2677 2678 2679 2680 2681 2682
	/*
	 * If this cfqq is shared between multiple processes, check to
	 * make sure that those processes are still issuing I/Os within
	 * the mean seek distance.  If not, it may be time to break the
	 * queues apart again.
	 */
	if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq))
		cfq_mark_cfqq_split_coop(cfqq);

2683
	/*
2684
	 * store what was left of this slice, if the queue idled/timed out
2685
	 */
2686 2687
	if (timed_out) {
		if (cfq_cfqq_slice_new(cfqq))
2688
			cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq);
2689
		else
2690
			cfqq->slice_resid = cfqq->slice_end - ktime_get_ns();
2691
		cfq_log_cfqq(cfqd, cfqq, "resid=%lld", cfqq->slice_resid);
2692
	}
2693

2694
	cfq_group_served(cfqd, cfqq->cfqg, cfqq);
2695

2696 2697 2698
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
		cfq_del_cfqq_rr(cfqd, cfqq);

2699
	cfq_resort_rr_list(cfqd, cfqq);
2700 2701 2702 2703 2704

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

	if (cfqd->active_cic) {
2705
		put_io_context(cfqd->active_cic->icq.ioc);
2706 2707 2708 2709
		cfqd->active_cic = NULL;
	}
}

2710
static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
2711 2712 2713 2714
{
	struct cfq_queue *cfqq = cfqd->active_queue;

	if (cfqq)
2715
		__cfq_slice_expired(cfqd, cfqq, timed_out);
2716 2717
}

2718 2719 2720 2721
/*
 * Get next queue for service. Unless we have a queue preemption,
 * we'll simply select the first cfqq in the service tree.
 */
J
Jens Axboe 已提交
2722
static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
2723
{
2724 2725
	struct cfq_rb_root *st = st_for(cfqd->serving_group,
			cfqd->serving_wl_class, cfqd->serving_wl_type);
2726

2727 2728 2729
	if (!cfqd->rq_queued)
		return NULL;

2730
	/* There is nothing to dispatch */
2731
	if (!st)
2732
		return NULL;
2733
	if (RB_EMPTY_ROOT(&st->rb))
2734
		return NULL;
2735
	return cfq_rb_first(st);
J
Jens Axboe 已提交
2736 2737
}

2738 2739
static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd)
{
2740
	struct cfq_group *cfqg;
2741 2742 2743 2744 2745 2746 2747
	struct cfq_queue *cfqq;
	int i, j;
	struct cfq_rb_root *st;

	if (!cfqd->rq_queued)
		return NULL;

2748 2749 2750 2751
	cfqg = cfq_get_next_cfqg(cfqd);
	if (!cfqg)
		return NULL;

2752 2753 2754
	for_each_cfqg_st(cfqg, i, j, st) {
		cfqq = cfq_rb_first(st);
		if (cfqq)
2755
			return cfqq;
2756
	}
2757 2758 2759
	return NULL;
}

2760 2761 2762
/*
 * Get and set a new active queue for service.
 */
2763 2764
static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
					      struct cfq_queue *cfqq)
J
Jens Axboe 已提交
2765
{
2766
	if (!cfqq)
2767
		cfqq = cfq_get_next_queue(cfqd);
J
Jens Axboe 已提交
2768

2769
	__cfq_set_active_queue(cfqd, cfqq);
J
Jens Axboe 已提交
2770
	return cfqq;
2771 2772
}

2773 2774 2775
static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
					  struct request *rq)
{
2776 2777
	if (blk_rq_pos(rq) >= cfqd->last_position)
		return blk_rq_pos(rq) - cfqd->last_position;
2778
	else
2779
		return cfqd->last_position - blk_rq_pos(rq);
2780 2781
}

2782
static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2783
			       struct request *rq)
J
Jens Axboe 已提交
2784
{
2785
	return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR;
J
Jens Axboe 已提交
2786 2787
}

2788 2789 2790
static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
				    struct cfq_queue *cur_cfqq)
{
2791
	struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802
	struct rb_node *parent, *node;
	struct cfq_queue *__cfqq;
	sector_t sector = cfqd->last_position;

	if (RB_EMPTY_ROOT(root))
		return NULL;

	/*
	 * First, if we find a request starting at the end of the last
	 * request, choose it.
	 */
2803
	__cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
2804 2805 2806 2807 2808 2809 2810 2811
	if (__cfqq)
		return __cfqq;

	/*
	 * If the exact sector wasn't found, the parent of the NULL leaf
	 * will contain the closest sector.
	 */
	__cfqq = rb_entry(parent, struct cfq_queue, p_node);
2812
	if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
2813 2814
		return __cfqq;

2815
	if (blk_rq_pos(__cfqq->next_rq) < sector)
2816 2817 2818 2819 2820 2821 2822
		node = rb_next(&__cfqq->p_node);
	else
		node = rb_prev(&__cfqq->p_node);
	if (!node)
		return NULL;

	__cfqq = rb_entry(node, struct cfq_queue, p_node);
2823
	if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839
		return __cfqq;

	return NULL;
}

/*
 * cfqd - obvious
 * cur_cfqq - passed in so that we don't decide that the current queue is
 * 	      closely cooperating with itself.
 *
 * So, basically we're assuming that that cur_cfqq has dispatched at least
 * one request, and that cfqd->last_position reflects a position on the disk
 * associated with the I/O issued by cur_cfqq.  I'm not sure this is a valid
 * assumption.
 */
static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
2840
					      struct cfq_queue *cur_cfqq)
J
Jens Axboe 已提交
2841
{
2842 2843
	struct cfq_queue *cfqq;

2844 2845
	if (cfq_class_idle(cur_cfqq))
		return NULL;
2846 2847 2848 2849 2850
	if (!cfq_cfqq_sync(cur_cfqq))
		return NULL;
	if (CFQQ_SEEKY(cur_cfqq))
		return NULL;

2851 2852 2853 2854 2855 2856
	/*
	 * Don't search priority tree if it's the only queue in the group.
	 */
	if (cur_cfqq->cfqg->nr_cfqq == 1)
		return NULL;

J
Jens Axboe 已提交
2857
	/*
2858 2859 2860
	 * We should notice if some of the queues are cooperating, eg
	 * working closely on the same area of the disk. In that case,
	 * we can group them together and don't waste time idling.
J
Jens Axboe 已提交
2861
	 */
2862 2863 2864 2865
	cfqq = cfqq_close(cfqd, cur_cfqq);
	if (!cfqq)
		return NULL;

2866 2867 2868 2869
	/* If new queue belongs to different cfq_group, don't choose it */
	if (cur_cfqq->cfqg != cfqq->cfqg)
		return NULL;

J
Jeff Moyer 已提交
2870 2871 2872 2873 2874
	/*
	 * It only makes sense to merge sync queues.
	 */
	if (!cfq_cfqq_sync(cfqq))
		return NULL;
2875 2876
	if (CFQQ_SEEKY(cfqq))
		return NULL;
J
Jeff Moyer 已提交
2877

2878 2879 2880 2881 2882 2883
	/*
	 * Do not merge queues of different priority classes
	 */
	if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
		return NULL;

2884
	return cfqq;
J
Jens Axboe 已提交
2885 2886
}

2887 2888 2889 2890 2891 2892
/*
 * Determine whether we should enforce idle window for this queue.
 */

static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
2893
	enum wl_class_t wl_class = cfqq_class(cfqq);
2894
	struct cfq_rb_root *st = cfqq->service_tree;
2895

2896 2897
	BUG_ON(!st);
	BUG_ON(!st->count);
2898

2899 2900 2901
	if (!cfqd->cfq_slice_idle)
		return false;

2902
	/* We never do for idle class queues. */
2903
	if (wl_class == IDLE_WORKLOAD)
2904 2905 2906
		return false;

	/* We do for queues that were marked with idle window flag. */
2907 2908
	if (cfq_cfqq_idle_window(cfqq) &&
	   !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag))
2909 2910 2911 2912 2913 2914
		return true;

	/*
	 * Otherwise, we do only if they are the last ones
	 * in their service tree.
	 */
2915 2916
	if (st->count == 1 && cfq_cfqq_sync(cfqq) &&
	   !cfq_io_thinktime_big(cfqd, &st->ttime, false))
S
Shaohua Li 已提交
2917
		return true;
2918
	cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count);
S
Shaohua Li 已提交
2919
	return false;
2920 2921
}

J
Jens Axboe 已提交
2922
static void cfq_arm_slice_timer(struct cfq_data *cfqd)
2923
{
2924
	struct cfq_queue *cfqq = cfqd->active_queue;
2925
	struct cfq_rb_root *st = cfqq->service_tree;
2926
	struct cfq_io_cq *cic;
2927 2928
	u64 sl, group_idle = 0;
	u64 now = ktime_get_ns();
2929

2930
	/*
J
Jens Axboe 已提交
2931 2932 2933
	 * SSD device without seek penalty, disable idling. But only do so
	 * for devices that support queuing, otherwise we still have a problem
	 * with sync vs async workloads.
2934
	 */
J
Jens Axboe 已提交
2935
	if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
2936 2937
		return;

2938
	WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
J
Jens Axboe 已提交
2939
	WARN_ON(cfq_cfqq_slice_new(cfqq));
2940 2941 2942 2943

	/*
	 * idle is disabled, either manually or by past process history
	 */
2944 2945 2946 2947 2948 2949 2950
	if (!cfq_should_idle(cfqd, cfqq)) {
		/* no queue idling. Check for group idling */
		if (cfqd->cfq_group_idle)
			group_idle = cfqd->cfq_group_idle;
		else
			return;
	}
J
Jens Axboe 已提交
2951

2952
	/*
2953
	 * still active requests from this queue, don't idle
2954
	 */
2955
	if (cfqq->dispatched)
2956 2957
		return;

2958 2959 2960
	/*
	 * task has exited, don't wait
	 */
2961
	cic = cfqd->active_cic;
T
Tejun Heo 已提交
2962
	if (!cic || !atomic_read(&cic->icq.ioc->active_ref))
J
Jens Axboe 已提交
2963 2964
		return;

2965 2966 2967 2968 2969
	/*
	 * If our average think time is larger than the remaining time
	 * slice, then don't idle. This avoids overrunning the allotted
	 * time slice.
	 */
2970
	if (sample_valid(cic->ttime.ttime_samples) &&
2971 2972
	    (cfqq->slice_end - now < cic->ttime.ttime_mean)) {
		cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%llu",
2973
			     cic->ttime.ttime_mean);
2974
		return;
2975
	}
2976

2977 2978 2979 2980 2981 2982 2983
	/*
	 * There are other queues in the group or this is the only group and
	 * it has too big thinktime, don't do group idle.
	 */
	if (group_idle &&
	    (cfqq->cfqg->nr_cfqq > 1 ||
	     cfq_io_thinktime_big(cfqd, &st->ttime, true)))
2984 2985
		return;

