cfq-iosched.c 98.1 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6
/*
 *  CFQ, or complete fairness queueing, disk scheduler.
 *
 *  Based on ideas from a previously unfinished io
 *  scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
 *
7
 *  Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
Linus Torvalds's avatar
Linus Torvalds committed
8 9
 */
#include <linux/module.h>
10
#include <linux/slab.h>
Al Viro's avatar
Al Viro committed
11 12
#include <linux/blkdev.h>
#include <linux/elevator.h>
Randy Dunlap's avatar
Randy Dunlap committed
13
#include <linux/jiffies.h>
Linus Torvalds's avatar
Linus Torvalds committed
14
#include <linux/rbtree.h>
15
#include <linux/ioprio.h>
16
#include <linux/blktrace_api.h>
17
#include "blk.h"
18
#include "cfq.h"
Linus Torvalds's avatar
Linus Torvalds committed
19

20 21
static struct blkio_policy_type blkio_policy_cfq;

Linus Torvalds's avatar
Linus Torvalds committed
22 23 24
/*
 * tunables
 */
25
/* max queue in one round of service */
Shaohua Li's avatar
Shaohua Li committed
26
static const int cfq_quantum = 8;
27
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
28 29 30 31
/* 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;
32
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
33
static int cfq_slice_async = HZ / 25;
34
static const int cfq_slice_async_rq = 2;
35
static int cfq_slice_idle = HZ / 125;
36
static int cfq_group_idle = HZ / 125;
37 38
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
39

40
/*
41
 * offset from end of service tree
42
 */
43
#define CFQ_IDLE_DELAY		(HZ / 5)
44 45 46 47 48 49

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

50
#define CFQ_SLICE_SCALE		(5)
51
#define CFQ_HW_QUEUE_MIN	(5)
52
#define CFQ_SERVICE_SHIFT       12
53

54
#define CFQQ_SEEK_THR		(sector_t)(8 * 100)
55
#define CFQQ_CLOSE_THR		(sector_t)(8 * 1024)
56
#define CFQQ_SECT_THR_NONROT	(sector_t)(2 * 32)
57
#define CFQQ_SEEKY(cfqq)	(hweight32(cfqq->seek_history) > 32/8)
58

59 60 61
#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])
Linus Torvalds's avatar
Linus Torvalds committed
62

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

65 66 67 68
#define CFQ_PRIO_LISTS		IOPRIO_BE_NR
#define cfq_class_idle(cfqq)	((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
#define cfq_class_rt(cfqq)	((cfqq)->ioprio_class == IOPRIO_CLASS_RT)

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

72 73 74 75 76 77 78 79
struct cfq_ttime {
	unsigned long last_end_request;

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

80 81 82 83 84 85 86 87 88
/*
 * 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;
89
	unsigned count;
90
	unsigned total_weight;
91
	u64 min_vdisktime;
92
	struct cfq_ttime ttime;
93
};
94 95
#define CFQ_RB_ROOT	(struct cfq_rb_root) { .rb = RB_ROOT, \
			.ttime = {.last_end_request = jiffies,},}
96

97 98 99 100 101
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
102
	int ref;
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
	/* 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;

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

135 136
	/* pending priority requests */
	int prio_pending;
137 138 139 140 141
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;

	/* io prio of this group */
	unsigned short ioprio, org_ioprio;
142
	unsigned short ioprio_class;
143

144 145
	pid_t pid;

146
	u32 seek_history;
147 148
	sector_t last_request_pos;

149
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
150
	struct cfq_queue *new_cfqq;
151
	struct cfq_group *cfqg;
152 153
	/* Number of sectors dispatched from queue in single dispatch round */
	unsigned long nr_sectors;
154 155
};

156
/*
157
 * First index in the service_trees.
158 159 160 161
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
162 163
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
164
	CFQ_PRIO_NR,
165 166
};

167 168 169 170 171 172 173 174 175
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

176 177
/* This is per cgroup per device grouping structure */
struct cfq_group {
178 179 180 181 182
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
183
	unsigned int weight;
184 185
	unsigned int new_weight;
	bool needs_update;
186 187 188 189

