cfq-iosched.c 83 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>
Al Viro's avatar
Al Viro committed
10 11
#include <linux/blkdev.h>
#include <linux/elevator.h>
Randy Dunlap's avatar
Randy Dunlap committed
12
#include <linux/jiffies.h>
Linus Torvalds's avatar
Linus Torvalds committed
13
#include <linux/rbtree.h>
14
#include <linux/ioprio.h>
15
#include <linux/blktrace_api.h>
16
#include "blk-cgroup.h"
Linus Torvalds's avatar
Linus Torvalds committed
17 18 19 20

/*
 * tunables
 */
21 22
/* max queue in one round of service */
static const int cfq_quantum = 4;
23
static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
24 25 26 27
/* 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;
28
static const int cfq_slice_sync = HZ / 10;
Jens Axboe's avatar
Jens Axboe committed
29
static int cfq_slice_async = HZ / 25;
30
static const int cfq_slice_async_rq = 2;
31
static int cfq_slice_idle = HZ / 125;
32 33
static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
static const int cfq_hist_divisor = 4;
34

35
/*
36
 * offset from end of service tree
37
 */
38
#define CFQ_IDLE_DELAY		(HZ / 5)
39 40 41 42 43 44

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

45 46 47 48 49 50
/*
 * Allow merged cfqqs to perform this amount of seeky I/O before
 * deciding to break the queues up again.
 */
#define CFQQ_COOP_TOUT		(HZ)

51
#define CFQ_SLICE_SCALE		(5)
52
#define CFQ_HW_QUEUE_MIN	(5)
53
#define CFQ_SERVICE_SHIFT       12
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)
Linus Torvalds's avatar
Linus Torvalds committed
58

59 60
static struct kmem_cache *cfq_pool;
static struct kmem_cache *cfq_ioc_pool;
Linus Torvalds's avatar
Linus Torvalds committed
61

62
static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
63
static struct completion *ioc_gone;
64
static DEFINE_SPINLOCK(ioc_gone_lock);
65

66 67 68 69
#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)

70
#define sample_valid(samples)	((samples) > 80)
71
#define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
72

73 74 75 76 77 78 79 80 81
/*
 * 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;
82
	unsigned count;
83
	u64 min_vdisktime;
84
	struct rb_node *active;
85
};
86
#define CFQ_RB_ROOT	(struct cfq_rb_root) { RB_ROOT, NULL, 0, 0, }
87

88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
/*
 * 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;

	unsigned long slice_end;
	long slice_resid;
	unsigned int slice_dispatch;

	/* 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;

130 131 132 133
	unsigned int seek_samples;
	u64 seek_total;
	sector_t seek_mean;
	sector_t last_request_pos;
134
	unsigned long seeky_start;
135

136
	pid_t pid;
Jeff Moyer's avatar
Jeff Moyer committed
137

138
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
139
	struct cfq_queue *new_cfqq;
140
	struct cfq_group *cfqg;
141 142
};

143
/*
144
 * First index in the service_trees.
145 146 147 148
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
149 150
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
151 152
};

153 154 155 156 157 158 159 160 161
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

162 163
/* This is per cgroup per device grouping structure */
struct cfq_group {
164 165 166 167 168
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
169
	unsigned int weight;
170 171 172 173 174
	bool on_st;

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

175 176 177 178 179 180 181
	/*
	 * 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;
};
182

183 184 185
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
186
struct cfq_data {
187
	struct request_queue *queue;
188 189
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
190
	struct cfq_group root_group;
191

192 193
	/*
	 * The priority currently being served
194
	 */
195
	enum wl_prio_t serving_prio;
196 197
	enum wl_type_t serving_type;
	unsigned long workload_expires;
198
	struct cfq_group *serving_group;
199
	bool noidle_tree_requires_idle;
200 201 202 203 204 205 206 207

	/*
	 * 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];

208
	unsigned int busy_queues;
209
	unsigned int busy_queues_avg[2];
210

211
	int rq_in_driver[2];
212
	int sync_flight;
213 214 215 216 217

	/*
	 * queue-depth detection
	 */
	int rq_queued;
218
	int hw_tag;
219 220 221 222 223 224 225 226
	/*
	 * 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
227

