builtin-lock.c 23.2 KB
Newer Older
1
#include <errno.h>
2
#include <inttypes.h>
3 4 5
#include "builtin.h"
#include "perf.h"

6
#include "util/evlist.h"
7
#include "util/evsel.h"
8 9 10 11 12 13
#include "util/util.h"
#include "util/cache.h"
#include "util/symbol.h"
#include "util/thread.h"
#include "util/header.h"

14
#include <subcmd/parse-options.h>
15 16 17 18
#include "util/trace-event.h"

#include "util/debug.h"
#include "util/session.h"
19
#include "util/tool.h"
20
#include "util/data.h"
21 22 23 24 25 26 27 28 29 30

#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>
31
#include <linux/kernel.h>
32

33 34
static struct perf_session *session;

35 36 37 38 39 40 41 42 43 44
/* 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 {
45 46
	struct list_head	hash_entry;
	struct rb_node		rb;		/* used for sorting */
47

48
	/*
49
	 * FIXME: perf_evsel__intval() returns u64,
50
	 * so address of lockdep_map should be dealed as 64bit.
51 52 53 54
	 * Is there more better solution?
	 */
	void			*addr;		/* address of lockdep_map, used as ID */
	char			*name;		/* for strcpy(), we cannot use const */
55

56
	unsigned int		nr_acquire;
57
	unsigned int		nr_acquired;
58 59
	unsigned int		nr_contended;
	unsigned int		nr_release;
60

61 62
	unsigned int		nr_readlock;
	unsigned int		nr_trylock;
63

64
	/* these times are in nano sec. */
65
	u64                     avg_wait_time;
66 67 68
	u64			wait_time_total;
	u64			wait_time_min;
	u64			wait_time_max;
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 99 100 101 102 103 104 105 106 107 108 109

	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;
110 111
};

112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
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));
171 172 173 174
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192

	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));
193 194 195 196
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
197 198 199 200 201 202 203 204 205 206
	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;
}

207
/* build simple key function one is bigger than two */
208
#define SINGLE_KEY(member)						\
209 210 211 212 213 214 215 216
	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)
217
SINGLE_KEY(avg_wait_time)
218 219 220
SINGLE_KEY(wait_time_total)
SINGLE_KEY(wait_time_max)

221 222 223 224 225 226 227 228 229 230 231 232
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;
}

233 234 235 236 237 238
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
	 */
239 240
	const char		*name;
	int			(*key)(struct lock_stat*, struct lock_stat*);
241 242
};

243 244 245 246 247
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 */
248 249 250 251 252 253

#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),
254
	DEF_KEY_LOCK(avg_wait, avg_wait_time),
255 256 257 258 259 260 261 262 263
	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 }
};

264
static int select_key(void)
265 266 267 268 269 270
{
	int i;

	for (i = 0; keys[i].name; i++) {
		if (!strcmp(keys[i].name, sort_key)) {
			compare = keys[i].key;
271
			return 0;
272 273 274
		}
	}

275 276 277
	pr_err("Unknown compare key: %s\n", sort_key);

	return -1;
278 279 280
}

static void insert_to_result(struct lock_stat *st,
281
			     int (*bigger)(struct lock_stat *, struct lock_stat *))
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315
{
	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);
}

316
static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
317 318 319 320 321 322 323 324 325 326 327 328 329 330 331
{
	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
332 333
	if (!new->name) {
		free(new);
334
		goto alloc_failed;
D
Davidlohr Bueso 已提交
335
	}
336

D
Davidlohr Bueso 已提交
337
	strcpy(new->name, name);
338 339 340 341 342 343
	new->wait_time_min = ULLONG_MAX;

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

alloc_failed:
344 345
	pr_err("memory allocation failed\n");
	return NULL;
346 347 348
}

struct trace_lock_handler {
349 350
	int (*acquire_event)(struct perf_evsel *evsel,
			     struct perf_sample *sample);
351

352 353
	int (*acquired_event)(struct perf_evsel *evsel,
			      struct perf_sample *sample);
354

355 356
	int (*contended_event)(struct perf_evsel *evsel,
			       struct perf_sample *sample);
357

