cfq-iosched.c 99.4 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>
R
Randy Dunlap 已提交
13
#include <linux/jiffies.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 "blk.h"
18
#include "cfq.h"
L
Linus Torvalds 已提交
19 20 21 22

/*
 * tunables
 */
23
/* max queue in one round of service */
S
Shaohua Li 已提交
24
static const int cfq_quantum = 8;
25
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
26 27 28 29
/* maximum backwards seek, in KiB */
static const int cfq_back_max = 16 * 1024;
/* penalty of a backwards seek */
static const int cfq_back_penalty = 2;
30
static const int cfq_slice_sync = HZ / 10;
J
Jens Axboe 已提交
31
static int cfq_slice_async = HZ / 25;
32
static const int cfq_slice_async_rq = 2;
33
static int cfq_slice_idle = HZ / 125;
34
static int cfq_group_idle = HZ / 125;
35 36
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
37

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

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

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

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

57 58 59
#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
60

61
static struct kmem_cache *cfq_pool;
L
Linus Torvalds 已提交
62

63 64 65 66
#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)

67
#define sample_valid(samples)	((samples) > 80)
68
#define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
69

70 71 72 73 74 75 76 77
struct cfq_ttime {
	unsigned long last_end_request;

	unsigned long ttime_total;
	unsigned long ttime_samples;
	unsigned long ttime_mean;
};

78 79 80 81 82 83 84 85 86
/*
 * Most of our rbtree usage is for sorting with min extraction, so
 * if we cache the leftmost node we don't have to walk down the tree
 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
 * move this into the elevator for the rq sorting as well.
 */
struct cfq_rb_root {
	struct rb_root rb;
	struct rb_node *left;
87
	unsigned count;
88
	unsigned total_weight;
89
	u64 min_vdisktime;
90
	struct cfq_ttime ttime;
91
};
92 93
#define CFQ_RB_ROOT	(struct cfq_rb_root) { .rb = RB_ROOT, \
			.ttime = {.last_end_request = jiffies,},}
94

95 96 97 98 99
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
100
	int ref;
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123
	/* various state flags, see below */
	unsigned int flags;
	/* parent cfq_data */
	struct cfq_data *cfqd;
	/* service_tree member */
	struct rb_node rb_node;
	/* service_tree key */
	unsigned long rb_key;
	/* prio tree member */
	struct rb_node p_node;
	/* prio tree root we belong to, if any */
	struct rb_root *p_root;
	/* sorted list of pending requests */
	struct rb_root sort_list;
	/* if fifo isn't expired, next request to serve */
	struct request *next_rq;
	/* requests queued in sort_list */
	int queued[2];
	/* currently allocated requests */
	int allocated[2];
	/* fifo list of requests in sort_list */
	struct list_head fifo;

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

133 134
	/* pending priority requests */
	int prio_pending;
135 136 137 138 139
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;

	/* io prio of this group */
	unsigned short ioprio, org_ioprio;
140
	unsigned short ioprio_class;
141

142 143
	pid_t pid;

144
	u32 seek_history;
145 146
	sector_t last_request_pos;

147
	struct cfq_rb_root *service_tree;
J
Jeff Moyer 已提交
148
	struct cfq_queue *new_cfqq;
149
	struct cfq_group *cfqg;
150 151
	/* Number of sectors dispatched from queue in single dispatch round */
	unsigned long nr_sectors;
152 153
};

154
/*
155
 * First index in the service_trees.
156 157 158 159
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
160 161
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
162
	CFQ_PRIO_NR,
163 164
};

165 166 167 168 169 170 171 172 173
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

174 175
/* This is per cgroup per device grouping structure */
struct cfq_group {
176 177 178 179 180
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
181
	unsigned int weight;
182 183
	unsigned int new_weight;
	bool needs_update;
184 185 186 187

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

188
	/*
189
	 * Per group busy queues average. Useful for workload slice calc. We
190 191 192 193 194 195 196 197 198 199 200
	 * 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.
201 202 203 204
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
205 206 207 208

	unsigned long saved_workload_slice;
	enum wl_type_t saved_workload;
	enum wl_prio_t saved_serving_prio;
209 210 211
	struct blkio_group blkg;
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	struct hlist_node cfqd_node;
212
	int ref;
213
#endif
214 215
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
S
Shaohua Li 已提交
216
	struct cfq_ttime ttime;
217
};
218

219 220 221 222 223 224
struct cfq_io_cq {
	struct io_cq		icq;		/* must be the first member */
	struct cfq_queue	*cfqq[2];
	struct cfq_ttime	ttime;
};

225 226 227
/*
 * Per block device queue structure
 */
L
Linus Torvalds 已提交
228
struct cfq_data {
229
	struct request_queue *queue;
230 231
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
232
	struct cfq_group root_group;
233

234 235
	/*
	 * The priority currently being served
236
	 */
237
	enum wl_prio_t serving_prio;
238 239
	enum wl_type_t serving_type;
	unsigned long workload_expires;
240
	struct cfq_group *serving_group;
241 242 243 244 245 246 247 248

	/*
	 * 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];

249
	unsigned int busy_queues;
250
	unsigned int busy_sync_queues;
251

252 253
	int rq_in_driver;
	int rq_in_flight[2];
254 255 256 257 258

	/*
	 * queue-depth detection
	 */
	int rq_queued;
259
	int hw_tag;
260 261 262 263 264 265 266 267
	/*
	 * 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
268

269 270 271 272
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
273
	struct work_struct unplug_work;
L
Linus Torvalds 已提交
274

275
	struct cfq_queue *active_queue;
276
	struct cfq_io_cq *active_cic;
277

278 279 280 281 282
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
283

J
Jens Axboe 已提交
284
	sector_t last_position;
L
Linus Torvalds 已提交
285 286 287 288 289

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
290
	unsigned int cfq_fifo_expire[2];
L
Linus Torvalds 已提交
291 292
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
293 294 295
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
296
	unsigned int cfq_group_idle;
297
	unsigned int cfq_latency;
298

299 300 301 302
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
303

304
	unsigned long last_delayed_sync;
305 306 307

	/* List of cfq groups being managed on this device*/
	struct hlist_head cfqg_list;
308 309 310

	/* Number of groups which are on blkcg->blkg_list */
	unsigned int nr_blkcg_linked_grps;
L
Linus Torvalds 已提交
311 312
};

313 314
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);

315 316
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
317
					    enum wl_type_t type)
318
{
319 320 321
	if (!cfqg)
		return NULL;

322
	if (prio == IDLE_WORKLOAD)
323
		return &cfqg->service_tree_idle;
324

325
	return &cfqg->service_trees[prio][type];
326 327
}

J
Jens Axboe 已提交
328
enum cfqq_state_flags {
329 330
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
331
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
332 333 334 335
	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 */
336
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
337
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
338
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
339
	CFQ_CFQQ_FLAG_split_coop,	/* shared cfqq will be splitted */
340
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
341
	CFQ_CFQQ_FLAG_wait_busy,	/* Waiting for next request */
J
Jens Axboe 已提交
342 343 344 345 346
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
347
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
J
Jens Axboe 已提交
348 349 350
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
351
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
J
Jens Axboe 已提交
352 353 354
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
355
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
J
Jens Axboe 已提交
356 357 358 359
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
360
CFQ_CFQQ_FNS(must_dispatch);
J
Jens Axboe 已提交
361 362 363 364
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
365
CFQ_CFQQ_FNS(slice_new);
366
CFQ_CFQQ_FNS(sync);
367
CFQ_CFQQ_FNS(coop);
368
CFQ_CFQQ_FNS(split_coop);
369
CFQ_CFQQ_FNS(deep);
370
CFQ_CFQQ_FNS(wait_busy);
J
Jens Axboe 已提交
371 372
#undef CFQ_CFQQ_FNS

373
#ifdef CONFIG_CFQ_GROUP_IOSCHED
V
Vivek Goyal 已提交
374 375 376
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \
			cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
377
			blkg_path(&(cfqq)->cfqg->blkg), ##args)
V
Vivek Goyal 已提交
378 379 380

#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)				\
	blk_add_trace_msg((cfqd)->queue, "%s " fmt,			\
381
				blkg_path(&(cfqg)->blkg), ##args)       \
V
Vivek Goyal 已提交
382 383

#else
384 385
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
386
#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0)
V
Vivek Goyal 已提交
387
#endif
388 389 390
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

391 392 393 394 395 396 397 398 399 400
/* 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) \

401 402 403 404 405 406 407 408 409 410 411 412
static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd,
	struct cfq_ttime *ttime, bool group_idle)
{
	unsigned long slice;
	if (!sample_valid(ttime->ttime_samples))
		return false;
	if (group_idle)
		slice = cfqd->cfq_group_idle;
	else
		slice = cfqd->cfq_slice_idle;
	return ttime->ttime_mean > slice;
}
413

414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
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;
}

429 430 431 432 433 434 435 436 437
static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq)
{
	if (cfq_class_idle(cfqq))
		return IDLE_WORKLOAD;
	if (cfq_class_rt(cfqq))
		return RT_WORKLOAD;
	return BE_WORKLOAD;
}

438 439 440 441 442 443 444 445 446 447

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;
}

448 449 450
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
451 452
{
	if (wl == IDLE_WORKLOAD)
453
		return cfqg->service_tree_idle.count;
454

455 456 457
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
458 459
}

460 461 462 463 464 465 466
static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg)
{
	return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
}

467
static void cfq_dispatch_insert(struct request_queue *, struct request *);
468
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
469
				       struct io_context *, gfp_t);
470

471 472 473 474 475 476
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);
}

477 478 479 480 481 482 483 484
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;
}

485
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync)
486
{
487
	return cic->cfqq[is_sync];
488 489
}

490 491
static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq,
				bool is_sync)
492
{
493
	cic->cfqq[is_sync] = cfqq;
494 495
}

496
static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic)
497
{
498
	return cic->icq.q->elevator->elevator_data;
499 500
}

501 502 503 504
/*
 * We regard a request as SYNC, if it's either a read or has the SYNC bit
 * set (in which case it could also be direct WRITE).
 */
505
static inline bool cfq_bio_sync(struct bio *bio)
506
{
507
	return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC);
508
}
L
Linus Torvalds 已提交
509

A
Andrew Morton 已提交
510 511 512 513
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
514
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
A
Andrew Morton 已提交
515
{
516 517
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
518
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
519
	}
A
Andrew Morton 已提交
520 521
}

522 523 524 525 526
/*
 * 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.
 */
527
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
528
				 unsigned short prio)
529
{
530
	const int base_slice = cfqd->cfq_slice[sync];
531

532 533 534 535
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
536

537 538 539 540
static inline int
cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
541 542
}

543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575
static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg)
{
	u64 d = delta << CFQ_SERVICE_SHIFT;

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

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

	return min_vdisktime;
}

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

	return min_vdisktime;
}

static void update_min_vdisktime(struct cfq_rb_root *st)
{
	struct cfq_group *cfqg;

	if (st->left) {
		cfqg = rb_entry_cfqg(st->left);
576 577
		st->min_vdisktime = max_vdisktime(st->min_vdisktime,
						  cfqg->vdisktime);
578 579 580
	}
}

581 582 583 584 585 586
/*
 * 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
 */

587 588
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
589
{
590 591 592
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
593
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
594

595 596 597
	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) /
598
		cfq_hist_divisor;
599 600 601 602 603 604 605 606 607
	return cfqg->busy_queues_avg[rt];
}

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

	return cfq_target_latency * cfqg->weight / st->total_weight;
608 609
}

610
static inline unsigned
611
cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
612
{
613 614
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
615 616 617 618 619 620
		/*
		 * 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));
621 622
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
623 624 625
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
626 627 628 629 630 631 632
			unsigned base_low_slice = 2 * cfqd->cfq_slice_idle;
			/* scale low_slice according to IO priority
			 * and sync vs async */
			unsigned low_slice =
				min(slice, base_low_slice * slice / sync_slice);
			/* the adapted slice value is scaled to fit all iqs
			 * into the target latency */
633
			slice = max(slice * group_slice / expect_latency,
634 635 636
				    low_slice);
		}
	}
637 638 639 640 641 642
	return slice;
}

static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
643
	unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
644

645
	cfqq->slice_start = jiffies;
646
	cfqq->slice_end = jiffies + slice;
647
	cfqq->allocated_slice = slice;
648
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
649 650 651 652 653 654 655
}

/*
 * 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.
 */
656
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
657 658
{
	if (cfq_cfqq_slice_new(cfqq))
S
Shaohua Li 已提交
659
		return false;
660
	if (time_before(jiffies, cfqq->slice_end))
S
Shaohua Li 已提交
661
		return false;
662

S
Shaohua Li 已提交
663
	return true;
664 665
}

L
Linus Torvalds 已提交
666
/*
J
Jens Axboe 已提交
667
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
L
Linus Torvalds 已提交
668
 * We choose the request that is closest to the head right now. Distance
669
 * behind the head is penalized and only allowed to a certain extent.
L
Linus Torvalds 已提交
670
 */
J
Jens Axboe 已提交
671
static struct request *
672
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
L
Linus Torvalds 已提交
673
{
674
	sector_t s1, s2, d1 = 0, d2 = 0;
L
Linus Torvalds 已提交
675
	unsigned long back_max;
676 677 678
#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
679

J
Jens Axboe 已提交
680 681 682 683
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
J
Jens Axboe 已提交
684

685 686 687
	if (rq_is_sync(rq1) != rq_is_sync(rq2))
		return rq_is_sync(rq1) ? rq1 : rq2;

688 689
	if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
		return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
690

