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 
thread_stat_find(u32 tid)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 
thread_stat_insert(struct thread_stat * new)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 
thread_stat_findnew_after_first(u32 tid)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 
thread_stat_findnew_first(u32 tid)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)
SINGLE_KEY(nr_contended)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 
select_key(void)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 
insert_to_result(struct lock_stat * st,int (* bigger)(struct lock_stat *,struct lock_stat *))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 */
pop_from_result(void)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 
lock_stat_findnew(void * addr,const char * name)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 struct trace_lock_handler {
339 	int (*acquire_event)(struct perf_evsel *evsel,
340 			     struct perf_sample *sample);
341 
342 	int (*acquired_event)(struct perf_evsel *evsel,
343 			      struct perf_sample *sample);
344 
345 	int (*contended_event)(struct perf_evsel *evsel,
346 			       struct perf_sample *sample);
347 
348 	int (*release_event)(struct perf_evsel *evsel,
349 			     struct perf_sample *sample);
350 };
351 
get_seq(struct thread_stat * ts,void * addr)352 static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
353 {
354 	struct lock_seq_stat *seq;
355 
356 	list_for_each_entry(seq, &ts->seq_list, list) {
357 		if (seq->addr == addr)
358 			return seq;
359 	}
360 
361 	seq = zalloc(sizeof(struct lock_seq_stat));
362 	if (!seq) {
363 		pr_err("memory allocation failed\n");
364 		return NULL;
365 	}
366 	seq->state = SEQ_STATE_UNINITIALIZED;
367 	seq->addr = addr;
368 
369 	list_add(&seq->list, &ts->seq_list);
370 	return seq;
371 }
372 
373 enum broken_state {
374 	BROKEN_ACQUIRE,
375 	BROKEN_ACQUIRED,
376 	BROKEN_CONTENDED,
377 	BROKEN_RELEASE,
378 	BROKEN_MAX,
379 };
380 
381 static int bad_hist[BROKEN_MAX];
382 
383 enum acquire_flags {
384 	TRY_LOCK = 1,
385 	READ_LOCK = 2,
386 };
387 
report_lock_acquire_event(struct perf_evsel * evsel,struct perf_sample * sample)388 static int report_lock_acquire_event(struct perf_evsel *evsel,
389 				     struct perf_sample *sample)
390 {
391 	void *addr;
392 	struct lock_stat *ls;
393 	struct thread_stat *ts;
394 	struct lock_seq_stat *seq;
395 	const char *name = perf_evsel__strval(evsel, sample, "name");
396 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
397 	int flag = perf_evsel__intval(evsel, sample, "flag");
398 
399 	memcpy(&addr, &tmp, sizeof(void *));
400 
401 	ls = lock_stat_findnew(addr, name);
402 	if (!ls)
403 		return -1;
404 	if (ls->discard)
405 		return 0;
406 
407 	ts = thread_stat_findnew(sample->tid);
408 	if (!ts)
409 		return -1;
410 
411 	seq = get_seq(ts, addr);
412 	if (!seq)
413 		return -1;
414 
415 	switch (seq->state) {
416 	case SEQ_STATE_UNINITIALIZED:
417 	case SEQ_STATE_RELEASED:
418 		if (!flag) {
419 			seq->state = SEQ_STATE_ACQUIRING;
420 		} else {
421 			if (flag & TRY_LOCK)
422 				ls->nr_trylock++;
423 			if (flag & READ_LOCK)
424 				ls->nr_readlock++;
425 			seq->state = SEQ_STATE_READ_ACQUIRED;
426 			seq->read_count = 1;
427 			ls->nr_acquired++;
428 		}
429 		break;
430 	case SEQ_STATE_READ_ACQUIRED:
431 		if (flag & READ_LOCK) {
432 			seq->read_count++;
433 			ls->nr_acquired++;
434 			goto end;
435 		} else {
436 			goto broken;
437 		}
438 		break;
439 	case SEQ_STATE_ACQUIRED:
440 	case SEQ_STATE_ACQUIRING:
441 	case SEQ_STATE_CONTENDED:
442 broken:
443 		/* broken lock sequence, discard it */
444 		ls->discard = 1;
445 		bad_hist[BROKEN_ACQUIRE]++;
446 		list_del(&seq->list);
447 		free(seq);
448 		goto end;
449 		break;
450 	default:
451 		BUG_ON("Unknown state of lock sequence found!\n");
452 		break;
453 	}
454 
455 	ls->nr_acquire++;
456 	seq->prev_event_time = sample->time;