J
Jens Axboe 已提交
2986
	cfq_mark_cfqq_wait_request(cfqq);
2987

2988 2989 2990 2991
	if (group_idle)
		sl = cfqd->cfq_group_idle;
	else
		sl = cfqd->cfq_slice_idle;
2992

2993 2994
	hrtimer_start(&cfqd->idle_slice_timer, ns_to_ktime(sl),
		      HRTIMER_MODE_REL);
2995
	cfqg_stats_set_start_idle_time(cfqq->cfqg);
2996
	cfq_log_cfqq(cfqd, cfqq, "arm_idle: %llu group_idle: %d", sl,
2997
			group_idle ? 1 : 0);
L
Linus Torvalds 已提交
2998 2999
}

3000 3001 3002
/*
 * Move request from internal lists to the request queue dispatch list.
 */
3003
static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
3004
{
3005
	struct cfq_data *cfqd = q->elevator->elevator_data;
J
Jens Axboe 已提交
3006
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
3007

3008 3009
	cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");

3010
	cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
3011
	cfq_remove_request(rq);
J
Jens Axboe 已提交
3012
	cfqq->dispatched++;
3013
	(RQ_CFQG(rq))->dispatched++;
3014
	elv_dispatch_sort(q, rq);
3015

3016
	cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++;
3017
	cfqq->nr_sectors += blk_rq_sectors(rq);
L
Linus Torvalds 已提交
3018 3019 3020 3021 3022
}

/*
 * return expired entry, or NULL to just start from scratch in rbtree
 */
J
Jens Axboe 已提交
3023
static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
3024
{
3025
	struct request *rq = NULL;
L
Linus Torvalds 已提交
3026

J
Jens Axboe 已提交
3027
	if (cfq_cfqq_fifo_expire(cfqq))
L
Linus Torvalds 已提交
3028
		return NULL;
3029 3030 3031

	cfq_mark_cfqq_fifo_expire(cfqq);

3032 3033
	if (list_empty(&cfqq->fifo))
		return NULL;
L
Linus Torvalds 已提交
3034

3035
	rq = rq_entry_fifo(cfqq->fifo.next);
3036
	if (ktime_get_ns() < rq->fifo_time)
3037
		rq = NULL;
L
Linus Torvalds 已提交
3038

J
Jens Axboe 已提交
3039
	return rq;
L
Linus Torvalds 已提交
3040 3041
}

3042 3043 3044 3045
static inline int
cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	const int base_rq = cfqd->cfq_slice_async_rq;
L
Linus Torvalds 已提交
3046

3047
	WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
L
Linus Torvalds 已提交
3048

3049
	return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
L
Linus Torvalds 已提交
3050 3051
}

J
Jeff Moyer 已提交
3052 3053 3054 3055 3056 3057 3058 3059
/*
 * Must be called with the queue_lock held.
 */
static int cfqq_process_refs(struct cfq_queue *cfqq)
{
	int process_refs, io_refs;

	io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE];
3060
	process_refs = cfqq->ref - io_refs;
J
Jeff Moyer 已提交
3061 3062 3063 3064 3065 3066
	BUG_ON(process_refs < 0);
	return process_refs;
}

static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
{
3067
	int process_refs, new_process_refs;
J
Jeff Moyer 已提交
3068 3069
	struct cfq_queue *__cfqq;

3070 3071 3072 3073 3074 3075 3076 3077 3078
	/*
	 * If there are no process references on the new_cfqq, then it is
	 * unsafe to follow the ->new_cfqq chain as other cfqq's in the
	 * chain may have dropped their last reference (not just their
	 * last process reference).
	 */
	if (!cfqq_process_refs(new_cfqq))
		return;

J
Jeff Moyer 已提交
3079 3080 3081 3082 3083 3084 3085 3086
	/* Avoid a circular list and skip interim queue merges */
	while ((__cfqq = new_cfqq->new_cfqq)) {
		if (__cfqq == cfqq)
			return;
		new_cfqq = __cfqq;
	}

	process_refs = cfqq_process_refs(cfqq);
3087
	new_process_refs = cfqq_process_refs(new_cfqq);
J
Jeff Moyer 已提交
3088 3089 3090 3091
	/*
	 * If the process for the cfqq has gone away, there is no
	 * sense in merging the queues.
	 */
3092
	if (process_refs == 0 || new_process_refs == 0)
J
Jeff Moyer 已提交
3093 3094
		return;

3095 3096 3097 3098 3099
	/*
	 * Merge in the direction of the lesser amount of work.
	 */
	if (new_process_refs >= process_refs) {
		cfqq->new_cfqq = new_cfqq;
3100
		new_cfqq->ref += process_refs;
3101 3102
	} else {
		new_cfqq->new_cfqq = cfqq;
3103
		cfqq->ref += new_process_refs;
3104
	}
J
Jeff Moyer 已提交
3105 3106
}

3107
static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd,
3108
			struct cfq_group *cfqg, enum wl_class_t wl_class)
3109 3110 3111 3112
{
	struct cfq_queue *queue;
	int i;
	bool key_valid = false;
3113
	u64 lowest_key = 0;
3114 3115
	enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;

3116 3117
	for (i = 0; i <= SYNC_WORKLOAD; ++i) {
		/* select the one with lowest rb_key */
3118
		queue = cfq_rb_first(st_for(cfqg, wl_class, i));
3119
		if (queue &&
3120
		    (!key_valid || queue->rb_key < lowest_key)) {
3121 3122 3123 3124 3125 3126 3127 3128 3129
			lowest_key = queue->rb_key;
			cur_best = i;
			key_valid = true;
		}
	}

	return cur_best;
}

3130 3131
static void
choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg)
3132
{
3133
	u64 slice;
3134
	unsigned count;
3135
	struct cfq_rb_root *st;
3136
	u64 group_slice;
3137
	enum wl_class_t original_class = cfqd->serving_wl_class;
3138
	u64 now = ktime_get_ns();
3139

3140
	/* Choose next priority. RT > BE > IDLE */
3141
	if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg))
3142
		cfqd->serving_wl_class = RT_WORKLOAD;
3143
	else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg))
3144
		cfqd->serving_wl_class = BE_WORKLOAD;
3145
	else {
3146
		cfqd->serving_wl_class = IDLE_WORKLOAD;
3147
		cfqd->workload_expires = now + jiffies_to_nsecs(1);
3148 3149 3150
		return;
	}

3151
	if (original_class != cfqd->serving_wl_class)
3152 3153
		goto new_workload;

3154 3155 3156 3157 3158
	/*
	 * For RT and BE, we have to choose also the type
	 * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
	 * expiration time
	 */
3159
	st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
3160
	count = st->count;
3161 3162

	/*
3163
	 * check workload expiration, and that we still have other queues ready
3164
	 */
3165
	if (count && !(now > cfqd->workload_expires))
3166 3167
		return;

3168
new_workload:
3169
	/* otherwise select new workload type */
3170
	cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg,
3171
					cfqd->serving_wl_class);
3172
	st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
3173
	count = st->count;
3174 3175 3176 3177 3178 3179

	/*
	 * the workload slice is computed as a fraction of target latency
	 * proportional to the number of queues in that workload, over
	 * all the queues in the same priority class
	 */
3180 3181
	group_slice = cfq_group_slice(cfqd, cfqg);

3182
	slice = div_u64(group_slice * count,
3183 3184
		max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class],
		      cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd,
3185
					cfqg)));
3186

3187
	if (cfqd->serving_wl_type == ASYNC_WORKLOAD) {
3188
		u64 tmp;
3189 3190 3191 3192 3193 3194 3195 3196

		/*
		 * Async queues are currently system wide. Just taking
		 * proportion of queues with-in same group will lead to higher
		 * async ratio system wide as generally root group is going
		 * to have higher weight. A more accurate thing would be to
		 * calculate system wide asnc/sync ratio.
		 */
3197 3198
		tmp = cfqd->cfq_target_latency *
			cfqg_busy_async_queues(cfqd, cfqg);
3199 3200
		tmp = div_u64(tmp, cfqd->busy_queues);
		slice = min_t(u64, slice, tmp);
3201

3202 3203
		/* async workload slice is scaled down according to
		 * the sync/async slice ratio. */
3204
		slice = div64_u64(slice*cfqd->cfq_slice[0], cfqd->cfq_slice[1]);
3205
	} else
3206 3207 3208
		/* sync workload slice is at least 2 * cfq_slice_idle */
		slice = max(slice, 2 * cfqd->cfq_slice_idle);

3209 3210 3211
	slice = max_t(u64, slice, CFQ_MIN_TT);
	cfq_log(cfqd, "workload slice:%llu", slice);
	cfqd->workload_expires = now + slice;
3212 3213
}

3214 3215 3216
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
3217
	struct cfq_group *cfqg;
3218 3219 3220

	if (RB_EMPTY_ROOT(&st->rb))
		return NULL;
3221 3222 3223
	cfqg = cfq_rb_first_group(st);
	update_min_vdisktime(st);
	return cfqg;
3224 3225
}

3226 3227
static void cfq_choose_cfqg(struct cfq_data *cfqd)
{
3228
	struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd);
3229
	u64 now = ktime_get_ns();
3230 3231

	cfqd->serving_group = cfqg;
3232 3233

	/* Restore the workload type data */
3234
	if (cfqg->saved_wl_slice) {
3235
		cfqd->workload_expires = now + cfqg->saved_wl_slice;
3236 3237
		cfqd->serving_wl_type = cfqg->saved_wl_type;
		cfqd->serving_wl_class = cfqg->saved_wl_class;
3238
	} else
3239
		cfqd->workload_expires = now - 1;
3240

3241
	choose_wl_class_and_type(cfqd, cfqg);
3242 3243
}

3244
/*
3245 3246
 * Select a queue for service. If we have a current active queue,
 * check whether to continue servicing it, or retrieve and set a new one.
3247
 */
3248
static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
L
Linus Torvalds 已提交
3249
{
3250
	struct cfq_queue *cfqq, *new_cfqq = NULL;
3251
	u64 now = ktime_get_ns();
L
Linus Torvalds 已提交
3252

3253 3254 3255
	cfqq = cfqd->active_queue;
	if (!cfqq)
		goto new_queue;
L
Linus Torvalds 已提交
3256

3257 3258
	if (!cfqd->rq_queued)
		return NULL;
3259 3260 3261 3262 3263 3264 3265

	/*
	 * We were waiting for group to get backlogged. Expire the queue
	 */
	if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list))
		goto expire;

3266
	/*
J
Jens Axboe 已提交
3267
	 * The active queue has run out of time, expire it and select new.
3268
	 */
3269 3270 3271 3272 3273 3274 3275 3276 3277 3278
	if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) {
		/*
		 * If slice had not expired at the completion of last request
		 * we might not have turned on wait_busy flag. Don't expire
		 * the queue yet. Allow the group to get backlogged.
		 *
		 * The very fact that we have used the slice, that means we
		 * have been idling all along on this queue and it should be
		 * ok to wait for this request to complete.
		 */
3279 3280 3281
		if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
		    && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
			cfqq = NULL;
3282
			goto keep_queue;
3283
		} else
3284
			goto check_group_idle;
3285
	}
L
Linus Torvalds 已提交
3286