691 692
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
L
Linus Torvalds 已提交
693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708

	/*
	 * 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
709
		wrap |= CFQ_RQ1_WRAP;
L
Linus Torvalds 已提交
710 711 712 713 714 715

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
716
		wrap |= CFQ_RQ2_WRAP;
L
Linus Torvalds 已提交
717 718

	/* Found required data */
719 720 721 722 723 724

	/*
	 * 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
725
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
726
		if (d1 < d2)
J
Jens Axboe 已提交
727
			return rq1;
728
		else if (d2 < d1)
J
Jens Axboe 已提交
729
			return rq2;
730 731
		else {
			if (s1 >= s2)
J
Jens Axboe 已提交
732
				return rq1;
733
			else
J
Jens Axboe 已提交
734
				return rq2;
735
		}
L
Linus Torvalds 已提交
736

737
	case CFQ_RQ2_WRAP:
J
Jens Axboe 已提交
738
		return rq1;
739
	case CFQ_RQ1_WRAP:
J
Jens Axboe 已提交
740 741
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
742 743 744 745 746 747 748 749
	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
750
			return rq1;
L
Linus Torvalds 已提交
751
		else
J
Jens Axboe 已提交
752
			return rq2;
L
Linus Torvalds 已提交
753 754 755
	}
}

756 757 758
/*
 * The below is leftmost cache rbtree addon
 */
759
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
760
{
761 762 763 764
	/* Service tree is empty */
	if (!root->count)
		return NULL;

765 766 767
	if (!root->left)
		root->left = rb_first(&root->rb);

768 769 770 771
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
772 773
}

774 775 776 777 778 779 780 781 782 783 784
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;
}

785 786 787 788 789 790
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

791 792 793 794
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
795
	rb_erase_init(n, &root->rb);
796
	--root->count;
797 798
}

L
Linus Torvalds 已提交
799 800 801
/*
 * would be nice to take fifo expire time into account as well
 */
J
Jens Axboe 已提交
802 803 804
static struct request *
cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		  struct request *last)
L
Linus Torvalds 已提交
805
{
806 807
	struct rb_node *rbnext = rb_next(&last->rb_node);
	struct rb_node *rbprev = rb_prev(&last->rb_node);
J
Jens Axboe 已提交
808
	struct request *next = NULL, *prev = NULL;
L
Linus Torvalds 已提交
809

810
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
L
Linus Torvalds 已提交
811 812

	if (rbprev)
J
Jens Axboe 已提交
813
		prev = rb_entry_rq(rbprev);
L
Linus Torvalds 已提交
814

815
	if (rbnext)
J
Jens Axboe 已提交
816
		next = rb_entry_rq(rbnext);
817 818 819
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
J
Jens Axboe 已提交
820
			next = rb_entry_rq(rbnext);
821
	}
L
Linus Torvalds 已提交
822

823
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
L
Linus Torvalds 已提交
824 825
}

826 827
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
828
{
829 830 831
	/*
	 * just an approximation, should be ok.
	 */
832
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
833
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
834 835
}

836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
static inline s64
cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	return cfqg->vdisktime - st->min_vdisktime;
}

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

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

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

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

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

static void
871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
cfq_update_group_weight(struct cfq_group *cfqg)
{
	BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
	if (cfqg->needs_update) {
		cfqg->weight = cfqg->new_weight;
		cfqg->needs_update = false;
	}
}

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

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

static void
cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
892 893 894 895 896 897
{
	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
898
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
899 900 901 902 903
		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
904
	 * if group does not loose all if it was not continuously backlogged.
905 906 907 908 909 910 911
	 */
	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;
912 913
	cfq_group_service_tree_add(st, cfqg);
}
914

915 916 917 918 919 920
static void
cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	st->total_weight -= cfqg->weight;
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
921 922 923
}

static void
924
cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
925 926 927 928 929
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
930

931 932 933 934
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

V
Vivek Goyal 已提交
935
	cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
936
	cfq_group_service_tree_del(st, cfqg);
937
	cfqg->saved_workload_slice = 0;
938
	cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1);
939 940
}

941 942
static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
						unsigned int *unaccounted_time)
943
{
944
	unsigned int slice_used;
945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960

	/*
	 * Queue got expired before even a single request completed or
	 * got expired immediately after first request completion.
	 */
	if (!cfqq->slice_start || cfqq->slice_start == jiffies) {
		/*
		 * Also charge the seek time incurred to the group, otherwise
		 * if there are mutiple queues in the group, each can dispatch
		 * a single request on seeky media and cause lots of seek time
		 * and group will never know it.
		 */
		slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start),
					1);
	} else {
		slice_used = jiffies - cfqq->slice_start;
961 962
		if (slice_used > cfqq->allocated_slice) {
			*unaccounted_time = slice_used - cfqq->allocated_slice;
963
			slice_used = cfqq->allocated_slice;
964 965 966 967
		}
		if (time_after(cfqq->slice_start, cfqq->dispatch_start))
			*unaccounted_time += cfqq->slice_start -
					cfqq->dispatch_start;
968 969 970 971 972 973
	}

	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
974
				struct cfq_queue *cfqq)
975 976
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
977
	unsigned int used_sl, charge, unaccounted_sl = 0;
978 979 980 981
	int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
			- cfqg->service_tree_idle.count;

	BUG_ON(nr_sync < 0);
982
	used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
983

984 985 986 987
	if (iops_mode(cfqd))
		charge = cfqq->slice_dispatch;
	else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
		charge = cfqq->allocated_slice;
988 989

	/* Can't update vdisktime while group is on service tree */
990
	cfq_group_service_tree_del(st, cfqg);
991
	cfqg->vdisktime += cfq_scale_slice(charge, cfqg);
992 993
	/* If a new weight was requested, update now, off tree */
	cfq_group_service_tree_add(st, cfqg);
994 995 996 997 998 999 1000 1001 1002

	/* This group is being expired. Save the context */
	if (time_after(cfqd->workload_expires, jiffies)) {
		cfqg->saved_workload_slice = cfqd->workload_expires
						- jiffies;
		cfqg->saved_workload = cfqd->serving_type;
		cfqg->saved_serving_prio = cfqd->serving_prio;
	} else
		cfqg->saved_workload_slice = 0;
V
Vivek Goyal 已提交
1003 1004 1005

	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
					st->min_vdisktime);
1006 1007 1008 1009
	cfq_log_cfqq(cfqq->cfqd, cfqq,
		     "sl_used=%u disp=%u charge=%u iops=%u sect=%lu",
		     used_sl, cfqq->slice_dispatch, charge,
		     iops_mode(cfqd), cfqq->nr_sectors);
1010 1011
	cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl,
					  unaccounted_sl);
1012
	cfq_blkiocg_set_start_empty_time(&cfqg->blkg);
1013 1014
}

1015 1016 1017 1018 1019 1020 1021 1022
#ifdef CONFIG_CFQ_GROUP_IOSCHED
static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg)
{
	if (blkg)
		return container_of(blkg, struct cfq_group, blkg);
	return NULL;
}

P
Paul Bolle 已提交
1023 1024
static void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg,
					  unsigned int weight)
1025
{
1026 1027 1028
	struct cfq_group *cfqg = cfqg_of_blkg(blkg);
	cfqg->new_weight = weight;
	cfqg->needs_update = true;
1029 1030
}

1031 1032
static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd,
			struct cfq_group *cfqg, struct blkio_cgroup *blkcg)
1033
{
1034 1035
	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
	unsigned int major, minor;
1036

1037 1038 1039 1040 1041 1042 1043
	/*
	 * Add group onto cgroup list. It might happen that bdi->dev is
	 * not initialized yet. Initialize this new group without major
	 * and minor info and this info will be filled in once a new thread
	 * comes for IO.
	 */
	if (bdi->dev) {
1044
		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
					(void *)cfqd, MKDEV(major, minor));
	} else
		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
					(void *)cfqd, 0);

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

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

/*
 * Should be called from sleepable context. No request queue lock as per
 * cpu stats are allocated dynamically and alloc_percpu needs to be called
 * from sleepable context.
 */
static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd)
{
	struct cfq_group *cfqg = NULL;
1066
	int i, j, ret;
1067
	struct cfq_rb_root *st;
1068 1069 1070

	cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node);
	if (!cfqg)
1071
		return NULL;
1072 1073 1074 1075 1076

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

S
Shaohua Li 已提交
1077 1078
	cfqg->ttime.last_end_request = jiffies;

1079 1080 1081 1082 1083 1084
	/*
	 * Take the initial reference that will be released on destroy
	 * This can be thought of a joint reference by cgroup and
	 * elevator which will be dropped by either elevator exit
	 * or cgroup deletion path depending on who is exiting first.
	 */
1085
	cfqg->ref = 1;
1086 1087 1088 1089 1090 1091 1092

	ret = blkio_alloc_blkg_stats(&cfqg->blkg);
	if (ret) {
		kfree(cfqg);
		return NULL;
	}

1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
	return cfqg;
}

static struct cfq_group *
cfq_find_cfqg(struct cfq_data *cfqd, struct blkio_cgroup *blkcg)
{
	struct cfq_group *cfqg = NULL;
	void *key = cfqd;
	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
	unsigned int major, minor;
1103

1104
	/*
1105 1106
	 * This is the common case when there are no blkio cgroups.
	 * Avoid lookup in this case
1107
	 */
1108 1109 1110 1111
	if (blkcg == &blkio_root_cgroup)
		cfqg = &cfqd->root_group;
	else
		cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
1112

1113 1114 1115 1116
	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
		cfqg->blkg.dev = MKDEV(major, minor);
	}
1117 1118 1119 1120 1121

	return cfqg;
}

/*
1122 1123
 * Search for the cfq group current task belongs to. request_queue lock must
 * be held.
1124
 */
1125
static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd)
1126
{
1127
	struct blkio_cgroup *blkcg;
1128 1129
	struct cfq_group *cfqg = NULL, *__cfqg = NULL;
	struct request_queue *q = cfqd->queue;
1130 1131

	rcu_read_lock();
1132
	blkcg = task_blkio_cgroup(current);
1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170
	cfqg = cfq_find_cfqg(cfqd, blkcg);
	if (cfqg) {
		rcu_read_unlock();
		return cfqg;
	}

	/*
	 * Need to allocate a group. Allocation of group also needs allocation
	 * of per cpu stats which in-turn takes a mutex() and can block. Hence
	 * we need to drop rcu lock and queue_lock before we call alloc.
	 *
	 * Not taking any queue reference here and assuming that queue is
	 * around by the time we return. CFQ queue allocation code does
	 * the same. It might be racy though.
	 */

	rcu_read_unlock();
	spin_unlock_irq(q->queue_lock);

	cfqg = cfq_alloc_cfqg(cfqd);

	spin_lock_irq(q->queue_lock);

	rcu_read_lock();
	blkcg = task_blkio_cgroup(current);

	/*
	 * If some other thread already allocated the group while we were
	 * not holding queue lock, free up the group
	 */
	__cfqg = cfq_find_cfqg(cfqd, blkcg);

	if (__cfqg) {
		kfree(cfqg);
		rcu_read_unlock();
		return __cfqg;
	}

1171
	if (!cfqg)
1172
		cfqg = &cfqd->root_group;
1173 1174

	cfq_init_add_cfqg_lists(cfqd, cfqg, blkcg);
1175 1176 1177 1178
	rcu_read_unlock();
	return cfqg;
}

1179 1180
static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
{
1181
	cfqg->ref++;
1182 1183 1184
	return cfqg;
}

1185 1186 1187 1188 1189 1190 1191
static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg)
{
	/* Currently, all async queues are mapped to root group */
	if (!cfq_cfqq_sync(cfqq))
		cfqg = &cfqq->cfqd->root_group;

	cfqq->cfqg = cfqg;
1192
	/* cfqq reference on cfqg */
1193
	cfqq->cfqg->ref++;
1194 1195 1196 1197 1198 1199 1200
}

static void cfq_put_cfqg(struct cfq_group *cfqg)
{
	struct cfq_rb_root *st;
	int i, j;

1201 1202 1203
	BUG_ON(cfqg->ref <= 0);
	cfqg->ref--;
	if (cfqg->ref)
1204 1205
		return;
	for_each_cfqg_st(cfqg, i, j, st)
G
Gui Jianfeng 已提交
1206
		BUG_ON(!RB_EMPTY_ROOT(&st->rb));
1207
	free_percpu(cfqg->blkg.stats_cpu);
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217
	kfree(cfqg);
}

static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
	/* Something wrong if we are trying to remove same group twice */
	BUG_ON(hlist_unhashed(&cfqg->cfqd_node));

	hlist_del_init(&cfqg->cfqd_node);

1218 1219 1220
	BUG_ON(cfqd->nr_blkcg_linked_grps <= 0);
	cfqd->nr_blkcg_linked_grps--;

1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
	/*
	 * Put the reference taken at the time of creation so that when all
	 * queues are gone, group can be destroyed.
	 */
	cfq_put_cfqg(cfqg);
}

static void cfq_release_cfq_groups(struct cfq_data *cfqd)
{
	struct hlist_node *pos, *n;
	struct cfq_group *cfqg;

	hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) {
		/*
		 * If cgroup removal path got to blk_group first and removed
		 * it from cgroup list, then it will take care of destroying
		 * cfqg also.
		 */
1239
		if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg))
1240 1241
			cfq_destroy_cfqg(cfqd, cfqg);
	}
1242
}
1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257

