builtin-lock.c 23.2 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2
#include <errno.h>
3
#include <inttypes.h>
4 5 6
#include "builtin.h"
#include "perf.h"

7
#include "util/evlist.h" // for struct evsel_str_handler
8
#include "util/evsel.h"
9 10 11 12 13
#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
#include <linux/zalloc.h>
33

34 35
static struct perf_session *session;

36 37 38 39 40 41 42 43 44 45
/* 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 {
46 47
	struct list_head	hash_entry;
	struct rb_node		rb;		/* used for sorting */
48

49
	/*
50
	 * FIXME: perf_evsel__intval() returns u64,
51
	 * so address of lockdep_map should be dealed as 64bit.
52 53 54 55
	 * Is there more better solution?
	 */
	void			*addr;		/* address of lockdep_map, used as ID */
	char			*name;		/* for strcpy(), we cannot use const */
56

57
	unsigned int		nr_acquire;
58
	unsigned int		nr_acquired;
59 60
	unsigned int		nr_contended;
	unsigned int		nr_release;
61

62 63
	unsigned int		nr_readlock;
	unsigned int		nr_trylock;
64

65
	/* these times are in nano sec. */
66
	u64                     avg_wait_time;
67 68 69
	u64			wait_time_total;
	u64			wait_time_min;
	u64			wait_time_max;
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 99 100 101 102 103 104 105 106 107 108 109 110

	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;
111 112
};

113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
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));
172 173 174 175
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193

	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));
194 195 196 197
	if (!st) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
198 199 200 201 202 203 204 205 206 207
	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;
}

208
/* build simple key function one is bigger than two */
209
#define SINGLE_KEY(member)						\
210 211 212 213 214 215 216 217
	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)
218
SINGLE_KEY(avg_wait_time)
219 220 221
SINGLE_KEY(wait_time_total)
SINGLE_KEY(wait_time_max)

222 223 224 225 226 227 228 229 230 231 232 233
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;
}

234 235 236 237 238 239
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
	 */
240 241
	const char		*name;
	int			(*key)(struct lock_stat*, struct lock_stat*);
242 243
};

244 245 246 247 248
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 */
249 250 251 252 253 254

#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),
255
	DEF_KEY_LOCK(avg_wait, avg_wait_time),
256 257 258 259 260 261 262 263 264
	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 }
};

265
static int select_key(void)
266 267 268 269 270 271
{
	int i;

	for (i = 0; keys[i].name; i++) {
		if (!strcmp(keys[i].name, sort_key)) {
			compare = keys[i].key;
272
			return 0;
273 274 275
		}
	}

276 277 278
	pr_err("Unknown compare key: %s\n", sort_key);

	return -1;
279 280 281
}

static void insert_to_result(struct lock_stat *st,
282
			     int (*bigger)(struct lock_stat *, struct lock_stat *))
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316
{
	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);
}

317
static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
{
	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
333 334
	if (!new->name) {
		free(new);
335
		goto alloc_failed;
D
Davidlohr Bueso 已提交
336
	}
337

D
Davidlohr Bueso 已提交
338
	strcpy(new->name, name);
339 340 341 342 343 344
	new->wait_time_min = ULLONG_MAX;

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

alloc_failed:
345 346
	pr_err("memory allocation failed\n");
	return NULL;
347 348 349
}

struct trace_lock_handler {
350
	int (*acquire_event)(struct evsel *evsel,
351
			     struct perf_sample *sample);
352

353
	int (*acquired_event)(struct evsel *evsel,
354
			      struct perf_sample *sample);
355

356
	int (*contended_event)(struct evsel *evsel,
357
			       struct perf_sample *sample);
358

359
	int (*release_event)(struct evsel *evsel,
360
			     struct perf_sample *sample);
361 362
};

363 364 365 366 367 368 369 370 371 372
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));
373 374 375 376
	if (!seq) {
		pr_err("memory allocation failed\n");
		return NULL;
	}
377 378 379 380 381 382 383
	seq->state = SEQ_STATE_UNINITIALIZED;
	seq->addr = addr;