457 end:
458 	return 0;
459 }
460 
report_lock_acquired_event(struct perf_evsel * evsel,struct perf_sample * sample)461 static int report_lock_acquired_event(struct perf_evsel *evsel,
462 				      struct perf_sample *sample)
463 {
464 	void *addr;
465 	struct lock_stat *ls;
466 	struct thread_stat *ts;
467 	struct lock_seq_stat *seq;
468 	u64 contended_term;
469 	const char *name = perf_evsel__strval(evsel, sample, "name");
470 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
471 
472 	memcpy(&addr, &tmp, sizeof(void *));
473 
474 	ls = lock_stat_findnew(addr, name);
475 	if (!ls)
476 		return -1;
477 	if (ls->discard)
478 		return 0;
479 
480 	ts = thread_stat_findnew(sample->tid);
481 	if (!ts)
482 		return -1;
483 
484 	seq = get_seq(ts, addr);
485 	if (!seq)
486 		return -1;
487 
488 	switch (seq->state) {
489 	case SEQ_STATE_UNINITIALIZED:
490 		/* orphan event, do nothing */
491 		return 0;
492 	case SEQ_STATE_ACQUIRING:
493 		break;
494 	case SEQ_STATE_CONTENDED:
495 		contended_term = sample->time - seq->prev_event_time;
496 		ls->wait_time_total += contended_term;
497 		if (contended_term < ls->wait_time_min)
498 			ls->wait_time_min = contended_term;
499 		if (ls->wait_time_max < contended_term)
500 			ls->wait_time_max = contended_term;
501 		break;
502 	case SEQ_STATE_RELEASED:
503 	case SEQ_STATE_ACQUIRED:
504 	case SEQ_STATE_READ_ACQUIRED:
505 		/* broken lock sequence, discard it */
506 		ls->discard = 1;
507 		bad_hist[BROKEN_ACQUIRED]++;
508 		list_del(&seq->list);
509 		free(seq);
510 		goto end;
511 		break;
512 
513 	default:
514 		BUG_ON("Unknown state of lock sequence found!\n");
515 		break;
516 	}
517 
518 	seq->state = SEQ_STATE_ACQUIRED;
519 	ls->nr_acquired++;
520 	seq->prev_event_time = sample->time;
521 end:
522 	return 0;
523 }
524 
report_lock_contended_event(struct perf_evsel * evsel,struct perf_sample * sample)525 static int report_lock_contended_event(struct perf_evsel *evsel,
526 				       struct perf_sample *sample)
527 {
528 	void *addr;
529 	struct lock_stat *ls;
530 	struct thread_stat *ts;
531 	struct lock_seq_stat *seq;
532 	const char *name = perf_evsel__strval(evsel, sample, "name");
533 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
534 
535 	memcpy(&addr, &tmp, sizeof(void *));
536 
537 	ls = lock_stat_findnew(addr, name);
538 	if (!ls)
539 		return -1;
540 	if (ls->discard)
541 		return 0;
542 
543 	ts = thread_stat_findnew(sample->tid);
544 	if (!ts)
545 		return -1;
546 
547 	seq = get_seq(ts, addr);
548 	if (!seq)
549 		return -1;
550 
551 	switch (seq->state) {
552 	case SEQ_STATE_UNINITIALIZED:
553 		/* orphan event, do nothing */
554 		return 0;
555 	case SEQ_STATE_ACQUIRING:
556 		break;
557 	case SEQ_STATE_RELEASED:
558 	case SEQ_STATE_ACQUIRED:
559 	case SEQ_STATE_READ_ACQUIRED:
560 	case SEQ_STATE_CONTENDED:
561 		/* broken lock sequence, discard it */
562 		ls->discard = 1;
563 		bad_hist[BROKEN_CONTENDED]++;
564 		list_del(&seq->list);
565 		free(seq);
566 		goto end;
567 		break;
568 	default:
569 		BUG_ON("Unknown state of lock sequence found!\n");
570 		break;
571 	}
572 
573 	seq->state = SEQ_STATE_CONTENDED;
574 	ls->nr_contended++;
575 	seq->prev_event_time = sample->time;
576 end:
577 	return 0;
578 }
579 
report_lock_release_event(struct perf_evsel * evsel,struct perf_sample * sample)580 static int report_lock_release_event(struct perf_evsel *evsel,
581 				     struct perf_sample *sample)
582 {
583 	void *addr;
584 	struct lock_stat *ls;
585 	struct thread_stat *ts;
586 	struct lock_seq_stat *seq;
587 	const char *name = perf_evsel__strval(evsel, sample, "name");
588 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
589 
590 	memcpy(&addr, &tmp, sizeof(void *));
591 
592 	ls = lock_stat_findnew(addr, name);
593 	if (!ls)
594 		return -1;
595 	if (ls->discard)
596 		return 0;
597 
598 	ts = thread_stat_findnew(sample->tid);
599 	if (!ts)
600 		return -1;
601 
602 	seq = get_seq(ts, addr);
