builtin-lock.c 22.9 KB
Newer Older
1 2 3
#include "builtin.h"
#include "perf.h"

4
#include "util/evlist.h"
5
#include "util/evsel.h"
6 7 8 9 10 11 12 13 14 15 16
#include "util/util.h"
#include "util/cache.h"
#include "util/symbol.h"
#include "util/thread.h"
#include "util/header.h"

#include "util/parse-options.h"
#include "util/trace-event.h"

#include "util/debug.h"
#include "util/session.h"
17
#include "util/tool.h"
18
#include "util/data.h"
19 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 31
static struct perf_session *session;

32 33 34 35 36 37 38 39 40 41
/* 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 {
42 43
	struct list_head	hash_entry;
	struct rb_node		rb;		/* used for sorting */
44

45
	/*
46
	 * FIXME: perf_evsel__intval() returns u64,
47
	 * so address of lockdep_map should be dealed as 64bit.
48 49 50 51
	 * Is there more better solution?
	 */
	void			*addr;		/* address of lockdep_map, used as ID */
	char			*name;		/* for strcpy(), we cannot use const */
52

53
	unsigned int		nr_acquire;
54
	unsigned int		nr_acquired;
55 56
	unsigned int		nr_contended;
	unsigned int		nr_release;
57

58 59
	unsigned int		nr_readlock;
	unsigned int		nr_trylock;
60

61
	/* these times are in nano sec. */
62
	u64                     avg_wait_time;
63 64 65
	u64			wait_time_total;
	u64			wait_time_min;
	u64			wait_time_max;
66 67 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

	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;
107 108
};

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
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));
168 169 170 171
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189

	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));
190 191 192 193
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
194 195 196 197 198 199 200 201 202 203
	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;
}

204
/* build simple key function one is bigger than two */
205
#define SINGLE_KEY(member)						\
206 207 208 209 210 211 212 213
	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)
214
SINGLE_KEY(avg_wait_time)
215 216 217
SINGLE_KEY(wait_time_total)
SINGLE_KEY(wait_time_max)

218 219 220 221 222 223 224 225 226 227 228 229
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;
}

230 231 232 233 234 235
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
	 */
236 237
	const char		*name;
	int			(*key)(struct lock_stat*, struct lock_stat*);
238 239
};

240 241 242 243 244
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 */
245 246 247 248 249 250

#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),
251
	DEF_KEY_LOCK(avg_wait, avg_wait_time),
252 253 254 255 256 257 258 259 260
	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 }
};

261
static int select_key(void)
262 263 264 265 266 267
{
	int i;

	for (i = 0; keys[i].name; i++) {
		if (!strcmp(keys[i].name, sort_key)) {
			compare = keys[i].key;
268
			return 0;
269 270 271
		}
	}

272 273 274
	pr_err("Unknown compare key: %s\n", sort_key);

	return -1;
275 276 277
}

static void insert_to_result(struct lock_stat *st,
278
			     int (*bigger)(struct lock_stat *, struct lock_stat *))
279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312
{
	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);
}

313
static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
{
	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
329 330
	if (!new->name) {
		free(new);
331
		goto alloc_failed;
D
Davidlohr Bueso 已提交
332
	}
333

D
Davidlohr Bueso 已提交
334
	strcpy(new->name, name);
335 336 337 338 339 340
	new->wait_time_min = ULLONG_MAX;

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

alloc_failed:
341 342
	pr_err("memory allocation failed\n");
	return NULL;
343 344 345
}

struct trace_lock_handler {
346 347
	int (*acquire_event)(struct perf_evsel *evsel,
			     struct perf_sample *sample);
348

349 350
	int (*acquired_event)(struct perf_evsel *evsel,
			      struct perf_sample *sample);
351

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

355 356
	int (*release_event)(struct perf_evsel *evsel,
			     struct perf_sample *sample);
357 358
};

359 360 361 362 363 364 365 366 367 368
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));
369 370 371 372
	if (!seq) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
373 374 375 376 377 378 379
	seq->state = SEQ_STATE_UNINITIALIZED;
	seq->addr = addr;

