xref: /linux/tools/perf/builtin-lock.c (revision f49f4ab95c301dbccad0efe85296d908b8ae7ad4)
1 #include "builtin.h"
2 #include "perf.h"
3 
4 #include "util/evlist.h"
5 #include "util/evsel.h"
6 #include "util/util.h"
7 #include "util/cache.h"
8 #include "util/symbol.h"
9 #include "util/thread.h"
10 #include "util/header.h"
11 
12 #include "util/parse-options.h"
13 #include "util/trace-event.h"
14 
15 #include "util/debug.h"
16 #include "util/session.h"
17 #include "util/tool.h"
18 
19 #include <sys/types.h>
20 #include <sys/prctl.h>
21 #include <semaphore.h>
22 #include <pthread.h>
23 #include <math.h>
24 #include <limits.h>
25 
26 #include <linux/list.h>
27 #include <linux/hash.h>
28 
29 static struct perf_session *session;
30 
31 /* based on kernel/lockdep.c */
32 #define LOCKHASH_BITS		12
33 #define LOCKHASH_SIZE		(1UL << LOCKHASH_BITS)
34 
35 static struct list_head lockhash_table[LOCKHASH_SIZE];
36 
37 #define __lockhashfn(key)	hash_long((unsigned long)key, LOCKHASH_BITS)
38 #define lockhashentry(key)	(lockhash_table + __lockhashfn((key)))
39 
40 struct lock_stat {
41 	struct list_head	hash_entry;
42 	struct rb_node		rb;		/* used for sorting */
43 
44 	/*
45 	 * FIXME: perf_evsel__intval() returns u64,
46 	 * so address of lockdep_map should be dealed as 64bit.
47 	 * Is there more better solution?
48 	 */
49 	void			*addr;		/* address of lockdep_map, used as ID */
50 	char			*name;		/* for strcpy(), we cannot use const */
51 
52 	unsigned int		nr_acquire;
53 	unsigned int		nr_acquired;
54 	unsigned int		nr_contended;
55 	unsigned int		nr_release;
56 
57 	unsigned int		nr_readlock;
58 	unsigned int		nr_trylock;
59 	/* these times are in nano sec. */
60 	u64			wait_time_total;
61 	u64			wait_time_min;
62 	u64			wait_time_max;
63 
64 	int			discard; /* flag of blacklist */
65 };
66 
67 /*
68  * States of lock_seq_stat
69  *
70  * UNINITIALIZED is required for detecting first event of acquire.
71  * As the nature of lock events, there is no guarantee
72  * that the first event for the locks are acquire,
73  * it can be acquired, contended or release.
74  */
75 #define SEQ_STATE_UNINITIALIZED      0	       /* initial state */
76 #define SEQ_STATE_RELEASED	1
77 #define SEQ_STATE_ACQUIRING	2
78 #define SEQ_STATE_ACQUIRED	3
79 #define SEQ_STATE_READ_ACQUIRED	4
80 #define SEQ_STATE_CONTENDED	5
81 
82 /*
83  * MAX_LOCK_DEPTH
84  * Imported from include/linux/sched.h.
85  * Should this be synchronized?
86  */
87 #define MAX_LOCK_DEPTH 48
88 
89 /*
90  * struct lock_seq_stat:
91  * Place to put on state of one lock sequence
92  * 1) acquire -> acquired -> release
93  * 2) acquire -> contended -> acquired -> release
94  * 3) acquire (with read or try) -> release
95  * 4) Are there other patterns?
