cfq-iosched.c 83.7 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
	unsigned total_weight;
86
};
87
#define CFQ_RB_ROOT	(struct cfq_rb_root) { RB_ROOT, NULL, 0, 0, }
88

89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130
/*
 * 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;

131 132 133 134
	unsigned int seek_samples;
	u64 seek_total;
	sector_t seek_mean;
	sector_t last_request_pos;
135
	unsigned long seeky_start;
136

137
	pid_t pid;
Jeff Moyer's avatar
Jeff Moyer committed
138

139
	struct cfq_rb_root *service_tree;
Jeff Moyer's avatar
Jeff Moyer committed
140
	struct cfq_queue *new_cfqq;
141
	struct cfq_group *cfqg;
142 143
};

144
/*
145
 * First index in the service_trees.
146 147 148 149
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
150 151
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
152 153
};

154 155 156 157 158 159 160 161 162
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

163 164
/* This is per cgroup per device grouping structure */
struct cfq_group {
165 166 167 168 169
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
170
	unsigned int weight;
171 172 173 174 175
	bool on_st;

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

176 177
	/* Per group busy queus average. Useful for workload slice calc. */
	unsigned int busy_queues_avg[2];
178 179 180 181 182 183 184
	/*
	 * 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;
};
185

186 187 188
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
189
struct cfq_data {
190
	struct request_queue *queue;
191 192
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
193
	struct cfq_group root_group;
194 195
	/* Number of active cfq groups on group service tree */
	int nr_groups;
196

197 198
	/*
	 * The priority currently being served
199
	 */
200
	enum wl_prio_t serving_prio;
201 202
	enum wl_type_t serving_type;
	unsigned long workload_expires;
203
	struct cfq_group *serving_group;
204
	bool noidle_tree_requires_idle;
205 206 207 208 209 210 211 212

	/*
	 * 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];

213 214
	unsigned int busy_queues;

215
	int rq_in_driver[2];
216
	int sync_flight;
217 218 219 220 221

	/*
	 * queue-depth detection
	 */
	int rq_queued;
222
	int hw_tag;
223 224 225 226 227 228 229 230
	/*
	 * 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
231

232 233 234 235
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
236
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
237

238 239 240
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

241 242 243 244 245
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
246

Jens Axboe's avatar
Jens Axboe committed
247
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
248 249 250 251 252

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
253
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
254 255
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
256 257 258
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
259
	unsigned int cfq_latency;
260 261

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
262

263 264 265 266
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
267 268

	unsigned long last_end_sync_rq;
Linus Torvalds's avatar
Linus Torvalds committed
269 270
};

271 272
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
273
					    enum wl_type_t type,
274 275
					    struct cfq_data *cfqd)
{
276 277 278
	if (!cfqg)
		return NULL;

279
	if (prio == IDLE_WORKLOAD)
280
		return &cfqg->service_tree_idle;
281

282
	return &cfqg->service_trees[prio][type];
283 284
}

Jens Axboe's avatar
Jens Axboe committed
285
enum cfqq_state_flags {
286 287
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
288
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
289 290 291 292
	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 */
293
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
294
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
295
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
296
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
Jens Axboe's avatar
Jens Axboe committed
297 298 299 300 301
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_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 void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
306
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
307 308 309
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
310
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
311 312 313 314
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
315
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
316 317 318 319
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
320
CFQ_CFQQ_FNS(slice_new);
321
CFQ_CFQQ_FNS(sync);
322
CFQ_CFQQ_FNS(coop);
323
CFQ_CFQQ_FNS(deep);
Jens Axboe's avatar
Jens Axboe committed
324 325
#undef CFQ_CFQQ_FNS

326 327 328 329 330
#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)

331 332 333 334 335 336 337 338 339 340 341
/* 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) \


342 343 344 345 346 347 348 349 350
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;
}

351 352 353 354 355 356 357 358 359 360

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;
}

361 362 363
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
364 365
{
	if (wl == IDLE_WORKLOAD)
366
		return cfqg->service_tree_idle.count;
367

368 369 370
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
371 372
}

373
static void cfq_dispatch_insert(struct request_queue *, struct request *);
374
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
375
				       struct io_context *, gfp_t);
376
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
377 378
						struct io_context *);

379 380 381 382 383
static inline int rq_in_driver(struct cfq_data *cfqd)
{
	return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
}