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

380 381 382 383 384 385 386 387 388
enum broken_state {
	BROKEN_ACQUIRE,
	BROKEN_ACQUIRED,
	BROKEN_CONTENDED,
	BROKEN_RELEASE,
	BROKEN_MAX,
};

static int bad_hist[BROKEN_MAX];
389

390 391 392 393 394
enum acquire_flags {
	TRY_LOCK = 1,
	READ_LOCK = 2,
};

395 396
static int report_lock_acquire_event(struct perf_evsel *evsel,
				     struct perf_sample *sample)
397
{
398
	void *addr;
399 400 401
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
402 403 404
	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");
405

406 407 408
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
409
	if (!ls)
410
		return -ENOMEM;
411
	if (ls->discard)
412
		return 0;
413

414
	ts = thread_stat_findnew(sample->tid);
415
	if (!ts)
416
		return -ENOMEM;
417

418
	seq = get_seq(ts, addr);
419
	if (!seq)
420
		return -ENOMEM;
421

422 423 424
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
	case SEQ_STATE_RELEASED:
425
		if (!flag) {
426 427
			seq->state = SEQ_STATE_ACQUIRING;
		} else {
428
			if (flag & TRY_LOCK)
429
				ls->nr_trylock++;
430
			if (flag & READ_LOCK)
431 432 433 434 435 436 437
				ls->nr_readlock++;
			seq->state = SEQ_STATE_READ_ACQUIRED;
			seq->read_count = 1;
			ls->nr_acquired++;
		}
		break;
	case SEQ_STATE_READ_ACQUIRED:
438
		if (flag & READ_LOCK) {
439 440 441 442 443 444
			seq->read_count++;
			ls->nr_acquired++;
			goto end;
		} else {
			goto broken;
		}
445
		break;
446 447 448 449 450 451
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
broken:
		/* broken lock sequence, discard it */
		ls->discard = 1;
452
		bad_hist[BROKEN_ACQUIRE]++;
453 454 455
		list_del(&seq->list);
		free(seq);
		goto end;
456
	default:
457
		BUG_ON("Unknown state of lock sequence found!\n");
458 459 460
		break;
	}

461
	ls->nr_acquire++;
462
	seq->prev_event_time = sample->time;
463
end:
464
	return 0;
465 466
}

467 468
static int report_lock_acquired_event(struct perf_evsel *evsel,
				      struct perf_sample *sample)
469
{
470
	void *addr;
471 472 473 474
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
	u64 contended_term;
475 476 477 478
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
479

480
	ls = lock_stat_findnew(addr, name);
481
	if (!ls)
482
		return -ENOMEM;
483
	if (ls->discard)
484
		return 0;
485

486
	ts = thread_stat_findnew(sample->tid);
487
	if (!ts)
488
		return -ENOMEM;
489

490
	seq = get_seq(ts, addr);
491
	if (!seq)
492
		return -ENOMEM;
493

494 495 496
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
497
		return 0;
498
	case SEQ_STATE_ACQUIRING:
499
		break;
500
	case SEQ_STATE_CONTENDED:
501
		contended_term = sample->time - seq->prev_event_time;
502 503 504
		ls->wait_time_total += contended_term;
		if (contended_term < ls->wait_time_min)
			ls->wait_time_min = contended_term;
505
		if (ls->wait_time_max < contended_term)
506
			ls->wait_time_max = contended_term;
507
		break;
508 509 510 511 512
	case SEQ_STATE_RELEASED:
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_READ_ACQUIRED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
513
		bad_hist[BROKEN_ACQUIRED]++;
514 515 516
		list_del(&seq->list);
		free(seq);
		goto end;
517
	default:
518
		BUG_ON("Unknown state of lock sequence found!\n");
519 520 521
		break;
	}

522 523
	seq->state = SEQ_STATE_ACQUIRED;
	ls->nr_acquired++;
524
	ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
525
	seq->prev_event_time = sample->time;
526
end:
527
	return 0;
528 529
}

