cfq-iosched.c 86.1 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6
/*
 *  CFQ, or complete fairness queueing, disk scheduler.
 *
 *  Based on ideas from a previously unfinished io
 *  scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
 *
7
 *  Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
Linus Torvalds's avatar
Linus Torvalds committed
8 9
 */
#include <linux/module.h>
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
/*
 * 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;

118 119 120 121
	/* time when queue got scheduled in to dispatch first request. */
	unsigned long dispatch_start;
	/* time when first request from queue completed and slice started. */
	unsigned long slice_start;
122 123 124 125 126 127 128 129 130 131 132 133 134
	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;

135 136 137 138
	unsigned int seek_samples;
	u64 seek_total;
	sector_t seek_mean;
	sector_t last_request_pos;
139
	unsigned long seeky_start;
140

141
	pid_t pid;
Jeff Moyer's avatar
Jeff Moyer committed
142

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;
146 147
};

148
/*
149
 * First index in the service_trees.
150 151 152 153
 * IDLE is handled separately, so it has negative index
 */
enum wl_prio_t {
	BE_WORKLOAD = 0,
154 155
	RT_WORKLOAD = 1,
	IDLE_WORKLOAD = 2,
156 157
};

158 159 160 161 162 163 164 165 166
/*
 * Second index in the service_trees.
 */
enum wl_type_t {
	ASYNC_WORKLOAD = 0,
	SYNC_NOIDLE_WORKLOAD = 1,
	SYNC_WORKLOAD = 2
};

167 168
/* This is per cgroup per device grouping structure */
struct cfq_group {
169 170 171 172 173
	/* group service_tree member */
	struct rb_node rb_node;

	/* group service_tree key */
	u64 vdisktime;
174
	unsigned int weight;
175 176 177 178 179
	bool on_st;

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

180 181
	/* Per group busy queus average. Useful for workload slice calc. */
	unsigned int busy_queues_avg[2];
182 183 184 185 186 187
	/*
	 * 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;
188 189 190 191

	unsigned long saved_workload_slice;
	enum wl_type_t saved_workload;
	enum wl_prio_t saved_serving_prio;
192
};
193

194 195 196
/*
 * Per block device queue structure
 */
Linus Torvalds's avatar
Linus Torvalds committed
197
struct cfq_data {
198
	struct request_queue *queue;
199 200
	/* Root service tree for cfq_groups */
	struct cfq_rb_root grp_service_tree;
201
	struct cfq_group root_group;
202 203
	/* Number of active cfq groups on group service tree */
	int nr_groups;
204

205 206
	/*
	 * The priority currently being served
207
	 */
208
	enum wl_prio_t serving_prio;
209 210
	enum wl_type_t serving_type;
	unsigned long workload_expires;
211
	struct cfq_group *serving_group;
212
	bool noidle_tree_requires_idle;
213 214 215 216 217 218 219 220

	/*
	 * 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];

221 222
	unsigned int busy_queues;

223
	int rq_in_driver[2];
224
	int sync_flight;
225 226 227 228 229

	/*
	 * queue-depth detection
	 */
	int rq_queued;
230
	int hw_tag;
231 232 233 234 235 236 237 238
	/*
	 * 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
239

240 241 242 243
	/*
	 * idle window management
	 */
	struct timer_list idle_slice_timer;
244
	struct work_struct unplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
245

246 247 248
	struct cfq_queue *active_queue;
	struct cfq_io_context *active_cic;

249 250 251 252 253
	/*
	 * async queue for each priority case
	 */
	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
	struct cfq_queue *async_idle_cfqq;
254

Jens Axboe's avatar
Jens Axboe committed
255
	sector_t last_position;
Linus Torvalds's avatar
Linus Torvalds committed
256 257 258 259 260

	/*
	 * tunables, see top of file
	 */
	unsigned int cfq_quantum;
261
	unsigned int cfq_fifo_expire[2];
Linus Torvalds's avatar
Linus Torvalds committed
262 263
	unsigned int cfq_back_penalty;
	unsigned int cfq_back_max;
264 265 266
	unsigned int cfq_slice[2];
	unsigned int cfq_slice_async_rq;
	unsigned int cfq_slice_idle;
267
	unsigned int cfq_latency;
268 269

	struct list_head cic_list;
Linus Torvalds's avatar
Linus Torvalds committed
270

271 272 273 274
	/*
	 * Fallback dummy cfqq for extreme OOM conditions
	 */
	struct cfq_queue oom_cfqq;
275 276

	unsigned long last_end_sync_rq;
Linus Torvalds's avatar
Linus Torvalds committed
277 278
};