/*
 * Blk cgroup controller notification saying that blkio_group object is being
 * delinked as associated cgroup object is going away. That also means that
 * no new IO will come in this group. So get rid of this group as soon as
 * any pending IO in the group is finished.
 *
 * This function is called under rcu_read_lock(). key is the rcu protected
 * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu
 * read lock.
 *
 * "key" was fetched from blkio_group under blkio_cgroup->lock. That means
 * it should not be NULL as even if elevator was exiting, cgroup deltion
 * path got to it first.
 */
P
Paul Bolle 已提交
1258
static void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg)
1259 1260 1261 1262 1263 1264 1265 1266 1267
{
	unsigned long  flags;
	struct cfq_data *cfqd = key;

	spin_lock_irqsave(cfqd->queue->queue_lock, flags);
	cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg));
	spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
}

1268
#else /* GROUP_IOSCHED */
1269
static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd)
1270 1271 1272
{
	return &cfqd->root_group;
}
1273 1274 1275

static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
{
1276
	return cfqg;
1277 1278
}

1279 1280 1281 1282 1283
static inline void
cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
	cfqq->cfqg = cfqg;
}

1284 1285 1286
static void cfq_release_cfq_groups(struct cfq_data *cfqd) {}
static inline void cfq_put_cfqg(struct cfq_group *cfqg) {}

1287 1288
#endif /* GROUP_IOSCHED */

1289
/*
1290
 * The cfqd->service_trees holds all pending cfq_queue's that have
1291 1292 1293
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
1294
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1295
				 bool add_front)
1296
{
1297 1298
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
1299
	unsigned long rb_key;
1300
	struct cfq_rb_root *service_tree;
1301
	int left;
1302
	int new_cfqq = 1;
1303

1304
	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
1305
						cfqq_type(cfqq));
1306 1307
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
1308
		parent = rb_last(&service_tree->rb);
1309 1310 1311 1312 1313 1314
		if (parent && parent != &cfqq->rb_node) {
			__cfqq = rb_entry(parent, struct cfq_queue, rb_node);
			rb_key += __cfqq->rb_key;
		} else
			rb_key += jiffies;
	} else if (!add_front) {
1315 1316 1317 1318 1319 1320
		/*
		 * 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.
		 */
1321
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
1322
		rb_key -= cfqq->slice_resid;
1323
		cfqq->slice_resid = 0;
1324 1325
	} else {
		rb_key = -HZ;
1326
		__cfqq = cfq_rb_first(service_tree);
1327 1328
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
L
Linus Torvalds 已提交
1329

1330
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
1331
		new_cfqq = 0;
1332
		/*
1333
		 * same position, nothing more to do
1334
		 */
1335 1336
		if (rb_key == cfqq->rb_key &&
		    cfqq->service_tree == service_tree)
1337
			return;
L
Linus Torvalds 已提交
1338

1339 1340
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
L
Linus Torvalds 已提交
1341
	}
1342

1343
	left = 1;
1344
	parent = NULL;
1345 1346
	cfqq->service_tree = service_tree;
	p = &service_tree->rb.rb_node;
1347
	while (*p) {
1348
		struct rb_node **n;
1349

1350 1351 1352
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

1353
		/*
1354
		 * sort by key, that represents service time.
1355
		 */
1356
		if (time_before(rb_key, __cfqq->rb_key))
1357
			n = &(*p)->rb_left;
1358
		else {
1359
			n = &(*p)->rb_right;
1360
			left = 0;
1361
		}
1362 1363

		p = n;
1364 1365
	}

1366
	if (left)
1367
		service_tree->left = &cfqq->rb_node;
1368

1369 1370
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
1371 1372
	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
	service_tree->count++;
1373
	if (add_front || !new_cfqq)
1374
		return;
1375
	cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
L
Linus Torvalds 已提交
1376 1377
}

1378
static struct cfq_queue *
1379 1380 1381
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)
1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397
{
	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.
		 */
1398
		if (sector > blk_rq_pos(cfqq->next_rq))
1399
			n = &(*p)->rb_right;
1400
		else if (sector < blk_rq_pos(cfqq->next_rq))
1401 1402 1403 1404
			n = &(*p)->rb_left;
		else
			break;
		p = n;
1405
		cfqq = NULL;
1406 1407 1408 1409 1410
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
1411
	return cfqq;
1412 1413 1414 1415 1416 1417 1418
}

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

1419 1420 1421 1422
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
1423 1424 1425 1426 1427 1428

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

1429
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
1430 1431
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
1432 1433
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
1434 1435 1436
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
1437 1438
}

1439 1440 1441
/*
 * Update cfqq's position in the service tree.
 */
1442
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
J
Jens Axboe 已提交
1443 1444 1445 1446
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
1447
	if (cfq_cfqq_on_rr(cfqq)) {
1448
		cfq_service_tree_add(cfqd, cfqq, 0);
1449 1450
		cfq_prio_tree_add(cfqd, cfqq);
	}
J
Jens Axboe 已提交
1451 1452
}

L
Linus Torvalds 已提交
1453 1454
/*
 * add to busy list of queues for service, trying to be fair in ordering
1455
 * the pending list according to last request service
L
Linus Torvalds 已提交
1456
 */
J
Jens Axboe 已提交
1457
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
1458
{
1459
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
J
Jens Axboe 已提交
1460 1461
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
L
Linus Torvalds 已提交
1462
	cfqd->busy_queues++;
1463 1464
	if (cfq_cfqq_sync(cfqq))
		cfqd->busy_sync_queues++;
L
Linus Torvalds 已提交
1465

1466
	cfq_resort_rr_list(cfqd, cfqq);
L
Linus Torvalds 已提交
1467 1468
}

1469 1470 1471 1472
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
J
Jens Axboe 已提交
1473
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
1474
{
1475
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
J
Jens Axboe 已提交
1476 1477
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
L
Linus Torvalds 已提交
1478

1479 1480 1481 1482
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
	}
1483 1484 1485 1486
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
1487

1488
	cfq_group_notify_queue_del(cfqd, cfqq->cfqg);
L
Linus Torvalds 已提交
1489 1490
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
1491 1492
	if (cfq_cfqq_sync(cfqq))
		cfqd->busy_sync_queues--;
L
Linus Torvalds 已提交
1493 1494 1495 1496 1497
}

/*
 * rb tree support functions
 */
J
Jens Axboe 已提交
1498
static void cfq_del_rq_rb(struct request *rq)
L
Linus Torvalds 已提交
1499
{
J
Jens Axboe 已提交
1500 1501
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
	const int sync = rq_is_sync(rq);
L
Linus Torvalds 已提交
1502

1503 1504
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
L
Linus Torvalds 已提交
1505

J
Jens Axboe 已提交
1506
	elv_rb_del(&cfqq->sort_list, rq);
L
Linus Torvalds 已提交
1507

1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518
	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
1519 1520
}

J
Jens Axboe 已提交
1521
static void cfq_add_rq_rb(struct request *rq)
L
Linus Torvalds 已提交
1522
{
J
Jens Axboe 已提交
1523
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
L
Linus Torvalds 已提交
1524
	struct cfq_data *cfqd = cfqq->cfqd;
1525
	struct request *prev;
L
Linus Torvalds 已提交
1526

1527
	cfqq->queued[rq_is_sync(rq)]++;
L
Linus Torvalds 已提交
1528

1529
	elv_rb_add(&cfqq->sort_list, rq);
1530 1531 1532

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
1533 1534 1535 1536

	/*
	 * check if this request is a better next-serve candidate
	 */
1537
	prev = cfqq->next_rq;
1538
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
1539 1540 1541 1542 1543 1544 1545

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

1546
	BUG_ON(!cfqq->next_rq);
L
Linus Torvalds 已提交
1547 1548
}

J
Jens Axboe 已提交
1549
static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
L
Linus Torvalds 已提交
1550
{
1551 1552
	elv_rb_del(&cfqq->sort_list, rq);
	cfqq->queued[rq_is_sync(rq)]--;
1553 1554
	cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg,
					rq_data_dir(rq), rq_is_sync(rq));
J
Jens Axboe 已提交
1555
	cfq_add_rq_rb(rq);
1556
	cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg,
1557 1558
			&cfqq->cfqd->serving_group->blkg, rq_data_dir(rq),
			rq_is_sync(rq));
L
Linus Torvalds 已提交
1559 1560
}

1561 1562
static struct request *
cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
L
Linus Torvalds 已提交
1563
{
1564
	struct task_struct *tsk = current;
1565
	struct cfq_io_cq *cic;
1566
	struct cfq_queue *cfqq;
L
Linus Torvalds 已提交
1567

1568
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
1569 1570 1571 1572
	if (!cic)
		return NULL;

	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
1573 1574 1575
	if (cfqq) {
		sector_t sector = bio->bi_sector + bio_sectors(bio);

1576
		return elv_rb_find(&cfqq->sort_list, sector);
1577
	}
L
Linus Torvalds 已提交
1578 1579 1580 1581

	return NULL;
}

1582
static void cfq_activate_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
1583
{
1584
	struct cfq_data *cfqd = q->elevator->elevator_data;
J
Jens Axboe 已提交
1585

1586
	cfqd->rq_in_driver++;
1587
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
1588
						cfqd->rq_in_driver);
1589

1590
	cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
L
Linus Torvalds 已提交
1591 1592
}

1593
static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
1594
{
1595 1596
	struct cfq_data *cfqd = q->elevator->elevator_data;

1597 1598
	WARN_ON(!cfqd->rq_in_driver);
	cfqd->rq_in_driver--;
1599
	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
1600
						cfqd->rq_in_driver);
L
Linus Torvalds 已提交
1601 1602
}

1603
static void cfq_remove_request(struct request *rq)
L
Linus Torvalds 已提交
1604
{
J
Jens Axboe 已提交
1605
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
1606

J
Jens Axboe 已提交
1607 1608
	if (cfqq->next_rq == rq)
		cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
L
Linus Torvalds 已提交
1609

1610
	list_del_init(&rq->queuelist);
J
Jens Axboe 已提交
1611
	cfq_del_rq_rb(rq);
1612

1613
	cfqq->cfqd->rq_queued--;
1614 1615
	cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg,
					rq_data_dir(rq), rq_is_sync(rq));
1616 1617 1618
	if (rq->cmd_flags & REQ_PRIO) {
		WARN_ON(!cfqq->prio_pending);
		cfqq->prio_pending--;
1619
	}
L
Linus Torvalds 已提交
1620 1621
}

1622 1623
static int cfq_merge(struct request_queue *q, struct request **req,
		     struct bio *bio)
L
Linus Torvalds 已提交
1624 1625 1626 1627
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request *__rq;

1628
	__rq = cfq_find_rq_fmerge(cfqd, bio);
1629
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
1630 1631
		*req = __rq;
		return ELEVATOR_FRONT_MERGE;
L
Linus Torvalds 已提交
1632 1633 1634 1635 1636
	}

	return ELEVATOR_NO_MERGE;
}

1637
static void cfq_merged_request(struct request_queue *q, struct request *req,
1638
			       int type)
L
Linus Torvalds 已提交
1639
{
1640
	if (type == ELEVATOR_FRONT_MERGE) {
J
Jens Axboe 已提交
1641
		struct cfq_queue *cfqq = RQ_CFQQ(req);
L
Linus Torvalds 已提交
1642

J
Jens Axboe 已提交
1643
		cfq_reposition_rq_rb(cfqq, req);
L
Linus Torvalds 已提交
1644 1645 1646
	}
}

D
Divyesh Shah 已提交
1647 1648 1649
static void cfq_bio_merged(struct request_queue *q, struct request *req,
				struct bio *bio)
{
1650 1651
	cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg,
					bio_data_dir(bio), cfq_bio_sync(bio));
D
Divyesh Shah 已提交
1652 1653
}

L
Linus Torvalds 已提交
1654
static void
1655
cfq_merged_requests(struct request_queue *q, struct request *rq,
L
Linus Torvalds 已提交
1656 1657
		    struct request *next)
{
1658
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
1659 1660
	struct cfq_data *cfqd = q->elevator->elevator_data;

1661 1662 1663 1664
	/*
	 * reposition in fifo if next is older than rq
	 */
	if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
1665
	    time_before(rq_fifo_time(next), rq_fifo_time(rq))) {
1666
		list_move(&rq->queuelist, &next->queuelist);
1667 1668
		rq_set_fifo_time(rq, rq_fifo_time(next));
	}
1669

1670 1671
	if (cfqq->next_rq == next)
		cfqq->next_rq = rq;
1672
	cfq_remove_request(next);
1673 1674
	cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg,
					rq_data_dir(next), rq_is_sync(next));
1675 1676 1677 1678 1679 1680 1681 1682 1683 1684

	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);
1685 1686
}

1687
static int cfq_allow_merge(struct request_queue *q, struct request *rq,
1688 1689 1690
			   struct bio *bio)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
1691
	struct cfq_io_cq *cic;
1692 1693 1694
	struct cfq_queue *cfqq;

	/*
1695
	 * Disallow merge of a sync bio into an async request.
1696
	 */
1697
	if (cfq_bio_sync(bio) && !rq_is_sync(rq))
1698
		return false;
1699 1700

	/*
T
Tejun Heo 已提交
1701
	 * Lookup the cfqq that this bio will be queued with and allow
1702
	 * merge only if rq is queued there.
T
Tejun Heo 已提交
1703
	 */
1704 1705 1706
	cic = cfq_cic_lookup(cfqd, current->io_context);
	if (!cic)
		return false;
1707

1708
	cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
1709
	return cfqq == RQ_CFQQ(rq);
1710 1711
}

1712 1713 1714
static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	del_timer(&cfqd->idle_slice_timer);
1715
	cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg);
1716 1717
}