530 531
static int report_lock_contended_event(struct perf_evsel *evsel,
				       struct perf_sample *sample)
532
{
533
	void *addr;
534 535 536
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
537 538 539 540
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
541

542
	ls = lock_stat_findnew(addr, name);
543
	if (!ls)
544
		return -ENOMEM;
545
	if (ls->discard)
546
		return 0;
547

548
	ts = thread_stat_findnew(sample->tid);
549
	if (!ts)
550
		return -ENOMEM;
551

552
	seq = get_seq(ts, addr);
553
	if (!seq)
554
		return -ENOMEM;
555

556 557 558
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
559
		return 0;
560
	case SEQ_STATE_ACQUIRING:
561
		break;
562 563 564 565 566 567
	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;
568
		bad_hist[BROKEN_CONTENDED]++;
569 570 571
		list_del(&seq->list);
		free(seq);
		goto end;
572
	default:
573
		BUG_ON("Unknown state of lock sequence found!\n");
574 575 576
		break;
	}

577 578
	seq->state = SEQ_STATE_CONTENDED;
	ls->nr_contended++;
579
	ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
580
	seq->prev_event_time = sample->time;
581
end:
582
	return 0;
583 584
}

585 586
static int report_lock_release_event(struct perf_evsel *evsel,
				     struct perf_sample *sample)
587
{
588
	void *addr;
589 590 591
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
592 593
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
594

595 596 597
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
598
	if (!ls)
599
		return -ENOMEM;
600
	if (ls->discard)
601
		return 0;
602

603
	ts = thread_stat_findnew(sample->tid);
604
	if (!ts)
605
		return -ENOMEM;
606

607
	seq = get_seq(ts, addr);
608
	if (!seq)
609
		return -ENOMEM;
610

611 612 613 614 615 616 617 618 619 620
	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++;
621 622
			goto end;
		}
623 624 625 626 627 628
		break;
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
	case SEQ_STATE_RELEASED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
629
		bad_hist[BROKEN_RELEASE]++;
630
		goto free_seq;
631
	default:
632
		BUG_ON("Unknown state of lock sequence found!\n");
633 634 635
		break;
	}

636 637 638 639
	ls->nr_release++;
free_seq:
	list_del(&seq->list);
	free(seq);
640
end:
641
	return 0;
642 643 644 645
}

/* lock oriented handlers */
/* TODO: handlers for CPU oriented, thread oriented */
646 647 648 649 650
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,
651 652 653 654
};

static struct trace_lock_handler *trace_handler;

655
static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
656
					     struct perf_sample *sample)
657
{
658
	if (trace_handler->acquire_event)
659 660
		return trace_handler->acquire_event(evsel, sample);
	return 0;
661 662
}

663
static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
664
					      struct perf_sample *sample)
665
{
666
	if (trace_handler->acquired_event)
667 668
		return trace_handler->acquired_event(evsel, sample);
	return 0;
669 670
}

671
static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
672
					      struct perf_sample *sample)
673
{
674
	if (trace_handler->contended_event)
675 676
		return trace_handler->contended_event(evsel, sample);
	return 0;
677 678
}

679
static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
680
					    struct perf_sample *sample)
681
{
682
	if (trace_handler->release_event)
683 684
		return trace_handler->release_event(evsel, sample);
	return 0;
685 686
}

687 688 689 690 691 692 693 694
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");
695
	pr_info("bad: %d, total: %d\n", bad, total);
696
	pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
697 698 699 700 701
	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]);
}

702 703 704 705 706
/* TODO: various way to print, coloring, nano or milli sec */
static void print_result(void)
{
	struct lock_stat *st;
	char cut_name[20];
707
	int bad, total;
708

709 710 711
	pr_info("%20s ", "Name");
	pr_info("%10s ", "acquired");
	pr_info("%10s ", "contended");
712

