builtin-lock.c 23.2 KB
Newer Older
1
#include <inttypes.h>
2 3 4
#include "builtin.h"
#include "perf.h"

5
#include "util/evlist.h"
6
#include "util/evsel.h"
7 8 9 10 11 12
#include "util/util.h"
#include "util/cache.h"
#include "util/symbol.h"
#include "util/thread.h"
#include "util/header.h"

13
#include <subcmd/parse-options.h>
14 15 16 17
#include "util/trace-event.h"

#include "util/debug.h"
#include "util/session.h"
18
#include "util/tool.h"
19
#include "util/data.h"
20 21 22 23 24 25 26 27 28 29

#include <sys/types.h>
#include <sys/prctl.h>
#include <semaphore.h>
#include <pthread.h>
#include <math.h>
#include <limits.h>

#include <linux/list.h>
#include <linux/hash.h>
30
#include <linux/kernel.h>
31

32 33
static struct perf_session *session;

34 35 36 37 38 39 40 41 42 43
/* based on kernel/lockdep.c */
#define LOCKHASH_BITS		12
#define LOCKHASH_SIZE		(1UL << LOCKHASH_BITS)

static struct list_head lockhash_table[LOCKHASH_SIZE];

#define __lockhashfn(key)	hash_long((unsigned long)key, LOCKHASH_BITS)
#define lockhashentry(key)	(lockhash_table + __lockhashfn((key)))

struct lock_stat {
44 45
	struct list_head	hash_entry;
	struct rb_node		rb;		/* used for sorting */
46

47
	/*
48
	 * FIXME: perf_evsel__intval() returns u64,
49
	 * so address of lockdep_map should be dealed as 64bit.
50 51 52 53
	 * Is there more better solution?
	 */
	void			*addr;		/* address of lockdep_map, used as ID */
	char			*name;		/* for strcpy(), we cannot use const */
54

55
	unsigned int		nr_acquire;
56
	unsigned int		nr_acquired;
57 58
	unsigned int		nr_contended;
	unsigned int		nr_release;
59

60 61
	unsigned int		nr_readlock;
	unsigned int		nr_trylock;
62

63
	/* these times are in nano sec. */
64
	u64                     avg_wait_time;
65 66 67
	u64			wait_time_total;
	u64			wait_time_min;
	u64			wait_time_max;
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108

	int			discard; /* flag of blacklist */
};

/*
 * States of lock_seq_stat
 *
 * UNINITIALIZED is required for detecting first event of acquire.
 * As the nature of lock events, there is no guarantee
 * that the first event for the locks are acquire,
 * it can be acquired, contended or release.
 */
#define SEQ_STATE_UNINITIALIZED      0	       /* initial state */
#define SEQ_STATE_RELEASED	1
#define SEQ_STATE_ACQUIRING	2
#define SEQ_STATE_ACQUIRED	3
#define SEQ_STATE_READ_ACQUIRED	4
#define SEQ_STATE_CONTENDED	5

/*
 * MAX_LOCK_DEPTH
 * Imported from include/linux/sched.h.
 * Should this be synchronized?
 */
#define MAX_LOCK_DEPTH 48

/*
 * struct lock_seq_stat:
 * Place to put on state of one lock sequence
 * 1) acquire -> acquired -> release
 * 2) acquire -> contended -> acquired -> release
 * 3) acquire (with read or try) -> release
 * 4) Are there other patterns?
 */
struct lock_seq_stat {
	struct list_head        list;
	int			state;
	u64			prev_event_time;
	void                    *addr;

	int                     read_count;
109 110
};

111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
struct thread_stat {
	struct rb_node		rb;

	u32                     tid;
	struct list_head        seq_list;
};

static struct rb_root		thread_stats;

static struct thread_stat *thread_stat_find(u32 tid)
{
	struct rb_node *node;
	struct thread_stat *st;

	node = thread_stats.rb_node;
	while (node) {
		st = container_of(node, struct thread_stat, rb);
		if (st->tid == tid)
			return st;
		else if (tid < st->tid)
			node = node->rb_left;
		else
			node = node->rb_right;
	}

	return NULL;
}

static void thread_stat_insert(struct thread_stat *new)
{
	struct rb_node **rb = &thread_stats.rb_node;
	struct rb_node *parent = NULL;
	struct thread_stat *p;