279 280
static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
					    enum wl_prio_t prio,
281
					    enum wl_type_t type,
282 283
					    struct cfq_data *cfqd)
{
284 285 286
	if (!cfqg)
		return NULL;

287
	if (prio == IDLE_WORKLOAD)
288
		return &cfqg->service_tree_idle;
289

290
	return &cfqg->service_trees[prio][type];
291 292
}

Jens Axboe's avatar
Jens Axboe committed
293
enum cfqq_state_flags {
294 295
	CFQ_CFQQ_FLAG_on_rr = 0,	/* on round-robin busy list */
	CFQ_CFQQ_FLAG_wait_request,	/* waiting for a request */
296
	CFQ_CFQQ_FLAG_must_dispatch,	/* must be allowed a dispatch */
297 298 299 300
	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 */
301
	CFQ_CFQQ_FLAG_slice_new,	/* no requests dispatched in slice */
302
	CFQ_CFQQ_FLAG_sync,		/* synchronous queue */
303
	CFQ_CFQQ_FLAG_coop,		/* cfqq is shared */
304
	CFQ_CFQQ_FLAG_deep,		/* sync cfqq experienced large depth */
Jens Axboe's avatar
Jens Axboe committed
305 306 307 308 309
};

#define CFQ_CFQQ_FNS(name)						\
static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)		\
{									\
310
	(cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
311 312 313
}									\
static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)	\
{									\
314
	(cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);			\
Jens Axboe's avatar
Jens Axboe committed
315 316 317
}									\
static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)		\
{									\
318
	return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;	\
Jens Axboe's avatar
Jens Axboe committed
319 320 321 322
}

CFQ_CFQQ_FNS(on_rr);
CFQ_CFQQ_FNS(wait_request);
323
CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe's avatar
Jens Axboe committed
324 325 326 327
CFQ_CFQQ_FNS(must_alloc_slice);
CFQ_CFQQ_FNS(fifo_expire);
CFQ_CFQQ_FNS(idle_window);
CFQ_CFQQ_FNS(prio_changed);
328
CFQ_CFQQ_FNS(slice_new);
329
CFQ_CFQQ_FNS(sync);
330
CFQ_CFQQ_FNS(coop);
331
CFQ_CFQQ_FNS(deep);
Jens Axboe's avatar
Jens Axboe committed
332 333
#undef CFQ_CFQQ_FNS

334 335 336 337 338
#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)

339 340 341 342 343 344 345 346 347 348 349
/* 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) \


350 351 352 353 354 355 356 357 358
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;
}

359 360 361 362 363 364 365 366 367 368

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;
}

369 370 371
static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
					struct cfq_data *cfqd,
					struct cfq_group *cfqg)
372 373
{
	if (wl == IDLE_WORKLOAD)
374
		return cfqg->service_tree_idle.count;
375

376 377 378
	return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
		+ cfqg->service_trees[wl][SYNC_WORKLOAD].count;
379 380
}

381
static void cfq_dispatch_insert(struct request_queue *, struct request *);
382
static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
383
				       struct io_context *, gfp_t);
384
static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
385 386
						struct io_context *);

387 388 389 390 391
static inline int rq_in_driver(struct cfq_data *cfqd)
{
	return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1];
}

392
static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
393
					    bool is_sync)
394
{
395
	return cic->cfqq[is_sync];
396 397 398
}

static inline void cic_set_cfqq(struct cfq_io_context *cic,
399
				struct cfq_queue *cfqq, bool is_sync)
400
{
401
	cic->cfqq[is_sync] = cfqq;
402 403 404 405 406 407
}

/*
 * 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).
 */
408
static inline bool cfq_bio_sync(struct bio *bio)
409
{
410
	return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO);
411
}
Linus Torvalds's avatar
Linus Torvalds committed
412

Andrew Morton's avatar
Andrew Morton committed
413 414 415 416
/*
 * scheduler run of queue, if there are requests pending and no one in the
 * driver that will restart queueing
 */
417
static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton's avatar
Andrew Morton committed
418
{
419 420
	if (cfqd->busy_queues) {
		cfq_log(cfqd, "schedule dispatch");
421
		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
422
	}
Andrew Morton's avatar
Andrew Morton committed
423 424
}