358 359
	int (*release_event)(struct perf_evsel *evsel,
			     struct perf_sample *sample);
360 361
};

362 363 364 365 366 367 368 369 370 371
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));
372 373 374 375
	if (!seq) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
376 377 378 379 380 381 382
	seq->state = SEQ_STATE_UNINITIALIZED;
	seq->addr = addr;

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

383 384 385 386 387 388 389 390 391
enum broken_state {
	BROKEN_ACQUIRE,
	BROKEN_ACQUIRED,
	BROKEN_CONTENDED,
	BROKEN_RELEASE,
	BROKEN_MAX,
};

static int bad_hist[BROKEN_MAX];
392

393 394 395 396 397
enum acquire_flags {
	TRY_LOCK = 1,
	READ_LOCK = 2,
};

398 399
static int report_lock_acquire_event(struct perf_evsel *evsel,
				     struct perf_sample *sample)
400
{
401
	void *addr;
402 403 404
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
405 406 407
	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");
408

409 410 411
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
412
	if (!ls)
413
		return -ENOMEM;
414
	if (ls->discard)
415
		return 0;
416

417
	ts = thread_stat_findnew(sample->tid);
418
	if (!ts)
419
		return -ENOMEM;
420

421
	seq = get_seq(ts, addr);
422
	if (!seq)
423
		return -ENOMEM;
424

425 426 427
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
	case SEQ_STATE_RELEASED:
428
		if (!flag) {
429 430
			seq->state = SEQ_STATE_ACQUIRING;
		} else {
431
			if (flag & TRY_LOCK)
432
				ls->nr_trylock++;
433
			if (flag & READ_LOCK)
434 435 436 437 438 439 440
				ls->nr_readlock++;
			seq->state = SEQ_STATE_READ_ACQUIRED;
			seq->read_count = 1;
			ls->nr_acquired++;
		}
		break;
	case SEQ_STATE_READ_ACQUIRED:
441
		if (flag & READ_LOCK) {
442 443 444 445 446 447
			seq->read_count++;
			ls->nr_acquired++;
			goto end;
		} else {
			goto broken;
		}
448
		break;
449 450 451 452 453 454
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
broken:
		/* broken lock sequence, discard it */
		ls->discard = 1;
455
		bad_hist[BROKEN_ACQUIRE]++;
456 457 458
		list_del(&seq->list);
		free(seq);
		goto end;
459
	default:
460
		BUG_ON("Unknown state of lock sequence found!\n");
461 462 463
		break;
	}

464
	ls->nr_acquire++;
465
	seq->prev_event_time = sample->time;
466
end:
467
	return 0;
468 469
}

470 471
static int report_lock_acquired_event(struct perf_evsel *evsel,
				      struct perf_sample *sample)
472
{
473
	void *addr;
474 475 476 477
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
	u64 contended_term;
478 479 480 481
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
482

483
	ls = lock_stat_findnew(addr, name);
484
	if (!ls)
485
		return -ENOMEM;
486
	if (ls->discard)
487
		return 0;
488

489
	ts = thread_stat_findnew(sample->tid);
490
	if (!ts)
491
		return -ENOMEM;
492

493
	seq = get_seq(ts, addr);
494
	if (!seq)
495
		return -ENOMEM;
496

497 498 499
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
500
		return 0;
501
	case SEQ_STATE_ACQUIRING:
502
		break;
503
	case SEQ_STATE_CONTENDED:
504
		contended_term = sample->time - seq->prev_event_time;
505 506 507
		ls->wait_time_total += contended_term;
		if (contended_term < ls->wait_time_min)
			ls->wait_time_min = contended_term;
508
		if (ls->wait_time_max < contended_term)
509
			ls->wait_time_max = contended_term;
510
		break;
511 512 513 514 515
	case SEQ_STATE_RELEASED:
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_READ_ACQUIRED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
516
		bad_hist[BROKEN_ACQUIRED]++;
517 518 519
		list_del(&seq->list);
		free(seq);
		goto end;
520
	default:
521
		BUG_ON("Unknown state of lock sequence found!\n");
522 523 524
		break;
	}