3287
	/*
J
Jens Axboe 已提交
3288 3289
	 * The active queue has requests and isn't expired, allow it to
	 * dispatch.
3290
	 */
3291
	if (!RB_EMPTY_ROOT(&cfqq->sort_list))
3292
		goto keep_queue;
J
Jens Axboe 已提交
3293

3294 3295 3296 3297
	/*
	 * If another queue has a request waiting within our mean seek
	 * distance, let it run.  The expire code will check for close
	 * cooperators and put the close queue at the front of the service
J
Jeff Moyer 已提交
3298
	 * tree.  If possible, merge the expiring queue with the new cfqq.
3299
	 */
3300
	new_cfqq = cfq_close_cooperator(cfqd, cfqq);
J
Jeff Moyer 已提交
3301 3302 3303
	if (new_cfqq) {
		if (!cfqq->new_cfqq)
			cfq_setup_merge(cfqq, new_cfqq);
3304
		goto expire;
J
Jeff Moyer 已提交
3305
	}
3306

J
Jens Axboe 已提交
3307 3308 3309 3310 3311
	/*
	 * No requests pending. If the active queue still has requests in
	 * flight or is idling for a new request, allow either of these
	 * conditions to happen (or time out) before selecting a new queue.
	 */
3312
	if (hrtimer_active(&cfqd->idle_slice_timer)) {
3313 3314 3315 3316
		cfqq = NULL;
		goto keep_queue;
	}

3317 3318 3319 3320 3321 3322
	/*
	 * This is a deep seek queue, but the device is much faster than
	 * the queue can deliver, don't idle
	 **/
	if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) &&
	    (cfq_cfqq_slice_new(cfqq) ||
3323
	    (cfqq->slice_end - now > now - cfqq->slice_start))) {
3324 3325 3326 3327
		cfq_clear_cfqq_deep(cfqq);
		cfq_clear_cfqq_idle_window(cfqq);
	}

3328 3329 3330 3331 3332 3333 3334 3335 3336 3337
	if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
		cfqq = NULL;
		goto keep_queue;
	}

	/*
	 * If group idle is enabled and there are requests dispatched from
	 * this group, wait for requests to complete.
	 */
check_group_idle:
S
Shaohua Li 已提交
3338 3339 3340
	if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 &&
	    cfqq->cfqg->dispatched &&
	    !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) {
3341 3342
		cfqq = NULL;
		goto keep_queue;
3343 3344
	}

J
Jens Axboe 已提交
3345
expire:
3346
	cfq_slice_expired(cfqd, 0);
J
Jens Axboe 已提交
3347
new_queue:
3348 3349 3350 3351 3352
	/*
	 * Current queue expired. Check if we have to switch to a new
	 * service tree
	 */
	if (!new_cfqq)
3353
		cfq_choose_cfqg(cfqd);
3354

3355
	cfqq = cfq_set_active_queue(cfqd, new_cfqq);
3356
keep_queue:
J
Jens Axboe 已提交
3357
	return cfqq;
3358 3359
}

J
Jens Axboe 已提交
3360
static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
3361 3362 3363 3364 3365 3366 3367 3368 3369
{
	int dispatched = 0;

	while (cfqq->next_rq) {
		cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
		dispatched++;
	}

	BUG_ON(!list_empty(&cfqq->fifo));
3370 3371

	/* By default cfqq is not expired if it is empty. Do it explicitly */
3372
	__cfq_slice_expired(cfqq->cfqd, cfqq, 0);
3373 3374 3375
	return dispatched;
}

3376 3377 3378 3379
/*
 * Drain our current requests. Used for barriers and when switching
 * io schedulers on-the-fly.
 */
3380
static int cfq_forced_dispatch(struct cfq_data *cfqd)
3381
{
3382
	struct cfq_queue *cfqq;
3383
	int dispatched = 0;
3384

3385
	/* Expire the timeslice of the current active queue first */
3386
	cfq_slice_expired(cfqd, 0);
3387 3388
	while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) {
		__cfq_set_active_queue(cfqd, cfqq);
3389
		dispatched += __cfq_forced_dispatch_cfqq(cfqq);
3390
	}
3391 3392 3393

	BUG_ON(cfqd->busy_queues);

3394
	cfq_log(cfqd, "forced_dispatch=%d", dispatched);
3395 3396 3397
	return dispatched;
}

S
Shaohua Li 已提交
3398 3399 3400
static inline bool cfq_slice_used_soon(struct cfq_data *cfqd,
	struct cfq_queue *cfqq)
{
3401 3402
	u64 now = ktime_get_ns();

S
Shaohua Li 已提交
3403 3404
	/* the queue hasn't finished any request, can't estimate */
	if (cfq_cfqq_slice_new(cfqq))
S
Shaohua Li 已提交
3405
		return true;
3406
	if (now + cfqd->cfq_slice_idle * cfqq->dispatched > cfqq->slice_end)
S
Shaohua Li 已提交
3407
		return true;
S
Shaohua Li 已提交
3408

S
Shaohua Li 已提交
3409
	return false;
S
Shaohua Li 已提交
3410 3411
}

3412
static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3413 3414
{
	unsigned int max_dispatch;
3415

3416 3417 3418
	if (cfq_cfqq_must_dispatch(cfqq))
		return true;

3419 3420 3421
	/*
	 * Drain async requests before we start sync IO
	 */
3422
	if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC])
3423
		return false;
3424

3425 3426 3427
	/*
	 * If this is an async queue and we have sync IO in flight, let it wait
	 */
3428
	if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq))
3429
		return false;
3430

S
Shaohua Li 已提交
3431
	max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1);
3432 3433
	if (cfq_class_idle(cfqq))
		max_dispatch = 1;
3434

3435 3436 3437 3438
	/*
	 * Does this cfqq already have too much IO in flight?
	 */
	if (cfqq->dispatched >= max_dispatch) {
3439
		bool promote_sync = false;
3440 3441 3442
		/*
		 * idle queue must always only have a single IO in flight
		 */
3443
		if (cfq_class_idle(cfqq))
3444
			return false;
3445

3446
		/*
3447 3448
		 * If there is only one sync queue
		 * we can ignore async queue here and give the sync
3449 3450 3451 3452
		 * queue no dispatch limit. The reason is a sync queue can
		 * preempt async queue, limiting the sync queue doesn't make
		 * sense. This is useful for aiostress test.
		 */
3453 3454
		if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1)
			promote_sync = true;
3455

3456 3457 3458
		/*
		 * We have other queues, don't allow more IO from this one
		 */
3459 3460
		if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) &&
				!promote_sync)
3461
			return false;
3462

3463
		/*
3464
		 * Sole queue user, no limit
3465
		 */
3466
		if (cfqd->busy_queues == 1 || promote_sync)
S
Shaohua Li 已提交
3467 3468 3469 3470 3471 3472 3473 3474 3475
			max_dispatch = -1;
		else
			/*
			 * Normally we start throttling cfqq when cfq_quantum/2
			 * requests have been dispatched. But we can drive
			 * deeper queue depths at the beginning of slice
			 * subjected to upper limit of cfq_quantum.
			 * */
			max_dispatch = cfqd->cfq_quantum;
3476 3477 3478 3479 3480 3481 3482
	}

	/*
	 * Async queues must wait a bit before being allowed dispatch.
	 * We also ramp up the dispatch depth gradually for async IO,
	 * based on the last sync IO we serviced
	 */
3483
	if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
3484
		u64 last_sync = ktime_get_ns() - cfqd->last_delayed_sync;
3485
		unsigned int depth;
3486

3487
		depth = div64_u64(last_sync, cfqd->cfq_slice[1]);
3488 3489
		if (!depth && !cfqq->dispatched)
			depth = 1;
3490 3491
		if (depth < max_dispatch)
			max_dispatch = depth;
3492
	}
3493

3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509
	/*
	 * If we're below the current max, allow a dispatch
	 */
	return cfqq->dispatched < max_dispatch;
}

/*
 * Dispatch a request from cfqq, moving them to the request queue
 * dispatch list.
 */
static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	struct request *rq;

	BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));

3510 3511 3512 3513
	rq = cfq_check_fifo(cfqq);
	if (rq)
		cfq_mark_cfqq_must_dispatch(cfqq);

3514 3515 3516 3517 3518 3519 3520 3521
	if (!cfq_may_dispatch(cfqd, cfqq))
		return false;

	/*
	 * follow expired path, else get first next available
	 */
	if (!rq)
		rq = cfqq->next_rq;
3522 3523
	else
		cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq);
3524 3525 3526 3527 3528 3529 3530

	/*
	 * insert request into driver dispatch list
	 */
	cfq_dispatch_insert(cfqd->queue, rq);

	if (!cfqd->active_cic) {
3531
		struct cfq_io_cq *cic = RQ_CIC(rq);
3532

3533
		atomic_long_inc(&cic->icq.ioc->refcount);
3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556
		cfqd->active_cic = cic;
	}

	return true;
}

/*
 * Find the cfqq that we need to service and move a request from that to the
 * dispatch list
 */
static int cfq_dispatch_requests(struct request_queue *q, int force)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct cfq_queue *cfqq;

	if (!cfqd->busy_queues)
		return 0;

	if (unlikely(force))
		return cfq_forced_dispatch(cfqd);

	cfqq = cfq_select_queue(cfqd);
	if (!cfqq)
3557 3558
		return 0;

3559
	/*
3560
	 * Dispatch a request from this cfqq, if it is allowed
3561
	 */
3562 3563 3564
	if (!cfq_dispatch_request(cfqd, cfqq))
		return 0;

3565
	cfqq->slice_dispatch++;
3566
	cfq_clear_cfqq_must_dispatch(cfqq);
3567

3568 3569 3570 3571 3572 3573 3574
	/*
	 * expire an async queue immediately if it has used up its slice. idle
	 * queue always expire after 1 dispatch round.
	 */
	if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
	    cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
	    cfq_class_idle(cfqq))) {
3575
		cfqq->slice_end = ktime_get_ns() + 1;
3576
		cfq_slice_expired(cfqd, 0);
L
Linus Torvalds 已提交
3577 3578
	}

3579
	cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
3580
	return 1;
L
Linus Torvalds 已提交
3581 3582 3583
}

/*
J
Jens Axboe 已提交
3584 3585
 * task holds one reference to the queue, dropped when task exits. each rq
 * in-flight on this queue also holds a reference, dropped when rq is freed.
L
Linus Torvalds 已提交
3586
 *
3587
 * Each cfq queue took a reference on the parent group. Drop it now.
L
Linus Torvalds 已提交
3588 3589 3590 3591
 * queue lock must be held here.
 */
static void cfq_put_queue(struct cfq_queue *cfqq)
{
3592
	struct cfq_data *cfqd = cfqq->cfqd;
3593
	struct cfq_group *cfqg;
3594