425
static int cfq_queue_empty(struct request_queue *q)
Andrew Morton's avatar
Andrew Morton committed
426 427 428
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

429
	return !cfqd->rq_queued;
Andrew Morton's avatar
Andrew Morton committed
430 431
}

432 433 434 435 436
/*
 * 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.
 */
437
static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
438
				 unsigned short prio)
439
{
440
	const int base_slice = cfqd->cfq_slice[sync];
441

442 443 444 445
	WARN_ON(prio >= IOPRIO_BE_NR);

	return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
}
446

447 448 449 450
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);
451 452
}

453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497
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);
}

498 499 500 501 502 503
/*
 * 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
 */

504 505
static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
					struct cfq_group *cfqg, bool rt)
506
{
507 508 509
	unsigned min_q, max_q;
	unsigned mult  = cfq_hist_divisor - 1;
	unsigned round = cfq_hist_divisor / 2;
510
	unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
511

512 513 514
	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) /
515
		cfq_hist_divisor;
516 517 518 519 520 521 522 523 524
	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;
525 526
}

527 528 529
static inline void
cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
530 531
	unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
	if (cfqd->cfq_latency) {
532 533 534 535 536 537
		/*
		 * 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));
538 539
		unsigned sync_slice = cfqd->cfq_slice[1];
		unsigned expect_latency = sync_slice * iq;
540 541 542
		unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);

		if (expect_latency > group_slice) {
543 544 545 546 547 548 549
			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 */
550
			slice = max(slice * group_slice / expect_latency,
551 552 553
				    low_slice);
		}
	}
554
	cfqq->slice_start = jiffies;
555
	cfqq->slice_end = jiffies + slice;
556
	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
557 558 559 560 561 562 563
}

/*
 * 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.
 */
564
static inline bool cfq_slice_used(struct cfq_queue *cfqq)
565 566 567 568 569 570 571 572 573
{
	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
574
/*
Jens Axboe's avatar
Jens Axboe committed
575
 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds's avatar
Linus Torvalds committed
576
 * We choose the request that is closest to the head right now. Distance
577
 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds's avatar
Linus Torvalds committed
578
 */
Jens Axboe's avatar
Jens Axboe committed
579
static struct request *
580
cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds's avatar
Linus Torvalds committed
581
{
582
	sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds's avatar
Linus Torvalds committed
583
	unsigned long back_max;
584 585 586
#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
587

Jens Axboe's avatar
Jens Axboe committed
588 589 590 591
	if (rq1 == NULL || rq1 == rq2)
		return rq2;
	if (rq2 == NULL)
		return rq1;
592

Jens Axboe's avatar
Jens Axboe committed
593 594 595 596
	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
		return rq1;
	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
		return rq2;
597 598 599 600
	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
601

602 603
	s1 = blk_rq_pos(rq1);
	s2 = blk_rq_pos(rq2);
Linus Torvalds's avatar
Linus Torvalds committed
604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619

	/*
	 * 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
620
		wrap |= CFQ_RQ1_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
621 622 623 624 625 626

	if (s2 >= last)
		d2 = s2 - last;
	else if (s2 + back_max >= last)
		d2 = (last - s2) * cfqd->cfq_back_penalty;
	else
627
		wrap |= CFQ_RQ2_WRAP;
Linus Torvalds's avatar
Linus Torvalds committed
628 629

	/* Found required data */
630 631 632 633 634 635

	/*
	 * 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
636
	case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
637
		if (d1 < d2)
Jens Axboe's avatar
Jens Axboe committed
638
			return rq1;
639
		else if (d2 < d1)
Jens Axboe's avatar
Jens Axboe committed
640
			return rq2;
641 642
		else {
			if (s1 >= s2)
Jens Axboe's avatar
Jens Axboe committed
643
				return rq1;
644
			else
Jens Axboe's avatar
Jens Axboe committed
645
				return rq2;
646
		}
Linus Torvalds's avatar
Linus Torvalds committed
647

648
	case CFQ_RQ2_WRAP:
Jens Axboe's avatar
Jens Axboe committed
649
		return rq1;
650
	case CFQ_RQ1_WRAP:
Jens Axboe's avatar
Jens Axboe committed
651 652
		return rq2;
	case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
653 654 655 656 657 658 659 660
	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
661
			return rq1;
Linus Torvalds's avatar
Linus Torvalds committed
662
		else
Jens Axboe's avatar
Jens Axboe committed
663
			return rq2;
Linus Torvalds's avatar
Linus Torvalds committed
664 665 666
	}
}