	while (*rb) {
		p = container_of(*rb, struct thread_stat, rb);
		parent = *rb;

		if (new->tid < p->tid)
			rb = &(*rb)->rb_left;
		else if (new->tid > p->tid)
			rb = &(*rb)->rb_right;
		else
			BUG_ON("inserting invalid thread_stat\n");
	}

	rb_link_node(&new->rb, parent, rb);
	rb_insert_color(&new->rb, &thread_stats);
}

static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
{
	struct thread_stat *st;

	st = thread_stat_find(tid);
	if (st)
		return st;

	st = zalloc(sizeof(struct thread_stat));
170 171 172 173
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191

	st->tid = tid;
	INIT_LIST_HEAD(&st->seq_list);

	thread_stat_insert(st);

	return st;
}

static struct thread_stat *thread_stat_findnew_first(u32 tid);
static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
	thread_stat_findnew_first;

static struct thread_stat *thread_stat_findnew_first(u32 tid)
{
	struct thread_stat *st;

	st = zalloc(sizeof(struct thread_stat));
192 193 194 195
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
196 197 198 199 200 201 202 203 204 205
	st->tid = tid;
	INIT_LIST_HEAD(&st->seq_list);

	rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
	rb_insert_color(&st->rb, &thread_stats);

	thread_stat_findnew = thread_stat_findnew_after_first;
	return st;
}

206
/* build simple key function one is bigger than two */
207
#define SINGLE_KEY(member)						\
208 209 210 211 212 213 214 215
	static int lock_stat_key_ ## member(struct lock_stat *one,	\
					 struct lock_stat *two)		\
	{								\
		return one->member > two->member;			\
	}

SINGLE_KEY(nr_acquired)
SINGLE_KEY(nr_contended)
216
SINGLE_KEY(avg_wait_time)
217 218 219
SINGLE_KEY(wait_time_total)
SINGLE_KEY(wait_time_max)

220 221 222 223 224 225 226 227 228 229 230 231
static int lock_stat_key_wait_time_min(struct lock_stat *one,
					struct lock_stat *two)
{
	u64 s1 = one->wait_time_min;
	u64 s2 = two->wait_time_min;
	if (s1 == ULLONG_MAX)
		s1 = 0;
	if (s2 == ULLONG_MAX)
		s2 = 0;
	return s1 > s2;
}

232 233 234 235 236 237
struct lock_key {
	/*
	 * name: the value for specify by user
	 * this should be simpler than raw name of member
	 * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
	 */
238 239
	const char		*name;
	int			(*key)(struct lock_stat*, struct lock_stat*);
240 241
};

242 243 244 245 246
static const char		*sort_key = "acquired";

static int			(*compare)(struct lock_stat *, struct lock_stat *);

static struct rb_root		result;	/* place to store sorted data */
247 248 249 250 251 252

#define DEF_KEY_LOCK(name, fn_suffix)	\
	{ #name, lock_stat_key_ ## fn_suffix }
struct lock_key keys[] = {
	DEF_KEY_LOCK(acquired, nr_acquired),
	DEF_KEY_LOCK(contended, nr_contended),
253
	DEF_KEY_LOCK(avg_wait, avg_wait_time),
254 255 256 257 258 259 260 261 262
	DEF_KEY_LOCK(wait_total, wait_time_total),
	DEF_KEY_LOCK(wait_min, wait_time_min),
	DEF_KEY_LOCK(wait_max, wait_time_max),

	/* extra comparisons much complicated should be here */

	{ NULL, NULL }
};

263
static int select_key(void)
264 265 266 267 268 269
{
	int i;

	for (i = 0; keys[i].name; i++) {
		if (!strcmp(keys[i].name, sort_key)) {
			compare = keys[i].key;
270
			return 0;
271 272 273
		}
	}