3595
	BUG_ON(cfqq->ref <= 0);
L
Linus Torvalds 已提交
3596

3597 3598
	cfqq->ref--;
	if (cfqq->ref)
L
Linus Torvalds 已提交
3599 3600
		return;

3601
	cfq_log_cfqq(cfqd, cfqq, "put_queue");
L
Linus Torvalds 已提交
3602
	BUG_ON(rb_first(&cfqq->sort_list));
3603
	BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
3604
	cfqg = cfqq->cfqg;
L
Linus Torvalds 已提交
3605

3606
	if (unlikely(cfqd->active_queue == cfqq)) {
3607
		__cfq_slice_expired(cfqd, cfqq, 0);
3608
		cfq_schedule_dispatch(cfqd);
3609
	}
3610

3611
	BUG_ON(cfq_cfqq_on_rr(cfqq));
L
Linus Torvalds 已提交
3612
	kmem_cache_free(cfq_pool, cfqq);
3613
	cfqg_put(cfqg);
L
Linus Torvalds 已提交
3614 3615
}

3616
static void cfq_put_cooperator(struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
3617
{
J
Jeff Moyer 已提交
3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634
	struct cfq_queue *__cfqq, *next;

	/*
	 * If this queue was scheduled to merge with another queue, be
	 * sure to drop the reference taken on that queue (and others in
	 * the merge chain).  See cfq_setup_merge and cfq_merge_cfqqs.
	 */
	__cfqq = cfqq->new_cfqq;
	while (__cfqq) {
		if (__cfqq == cfqq) {
			WARN(1, "cfqq->new_cfqq loop detected\n");
			break;
		}
		next = __cfqq->new_cfqq;
		cfq_put_queue(__cfqq);
		__cfqq = next;
	}
3635 3636 3637 3638 3639 3640 3641 3642 3643 3644
}

static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	if (unlikely(cfqq == cfqd->active_queue)) {
		__cfq_slice_expired(cfqd, cfqq, 0);
		cfq_schedule_dispatch(cfqd);
	}

	cfq_put_cooperator(cfqq);
J
Jeff Moyer 已提交
3645

3646 3647
	cfq_put_queue(cfqq);
}
3648

3649 3650 3651 3652
static void cfq_init_icq(struct io_cq *icq)
{
	struct cfq_io_cq *cic = icq_to_cic(icq);

3653
	cic->ttime.last_end_request = ktime_get_ns();
3654 3655
}

3656
static void cfq_exit_icq(struct io_cq *icq)
3657
{
3658
	struct cfq_io_cq *cic = icq_to_cic(icq);
3659
	struct cfq_data *cfqd = cic_to_cfqd(cic);
3660

T
Tejun Heo 已提交
3661 3662 3663
	if (cic_to_cfqq(cic, false)) {
		cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false));
		cic_set_cfqq(cic, NULL, false);
3664 3665
	}

T
Tejun Heo 已提交
3666 3667 3668
	if (cic_to_cfqq(cic, true)) {
		cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true));
		cic_set_cfqq(cic, NULL, true);
3669
	}
3670 3671
}

3672
static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic)
3673 3674 3675 3676
{
	struct task_struct *tsk = current;
	int ioprio_class;

J
Jens Axboe 已提交
3677
	if (!cfq_cfqq_prio_changed(cfqq))
3678 3679
		return;

T
Tejun Heo 已提交
3680
	ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
3681
	switch (ioprio_class) {
3682 3683 3684 3685
	default:
		printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
	case IOPRIO_CLASS_NONE:
		/*
3686
		 * no prio set, inherit CPU scheduling settings
3687 3688
		 */
		cfqq->ioprio = task_nice_ioprio(tsk);
3689
		cfqq->ioprio_class = task_nice_ioclass(tsk);
3690 3691
		break;
	case IOPRIO_CLASS_RT:
T
Tejun Heo 已提交
3692
		cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
3693 3694 3695
		cfqq->ioprio_class = IOPRIO_CLASS_RT;
		break;
	case IOPRIO_CLASS_BE:
T
Tejun Heo 已提交
3696
		cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
3697 3698 3699 3700 3701 3702 3703
		cfqq->ioprio_class = IOPRIO_CLASS_BE;
		break;
	case IOPRIO_CLASS_IDLE:
		cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
		cfqq->ioprio = 7;
		cfq_clear_cfqq_idle_window(cfqq);
		break;
3704 3705 3706 3707 3708 3709 3710
	}

	/*
	 * keep track of original prio settings in case we have to temporarily
	 * elevate the priority of this queue
	 */
	cfqq->org_ioprio = cfqq->ioprio;
3711
	cfqq->org_ioprio_class = cfqq->ioprio_class;
J
Jens Axboe 已提交
3712
	cfq_clear_cfqq_prio_changed(cfqq);
3713 3714
}

T
Tejun Heo 已提交
3715
static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio)
3716
{
T
Tejun Heo 已提交
3717
	int ioprio = cic->icq.ioc->ioprio;
3718
	struct cfq_data *cfqd = cic_to_cfqd(cic);
3719
	struct cfq_queue *cfqq;
3720

T
Tejun Heo 已提交
3721 3722 3723 3724 3725
	/*
	 * Check whether ioprio has changed.  The condition may trigger
	 * spuriously on a newly created cic but there's no harm.
	 */
	if (unlikely(!cfqd) || likely(cic->ioprio == ioprio))
3726 3727
		return;

T
Tejun Heo 已提交
3728
	cfqq = cic_to_cfqq(cic, false);
3729
	if (cfqq) {
T
Tejun Heo 已提交
3730
		cfq_put_queue(cfqq);
3731
		cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio);
T
Tejun Heo 已提交
3732
		cic_set_cfqq(cic, cfqq, false);
3733
	}
3734

T
Tejun Heo 已提交
3735
	cfqq = cic_to_cfqq(cic, true);
3736 3737
	if (cfqq)
		cfq_mark_cfqq_prio_changed(cfqq);
T
Tejun Heo 已提交
3738 3739

	cic->ioprio = ioprio;
3740 3741
}

3742
static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3743
			  pid_t pid, bool is_sync)
3744 3745 3746 3747 3748
{
	RB_CLEAR_NODE(&cfqq->rb_node);
	RB_CLEAR_NODE(&cfqq->p_node);
	INIT_LIST_HEAD(&cfqq->fifo);

3749
	cfqq->ref = 0;
3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761
	cfqq->cfqd = cfqd;

	cfq_mark_cfqq_prio_changed(cfqq);

	if (is_sync) {
		if (!cfq_class_idle(cfqq))
			cfq_mark_cfqq_idle_window(cfqq);
		cfq_mark_cfqq_sync(cfqq);
	}
	cfqq->pid = pid;
}

3762
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
3763
static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio)
3764
{
3765
	struct cfq_data *cfqd = cic_to_cfqd(cic);
3766
	struct cfq_queue *cfqq;
T
Tejun Heo 已提交
3767
	uint64_t serial_nr;
J
Jens Axboe 已提交
3768
	bool nonroot_cg;
3769

T
Tejun Heo 已提交
3770
	rcu_read_lock();
T
Tejun Heo 已提交
3771
	serial_nr = bio_blkcg(bio)->css.serial_nr;
J
Jens Axboe 已提交
3772
	nonroot_cg = bio_blkcg(bio) != &blkcg_root;
T
Tejun Heo 已提交
3773
	rcu_read_unlock();
3774

T
Tejun Heo 已提交
3775 3776 3777 3778
	/*
	 * Check whether blkcg has changed.  The condition may trigger
	 * spuriously on a newly created cic but there's no harm.
	 */
T
Tejun Heo 已提交
3779
	if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr))
T
Tejun Heo 已提交
3780
		return;
3781

J
Jens Axboe 已提交
3782 3783 3784
	/*
	 * If we have a non-root cgroup, we can depend on that to
	 * do proper throttling of writes. Turn off wbt for that
3785
	 * case, if it was enabled by default.
J
Jens Axboe 已提交
3786
	 */
3787 3788
	if (nonroot_cg)
		wbt_disable_default(cfqd->queue);
J
Jens Axboe 已提交
3789

3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805
	/*
	 * Drop reference to queues.  New queues will be assigned in new
	 * group upon arrival of fresh requests.
	 */
	cfqq = cic_to_cfqq(cic, false);
	if (cfqq) {
		cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
		cic_set_cfqq(cic, NULL, false);
		cfq_put_queue(cfqq);
	}

	cfqq = cic_to_cfqq(cic, true);
	if (cfqq) {
		cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
		cic_set_cfqq(cic, NULL, true);
		cfq_put_queue(cfqq);
3806
	}
T
Tejun Heo 已提交
3807

T
Tejun Heo 已提交
3808
	cic->blkcg_serial_nr = serial_nr;
3809
}
T
Tejun Heo 已提交
3810 3811
#else
static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { }
3812 3813
#endif  /* CONFIG_CFQ_GROUP_IOSCHED */

3814
static struct cfq_queue **
3815
cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio)
3816
{
3817
	switch (ioprio_class) {
3818
	case IOPRIO_CLASS_RT:
3819
		return &cfqg->async_cfqq[0][ioprio];
T
Tejun Heo 已提交
3820 3821 3822
	case IOPRIO_CLASS_NONE:
		ioprio = IOPRIO_NORM;
		/* fall through */
3823
	case IOPRIO_CLASS_BE:
3824
		return &cfqg->async_cfqq[1][ioprio];
3825
	case IOPRIO_CLASS_IDLE:
3826
		return &cfqg->async_idle_cfqq;
3827 3828 3829 3830 3831
	default:
		BUG();
	}
}

3832
static struct cfq_queue *
3833
cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic,
3834
	      struct bio *bio)
3835
{
3836 3837
	int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
	int ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
3838
	struct cfq_queue **async_cfqq = NULL;
3839
	struct cfq_queue *cfqq;
3840 3841 3842
	struct cfq_group *cfqg;

	rcu_read_lock();
3843
	cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio));
3844 3845 3846 3847
	if (!cfqg) {
		cfqq = &cfqd->oom_cfqq;
		goto out;
	}
3848

3849
	if (!is_sync) {
3850 3851 3852 3853 3854
		if (!ioprio_valid(cic->ioprio)) {
			struct task_struct *tsk = current;
			ioprio = task_nice_ioprio(tsk);
			ioprio_class = task_nice_ioclass(tsk);
		}
3855
		async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio);
3856
		cfqq = *async_cfqq;
3857 3858
		if (cfqq)
			goto out;
3859 3860
	}