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

190
	/*
191
	 * Per group busy queues average. Useful for workload slice calc. We
192 193 194 195 196 197 198 199 200 201 202
	 * 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.
203 204 205 206
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
207 208 209 210

	unsigned long saved_workload_slice;
	enum wl_type_t saved_workload;
	enum wl_prio_t saved_serving_prio;
211 212 213
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	struct hlist_node cfqd_node;
#endif
214 215
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;
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
 */
Linus Torvalds's avatar
Linus Torvalds committed
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;
Linus Torvalds's avatar
Linus Torvalds committed
268

269 270 271 272
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
273
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
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

Jens Axboe's avatar
Jens Axboe committed
284
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
285 286 287 288 289

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
290
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
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;
Linus Torvalds's avatar
Linus Torvalds committed
311 312
};

313 314 315 316 317 318 319 320 321 322
static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg)
{
	return blkg_to_pdata(blkg, &blkio_policy_cfq);
}

static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg)
{
	return pdata_to_blkg(cfqg, &blkio_policy_cfq);
}

323 324
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);

325 326
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
327
					    enum wl_type_t type)
328
{
329 330 331
	if (!cfqg)
		return NULL;

332
	if (prio == IDLE_WORKLOAD)
333
		return &cfqg->service_tree_idle;
334

335
	return &cfqg->service_trees[prio][type];
336 337
}

Jens Axboe's avatar
Jens Axboe committed
338
enum cfqq_state_flags {
339 340
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
341
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
342 343 344 345
	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 */
346
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
347
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
348
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
349
	CFQ_CFQQ_FLAG_split_coop,	/* shared cfqq will be splitted */
350
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
351
	CFQ_CFQQ_FLAG_wait_busy,	/* Waiting for next request */
Jens Axboe's avatar
Jens Axboe committed
352 353 354 355 356
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
357
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
358 359 360
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
361
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
362 363 364
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
365
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
366 367 368 369
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
370
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
371 372 373 374
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
375
CFQ_CFQQ_FNS(slice_new);
376
CFQ_CFQQ_FNS(sync);
377
CFQ_CFQQ_FNS(coop);
378
CFQ_CFQQ_FNS(split_coop);
379
CFQ_CFQQ_FNS(deep);
380
CFQ_CFQQ_FNS(wait_busy);
Jens Axboe's avatar
Jens Axboe committed
381 382
#undef CFQ_CFQQ_FNS

383
#ifdef CONFIG_CFQ_GROUP_IOSCHED
384 385 386
#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', \
387
			blkg_path(cfqg_to_blkg((cfqq)->cfqg)), ##args)
388 389 390

#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)				\
	blk_add_trace_msg((cfqd)->queue, "%s " fmt,			\
391
			blkg_path(cfqg_to_blkg((cfqg))), ##args)	\
392 393

#else
394 395
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
396
#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0)
397
#endif
398 399 400
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

401 402 403 404 405 406 407 408 409 410
/* 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) \

411 412 413 414 415 416 417 418 419 420 421 422
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;
}
423

424 425 426 427 428 429 430 431 432 433 434 435 436 437 438
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;
}

439 440 441 442 443 444 445 446 447
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;
}

448 449 450 451 452 453 454 455 456 457

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;
}

458 459 460
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
461 462
{
	if (wl == IDLE_WORKLOAD)
463
		return cfqg->service_tree_idle.count;
464

465 466 467
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
468 469
}

470 471 472 473 474 475 476
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;
}

477
static void cfq_dispatch_insert(struct request_queue *, struct request *);
478
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
479
				       struct io_context *, gfp_t);
480

481 482 483 484 485 486
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);
}

487 488 489 490 491 492 493 494
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;
}

495
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync)
496
{
497
	return cic->cfqq[is_sync];
498 499
}

500 501
static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq,
				bool is_sync)
502
{
503
	cic->cfqq[is_sync] = cfqq;
504 505
}

506
static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic)
507
{
508
	return cic->icq.q->elevator->elevator_data;
509 510
}

511 512 513 514
/*
 * 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).
 */
515
static inline bool cfq_bio_sync(struct bio *bio)
516
{
517
	return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC);
518
}
Linus Torvalds's avatar
Linus Torvalds committed
519

Andrew Morton's avatar
Andrew Morton committed
520 521 522 523
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
524
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
525
{
526 527
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
528
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
529
	}
Andrew Morton's avatar
Andrew Morton committed
530 531
}

532 533 534 535 536
/*
 * 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.
 */
537
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
538
				 unsigned short prio)
539
{
540
	const int base_slice = cfqd->cfq_slice[sync];
541