274 275 276
	pr_err("Unknown compare key: %s\n", sort_key);

	return -1;
277 278 279
}

static void insert_to_result(struct lock_stat *st,
280
			     int (*bigger)(struct lock_stat *, struct lock_stat *))
281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314
{
	struct rb_node **rb = &result.rb_node;
	struct rb_node *parent = NULL;
	struct lock_stat *p;

	while (*rb) {
		p = container_of(*rb, struct lock_stat, rb);
		parent = *rb;

		if (bigger(st, p))
			rb = &(*rb)->rb_left;
		else
			rb = &(*rb)->rb_right;
	}

	rb_link_node(&st->rb, parent, rb);
	rb_insert_color(&st->rb, &result);
}

/* returns left most element of result, and erase it */
static struct lock_stat *pop_from_result(void)
{
	struct rb_node *node = result.rb_node;

	if (!node)
		return NULL;

	while (node->rb_left)
		node = node->rb_left;

	rb_erase(node, &result);
	return container_of(node, struct lock_stat, rb);
}

315
static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
{
	struct list_head *entry = lockhashentry(addr);
	struct lock_stat *ret, *new;

	list_for_each_entry(ret, entry, hash_entry) {
		if (ret->addr == addr)
			return ret;
	}

	new = zalloc(sizeof(struct lock_stat));
	if (!new)
		goto alloc_failed;

	new->addr = addr;
	new->name = zalloc(sizeof(char) * strlen(name) + 1);
D
Davidlohr Bueso 已提交
331 332
	if (!new->name) {
		free(new);
333
		goto alloc_failed;
D
Davidlohr Bueso 已提交
334
	}
335

D
Davidlohr Bueso 已提交
336
	strcpy(new->name, name);
337 338 339 340 341 342
	new->wait_time_min = ULLONG_MAX;

	list_add(&new->hash_entry, entry);
	return new;

alloc_failed:
343 344
	pr_err("memory allocation failed\n");
	return NULL;
345 346 347
}

struct trace_lock_handler {
348 349
	int (*acquire_event)(struct perf_evsel *evsel,
			     struct perf_sample *sample);
350

351 352
	int (*acquired_event)(struct perf_evsel *evsel,
			      struct perf_sample *sample);
353

354 355
	int (*contended_event)(struct perf_evsel *evsel,
			       struct perf_sample *sample);
356

357 358
	int (*release_event)(struct perf_evsel *evsel,
			     struct perf_sample *sample);
359 360
};

361 362 363 364 365 366 367 368 369 370
static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
{
	struct lock_seq_stat *seq;

	list_for_each_entry(seq, &ts->seq_list, list) {
		if (seq->addr == addr)
			return seq;
	}

	seq = zalloc(sizeof(struct lock_seq_stat));
371 372 373 374
	if (!seq) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
375 376 377 378 379 380 381
	seq->state = SEQ_STATE_UNINITIALIZED;
	seq->addr = addr;

	list_add(&seq->list, &ts->seq_list);
	return seq;
}

382 383 384 385 386 387 388 389 390
enum broken_state {
	BROKEN_ACQUIRE,
	BROKEN_ACQUIRED,
	BROKEN_CONTENDED,
	BROKEN_RELEASE,
	BROKEN_MAX,
};

static int bad_hist[BROKEN_MAX];
391

392 393 394 395 396
enum acquire_flags {
	TRY_LOCK = 1,
	READ_LOCK = 2,
};

397 398
static int report_lock_acquire_event(struct perf_evsel *evsel,
				     struct perf_sample *sample)
399
{
400
	void *addr;
401 402 403
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
404 405 406
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
	int flag = perf_evsel__intval(evsel, sample, "flag");
407

408 409 410
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
411
	if (!ls)
412
		return -ENOMEM;
413
	if (ls->discard)
414
		return 0;
415

416
	ts = thread_stat_findnew(sample->tid);
417
	if (!ts)
418
		return -ENOMEM;
419

420
	seq = get_seq(ts, addr);
421
	if (!seq)
422
		return -ENOMEM;
423

424 425 426
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
	case SEQ_STATE_RELEASED:
427
		if (!flag) {
428 429
			seq->state = SEQ_STATE_ACQUIRING;
		} else {
430
			if (flag & TRY_LOCK)
431
				ls->nr_trylock++;
432
			if (flag & READ_LOCK)
433 434 435 436 437 438 439
				ls->nr_readlock++;
			seq->state = SEQ_STATE_READ_ACQUIRED;
			seq->read_count = 1;
			ls->nr_acquired++;
		}
		break;
	case SEQ_STATE_READ_ACQUIRED:
440
		if (flag & READ_LOCK) {
441 442 443 444 445 446
			seq->read_count++;
			ls->nr_acquired++;
			goto end;
		} else {
			goto broken;
		}
447
		break;
448 449 450 451 452 453
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
broken:
		/* broken lock sequence, discard it */
		ls->discard = 1;
454
		bad_hist[BROKEN_ACQUIRE]++;
455 456 457
		list_del(&seq->list);
		free(seq);
		goto end;
458
	default:
459
		BUG_ON("Unknown state of lock sequence found!\n");
460 461 462
		break;
	}