228 229 230 231
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
232
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
233

234 235 236
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

237 238 239 240 241
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
242

Jens Axboe's avatar
Jens Axboe committed
243
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
244 245 246 247 248

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
249
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
250 251
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
252 253 254
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
255
	unsigned int cfq_latency;
256 257

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
258

259 260 261 262
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
263 264

	unsigned long last_end_sync_rq;
Linus Torvalds's avatar
Linus Torvalds committed
265 266
};

267 268
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
269
					    enum wl_type_t type,
270 271
					    struct cfq_data *cfqd)
{
272 273 274
	if (!cfqg)
		return NULL;

275
	if (prio == IDLE_WORKLOAD)
276
		return &cfqg->service_tree_idle;
277

278
	return &cfqg->service_trees[prio][type];
279 280
}

Jens Axboe's avatar
Jens Axboe committed
281
enum cfqq_state_flags {
282 283
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
284
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
285 286 287 288
	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 */
289
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
290
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
291
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
292
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
Jens Axboe's avatar
Jens Axboe committed
293 294 295 296 297
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
298
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
299 300 301
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
302
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
303 304 305
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
306
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
307 308 309 310
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
311
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
312 313 314 315
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
316
CFQ_CFQQ_FNS(slice_new);
317
CFQ_CFQQ_FNS(sync);
318
CFQ_CFQQ_FNS(coop);
319
CFQ_CFQQ_FNS(deep);
Jens Axboe's avatar
Jens Axboe committed
320 321
#undef CFQ_CFQQ_FNS

322 323 324 325 326
#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
#define cfq_log(cfqd, fmt, args...)	\
	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)

327 328 329 330 331 332 333 334 335 336 337
/* 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) \


338 339 340 341 342 343 344 345 346
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;
}

347 348 349 350 351 352 353 354 355 356

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;
}

357 358
static inline int cfq_busy_queues_wl(enum wl_prio_t wl, struct cfq_data *cfqd)
{
359 360
	struct cfq_group *cfqg = &cfqd->root_group;

361
	if (wl == IDLE_WORKLOAD)
362
		return cfqg->service_tree_idle.count;
363

364 365 366
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
367 368
}

369
static void cfq_dispatch_insert(struct request_queue *, struct request *);
370
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
371
				       struct io_context *, gfp_t);
372
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
373 374
						struct io_context *);

375 376 377 378 379
static inline int rq_in_driver(struct cfq_data *cfqd)
{
	return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
}

380
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
381
					    bool is_sync)
382
{
383
	return cic->cfqq[is_sync];
384 385 386
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
387
				struct cfq_queue *cfqq, bool is_sync)
388
{
389
	cic->cfqq[is_sync] = cfqq;
390 391 392 393 394 395
}

/*
 * 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).
 */
396
static inline bool cfq_bio_sync(struct bio *bio)
397
{
398
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
399
}
Linus Torvalds's avatar
Linus Torvalds committed
400

Andrew Morton's avatar
Andrew Morton committed
401 402 403 404
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
405
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
406
{
407 408
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
409
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
410
	}
Andrew Morton's avatar
Andrew Morton committed
411 412
}

413
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
414 415 416
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

417
	return !cfqd->rq_queued;
Andrew Morton's avatar
Andrew Morton committed
418 419
}

420 421 422 423 424
/*
 * 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.
 */
425
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
426
				 unsigned short prio)
427
{
428
	const int base_slice = cfqd->cfq_slice[sync];
429

430 431 432 433
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
434

435 436 437 438
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);
439 440
}

441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
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);
}

486 487 488 489 490 491
/*
 * 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
 */

492 493
static inline unsigned cfq_get_avg_queues(struct cfq_data *cfqd, bool rt)
{
494 495 496
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
497
	unsigned busy = cfq_busy_queues_wl(rt, cfqd);
498 499 500 501 502 503 504 505

	min_q = min(cfqd->busy_queues_avg[rt], busy);
	max_q = max(cfqd->busy_queues_avg[rt], busy);
	cfqd->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
		cfq_hist_divisor;
	return cfqd->busy_queues_avg[rt];
}