J
Jens Axboe 已提交
1718 1719
static void __cfq_set_active_queue(struct cfq_data *cfqd,
				   struct cfq_queue *cfqq)
1720 1721
{
	if (cfqq) {
1722 1723
		cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d",
				cfqd->serving_prio, cfqd->serving_type);
1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738
		cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg);
		cfqq->slice_start = 0;
		cfqq->dispatch_start = jiffies;
		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);
1739 1740 1741 1742 1743
	}

	cfqd->active_queue = cfqq;
}

1744 1745 1746 1747 1748
/*
 * 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,
1749
		    bool timed_out)
1750
{
1751 1752
	cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);

1753
	if (cfq_cfqq_wait_request(cfqq))
1754
		cfq_del_timer(cfqd, cfqq);
1755 1756

	cfq_clear_cfqq_wait_request(cfqq);
1757
	cfq_clear_cfqq_wait_busy(cfqq);
1758

1759 1760 1761 1762 1763 1764 1765 1766 1767
	/*
	 * 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);

1768
	/*
1769
	 * store what was left of this slice, if the queue idled/timed out
1770
	 */
1771 1772
	if (timed_out) {
		if (cfq_cfqq_slice_new(cfqq))
1773
			cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq);
1774 1775
		else
			cfqq->slice_resid = cfqq->slice_end - jiffies;
1776 1777
		cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
	}
1778

1779
	cfq_group_served(cfqd, cfqq->cfqg, cfqq);
1780

1781 1782 1783
	if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
		cfq_del_cfqq_rr(cfqd, cfqq);

1784
	cfq_resort_rr_list(cfqd, cfqq);
1785 1786 1787 1788 1789

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

	if (cfqd->active_cic) {
1790
		put_io_context(cfqd->active_cic->icq.ioc);
1791 1792 1793 1794
		cfqd->active_cic = NULL;
	}
}

1795
static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
1796 1797 1798 1799
{
	struct cfq_queue *cfqq = cfqd->active_queue;

	if (cfqq)
1800
		__cfq_slice_expired(cfqd, cfqq, timed_out);
1801 1802
}

1803 1804 1805 1806
/*
 * 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
1807
static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
1808
{
1809
	struct cfq_rb_root *service_tree =
1810
		service_tree_for(cfqd->serving_group, cfqd->serving_prio,
1811
					cfqd->serving_type);
1812

1813 1814 1815
	if (!cfqd->rq_queued)
		return NULL;

1816 1817 1818
	/* There is nothing to dispatch */
	if (!service_tree)
		return NULL;
1819 1820 1821
	if (RB_EMPTY_ROOT(&service_tree->rb))
		return NULL;
	return cfq_rb_first(service_tree);
J
Jens Axboe 已提交
1822 1823
}

1824 1825
static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd)
{
1826
	struct cfq_group *cfqg;
1827 1828 1829 1830 1831 1832 1833
	struct cfq_queue *cfqq;
	int i, j;
	struct cfq_rb_root *st;

	if (!cfqd->rq_queued)
		return NULL;

1834 1835 1836 1837
	cfqg = cfq_get_next_cfqg(cfqd);
	if (!cfqg)
		return NULL;

1838 1839 1840 1841 1842 1843
	for_each_cfqg_st(cfqg, i, j, st)
		if ((cfqq = cfq_rb_first(st)) != NULL)
			return cfqq;
	return NULL;
}

1844 1845 1846
/*
 * Get and set a new active queue for service.
 */
1847 1848
static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
					      struct cfq_queue *cfqq)
J
Jens Axboe 已提交
1849
{
1850
	if (!cfqq)
1851
		cfqq = cfq_get_next_queue(cfqd);
J
Jens Axboe 已提交
1852

1853
	__cfq_set_active_queue(cfqd, cfqq);
J
Jens Axboe 已提交
1854
	return cfqq;
1855 1856
}

1857 1858 1859
static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
					  struct request *rq)
{
1860 1861
	if (blk_rq_pos(rq) >= cfqd->last_position)
		return blk_rq_pos(rq) - cfqd->last_position;
1862
	else
1863
		return cfqd->last_position - blk_rq_pos(rq);
1864 1865
}

1866
static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1867
			       struct request *rq)
J
Jens Axboe 已提交
1868
{
1869
	return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR;
J
Jens Axboe 已提交
1870 1871
}

1872 1873 1874
static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
				    struct cfq_queue *cur_cfqq)
{
1875
	struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886
	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.
	 */
1887
	__cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
1888 1889 1890 1891 1892 1893 1894 1895
	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);
1896
	if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
1897 1898
		return __cfqq;

1899
	if (blk_rq_pos(__cfqq->next_rq) < sector)
1900 1901 1902 1903 1904 1905 1906
		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);
1907
	if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923
		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,
1924
					      struct cfq_queue *cur_cfqq)
J
Jens Axboe 已提交
1925
{
1926 1927
	struct cfq_queue *cfqq;

1928 1929
	if (cfq_class_idle(cur_cfqq))
		return NULL;
1930 1931 1932 1933 1934
	if (!cfq_cfqq_sync(cur_cfqq))
		return NULL;
	if (CFQQ_SEEKY(cur_cfqq))
		return NULL;

1935 1936 1937 1938 1939 1940
	/*
	 * 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
1941
	/*
1942 1943 1944
	 * 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
1945
	 */
1946 1947 1948 1949
	cfqq = cfqq_close(cfqd, cur_cfqq);
	if (!cfqq)
		return NULL;

1950 1951 1952 1953
	/* 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
1954 1955 1956 1957 1958
	/*
	 * It only makes sense to merge sync queues.
	 */
	if (!cfq_cfqq_sync(cfqq))
		return NULL;
1959 1960
	if (CFQQ_SEEKY(cfqq))
		return NULL;
J
Jeff Moyer 已提交
1961

1962 1963 1964 1965 1966 1967
	/*
	 * Do not merge queues of different priority classes
	 */
	if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
		return NULL;

1968
	return cfqq;
J
Jens Axboe 已提交
1969 1970
}

1971 1972 1973 1974 1975 1976 1977
/*
 * Determine whether we should enforce idle window for this queue.
 */

static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	enum wl_prio_t prio = cfqq_prio(cfqq);
1978
	struct cfq_rb_root *service_tree = cfqq->service_tree;
1979

1980 1981 1982
	BUG_ON(!service_tree);
	BUG_ON(!service_tree->count);

1983 1984 1985
	if (!cfqd->cfq_slice_idle)
		return false;

1986 1987 1988 1989 1990
	/* We never do for idle class queues. */
	if (prio == IDLE_WORKLOAD)
		return false;

	/* We do for queues that were marked with idle window flag. */
1991 1992
	if (cfq_cfqq_idle_window(cfqq) &&
	   !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag))
1993 1994 1995 1996 1997 1998
		return true;

	/*
	 * Otherwise, we do only if they are the last ones
	 * in their service tree.
	 */
1999 2000
	if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) &&
	   !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false))
S
Shaohua Li 已提交
2001
		return true;
2002 2003
	cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d",
			service_tree->count);
S
Shaohua Li 已提交
2004
	return false;
2005 2006
}

J
Jens Axboe 已提交
2007
static void cfq_arm_slice_timer(struct cfq_data *cfqd)
2008
{
2009
	struct cfq_queue *cfqq = cfqd->active_queue;
2010
	struct cfq_io_cq *cic;
2011
	unsigned long sl, group_idle = 0;
2012

2013
	/*
J
Jens Axboe 已提交
2014 2015 2016
	 * 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.
2017
	 */
J
Jens Axboe 已提交
2018
	if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
2019 2020
		return;

2021
	WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
J
Jens Axboe 已提交
2022
	WARN_ON(cfq_cfqq_slice_new(cfqq));
2023 2024 2025 2026

	/*
	 * idle is disabled, either manually or by past process history
	 */
2027 2028 2029 2030 2031 2032 2033
	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
2034

2035
	/*
2036
	 * still active requests from this queue, don't idle
2037
	 */
2038
	if (cfqq->dispatched)
2039 2040
		return;

2041 2042 2043
	/*
	 * task has exited, don't wait
	 */
2044
	cic = cfqd->active_cic;
2045
	if (!cic || !atomic_read(&cic->icq.ioc->nr_tasks))
J
Jens Axboe 已提交
2046 2047
		return;

2048 2049 2050 2051 2052
	/*
	 * If our average think time is larger than the remaining time
	 * slice, then don't idle. This avoids overrunning the allotted
	 * time slice.
	 */
2053 2054
	if (sample_valid(cic->ttime.ttime_samples) &&
	    (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) {
2055
		cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu",
2056
			     cic->ttime.ttime_mean);
2057
		return;
2058
	}
2059

2060 2061 2062 2063
	/* There are other queues in the group, don't do group idle */
	if (group_idle && cfqq->cfqg->nr_cfqq > 1)
		return;

J
Jens Axboe 已提交
2064
	cfq_mark_cfqq_wait_request(cfqq);
2065

2066 2067 2068 2069
	if (group_idle)
		sl = cfqd->cfq_group_idle;
	else
		sl = cfqd->cfq_slice_idle;
2070

2071
	mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
2072
	cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg);
2073 2074
	cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl,
			group_idle ? 1 : 0);
L
Linus Torvalds 已提交
2075 2076
}

2077 2078 2079
/*
 * Move request from internal lists to the request queue dispatch list.
 */
2080
static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
2081
{
2082
	struct cfq_data *cfqd = q->elevator->elevator_data;
J
Jens Axboe 已提交
2083
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
2084

2085 2086
	cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");

2087
	cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
2088
	cfq_remove_request(rq);
J
Jens Axboe 已提交
2089
	cfqq->dispatched++;
2090
	(RQ_CFQG(rq))->dispatched++;
2091
	elv_dispatch_sort(q, rq);
2092

2093
	cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++;
2094
	cfqq->nr_sectors += blk_rq_sectors(rq);
2095
	cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq),
2096
					rq_data_dir(rq), rq_is_sync(rq));
L
Linus Torvalds 已提交
2097 2098 2099 2100 2101
}

/*
 * return expired entry, or NULL to just start from scratch in rbtree
 */
J
Jens Axboe 已提交
2102
static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
2103
{
2104
	struct request *rq = NULL;
L
Linus Torvalds 已提交
2105

J
Jens Axboe 已提交
2106
	if (cfq_cfqq_fifo_expire(cfqq))
L
Linus Torvalds 已提交
2107
		return NULL;
2108 2109 2110

	cfq_mark_cfqq_fifo_expire(cfqq);

2111 2112
	if (list_empty(&cfqq->fifo))
		return NULL;
L
Linus Torvalds 已提交
2113

2114
	rq = rq_entry_fifo(cfqq->fifo.next);
2115
	if (time_before(jiffies, rq_fifo_time(rq)))
2116
		rq = NULL;
L
Linus Torvalds 已提交
2117

2118
	cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq);
J
Jens Axboe 已提交
2119
	return rq;
L
Linus Torvalds 已提交
2120 2121
}

2122 2123 2124 2125
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
2126

2127
	WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
L
Linus Torvalds 已提交
2128

2129
	return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
L
Linus Torvalds 已提交
2130 2131
}

J
Jeff Moyer 已提交
2132 2133 2134 2135 2136 2137 2138 2139
/*
 * 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];
2140
	process_refs = cfqq->ref - io_refs;
J
Jeff Moyer 已提交
2141 2142 2143 2144 2145 2146
	BUG_ON(process_refs < 0);
	return process_refs;
}

static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
{
2147
	int process_refs, new_process_refs;
J
Jeff Moyer 已提交
2148 2149
	struct cfq_queue *__cfqq;

2150 2151 2152 2153 2154 2155 2156 2157 2158
	/*
	 * 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
2159 2160 2161 2162 2163 2164 2165 2166
	/* 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);
2167
	new_process_refs = cfqq_process_refs(new_cfqq);
J
Jeff Moyer 已提交
2168 2169 2170 2171
	/*
	 * If the process for the cfqq has gone away, there is no
	 * sense in merging the queues.
	 */
2172
	if (process_refs == 0 || new_process_refs == 0)
J
Jeff Moyer 已提交
2173 2174
		return;

2175 2176 2177 2178 2179
	/*
	 * Merge in the direction of the lesser amount of work.
	 */
	if (new_process_refs >= process_refs) {
		cfqq->new_cfqq = new_cfqq;
2180
		new_cfqq->ref += process_refs;
2181 2182
	} else {
		new_cfqq->new_cfqq = cfqq;
2183
		cfqq->ref += new_process_refs;
2184
	}
J
Jeff Moyer 已提交
2185 2186
}

2187
static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd,
2188
				struct cfq_group *cfqg, enum wl_prio_t prio)
2189 2190 2191 2192 2193 2194 2195
{
	struct cfq_queue *queue;
	int i;
	bool key_valid = false;
	unsigned long lowest_key = 0;
	enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;

2196 2197 2198
	for (i = 0; i <= SYNC_WORKLOAD; ++i) {
		/* select the one with lowest rb_key */
		queue = cfq_rb_first(service_tree_for(cfqg, prio, i));
2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209
		if (queue &&
		    (!key_valid || time_before(queue->rb_key, lowest_key))) {
			lowest_key = queue->rb_key;
			cur_best = i;
			key_valid = true;
		}
	}

	return cur_best;
}

2210
static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg)
2211 2212 2213
{
	unsigned slice;
	unsigned count;
2214
	struct cfq_rb_root *st;
2215
	unsigned group_slice;
2216
	enum wl_prio_t original_prio = cfqd->serving_prio;
2217