96  */
97 struct lock_seq_stat {
98 	struct list_head        list;
99 	int			state;
100 	u64			prev_event_time;
101 	void                    *addr;
102 
103 	int                     read_count;
104 };
105 
106 struct thread_stat {
107 	struct rb_node		rb;
108 
109 	u32                     tid;
110 	struct list_head        seq_list;
111 };
112 
113 static struct rb_root		thread_stats;
114 
115 static struct thread_stat *thread_stat_find(u32 tid)
116 {
117 	struct rb_node *node;
118 	struct thread_stat *st;
119 
120 	node = thread_stats.rb_node;
121 	while (node) {
122 		st = container_of(node, struct thread_stat, rb);
123 		if (st->tid == tid)
124 			return st;
125 		else if (tid < st->tid)
126 			node = node->rb_left;
127 		else
128 			node = node->rb_right;
129 	}
130 
131 	return NULL;
132 }
133 
134 static void thread_stat_insert(struct thread_stat *new)
135 {
136 	struct rb_node **rb = &thread_stats.rb_node;
137 	struct rb_node *parent = NULL;
138 	struct thread_stat *p;
139 
140 	while (*rb) {
141 		p = container_of(*rb, struct thread_stat, rb);
142 		parent = *rb;
143 
144 		if (new->tid < p->tid)
145 			rb = &(*rb)->rb_left;
146 		else if (new->tid > p->tid)
147 			rb = &(*rb)->rb_right;
148 		else
149 			BUG_ON("inserting invalid thread_stat\n");
150 	}
151 
152 	rb_link_node(&new->rb, parent, rb);
153 	rb_insert_color(&new->rb, &thread_stats);
154 }
155 
156 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
157 {
158 	struct thread_stat *st;
159 
160 	st = thread_stat_find(tid);
161 	if (st)
162 		return st;
163 
164 	st = zalloc(sizeof(struct thread_stat));
165 	if (!st) {
166 		pr_err("memory allocation failed\n");
167 		return NULL;
168 	}
169 
170 	st->tid = tid;
171 	INIT_LIST_HEAD(&st->seq_list);
172 
173 	thread_stat_insert(st);
174 
175 	return st;
176 }
177 
178 static struct thread_stat *thread_stat_findnew_first(u32 tid);
179 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
180 	thread_stat_findnew_first;
181 
182 static struct thread_stat *thread_stat_findnew_first(u32 tid)
183 {
184 	struct thread_stat *st;
185 
186 	st = zalloc(sizeof(struct thread_stat));
187 	if (!st) {
188 		pr_err("memory allocation failed\n");
189 		return NULL;
190 	}
191 	st->tid = tid;
192 	INIT_LIST_HEAD(&st->seq_list);
193 
194 	rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
195 	rb_insert_color(&st->rb, &thread_stats);
196 
197 	thread_stat_findnew = thread_stat_findnew_after_first;
198 	return st;
199 }
200 
201 /* build simple key function one is bigger than two */
202 #define SINGLE_KEY(member)						\
203 	static int lock_stat_key_ ## member(struct lock_stat *one,	\
204 					 struct lock_stat *two)		\
205 	{								\
206 		return one->member > two->member;			\
207 	}
208 
209 SINGLE_KEY(nr_acquired)
210 SINGLE_KEY(nr_contended)
211 SINGLE_KEY(wait_time_total)
212 SINGLE_KEY(wait_time_max)
213 
214 static int lock_stat_key_wait_time_min(struct lock_stat *one,
215 					struct lock_stat *two)
216 {
217 	u64 s1 = one->wait_time_min;
218 	u64 s2 = two->wait_time_min;
219 	if (s1 == ULLONG_MAX)
220 		s1 = 0;
221 	if (s2 == ULLONG_MAX)
222 		s2 = 0;
223 	return s1 > s2;
224 }
225 
226 struct lock_key {
227 	/*
228 	 * name: the value for specify by user
229 	 * this should be simpler than raw name of member
230 	 * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
231 	 */
232 	const char		*name;