506 507 508
static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
		/* interested queues (we consider only the ones with the same
		 * priority class) */
		unsigned iq = cfq_get_avg_queues(cfqd, cfq_class_rt(cfqq));
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
		if (expect_latency > cfq_target_latency) {
			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 */
			slice = max(slice * cfq_target_latency / expect_latency,
				    low_slice);
		}
	}
	cfqq->slice_end = jiffies + slice;
529
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
530 531 532 533 534 535 536
}

/*
 * 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.
 */
537
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
538 539 540 541 542 543 544 545 546
{
	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
547
/*
Jens Axboe's avatar
Jens Axboe committed
548
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
549
 * We choose the request that is closest to the head right now. Distance
550
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
551
 */
Jens Axboe's avatar
Jens Axboe committed
552
static struct request *
553
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
554
{
555
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
556
	unsigned long back_max;
557 558 559
#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
560

Jens Axboe's avatar
Jens Axboe committed
561 562 563 564
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
565

Jens Axboe's avatar
Jens Axboe committed
566 567 568 569
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
570 571 572 573
	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
574

575 576
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592

	/*
	 * 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
593
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
594 595 596 597 598 599

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
600
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
601 602

	/* Found required data */
603 604 605 606 607 608

	/*
	 * 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
609
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
610
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
611
			return rq1;
612
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
613
			return rq2;
614 615
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
616
				return rq1;
617
			else
Jens Axboe's avatar
Jens Axboe committed
618
				return rq2;
619
		}
Linus Torvalds's avatar
Linus Torvalds committed
620

621
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
622
		return rq1;
623
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
624 625
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
626 627 628 629 630 631 632 633
	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
634
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
635
		else
Jens Axboe's avatar
Jens Axboe committed
636
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
637 638 639
	}
}

640 641 642
/*
 * The below is leftmost cache rbtree addon
 */
643
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
644
{
645 646 647 648
	/* Service tree is empty */
	if (!root->count)
		return NULL;

649 650 651
	if (!root->left)
		root->left = rb_first(&root->rb);

652 653 654 655
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
656 657
}

658 659 660 661 662 663 664 665 666 667 668
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;
}

669 670 671 672 673 674
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

675 676 677 678
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
679
	rb_erase_init(n, &root->rb);
680
	--root->count;
681 682
}

Linus Torvalds's avatar
Linus Torvalds committed
683 684 685
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
686 687 688
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
689
{
690 691
	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
692
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
693

694
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
695 696

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
697
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
698

699
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
700
		next = rb_entry_rq(rbnext);
701 702 703
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
704
			next = rb_entry_rq(rbnext);
705
	}
Linus Torvalds's avatar
Linus Torvalds committed
706

707
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
708 709
}

710 711
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
712
{
713 714 715
	/*
	 * just an approximation, should be ok.
	 */
716
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
717
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
718 719
}

720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785
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;
}

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

786 787 788
	if (st->active == &cfqg->rb_node)
		st->active = NULL;

789 790
	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
791

792 793 794 795 796 797 798 799 800
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

	cfqg->on_st = false;
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
}

801
/*
802
 * The cfqd->service_trees holds all pending cfq_queue's that have
803 804 805
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
806
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
807
				 bool add_front)
808
{
809 810
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
811
	unsigned long rb_key;
812
	struct cfq_rb_root *service_tree;
813
	int left;
814

815 816
	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
						cfqq_type(cfqq), cfqd);
817 818
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
819
		parent = rb_last(&service_tree->rb);
820 821 822 823 824 825
		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) {
826 827 828 829 830 831
		/*
		 * 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.
		 */
832
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
833
		rb_key -= cfqq->slice_resid;
834
		cfqq->slice_resid = 0;
835 836
	} else {
		rb_key = -HZ;
837
		__cfqq = cfq_rb_first(service_tree);
838 839
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
Linus Torvalds's avatar
Linus Torvalds committed
840