2218
	/* Choose next priority. RT > BE > IDLE */
2219
	if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg))
2220
		cfqd->serving_prio = RT_WORKLOAD;
2221
	else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg))
2222 2223 2224 2225 2226 2227 2228
		cfqd->serving_prio = BE_WORKLOAD;
	else {
		cfqd->serving_prio = IDLE_WORKLOAD;
		cfqd->workload_expires = jiffies + 1;
		return;
	}

2229 2230 2231
	if (original_prio != cfqd->serving_prio)
		goto new_workload;

2232 2233 2234 2235 2236
	/*
	 * For RT and BE, we have to choose also the type
	 * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
	 * expiration time
	 */
2237
	st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type);
2238
	count = st->count;
2239 2240

	/*
2241
	 * check workload expiration, and that we still have other queues ready
2242
	 */
2243
	if (count && !time_after(jiffies, cfqd->workload_expires))
2244 2245
		return;

2246
new_workload:
2247 2248
	/* otherwise select new workload type */
	cfqd->serving_type =
2249 2250
		cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio);
	st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type);
2251
	count = st->count;
2252 2253 2254 2255 2256 2257

	/*
	 * 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
	 */
2258 2259 2260 2261 2262
	group_slice = cfq_group_slice(cfqd, cfqg);

	slice = group_slice * count /
		max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio],
		      cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg));
2263

2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277
	if (cfqd->serving_type == ASYNC_WORKLOAD) {
		unsigned int tmp;

		/*
		 * 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.
		 */
		tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg);
		tmp = tmp/cfqd->busy_queues;
		slice = min_t(unsigned, slice, tmp);

2278 2279 2280
		/* async workload slice is scaled down according to
		 * the sync/async slice ratio. */
		slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1];
2281
	} else
2282 2283 2284 2285
		/* sync workload slice is at least 2 * cfq_slice_idle */
		slice = max(slice, 2 * cfqd->cfq_slice_idle);

	slice = max_t(unsigned, slice, CFQ_MIN_TT);
2286
	cfq_log(cfqd, "workload slice:%d", slice);
2287 2288 2289
	cfqd->workload_expires = jiffies + slice;
}

2290 2291 2292
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
2293
	struct cfq_group *cfqg;
2294 2295 2296

	if (RB_EMPTY_ROOT(&st->rb))
		return NULL;
2297 2298 2299
	cfqg = cfq_rb_first_group(st);
	update_min_vdisktime(st);
	return cfqg;
2300 2301
}

2302 2303
static void cfq_choose_cfqg(struct cfq_data *cfqd)
{
2304 2305 2306
	struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd);

	cfqd->serving_group = cfqg;
2307 2308 2309 2310 2311 2312

	/* Restore the workload type data */
	if (cfqg->saved_workload_slice) {
		cfqd->workload_expires = jiffies + cfqg->saved_workload_slice;
		cfqd->serving_type = cfqg->saved_workload;
		cfqd->serving_prio = cfqg->saved_serving_prio;
2313 2314 2315
	} else
		cfqd->workload_expires = jiffies - 1;

2316
	choose_service_tree(cfqd, cfqg);
2317 2318
}

2319
/*
2320 2321
 * 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.
2322
 */
2323
static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
L
Linus Torvalds 已提交
2324
{
2325
	struct cfq_queue *cfqq, *new_cfqq = NULL;
L
Linus Torvalds 已提交
2326

2327 2328 2329
	cfqq = cfqd->active_queue;
	if (!cfqq)
		goto new_queue;
L
Linus Torvalds 已提交
2330

2331 2332
	if (!cfqd->rq_queued)
		return NULL;
2333 2334 2335 2336 2337 2338 2339

	/*
	 * 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;

2340
	/*
J
Jens Axboe 已提交
2341
	 * The active queue has run out of time, expire it and select new.
2342
	 */
2343 2344 2345 2346 2347 2348 2349 2350 2351 2352
	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.
		 */
2353 2354 2355
		if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
		    && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
			cfqq = NULL;
2356
			goto keep_queue;
2357
		} else
2358
			goto check_group_idle;
2359
	}
L
Linus Torvalds 已提交
2360

2361
	/*
J
Jens Axboe 已提交
2362 2363
	 * The active queue has requests and isn't expired, allow it to
	 * dispatch.
2364
	 */
2365
	if (!RB_EMPTY_ROOT(&cfqq->sort_list))
2366
		goto keep_queue;
J
Jens Axboe 已提交
2367

2368 2369 2370 2371
	/*
	 * 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
2372
	 * tree.  If possible, merge the expiring queue with the new cfqq.
2373
	 */
2374
	new_cfqq = cfq_close_cooperator(cfqd, cfqq);
J
Jeff Moyer 已提交
2375 2376 2377
	if (new_cfqq) {
		if (!cfqq->new_cfqq)
			cfq_setup_merge(cfqq, new_cfqq);
2378
		goto expire;
J
Jeff Moyer 已提交
2379
	}
2380

J
Jens Axboe 已提交
2381 2382 2383 2384 2385
	/*
	 * 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.
	 */
2386 2387 2388 2389 2390
	if (timer_pending(&cfqd->idle_slice_timer)) {
		cfqq = NULL;
		goto keep_queue;
	}

2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401
	/*
	 * 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) ||
	    (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) {
		cfq_clear_cfqq_deep(cfqq);
		cfq_clear_cfqq_idle_window(cfqq);
	}

2402 2403 2404 2405 2406 2407 2408 2409 2410 2411
	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
2412 2413 2414
	if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 &&
	    cfqq->cfqg->dispatched &&
	    !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) {
2415 2416
		cfqq = NULL;
		goto keep_queue;
2417 2418
	}

J
Jens Axboe 已提交
2419
expire:
2420
	cfq_slice_expired(cfqd, 0);
J
Jens Axboe 已提交
2421
new_queue:
2422 2423 2424 2425 2426
	/*
	 * Current queue expired. Check if we have to switch to a new
	 * service tree
	 */
	if (!new_cfqq)
2427
		cfq_choose_cfqg(cfqd);
2428

2429
	cfqq = cfq_set_active_queue(cfqd, new_cfqq);
2430
keep_queue:
J
Jens Axboe 已提交
2431
	return cfqq;
2432 2433
}

J
Jens Axboe 已提交
2434
static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
2435 2436 2437 2438 2439 2440 2441 2442 2443
{
	int dispatched = 0;

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

	BUG_ON(!list_empty(&cfqq->fifo));
2444 2445

	/* By default cfqq is not expired if it is empty. Do it explicitly */
2446
	__cfq_slice_expired(cfqq->cfqd, cfqq, 0);
2447 2448 2449
	return dispatched;
}

2450 2451 2452 2453
/*
 * Drain our current requests. Used for barriers and when switching
 * io schedulers on-the-fly.
 */
2454
static int cfq_forced_dispatch(struct cfq_data *cfqd)
2455
{
2456
	struct cfq_queue *cfqq;
2457
	int dispatched = 0;
2458

2459
	/* Expire the timeslice of the current active queue first */
2460
	cfq_slice_expired(cfqd, 0);
2461 2462
	while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) {
		__cfq_set_active_queue(cfqd, cfqq);
2463
		dispatched += __cfq_forced_dispatch_cfqq(cfqq);
2464
	}
2465 2466 2467

	BUG_ON(cfqd->busy_queues);

2468
	cfq_log(cfqd, "forced_dispatch=%d", dispatched);
2469 2470 2471
	return dispatched;
}

S
Shaohua Li 已提交
2472 2473 2474 2475 2476
static inline bool cfq_slice_used_soon(struct cfq_data *cfqd,
	struct cfq_queue *cfqq)
{
	/* the queue hasn't finished any request, can't estimate */
	if (cfq_cfqq_slice_new(cfqq))
S
Shaohua Li 已提交
2477
		return true;
S
Shaohua Li 已提交
2478 2479
	if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched,
		cfqq->slice_end))
S
Shaohua Li 已提交
2480
		return true;
S
Shaohua Li 已提交
2481

S
Shaohua Li 已提交
2482
	return false;
S
Shaohua Li 已提交
2483 2484
}

2485
static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2486 2487
{
	unsigned int max_dispatch;
2488

2489 2490 2491
	/*
	 * Drain async requests before we start sync IO
	 */
2492
	if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC])
2493
		return false;
2494

2495 2496 2497
	/*
	 * If this is an async queue and we have sync IO in flight, let it wait
	 */
2498
	if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq))
2499
		return false;
2500

S
Shaohua Li 已提交
2501
	max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1);
2502 2503
	if (cfq_class_idle(cfqq))
		max_dispatch = 1;
2504

2505 2506 2507 2508
	/*
	 * Does this cfqq already have too much IO in flight?
	 */
	if (cfqq->dispatched >= max_dispatch) {
2509
		bool promote_sync = false;
2510 2511 2512
		/*
		 * idle queue must always only have a single IO in flight
		 */
2513
		if (cfq_class_idle(cfqq))
2514
			return false;
2515

2516
		/*
2517 2518
		 * If there is only one sync queue
		 * we can ignore async queue here and give the sync
2519 2520 2521 2522
		 * 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.
		 */
2523 2524
		if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1)
			promote_sync = true;
2525

2526 2527 2528
		/*
		 * We have other queues, don't allow more IO from this one
		 */
2529 2530
		if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) &&
				!promote_sync)
2531
			return false;
2532

2533
		/*
2534
		 * Sole queue user, no limit
2535
		 */
2536
		if (cfqd->busy_queues == 1 || promote_sync)
S
Shaohua Li 已提交
2537 2538 2539 2540 2541 2542 2543 2544 2545
			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;
2546 2547 2548 2549 2550 2551 2552
	}

	/*
	 * 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
	 */
2553
	if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
2554
		unsigned long last_sync = jiffies - cfqd->last_delayed_sync;
2555
		unsigned int depth;
2556

2557
		depth = last_sync / cfqd->cfq_slice[1];
2558 2559
		if (!depth && !cfqq->dispatched)
			depth = 1;
2560 2561
		if (depth < max_dispatch)
			max_dispatch = depth;
2562
	}
2563

2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595
	/*
	 * 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));

	if (!cfq_may_dispatch(cfqd, cfqq))
		return false;

	/*
	 * follow expired path, else get first next available
	 */
	rq = cfq_check_fifo(cfqq);
	if (!rq)
		rq = cfqq->next_rq;

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

	if (!cfqd->active_cic) {
2596
		struct cfq_io_cq *cic = RQ_CIC(rq);
2597

2598
		atomic_long_inc(&cic->icq.ioc->refcount);
2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621
		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)
2622 2623
		return 0;

2624
	/*
2625
	 * Dispatch a request from this cfqq, if it is allowed
2626
	 */
2627 2628 2629
	if (!cfq_dispatch_request(cfqd, cfqq))
		return 0;

2630
	cfqq->slice_dispatch++;
2631
	cfq_clear_cfqq_must_dispatch(cfqq);
2632

2633 2634 2635 2636 2637 2638 2639 2640
	/*
	 * 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))) {
		cfqq->slice_end = jiffies + 1;
2641
		cfq_slice_expired(cfqd, 0);
L
Linus Torvalds 已提交
2642 2643
	}

2644
	cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
2645
	return 1;
L
Linus Torvalds 已提交
2646 2647 2648
}

/*
J
Jens Axboe 已提交
2649 2650
 * 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
2651
 *
2652
 * Each cfq queue took a reference on the parent group. Drop it now.
L
Linus Torvalds 已提交
2653 2654 2655 2656
 * queue lock must be held here.
 */
static void cfq_put_queue(struct cfq_queue *cfqq)
{
2657
	struct cfq_data *cfqd = cfqq->cfqd;
2658
	struct cfq_group *cfqg;
2659

2660
	BUG_ON(cfqq->ref <= 0);
L
Linus Torvalds 已提交
2661

2662 2663
	cfqq->ref--;
	if (cfqq->ref)
L
Linus Torvalds 已提交
2664 2665
		return;

2666
	cfq_log_cfqq(cfqd, cfqq, "put_queue");
L
Linus Torvalds 已提交
2667
	BUG_ON(rb_first(&cfqq->sort_list));
2668
	BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
2669
	cfqg = cfqq->cfqg;
L
Linus Torvalds 已提交
2670

2671
	if (unlikely(cfqd->active_queue == cfqq)) {
2672
		__cfq_slice_expired(cfqd, cfqq, 0);
2673
		cfq_schedule_dispatch(cfqd);
2674
	}
2675

2676
	BUG_ON(cfq_cfqq_on_rr(cfqq));
L
Linus Torvalds 已提交
2677
	kmem_cache_free(cfq_pool, cfqq);
2678
	cfq_put_cfqg(cfqg);
L
Linus Torvalds 已提交
2679 2680
}

2681
static void cfq_put_cooperator(struct cfq_queue *cfqq)
L
Linus Torvalds 已提交
2682
{
J
Jeff Moyer 已提交
2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699
	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;
	}
2700 2701 2702 2703 2704 2705 2706 2707 2708 2709
}

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
2710

2711 2712
	cfq_put_queue(cfqq);
}
2713

2714 2715 2716 2717 2718 2719 2720
static void cfq_init_icq(struct io_cq *icq)
{
	struct cfq_io_cq *cic = icq_to_cic(icq);

	cic->ttime.last_end_request = jiffies;
}

2721
static void cfq_exit_icq(struct io_cq *icq)
2722
{
2723
	struct cfq_io_cq *cic = icq_to_cic(icq);
2724
	struct cfq_data *cfqd = cic_to_cfqd(cic);
2725