233 	int			(*key)(struct lock_stat*, struct lock_stat*);
234 };
235 
236 static const char		*sort_key = "acquired";
237 
238 static int			(*compare)(struct lock_stat *, struct lock_stat *);
239 
240 static struct rb_root		result;	/* place to store sorted data */
241 
242 #define DEF_KEY_LOCK(name, fn_suffix)	\
243 	{ #name, lock_stat_key_ ## fn_suffix }
244 struct lock_key keys[] = {
245 	DEF_KEY_LOCK(acquired, nr_acquired),
246 	DEF_KEY_LOCK(contended, nr_contended),
247 	DEF_KEY_LOCK(wait_total, wait_time_total),
248 	DEF_KEY_LOCK(wait_min, wait_time_min),
249 	DEF_KEY_LOCK(wait_max, wait_time_max),
250 
251 	/* extra comparisons much complicated should be here */
252 
253 	{ NULL, NULL }
254 };
255 
256 static int select_key(void)
257 {
258 	int i;
259 
260 	for (i = 0; keys[i].name; i++) {
261 		if (!strcmp(keys[i].name, sort_key)) {
262 			compare = keys[i].key;
263 			return 0;
264 		}
265 	}
266 
267 	pr_err("Unknown compare key: %s\n", sort_key);
268 
269 	return -1;
270 }
271 
272 static void insert_to_result(struct lock_stat *st,
273 			     int (*bigger)(struct lock_stat *, struct lock_stat *))
274 {
275 	struct rb_node **rb = &result.rb_node;
276 	struct rb_node *parent = NULL;
277 	struct lock_stat *p;
278 
279 	while (*rb) {
280 		p = container_of(*rb, struct lock_stat, rb);
281 		parent = *rb;
282 
283 		if (bigger(st, p))
284 			rb = &(*rb)->rb_left;
285 		else
286 			rb = &(*rb)->rb_right;
287 	}
288 
289 	rb_link_node(&st->rb, parent, rb);
290 	rb_insert_color(&st->rb, &result);
291 }
292 
293 /* returns left most element of result, and erase it */
294 static struct lock_stat *pop_from_result(void)
295 {
296 	struct rb_node *node = result.rb_node;
297 
298 	if (!node)
299 		return NULL;
300 
301 	while (node->rb_left)
302 		node = node->rb_left;
303 
304 	rb_erase(node, &result);
305 	return container_of(node, struct lock_stat, rb);
306 }
307 
308 static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
309 {
310 	struct list_head *entry = lockhashentry(addr);
311 	struct lock_stat *ret, *new;
312 
313 	list_for_each_entry(ret, entry, hash_entry) {
314 		if (ret->addr == addr)
315 			return ret;
316 	}
317 
318 	new = zalloc(sizeof(struct lock_stat));
319 	if (!new)
320 		goto alloc_failed;
321 
322 	new->addr = addr;
323 	new->name = zalloc(sizeof(char) * strlen(name) + 1);
324 	if (!new->name)
325 		goto alloc_failed;
326 	strcpy(new->name, name);
327 
328 	new->wait_time_min = ULLONG_MAX;
329 
330 	list_add(&new->hash_entry, entry);
331 	return new;
332 
333 alloc_failed:
334 	pr_err("memory allocation failed\n");
335 	return NULL;
336 }
337 
338 static const char *input_name;
339 
340 struct trace_lock_handler {
341 	int (*acquire_event)(struct perf_evsel *evsel,
342 			     struct perf_sample *sample);
343 
344 	int (*acquired_event)(struct perf_evsel *evsel,
345 			      struct perf_sample *sample);
346 
347 	int (*contended_event)(struct perf_evsel *evsel,
348 			       struct perf_sample *sample);
349 
350 	int (*release_event)(struct perf_evsel *evsel,
351 			     struct perf_sample *sample);
352 };
353 
354 static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
355 {
356 	struct lock_seq_stat *seq;
357 
358 	list_for_each_entry(seq, &ts->seq_list, list) {
359 		if (seq->addr == addr)
360 			return seq;
361 	}
362 
363 	seq = zalloc(sizeof(struct lock_seq_stat));
364 	if (!seq) {
365 		pr_err("memory allocation failed\n");