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

384 385 386 387 388 389 390 391 392
enum broken_state {
	BROKEN_ACQUIRE,
	BROKEN_ACQUIRED,
	BROKEN_CONTENDED,
	BROKEN_RELEASE,
	BROKEN_MAX,
};

static int bad_hist[BROKEN_MAX];
393

394 395 396 397 398
enum acquire_flags {
	TRY_LOCK = 1,
	READ_LOCK = 2,
};

399
static int report_lock_acquire_event(struct evsel *evsel,
400
				     struct perf_sample *sample)
401
{
402
	void *addr;
403 404 405
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
406 407 408
	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");
409

410 411 412
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
413
	if (!ls)
414
		return -ENOMEM;
415
	if (ls->discard)
416
		return 0;
417

418
	ts = thread_stat_findnew(sample->tid);
419
	if (!ts)
420
		return -ENOMEM;
421

422
	seq = get_seq(ts, addr);
423
	if (!seq)
424
		return -ENOMEM;
425

426 427 428
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
	case SEQ_STATE_RELEASED:
429
		if (!flag) {
430 431
			seq->state = SEQ_STATE_ACQUIRING;
		} else {
432
			if (flag & TRY_LOCK)
433
				ls->nr_trylock++;
434
			if (flag & READ_LOCK)
435 436 437 438 439 440 441
				ls->nr_readlock++;
			seq->state = SEQ_STATE_READ_ACQUIRED;
			seq->read_count = 1;
			ls->nr_acquired++;
		}
		break;
	case SEQ_STATE_READ_ACQUIRED:
442
		if (flag & READ_LOCK) {
443 444 445 446 447 448
			seq->read_count++;
			ls->nr_acquired++;
			goto end;
		} else {
			goto broken;
		}
449
		break;
450 451 452 453 454 455
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
broken:
		/* broken lock sequence, discard it */
		ls->discard = 1;
456
		bad_hist[BROKEN_ACQUIRE]++;
457
		list_del_init(&seq->list);
458 459
		free(seq);
		goto end;
460
	default:
461
		BUG_ON("Unknown state of lock sequence found!\n");
462 463 464
		break;
	}

465
	ls->nr_acquire++;
466
	seq->prev_event_time = sample->time;
467
end:
468
	return 0;
469 470
}

471
static int report_lock_acquired_event(struct evsel *evsel,
472
				      struct perf_sample *sample)
473
{
474
	void *addr;
475 476 477 478
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
	u64 contended_term;
479 480 481 482
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
483

484
	ls = lock_stat_findnew(addr, name);
485
	if (!ls)
486
		return -ENOMEM;
487
	if (ls->discard)
488
		return 0;
489

490
	ts = thread_stat_findnew(sample->tid);
491
	if (!ts)
492
		return -ENOMEM;
493

494
	seq = get_seq(ts, addr);
495
	if (!seq)
496
		return -ENOMEM;
497

498 499 500
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
501
		return 0;
502
	case SEQ_STATE_ACQUIRING:
503
		break;
504
	case SEQ_STATE_CONTENDED:
505
		contended_term = sample->time - seq->prev_event_time;
506 507 508
		ls->wait_time_total += contended_term;
		if (contended_term < ls->wait_time_min)
			ls->wait_time_min = contended_term;
509
		if (ls->wait_time_max < contended_term)
510
			ls->wait_time_max = contended_term;
511
		break;
512 513 514 515 516
	case SEQ_STATE_RELEASED:
	case SEQ_STATE_ACQUIRED:
	case SEQ_STATE_READ_ACQUIRED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
517
		bad_hist[BROKEN_ACQUIRED]++;
518
		list_del_init(&seq->list);
519 520
		free(seq);
		goto end;
521
	default:
522
		BUG_ON("Unknown state of lock sequence found!\n");
523 524 525
		break;
	}

526 527
	seq->state = SEQ_STATE_ACQUIRED;
	ls->nr_acquired++;
528
	ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
529
	seq->prev_event_time = sample->time;
530
end:
531
	return 0;
532 533
}

534
static int report_lock_contended_event(struct evsel *evsel,
535
				       struct perf_sample *sample)
536
{
537
	void *addr;
538 539 540
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
541 542 543 544
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");

	memcpy(&addr, &tmp, sizeof(void *));
545

546
	ls = lock_stat_findnew(addr, name);
547
	if (!ls)
548
		return -ENOMEM;
549
	if (ls->discard)
550
		return 0;
551