2726 2727 2728
	if (cic->cfqq[BLK_RW_ASYNC]) {
		cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]);
		cic->cfqq[BLK_RW_ASYNC] = NULL;
2729 2730
	}

2731 2732 2733
	if (cic->cfqq[BLK_RW_SYNC]) {
		cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]);
		cic->cfqq[BLK_RW_SYNC] = NULL;
2734
	}
2735 2736
}

2737
static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
2738 2739 2740 2741
{
	struct task_struct *tsk = current;
	int ioprio_class;

J
Jens Axboe 已提交
2742
	if (!cfq_cfqq_prio_changed(cfqq))
2743 2744
		return;

2745
	ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio);
2746
	switch (ioprio_class) {
2747 2748 2749 2750
	default:
		printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
	case IOPRIO_CLASS_NONE:
		/*
2751
		 * no prio set, inherit CPU scheduling settings
2752 2753
		 */
		cfqq->ioprio = task_nice_ioprio(tsk);
2754
		cfqq->ioprio_class = task_nice_ioclass(tsk);
2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768
		break;
	case IOPRIO_CLASS_RT:
		cfqq->ioprio = task_ioprio(ioc);
		cfqq->ioprio_class = IOPRIO_CLASS_RT;
		break;
	case IOPRIO_CLASS_BE:
		cfqq->ioprio = task_ioprio(ioc);
		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;
2769 2770 2771 2772 2773 2774 2775
	}

	/*
	 * keep track of original prio settings in case we have to temporarily
	 * elevate the priority of this queue
	 */
	cfqq->org_ioprio = cfqq->ioprio;
J
Jens Axboe 已提交
2776
	cfq_clear_cfqq_prio_changed(cfqq);
2777 2778
}

2779
static void changed_ioprio(struct cfq_io_cq *cic)
2780
{
2781
	struct cfq_data *cfqd = cic_to_cfqd(cic);
2782
	struct cfq_queue *cfqq;
2783

2784 2785 2786
	if (unlikely(!cfqd))
		return;

2787
	cfqq = cic->cfqq[BLK_RW_ASYNC];
2788 2789
	if (cfqq) {
		struct cfq_queue *new_cfqq;
2790
		new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->icq.ioc,
2791
						GFP_ATOMIC);
2792
		if (new_cfqq) {
2793
			cic->cfqq[BLK_RW_ASYNC] = new_cfqq;
2794 2795
			cfq_put_queue(cfqq);
		}
2796
	}
2797

2798
	cfqq = cic->cfqq[BLK_RW_SYNC];
2799 2800
	if (cfqq)
		cfq_mark_cfqq_prio_changed(cfqq);
2801 2802
}

2803
static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2804
			  pid_t pid, bool is_sync)
2805 2806 2807 2808 2809
{
	RB_CLEAR_NODE(&cfqq->rb_node);
	RB_CLEAR_NODE(&cfqq->p_node);
	INIT_LIST_HEAD(&cfqq->fifo);

2810
	cfqq->ref = 0;
2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822
	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;
}

2823
#ifdef CONFIG_CFQ_GROUP_IOSCHED
2824
static void changed_cgroup(struct cfq_io_cq *cic)
2825 2826
{
	struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1);
2827
	struct cfq_data *cfqd = cic_to_cfqd(cic);
2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846
	struct request_queue *q;

	if (unlikely(!cfqd))
		return;

	q = cfqd->queue;

	if (sync_cfqq) {
		/*
		 * Drop reference to sync queue. A new sync queue will be
		 * assigned in new group upon arrival of a fresh request.
		 */
		cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup");
		cic_set_cfqq(cic, NULL, 1);
		cfq_put_queue(sync_cfqq);
	}
}
#endif  /* CONFIG_CFQ_GROUP_IOSCHED */

2847
static struct cfq_queue *
2848
cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync,
2849
		     struct io_context *ioc, gfp_t gfp_mask)
2850 2851
{
	struct cfq_queue *cfqq, *new_cfqq = NULL;
2852
	struct cfq_io_cq *cic;
2853
	struct cfq_group *cfqg;
2854 2855

retry:
2856
	cfqg = cfq_get_cfqg(cfqd);
2857
	cic = cfq_cic_lookup(cfqd, ioc);
2858 2859
	/* cic always exists here */
	cfqq = cic_to_cfqq(cic, is_sync);
2860

2861 2862 2863 2864 2865 2866
	/*
	 * Always try a new alloc if we fell back to the OOM cfqq
	 * originally, since it should just be a temporary situation.
	 */
	if (!cfqq || cfqq == &cfqd->oom_cfqq) {
		cfqq = NULL;
2867 2868 2869 2870 2871
		if (new_cfqq) {
			cfqq = new_cfqq;
			new_cfqq = NULL;
		} else if (gfp_mask & __GFP_WAIT) {
			spin_unlock_irq(cfqd->queue->queue_lock);
2872
			new_cfqq = kmem_cache_alloc_node(cfq_pool,
2873
					gfp_mask | __GFP_ZERO,
2874
					cfqd->queue->node);
2875
			spin_lock_irq(cfqd->queue->queue_lock);
2876 2877
			if (new_cfqq)
				goto retry;
2878
		} else {
2879 2880 2881
			cfqq = kmem_cache_alloc_node(cfq_pool,
					gfp_mask | __GFP_ZERO,
					cfqd->queue->node);
2882 2883
		}

2884 2885 2886
		if (cfqq) {
			cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
			cfq_init_prio_data(cfqq, ioc);
2887
			cfq_link_cfqq_cfqg(cfqq, cfqg);
2888 2889 2890
			cfq_log_cfqq(cfqd, cfqq, "alloced");
		} else
			cfqq = &cfqd->oom_cfqq;
2891 2892 2893 2894 2895 2896 2897 2898
	}

	if (new_cfqq)
		kmem_cache_free(cfq_pool, new_cfqq);

	return cfqq;
}

2899 2900 2901
static struct cfq_queue **
cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio)
{
2902
	switch (ioprio_class) {
2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913
	case IOPRIO_CLASS_RT:
		return &cfqd->async_cfqq[0][ioprio];
	case IOPRIO_CLASS_BE:
		return &cfqd->async_cfqq[1][ioprio];
	case IOPRIO_CLASS_IDLE:
		return &cfqd->async_idle_cfqq;
	default:
		BUG();
	}
}

2914
static struct cfq_queue *
2915
cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc,
2916 2917
	      gfp_t gfp_mask)
{
2918 2919
	const int ioprio = task_ioprio(ioc);
	const int ioprio_class = task_ioprio_class(ioc);
2920
	struct cfq_queue **async_cfqq = NULL;
2921 2922
	struct cfq_queue *cfqq = NULL;

2923 2924 2925 2926 2927
	if (!is_sync) {
		async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio);
		cfqq = *async_cfqq;
	}

2928
	if (!cfqq)
2929
		cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask);
2930 2931 2932 2933

	/*
	 * pin the queue now that it's allocated, scheduler exit will prune it
	 */
2934
	if (!is_sync && !(*async_cfqq)) {
2935
		cfqq->ref++;
2936
		*async_cfqq = cfqq;
2937 2938
	}

2939
	cfqq->ref++;
2940 2941 2942
	return cfqq;
}

2943
static void
2944
__cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle)
L
Linus Torvalds 已提交
2945
{
2946 2947
	unsigned long elapsed = jiffies - ttime->last_end_request;
	elapsed = min(elapsed, 2UL * slice_idle);
2948

2949 2950 2951 2952 2953 2954 2955
	ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
	ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8;
	ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples;
}

static void
cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2956
			struct cfq_io_cq *cic)
2957
{
2958
	if (cfq_cfqq_sync(cfqq)) {
2959
		__cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle);
2960 2961 2962
		__cfq_update_io_thinktime(&cfqq->service_tree->ttime,
			cfqd->cfq_slice_idle);
	}
S
Shaohua Li 已提交
2963 2964 2965
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	__cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle);
#endif
2966
}
L
Linus Torvalds 已提交
2967

2968
static void
2969
cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
J
Jens Axboe 已提交
2970
		       struct request *rq)
2971
{
2972
	sector_t sdist = 0;
2973
	sector_t n_sec = blk_rq_sectors(rq);
2974 2975 2976 2977 2978 2979
	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);
	}
2980

2981
	cfqq->seek_history <<= 1;
2982 2983 2984 2985
	if (blk_queue_nonrot(cfqd->queue))
		cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT);
	else
		cfqq->seek_history |= (sdist > CFQQ_SEEK_THR);
2986
}
L
Linus Torvalds 已提交
2987

2988 2989 2990 2991 2992 2993
/*
 * 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,
2994
		       struct cfq_io_cq *cic)
2995
{
2996
	int old_idle, enable_idle;
2997

2998 2999 3000 3001
	/*
	 * Don't idle for async or idle io prio class
	 */
	if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
3002 3003
		return;

3004
	enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
L
Linus Torvalds 已提交
3005

3006 3007 3008
	if (cfqq->queued[0] + cfqq->queued[1] >= 4)
		cfq_mark_cfqq_deep(cfqq);

3009 3010
	if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE))
		enable_idle = 0;
3011 3012 3013
	else if (!atomic_read(&cic->icq.ioc->nr_tasks) ||
		 !cfqd->cfq_slice_idle ||
		 (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq)))
3014
		enable_idle = 0;
3015 3016
	else if (sample_valid(cic->ttime.ttime_samples)) {
		if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle)
3017 3018 3019
			enable_idle = 0;
		else
			enable_idle = 1;
L
Linus Torvalds 已提交
3020 3021
	}

3022 3023 3024 3025 3026 3027 3028
	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);
	}
3029
}
L
Linus Torvalds 已提交
3030

3031 3032 3033 3034
/*
 * 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.
 */
3035
static bool
3036
cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
J
Jens Axboe 已提交
3037
		   struct request *rq)
3038
{
J
Jens Axboe 已提交
3039
	struct cfq_queue *cfqq;
3040

J
Jens Axboe 已提交
3041 3042
	cfqq = cfqd->active_queue;
	if (!cfqq)
3043
		return false;
3044

J
Jens Axboe 已提交
3045
	if (cfq_class_idle(new_cfqq))
3046
		return false;
3047 3048

	if (cfq_class_idle(cfqq))
3049
		return true;
3050

3051 3052 3053 3054 3055 3056
	/*
	 * 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;

3057 3058 3059 3060
	/*
	 * if the new request is sync, but the currently running queue is
	 * not, let the sync request have priority.
	 */
J
Jens Axboe 已提交
3061
	if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
3062
		return true;
3063

3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076
	if (new_cfqq->cfqg != cfqq->cfqg)
		return false;

	if (cfq_slice_used(cfqq))
		return true;

	/* Allow preemption only if we are idling on sync-noidle tree */
	if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD &&
	    cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD &&
	    new_cfqq->service_tree->count == 2 &&
	    RB_EMPTY_ROOT(&cfqq->sort_list))
		return true;

3077 3078 3079 3080
	/*
	 * 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.
	 */
3081
	if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending)
3082 3083
		return true;

3084 3085 3086 3087
	/*
	 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
	 */
	if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
3088
		return true;
3089

3090 3091 3092 3093
	/* 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;

3094
	if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
3095
		return false;
3096 3097 3098 3099 3100

	/*
	 * if this request is as-good as one we would expect from the
	 * current cfqq, let it preempt
	 */
3101
	if (cfq_rq_close(cfqd, cfqq, rq))
3102
		return true;
3103

3104
	return false;
3105 3106 3107 3108 3109 3110 3111 3112
}

/*
 * 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
3113 3114
	enum wl_type_t old_type = cfqq_type(cfqd->active_queue);

3115
	cfq_log_cfqq(cfqd, cfqq, "preempt");
S
Shaohua Li 已提交
3116
	cfq_slice_expired(cfqd, 1);
3117

3118 3119 3120 3121
	/*
	 * workload type is changed, don't save slice, otherwise preempt
	 * doesn't happen
	 */
S
Shaohua Li 已提交
3122
	if (old_type != cfqq_type(cfqq))
3123 3124
		cfqq->cfqg->saved_workload_slice = 0;

3125 3126 3127 3128 3129
	/*
	 * 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));
3130 3131

	cfq_service_tree_add(cfqd, cfqq, 1);
3132

3133 3134
	cfqq->slice_end = 0;
	cfq_mark_cfqq_slice_new(cfqq);
3135 3136 3137
}

/*
J
Jens Axboe 已提交
3138
 * Called when a new fs request (rq) is added (to cfqq). Check if there's
3139 3140 3141
 * something we should do about it
 */
static void
J
Jens Axboe 已提交
3142 3143
cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		struct request *rq)
3144
{
3145
	struct cfq_io_cq *cic = RQ_CIC(rq);
3146

3147
	cfqd->rq_queued++;
3148 3149
	if (rq->cmd_flags & REQ_PRIO)
		cfqq->prio_pending++;
3150

3151
	cfq_update_io_thinktime(cfqd, cfqq, cic);
3152
	cfq_update_io_seektime(cfqd, cfqq, rq);
J
Jens Axboe 已提交
3153 3154
	cfq_update_idle_window(cfqd, cfqq, cic);