366 		return NULL;
367 	}
368 	seq->state = SEQ_STATE_UNINITIALIZED;
369 	seq->addr = addr;
370 
371 	list_add(&seq->list, &ts->seq_list);
372 	return seq;
373 }
374 
375 enum broken_state {
376 	BROKEN_ACQUIRE,
377 	BROKEN_ACQUIRED,
378 	BROKEN_CONTENDED,
379 	BROKEN_RELEASE,
380 	BROKEN_MAX,
381 };
382 
383 static int bad_hist[BROKEN_MAX];
384 
385 enum acquire_flags {
386 	TRY_LOCK = 1,
387 	READ_LOCK = 2,
388 };
389 
390 static int report_lock_acquire_event(struct perf_evsel *evsel,
391 				     struct perf_sample *sample)
392 {
393 	void *addr;
394 	struct lock_stat *ls;
395 	struct thread_stat *ts;
396 	struct lock_seq_stat *seq;
397 	const char *name = perf_evsel__strval(evsel, sample, "name");
398 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
399 	int flag = perf_evsel__intval(evsel, sample, "flag");
400 
401 	memcpy(&addr, &tmp, sizeof(void *));
402 
403 	ls = lock_stat_findnew(addr, name);
404 	if (!ls)
405 		return -1;
406 	if (ls->discard)
407 		return 0;
408 
409 	ts = thread_stat_findnew(sample->tid);
410 	if (!ts)
411 		return -1;
412 
413 	seq = get_seq(ts, addr);
414 	if (!seq)
415 		return -1;
416 
417 	switch (seq->state) {
418 	case SEQ_STATE_UNINITIALIZED:
419 	case SEQ_STATE_RELEASED:
420 		if (!flag) {
421 			seq->state = SEQ_STATE_ACQUIRING;
422 		} else {
423 			if (flag & TRY_LOCK)
424 				ls->nr_trylock++;
425 			if (flag & READ_LOCK)
426 				ls->nr_readlock++;
427 			seq->state = SEQ_STATE_READ_ACQUIRED;
428 			seq->read_count = 1;
429 			ls->nr_acquired++;
430 		}
431 		break;
432 	case SEQ_STATE_READ_ACQUIRED:
433 		if (flag & READ_LOCK) {
434 			seq->read_count++;
435 			ls->nr_acquired++;
436 			goto end;
437 		} else {
438 			goto broken;
439 		}
440 		break;
441 	case SEQ_STATE_ACQUIRED:
442 	case SEQ_STATE_ACQUIRING:
443 	case SEQ_STATE_CONTENDED:
444 broken:
445 		/* broken lock sequence, discard it */
446 		ls->discard = 1;
447 		bad_hist[BROKEN_ACQUIRE]++;
448 		list_del(&seq->list);
449 		free(seq);
450 		goto end;
451 		break;
452 	default:
453 		BUG_ON("Unknown state of lock sequence found!\n");
454 		break;
455 	}
456 
457 	ls->nr_acquire++;
458 	seq->prev_event_time = sample->time;
459 end:
460 	return 0;
461 }
462 
463 static int report_lock_acquired_event(struct perf_evsel *evsel,
464 				      struct perf_sample *sample)
465 {
466 	void *addr;
467 	struct lock_stat *ls;
468 	struct thread_stat *ts;
469 	struct lock_seq_stat *seq;
470 	u64 contended_term;
471 	const char *name = perf_evsel__strval(evsel, sample, "name");
472 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
473 
474 	memcpy(&addr, &tmp, sizeof(void *));
475 
476 	ls = lock_stat_findnew(addr, name);
477 	if (!ls)
478 		return -1;
479 	if (ls->discard)
480 		return 0;
481 
482 	ts = thread_stat_findnew(sample->tid);
483 	if (!ts)
484 		return -1;
485 
486 	seq = get_seq(ts, addr);
487 	if (!seq)
488 		return -1;
489 
490 	switch (seq->state) {
491 	case SEQ_STATE_UNINITIALIZED:
492 		/* orphan event, do nothing */
493 		return 0;
494 	case SEQ_STATE_ACQUIRING:
495 		break;
496 	case SEQ_STATE_CONTENDED:
497 		contended_term = sample->time - seq->prev_event_time;
498 		ls->wait_time_total += contended_term;
499 		if (contended_term < ls->wait_time_min)
500 			ls->wait_time_min = contended_term;
501 		if (ls->wait_time_max < contended_term)
502 			ls->wait_time_max = contended_term;
503 		break;