525 526
	seq->state = SEQ_STATE_ACQUIRED;
	ls->nr_acquired++;
527
	ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
528
	seq->prev_event_time = sample->time;
529
end:
530
	return 0;
531 532
}

533 534
static int report_lock_contended_event(struct perf_evsel *evsel,
				       struct perf_sample *sample)
535
{
536
	void *addr;
537 538 539
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
540 541 542 543
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
544

545
	ls = lock_stat_findnew(addr, name);
546
	if (!ls)
547
		return -ENOMEM;
548
	if (ls->discard)
549
		return 0;
550

551
	ts = thread_stat_findnew(sample->tid);
552
	if (!ts)
553
		return -ENOMEM;
554

555
	seq = get_seq(ts, addr);
556
	if (!seq)
557
		return -ENOMEM;
558

559 560 561
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
562
		return 0;
563
	case SEQ_STATE_ACQUIRING:
564
		break;
565 566 567 568 569 570
	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;
571
		bad_hist[BROKEN_CONTENDED]++;
572 573 574
		list_del(&seq->list);
		free(seq);
		goto end;
575
	default:
576
		BUG_ON("Unknown state of lock sequence found!\n");
577 578 579
		break;
	}

580 581
	seq->state = SEQ_STATE_CONTENDED;
	ls->nr_contended++;
582
	ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
583
	seq->prev_event_time = sample->time;
584
end:
585
	return 0;
586 587
}

588 589
static int report_lock_release_event(struct perf_evsel *evsel,
				     struct perf_sample *sample)
590
{
591
	void *addr;
592 593 594
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
595 596
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
597

598 599 600
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
601
	if (!ls)
602
		return -ENOMEM;
603
	if (ls->discard)
604
		return 0;
605

606
	ts = thread_stat_findnew(sample->tid);
607
	if (!ts)
608
		return -ENOMEM;
609

610
	seq = get_seq(ts, addr);
611
	if (!seq)
612
		return -ENOMEM;
613

614 615 616 617 618 619 620 621 622 623
	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++;
624 625
			goto end;
		}
626 627 628 629 630 631
		break;
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
	case SEQ_STATE_RELEASED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
632
		bad_hist[BROKEN_RELEASE]++;
633
		goto free_seq;
634
	default:
635
		BUG_ON("Unknown state of lock sequence found!\n");
636 637 638
		break;
	}

639 640 641 642
	ls->nr_release++;
free_seq:
	list_del(&seq->list);
	free(seq);
643
end:
644
	return 0;
645 646 647 648
}

/* lock oriented handlers */
/* TODO: handlers for CPU oriented, thread oriented */
649 650 651 652 653
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,
654 655 656 657
};

static struct trace_lock_handler *trace_handler;

658
static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
659
					     struct perf_sample *sample)
660
{
661
	if (trace_handler->acquire_event)
662 663
		return trace_handler->acquire_event(evsel, sample);
	return 0;
664 665
}

666
static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
667
					      struct perf_sample *sample)
668
{
669
	if (trace_handler->acquired_event)
670 671
		return trace_handler->acquired_event(evsel, sample);
	return 0;
672 673
}

674
static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
675
					      struct perf_sample *sample)
676
{
677
	if (trace_handler->contended_event)
678 679
		return trace_handler->contended_event(evsel, sample);
	return 0;
680 681
}

682
static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
683
					    struct perf_sample *sample)
684
{
685
	if (trace_handler->release_event)
686 687
		return trace_handler->release_event(evsel, sample);
	return 0;
688 689
}

690 691 692 693 694 695 696 697
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");
698
	pr_info("bad: %d, total: %d\n", bad, total);
699
	pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
700 701 702 703 704
	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]);
}

705 706 707 708 709
/* TODO: various way to print, coloring, nano or milli sec */
static void print_result(void)
{
	struct lock_stat *st;
	char cut_name[20];
710
	int bad, total;
711

712 713 714
	pr_info("%20s ", "Name");
	pr_info("%10s ", "acquired");
	pr_info("%10s ", "contended");
715