667 668 669
/*
 * The below is leftmost cache rbtree addon
 */
670
static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
671
{
672 673 674 675
	/* Service tree is empty */
	if (!root->count)
		return NULL;

676 677 678
	if (!root->left)
		root->left = rb_first(&root->rb);

679 680 681 682
	if (root->left)
		return rb_entry(root->left, struct cfq_queue, rb_node);

	return NULL;
683 684
}

685 686 687 688 689 690 691 692 693 694 695
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;
}

696 697 698 699 700 701
static void rb_erase_init(struct rb_node *n, struct rb_root *root)
{
	rb_erase(n, root);
	RB_CLEAR_NODE(n);
}

702 703 704 705
static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
{
	if (root->left == n)
		root->left = NULL;
706
	rb_erase_init(n, &root->rb);
707
	--root->count;
708 709
}

Linus Torvalds's avatar
Linus Torvalds committed
710 711 712
/*
 * would be nice to take fifo expire time into account as well
 */
Jens Axboe's avatar
Jens Axboe committed
713 714 715
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
716
{
717 718
	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
719
	struct request *next = NULL, *prev = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
720

721
	BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds's avatar
Linus Torvalds committed
722 723

	if (rbprev)
Jens Axboe's avatar
Jens Axboe committed
724
		prev = rb_entry_rq(rbprev);
Linus Torvalds's avatar
Linus Torvalds committed
725

726
	if (rbnext)
Jens Axboe's avatar
Jens Axboe committed
727
		next = rb_entry_rq(rbnext);
728 729 730
	else {
		rbnext = rb_first(&cfqq->sort_list);
		if (rbnext && rbnext != &last->rb_node)
Jens Axboe's avatar
Jens Axboe committed
731
			next = rb_entry_rq(rbnext);
732
	}
Linus Torvalds's avatar
Linus Torvalds committed
733

734
	return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds's avatar
Linus Torvalds committed
735 736
}

737 738
static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
				      struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
739
{
740 741 742
	/*
	 * just an approximation, should be ok.
	 */
743
	return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
744
		       cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
745 746
}

747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805
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;
806 807
	cfqd->nr_groups++;
	st->total_weight += cfqg->weight;
808 809 810 811 812 813 814
}

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

815 816 817
	if (st->active == &cfqg->rb_node)
		st->active = NULL;

818 819
	BUG_ON(cfqg->nr_cfqq < 1);
	cfqg->nr_cfqq--;
820

821 822 823 824 825
	/* If there are other cfq queues under this group, don't delete it */
	if (cfqg->nr_cfqq)
		return;

	cfqg->on_st = false;
826 827
	cfqd->nr_groups--;
	st->total_weight -= cfqg->weight;
828 829
	if (!RB_EMPTY_NODE(&cfqg->rb_node))
		cfq_rb_erase(&cfqg->rb_node, st);
830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881
	cfqg->saved_workload_slice = 0;
}

static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq)
{
	unsigned int slice_used, allocated_slice;

	/*
	 * 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;
		allocated_slice = cfqq->slice_end - cfqq->slice_start;
		if (slice_used > allocated_slice)
			slice_used = allocated_slice;
	}

	cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u", slice_used);
	return slice_used;
}

static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
				struct cfq_queue *cfqq)
{
	struct cfq_rb_root *st = &cfqd->grp_service_tree;
	unsigned int used_sl;

	used_sl = cfq_cfqq_slice_usage(cfqq);

	/* Can't update vdisktime while group is on service tree */
	cfq_rb_erase(&cfqg->rb_node, st);
	cfqg->vdisktime += cfq_scale_slice(used_sl, cfqg);
	__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;
882 883
}

884
/*
885
 * The cfqd->service_trees holds all pending cfq_queue's that have
886 887 888
 * requests waiting to be processed. It is sorted in the order that
 * we will service the queues.
 */
889
static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
890
				 bool add_front)
891
{
892 893
	struct rb_node **p, *parent;
	struct cfq_queue *__cfqq;
894
	unsigned long rb_key;
895
	struct cfq_rb_root *service_tree;
896
	int left;
897
	int new_cfqq = 1;
898