504 	case SEQ_STATE_RELEASED:
505 	case SEQ_STATE_ACQUIRED:
506 	case SEQ_STATE_READ_ACQUIRED:
507 		/* broken lock sequence, discard it */
508 		ls->discard = 1;
509 		bad_hist[BROKEN_ACQUIRED]++;
510 		list_del(&seq->list);
511 		free(seq);
512 		goto end;
513 		break;
514 
515 	default:
516 		BUG_ON("Unknown state of lock sequence found!\n");
517 		break;
518 	}
519 
520 	seq->state = SEQ_STATE_ACQUIRED;
521 	ls->nr_acquired++;
522 	seq->prev_event_time = sample->time;
523 end:
524 	return 0;
525 }
526 
527 static int report_lock_contended_event(struct perf_evsel *evsel,
528 				       struct perf_sample *sample)
529 {
530 	void *addr;
531 	struct lock_stat *ls;
532 	struct thread_stat *ts;
533 	struct lock_seq_stat *seq;
534 	const char *name = perf_evsel__strval(evsel, sample, "name");
535 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
536 
537 	memcpy(&addr, &tmp, sizeof(void *));
538 
539 	ls = lock_stat_findnew(addr, name);
540 	if (!ls)
541 		return -1;
542 	if (ls->discard)
543 		return 0;
544 
545 	ts = thread_stat_findnew(sample->tid);
546 	if (!ts)
547 		return -1;
548 
549 	seq = get_seq(ts, addr);
550 	if (!seq)
551 		return -1;
552 
553 	switch (seq->state) {
554 	case SEQ_STATE_UNINITIALIZED:
555 		/* orphan event, do nothing */
556 		return 0;
557 	case SEQ_STATE_ACQUIRING:
558 		break;
559 	case SEQ_STATE_RELEASED:
560 	case SEQ_STATE_ACQUIRED:
561 	case SEQ_STATE_READ_ACQUIRED:
562 	case SEQ_STATE_CONTENDED:
563 		/* broken lock sequence, discard it */
564 		ls->discard = 1;
565 		bad_hist[BROKEN_CONTENDED]++;
566 		list_del(&seq->list);
567 		free(seq);
568 		goto end;
569 		break;
570 	default:
571 		BUG_ON("Unknown state of lock sequence found!\n");
572 		break;
573 	}
574 
575 	seq->state = SEQ_STATE_CONTENDED;
576 	ls->nr_contended++;
577 	seq->prev_event_time = sample->time;
578 end:
579 	return 0;
580 }
581 
582 static int report_lock_release_event(struct perf_evsel *evsel,
583 				     struct perf_sample *sample)
584 {
585 	void *addr;
586 	struct lock_stat *ls;
587 	struct thread_stat *ts;
588 	struct lock_seq_stat *seq;
589 	const char *name = perf_evsel__strval(evsel, sample, "name");
590 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
591 
592 	memcpy(&addr, &tmp, sizeof(void *));
593 
594 	ls = lock_stat_findnew(addr, name);
595 	if (!ls)
596 		return -1;
597 	if (ls->discard)
598 		return 0;
599 
600 	ts = thread_stat_findnew(sample->tid);
601 	if (!ts)
602 		return -1;
603 
604 	seq = get_seq(ts, addr);
605 	if (!seq)
606 		return -1;
607 
608 	switch (seq->state) {
609 	case SEQ_STATE_UNINITIALIZED:
610 		goto end;
611 		break;
612 	case SEQ_STATE_ACQUIRED:
613 		break;
614 	case SEQ_STATE_READ_ACQUIRED:
615 		seq->read_count--;
616 		BUG_ON(seq->read_count < 0);
617 		if (!seq->read_count) {
618 			ls->nr_release++;
619 			goto end;
620 		}
621 		break;
622 	case SEQ_STATE_ACQUIRING:
623 	case SEQ_STATE_CONTENDED:
624 	case SEQ_STATE_RELEASED:
625 		/* broken lock sequence, discard it */
626 		ls->discard = 1;
627 		bad_hist[BROKEN_RELEASE]++;
628 		goto free_seq;
629 		break;
630 	default:
631 		BUG_ON("Unknown state of lock sequence found!\n");
632 		break;
633 	}
634 
635 	ls->nr_release++;
636 free_seq:
637 	list_del(&seq->list);
638 	free(seq);
639 end:
640 	return 0;
641 }
642 
643 /* lock oriented handlers */
644 /* TODO: handlers for CPU oriented, thread oriented */