716
	pr_info("%15s ", "avg wait (ns)");
717 718 719
	pr_info("%15s ", "total wait (ns)");
	pr_info("%15s ", "max wait (ns)");
	pr_info("%15s ", "min wait (ns)");
720

721
	pr_info("\n\n");
722

723
	bad = total = 0;
724
	while ((st = pop_from_result())) {
725 726 727 728 729
		total++;
		if (st->discard) {
			bad++;
			continue;
		}
730 731 732 733
		bzero(cut_name, 20);

		if (strlen(st->name) < 16) {
			/* output raw name */
734
			pr_info("%20s ", st->name);
735 736 737 738 739 740 741
		} 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 */
742
			pr_info("%20s ", cut_name);
743 744
		}

745 746
		pr_info("%10u ", st->nr_acquired);
		pr_info("%10u ", st->nr_contended);
747

748
		pr_info("%15" PRIu64 " ", st->avg_wait_time);
749 750 751
		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 ?
752
		       0 : st->wait_time_min);
753
		pr_info("\n");
754
	}
755

756
	print_bad_events(bad, total);
757 758
}

759
static bool info_threads, info_map;
760 761 762 763 764 765 766 767 768 769 770 771 772

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);
773
		pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
774
		node = rb_next(node);
775
		thread__put(t);
776 777 778
	};
}

779 780 781 782 783
static void dump_map(void)
{
	unsigned int i;
	struct lock_stat *st;

784
	pr_info("Address of instance: name of class\n");
785 786
	for (i = 0; i < LOCKHASH_SIZE; i++) {
		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
787
			pr_info(" %p: %s\n", st->addr, st->name);
788 789 790 791
		}
	}
}

792
static int dump_info(void)
793
{
794 795
	int rc = 0;

796 797 798 799
	if (info_threads)
		dump_threads();
	else if (info_map)
		dump_map();
800 801 802 803 804 805
	else {
		rc = -1;
		pr_err("Unknown type of information\n");
	}

	return rc;
806 807
}

808 809 810
typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
				  struct perf_sample *sample);

811
static int process_sample_event(struct perf_tool *tool __maybe_unused,
812
				union perf_event *event,
813
				struct perf_sample *sample,
814
				struct perf_evsel *evsel,
815
				struct machine *machine)
816
{
817
	int err = 0;
818 819
	struct thread *thread = machine__findnew_thread(machine, sample->pid,
							sample->tid);
820 821 822

	if (thread == NULL) {
		pr_debug("problem processing %d event, skipping it.\n",
823
			event->header.type);
824 825 826
		return -1;
	}

827 828
	if (evsel->handler != NULL) {
		tracepoint_handler f = evsel->handler;
829
		err = f(evsel, sample);
830 831
	}

832 833 834
	thread__put(thread);

	return err;
835 836
}

D
Davidlohr Bueso 已提交
837 838 839 840 841 842 843 844 845 846 847 848
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);
		}
	}
}

849 850 851 852 853 854 855
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 */
};

856 857
static bool force;

D
Davidlohr Bueso 已提交
858
static int __cmd_report(bool display_info)
859
{
D
Davidlohr Bueso 已提交
860
	int err = -EINVAL;
861 862 863
	struct perf_tool eops = {
		.sample		 = process_sample_event,
		.comm		 = perf_event__process_comm,
864
		.namespaces	 = perf_event__process_namespaces,
865
		.ordered_events	 = true,
866
	};
867 868 869
	struct perf_data_file file = {
		.path = input_name,
		.mode = PERF_DATA_MODE_READ,
870
		.force = force,
871
	};
D
Davidlohr Bueso 已提交
872

873
	session = perf_session__new(&file, false, &eops);
874 875
	if (!session) {
		pr_err("Initializing perf session failed\n");
876
		return -1;
877
	}
878

879
	symbol__init(&session->header.env);
880

D
Davidlohr Bueso 已提交
881 882 883
	if (!perf_session__has_traces(session, "lock record"))
		goto out_delete;

884 885
	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
		pr_err("Initializing perf session tracepoint handlers failed\n");
D
Davidlohr Bueso 已提交
886
		goto out_delete;
887 888
	}