463
	ls->nr_acquire++;
464
	seq->prev_event_time = sample->time;
465
end:
466
	return 0;
467 468
}

469 470
static int report_lock_acquired_event(struct perf_evsel *evsel,
				      struct perf_sample *sample)
471
{
472
	void *addr;
473 474 475 476
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
	u64 contended_term;
477 478 479 480
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
481

482
	ls = lock_stat_findnew(addr, name);
483
	if (!ls)
484
		return -ENOMEM;
485
	if (ls->discard)
486
		return 0;
487

488
	ts = thread_stat_findnew(sample->tid);
489
	if (!ts)
490
		return -ENOMEM;
491

492
	seq = get_seq(ts, addr);
493
	if (!seq)
494
		return -ENOMEM;
495

496 497 498
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
499
		return 0;
500
	case SEQ_STATE_ACQUIRING:
501
		break;
502
	case SEQ_STATE_CONTENDED:
503
		contended_term = sample->time - seq->prev_event_time;
504 505 506
		ls->wait_time_total += contended_term;
		if (contended_term < ls->wait_time_min)
			ls->wait_time_min = contended_term;
507
		if (ls->wait_time_max < contended_term)
508
			ls->wait_time_max = contended_term;
509
		break;
510 511 512 513 514
	case SEQ_STATE_RELEASED:
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_READ_ACQUIRED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
515
		bad_hist[BROKEN_ACQUIRED]++;
516 517 518
		list_del(&seq->list);
		free(seq);
		goto end;
519
	default:
520
		BUG_ON("Unknown state of lock sequence found!\n");
521 522 523
		break;
	}

524 525
	seq->state = SEQ_STATE_ACQUIRED;
	ls->nr_acquired++;
526
	ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
527
	seq->prev_event_time = sample->time;
528
end:
529
	return 0;
530 531
}

532 533
static int report_lock_contended_event(struct perf_evsel *evsel,
				       struct perf_sample *sample)
534
{
535
	void *addr;
536 537 538
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
539 540 541 542
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
543

544
	ls = lock_stat_findnew(addr, name);
545
	if (!ls)
546
		return -ENOMEM;
547
	if (ls->discard)
548
		return 0;
549

550
	ts = thread_stat_findnew(sample->tid);
551
	if (!ts)
552
		return -ENOMEM;
553

554
	seq = get_seq(ts, addr);
555
	if (!seq)
556
		return -ENOMEM;
557

558 559 560
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
561
		return 0;
562
	case SEQ_STATE_ACQUIRING:
563
		break;
564 565 566 567 568 569
	case SEQ_STATE_RELEASED:
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_READ_ACQUIRED:
	case SEQ_STATE_CONTENDED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
570
		bad_hist[BROKEN_CONTENDED]++;
571 572 573
		list_del(&seq->list);
		free(seq);
		goto end;
574
	default:
575
		BUG_ON("Unknown state of lock sequence found!\n");
576 577 578
		break;
	}

579 580
	seq->state = SEQ_STATE_CONTENDED;
	ls->nr_contended++;
581
	ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
582
	seq->prev_event_time = sample->time;
583
end:
584
	return 0;
585 586
}

587 588
static int report_lock_release_event(struct perf_evsel *evsel,
				     struct perf_sample *sample)
589
{
590
	void *addr;
591 592 593
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
594 595
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
596

597 598 599
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
600
	if (!ls)
601
		return -ENOMEM;
602
	if (ls->discard)
603
		return 0;
604

605
	ts = thread_stat_findnew(sample->tid);
606
	if (!ts)
607
		return -ENOMEM;
608

