cfq-iosched.c 100 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-cgroup.h"
Linus Torvalds's avatar
Linus Torvalds committed
18 19 20 21

/*
 * tunables
 */
22
/* max queue in one round of service */
Shaohua Li's avatar
Shaohua Li committed
23
static const int cfq_quantum = 8;
24
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
25 26 27 28
/* maximum backwards seek, in KiB */
static const int cfq_back_max = 16 * 1024;
/* penalty of a backwards seek */
static const int cfq_back_penalty = 2;
29
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
30
static int cfq_slice_async = HZ / 25;
31
static const int cfq_slice_async_rq = 2;
32
static int cfq_slice_idle = HZ / 125;
33 34
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
35

36
/*
37
 * offset from end of service tree
38
 */
39
#define CFQ_IDLE_DELAY		(HZ / 5)
40 41 42 43 44 45

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

46
#define CFQ_SLICE_SCALE		(5)
47
#define CFQ_HW_QUEUE_MIN	(5)
48
#define CFQ_SERVICE_SHIFT       12
49

50
#define CFQQ_SEEK_THR		(sector_t)(8 * 100)
51
#define CFQQ_CLOSE_THR		(sector_t)(8 * 1024)
52
#define CFQQ_SECT_THR_NONROT	(sector_t)(2 * 32)
53
#define CFQQ_SEEKY(cfqq)	(hweight32(cfqq->seek_history) > 32/8)
54

55 56
#define RQ_CIC(rq)		\
	((struct cfq_io_context *) (rq)->elevator_private)
57
#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
58
#define RQ_CFQG(rq)		(struct cfq_group *) ((rq)->elevator_private3)
Linus Torvalds's avatar
Linus Torvalds committed
59

60 61
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
62

63
static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
64
static struct completion *ioc_gone;
65
static DEFINE_SPINLOCK(ioc_gone_lock);
66

67 68 69 70
#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)

71
#define sample_valid(samples)	((samples) > 80)
72
#define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
73

74 75 76 77 78 79 80 81 82
/*
 * 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;
83
	unsigned count;
84
	unsigned total_weight;
85
	u64 min_vdisktime;
86
	struct rb_node *active;
87
};
88 89
#define CFQ_RB_ROOT	(struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \
			.count = 0, .min_vdisktime = 0, }
90

91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119
/*
 * Per process-grouping structure
 */
struct cfq_queue {
	/* reference count */
	atomic_t ref;
	/* 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;

120 121
	/* time when queue got scheduled in to dispatch first request. */
	unsigned long dispatch_start;
122
	unsigned int allocated_slice;
123
	unsigned int slice_dispatch;
124 125
	/* time when first request from queue completed and slice started. */
	unsigned long slice_start;
126 127 128 129 130 131 132 133 134 135 136 137
	unsigned long slice_end;
	long slice_resid;

	/* pending metadata requests */
	int meta_pending;
	/* number of requests that are on the dispatch list or inside driver */
	int dispatched;

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

138 139
	pid_t pid;

140
	u32 seek_history;
141 142
	sector_t last_request_pos;

143
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
144
	struct cfq_queue *new_cfqq;
145
	struct cfq_group *cfqg;
146
	struct cfq_group *orig_cfqg;
147 148
};

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

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

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

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

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

181 182
	/* Per group busy queus average. Useful for workload slice calc. */
	unsigned int busy_queues_avg[2];
183 184 185 186 187 188
	/*
	 * rr lists of queues with requests, onle rr for each priority class.
	 * Counts are embedded in the cfq_rb_root
	 */
	struct cfq_rb_root service_trees[2][3];
	struct cfq_rb_root service_tree_idle;
189 190 191 192

	unsigned long saved_workload_slice;
	enum wl_type_t saved_workload;
	enum wl_prio_t saved_serving_prio;
193 194 195
	struct blkio_group blkg;
#ifdef CONFIG_CFQ_GROUP_IOSCHED
	struct hlist_node cfqd_node;
196
	atomic_t ref;
197
#endif
198
};
199

200 201 202
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
203
struct cfq_data {
204
	struct request_queue *queue;
205 206
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
207
	struct cfq_group root_group;
208

209 210
	/*
	 * The priority currently being served
211
	 */
212
	enum wl_prio_t serving_prio;
213 214
	enum wl_type_t serving_type;
	unsigned long workload_expires;
215
	struct cfq_group *serving_group;
216
	bool noidle_tree_requires_idle;
217 218 219 220 221 222 223 224

	/*
	 * 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];

225 226
	unsigned int busy_queues;