603 	if (!seq)
604 		return -1;
605 
606 	switch (seq->state) {
607 	case SEQ_STATE_UNINITIALIZED:
608 		goto end;
609 		break;
610 	case SEQ_STATE_ACQUIRED:
611 		break;
612 	case SEQ_STATE_READ_ACQUIRED:
613 		seq->read_count--;
614 		BUG_ON(seq->read_count < 0);
615 		if (!seq->read_count) {
616 			ls->nr_release++;
617 			goto end;
618 		}
619 		break;
620 	case SEQ_STATE_ACQUIRING:
621 	case SEQ_STATE_CONTENDED:
622 	case SEQ_STATE_RELEASED:
623 		/* broken lock sequence, discard it */
624 		ls->discard = 1;
625 		bad_hist[BROKEN_RELEASE]++;
626 		goto free_seq;
627 		break;
628 	default:
629 		BUG_ON("Unknown state of lock sequence found!\n");
630 		break;
631 	}
632 
633 	ls->nr_release++;
634 free_seq:
635 	list_del(&seq->list);
636 	free(seq);
637 end:
638 	return 0;
639 }
640 
641 /* lock oriented handlers */
642 /* TODO: handlers for CPU oriented, thread oriented */
643 static struct trace_lock_handler report_lock_ops  = {
644 	.acquire_event		= report_lock_acquire_event,
645 	.acquired_event		= report_lock_acquired_event,
646 	.contended_event	= report_lock_contended_event,
647 	.release_event		= report_lock_release_event,
648 };
649 
650 static struct trace_lock_handler *trace_handler;
651 
perf_evsel__process_lock_acquire(struct perf_evsel * evsel,struct perf_sample * sample)652 static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
653 					     struct perf_sample *sample)
654 {
655 	if (trace_handler->acquire_event)
656 		return trace_handler->acquire_event(evsel, sample);
657 	return 0;
658 }
659 
perf_evsel__process_lock_acquired(struct perf_evsel * evsel,struct perf_sample * sample)660 static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
661 					      struct perf_sample *sample)
662 {
663 	if (trace_handler->acquired_event)
664 		return trace_handler->acquired_event(evsel, sample);
665 	return 0;
666 }
667 
perf_evsel__process_lock_contended(struct perf_evsel * evsel,struct perf_sample * sample)668 static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
669 					      struct perf_sample *sample)
670 {
671 	if (trace_handler->contended_event)
672 		return trace_handler->contended_event(evsel, sample);
673 	return 0;
674 }
675 
perf_evsel__process_lock_release(struct perf_evsel * evsel,struct perf_sample * sample)676 static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
677 					    struct perf_sample *sample)
678 {
679 	if (trace_handler->release_event)
680 		return trace_handler->release_event(evsel, sample);
681 	return 0;
682 }
683 
print_bad_events(int bad,int total)684 static void print_bad_events(int bad, int total)
685 {
686 	/* Output for debug, this have to be removed */
687 	int i;
688 	const char *name[4] =
689 		{ "acquire", "acquired", "contended", "release" };
690 
691 	pr_info("\n=== output for debug===\n\n");
692 	pr_info("bad: %d, total: %d\n", bad, total);
693 	pr_info("bad rate: %f %%\n", (double)bad / (double)total * 100);
694 	pr_info("histogram of events caused bad sequence\n");
695 	for (i = 0; i < BROKEN_MAX; i++)
696 		pr_info(" %10s: %d\n", name[i], bad_hist[i]);
697 }
698 
699 /* TODO: various way to print, coloring, nano or milli sec */
print_result(void)700 static void print_result(void)
701 {
702 	struct lock_stat *st;
703 	char cut_name[20];
704 	int bad, total;
705 
706 	pr_info("%20s ", "Name");
707 	pr_info("%10s ", "acquired");
708 	pr_info("%10s ", "contended");
709 
710 	pr_info("%15s ", "total wait (ns)");
711 	pr_info("%15s ", "max wait (ns)");
712 	pr_info("%15s ", "min wait (ns)");
713 
714 	pr_info("\n\n");
715 
716 	bad = total = 0;
717 	while ((st = pop_from_result())) {
718 		total++;
719 		if (st->discard) {
720 			bad++;
721 			continue;
722 		}
723 		bzero(cut_name, 20);
724 
725 		if (strlen(st->name) < 16) {
726 			/* output raw name */
727 			pr_info("%20s ", st->name);
728 		} else {
729 			strncpy(cut_name, st->name, 16);
730 			cut_name[16] = '.';
731 			cut_name[17] = '.';