645 static struct trace_lock_handler report_lock_ops  = {
646 	.acquire_event		= report_lock_acquire_event,
647 	.acquired_event		= report_lock_acquired_event,
648 	.contended_event	= report_lock_contended_event,
649 	.release_event		= report_lock_release_event,
650 };
651 
652 static struct trace_lock_handler *trace_handler;
653 
654 static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
655 					     struct perf_sample *sample)
656 {
657 	if (trace_handler->acquire_event)
658 		return trace_handler->acquire_event(evsel, sample);
659 	return 0;
660 }
661 
662 static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
663 					      struct perf_sample *sample)
664 {
665 	if (trace_handler->acquired_event)
666 		return trace_handler->acquired_event(evsel, sample);
667 	return 0;
668 }
669 
670 static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
671 					      struct perf_sample *sample)
672 {
673 	if (trace_handler->contended_event)
674 		return trace_handler->contended_event(evsel, sample);
675 	return 0;
676 }
677 
678 static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
679 					    struct perf_sample *sample)
680 {
681 	if (trace_handler->release_event)
682 		return trace_handler->release_event(evsel, sample);
683 	return 0;
684 }
685 
686 static void print_bad_events(int bad, int total)
687 {
688 	/* Output for debug, this have to be removed */
689 	int i;
690 	const char *name[4] =
691 		{ "acquire", "acquired", "contended", "release" };
692 
693 	pr_info("\n=== output for debug===\n\n");
694 	pr_info("bad: %d, total: %d\n", bad, total);
695 	pr_info("bad rate: %f %%\n", (double)bad / (double)total * 100);
696 	pr_info("histogram of events caused bad sequence\n");
697 	for (i = 0; i < BROKEN_MAX; i++)
698 		pr_info(" %10s: %d\n", name[i], bad_hist[i]);
699 }
700 
701 /* TODO: various way to print, coloring, nano or milli sec */
702 static void print_result(void)
703 {
704 	struct lock_stat *st;
705 	char cut_name[20];
706 	int bad, total;
707 
708 	pr_info("%20s ", "Name");
709 	pr_info("%10s ", "acquired");
710 	pr_info("%10s ", "contended");
711 
712 	pr_info("%15s ", "total wait (ns)");
713 	pr_info("%15s ", "max wait (ns)");
714 	pr_info("%15s ", "min wait (ns)");
715 
716 	pr_info("\n\n");
717 
718 	bad = total = 0;
719 	while ((st = pop_from_result())) {
720 		total++;
721 		if (st->discard) {
722 			bad++;
723 			continue;
724 		}
725 		bzero(cut_name, 20);
726 
727 		if (strlen(st->name) < 16) {
728 			/* output raw name */
729 			pr_info("%20s ", st->name);
730 		} else {
731 			strncpy(cut_name, st->name, 16);
732 			cut_name[16] = '.';
733 			cut_name[17] = '.';
734 			cut_name[18] = '.';
735 			cut_name[19] = '\0';
736 			/* cut off name for saving output style */
737 			pr_info("%20s ", cut_name);
738 		}
739 
740 		pr_info("%10u ", st->nr_acquired);
741 		pr_info("%10u ", st->nr_contended);
742 
743 		pr_info("%15" PRIu64 " ", st->wait_time_total);
744 		pr_info("%15" PRIu64 " ", st->wait_time_max);
745 		pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
746 		       0 : st->wait_time_min);
747 		pr_info("\n");
748 	}
749 
750 	print_bad_events(bad, total);
751 }
752 
753 static bool info_threads, info_map;
754 
755 static void dump_threads(void)
756 {
757 	struct thread_stat *st;
758 	struct rb_node *node;
759 	struct thread *t;
760 
761 	pr_info("%10s: comm\n", "Thread ID");
762 
763 	node = rb_first(&thread_stats);