384
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
385
					    bool is_sync)
386
{
387
	return cic->cfqq[is_sync];
388 389 390
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
391
				struct cfq_queue *cfqq, bool is_sync)
392
{
393
	cic->cfqq[is_sync] = cfqq;
394 395 396 397 398 399
}

/*
 * 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).
 */
400
static inline bool cfq_bio_sync(struct bio *bio)
401
{
402
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
403
}
Linus Torvalds's avatar
Linus Torvalds committed
404

Andrew Morton's avatar
Andrew Morton committed
405 406 407 408
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
409
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
410
{
411 412
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
413
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
414
	}
Andrew Morton's avatar
Andrew Morton committed
415 416
}

417
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
418 419 420
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

421
	return !cfqd->rq_queued;
Andrew Morton's avatar
Andrew Morton committed
422 423
}

424 425 426 427 428
/*
 * 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.
 */
429
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
430
				 unsigned short prio)
431
{
432
	const int base_slice = cfqd->cfq_slice[sync];
433

434 435 436 437
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
438

439 440 441 442
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);
443 444
}

445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
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);
}

490 491 492 493 494 495
/*
 * 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
 */

496 497
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
498
{
499 500 501
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
502
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
503

504 505 506
	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) /
507
		cfq_hist_divisor;
508 509 510 511 512 513 514 515 516
	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;
517 518
}

519 520 521
static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
522 523
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
524 525 526 527 528 529
		/*
		 * 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));
530 531
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
532 533 534
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
535 536 537 538 539 540 541
			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 */
542
			slice = max(slice * group_slice / expect_latency,
543 544 545 546
				    low_slice);
		}
	}
	cfqq->slice_end = jiffies + slice;
547
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
548 549 550 551 552 553 554
}

/*
 * 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.
 */
555
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
556 557 558 559 560 561 562 563 564
{
	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
565
/*
Jens Axboe's avatar
Jens Axboe committed
566
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
567
 * We choose the request that is closest to the head right now. Distance
568
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
569
 */
Jens Axboe's avatar
Jens Axboe committed
570
static struct request *
571
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
572
{
573
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
574
	unsigned long back_max;
575 576 577
#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
578

Jens Axboe's avatar
Jens Axboe committed
579 580 581 582
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
583

Jens Axboe's avatar
Jens Axboe committed
584 585 586 587
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
588 589 590 591
	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
592

593 594
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610

	/*
	 * 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
611
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
612 613 614 615 616 617

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
618
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
619 620

	/* Found required data */
621 622 623 624 625 626

	/*
	 * 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
627
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
628
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
629
			return rq1;
630
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
631
			return rq2;
632 633
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
634
				return rq1;
635
			else
Jens Axboe's avatar
Jens Axboe committed
636
				return rq2;
637
		}
Linus Torvalds's avatar
Linus Torvalds committed
638

639
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
640
		return rq1;
641
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
642 643
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
644 645 646 647 648 649 650 651
	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
652
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
653
		else
Jens Axboe's avatar
Jens Axboe committed
654
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
655 656 657
	}
}

658 659 660
/*
 * The below is leftmost cache rbtree addon
 */
661
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
662
{
663 664 665 666
	/* Service tree is empty */
	if (!root->count)
		return NULL;

667 668 669
	if (!root->left)
		root->left = rb_first(&root->rb);

670 671 672 673
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
674 675
}

676 677 678 679 680 681 682 683 684 685 686
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;
}

687 688 689 690 691 692
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

693 694 695 696
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
697
	rb_erase_init(n, &root->rb);
698
	--root->count;
699 700
}

Linus Torvalds's avatar
Linus Torvalds committed
701 702 703
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
704 705 706
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
707
{
708 709
	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
710
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
711

712
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
713 714

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
715
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
716

717
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
718
		next = rb_entry_rq(rbnext);
719 720 721
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
722
			next = rb_entry_rq(rbnext);
723
	}
Linus Torvalds's avatar
Linus Torvalds committed
724

725
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
726 727
}

728 729
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
730
{
731 732 733
	/*
	 * just an approximation, should be ok.
	 */
734
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
735
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
736 737
}

738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796
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;
797 798
	cfqd->nr_groups++;
	st->total_weight += cfqg->weight;
799 800 801 802 803 804 805
}

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

806 807 808
	if (st->active == &cfqg->rb_node)
		st->active = NULL;

809 810
	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
811

812 813 814 815 816
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

	cfqg->on_st = false;
817 818
	cfqd->nr_groups--;
	st->total_weight -= cfqg->weight;
819 820 821 822
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
}