713
	pr_info("%15s ", "avg wait (ns)");
714 715 716
	pr_info("%15s ", "total wait (ns)");
	pr_info("%15s ", "max wait (ns)");
	pr_info("%15s ", "min wait (ns)");
717

718
	pr_info("\n\n");
719

720
	bad = total = 0;
721
	while ((st = pop_from_result())) {
722 723 724 725 726
		total++;
		if (st->discard) {
			bad++;
			continue;
		}
727 728 729 730
		bzero(cut_name, 20);

		if (strlen(st->name) < 16) {
			/* output raw name */
731
			pr_info("%20s ", st->name);
732 733 734 735 736 737 738
		} 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 */
739
			pr_info("%20s ", cut_name);
740 741
		}

742 743
		pr_info("%10u ", st->nr_acquired);
		pr_info("%10u ", st->nr_contended);
744

745
		pr_info("%15" PRIu64 " ", st->avg_wait_time);
746 747 748
		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 ?
749
		       0 : st->wait_time_min);
750
		pr_info("\n");
751
	}
752

753
	print_bad_events(bad, total);
754 755
}

756
static bool info_threads, info_map;
757 758 759 760 761 762 763 764 765 766 767 768 769

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);
770
		pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
771 772 773 774
		node = rb_next(node);
	};
}

775 776 777 778 779
static void dump_map(void)
{
	unsigned int i;
	struct lock_stat *st;

780
	pr_info("Address of instance: name of class\n");
781 782
	for (i = 0; i < LOCKHASH_SIZE; i++) {
		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
783
			pr_info(" %p: %s\n", st->addr, st->name);
784 785 786 787
		}
	}
}

788
static int dump_info(void)
789
{
790 791
	int rc = 0;

792 793 794 795
	if (info_threads)
		dump_threads();
	else if (info_map)
		dump_map();
796 797 798 799 800 801
	else {
		rc = -1;
		pr_err("Unknown type of information\n");
	}

	return rc;
802 803
}

804 805 806
typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
				  struct perf_sample *sample);

807
static int process_sample_event(struct perf_tool *tool __maybe_unused,
808
				union perf_event *event,
809
				struct perf_sample *sample,
810
				struct perf_evsel *evsel,
811
				struct machine *machine)
812
{
813 814
	struct thread *thread = machine__findnew_thread(machine, sample->pid,
							sample->tid);
815 816 817

	if (thread == NULL) {
		pr_debug("problem processing %d event, skipping it.\n",
818
			event->header.type);
819 820 821
		return -1;
	}

822 823
	if (evsel->handler != NULL) {
		tracepoint_handler f = evsel->handler;
824 825 826 827
		return f(evsel, sample);
	}

	return 0;
828 829
}

D
Davidlohr Bueso 已提交
830 831 832 833 834 835 836 837 838 839 840 841
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);
		}
	}
}

842 843 844 845 846 847 848
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 */
};

D
Davidlohr Bueso 已提交
849
static int __cmd_report(bool display_info)
850
{
D
Davidlohr Bueso 已提交
851
	int err = -EINVAL;
852 853 854
	struct perf_tool eops = {
		.sample		 = process_sample_event,
		.comm		 = perf_event__process_comm,
855
		.ordered_events	 = true,
856
	};
857 858 859 860
	struct perf_data_file file = {
		.path = input_name,
		.mode = PERF_DATA_MODE_READ,
	};
D
Davidlohr Bueso 已提交
861

862
	session = perf_session__new(&file, false, &eops);
863 864
	if (!session) {
		pr_err("Initializing perf session failed\n");
865
		return -1;
866
	}
867

868
	symbol__init(&session->header.env);
869

D
Davidlohr Bueso 已提交
870 871 872
	if (!perf_session__has_traces(session, "lock record"))
		goto out_delete;

873 874
	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
		pr_err("Initializing perf session tracepoint handlers failed\n");
D
Davidlohr Bueso 已提交
875
		goto out_delete;
876 877
	}