552
	ts = thread_stat_findnew(sample->tid);
553
	if (!ts)
554
		return -ENOMEM;
555

556
	seq = get_seq(ts, addr);
557
	if (!seq)
558
		return -ENOMEM;
559

560 561 562
	switch (seq->state) {
	case SEQ_STATE_UNINITIALIZED:
		/* orphan event, do nothing */
563
		return 0;
564
	case SEQ_STATE_ACQUIRING:
565
		break;
566 567 568 569 570 571
	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;
572
		bad_hist[BROKEN_CONTENDED]++;
573
		list_del_init(&seq->list);
574 575
		free(seq);
		goto end;
576
	default:
577
		BUG_ON("Unknown state of lock sequence found!\n");
578 579 580
		break;
	}

581 582
	seq->state = SEQ_STATE_CONTENDED;
	ls->nr_contended++;
583
	ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
584
	seq->prev_event_time = sample->time;
585
end:
586
	return 0;
587 588
}

589
static int report_lock_release_event(struct evsel *evsel,
590
				     struct perf_sample *sample)
591
{
592
	void *addr;
593 594 595
	struct lock_stat *ls;
	struct thread_stat *ts;
	struct lock_seq_stat *seq;
596 597
	const char *name = perf_evsel__strval(evsel, sample, "name");
	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
598

599 600 601
	memcpy(&addr, &tmp, sizeof(void *));

	ls = lock_stat_findnew(addr, name);
602
	if (!ls)
603
		return -ENOMEM;
604
	if (ls->discard)
605
		return 0;
606

607
	ts = thread_stat_findnew(sample->tid);
608
	if (!ts)
609
		return -ENOMEM;
610

611
	seq = get_seq(ts, addr);
612
	if (!seq)
613
		return -ENOMEM;
614

615 616 617 618 619 620 621 622 623 624
	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++;
625 626
			goto end;
		}
627 628 629 630 631 632
		break;
	case SEQ_STATE_ACQUIRING:
	case SEQ_STATE_CONTENDED:
	case SEQ_STATE_RELEASED:
		/* broken lock sequence, discard it */
		ls->discard = 1;
633
		bad_hist[BROKEN_RELEASE]++;
634
		goto free_seq;
635
	default:
636
		BUG_ON("Unknown state of lock sequence found!\n");
637 638 639
		break;
	}

640 641
	ls->nr_release++;
free_seq:
642
	list_del_init(&seq->list);
643
	free(seq);
644
end:
645
	return 0;
646 647 648 649
}

/* lock oriented handlers */
/* TODO: handlers for CPU oriented, thread oriented */
650 651 652 653 654
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,
655 656 657 658
};

static struct trace_lock_handler *trace_handler;

659
static int perf_evsel__process_lock_acquire(struct evsel *evsel,
660
					     struct perf_sample *sample)
661
{
662
	if (trace_handler->acquire_event)
663 664
		return trace_handler->acquire_event(evsel, sample);
	return 0;
665 666
}

667
static int perf_evsel__process_lock_acquired(struct evsel *evsel,
668
					      struct perf_sample *sample)
669
{
670
	if (trace_handler->acquired_event)
671 672
		return trace_handler->acquired_event(evsel, sample);
	return 0;
673 674
}

675
static int perf_evsel__process_lock_contended(struct evsel *evsel,
676
					      struct perf_sample *sample)
677
{
678
	if (trace_handler->contended_event)
679 680
		return trace_handler->contended_event(evsel, sample);
	return 0;
681 682
}

683
static int perf_evsel__process_lock_release(struct evsel *evsel,
684
					    struct perf_sample *sample)
685
{
686
	if (trace_handler->release_event)
687 688
		return trace_handler->release_event(evsel, sample);
	return 0;
689 690
}

691 692 693 694 695 696 697 698
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");
699
	pr_info("bad: %d, total: %d\n", bad, total);
700
	pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
701 702 703 704 705
	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]);
}

706 707 708 709 710
/* TODO: various way to print, coloring, nano or milli sec */
static void print_result(void)
{
	struct lock_stat *st;
	char cut_name[20];
711
	int bad, total;
712

713 714 715
	pr_info("%20s ", "Name");
	pr_info("%10s ", "acquired");
	pr_info("%10s ", "contended");
716