732 			cut_name[18] = '.';
733 			cut_name[19] = '\0';
734 			/* cut off name for saving output style */
735 			pr_info("%20s ", cut_name);
736 		}
737 
738 		pr_info("%10u ", st->nr_acquired);
739 		pr_info("%10u ", st->nr_contended);
740 
741 		pr_info("%15" PRIu64 " ", st->wait_time_total);
742 		pr_info("%15" PRIu64 " ", st->wait_time_max);
743 		pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
744 		       0 : st->wait_time_min);
745 		pr_info("\n");
746 	}
747 
748 	print_bad_events(bad, total);
749 }
750 
751 static bool info_threads, info_map;
752 
dump_threads(void)753 static void dump_threads(void)
754 {
755 	struct thread_stat *st;
756 	struct rb_node *node;
757 	struct thread *t;
758 
759 	pr_info("%10s: comm\n", "Thread ID");
760 
761 	node = rb_first(&thread_stats);
762 	while (node) {
763 		st = container_of(node, struct thread_stat, rb);
764 		t = perf_session__findnew(session, st->tid);
765 		pr_info("%10d: %s\n", st->tid, t->comm);
766 		node = rb_next(node);
767 	};
768 }
769 
dump_map(void)770 static void dump_map(void)
771 {
772 	unsigned int i;
773 	struct lock_stat *st;
774 
775 	pr_info("Address of instance: name of class\n");
776 	for (i = 0; i < LOCKHASH_SIZE; i++) {
777 		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
778 			pr_info(" %p: %s\n", st->addr, st->name);
779 		}
780 	}
781 }
782 
dump_info(void)783 static int dump_info(void)
784 {
785 	int rc = 0;
786 
787 	if (info_threads)
788 		dump_threads();
789 	else if (info_map)
790 		dump_map();
791 	else {
792 		rc = -1;
793 		pr_err("Unknown type of information\n");
794 	}
795 
796 	return rc;
797 }
798 
799 typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
800 				  struct perf_sample *sample);
801 
process_sample_event(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct perf_evsel * evsel,struct machine * machine)802 static int process_sample_event(struct perf_tool *tool __maybe_unused,
803 				union perf_event *event,
804 				struct perf_sample *sample,
805 				struct perf_evsel *evsel,
806 				struct machine *machine)
807 {
808 	struct thread *thread = machine__findnew_thread(machine, sample->pid,
809 							sample->tid);
810 
811 	if (thread == NULL) {
812 		pr_debug("problem processing %d event, skipping it.\n",
813 			event->header.type);
814 		return -1;
815 	}
816 
817 	if (evsel->handler.func != NULL) {
818 		tracepoint_handler f = evsel->handler.func;
819 		return f(evsel, sample);
820 	}
821 
822 	return 0;
823 }
824 
825 static const struct perf_evsel_str_handler lock_tracepoints[] = {
826 	{ "lock:lock_acquire",	 perf_evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
827 	{ "lock:lock_acquired",	 perf_evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
828 	{ "lock:lock_contended", perf_evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
829 	{ "lock:lock_release",	 perf_evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
830 };
831 
read_events(void)832 static int read_events(void)
833 {
834 	struct perf_tool eops = {
835 		.sample		 = process_sample_event,
836 		.comm		 = perf_event__process_comm,
837 		.ordered_samples = true,
838 	};
839 	session = perf_session__new(input_name, O_RDONLY, 0, false, &eops);
840 	if (!session) {
841 		pr_err("Initializing perf session failed\n");
842 		return -1;
843 	}
844 
845 	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
846 		pr_err("Initializing perf session tracepoint handlers failed\n");
847 		return -1;
848 	}
849 
850 	return perf_session__process_events(session, &eops);
851 }
852 
sort_result(void)853 static void sort_result(void)
854 {
855 	unsigned int i;
856 	struct lock_stat *st;
857 
858 	for (i = 0; i < LOCKHASH_SIZE; i++) {
859 		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
860 			insert_to_result(st, compare);
861 		}
862 	}
863 }
864 
__cmd_report(void)865 static int __cmd_report(void)
866 {
867 	setup_pager();
868 
869 	if ((select_key() != 0) ||
870 	    (read_events() != 0))