227 228
	int rq_in_driver;
	int rq_in_flight[2];
229 230 231 232 233

	/*
	 * queue-depth detection
	 */
	int rq_queued;
234
	int hw_tag;
235 236 237 238 239 240 241 242
	/*
	 * 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
243

244 245 246 247
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
248
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
249

250 251 252
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

253 254 255 256 257
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
258

Jens Axboe's avatar
Jens Axboe committed
259
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
260 261 262 263 264

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
265
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
266 267
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
268 269 270
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
271
	unsigned int cfq_latency;
272
	unsigned int cfq_group_isolation;
273 274

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
275

276 277 278 279
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
280

281
	unsigned long last_delayed_sync;
282 283 284

	/* List of cfq groups being managed on this device*/
	struct hlist_head cfqg_list;
285
	struct rcu_head rcu;
Linus Torvalds's avatar
Linus Torvalds committed
286 287
};

288 289
static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);

290 291
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
292
					    enum wl_type_t type)
293
{
294 295 296
	if (!cfqg)
		return NULL;

297
	if (prio == IDLE_WORKLOAD)
298
		return &cfqg->service_tree_idle;
299

300
	return &cfqg->service_trees[prio][type];
301 302
}

Jens Axboe's avatar
Jens Axboe committed
303
enum cfqq_state_flags {
304 305
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
306
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
307 308 309 310
	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 */
311
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
312
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
313
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
314
	CFQ_CFQQ_FLAG_split_coop,	/* shared cfqq will be splitted */
315
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
316
	CFQ_CFQQ_FLAG_wait_busy,	/* Waiting for next request */
Jens Axboe's avatar
Jens Axboe committed
317 318 319 320 321
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
322
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
323 324 325
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
326
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
327 328 329
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
330
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
331 332 333 334
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
335
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
336 337 338 339
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
340
CFQ_CFQQ_FNS(slice_new);
341
CFQ_CFQQ_FNS(sync);
342
CFQ_CFQQ_FNS(coop);
343
CFQ_CFQQ_FNS(split_coop);
344
CFQ_CFQQ_FNS(deep);
345
CFQ_CFQQ_FNS(wait_busy);
Jens Axboe's avatar
Jens Axboe committed
346 347
#undef CFQ_CFQQ_FNS

348
#ifdef CONFIG_CFQ_GROUP_IOSCHED
349 350 351 352 353 354 355 356 357 358
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \
			cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
			blkg_path(&(cfqq)->cfqg->blkg), ##args);

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

#else
359 360
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
361 362
#define cfq_log_cfqg(cfqd, cfqg, fmt, args...)		do {} while (0);
#endif
363 364 365
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

366 367 368 369 370 371 372 373 374 375 376
/* 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) \


377 378 379 380 381 382 383 384 385
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;
}

386 387 388 389 390 391 392 393 394 395

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;
}

396 397 398
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
399 400
{
	if (wl == IDLE_WORKLOAD)
401
		return cfqg->service_tree_idle.count;
402

403 404 405
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
406 407
}

408 409 410 411 412 413 414
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;
}

415
static void cfq_dispatch_insert(struct request_queue *, struct request *);
416
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
417
				       struct io_context *, gfp_t);
418
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
419 420 421
						struct io_context *);

static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
422
					    bool is_sync)
423
{
424
	return cic->cfqq[is_sync];
425 426 427
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
428
				struct cfq_queue *cfqq, bool is_sync)
429
{
430
	cic->cfqq[is_sync] = cfqq;
431 432 433 434 435 436
}

/*
 * 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).
 */
437
static inline bool cfq_bio_sync(struct bio *bio)
438
{
439
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
440
}
Linus Torvalds's avatar
Linus Torvalds committed
441

Andrew Morton's avatar
Andrew Morton committed
442 443 444 445
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
446
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
447
{
448 449
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
450
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
451
	}
Andrew Morton's avatar
Andrew Morton committed
452 453
}

454
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
455 456 457
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

458
	return !cfqd->rq_queued;
Andrew Morton's avatar
Andrew Morton committed
459 460
}

461 462 463 464 465
/*
 * 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.
 */
466
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
467
				 unsigned short prio)
468
{
469
	const int base_slice = cfqd->cfq_slice[sync];
470

471 472 473 474
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
475

476 477 478 479
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);
480 481
}

482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg)
{
	u64 d = delta << CFQ_SERVICE_SHIFT;

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

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

	return min_vdisktime;
}

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

	return min_vdisktime;
}

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

	if (st->active) {
		cfqg = rb_entry_cfqg(st->active);
		vdisktime = cfqg->vdisktime;
	}

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

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

527 528 529 530 531 532
/*
 * 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
 */