3861 3862
	cfqq = kmem_cache_alloc_node(cfq_pool,
				     GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN,
3863 3864 3865 3866 3867 3868
				     cfqd->queue->node);
	if (!cfqq) {
		cfqq = &cfqd->oom_cfqq;
		goto out;
	}

3869 3870
	/* cfq_init_cfqq() assumes cfqq->ioprio_class is initialized. */
	cfqq->ioprio_class = IOPRIO_CLASS_NONE;
3871 3872 3873 3874
	cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
	cfq_init_prio_data(cfqq, cic);
	cfq_link_cfqq_cfqg(cfqq, cfqg);
	cfq_log_cfqq(cfqd, cfqq, "alloced");
3875

3876 3877
	if (async_cfqq) {
		/* a new async queue is created, pin and remember */
3878
		cfqq->ref++;
3879
		*async_cfqq = cfqq;
3880
	}
3881
out:
3882
	cfqq->ref++;
3883
	rcu_read_unlock();
3884 3885 3886
	return cfqq;
}

3887
static void
3888
__cfq_update_io_thinktime(struct cfq_ttime *ttime, u64 slice_idle)
L
Linus Torvalds 已提交
3889
{
3890
	u64 elapsed = ktime_get_ns() - ttime->last_end_request;
3891
	elapsed = min(elapsed, 2UL * slice_idle);
3892

3893
	ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
3894 3895 3896
	ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed,  8);
	ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,
				     ttime->ttime_samples);
3897 3898 3899 3900
}

static void
cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3901
			struct cfq_io_cq *cic)
3902
{
3903
	if (cfq_cfqq_sync(cfqq)) {
3904
		__cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle);
3905 3906 3907
		__cfq_update_io_thinktime(&cfqq->service_tree->ttime,
			cfqd->cfq_slice_idle);
	}
S
Shaohua Li 已提交
3908 3909 3910
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	__cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle);
#endif
3911
}
L
Linus Torvalds 已提交
3912

3913
static void
3914
cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
J
Jens Axboe 已提交
3915
		       struct request *rq)
3916
{
3917
	sector_t sdist = 0;
3918
	sector_t n_sec = blk_rq_sectors(rq);
3919 3920 3921 3922 3923 3924
	if (cfqq->last_request_pos) {
		if (cfqq->last_request_pos < blk_rq_pos(rq))
			sdist = blk_rq_pos(rq) - cfqq->last_request_pos;
		else
			sdist = cfqq->last_request_pos - blk_rq_pos(rq);
	}
3925

3926
	cfqq->seek_history <<= 1;
3927 3928 3929 3930
	if (blk_queue_nonrot(cfqd->queue))
		cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT);
	else
		cfqq->seek_history |= (sdist > CFQQ_SEEK_THR);
3931
}
L
Linus Torvalds 已提交
3932

3933 3934 3935 3936 3937 3938
static inline bool req_noidle(struct request *req)
{
	return req_op(req) == REQ_OP_WRITE &&
		(req->cmd_flags & (REQ_SYNC | REQ_IDLE)) == REQ_SYNC;
}

3939 3940 3941 3942 3943 3944
/*
 * Disable idle window if the process thinks too long or seeks so much that
 * it doesn't matter
 */
static void
cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3945
		       struct cfq_io_cq *cic)
3946
{
3947
	int old_idle, enable_idle;
3948

3949 3950 3951 3952
	/*
	 * Don't idle for async or idle io prio class
	 */
	if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
3953 3954
		return;

3955
	enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
L
Linus Torvalds 已提交
3956

3957 3958 3959
	if (cfqq->queued[0] + cfqq->queued[1] >= 4)
		cfq_mark_cfqq_deep(cfqq);

3960
	if (cfqq->next_rq && req_noidle(cfqq->next_rq))
3961
		enable_idle = 0;
T
Tejun Heo 已提交
3962
	else if (!atomic_read(&cic->icq.ioc->active_ref) ||
3963 3964
		 !cfqd->cfq_slice_idle ||
		 (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq)))
3965
		enable_idle = 0;
3966 3967
	else if (sample_valid(cic->ttime.ttime_samples)) {
		if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle)
3968 3969 3970
			enable_idle = 0;
		else
			enable_idle = 1;
L
Linus Torvalds 已提交
3971 3972
	}

3973 3974 3975 3976 3977 3978 3979
	if (old_idle != enable_idle) {
		cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
		if (enable_idle)
			cfq_mark_cfqq_idle_window(cfqq);
		else
			cfq_clear_cfqq_idle_window(cfqq);
	}
3980
}
L
Linus Torvalds 已提交
3981

3982 3983 3984 3985
/*
 * Check if new_cfqq should preempt the currently active queue. Return 0 for
 * no or if we aren't sure, a 1 will cause a preempt.
 */
3986
static bool
3987
cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
J
Jens Axboe 已提交
3988
		   struct request *rq)
3989
{
J
Jens Axboe 已提交
3990
	struct cfq_queue *cfqq;
3991

J
Jens Axboe 已提交
3992 3993
	cfqq = cfqd->active_queue;
	if (!cfqq)
3994
		return false;
3995

J
Jens Axboe 已提交
3996
	if (cfq_class_idle(new_cfqq))
3997
		return false;
3998 3999

	if (cfq_class_idle(cfqq))
4000
		return true;
4001

4002 4003 4004 4005 4006 4007
	/*
	 * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice.
	 */
	if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq))
		return false;

4008 4009 4010 4011
	/*
	 * if the new request is sync, but the currently running queue is
	 * not, let the sync request have priority.
	 */
4012
	if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq))
4013
		return true;
4014

4015 4016 4017 4018 4019 4020
	/*
	 * Treat ancestors of current cgroup the same way as current cgroup.
	 * For anybody else we disallow preemption to guarantee service
	 * fairness among cgroups.
	 */
	if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg))
4021 4022 4023 4024 4025
		return false;

	if (cfq_slice_used(cfqq))
		return true;

4026 4027 4028 4029 4030 4031 4032
	/*
	 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
	 */
	if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
		return true;

	WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class);
4033
	/* Allow preemption only if we are idling on sync-noidle tree */
4034
	if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD &&
4035 4036 4037 4038
	    cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD &&
	    RB_EMPTY_ROOT(&cfqq->sort_list))
		return true;

4039 4040 4041 4042
	/*
	 * So both queues are sync. Let the new request get disk time if
	 * it's a metadata request and the current queue is doing regular IO.
	 */
4043
	if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending)
4044 4045
		return true;

4046 4047 4048 4049
	/* An idle queue should not be idle now for some reason */
	if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq))
		return true;

4050
	if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
4051
		return false;
4052 4053 4054 4055 4056

	/*
	 * if this request is as-good as one we would expect from the
	 * current cfqq, let it preempt
	 */
4057
	if (cfq_rq_close(cfqd, cfqq, rq))
4058
		return true;
4059

4060
	return false;
4061 4062 4063 4064 4065 4066 4067 4068
}

/*
 * cfqq preempts the active queue. if we allowed preempt with no slice left,
 * let it have half of its nominal slice.
 */
static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
S
Shaohua Li 已提交
4069 4070
	enum wl_type_t old_type = cfqq_type(cfqd->active_queue);

4071
	cfq_log_cfqq(cfqd, cfqq, "preempt");
S
Shaohua Li 已提交
4072
	cfq_slice_expired(cfqd, 1);
4073

4074 4075 4076 4077
	/*
	 * workload type is changed, don't save slice, otherwise preempt
	 * doesn't happen
	 */
S
Shaohua Li 已提交
4078
	if (old_type != cfqq_type(cfqq))
4079
		cfqq->cfqg->saved_wl_slice = 0;
4080

4081 4082 4083 4084 4085
	/*
	 * Put the new queue at the front of the of the current list,
	 * so we know that it will be selected next.
	 */
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
4086 4087

	cfq_service_tree_add(cfqd, cfqq, 1);
4088

4089 4090
	cfqq->slice_end = 0;
	cfq_mark_cfqq_slice_new(cfqq);
4091 4092 4093
}

/*
J
Jens Axboe 已提交
4094
 * Called when a new fs request (rq) is added (to cfqq). Check if there's
4095 4096 4097
 * something we should do about it
 */
static void
J
Jens Axboe 已提交
4098 4099
cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		struct request *rq)
4100
{
4101
	struct cfq_io_cq *cic = RQ_CIC(rq);
4102

4103
	cfqd->rq_queued++;
4104 4105
	if (rq->cmd_flags & REQ_PRIO)
		cfqq->prio_pending++;
4106

4107
	cfq_update_io_thinktime(cfqd, cfqq, cic);
4108
	cfq_update_io_seektime(cfqd, cfqq, rq);
J
Jens Axboe 已提交
4109 4110
	cfq_update_idle_window(cfqd, cfqq, cic);

4111
	cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
4112 4113 4114

	if (cfqq == cfqd->active_queue) {
		/*
4115 4116 4117
		 * Remember that we saw a request from this process, but
		 * don't start queuing just yet. Otherwise we risk seeing lots
		 * of tiny requests, because we disrupt the normal plugging
4118 4119
		 * and merging. If the request is already larger than a single
		 * page, let it rip immediately. For that case we assume that
4120 4121 4122
		 * merging is already done. Ditto for a busy system that
		 * has other work pending, don't risk delaying until the
		 * idle timer unplug to continue working.
4123
		 */
4124
		if (cfq_cfqq_wait_request(cfqq)) {
4125
			if (blk_rq_bytes(rq) > PAGE_SIZE ||
4126
			    cfqd->busy_queues > 1) {
4127
				cfq_del_timer(cfqd, cfqq);
4128
				cfq_clear_cfqq_wait_request(cfqq);
4129
				__blk_run_queue(cfqd->queue);
4130
			} else {
4131
				cfqg_stats_update_idle_time(cfqq->cfqg);
4132
				cfq_mark_cfqq_must_dispatch(cfqq);
4133
			}
4134
		}
J
Jens Axboe 已提交
4135
	} else if (cfq_should_preempt(cfqd, cfqq, rq)) {
4136 4137 4138
		/*
		 * not the active queue - expire current slice if it is
		 * idle and has expired it's mean thinktime or this new queue
4139 4140
		 * has some old slice time left and is of higher priority or
		 * this new queue is RT and the current one is BE
4141 4142
		 */
		cfq_preempt_queue(cfqd, cfqq);
4143
		__blk_run_queue(cfqd->queue);
4144
	}
L
Linus Torvalds 已提交
4145 4146
}

4147
static void cfq_insert_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
4148
{
4149
	struct cfq_data *cfqd = q->elevator->elevator_data;
J
Jens Axboe 已提交
4150
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
4151

4152
	cfq_log_cfqq(cfqd, cfqq, "insert_request");
4153
	cfq_init_prio_data(cfqq, RQ_CIC(rq));
L
Linus Torvalds 已提交
4154