3155
	cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
3156 3157 3158

	if (cfqq == cfqd->active_queue) {
		/*
3159 3160 3161
		 * 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
3162 3163
		 * and merging. If the request is already larger than a single
		 * page, let it rip immediately. For that case we assume that
3164 3165 3166
		 * 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.
3167
		 */
3168
		if (cfq_cfqq_wait_request(cfqq)) {
3169 3170
			if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE ||
			    cfqd->busy_queues > 1) {
3171
				cfq_del_timer(cfqd, cfqq);
3172
				cfq_clear_cfqq_wait_request(cfqq);
3173
				__blk_run_queue(cfqd->queue);
3174
			} else {
3175
				cfq_blkiocg_update_idle_time_stats(
3176
						&cfqq->cfqg->blkg);
3177
				cfq_mark_cfqq_must_dispatch(cfqq);
3178
			}
3179
		}
J
Jens Axboe 已提交
3180
	} else if (cfq_should_preempt(cfqd, cfqq, rq)) {
3181 3182 3183
		/*
		 * not the active queue - expire current slice if it is
		 * idle and has expired it's mean thinktime or this new queue
3184 3185
		 * has some old slice time left and is of higher priority or
		 * this new queue is RT and the current one is BE
3186 3187
		 */
		cfq_preempt_queue(cfqd, cfqq);
3188
		__blk_run_queue(cfqd->queue);
3189
	}
L
Linus Torvalds 已提交
3190 3191
}

3192
static void cfq_insert_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
3193
{
3194
	struct cfq_data *cfqd = q->elevator->elevator_data;
J
Jens Axboe 已提交
3195
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
3196

3197
	cfq_log_cfqq(cfqd, cfqq, "insert_request");
3198
	cfq_init_prio_data(cfqq, RQ_CIC(rq)->icq.ioc);
L
Linus Torvalds 已提交
3199

3200
	rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]);
3201
	list_add_tail(&rq->queuelist, &cfqq->fifo);
3202
	cfq_add_rq_rb(rq);
3203
	cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg,
3204 3205
			&cfqd->serving_group->blkg, rq_data_dir(rq),
			rq_is_sync(rq));
J
Jens Axboe 已提交
3206
	cfq_rq_enqueued(cfqd, cfqq, rq);
L
Linus Torvalds 已提交
3207 3208
}

3209 3210 3211 3212 3213 3214
/*
 * 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
3215 3216
	struct cfq_queue *cfqq = cfqd->active_queue;

3217 3218
	if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth)
		cfqd->hw_tag_est_depth = cfqd->rq_in_driver;
3219 3220 3221

	if (cfqd->hw_tag == 1)
		return;
3222 3223

	if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
3224
	    cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
3225 3226
		return;

S
Shaohua Li 已提交
3227 3228 3229 3230 3231 3232 3233
	/*
	 * 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] <
3234
	    CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN)
S
Shaohua Li 已提交
3235 3236
		return;

3237 3238 3239
	if (cfqd->hw_tag_samples++ < 50)
		return;

3240
	if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN)
3241 3242 3243 3244 3245
		cfqd->hw_tag = 1;
	else
		cfqd->hw_tag = 0;
}

3246 3247
static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
3248
	struct cfq_io_cq *cic = cfqd->active_cic;
3249

3250 3251 3252 3253
	/* If the queue already has requests, don't wait */
	if (!RB_EMPTY_ROOT(&cfqq->sort_list))
		return false;

3254 3255 3256 3257
	/* 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
3258 3259 3260 3261
	/* the only queue in the group, but think time is big */
	if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true))
		return false;

3262 3263 3264 3265
	if (cfq_slice_used(cfqq))
		return true;

	/* if slice left is less than think time, wait busy */
3266 3267
	if (cic && sample_valid(cic->ttime.ttime_samples)
	    && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean))
3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282
		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.
	 */
	if (cfqq->slice_end - jiffies == 1)
		return true;

	return false;
}

3283
static void cfq_completed_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
3284
{
J
Jens Axboe 已提交
3285
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
3286
	struct cfq_data *cfqd = cfqq->cfqd;
3287
	const int sync = rq_is_sync(rq);
3288
	unsigned long now;
L
Linus Torvalds 已提交
3289

3290
	now = jiffies;
3291 3292
	cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d",
		     !!(rq->cmd_flags & REQ_NOIDLE));
L
Linus Torvalds 已提交
3293

3294 3295
	cfq_update_hw_tag(cfqd);

3296
	WARN_ON(!cfqd->rq_in_driver);
J
Jens Axboe 已提交
3297
	WARN_ON(!cfqq->dispatched);
3298
	cfqd->rq_in_driver--;
J
Jens Axboe 已提交
3299
	cfqq->dispatched--;
3300
	(RQ_CFQG(rq))->dispatched--;
3301 3302 3303
	cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg,
			rq_start_time_ns(rq), rq_io_start_time_ns(rq),
			rq_data_dir(rq), rq_is_sync(rq));
L
Linus Torvalds 已提交
3304

3305
	cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--;
3306

3307
	if (sync) {
3308 3309
		struct cfq_rb_root *service_tree;

3310
		RQ_CIC(rq)->ttime.last_end_request = now;
3311 3312 3313 3314 3315 3316 3317

		if (cfq_cfqq_on_rr(cfqq))
			service_tree = cfqq->service_tree;
		else
			service_tree = service_tree_for(cfqq->cfqg,
				cfqq_prio(cfqq), cfqq_type(cfqq));
		service_tree->ttime.last_end_request = now;
3318 3319
		if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now))
			cfqd->last_delayed_sync = now;
3320
	}
3321

S
Shaohua Li 已提交
3322 3323 3324 3325
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	cfqq->cfqg->ttime.last_end_request = now;
#endif

3326 3327 3328 3329 3330
	/*
	 * 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) {
3331 3332
		const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);

3333 3334 3335 3336
		if (cfq_cfqq_slice_new(cfqq)) {
			cfq_set_prio_slice(cfqd, cfqq);
			cfq_clear_cfqq_slice_new(cfqq);
		}
3337 3338

		/*
3339 3340
		 * Should we wait for next request to come in before we expire
		 * the queue.
3341
		 */
3342
		if (cfq_should_wait_busy(cfqd, cfqq)) {
3343 3344 3345 3346
			unsigned long extend_sl = cfqd->cfq_slice_idle;
			if (!cfqd->cfq_slice_idle)
				extend_sl = cfqd->cfq_group_idle;
			cfqq->slice_end = jiffies + extend_sl;
3347
			cfq_mark_cfqq_wait_busy(cfqq);
3348
			cfq_log_cfqq(cfqd, cfqq, "will busy wait");
3349 3350
		}

3351
		/*
3352 3353 3354 3355 3356 3357
		 * Idling is not enabled on:
		 * - expired queues
		 * - idle-priority queues
		 * - async queues
		 * - queues with still some requests queued
		 * - when there is a close cooperator
3358
		 */
3359
		if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
3360
			cfq_slice_expired(cfqd, 1);
3361 3362
		else if (sync && cfqq_empty &&
			 !cfq_close_cooperator(cfqd, cfqq)) {
3363
			cfq_arm_slice_timer(cfqd);
3364
		}
3365
	}
J
Jens Axboe 已提交
3366

3367
	if (!cfqd->rq_in_driver)
3368
		cfq_schedule_dispatch(cfqd);
L
Linus Torvalds 已提交
3369 3370
}

3371
static inline int __cfq_may_queue(struct cfq_queue *cfqq)
3372
{
3373
	if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
J
Jens Axboe 已提交
3374
		cfq_mark_cfqq_must_alloc_slice(cfqq);
3375
		return ELV_MQUEUE_MUST;
J
Jens Axboe 已提交
3376
	}
L
Linus Torvalds 已提交
3377

3378 3379 3380
	return ELV_MQUEUE_MAY;
}

3381
static int cfq_may_queue(struct request_queue *q, int rw)
3382 3383 3384
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct task_struct *tsk = current;
3385
	struct cfq_io_cq *cic;
3386 3387 3388 3389 3390 3391 3392 3393
	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
	 */
3394
	cic = cfq_cic_lookup(cfqd, tsk->io_context);
3395 3396 3397
	if (!cic)
		return ELV_MQUEUE_MAY;

3398
	cfqq = cic_to_cfqq(cic, rw_is_sync(rw));
3399
	if (cfqq) {
3400
		cfq_init_prio_data(cfqq, cic->icq.ioc);
3401

3402
		return __cfq_may_queue(cfqq);
3403 3404 3405
	}

	return ELV_MQUEUE_MAY;
L
Linus Torvalds 已提交
3406 3407 3408 3409 3410
}

/*
 * queue lock held here
 */
3411
static void cfq_put_request(struct request *rq)
L
Linus Torvalds 已提交
3412
{
J
Jens Axboe 已提交
3413
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
L
Linus Torvalds 已提交
3414

J
Jens Axboe 已提交
3415
	if (cfqq) {
3416
		const int rw = rq_data_dir(rq);
L
Linus Torvalds 已提交
3417

3418 3419
		BUG_ON(!cfqq->allocated[rw]);
		cfqq->allocated[rw]--;
L
Linus Torvalds 已提交
3420

3421 3422
		/* Put down rq reference on cfqg */
		cfq_put_cfqg(RQ_CFQG(rq));
3423 3424
		rq->elv.priv[0] = NULL;
		rq->elv.priv[1] = NULL;
3425

L
Linus Torvalds 已提交
3426 3427 3428 3429
		cfq_put_queue(cfqq);
	}
}

J
Jeff Moyer 已提交
3430
static struct cfq_queue *
3431
cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic,
J
Jeff Moyer 已提交
3432 3433 3434 3435
		struct cfq_queue *cfqq)
{
	cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
	cic_set_cfqq(cic, cfqq->new_cfqq, 1);
3436
	cfq_mark_cfqq_coop(cfqq->new_cfqq);
J
Jeff Moyer 已提交
3437 3438 3439 3440
	cfq_put_queue(cfqq);
	return cic_to_cfqq(cic, 1);
}

3441 3442 3443 3444 3445
/*
 * 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 *
3446
split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq)
3447 3448 3449 3450
{
	if (cfqq_process_refs(cfqq) == 1) {
		cfqq->pid = current->pid;
		cfq_clear_cfqq_coop(cfqq);
3451
		cfq_clear_cfqq_split_coop(cfqq);
3452 3453 3454 3455
		return cfqq;
	}

	cic_set_cfqq(cic, NULL, 1);
3456 3457 3458

	cfq_put_cooperator(cfqq);

3459 3460 3461
	cfq_put_queue(cfqq);
	return NULL;
}
L
Linus Torvalds 已提交
3462
/*
3463
 * Allocate cfq data structures associated with this request.
L
Linus Torvalds 已提交
3464
 */
3465
static int
3466
cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
L
Linus Torvalds 已提交
3467 3468
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
3469
	struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq);
L
Linus Torvalds 已提交
3470
	const int rw = rq_data_dir(rq);
3471
	const bool is_sync = rq_is_sync(rq);
3472
	struct cfq_queue *cfqq;
L
Linus Torvalds 已提交
3473 3474 3475

	might_sleep_if(gfp_mask & __GFP_WAIT);

3476
	spin_lock_irq(q->queue_lock);
3477 3478 3479 3480 3481 3482 3483 3484 3485 3486

	/* handle changed notifications */
	if (unlikely(cic->icq.changed)) {
		if (test_and_clear_bit(ICQ_IOPRIO_CHANGED, &cic->icq.changed))
			changed_ioprio(cic);
#ifdef CONFIG_CFQ_GROUP_IOSCHED
		if (test_and_clear_bit(ICQ_CGROUP_CHANGED, &cic->icq.changed))
			changed_cgroup(cic);
#endif
	}
3487

3488
new_queue:
3489
	cfqq = cic_to_cfqq(cic, is_sync);
3490
	if (!cfqq || cfqq == &cfqd->oom_cfqq) {
3491
		cfqq = cfq_get_queue(cfqd, is_sync, cic->icq.ioc, gfp_mask);
3492
		cic_set_cfqq(cic, cfqq, is_sync);
J
Jeff Moyer 已提交
3493
	} else {
3494 3495 3496
		/*
		 * If the queue was seeky for too long, break it apart.
		 */
3497
		if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) {
3498 3499 3500 3501 3502 3503
			cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
			cfqq = split_cfqq(cic, cfqq);
			if (!cfqq)
				goto new_queue;
		}

J
Jeff Moyer 已提交
3504 3505 3506 3507 3508 3509 3510 3511
		/*
		 * 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);
3512
	}
L
Linus Torvalds 已提交
3513 3514 3515

	cfqq->allocated[rw]++;

3516
	cfqq->ref++;
3517 3518
	rq->elv.priv[0] = cfqq;
	rq->elv.priv[1] = cfq_ref_get_cfqg(cfqq->cfqg);
3519
	spin_unlock_irq(q->queue_lock);
J
Jens Axboe 已提交
3520
	return 0;
L
Linus Torvalds 已提交
3521 3522
}

3523
static void cfq_kick_queue(struct work_struct *work)
3524
{
3525
	struct cfq_data *cfqd =
3526
		container_of(work, struct cfq_data, unplug_work);
3527
	struct request_queue *q = cfqd->queue;
3528

3529
	spin_lock_irq(q->queue_lock);
3530
	__blk_run_queue(cfqd->queue);
3531
	spin_unlock_irq(q->queue_lock);
3532 3533 3534 3535 3536 3537 3538 3539 3540 3541
}

/*
 * Timer running if the active_queue is currently idling inside its time slice
 */
static void cfq_idle_slice_timer(unsigned long data)
{
	struct cfq_data *cfqd = (struct cfq_data *) data;
	struct cfq_queue *cfqq;
	unsigned long flags;
3542
	int timed_out = 1;
3543

3544 3545
	cfq_log(cfqd, "idle timer fired");