841
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
842
		/*
843
		 * same position, nothing more to do
844
		 */
845 846
		if (rb_key == cfqq->rb_key &&
		    cfqq->service_tree == service_tree)
847
			return;
Linus Torvalds's avatar
Linus Torvalds committed
848

849 850
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
851
	}
852

853
	left = 1;
854
	parent = NULL;
855 856
	cfqq->service_tree = service_tree;
	p = &service_tree->rb.rb_node;
857
	while (*p) {
858
		struct rb_node **n;
859

860 861 862
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

863
		/*
864
		 * sort by key, that represents service time.
865
		 */
866
		if (time_before(rb_key, __cfqq->rb_key))
867
			n = &(*p)->rb_left;
868
		else {
869
			n = &(*p)->rb_right;
870
			left = 0;
871
		}
872 873

		p = n;
874 875
	}

876
	if (left)
877
		service_tree->left = &cfqq->rb_node;
878

879 880
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
881 882
	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
	service_tree->count++;
883
	cfq_group_service_tree_add(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
884 885
}

886
static struct cfq_queue *
887 888 889
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)
890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905
{
	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.
		 */
906
		if (sector > blk_rq_pos(cfqq->next_rq))
907
			n = &(*p)->rb_right;
908
		else if (sector < blk_rq_pos(cfqq->next_rq))
909 910 911 912
			n = &(*p)->rb_left;
		else
			break;
		p = n;
913
		cfqq = NULL;
914 915 916 917 918
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
919
	return cfqq;
920 921 922 923 924 925 926
}

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

927 928 929 930
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
931 932 933 934 935 936

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

937
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
938 939
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
940 941
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
942 943 944
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
945 946
}

947 948 949
/*
 * Update cfqq's position in the service tree.
 */
950
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
951 952 953 954
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
955
	if (cfq_cfqq_on_rr(cfqq)) {
956
		cfq_service_tree_add(cfqd, cfqq, 0);
957 958
		cfq_prio_tree_add(cfqd, cfqq);
	}
Jens Axboe's avatar
Jens Axboe committed
959 960
}

Linus Torvalds's avatar
Linus Torvalds committed
961 962
/*
 * add to busy list of queues for service, trying to be fair in ordering
963
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
964
 */
965
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
966
{
967
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
968 969
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
970 971
	cfqd->busy_queues++;

972
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
973 974
}

975 976 977 978
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
979
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
980
{
981
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
982 983
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
984

985 986 987 988
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
	}
989 990 991 992
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
993

994
	cfq_group_service_tree_del(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
995 996 997 998 999 1000 1001
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
}

/*
 * rb tree support functions
 */
1002
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
1003
{
Jens Axboe's avatar
Jens Axboe committed
1004 1005
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
1006

1007 1008
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
1009

Jens Axboe's avatar
Jens Axboe committed
1010
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
1011

1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022
	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;
		}
	}
Linus Torvalds's avatar
Linus Torvalds committed
1023 1024
}

Jens Axboe's avatar
Jens Axboe committed
1025
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
1026
{
Jens Axboe's avatar
Jens Axboe committed
1027
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds's avatar
Linus Torvalds committed
1028
	struct cfq_data *cfqd = cfqq->cfqd;
1029
	struct request *__alias, *prev;
Linus Torvalds's avatar
Linus Torvalds committed
1030

1031
	cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds's avatar
Linus Torvalds committed
1032 1033 1034 1035 1036

	/*
	 * looks a little odd, but the first insert might return an alias.
	 * if that happens, put the alias on the dispatch list
	 */
1037
	while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe's avatar
Jens Axboe committed
1038
		cfq_dispatch_insert(cfqd->queue, __alias);
1039 1040 1041

	if (!cfq_cfqq_on_rr(cfqq))
		cfq_add_cfqq_rr(cfqd, cfqq);
1042 1043 1044 1045

	/*
	 * check if this request is a better next-serve candidate
	 */
1046
	prev = cfqq->next_rq;
1047
	cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
1048 1049 1050 1051 1052 1053 1054

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