4155
	rq->fifo_time = ktime_get_ns() + cfqd->cfq_fifo_expire[rq_is_sync(rq)];
4156
	list_add_tail(&rq->queuelist, &cfqq->fifo);
4157
	cfq_add_rq_rb(rq);
4158
	cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group,
4159
				 rq->cmd_flags);
J
Jens Axboe 已提交
4160
	cfq_rq_enqueued(cfqd, cfqq, rq);
L
Linus Torvalds 已提交
4161 4162
}

4163 4164 4165 4166 4167 4168
/*
 * Update hw_tag based on peak queue depth over 50 samples under
 * sufficient load.
 */
static void cfq_update_hw_tag(struct cfq_data *cfqd)
{
S
Shaohua Li 已提交
4169 4170
	struct cfq_queue *cfqq = cfqd->active_queue;

4171 4172
	if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth)
		cfqd->hw_tag_est_depth = cfqd->rq_in_driver;
4173 4174 4175

	if (cfqd->hw_tag == 1)
		return;
4176 4177

	if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
4178
	    cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
4179 4180
		return;

S
Shaohua Li 已提交
4181 4182 4183 4184 4185 4186 4187
	/*
	 * If active queue hasn't enough requests and can idle, cfq might not
	 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
	 * case
	 */
	if (cfqq && cfq_cfqq_idle_window(cfqq) &&
	    cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] <
4188
	    CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN)
S
Shaohua Li 已提交
4189 4190
		return;

4191 4192 4193
	if (cfqd->hw_tag_samples++ < 50)
		return;

4194
	if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN)
4195 4196 4197 4198 4199
		cfqd->hw_tag = 1;
	else
		cfqd->hw_tag = 0;
}

4200 4201
static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
4202
	struct cfq_io_cq *cic = cfqd->active_cic;
4203
	u64 now = ktime_get_ns();
4204

4205 4206 4207 4208
	/* If the queue already has requests, don't wait */
	if (!RB_EMPTY_ROOT(&cfqq->sort_list))
		return false;

4209 4210 4211 4212
	/* If there are other queues in the group, don't wait */
	if (cfqq->cfqg->nr_cfqq > 1)
		return false;

S
Shaohua Li 已提交
4213 4214 4215 4216
	/* the only queue in the group, but think time is big */
	if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true))
		return false;

4217 4218 4219 4220
	if (cfq_slice_used(cfqq))
		return true;

	/* if slice left is less than think time, wait busy */
4221
	if (cic && sample_valid(cic->ttime.ttime_samples)
4222
	    && (cfqq->slice_end - now < cic->ttime.ttime_mean))
4223 4224 4225 4226 4227 4228 4229 4230 4231
		return true;

	/*
	 * If think times is less than a jiffy than ttime_mean=0 and above
	 * will not be true. It might happen that slice has not expired yet
	 * but will expire soon (4-5 ns) during select_queue(). To cover the
	 * case where think time is less than a jiffy, mark the queue wait
	 * busy if only 1 jiffy is left in the slice.
	 */
4232
	if (cfqq->slice_end - now <= jiffies_to_nsecs(1))
4233 4234 4235 4236 4237
		return true;

	return false;
}

4238
static void cfq_completed_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
4239
{
J
Jens Axboe 已提交
4240
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
4241
	struct cfq_data *cfqd = cfqq->cfqd;
4242
	const int sync = rq_is_sync(rq);
4243
	u64 now = ktime_get_ns();
L
Linus Torvalds 已提交
4244

4245
	cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", req_noidle(rq));
L
Linus Torvalds 已提交
4246

4247 4248
	cfq_update_hw_tag(cfqd);

4249
	WARN_ON(!cfqd->rq_in_driver);
J
Jens Axboe 已提交
4250
	WARN_ON(!cfqq->dispatched);
4251
	cfqd->rq_in_driver--;
J
Jens Axboe 已提交
4252
	cfqq->dispatched--;
4253
	(RQ_CFQG(rq))->dispatched--;
4254
	cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq),
4255
				     rq_io_start_time_ns(rq), rq->cmd_flags);
L
Linus Torvalds 已提交
4256

4257
	cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--;
4258

4259
	if (sync) {
4260
		struct cfq_rb_root *st;
4261

4262
		RQ_CIC(rq)->ttime.last_end_request = now;
4263 4264

		if (cfq_cfqq_on_rr(cfqq))
4265
			st = cfqq->service_tree;
4266
		else
4267 4268 4269 4270
			st = st_for(cfqq->cfqg, cfqq_class(cfqq),
					cfqq_type(cfqq));

		st->ttime.last_end_request = now;
4271 4272 4273 4274 4275 4276 4277 4278 4279 4280
		/*
		 * We have to do this check in jiffies since start_time is in
		 * jiffies and it is not trivial to convert to ns. If
		 * cfq_fifo_expire[1] ever comes close to 1 jiffie, this test
		 * will become problematic but so far we are fine (the default
		 * is 128 ms).
		 */
		if (!time_after(rq->start_time +
				  nsecs_to_jiffies(cfqd->cfq_fifo_expire[1]),
				jiffies))
4281
			cfqd->last_delayed_sync = now;
4282
	}
4283

S
Shaohua Li 已提交
4284 4285 4286 4287
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	cfqq->cfqg->ttime.last_end_request = now;
#endif

4288 4289 4290 4291 4292
	/*
	 * If this is the active queue, check if it needs to be expired,
	 * or if we want to idle in case it has no pending requests.
	 */
	if (cfqd->active_queue == cfqq) {
4293 4294
		const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);

4295 4296 4297 4298
		if (cfq_cfqq_slice_new(cfqq)) {
			cfq_set_prio_slice(cfqd, cfqq);
			cfq_clear_cfqq_slice_new(cfqq);
		}
4299 4300

		/*
4301 4302
		 * Should we wait for next request to come in before we expire
		 * the queue.
4303
		 */
4304
		if (cfq_should_wait_busy(cfqd, cfqq)) {
4305
			u64 extend_sl = cfqd->cfq_slice_idle;
4306 4307
			if (!cfqd->cfq_slice_idle)
				extend_sl = cfqd->cfq_group_idle;
4308
			cfqq->slice_end = now + extend_sl;
4309
			cfq_mark_cfqq_wait_busy(cfqq);
4310
			cfq_log_cfqq(cfqd, cfqq, "will busy wait");
4311 4312
		}

4313
		/*
4314 4315 4316 4317 4318 4319
		 * Idling is not enabled on:
		 * - expired queues
		 * - idle-priority queues
		 * - async queues
		 * - queues with still some requests queued
		 * - when there is a close cooperator
4320
		 */
4321
		if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
4322
			cfq_slice_expired(cfqd, 1);
4323 4324
		else if (sync && cfqq_empty &&
			 !cfq_close_cooperator(cfqd, cfqq)) {
4325
			cfq_arm_slice_timer(cfqd);
4326
		}
4327
	}
J
Jens Axboe 已提交
4328

4329
	if (!cfqd->rq_in_driver)
4330
		cfq_schedule_dispatch(cfqd);
L
Linus Torvalds 已提交
4331 4332
}

4333
static void cfqq_boost_on_prio(struct cfq_queue *cfqq, unsigned int op)
4334 4335 4336 4337 4338 4339
{
	/*
	 * If REQ_PRIO is set, boost class and prio level, if it's below
	 * BE/NORM. If prio is not set, restore the potentially boosted
	 * class/prio level.
	 */
4340
	if (!(op & REQ_PRIO)) {
4341 4342 4343 4344 4345 4346 4347 4348 4349 4350
		cfqq->ioprio_class = cfqq->org_ioprio_class;
		cfqq->ioprio = cfqq->org_ioprio;
	} else {
		if (cfq_class_idle(cfqq))
			cfqq->ioprio_class = IOPRIO_CLASS_BE;
		if (cfqq->ioprio > IOPRIO_NORM)
			cfqq->ioprio = IOPRIO_NORM;
	}
}

4351
static inline int __cfq_may_queue(struct cfq_queue *cfqq)
4352
{
4353
	if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
J
Jens Axboe 已提交
4354
		cfq_mark_cfqq_must_alloc_slice(cfqq);
4355
		return ELV_MQUEUE_MUST;
J
Jens Axboe 已提交
4356
	}
L
Linus Torvalds 已提交
4357

4358 4359 4360
	return ELV_MQUEUE_MAY;
}

4361
static int cfq_may_queue(struct request_queue *q, unsigned int op)
4362 4363 4364
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct task_struct *tsk = current;
4365
	struct cfq_io_cq *cic;
4366 4367 4368 4369 4370 4371 4372 4373
	struct cfq_queue *cfqq;

	/*
	 * don't force setup of a queue from here, as a call to may_queue
	 * does not necessarily imply that a request actually will be queued.
	 * so just lookup a possibly existing queue, or return 'may queue'
	 * if that fails
	 */
4374
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
4375 4376 4377
	if (!cic)
		return ELV_MQUEUE_MAY;

4378
	cfqq = cic_to_cfqq(cic, op_is_sync(op));
4379
	if (cfqq) {
4380
		cfq_init_prio_data(cfqq, cic);
4381
		cfqq_boost_on_prio(cfqq, op);
4382

4383
		return __cfq_may_queue(cfqq);
4384 4385 4386
	}

	return ELV_MQUEUE_MAY;
L
Linus Torvalds 已提交
4387 4388 4389 4390 4391
}

/*
 * queue lock held here
 */
4392
static void cfq_put_request(struct request *rq)
L
Linus Torvalds 已提交
4393
{
J
Jens Axboe 已提交
4394
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
L
Linus Torvalds 已提交
4395

J
Jens Axboe 已提交
4396
	if (cfqq) {
4397
		const int rw = rq_data_dir(rq);
L
Linus Torvalds 已提交
4398

4399 4400
		BUG_ON(!cfqq->allocated[rw]);
		cfqq->allocated[rw]--;
L
Linus Torvalds 已提交
4401

4402
		/* Put down rq reference on cfqg */
4403
		cfqg_put(RQ_CFQG(rq));
4404 4405
		rq->elv.priv[0] = NULL;
		rq->elv.priv[1] = NULL;
4406

L
Linus Torvalds 已提交
4407 4408 4409 4410
		cfq_put_queue(cfqq);
	}
}

J
Jeff Moyer 已提交
4411
static struct cfq_queue *
4412
cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic,
J
Jeff Moyer 已提交
4413 4414 4415 4416
		struct cfq_queue *cfqq)
{
	cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
	cic_set_cfqq(cic, cfqq->new_cfqq, 1);
4417
	cfq_mark_cfqq_coop(cfqq->new_cfqq);
J
Jeff Moyer 已提交
4418 4419 4420 4421
	cfq_put_queue(cfqq);
	return cic_to_cfqq(cic, 1);
}