609
	seq = get_seq(ts, addr);
610
	if (!seq)
611
		return -ENOMEM;
612

613 614 615 616 617 618 619 620 621 622
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		goto end;
	case SEQ_STATE_ACQUIRED:
		break;
	case SEQ_STATE_READ_ACQUIRED:
		seq->read_count--;
		BUG_ON(seq->read_count < 0);
		if (!seq->read_count) {
			ls->nr_release++;
623 624
			goto end;
		}
625 626 627 628 629 630
		break;
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
	case SEQ_STATE_RELEASED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
631
		bad_hist[BROKEN_RELEASE]++;
632
		goto free_seq;
633
	default:
634
		BUG_ON("Unknown state of lock sequence found!\n");
635 636 637
		break;
	}

638 639 640 641
	ls->nr_release++;
free_seq:
	list_del(&seq->list);
	free(seq);
642
end:
643
	return 0;
644 645 646 647
}

/* lock oriented handlers */
/* TODO: handlers for CPU oriented, thread oriented */
648 649 650 651 652
static struct trace_lock_handler report_lock_ops  = {
	.acquire_event		= report_lock_acquire_event,
	.acquired_event		= report_lock_acquired_event,
	.contended_event	= report_lock_contended_event,
	.release_event		= report_lock_release_event,
653 654 655 656
};

static struct trace_lock_handler *trace_handler;

657
static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
658
					     struct perf_sample *sample)
659
{
660
	if (trace_handler->acquire_event)
661 662
		return trace_handler->acquire_event(evsel, sample);
	return 0;
663 664
}

665
static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
666
					      struct perf_sample *sample)
667
{
668
	if (trace_handler->acquired_event)
669 670
		return trace_handler->acquired_event(evsel, sample);
	return 0;
671 672
}

673
static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
674
					      struct perf_sample *sample)
675
{
676
	if (trace_handler->contended_event)
677 678
		return trace_handler->contended_event(evsel, sample);
	return 0;
679 680
}

681
static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
682
					    struct perf_sample *sample)
683
{
684
	if (trace_handler->release_event)
685 686
		return trace_handler->release_event(evsel, sample);
	return 0;
687 688
}

689 690 691 692 693 694 695 696
static void print_bad_events(int bad, int total)
{
	/* Output for debug, this have to be removed */
	int i;
	const char *name[4] =
		{ "acquire", "acquired", "contended", "release" };

	pr_info("\n=== output for debug===\n\n");
697
	pr_info("bad: %d, total: %d\n", bad, total);
698
	pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
699 700 701 702 703
	pr_info("histogram of events caused bad sequence\n");
	for (i = 0; i < BROKEN_MAX; i++)
		pr_info(" %10s: %d\n", name[i], bad_hist[i]);
}

704 705 706 707 708
/* TODO: various way to print, coloring, nano or milli sec */
static void print_result(void)
{
	struct lock_stat *st;
	char cut_name[20];
709
	int bad, total;
710

711 712 713
	pr_info("%20s ", "Name");
	pr_info("%10s ", "acquired");
	pr_info("%10s ", "contended");
714

715
	pr_info("%15s ", "avg wait (ns)");
716 717 718
	pr_info("%15s ", "total wait (ns)");
	pr_info("%15s ", "max wait (ns)");
	pr_info("%15s ", "min wait (ns)");
719

720
	pr_info("\n\n");
721

722
	bad = total = 0;
723
	while ((st = pop_from_result())) {
724 725 726 727 728
		total++;
		if (st->discard) {
			bad++;
			continue;
		}
729 730 731 732
		bzero(cut_name, 20);

		if (strlen(st->name) < 16) {
			/* output raw name */
733
			pr_info("%20s ", st->name);
734 735 736 737 738 739 740
		} else {
			strncpy(cut_name, st->name, 16);
			cut_name[16] = '.';
			cut_name[17] = '.';
			cut_name[18] = '.';
			cut_name[19] = '\0';
			/* cut off name for saving output style */
741
			pr_info("%20s ", cut_name);
742 743
		}

744 745
		pr_info("%10u ", st->nr_acquired);
		pr_info("%10u ", st->nr_contended);
746

747
		pr_info("%15" PRIu64 " ", st->avg_wait_time);
748 749 750
		pr_info("%15" PRIu64 " ", st->wait_time_total);
		pr_info("%15" PRIu64 " ", st->wait_time_max);
		pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
751
		       0 : st->wait_time_min);
752
		pr_info("\n");
753
	}
754