533 534
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
535
{
536 537 538
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
539
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
540

541 542 543
	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) /
544
		cfq_hist_divisor;
545 546 547 548 549 550 551 552 553
	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;
554 555
}

556 557 558
static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
559 560
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
561 562 563 564 565 566
		/*
		 * 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));
567 568
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
569 570 571
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
572 573 574 575 576 577 578
			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 */
579
			slice = max(slice * group_slice / expect_latency,
580 581 582
				    low_slice);
		}
	}
583
	cfqq->slice_start = jiffies;
584
	cfqq->slice_end = jiffies + slice;
585
	cfqq->allocated_slice = slice;
586
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
587 588 589 590 591 592 593
}

/*
 * 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.
 */
594
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
595 596 597 598 599 600 601 602 603
{
	if (cfq_cfqq_slice_new(cfqq))
		return 0;
	if (time_before(jiffies, cfqq->slice_end))
		return 0;

	return 1;
}

Linus Torvalds's avatar
Linus Torvalds committed
604
/*
Jens Axboe's avatar
Jens Axboe committed
605
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
606
 * We choose the request that is closest to the head right now. Distance
607
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
608
 */
Jens Axboe's avatar
Jens Axboe committed
609
static struct request *
610
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
611
{
612
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
613
	unsigned long back_max;
614 615 616
#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
617

Jens Axboe's avatar
Jens Axboe committed
618 619 620 621
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
622

Jens Axboe's avatar
Jens Axboe committed
623 624 625 626
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
627 628 629 630
	if (rq_is_meta(rq1) && !rq_is_meta(rq2))
		return rq1;
	else if (rq_is_meta(rq2) && !rq_is_meta(rq1))
		return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
631

632 633
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649

	/*
	 * 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
650
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
651 652 653 654 655 656

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
657
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
658 659

	/* Found required data */
660 661 662 663 664 665

	/*
	 * 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
666
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
667
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
668
			return rq1;
669
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
670
			return rq2;
671 672
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
673
				return rq1;
674
			else
Jens Axboe's avatar
Jens Axboe committed
675
				return rq2;
676
		}
Linus Torvalds's avatar
Linus Torvalds committed
677

678
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
679
		return rq1;
680
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
681 682
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
683 684 685 686 687 688 689 690
	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
691
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
692
		else
Jens Axboe's avatar
Jens Axboe committed
693
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
694 695 696
	}
}

697 698 699
/*
 * The below is leftmost cache rbtree addon
 */
700
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
701
{
702 703 704 705
	/* Service tree is empty */
	if (!root->count)
		return NULL;

706 707 708
	if (!root->left)
		root->left = rb_first(&root->rb);

709 710 711 712
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
713 714
}

715 716 717 718 719 720 721 722 723 724 725
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;
}

726 727 728 729 730 731
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

732 733 734 735
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
736
	rb_erase_init(n, &root->rb);
737
	--root->count;
738 739
}

Linus Torvalds's avatar
Linus Torvalds committed
740 741 742
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
743 744 745
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
746
{
747 748
	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
749
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
750

751
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
752 753

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
754
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
755

756
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
757
		next = rb_entry_rq(rbnext);
758 759 760
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
761
			next = rb_entry_rq(rbnext);
762
	}
Linus Torvalds's avatar
Linus Torvalds committed
763

764
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
765 766
}

767 768
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
769
{
770 771 772
	/*
	 * just an approximation, should be ok.
	 */
773
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
774
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
775 776
}

777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835
static inline s64
cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
{
	return cfqg->vdisktime - st->min_vdisktime;
}

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

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

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

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

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

static void
cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
	struct cfq_group *__cfqg;
	struct rb_node *n;

	cfqg->nr_cfqq++;
	if (cfqg->on_st)
		return;

	/*
	 * Currently put the group at the end. Later implement something
	 * so that groups get lesser vtime based on their weights, so that
	 * if group does not loose all if it was not continously backlogged.
	 */
	n = rb_last(&st->rb);
	if (n) {
		__cfqg = rb_entry_cfqg(n);
		cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY;
	} else
		cfqg->vdisktime = st->min_vdisktime;

	__cfq_group_service_tree_add(st, cfqg);
	cfqg->on_st = true;
836
	st->total_weight += cfqg->weight;
837 838 839 840 841 842 843
}

static void
cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;

844 845 846
	if (st->active == &cfqg->rb_node)
		st->active = NULL;