4422 4423 4424 4425 4426
/*
 * Returns NULL if a new cfqq should be allocated, or the old cfqq if this
 * was the last process referring to said cfqq.
 */
static struct cfq_queue *
4427
split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq)
4428 4429 4430 4431
{
	if (cfqq_process_refs(cfqq) == 1) {
		cfqq->pid = current->pid;
		cfq_clear_cfqq_coop(cfqq);
4432
		cfq_clear_cfqq_split_coop(cfqq);
4433 4434 4435 4436
		return cfqq;
	}

	cic_set_cfqq(cic, NULL, 1);
4437 4438 4439

	cfq_put_cooperator(cfqq);

4440 4441 4442
	cfq_put_queue(cfqq);
	return NULL;
}
L
Linus Torvalds 已提交
4443
/*
4444
 * Allocate cfq data structures associated with this request.
L
Linus Torvalds 已提交
4445
 */
4446
static int
4447 4448
cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio,
		gfp_t gfp_mask)
L
Linus Torvalds 已提交
4449 4450
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
4451
	struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq);
L
Linus Torvalds 已提交
4452
	const int rw = rq_data_dir(rq);
4453
	const bool is_sync = rq_is_sync(rq);
4454
	struct cfq_queue *cfqq;
L
Linus Torvalds 已提交
4455

4456
	spin_lock_irq(q->queue_lock);
4457

T
Tejun Heo 已提交
4458 4459
	check_ioprio_changed(cic, bio);
	check_blkcg_changed(cic, bio);
4460
new_queue:
4461
	cfqq = cic_to_cfqq(cic, is_sync);
4462
	if (!cfqq || cfqq == &cfqd->oom_cfqq) {
4463 4464
		if (cfqq)
			cfq_put_queue(cfqq);
4465
		cfqq = cfq_get_queue(cfqd, is_sync, cic, bio);
4466
		cic_set_cfqq(cic, cfqq, is_sync);
J
Jeff Moyer 已提交
4467
	} else {
4468 4469 4470
		/*
		 * If the queue was seeky for too long, break it apart.
		 */
4471
		if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) {
4472 4473 4474 4475 4476 4477
			cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
			cfqq = split_cfqq(cic, cfqq);
			if (!cfqq)
				goto new_queue;
		}

J
Jeff Moyer 已提交
4478 4479 4480 4481 4482 4483 4484 4485
		/*
		 * Check to see if this queue is scheduled to merge with
		 * another, closely cooperating queue.  The merging of
		 * queues happens here as it must be done in process context.
		 * The reference on new_cfqq was taken in merge_cfqqs.
		 */
		if (cfqq->new_cfqq)
			cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq);
4486
	}
L
Linus Torvalds 已提交
4487 4488 4489

	cfqq->allocated[rw]++;

4490
	cfqq->ref++;
4491
	cfqg_get(cfqq->cfqg);
4492
	rq->elv.priv[0] = cfqq;
T
Tejun Heo 已提交
4493
	rq->elv.priv[1] = cfqq->cfqg;
4494
	spin_unlock_irq(q->queue_lock);
J
Jens Axboe 已提交
4495
	return 0;
L
Linus Torvalds 已提交
4496 4497
}

4498
static void cfq_kick_queue(struct work_struct *work)
4499
{
4500
	struct cfq_data *cfqd =
4501
		container_of(work, struct cfq_data, unplug_work);
4502
	struct request_queue *q = cfqd->queue;
4503

4504
	spin_lock_irq(q->queue_lock);
4505
	__blk_run_queue(cfqd->queue);
4506
	spin_unlock_irq(q->queue_lock);
4507 4508 4509 4510 4511
}

/*
 * Timer running if the active_queue is currently idling inside its time slice
 */
4512
static enum hrtimer_restart cfq_idle_slice_timer(struct hrtimer *timer)
4513
{
4514 4515
	struct cfq_data *cfqd = container_of(timer, struct cfq_data,
					     idle_slice_timer);
4516 4517
	struct cfq_queue *cfqq;
	unsigned long flags;
4518
	int timed_out = 1;
4519

4520 4521
	cfq_log(cfqd, "idle timer fired");

4522 4523
	spin_lock_irqsave(cfqd->queue->queue_lock, flags);

4524 4525
	cfqq = cfqd->active_queue;
	if (cfqq) {
4526 4527
		timed_out = 0;

4528 4529 4530 4531 4532 4533
		/*
		 * We saw a request before the queue expired, let it through
		 */
		if (cfq_cfqq_must_dispatch(cfqq))
			goto out_kick;

4534 4535 4536
		/*
		 * expired
		 */
4537
		if (cfq_slice_used(cfqq))
4538 4539 4540 4541 4542 4543
			goto expire;

		/*
		 * only expire and reinvoke request handler, if there are
		 * other queues with pending requests
		 */
4544
		if (!cfqd->busy_queues)
4545 4546 4547 4548 4549
			goto out_cont;

		/*
		 * not expired and it has a request pending, let it dispatch
		 */
4550
		if (!RB_EMPTY_ROOT(&cfqq->sort_list))
4551
			goto out_kick;
4552 4553 4554 4555 4556

		/*
		 * Queue depth flag is reset only when the idle didn't succeed
		 */
		cfq_clear_cfqq_deep(cfqq);
4557 4558
	}
expire:
4559
	cfq_slice_expired(cfqd, timed_out);
4560
out_kick:
4561
	cfq_schedule_dispatch(cfqd);
4562 4563
out_cont:
	spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
4564
	return HRTIMER_NORESTART;
4565 4566
}

J
Jens Axboe 已提交
4567 4568
static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
{
4569
	hrtimer_cancel(&cfqd->idle_slice_timer);
4570
	cancel_work_sync(&cfqd->unplug_work);
J
Jens Axboe 已提交
4571
}
4572

J
Jens Axboe 已提交
4573
static void cfq_exit_queue(struct elevator_queue *e)
L
Linus Torvalds 已提交
4574
{
4575
	struct cfq_data *cfqd = e->elevator_data;
4576
	struct request_queue *q = cfqd->queue;
4577

J
Jens Axboe 已提交
4578
	cfq_shutdown_timer_wq(cfqd);
4579

4580
	spin_lock_irq(q->queue_lock);
4581

4582
	if (cfqd->active_queue)
4583
		__cfq_slice_expired(cfqd, cfqd->active_queue, 0);
4584

4585 4586
	spin_unlock_irq(q->queue_lock);

4587 4588
	cfq_shutdown_timer_wq(cfqd);

4589 4590 4591
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	blkcg_deactivate_policy(q, &blkcg_policy_cfq);
#else
4592
	kfree(cfqd->root_group);
4593
#endif
4594
	kfree(cfqd);
L
Linus Torvalds 已提交
4595 4596
}

4597
static int cfq_init_queue(struct request_queue *q, struct elevator_type *e)
L
Linus Torvalds 已提交
4598 4599
{
	struct cfq_data *cfqd;
T
Tejun Heo 已提交
4600
	struct blkcg_gq *blkg __maybe_unused;
4601
	int i, ret;
4602 4603 4604 4605 4606
	struct elevator_queue *eq;

	eq = elevator_alloc(q, e);
	if (!eq)
		return -ENOMEM;
L
Linus Torvalds 已提交
4607

4608
	cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node);
4609 4610
	if (!cfqd) {
		kobject_put(&eq->kobj);
4611
		return -ENOMEM;
4612 4613
	}
	eq->elevator_data = cfqd;
4614

4615
	cfqd->queue = q;
4616 4617 4618
	spin_lock_irq(q->queue_lock);
	q->elevator = eq;
	spin_unlock_irq(q->queue_lock);
4619

4620 4621 4622
	/* Init root service tree */
	cfqd->grp_service_tree = CFQ_RB_ROOT;

4623
	/* Init root group and prefer root group over other groups by default */
4624
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
4625
	ret = blkcg_activate_policy(q, &blkcg_policy_cfq);
4626 4627
	if (ret)
		goto out_free;
4628

4629
	cfqd->root_group = blkg_to_cfqg(q->root_blkg);
4630
#else
4631
	ret = -ENOMEM;
4632 4633
	cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group),
					GFP_KERNEL, cfqd->queue->node);
4634 4635
	if (!cfqd->root_group)
		goto out_free;
4636

4637
	cfq_init_cfqg_base(cfqd->root_group);
4638 4639
	cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
	cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
4640
#endif
4641

4642 4643 4644 4645 4646 4647 4648 4649
	/*
	 * Not strictly needed (since RB_ROOT just clears the node and we
	 * zeroed cfqd on alloc), but better be safe in case someone decides
	 * to add magic to the rb code
	 */
	for (i = 0; i < CFQ_PRIO_LISTS; i++)
		cfqd->prio_trees[i] = RB_ROOT;

4650
	/*
4651
	 * Our fallback cfqq if cfq_get_queue() runs into OOM issues.
4652
	 * Grab a permanent reference to it, so that the normal code flow
4653 4654 4655
	 * will not attempt to free it.  oom_cfqq is linked to root_group
	 * but shouldn't hold a reference as it'll never be unlinked.  Lose
	 * the reference from linking right away.
4656 4657
	 */
	cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
4658
	cfqd->oom_cfqq.ref++;
T
Tejun Heo 已提交
4659 4660

	spin_lock_irq(q->queue_lock);
4661
	cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group);
4662
	cfqg_put(cfqd->root_group);
T
Tejun Heo 已提交
4663
	spin_unlock_irq(q->queue_lock);
L
Linus Torvalds 已提交
4664

4665 4666
	hrtimer_init(&cfqd->idle_slice_timer, CLOCK_MONOTONIC,
		     HRTIMER_MODE_REL);
4667 4668
	cfqd->idle_slice_timer.function = cfq_idle_slice_timer;

4669
	INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
4670

L
Linus Torvalds 已提交
4671
	cfqd->cfq_quantum = cfq_quantum;
4672 4673
	cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
	cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
L
Linus Torvalds 已提交
4674 4675
	cfqd->cfq_back_max = cfq_back_max;
	cfqd->cfq_back_penalty = cfq_back_penalty;
4676 4677
	cfqd->cfq_slice[0] = cfq_slice_async;
	cfqd->cfq_slice[1] = cfq_slice_sync;
4678
	cfqd->cfq_target_latency = cfq_target_latency;
4679
	cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
4680
	cfqd->cfq_slice_idle = cfq_slice_idle;
4681
	cfqd->cfq_group_idle = cfq_group_idle;
4682
	cfqd->cfq_latency = 1;
4683
	cfqd->hw_tag = -1;
4684 4685 4686 4687
	/*
	 * we optimistically start assuming sync ops weren't delayed in last
	 * second, in order to have larger depth for async operations.
	 */
4688
	cfqd->last_delayed_sync = ktime_get_ns() - NSEC_PER_SEC;
4689
	return 0;
4690 4691 4692

out_free:
	kfree(cfqd);
4693
	kobject_put(&eq->kobj);
4694
	return ret;
L
Linus Torvalds 已提交
4695 4696
}