764 	while (node) {
765 		st = container_of(node, struct thread_stat, rb);
766 		t = perf_session__findnew(session, st->tid);
767 		pr_info("%10d: %s\n", st->tid, t->comm);
768 		node = rb_next(node);
769 	};
770 }
771 
772 static void dump_map(void)
773 {
774 	unsigned int i;
775 	struct lock_stat *st;
776 
777 	pr_info("Address of instance: name of class\n");
778 	for (i = 0; i < LOCKHASH_SIZE; i++) {
779 		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
780 			pr_info(" %p: %s\n", st->addr, st->name);
781 		}
782 	}
783 }
784 
785 static int dump_info(void)
786 {
787 	int rc = 0;
788 
789 	if (info_threads)
790 		dump_threads();
791 	else if (info_map)
792 		dump_map();
793 	else {
794 		rc = -1;
795 		pr_err("Unknown type of information\n");
796 	}
797 
798 	return rc;
799 }
800 
801 typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
802 				  struct perf_sample *sample);
803 
804 static int process_sample_event(struct perf_tool *tool __maybe_unused,
805 				union perf_event *event,
806 				struct perf_sample *sample,
807 				struct perf_evsel *evsel,
808 				struct machine *machine)
809 {
810 	struct thread *thread = machine__findnew_thread(machine, sample->tid);
811 
812 	if (thread == NULL) {
813 		pr_debug("problem processing %d event, skipping it.\n",
814 			event->header.type);
815 		return -1;
816 	}
817 
818 	if (evsel->handler.func != NULL) {
819 		tracepoint_handler f = evsel->handler.func;
820 		return f(evsel, sample);
821 	}
822 
823 	return 0;
824 }
825 
826 static const struct perf_evsel_str_handler lock_tracepoints[] = {
827 	{ "lock:lock_acquire",	 perf_evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
828 	{ "lock:lock_acquired",	 perf_evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
829 	{ "lock:lock_contended", perf_evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
830 	{ "lock:lock_release",	 perf_evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
831 };
832 
833 static int read_events(void)
834 {
835 	struct perf_tool eops = {
836 		.sample		 = process_sample_event,
837 		.comm		 = perf_event__process_comm,
838 		.ordered_samples = true,
839 	};
840 	session = perf_session__new(input_name, O_RDONLY, 0, false, &eops);
841 	if (!session) {
842 		pr_err("Initializing perf session failed\n");
843 		return -1;
844 	}
845 
846 	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
847 		pr_err("Initializing perf session tracepoint handlers failed\n");
848 		return -1;
849 	}
850 
851 	return perf_session__process_events(session, &eops);
852 }
853 
854 static void sort_result(void)
855 {
856 	unsigned int i;
857 	struct lock_stat *st;
858 
859 	for (i = 0; i < LOCKHASH_SIZE; i++) {
860 		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
861 			insert_to_result(st, compare);
862 		}
863 	}
864 }
865 
866 static int __cmd_report(void)
867 {
868 	setup_pager();
869 
870 	if ((select_key() != 0) ||
871 	    (read_events() != 0))
872 		return -1;
873 
874 	sort_result();
875 	print_result();
876 
877 	return 0;
878 }
879 
880 static int __cmd_record(int argc, const char **argv)
881 {
882 	const char *record_args[] = {
883 		"record", "-R", "-f", "-m", "1024", "-c", "1",
884 	};
885 	unsigned int rec_argc, i, j;
886 	const char **rec_argv;
887 
888 	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
889 		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