899 900
	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
						cfqq_type(cfqq), cfqd);
901 902
	if (cfq_class_idle(cfqq)) {
		rb_key = CFQ_IDLE_DELAY;
903
		parent = rb_last(&service_tree->rb);
904 905 906 907 908 909
		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) {
910 911 912 913 914 915
		/*
		 * 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.
		 */
916
		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
917
		rb_key -= cfqq->slice_resid;
918
		cfqq->slice_resid = 0;
919 920
	} else {
		rb_key = -HZ;
921
		__cfqq = cfq_rb_first(service_tree);
922 923
		rb_key += __cfqq ? __cfqq->rb_key : jiffies;
	}
Linus Torvalds's avatar
Linus Torvalds committed
924

925
	if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
926
		new_cfqq = 0;
927
		/*
928
		 * same position, nothing more to do
929
		 */
930 931
		if (rb_key == cfqq->rb_key &&
		    cfqq->service_tree == service_tree)
932
			return;
Linus Torvalds's avatar
Linus Torvalds committed
933

934 935
		cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
		cfqq->service_tree = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
936
	}
937

938
	left = 1;
939
	parent = NULL;
940 941
	cfqq->service_tree = service_tree;
	p = &service_tree->rb.rb_node;
942
	while (*p) {
943
		struct rb_node **n;
944

945 946 947
		parent = *p;
		__cfqq = rb_entry(parent, struct cfq_queue, rb_node);

948
		/*
949
		 * sort by key, that represents service time.
950
		 */
951
		if (time_before(rb_key, __cfqq->rb_key))
952
			n = &(*p)->rb_left;
953
		else {
954
			n = &(*p)->rb_right;
955
			left = 0;
956
		}
957 958

		p = n;
959 960
	}

961
	if (left)
962
		service_tree->left = &cfqq->rb_node;
963

964 965
	cfqq->rb_key = rb_key;
	rb_link_node(&cfqq->rb_node, parent, p);
966 967
	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
	service_tree->count++;
968 969
	if (add_front || !new_cfqq)
		return;
970
	cfq_group_service_tree_add(cfqd, cfqq->cfqg);
Linus Torvalds's avatar
Linus Torvalds committed
971 972
}

973
static struct cfq_queue *
974 975 976
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)
977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992
{
	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.
		 */
993
		if (sector > blk_rq_pos(cfqq->next_rq))
994
			n = &(*p)->rb_right;
995
		else if (sector < blk_rq_pos(cfqq->next_rq))
996 997 998 999
			n = &(*p)->rb_left;
		else
			break;
		p = n;
1000
		cfqq = NULL;
1001 1002 1003 1004 1005
	}

	*ret_parent = parent;
	if (rb_link)
		*rb_link = p;
1006
	return cfqq;
1007 1008 1009 1010 1011 1012 1013
}

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

1014 1015 1016 1017
	if (cfqq->p_root) {
		rb_erase(&cfqq->p_node, cfqq->p_root);
		cfqq->p_root = NULL;
	}
1018 1019 1020 1021 1022 1023

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

1024
	cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
1025 1026
	__cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
				      blk_rq_pos(cfqq->next_rq), &parent, &p);
1027 1028
	if (!__cfqq) {
		rb_link_node(&cfqq->p_node, parent, p);
1029 1030 1031
		rb_insert_color(&cfqq->p_node, cfqq->p_root);
	} else
		cfqq->p_root = NULL;
1032 1033
}

1034 1035 1036
/*
 * Update cfqq's position in the service tree.
 */
1037
static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe's avatar
Jens Axboe committed
1038 1039 1040 1041
{
	/*
	 * Resorting requires the cfqq to be on the RR list already.
	 */
1042
	if (cfq_cfqq_on_rr(cfqq)) {
1043
		cfq_service_tree_add(cfqd, cfqq, 0);
1044 1045
		cfq_prio_tree_add(cfqd, cfqq);
	}
Jens Axboe's avatar
Jens Axboe committed
1046 1047
}

Linus Torvalds's avatar
Linus Torvalds committed
1048 1049
/*
 * add to busy list of queues for service, trying to be fair in ordering
1050
 * the pending list according to last request service
Linus Torvalds's avatar
Linus Torvalds committed
1051
 */
1052
static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
1053
{
1054
	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe's avatar
Jens Axboe committed
1055 1056
	BUG_ON(cfq_cfqq_on_rr(cfqq));
	cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
1057 1058
	cfqd->busy_queues++;

1059
	cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds's avatar
Linus Torvalds committed
1060 1061
}

1062 1063 1064 1065
/*
 * Called when the cfqq no longer has requests pending, remove it from
 * the service tree.
 */
1066
static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds's avatar
Linus Torvalds committed
1067
{