847 848
	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
849

850 851 852 853
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

854
	cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
855
	cfqg->on_st = false;
856
	st->total_weight -= cfqg->weight;
857 858
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
859
	cfqg->saved_workload_slice = 0;
860
	blkiocg_update_dequeue_stats(&cfqg->blkg, 1);
861 862 863 864
}

static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq)
{
865
	unsigned int slice_used;
866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881

	/*
	 * 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;
882 883
		if (slice_used > cfqq->allocated_slice)
			slice_used = cfqq->allocated_slice;
884 885
	}

886
	cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u", slice_used);
887 888 889 890
	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
891
				struct cfq_queue *cfqq)
892 893
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
894 895 896 897 898 899
	unsigned int used_sl, charge_sl;
	int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
			- cfqg->service_tree_idle.count;

	BUG_ON(nr_sync < 0);
	used_sl = charge_sl = cfq_cfqq_slice_usage(cfqq);
900

901 902
	if (!cfq_cfqq_sync(cfqq) && !nr_sync)
		charge_sl = cfqq->allocated_slice;
903 904 905

	/* Can't update vdisktime while group is on service tree */
	cfq_rb_erase(&cfqg->rb_node, st);
906
	cfqg->vdisktime += cfq_scale_slice(charge_sl, cfqg);
907 908 909 910 911 912 913 914 915 916
	__cfq_group_service_tree_add(st, cfqg);

	/* This group is being expired. Save the context */
	if (time_after(cfqd->workload_expires, jiffies)) {
		cfqg->saved_workload_slice = cfqd->workload_expires
						- jiffies;
		cfqg->saved_workload = cfqd->serving_type;
		cfqg->saved_serving_prio = cfqd->serving_prio;
	} else
		cfqg->saved_workload_slice = 0;
917 918 919

	cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
					st->min_vdisktime);
920
	blkiocg_update_timeslice_used(&cfqg->blkg, used_sl);
921
	blkiocg_set_start_empty_time(&cfqg->blkg);
922 923
}

924 925 926 927 928 929 930 931
#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;
}

932 933 934 935 936 937
void
cfq_update_blkio_group_weight(struct blkio_group *blkg, unsigned int weight)
{
	cfqg_of_blkg(blkg)->weight = weight;
}

938 939 940 941 942 943 944 945
static struct cfq_group *
cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create)
{
	struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
	struct cfq_group *cfqg = NULL;
	void *key = cfqd;
	int i, j;
	struct cfq_rb_root *st;
946 947
	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
	unsigned int major, minor;
948 949

	cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
950 951 952 953 954
	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
		cfqg->blkg.dev = MKDEV(major, minor);
		goto done;
	}
955 956 957 958 959 960 961 962 963 964 965
	if (cfqg || !create)
		goto done;

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

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

966 967 968 969 970 971 972 973
	/*
	 * 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.
	 */
	atomic_set(&cfqg->ref, 1);

974
	/* Add group onto cgroup list */
975 976 977
	sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
	blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
					MKDEV(major, minor));
978
	cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004

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

done:
	return cfqg;
}

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

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

1005 1006 1007 1008 1009 1010
static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
{
	atomic_inc(&cfqg->ref);
	return cfqg;
}

1011 1012 1013 1014 1015 1016 1017
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;
1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062
	/* cfqq reference on cfqg */
	atomic_inc(&cfqq->cfqg->ref);
}

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

	BUG_ON(atomic_read(&cfqg->ref) <= 0);
	if (!atomic_dec_and_test(&cfqg->ref))
		return;
	for_each_cfqg_st(cfqg, i, j, st)
		BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL);
	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);

	/*
	 * 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.
		 */
		if (!blkiocg_del_blkio_group(&cfqg->blkg))
			cfq_destroy_cfqg(cfqd, cfqg);
	}
1063
}
1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088

/*
 * 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.
 */
void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg)
{
	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);
}

1089 1090 1091 1092 1093
#else /* GROUP_IOSCHED */
static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
{
	return &cfqd->root_group;
}
1094 1095 1096

static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
{
1097
	return cfqg;
1098 1099
}

1100 1101 1102 1103 1104
static inline void
cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
	cfqq->cfqg = cfqg;
}

1105 1106 1107
static void cfq_release_cfq_groups(struct cfq_data *cfqd) {}
static inline void cfq_put_cfqg(struct cfq_group *cfqg) {}

1108 1109
#endif /* GROUP_IOSCHED */

1110
/*
1111
 * The cfqd->service_trees holds all pending cfq_queue's that have
1112 1113 1114
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
1115
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1116
				 bool add_front)
1117
{
1118 1119
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
1120
	unsigned long rb_key;
1121
	struct cfq_rb_root *service_tree;
1122
	int left;
1123
	int new_