D
Davidlohr Bueso 已提交
889 890
	if (select_key())
		goto out_delete;
891

892
	err = perf_session__process_events(session);
D
Davidlohr Bueso 已提交
893 894
	if (err)
		goto out_delete;
895 896

	setup_pager();
D
Davidlohr Bueso 已提交
897 898 899 900 901 902
	if (display_info) /* used for info subcommand */
		err = dump_info();
	else {
		sort_result();
		print_result();
	}
903

D
Davidlohr Bueso 已提交
904 905 906
out_delete:
	perf_session__delete(session);
	return err;
907 908 909 910
}

static int __cmd_record(int argc, const char **argv)
{
911
	const char *record_args[] = {
912
		"record", "-R", "-m", "1024", "-c", "1",
913
	};
D
Davidlohr Bueso 已提交
914
	unsigned int rec_argc, i, j, ret;
915 916
	const char **rec_argv;

917
	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
918
		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
919 920
				pr_err("tracepoint %s is not enabled. "
				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
921
				       lock_tracepoints[i].name);
922 923 924 925
				return 1;
		}
	}

926
	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
927 928
	/* factor of 2 is for -e in front of each tracepoint */
	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
929

930
	rec_argv = calloc(rec_argc + 1, sizeof(char *));
D
Davidlohr Bueso 已提交
931
	if (!rec_argv)
932 933
		return -ENOMEM;

934 935 936
	for (i = 0; i < ARRAY_SIZE(record_args); i++)
		rec_argv[i] = strdup(record_args[i]);

937 938
	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
		rec_argv[i++] = "-e";
939
		rec_argv[i++] = strdup(lock_tracepoints[j].name);
940 941
	}

942 943 944 945 946
	for (j = 1; j < (unsigned int)argc; j++, i++)
		rec_argv[i] = argv[j];

	BUG_ON(i != rec_argc);

947
	ret = cmd_record(i, rec_argv);
D
Davidlohr Bueso 已提交
948 949
	free(rec_argv);
	return ret;
950 951
}

952
int cmd_lock(int argc, const char **argv)
953
{
954 955 956 957
	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"),
958
	OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
959 960 961
	OPT_END()
	};

962 963 964 965 966
	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)"),
967
	OPT_PARENT(lock_options)
968
	};
969

970 971
	const struct option report_options[] = {
	OPT_STRING('k', "key", &sort_key, "acquired",
972
		    "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
973
	/* TODO: type */
974
	OPT_PARENT(lock_options)
975
	};
976

977 978 979 980
	const char * const info_usage[] = {
		"perf lock info [<options>]",
		NULL
	};
981 982 983 984
	const char *const lock_subcommands[] = { "record", "report", "script",
						 "info", NULL };
	const char *lock_usage[] = {
		NULL,
985 986 987 988 989 990
		NULL
	};
	const char * const report_usage[] = {
		"perf lock report [<options>]",
		NULL
	};
991
	unsigned int i;
992
	int rc = 0;
993 994 995 996

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

997 998
	argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
					lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
999 1000 1001 1002 1003
	if (!argc)
		usage_with_options(lock_usage, lock_options);

	if (!strncmp(argv[0], "rec", 3)) {
		return __cmd_record(argc, argv);
1004 1005
	} else if (!strncmp(argv[0], "report", 6)) {
		trace_handler = &report_lock_ops;
1006 1007
		if (argc) {
			argc = parse_options(argc, argv,
1008
					     report_options, report_usage, 0);
1009
			if (argc)
1010
				usage_with_options(report_usage, report_options);
1011
		}
D
Davidlohr Bueso 已提交
1012
		rc = __cmd_report(false);
1013 1014
	} else if (!strcmp(argv[0], "script")) {
		/* Aliased to 'perf script' */
1015
		return cmd_script(argc, argv);
1016 1017 1018 1019 1020 1021 1022
	} 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);
		}
1023 1024
		/* recycling report_lock_ops */
		trace_handler = &report_lock_ops;
D
Davidlohr Bueso 已提交
1025
		rc = __cmd_report(true);
1026 1027 1028 1029
	} else {
		usage_with_options(lock_usage, lock_options);
	}

1030
	return rc;
1031
}