755
	print_bad_events(bad, total);
756 757
}

758
static bool info_threads, info_map;
759 760 761 762 763 764 765 766 767 768 769 770 771

static void dump_threads(void)
{
	struct thread_stat *st;
	struct rb_node *node;
	struct thread *t;

	pr_info("%10s: comm\n", "Thread ID");

	node = rb_first(&thread_stats);
	while (node) {
		st = container_of(node, struct thread_stat, rb);
		t = perf_session__findnew(session, st->tid);
772
		pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
773
		node = rb_next(node);
774
		thread__put(t);
775 776 777
	};
}

778 779 780 781 782
static void dump_map(void)
{
	unsigned int i;
	struct lock_stat *st;

783
	pr_info("Address of instance: name of class\n");
784 785
	for (i = 0; i < LOCKHASH_SIZE; i++) {
		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
786
			pr_info(" %p: %s\n", st->addr, st->name);
787 788 789 790
		}
	}
}

791
static int dump_info(void)
792
{
793 794
	int rc = 0;

795 796 797 798
	if (info_threads)
		dump_threads();
	else if (info_map)
		dump_map();
799 800 801 802 803 804
	else {
		rc = -1;
		pr_err("Unknown type of information\n");
	}

	return rc;
805 806
}

807 808 809
typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
				  struct perf_sample *sample);

810
static int process_sample_event(struct perf_tool *tool __maybe_unused,
811
				union perf_event *event,
812
				struct perf_sample *sample,
813
				struct perf_evsel *evsel,
814
				struct machine *machine)
815
{
816
	int err = 0;
817 818
	struct thread *thread = machine__findnew_thread(machine, sample->pid,
							sample->tid);
819 820 821

	if (thread == NULL) {
		pr_debug("problem processing %d event, skipping it.\n",
822
			event->header.type);
823 824 825
		return -1;
	}

826 827
	if (evsel->handler != NULL) {
		tracepoint_handler f = evsel->handler;
828
		err = f(evsel, sample);
829 830
	}

831 832 833
	thread__put(thread);

	return err;
834 835
}

D
Davidlohr Bueso 已提交
836 837 838 839 840 841 842 843 844 845 846 847
static void sort_result(void)
{
	unsigned int i;
	struct lock_stat *st;

	for (i = 0; i < LOCKHASH_SIZE; i++) {
		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
			insert_to_result(st, compare);
		}
	}
}

848 849 850 851 852 853 854
static const struct perf_evsel_str_handler lock_tracepoints[] = {
	{ "lock:lock_acquire",	 perf_evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
	{ "lock:lock_acquired",	 perf_evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
	{ "lock:lock_contended", perf_evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
	{ "lock:lock_release",	 perf_evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
};

855 856
static bool force;

D
Davidlohr Bueso 已提交
857
static int __cmd_report(bool display_info)
858
{
D
Davidlohr Bueso 已提交
859
	int err = -EINVAL;
860 861 862
	struct perf_tool eops = {
		.sample		 = process_sample_event,
		.comm		 = perf_event__process_comm,
863
		.namespaces	 = perf_event__process_namespaces,
864
		.ordered_events	 = true,
865
	};
866 867 868
	struct perf_data_file file = {
		.path = input_name,
		.mode = PERF_DATA_MODE_READ,
869
		.force = force,
870
	};
D
Davidlohr Bueso 已提交
871

872
	session = perf_session__new(&file, false, &eops);
873 874
	if (!session) {
		pr_err("Initializing perf session failed\n");
875
		return -1;
876
	}
877

878
	symbol__init(&session->header.env);
879

D
Davidlohr Bueso 已提交
880 881 882
	if (!perf_session__has_traces(session, "lock record"))
		goto out_delete;

883 884
	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
		pr_err("Initializing perf session tracepoint handlers failed\n");
D
Davidlohr Bueso 已提交
885
		goto out_delete;
886 887
	}

D
Davidlohr Bueso 已提交
888 889
	if (select_key())
		goto out_delete;
890

891
	err = perf_session__process_events(session);
D
Davidlohr Bueso 已提交
892 893
	if (err)
		goto out_delete;
894 895

	setup_pager();
D
Davidlohr Bueso 已提交
896 897 898 899 900 901
	if (display_info) /* used for info subcommand */
		err = dump_info();
	else {
		sort_result();
		print_result();
	}
902