717
	pr_info("%15s ", "avg wait (ns)");
718 719 720
	pr_info("%15s ", "total wait (ns)");
	pr_info("%15s ", "max wait (ns)");
	pr_info("%15s ", "min wait (ns)");
721

722
	pr_info("\n\n");
723

724
	bad = total = 0;
725
	while ((st = pop_from_result())) {
726 727 728 729 730
		total++;
		if (st->discard) {
			bad++;
			continue;
		}
731 732 733 734
		bzero(cut_name, 20);

		if (strlen(st->name) < 16) {
			/* output raw name */
735
			pr_info("%20s ", st->name);
736 737 738 739 740 741 742
		} 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 */
743
			pr_info("%20s ", cut_name);
744 745
		}

746 747
		pr_info("%10u ", st->nr_acquired);
		pr_info("%10u ", st->nr_contended);
748

749
		pr_info("%15" PRIu64 " ", st->avg_wait_time);
750 751 752
		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 ?
753
		       0 : st->wait_time_min);
754
		pr_info("\n");
755
	}
756

757
	print_bad_events(bad, total);
758 759
}

760
static bool info_threads, info_map;
761 762 763 764 765 766 767 768 769 770 771 772 773

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);
774
		pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
775
		node = rb_next(node);
776
		thread__put(t);
777 778 779
	};
}

780 781 782 783 784
static void dump_map(void)
{
	unsigned int i;
	struct lock_stat *st;

785
	pr_info("Address of instance: name of class\n");
786 787
	for (i = 0; i < LOCKHASH_SIZE; i++) {
		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
788
			pr_info(" %p: %s\n", st->addr, st->name);
789 790 791 792
		}
	}
}

793
static int dump_info(void)
794
{
795 796
	int rc = 0;

797 798 799 800
	if (info_threads)
		dump_threads();
	else if (info_map)
		dump_map();
801 802 803 804 805 806
	else {
		rc = -1;
		pr_err("Unknown type of information\n");
	}

	return rc;
807 808
}

809
typedef int (*tracepoint_handler)(struct evsel *evsel,
810 811
				  struct perf_sample *sample);

812
static int process_sample_event(struct perf_tool *tool __maybe_unused,
813
				union perf_event *event,
814
				struct perf_sample *sample,
815
				struct evsel *evsel,
816
				struct machine *machine)
817
{
818
	int err = 0;
819 820
	struct thread *thread = machine__findnew_thread(machine, sample->pid,
							sample->tid);
821 822 823

	if (thread == NULL) {
		pr_debug("problem processing %d event, skipping it.\n",
824
			event->header.type);
825 826 827
		return -1;
	}

828 829
	if (evsel->handler != NULL) {
		tracepoint_handler f = evsel->handler;
830
		err = f(evsel, sample);
831 832
	}

833 834 835
	thread__put(thread);

	return err;
836 837
}

D
Davidlohr Bueso 已提交
838 839 840 841 842 843 844 845 846 847 848 849
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);
		}
	}
}

850
static const struct evsel_str_handler lock_tracepoints[] = {
851 852 853 854 855 856
	{ "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 */
};

857 858
static bool force;

D
Davidlohr Bueso 已提交
859
static int __cmd_report(bool display_info)
860
{
D
Davidlohr Bueso 已提交
861
	int err = -EINVAL;
862 863 864
	struct perf_tool eops = {
		.sample		 = process_sample_event,
		.comm		 = perf_event__process_comm,
865
		.namespaces	 = perf_event__process_namespaces,
866
		.ordered_events	 = true,
867
	};
868
	struct perf_data data = {
J
Jiri Olsa 已提交
869 870 871
		.path  = input_name,
		.mode  = PERF_DATA_MODE_READ,
		.force = force,
872
	};
D
Davidlohr Bueso 已提交
873

874
	session = perf_session__new(&data, false, &eops);
875 876
	if (!session) {
		pr_err("Initializing perf session failed\n");
877
		return -1;
878
	}
879

880
	symbol__init(&session->header.env);
881

D
Davidlohr Bueso 已提交
882 883 884
	if (!perf_session__has_traces(session, "lock record"))
		goto out_delete;

885 886
	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
		pr_err("Initializing perf session tracepoint handlers failed\n");
D
Davidlohr Bueso 已提交
887
		goto out_delete;
888 889
	}