823
/*
824
 * The cfqd->service_trees holds all pending cfq_queue's that have
825 826 827
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
828
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
829
				 bool add_front)
830
{
831 832
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
833
	unsigned long rb_key;
834
	struct cfq_rb_root *service_tree;
835
	int left;
836

837 838
	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
						cfqq_type(cfqq), cfqd);
839 840
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
841
		parent = rb_last(&service_tree->rb);
842 843 844 845 846 847
		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) {
848 849 850 851 852 853
		/*
		 * 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.
		 */
854
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
855
		rb_key -= cfqq->slice_resid;
856
		cfqq->slice_resid = 0;
857 858
	} else {
		rb_key = -HZ;
859
		__cfqq = cfq_rb_first(service_tree);
860 861
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
Linus Torvalds's avatar
Linus Torvalds committed
862

863
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
864
		/*
865
		 * same position, nothing more to do
866
		 */
867 868
		if (rb_key == cfqq->rb_key &&
		    cfqq->service_tree == service_tree)
869
			return;
Linus Torvalds's avatar
Linus Torvalds committed
870

871 872
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
873
	}
874

875
	left = 1;
876
	parent = NULL;
877 878
	cfqq->service_tree = service_tree;
	p = &service_tree->rb.rb_node;
879
	while (*p) {
880
		struct rb_node **n;
881

882 883 884
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

885
		/*
886
		 * sort by key, that represents service time.
887
		 */
888
		if (time_before(rb_key, __cfqq->rb_key))
889
			n = &(*p)->rb_left;
890
		else {
891
			n = &(*p)->rb_right;
892
			left = 0;
893
		}
894 895

		p = n;
896 897
	}

898
	if (left)
899
		service_tree->left = &cfqq->rb_node;
900

901 902
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
903 904
	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
	service_tree->count++;
905
	cfq_group_service_tree_add(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
906 907
}

908
static struct cfq_queue *
909 910 911
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)
912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927
{
	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.
		 */
928
		if (sector > blk_rq_pos(cfqq->next_rq))
929
			n = &(*p)->rb_right;
930
		else if (sector < blk_rq_pos(cfqq->next_rq))
931 932 933 934
			n = &(*p)->rb_left;
		else
			break;
		p = n;
935
		cfqq = NULL;
936 937 938 939 940
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
941
	return cfqq;
942 943 944 945 946 947 948
}

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

949 950 951 952
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
953 954 955 956 957 958

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

959
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
960 961
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
962 963
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
964 965 966
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
967 968
}

969 970 971
/*
 * Update cfqq's position in the service tree.
 */
972
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
973 974 975 976
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
977
	if (cfq_cfqq_on_rr(cfqq)) {
978
		cfq_service_tree_add(cfqd, cfqq, 0);
979 980
		cfq_prio_tree_add(cfqd, cfqq);
	}
Jens Axboe's avatar
Jens Axboe committed
981 982
}

Linus Torvalds's avatar
Linus Torvalds committed
983 984
/*
 * add to busy list of queues for service, trying to be fair in ordering
985
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
986
 */
987
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
988
{
989
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
990 991
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
992 993
	cfqd->busy_queues++;

994
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
995 996
}

997 998 999 1000
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
1001
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
1002
{
1003
	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe's avatar
Jens Axboe committed
1004 1005
	BUG_ON(!cfq_cfqq_on_rr(cfqq));
	cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
1006

1007 1008 1009 1010
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
	}
1011 1012 1013 1014
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
1015

1016
	cfq_group_service_tree_del(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
1017 1018 1019 1020 1021 1022 1023
	BUG_ON(!cfqd->busy_queues);
	cfqd->busy_queues--;
}

/*
 * rb tree support functions
 */
1024
static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
1025
{
Jens Axboe's avatar
Jens Axboe committed
1026 1027
	struct cfq_queue *cfqq = RQ_CFQQ(rq);
	const int sync = rq_is_sync(rq);
Linus Torvalds's avatar
Linus Torvalds committed
1028

1029 1030
	BUG_ON(!cfqq->queued[sync]);
	cfqq->queued[sync]--;
Linus Torvalds's avatar
Linus Torvalds committed
1031

Jens Axboe's avatar
Jens Axboe committed
1032
	elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds's avatar
Linus Torvalds committed
1033

1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
	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
1045 1046
}

Jens Axboe's avatar
Jens Axboe committed
1047
static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds's avatar