890 				pr_err("tracepoint %s is not enabled. "
891 				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
892 				       lock_tracepoints[i].name);
893 				return 1;
894 		}
895 	}
896 
897 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
898 	/* factor of 2 is for -e in front of each tracepoint */
899 	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
900 
901 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
902 	if (rec_argv == NULL)
903 		return -ENOMEM;
904 
905 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
906 		rec_argv[i] = strdup(record_args[i]);
907 
908 	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
909 		rec_argv[i++] = "-e";
910 		rec_argv[i++] = strdup(lock_tracepoints[j].name);
911 	}
912 
913 	for (j = 1; j < (unsigned int)argc; j++, i++)
914 		rec_argv[i] = argv[j];
915 
916 	BUG_ON(i != rec_argc);
917 
918 	return cmd_record(i, rec_argv, NULL);
919 }
920 
921 int cmd_lock(int argc, const char **argv, const char *prefix __maybe_unused)
922 {
923 	const struct option info_options[] = {
924 	OPT_BOOLEAN('t', "threads", &info_threads,
925 		    "dump thread list in perf.data"),
926 	OPT_BOOLEAN('m', "map", &info_map,
927 		    "map of lock instances (address:name table)"),
928 	OPT_END()
929 	};
930 	const struct option lock_options[] = {
931 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
932 	OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
933 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
934 	OPT_END()
935 	};
936 	const struct option report_options[] = {
937 	OPT_STRING('k', "key", &sort_key, "acquired",
938 		    "key for sorting (acquired / contended / wait_total / wait_max / wait_min)"),
939 	/* TODO: type */
940 	OPT_END()
941 	};
942 	const char * const info_usage[] = {
943 		"perf lock info [<options>]",
944 		NULL
945 	};
946 	const char * const lock_usage[] = {
947 		"perf lock [<options>] {record|report|script|info}",
948 		NULL
949 	};
950 	const char * const report_usage[] = {
951 		"perf lock report [<options>]",
952 		NULL
953 	};
954 	unsigned int i;
955 	int rc = 0;
956 
957 	symbol__init();
958 	for (i = 0; i < LOCKHASH_SIZE; i++)
959 		INIT_LIST_HEAD(lockhash_table + i);
960 
961 	argc = parse_options(argc, argv, lock_options, lock_usage,
962 			     PARSE_OPT_STOP_AT_NON_OPTION);
963 	if (!argc)
964 		usage_with_options(lock_usage, lock_options);
965 
966 	if (!strncmp(argv[0], "rec", 3)) {
967 		return __cmd_record(argc, argv);
968 	} else if (!strncmp(argv[0], "report", 6)) {
969 		trace_handler = &report_lock_ops;
970 		if (argc) {
971 			argc = parse_options(argc, argv,
972 					     report_options, report_usage, 0);
973 			if (argc)
974 				usage_with_options(report_usage, report_options);
975 		}
976 		__cmd_report();
977 	} else if (!strcmp(argv[0], "script")) {
978 		/* Aliased to 'perf script' */
979 		return cmd_script(argc, argv, prefix);
980 	} else if (!strcmp(argv[0], "info")) {
981 		if (argc) {
982 			argc = parse_options(argc, argv,
983 					     info_options, info_usage, 0);
984 			if (argc)
985 				usage_with_options(info_usage, info_options);
986 		}
987 		/* recycling report_lock_ops */
988 		trace_handler = &report_lock_ops;
989 		setup_pager();
990 		if (read_events() != 0)
991 			rc = -1;
992 		else
993 			rc = dump_info();
994 	} else {
995 		usage_with_options(lock_usage, lock_options);
996 	}
997 
998 	return rc;
999 }
1000