D
Davidlohr Bueso 已提交
890 891
	if (select_key())
		goto out_delete;
892

893
	err = perf_session__process_events(session);
D
Davidlohr Bueso 已提交
894 895
	if (err)
		goto out_delete;
896 897

	setup_pager();
D
Davidlohr Bueso 已提交
898 899 900 901 902 903
	if (display_info) /* used for info subcommand */
		err = dump_info();
	else {
		sort_result();
		print_result();
	}
904

D
Davidlohr Bueso 已提交
905 906 907
out_delete:
	perf_session__delete(session);
	return err;
908 909 910 911
}

static int __cmd_record(int argc, const char **argv)
{
912
	const char *record_args[] = {
913
		"record", "-R", "-m", "1024", "-c", "1",
914
	};
D
Davidlohr Bueso 已提交
915
	unsigned int rec_argc, i, j, ret;
916 917
	const char **rec_argv;

918
	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
919
		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
920 921
				pr_err("tracepoint %s is not enabled. "
				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
922
				       lock_tracepoints[i].name);
923 924 925 926
				return 1;
		}
	}

927
	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
928 929
	/* factor of 2 is for -e in front of each tracepoint */
	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
930

931
	rec_argv = calloc(rec_argc + 1, sizeof(char *));
D
Davidlohr Bueso 已提交
932
	if (!rec_argv)
933 934
		return -ENOMEM;

935 936 937
	for (i = 0; i < ARRAY_SIZE(record_args); i++)
		rec_argv[i] = strdup(record_args[i]);

938 939
	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
		rec_argv[i++] = "-e";
940
		rec_argv[i++] = strdup(lock_tracepoints[j].name);
941 942
	}

943 944 945 946 947
	for (j = 1; j < (unsigned int)argc; j++, i++)
		rec_argv[i] = argv[j];

	BUG_ON(i != rec_argc);

948
	ret = cmd_record(i, rec_argv);
D
Davidlohr Bueso 已提交
949 950
	free(rec_argv);
	return ret;
951 952
}

953
int cmd_lock(int argc, const char **argv)
954
{
955 956 957 958
	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"),
959
	OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
960 961 962
	OPT_END()
	};

963 964 965 966 967
	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)"),
968
	OPT_PARENT(lock_options)
969
	};
970

971 972
	const struct option report_options[] = {
	OPT_STRING('k', "key", &sort_key, "acquired",
973
		    "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
974
	/* TODO: type */
975
	OPT_PARENT(lock_options)
976
	};
977

978 979 980 981
	const char * const info_usage[] = {
		"perf lock info [<options>]",
		NULL
	};
982 983 984 985
	const char *const lock_subcommands[] = { "record", "report", "script",
						 "info", NULL };
	const char *lock_usage[] = {
		NULL,
986 987 988 989 990 991
		NULL
	};
	const char * const report_usage[] = {
		"perf lock report [<options>]",
		NULL
	};
992
	unsigned int i;
993
	int rc = 0;
994 995 996 997

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

998 999
	argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
					lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1000 1001 1002 1003 1004
	if (!argc)
		usage_with_options(lock_usage, lock_options);

	if (!strncmp(argv[0], "rec", 3)) {
		return __cmd_record(argc, argv);
1005 1006
	} else if (!strncmp(argv[0], "report", 6)) {
		trace_handler = &report_lock_ops;
1007 1008
		if (argc) {
			argc = parse_options(argc, argv,
1009
					     report_options, report_usage, 0);
1010
			if (argc)
1011
				usage_with_options(report_usage, report_options);
1012
		}
D
Davidlohr Bueso 已提交
1013
		rc = __cmd_report(false);
1014 1015
	} else if (!strcmp(argv[0], "script")) {
		/* Aliased to 'perf script' */
1016
		return cmd_script(argc, argv);
1017 1018 1019 1020 1021 1022 1023
	} 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);
		}
1024 1025
		/* recycling report_lock_ops */
		trace_handler = &report_lock_ops;
D
Davidlohr Bueso 已提交
1026
		rc = __cmd_report(true);
1027 1028 1029 1030
	} else {
		usage_with_options(lock_usage, lock_options);
	}

1031
	return rc;
1032
}