D
Davidlohr Bueso 已提交
878 879
	if (select_key())
		goto out_delete;
880

881
	err = perf_session__process_events(session);
D
Davidlohr Bueso 已提交
882 883
	if (err)
		goto out_delete;
884 885

	setup_pager();
D
Davidlohr Bueso 已提交
886 887 888 889 890 891
	if (display_info) /* used for info subcommand */
		err = dump_info();
	else {
		sort_result();
		print_result();
	}
892

D
Davidlohr Bueso 已提交
893 894 895
out_delete:
	perf_session__delete(session);
	return err;
896 897 898 899
}

static int __cmd_record(int argc, const char **argv)
{
900
	const char *record_args[] = {
901
		"record", "-R", "-m", "1024", "-c", "1",
902
	};
D
Davidlohr Bueso 已提交
903
	unsigned int rec_argc, i, j, ret;
904 905
	const char **rec_argv;

906
	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
907
		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
908 909
				pr_err("tracepoint %s is not enabled. "
				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
910
				       lock_tracepoints[i].name);
911 912 913 914
				return 1;
		}
	}

915
	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
916 917
	/* factor of 2 is for -e in front of each tracepoint */
	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
918

919
	rec_argv = calloc(rec_argc + 1, sizeof(char *));
D
Davidlohr Bueso 已提交
920
	if (!rec_argv)
921 922
		return -ENOMEM;

923 924 925
	for (i = 0; i < ARRAY_SIZE(record_args); i++)
		rec_argv[i] = strdup(record_args[i]);

926 927
	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
		rec_argv[i++] = "-e";
928
		rec_argv[i++] = strdup(lock_tracepoints[j].name);
929 930
	}

931 932 933 934 935
	for (j = 1; j < (unsigned int)argc; j++, i++)
		rec_argv[i] = argv[j];

	BUG_ON(i != rec_argc);

D
Davidlohr Bueso 已提交
936 937 938
	ret = cmd_record(i, rec_argv, NULL);
	free(rec_argv);
	return ret;
939 940
}

941
int cmd_lock(int argc, const char **argv, const char *prefix __maybe_unused)
942
{
943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
	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)"),
	OPT_END()
	};
	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"),
	OPT_END()
	};
	const struct option report_options[] = {
	OPT_STRING('k', "key", &sort_key, "acquired",
958
		    "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
959 960 961 962 963 964 965
	/* TODO: type */
	OPT_END()
	};
	const char * const info_usage[] = {
		"perf lock info [<options>]",
		NULL
	};
966 967 968 969
	const char *const lock_subcommands[] = { "record", "report", "script",
						 "info", NULL };
	const char *lock_usage[] = {
		NULL,
970 971 972 973 974 975
		NULL
	};
	const char * const report_usage[] = {
		"perf lock report [<options>]",
		NULL
	};
976
	unsigned int i;
977
	int rc = 0;
978 979 980 981

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

982 983
	argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
					lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
984 985 986 987 988
	if (!argc)
		usage_with_options(lock_usage, lock_options);

	if (!strncmp(argv[0], "rec", 3)) {
		return __cmd_record(argc, argv);
989 990
	} else if (!strncmp(argv[0], "report", 6)) {
		trace_handler = &report_lock_ops;
991 992
		if (argc) {
			argc = parse_options(argc, argv,
993
					     report_options, report_usage, 0);
994
			if (argc)
995
				usage_with_options(report_usage, report_options);
996
		}
D
Davidlohr Bueso 已提交
997
		rc = __cmd_report(false);
998 999 1000
	} else if (!strcmp(argv[0], "script")) {
		/* Aliased to 'perf script' */
		return cmd_script(argc, argv, prefix);
1001 1002 1003 1004 1005 1006 1007
	} 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);
		}
1008 1009
		/* recycling report_lock_ops */
		trace_handler = &report_lock_ops;
D
Davidlohr Bueso 已提交
1010
		rc = __cmd_report(true);
1011 1012 1013 1014
	} else {
		usage_with_options(lock_usage, lock_options);
	}

1015
	return rc;
1016
}