542 543 544 545
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
546

547 548 549 550
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);
551 552
}

553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585
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);
586 587
		st->min_vdisktime = max_vdisktime(st->min_vdisktime,
						  cfqg->vdisktime);
588 589 590
	}
}

591 592 593 594 595 596
/*
 * 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
 */

597 598
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
599
{
600 601 602
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
603
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
604

605 606 607
	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) /
608
		cfq_hist_divisor;
609 610 611 612 613 614 615 616 617
	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;
618 619
}

620
static inline unsigned
621
cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
622
{
623 624
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
625 626 627 628 629 630
		/*
		 * 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));
631 632
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
633 634 635
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
636 637 638 639 640 641 642
			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 */
643
			slice = max(slice * group_slice / expect_latency,
644 645 646
				    low_slice);
		}
	}
647 648 649 650 651 652
	return slice;
}

static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
653
	unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
654

655
	cfqq->slice_start = jiffies;
656
	cfqq->slice_end = jiffies + slice;
657
	cfqq->allocated_slice = slice;
658
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
659 660 661 662 663 664 665
}

/*
 * 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.
 */
666
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
667 668
{
	if (cfq_cfqq_slice_new(cfqq))
Shaohua Li's avatar
Shaohua Li committed
669
		return false;
670
	if (time_before(jiffies, cfqq->slice_end))
Shaohua Li's avatar
Shaohua Li committed
671
		return false;
672

Shaohua Li's avatar
Shaohua Li committed
673
	return true;
674 675
}

Linus Torvalds's avatar
Linus Torvalds committed
676
/*
Jens Axboe's avatar
Jens Axboe committed
677
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
678
 * We choose the request that is closest to the head right now. Distance
679
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
680
 */
Jens Axboe's avatar
Jens Axboe committed
681
static struct request *
682
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
683
{
684
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
685
	unsigned long back_max;
686 687 688
#define CFQ_RQ1_WRAP	0x01 /* request 1 wraps */
#define CFQ_RQ2_WRAP	0x02 /* request 2 wraps */
	unsigned wrap = 0; /* bit mask: requests behind the disk head? */
Linus Torvalds's avatar
Linus Torvalds committed
689

Jens Axboe's avatar
Jens Axboe committed
690 691 692 693
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
694

695 696 697
	if (rq_is_sync(rq1) != rq_is_sync(rq2))
		return rq_is_sync(rq1) ? rq1 : rq2;

698 699
	if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
		return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
700

701 702
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718

	/*
	 * 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
719
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
720 721 722 723 724 725

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
726
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
727 728

	/* Found required data */
729 730 731 732 733 734

	/*
	 * By doing switch() on the bit mask "wrap" we avoid having to
	 * check two variables for all permutations: --> faster!
	 */
	switch (wrap) {
Jens Axboe's avatar
Jens Axboe committed
735
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
736
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
737
			return rq1;
738
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
739
			return rq2;
740 741
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
742
				return rq1;
743
			else
Jens Axboe's avatar
Jens Axboe committed
744
				return rq2;
745
		}
Linus Torvalds's avatar
Linus Torvalds committed
746

747
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
748
		return rq1;
749
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
750 751
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
752 753 754 755 756 757 758 759
	default:
		/*
		 * Since both rqs are wrapped,
		 * start with the one that's further behind head
		 * (--> only *one* back seek required),
		 * since back seek takes more time than forward.
		 */
		if (s1 <= s2)
Jens Axboe's avatar
Jens Axboe committed
760
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
761
		else
Jens Axboe's avatar
Jens Axboe committed
762
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
763 764 765
	}
}

766 767 768
/*
 * The below is leftmost cache rbtree addon
 */
769
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
770
{
771 772 773 774
	/* Service tree is empty */
	if (!root->count)
		return NULL;

775 776 777
	if (!root->left)
		root->left = rb_first(&root->rb);

778 779 780 781
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
782 783
}

784 785 786 787 788 789 790 791 792 793 794
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;
}

795 796 797 798 799 800
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

801 802 803 804
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
805
	rb_erase_init(n, &root->rb);
806
	--root->count;
807 808
}

Linus Torvalds's avatar
Linus Torvalds committed
809 810 811
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
812 813 814
static struct request *
cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		  struct request *last)
Linus Torvalds's avatar
Linus Torvalds committed
815
{
816 817
	struct rb_node *rbnext = rb_next(&last->rb_node);
	struct rb_node *rbprev = rb_prev(&last->rb_node);
Jens Axboe's avatar
Jens Axboe committed
818
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
819