3546 3547
	spin_lock_irqsave(cfqd->queue->queue_lock, flags);

3548 3549
	cfqq = cfqd->active_queue;
	if (cfqq) {
3550 3551
		timed_out = 0;

3552 3553 3554 3555 3556 3557
		/*
		 * We saw a request before the queue expired, let it through
		 */
		if (cfq_cfqq_must_dispatch(cfqq))
			goto out_kick;

3558 3559 3560
		/*
		 * expired
		 */
3561
		if (cfq_slice_used(cfqq))
3562 3563 3564 3565 3566 3567
			goto expire;

		/*
		 * only expire and reinvoke request handler, if there are
		 * other queues with pending requests
		 */
3568
		if (!cfqd->busy_queues)
3569 3570 3571 3572 3573
			goto out_cont;

		/*
		 * not expired and it has a request pending, let it dispatch
		 */
3574
		if (!RB_EMPTY_ROOT(&cfqq->sort_list))
3575
			goto out_kick;
3576 3577 3578 3579 3580

		/*
		 * Queue depth flag is reset only when the idle didn't succeed
		 */
		cfq_clear_cfqq_deep(cfqq);
3581 3582
	}
expire:
3583
	cfq_slice_expired(cfqd, timed_out);
3584
out_kick:
3585
	cfq_schedule_dispatch(cfqd);
3586 3587 3588 3589
out_cont:
	spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
}

J
Jens Axboe 已提交
3590 3591 3592
static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
{
	del_timer_sync(&cfqd->idle_slice_timer);
3593
	cancel_work_sync(&cfqd->unplug_work);
J
Jens Axboe 已提交
3594
}
3595

3596 3597 3598 3599 3600 3601 3602 3603 3604 3605
static void cfq_put_async_queues(struct cfq_data *cfqd)
{
	int i;

	for (i = 0; i < IOPRIO_BE_NR; i++) {
		if (cfqd->async_cfqq[0][i])
			cfq_put_queue(cfqd->async_cfqq[0][i]);
		if (cfqd->async_cfqq[1][i])
			cfq_put_queue(cfqd->async_cfqq[1][i]);
	}
3606 3607 3608

	if (cfqd->async_idle_cfqq)
		cfq_put_queue(cfqd->async_idle_cfqq);
3609 3610
}

J
Jens Axboe 已提交
3611
static void cfq_exit_queue(struct elevator_queue *e)
L
Linus Torvalds 已提交
3612
{
3613
	struct cfq_data *cfqd = e->elevator_data;
3614
	struct request_queue *q = cfqd->queue;
3615
	bool wait = false;
3616

J
Jens Axboe 已提交
3617
	cfq_shutdown_timer_wq(cfqd);
3618

3619
	spin_lock_irq(q->queue_lock);
3620

3621
	if (cfqd->active_queue)
3622
		__cfq_slice_expired(cfqd, cfqd->active_queue, 0);
3623

3624
	cfq_put_async_queues(cfqd);
3625
	cfq_release_cfq_groups(cfqd);
3626 3627 3628 3629 3630 3631 3632

	/*
	 * If there are groups which we could not unlink from blkcg list,
	 * wait for a rcu period for them to be freed.
	 */
	if (cfqd->nr_blkcg_linked_grps)
		wait = true;
3633

3634
	spin_unlock_irq(q->queue_lock);
3635 3636 3637

	cfq_shutdown_timer_wq(cfqd);

3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650
	/*
	 * Wait for cfqg->blkg->key accessors to exit their grace periods.
	 * Do this wait only if there are other unlinked groups out
	 * there. This can happen if cgroup deletion path claimed the
	 * responsibility of cleaning up a group before queue cleanup code
	 * get to the group.
	 *
	 * Do not call synchronize_rcu() unconditionally as there are drivers
	 * which create/delete request queue hundreds of times during scan/boot
	 * and synchronize_rcu() can take significant time and slow down boot.
	 */
	if (wait)
		synchronize_rcu();
3651 3652 3653 3654 3655

#ifdef CONFIG_CFQ_GROUP_IOSCHED
	/* Free up per cpu stats for root group */
	free_percpu(cfqd->root_group.blkg.stats_cpu);
#endif
3656
	kfree(cfqd);
L
Linus Torvalds 已提交
3657 3658
}

3659
static void *cfq_init_queue(struct request_queue *q)
L
Linus Torvalds 已提交
3660 3661
{
	struct cfq_data *cfqd;
3662
	int i, j;
3663
	struct cfq_group *cfqg;
3664
	struct cfq_rb_root *st;
L
Linus Torvalds 已提交
3665

3666
	cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
3667
	if (!cfqd)
J
Jens Axboe 已提交
3668
		return NULL;
3669

3670 3671 3672
	/* Init root service tree */
	cfqd->grp_service_tree = CFQ_RB_ROOT;

3673 3674
	/* Init root group */
	cfqg = &cfqd->root_group;
3675 3676
	for_each_cfqg_st(cfqg, i, j, st)
		*st = CFQ_RB_ROOT;
3677
	RB_CLEAR_NODE(&cfqg->rb_node);
3678

3679 3680 3681
	/* Give preference to root group over other groups */
	cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT;

3682
#ifdef CONFIG_CFQ_GROUP_IOSCHED
3683
	/*
3684 3685 3686 3687 3688
	 * Set root group reference to 2. One reference will be dropped when
	 * all groups on cfqd->cfqg_list are being deleted during queue exit.
	 * Other reference will remain there as we don't want to delete this
	 * group as it is statically allocated and gets destroyed when
	 * throtl_data goes away.
3689
	 */
3690
	cfqg->ref = 2;
3691 3692 3693 3694 3695 3696 3697

	if (blkio_alloc_blkg_stats(&cfqg->blkg)) {
		kfree(cfqg);
		kfree(cfqd);
		return NULL;
	}

3698
	rcu_read_lock();
3699

3700 3701
	cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg,
					(void *)cfqd, 0);
3702
	rcu_read_unlock();
3703 3704 3705 3706
	cfqd->nr_blkcg_linked_grps++;

	/* Add group on cfqd->cfqg_list */
	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
3707
#endif
3708 3709 3710 3711 3712 3713 3714 3715
	/*
	 * 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;

3716 3717 3718 3719 3720 3721
	/*
	 * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues.
	 * Grab a permanent reference to it, so that the normal code flow
	 * will not attempt to free it.
	 */
	cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
3722
	cfqd->oom_cfqq.ref++;
3723
	cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group);
3724

L
Linus Torvalds 已提交
3725 3726
	cfqd->queue = q;

3727 3728 3729 3730
	init_timer(&cfqd->idle_slice_timer);
	cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
	cfqd->idle_slice_timer.data = (unsigned long) cfqd;

3731
	INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
3732

L
Linus Torvalds 已提交
3733
	cfqd->cfq_quantum = cfq_quantum;
3734 3735
	cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
	cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
L
Linus Torvalds 已提交
3736 3737
	cfqd->cfq_back_max = cfq_back_max;
	cfqd->cfq_back_penalty = cfq_back_penalty;
3738 3739 3740 3741
	cfqd->cfq_slice[0] = cfq_slice_async;
	cfqd->cfq_slice[1] = cfq_slice_sync;
	cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
	cfqd->cfq_slice_idle = cfq_slice_idle;
3742
	cfqd->cfq_group_idle = cfq_group_idle;
3743
	cfqd->cfq_latency = 1;
3744
	cfqd->hw_tag = -1;
3745 3746 3747 3748
	/*
	 * we optimistically start assuming sync ops weren't delayed in last
	 * second, in order to have larger depth for async operations.
	 */
3749
	cfqd->last_delayed_sync = jiffies - HZ;
J
Jens Axboe 已提交
3750
	return cfqd;
L
Linus Torvalds 已提交
3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771
}

/*
 * sysfs parts below -->
 */
static ssize_t
cfq_var_show(unsigned int var, char *page)
{
	return sprintf(page, "%d\n", var);
}

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
3772
static ssize_t __FUNC(struct elevator_queue *e, char *page)		\
L
Linus Torvalds 已提交
3773
{									\
3774
	struct cfq_data *cfqd = e->elevator_data;			\
L
Linus Torvalds 已提交
3775 3776 3777 3778 3779 3780
	unsigned int __data = __VAR;					\
	if (__CONV)							\
		__data = jiffies_to_msecs(__data);			\
	return cfq_var_show(__data, (page));				\
}
SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
3781 3782
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);
3783 3784
SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
3785
SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
3786
SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1);
3787 3788 3789
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);
3790
SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
L
Linus Torvalds 已提交
3791 3792 3793
#undef SHOW_FUNCTION

#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)			\
J
Jens Axboe 已提交
3794
static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)	\
L
Linus Torvalds 已提交
3795
{									\
3796
	struct cfq_data *cfqd = e->elevator_data;			\
L
Linus Torvalds 已提交
3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809
	unsigned int __data;						\
	int ret = cfq_var_store(&__data, (page), count);		\
	if (__data < (MIN))						\
		__data = (MIN);						\
	else if (__data > (MAX))					\
		__data = (MAX);						\
	if (__CONV)							\
		*(__PTR) = msecs_to_jiffies(__data);			\
	else								\
		*(__PTR) = __data;					\
	return ret;							\
}
STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
3810 3811 3812 3813
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);
3814
STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
3815 3816
STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
		UINT_MAX, 0);
3817
STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
3818
STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1);
3819 3820
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);
3821 3822
STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
		UINT_MAX, 0);
3823
STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
L
Linus Torvalds 已提交
3824 3825
#undef STORE_FUNCTION

3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838
#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),
	CFQ_ATTR(slice_async),
	CFQ_ATTR(slice_async_rq),
	CFQ_ATTR(slice_idle),
3839
	CFQ_ATTR(group_idle),
3840
	CFQ_ATTR(low_latency),
3841
	__ATTR_NULL
L
Linus Torvalds 已提交
3842 3843 3844 3845 3846 3847 3848
};

static struct elevator_type iosched_cfq = {
	.ops = {
		.elevator_merge_fn = 		cfq_merge,
		.elevator_merged_fn =		cfq_merged_request,
		.elevator_merge_req_fn =	cfq_merged_requests,
3849
		.elevator_allow_merge_fn =	cfq_allow_merge,
D
Divyesh Shah 已提交
3850
		.elevator_bio_merged_fn =	cfq_bio_merged,
3851
		.elevator_dispatch_fn =		cfq_dispatch_requests,
L
Linus Torvalds 已提交
3852
		.elevator_add_req_fn =		cfq_insert_request,
3853
		.elevator_activate_req_fn =	cfq_activate_request,
L
Linus Torvalds 已提交
3854 3855
		.elevator_deactivate_req_fn =	cfq_deactivate_request,
		.elevator_completed_req_fn =	cfq_completed_request,
3856 3857
		.elevator_former_req_fn =	elv_rb_former_request,
		.elevator_latter_req_fn =	elv_rb_latter_request,
3858
		.elevator_init_icq_fn =		cfq_init_icq,
3859
		.elevator_exit_icq_fn =		cfq_exit_icq,
L
Linus Torvalds 已提交
3860 3861 3862 3863 3864 3865
		.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,
	},
3866 3867
	.icq_size	=	sizeof(struct cfq_io_cq),
	.icq_align	=	__alignof__(struct cfq_io_cq),
3868
	.elevator_attrs =	cfq_attrs,
3869
	.elevator_name	=	"cfq",
L
Linus Torvalds 已提交
3870 3871 3872
	.elevator_owner =	THIS_MODULE,
};

3873 3874 3875 3876 3877 3878
#ifdef CONFIG_CFQ_GROUP_IOSCHED
static struct blkio_policy_type blkio_policy_cfq = {
	.ops = {
		.blkio_unlink_group_fn =	cfq_unlink_blkio_group,
		.blkio_update_group_weight_fn =	cfq_update_blkio_group_weight,
	},
3879
	.plid = BLKIO_POLICY_PROP,
3880 3881 3882 3883 3884
};
#else
static struct blkio_policy_type blkio_policy_cfq;
#endif

L
Linus Torvalds 已提交
3885 3886
static int __init cfq_init(void)
{
3887 3888
	int ret;

3889 3890 3891 3892 3893 3894 3895 3896
	/*
	 * could be 0 on HZ < 1000 setups
	 */
	if (!cfq_slice_async)
		cfq_slice_async = 1;
	if (!cfq_slice_idle)
		cfq_slice_idle = 1;

3897 3898 3899 3900 3901 3902
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	if (!cfq_group_idle)
		cfq_group_idle = 1;
#else
		cfq_group_idle = 0;
#endif
3903 3904
	cfq_pool = KMEM_CACHE(cfq_queue, 0);
	if (!cfq_pool)
L
Linus Torvalds 已提交
3905 3906
		return -ENOMEM;

3907 3908 3909 3910 3911 3912
	ret = elv_register(&iosched_cfq);
	if (ret) {
		kmem_cache_destroy(cfq_pool);
		return ret;
	}

3913
	blkio_policy_register(&blkio_policy_cfq);
L
Linus Torvalds 已提交
3914

3915
	return 0;
L
Linus Torvalds 已提交
3916 3917 3918 3919
}

static void __exit cfq_exit(void)
{
3920
	blkio_policy_unregister(&blkio_policy_cfq);
L
Linus Torvalds 已提交
3921
	elv_unregister(&iosched_cfq);
3922
	kmem_cache_destroy(cfq_pool);
L
Linus Torvalds 已提交
3923 3924 3925 3926 3927 3928 3929 3930
}

module_init(cfq_init);
module_exit(cfq_exit);

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