871 		return -1;
872 
873 	sort_result();
874 	print_result();
875 
876 	return 0;
877 }
878 
__cmd_record(int argc,const char ** argv)879 static int __cmd_record(int argc, const char **argv)
880 {
881 	const char *record_args[] = {
882 		"record", "-R", "-m", "1024", "-c", "1",
883 	};
884 	unsigned int rec_argc, i, j;
885 	const char **rec_argv;
886 
887 	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
888 		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
889 				pr_err("tracepoint %s is not enabled. "
890 				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
891 				       lock_tracepoints[i].name);
892 				return 1;
893 		}
894 	}
895 
896 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
897 	/* factor of 2 is for -e in front of each tracepoint */
898 	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
899 
900 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
901 	if (rec_argv == NULL)
902 		return -ENOMEM;
903 
904 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
905 		rec_argv[i] = strdup(record_args[i]);
906 
907 	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
908 		rec_argv[i++] = "-e";
909 		rec_argv[i++] = strdup(lock_tracepoints[j].name);
910 	}
911 
912 	for (j = 1; j < (unsigned int)argc; j++, i++)
913 		rec_argv[i] = argv[j];
914 
915 	BUG_ON(i != rec_argc);
916 
917 	return cmd_record(i, rec_argv, NULL);
918 }
919 
cmd_lock(int argc,const char ** argv,const char * prefix __maybe_unused)920 int cmd_lock(int argc, const char **argv, const char *prefix __maybe_unused)
921 {
922 	const struct option info_options[] = {
923 	OPT_BOOLEAN('t', "threads", &info_threads,
924 		    "dump thread list in perf.data"),
925 	OPT_BOOLEAN('m', "map", &info_map,
926 		    "map of lock instances (address:name table)"),
927 	OPT_END()
928 	};
929 	const struct option lock_options[] = {
930 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
931 	OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
932 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
933 	OPT_END()
934 	};
935 	const struct option report_options[] = {
936 	OPT_STRING('k', "key", &sort_key, "acquired",
937 		    "key for sorting (acquired / contended / wait_total / wait_max / wait_min)"),
938 	/* TODO: type */
939 	OPT_END()
940 	};
941 	const char * const info_usage[] = {
942 		"perf lock info [<options>]",
943 		NULL
944 	};
945 	const char * const lock_usage[] = {
946 		"perf lock [<options>] {record|report|script|info}",
947 		NULL
948 	};
949 	const char * const report_usage[] = {
950 		"perf lock report [<options>]",
951 		NULL
952 	};
953 	unsigned int i;
954 	int rc = 0;
955 
956 	symbol__init();
957 	for (i = 0; i < LOCKHASH_SIZE; i++)
958 		INIT_LIST_HEAD(lockhash_table + i);
959 
960 	argc = parse_options(argc, argv, lock_options, lock_usage,
961 			     PARSE_OPT_STOP_AT_NON_OPTION);
962 	if (!argc)
963 		usage_with_options(lock_usage, lock_options);
964 
965 	if (!strncmp(argv[0], "rec", 3)) {
966 		return __cmd_record(argc, argv);
967 	} else if (!strncmp(argv[0], "report", 6)) {
968 		trace_handler = &report_lock_ops;
969 		if (argc) {
970 			argc = parse_options(argc, argv,
971 					     report_options, report_usage, 0);
972 			if (argc)
973 				usage_with_options(report_usage, report_options);
974 		}
975 		__cmd_report();
976 	} else if (!strcmp(argv[0], "script")) {
977 		/* Aliased to 'perf script' */
978 		return cmd_script(argc, argv, prefix);
979 	} else if (!strcmp(argv[0], "info")) {
980 		if (argc) {
981 			argc = parse_options(argc, argv,
982 					     info_options, info_usage, 0);
983 			if (argc)
984 				usage_with_options(info_usage, info_options);
985 		}
986 		/* recycling report_lock_ops */
987 		trace_handler = &report_lock_ops;
988 		setup_pager();
989 		if (read_events() != 0)
990 			rc = -1;
991 		else
992 			rc = dump_info();
993 	} else {
994 		usage_with_options(lock_usage, lock_options);
995 	}
996 
997 	return rc;
998 }
999