820
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
821 822

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
823
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
824

825
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
826
		next = rb_entry_rq(rbnext);
827 828 829
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
830
			next = rb_entry_rq(rbnext);
831
	}
Linus Torvalds's avatar
Linus Torvalds committed
832

833
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
834 835
}

836 837
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
838
{
839 840 841
	/*
	 * just an approximation, should be ok.
	 */
842
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
843
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
844 845
}

846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880
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
881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901
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)
902 903 904 905 906 907
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
	struct cfq_group *__cfqg;
	struct rb_node *n;

	cfqg->nr_cfqq++;
908
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
909 910 911 912 913
		return;

	/*
	 * Currently put the group at the end. Later implement something
	 * so that groups get lesser vtime based on their weights, so that
Lucas De Marchi's avatar
Lucas De Marchi committed
914
	 * if group does not loose all if it was not continuously backlogged.
915 916 917 918 919 920 921
	 */
	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;
922 923
	cfq_group_service_tree_add(st, cfqg);
}
924

925 926 927 928 929 930
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);
931 932 933
}

static void
934
cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
935 936 937 938 939
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
940

941 942 943 944
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

945
	cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
946
	cfq_group_service_tree_del(st, cfqg);
947
	cfqg->saved_workload_slice = 0;
948 949
	cfq_blkiocg_update_dequeue_stats(cfqg_to_blkg(cfqg),
					 &blkio_policy_cfq, 1);
950 951
}

952 953
static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
						unsigned int *unaccounted_time)
954
{
955
	unsigned int slice_used;
956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971

	/*
	 * 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;
972 973
		if (slice_used > cfqq->allocated_slice) {
			*unaccounted_time = slice_used - cfqq->allocated_slice;
974
			slice_used = cfqq->allocated_slice;
975 976 977 978
		}
		if (time_after(cfqq->slice_start, cfqq->dispatch_start))
			*unaccounted_time += cfqq->slice_start -
					cfqq->dispatch_start;
979 980 981 982 983 984
	}

	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
985
				struct cfq_queue *cfqq)
986 987
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
988
	unsigned int used_sl, charge, unaccounted_sl = 0;
989 990 991 992
	int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
			- cfqg->service_tree_idle.count;

	BUG_ON(nr_sync < 0);
993
	used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
994

995 996 997 998
	if (iops_mode(cfqd))
		charge = cfqq->slice_dispatch;
	else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
		charge = cfqq->allocated_slice;
999 1000

	/* Can't update vdisktime while group is on service tree */
1001
	cfq_group_service_tree_del(st, cfqg);
1002
	cfqg->vdisktime += cfq_scale_slice(charge, cfqg);
1003 1004
	/* If a new weight was requested, update now, off tree */
	cfq_group_service_tree_add(st, cfqg);
1005 1006 1007 1008 1009 1010 1011 1012 1013

	/* 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;
1014 1015 1016

	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
					st->min_vdisktime);
1017 1018 1019 1020
	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);
1021 1022 1023
	cfq_blkiocg_update_timeslice_used(cfqg_to_blkg(cfqg), &blkio_policy_cfq,
					  used_sl, unaccounted_sl);
	cfq_blkiocg_set_start_empty_time(cfqg_to_blkg(cfqg), &blkio_policy_cfq);
1024 1025
}

1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
/**
 * cfq_init_cfqg_base - initialize base part of a cfq_group
 * @cfqg: cfq_group to initialize
 *
 * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED
 * is enabled or not.
 */
static void cfq_init_cfqg_base(struct cfq_group *cfqg)
{
	struct cfq_rb_root *st;
	int i, j;

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

	cfqg->ttime.last_end_request = jiffies;
}

1045
#ifdef CONFIG_CFQ_GROUP_IOSCHED
1046 1047
static void cfq_update_blkio_group_weight(struct request_queue *q,
					  struct blkio_group *blkg,
Paul Bolle's avatar
Paul Bolle committed
1048
					  unsigned int weight)
1049
{
1050 1051
	struct cfq_group *cfqg = blkg_to_cfqg(blkg);

1052 1053
	cfqg->new_weight = weight;
	cfqg->needs_update = true;