4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708
static void cfq_registered_queue(struct request_queue *q)
{
	struct elevator_queue *e = q->elevator;
	struct cfq_data *cfqd = e->elevator_data;

	/*
	 * Default to IOPS mode with no idling for SSDs
	 */
	if (blk_queue_nonrot(q))
		cfqd->cfq_slice_idle = 0;
}

L
Linus Torvalds 已提交
4709 4710 4711 4712 4713 4714
/*
 * sysfs parts below -->
 */
static ssize_t
cfq_var_show(unsigned int var, char *page)
{
4715
	return sprintf(page, "%u\n", var);
L
Linus Torvalds 已提交
4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727
}

static ssize_t
cfq_var_store(unsigned int *var, const char *page, size_t count)
{
	char *p = (char *) page;

	*var = simple_strtoul(p, &p, 10);
	return count;
}

#define SHOW_FUNCTION(__FUNC, __VAR, __CONV)				\
J
Jens Axboe 已提交
4728
static ssize_t __FUNC(struct elevator_queue *e, char *page)		\
L
Linus Torvalds 已提交
4729
{									\
4730
	struct cfq_data *cfqd = e->elevator_data;			\
4731
	u64 __data = __VAR;						\
L
Linus Torvalds 已提交
4732
	if (__CONV)							\
4733
		__data = div_u64(__data, NSEC_PER_MSEC);			\
L
Linus Torvalds 已提交
4734 4735 4736
	return cfq_var_show(__data, (page));				\
}
SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
4737 4738
SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
4739 4740
SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
4741
SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
4742
SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1);
4743 4744 4745
SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
4746
SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
4747
SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1);
L
Linus Torvalds 已提交
4748 4749
#undef SHOW_FUNCTION

4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764
#define USEC_SHOW_FUNCTION(__FUNC, __VAR)				\
static ssize_t __FUNC(struct elevator_queue *e, char *page)		\
{									\
	struct cfq_data *cfqd = e->elevator_data;			\
	u64 __data = __VAR;						\
	__data = div_u64(__data, NSEC_PER_USEC);			\
	return cfq_var_show(__data, (page));				\
}
USEC_SHOW_FUNCTION(cfq_slice_idle_us_show, cfqd->cfq_slice_idle);
USEC_SHOW_FUNCTION(cfq_group_idle_us_show, cfqd->cfq_group_idle);
USEC_SHOW_FUNCTION(cfq_slice_sync_us_show, cfqd->cfq_slice[1]);
USEC_SHOW_FUNCTION(cfq_slice_async_us_show, cfqd->cfq_slice[0]);
USEC_SHOW_FUNCTION(cfq_target_latency_us_show, cfqd->cfq_target_latency);
#undef USEC_SHOW_FUNCTION

L
Linus Torvalds 已提交
4765
#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)			\
J
Jens Axboe 已提交
4766
static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)	\
L
Linus Torvalds 已提交
4767
{									\
4768
	struct cfq_data *cfqd = e->elevator_data;			\
L
Linus Torvalds 已提交
4769 4770 4771 4772 4773 4774 4775
	unsigned int __data;						\
	int ret = cfq_var_store(&__data, (page), count);		\
	if (__data < (MIN))						\
		__data = (MIN);						\
	else if (__data > (MAX))					\
		__data = (MAX);						\
	if (__CONV)							\
4776
		*(__PTR) = (u64)__data * NSEC_PER_MSEC;			\
L
Linus Torvalds 已提交
4777 4778 4779 4780 4781
	else								\
		*(__PTR) = __data;					\
	return ret;							\
}
STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
4782 4783 4784 4785
STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
		UINT_MAX, 1);
STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
		UINT_MAX, 1);
4786
STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
4787 4788
STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
		UINT_MAX, 0);
4789
STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
4790
STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1);
4791 4792
STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
4793 4794
STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
		UINT_MAX, 0);
4795
STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
4796
STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1);
L
Linus Torvalds 已提交
4797 4798
#undef STORE_FUNCTION

4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818
#define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX)			\
static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)	\
{									\
	struct cfq_data *cfqd = e->elevator_data;			\
	unsigned int __data;						\
	int ret = cfq_var_store(&__data, (page), count);		\
	if (__data < (MIN))						\
		__data = (MIN);						\
	else if (__data > (MAX))					\
		__data = (MAX);						\
	*(__PTR) = (u64)__data * NSEC_PER_USEC;				\
	return ret;							\
}
USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX);
USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX);
USEC_STORE_FUNCTION(cfq_slice_sync_us_store, &cfqd->cfq_slice[1], 1, UINT_MAX);
USEC_STORE_FUNCTION(cfq_slice_async_us_store, &cfqd->cfq_slice[0], 1, UINT_MAX);
USEC_STORE_FUNCTION(cfq_target_latency_us_store, &cfqd->cfq_target_latency, 1, UINT_MAX);
#undef USEC_STORE_FUNCTION

4819 4820 4821 4822 4823 4824 4825 4826 4827 4828
#define CFQ_ATTR(name) \
	__ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)

static struct elv_fs_entry cfq_attrs[] = {
	CFQ_ATTR(quantum),
	CFQ_ATTR(fifo_expire_sync),
	CFQ_ATTR(fifo_expire_async),
	CFQ_ATTR(back_seek_max),
	CFQ_ATTR(back_seek_penalty),
	CFQ_ATTR(slice_sync),
4829
	CFQ_ATTR(slice_sync_us),
4830
	CFQ_ATTR(slice_async),
4831
	CFQ_ATTR(slice_async_us),
4832 4833
	CFQ_ATTR(slice_async_rq),
	CFQ_ATTR(slice_idle),
4834
	CFQ_ATTR(slice_idle_us),
4835
	CFQ_ATTR(group_idle),
4836
	CFQ_ATTR(group_idle_us),
4837
	CFQ_ATTR(low_latency),
4838
	CFQ_ATTR(target_latency),
4839
	CFQ_ATTR(target_latency_us),
4840
	__ATTR_NULL
L
Linus Torvalds 已提交
4841 4842 4843
};

static struct elevator_type iosched_cfq = {
4844
	.ops.sq = {
L
Linus Torvalds 已提交
4845 4846 4847
		.elevator_merge_fn = 		cfq_merge,
		.elevator_merged_fn =		cfq_merged_request,
		.elevator_merge_req_fn =	cfq_merged_requests,
4848 4849
		.elevator_allow_bio_merge_fn =	cfq_allow_bio_merge,
		.elevator_allow_rq_merge_fn =	cfq_allow_rq_merge,
D
Divyesh Shah 已提交
4850
		.elevator_bio_merged_fn =	cfq_bio_merged,
4851
		.elevator_dispatch_fn =		cfq_dispatch_requests,
L
Linus Torvalds 已提交
4852
		.elevator_add_req_fn =		cfq_insert_request,
4853
		.elevator_activate_req_fn =	cfq_activate_request,
L
Linus Torvalds 已提交
4854 4855
		.elevator_deactivate_req_fn =	cfq_deactivate_request,
		.elevator_completed_req_fn =	cfq_completed_request,
4856 4857
		.elevator_former_req_fn =	elv_rb_former_request,
		.elevator_latter_req_fn =	elv_rb_latter_request,
4858
		.elevator_init_icq_fn =		cfq_init_icq,
4859
		.elevator_exit_icq_fn =		cfq_exit_icq,
L
Linus Torvalds 已提交
4860 4861 4862 4863 4864
		.elevator_set_req_fn =		cfq_set_request,
		.elevator_put_req_fn =		cfq_put_request,
		.elevator_may_queue_fn =	cfq_may_queue,
		.elevator_init_fn =		cfq_init_queue,
		.elevator_exit_fn =		cfq_exit_queue,
4865
		.elevator_registered_fn =	cfq_registered_queue,
L
Linus Torvalds 已提交
4866
	},
4867 4868
	.icq_size	=	sizeof(struct cfq_io_cq),
	.icq_align	=	__alignof__(struct cfq_io_cq),
4869
	.elevator_attrs =	cfq_attrs,
4870
	.elevator_name	=	"cfq",
L
Linus Torvalds 已提交
4871 4872 4873
	.elevator_owner =	THIS_MODULE,
};

4874
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
4875
static struct blkcg_policy blkcg_policy_cfq = {
4876
	.dfl_cftypes		= cfq_blkcg_files,
4877
	.legacy_cftypes		= cfq_blkcg_legacy_files,
4878

4879
	.cpd_alloc_fn		= cfq_cpd_alloc,
4880
	.cpd_init_fn		= cfq_cpd_init,
4881
	.cpd_free_fn		= cfq_cpd_free,
4882
	.cpd_bind_fn		= cfq_cpd_bind,
4883

4884
	.pd_alloc_fn		= cfq_pd_alloc,
4885
	.pd_init_fn		= cfq_pd_init,
4886
	.pd_offline_fn		= cfq_pd_offline,
4887
	.pd_free_fn		= cfq_pd_free,
4888
	.pd_reset_stats_fn	= cfq_pd_reset_stats,
4889 4890 4891
};
#endif

L
Linus Torvalds 已提交
4892 4893
static int __init cfq_init(void)
{
4894 4895
	int ret;

4896
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
4897
	ret = blkcg_policy_register(&blkcg_policy_cfq);
T
Tejun Heo 已提交
4898 4899
	if (ret)
		return ret;
4900 4901 4902
#else
	cfq_group_idle = 0;
#endif
T
Tejun Heo 已提交
4903

4904
	ret = -ENOMEM;
4905 4906
	cfq_pool = KMEM_CACHE(cfq_queue, 0);
	if (!cfq_pool)
T
Tejun Heo 已提交
4907
		goto err_pol_unreg;
L
Linus Torvalds 已提交
4908

4909
	ret = elv_register(&iosched_cfq);
T
Tejun Heo 已提交
4910 4911
	if (ret)
		goto err_free_pool;
4912

4913
	return 0;
T
Tejun Heo 已提交
4914 4915 4916 4917

err_free_pool:
	kmem_cache_destroy(cfq_pool);
err_pol_unreg:
4918
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
4919
	blkcg_policy_unregister(&blkcg_policy_cfq);
4920
#endif
T
Tejun Heo 已提交
4921
	return ret;
L
Linus Torvalds 已提交
4922 4923 4924 4925
}

static void __exit cfq_exit(void)
{
4926
#ifdef CONFIG_CFQ_GROUP_IOSCHED
T
Tejun Heo 已提交
4927
	blkcg_policy_unregister(&blkcg_policy_cfq);
4928
#endif
L
Linus Torvalds 已提交
4929
	elv_unregister(&iosched_cfq);
4930
	kmem_cache_destroy(cfq_pool);
L
Linus Torvalds 已提交
4931 4932 4933 4934 4935 4936 4937 4938
}

module_init(cfq_init);
module_exit(cfq_exit);

MODULE_AUTHOR("Jens Axboe");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");