D
Davidlohr Bueso 已提交
903 904 905
out_delete:
	perf_session__delete(session);
	return err;
906 907 908 909
}

static int __cmd_record(int argc, const char **argv)
{
910
	const char *record_args[] = {
911
		"record", "-R", "-m", "1024", "-c", "1",
912
	};
D
Davidlohr Bueso 已提交
913
	unsigned int rec_argc, i, j, ret;
914 915
	const char **rec_argv;

916
	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
917
		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
918 919
				pr_err("tracepoint %s is not enabled. "
				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
920
				       lock_tracepoints[i].name);
921 922 923 924
				return 1;
		}
	}

925
	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
926 927
	/* factor of 2 is for -e in front of each tracepoint */
	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
928

929
	rec_argv = calloc(rec_argc + 1, sizeof(char *));
D
Davidlohr Bueso 已提交
930
	if (!rec_argv)
931 932
		return -ENOMEM;

933 934 935
	for (i = 0; i < ARRAY_SIZE(record_args); i++)
		rec_argv[i] = strdup(record_args[i]);

936 937
	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
		rec_argv[i++] = "-e";
938
		rec_argv[i++] = strdup(lock_tracepoints[j].name);
939 940
	}

941 942 943 944 945
	for (j = 1; j < (unsigned int)argc; j++, i++)
		rec_argv[i] = argv[j];

	BUG_ON(i != rec_argc);

946
	ret = cmd_record(i, rec_argv);
D
Davidlohr Bueso 已提交
947 948
	free(rec_argv);
	return ret;
949 950
}

951
int cmd_lock(int argc, const char **argv)
952
{
953 954 955 956
	const struct option lock_options[] = {
	OPT_STRING('i', "input", &input_name, "file", "input file name"),
	OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
957
	OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
958 959 960
	OPT_END()
	};

961 962 963 964 965
	const struct option info_options[] = {
	OPT_BOOLEAN('t', "threads", &info_threads,
		    "dump thread list in perf.data"),
	OPT_BOOLEAN('m', "map", &info_map,
		    "map of lock instances (address:name table)"),
966
	OPT_PARENT(lock_options)
967
	};
968

969 970
	const struct option report_options[] = {
	OPT_STRING('k', "key", &sort_key, "acquired",
971
		    "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
972
	/* TODO: type */
973
	OPT_PARENT(lock_options)
974
	};
975

976 977 978 979
	const char * const info_usage[] = {
		"perf lock info [<options>]",
		NULL
	};
980 981 982 983
	const char *const lock_subcommands[] = { "record", "report", "script",
						 "info", NULL };
	const char *lock_usage[] = {
		NULL,
984 985 986 987 988 989
		NULL
	};
	const char * const report_usage[] = {
		"perf lock report [<options>]",
		NULL
	};
990
	unsigned int i;
991
	int rc = 0;
992 993 994 995

	for (i = 0; i < LOCKHASH_SIZE; i++)
		INIT_LIST_HEAD(lockhash_table + i);

996 997
	argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
					lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
998 999 1000 1001 1002
	if (!argc)
		usage_with_options(lock_usage, lock_options);

	if (!strncmp(argv[0], "rec", 3)) {
		return __cmd_record(argc, argv);
1003 1004
	} else if (!strncmp(argv[0], "report", 6)) {
		trace_handler = &report_lock_ops;
1005 1006
		if (argc) {
			argc = parse_options(argc, argv,
1007
					     report_options, report_usage, 0);
1008
			if (argc)
1009
				usage_with_options(report_usage, report_options);
1010
		}
D
Davidlohr Bueso 已提交
1011
		rc = __cmd_report(false);
1012 1013
	} else if (!strcmp(argv[0], "script")) {
		/* Aliased to 'perf script' */
1014
		return cmd_script(argc, argv);
1015 1016 1017 1018 1019 1020 1021
	} else if (!strcmp(argv[0], "info")) {
		if (argc) {
			argc = parse_options(argc, argv,
					     info_options, info_usage, 0);
			if (argc)
				usage_with_options(info_usage, info_options);
		}
1022 1023
		/* recycling report_lock_ops */
		trace_handler = &report_lock_ops;
D
Davidlohr Bueso 已提交
1024
		rc = __cmd_report(true);
1025 1026 1027 1028
	} else {
		usage_with_options(lock_usage, lock_options);
	}

1029
	return rc;
1030
}