xref: /openbmc/linux/tools/perf/builtin-lock.c (revision dea54fba)
1 #include <errno.h>
2 #include <inttypes.h>
3 #include "builtin.h"
4 #include "perf.h"
5 
6 #include "util/evlist.h"
7 #include "util/evsel.h"
8 #include "util/util.h"
9 #include "util/cache.h"
10 #include "util/symbol.h"
11 #include "util/thread.h"
12 #include "util/header.h"
13 
14 #include <subcmd/parse-options.h>
15 #include "util/trace-event.h"
16 
17 #include "util/debug.h"
18 #include "util/session.h"
19 #include "util/tool.h"
20 #include "util/data.h"
21 
22 #include <sys/types.h>
23 #include <sys/prctl.h>
24 #include <semaphore.h>
25 #include <pthread.h>
26 #include <math.h>
27 #include <limits.h>
28 
29 #include <linux/list.h>
30 #include <linux/hash.h>
31 #include <linux/kernel.h>
32 
33 static struct perf_session *session;
34 
35 /* based on kernel/lockdep.c */
36 #define LOCKHASH_BITS		12
37 #define LOCKHASH_SIZE		(1UL << LOCKHASH_BITS)
38 
39 static struct list_head lockhash_table[LOCKHASH_SIZE];
40 
41 #define __lockhashfn(key)	hash_long((unsigned long)key, LOCKHASH_BITS)
42 #define lockhashentry(key)	(lockhash_table + __lockhashfn((key)))
43 
44 struct lock_stat {
45 	struct list_head	hash_entry;
46 	struct rb_node		rb;		/* used for sorting */
47 
48 	/*
49 	 * FIXME: perf_evsel__intval() returns u64,
50 	 * so address of lockdep_map should be dealed as 64bit.
51 	 * Is there more better solution?
52 	 */
53 	void			*addr;		/* address of lockdep_map, used as ID */
54 	char			*name;		/* for strcpy(), we cannot use const */
55 
56 	unsigned int		nr_acquire;
57 	unsigned int		nr_acquired;
58 	unsigned int		nr_contended;
59 	unsigned int		nr_release;
60 
61 	unsigned int		nr_readlock;
62 	unsigned int		nr_trylock;
63 
64 	/* these times are in nano sec. */
65 	u64                     avg_wait_time;
66 	u64			wait_time_total;
67 	u64			wait_time_min;
68 	u64			wait_time_max;
69 
70 	int			discard; /* flag of blacklist */
71 };
72 
73 /*
74  * States of lock_seq_stat
75  *
76  * UNINITIALIZED is required for detecting first event of acquire.
77  * As the nature of lock events, there is no guarantee
78  * that the first event for the locks are acquire,
79  * it can be acquired, contended or release.
80  */
81 #define SEQ_STATE_UNINITIALIZED      0	       /* initial state */
82 #define SEQ_STATE_RELEASED	1
83 #define SEQ_STATE_ACQUIRING	2
84 #define SEQ_STATE_ACQUIRED	3
85 #define SEQ_STATE_READ_ACQUIRED	4
86 #define SEQ_STATE_CONTENDED	5
87 
88 /*
89  * MAX_LOCK_DEPTH
90  * Imported from include/linux/sched.h.
91  * Should this be synchronized?
92  */
93 #define MAX_LOCK_DEPTH 48
94 
95 /*
96  * struct lock_seq_stat:
97  * Place to put on state of one lock sequence
98  * 1) acquire -> acquired -> release
99  * 2) acquire -> contended -> acquired -> release
100  * 3) acquire (with read or try) -> release
101  * 4) Are there other patterns?
102  */
103 struct lock_seq_stat {
104 	struct list_head        list;
105 	int			state;
106 	u64			prev_event_time;
107 	void                    *addr;
108 
109 	int                     read_count;
110 };
111 
112 struct thread_stat {
113 	struct rb_node		rb;
114 
115 	u32                     tid;
116 	struct list_head        seq_list;
117 };
118 
119 static struct rb_root		thread_stats;
120 
121 static struct thread_stat *thread_stat_find(u32 tid)
122 {
123 	struct rb_node *node;
124 	struct thread_stat *st;
125 
126 	node = thread_stats.rb_node;
127 	while (node) {
128 		st = container_of(node, struct thread_stat, rb);
129 		if (st->tid == tid)
130 			return st;
131 		else if (tid < st->tid)
132 			node = node->rb_left;
133 		else
134 			node = node->rb_right;
135 	}
136 
137 	return NULL;
138 }
139 
140 static void thread_stat_insert(struct thread_stat *new)
141 {
142 	struct rb_node **rb = &thread_stats.rb_node;
143 	struct rb_node *parent = NULL;
144 	struct thread_stat *p;
145 
146 	while (*rb) {
147 		p = container_of(*rb, struct thread_stat, rb);
148 		parent = *rb;
149 
150 		if (new->tid < p->tid)
151 			rb = &(*rb)->rb_left;
152 		else if (new->tid > p->tid)
153 			rb = &(*rb)->rb_right;
154 		else
155 			BUG_ON("inserting invalid thread_stat\n");
156 	}
157 
158 	rb_link_node(&new->rb, parent, rb);
159 	rb_insert_color(&new->rb, &thread_stats);
160 }
161 
162 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
163 {
164 	struct thread_stat *st;
165 
166 	st = thread_stat_find(tid);
167 	if (st)
168 		return st;
169 
170 	st = zalloc(sizeof(struct thread_stat));
171 	if (!st) {
172 		pr_err("memory allocation failed\n");
173 		return NULL;
174 	}
175 
176 	st->tid = tid;
177 	INIT_LIST_HEAD(&st->seq_list);
178 
179 	thread_stat_insert(st);
180 
181 	return st;
182 }
183 
184 static struct thread_stat *thread_stat_findnew_first(u32 tid);
185 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
186 	thread_stat_findnew_first;
187 
188 static struct thread_stat *thread_stat_findnew_first(u32 tid)
189 {
190 	struct thread_stat *st;
191 
192 	st = zalloc(sizeof(struct thread_stat));
193 	if (!st) {
194 		pr_err("memory allocation failed\n");
195 		return NULL;
196 	}
197 	st->tid = tid;
198 	INIT_LIST_HEAD(&st->seq_list);
199 
200 	rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
201 	rb_insert_color(&st->rb, &thread_stats);
202 
203 	thread_stat_findnew = thread_stat_findnew_after_first;
204 	return st;
205 }
206 
207 /* build simple key function one is bigger than two */
208 #define SINGLE_KEY(member)						\
209 	static int lock_stat_key_ ## member(struct lock_stat *one,	\
210 					 struct lock_stat *two)		\
211 	{								\
212 		return one->member > two->member;			\
213 	}
214 
215 SINGLE_KEY(nr_acquired)
216 SINGLE_KEY(nr_contended)
217 SINGLE_KEY(avg_wait_time)
218 SINGLE_KEY(wait_time_total)
219 SINGLE_KEY(wait_time_max)
220 
221 static int lock_stat_key_wait_time_min(struct lock_stat *one,
222 					struct lock_stat *two)
223 {
224 	u64 s1 = one->wait_time_min;
225 	u64 s2 = two->wait_time_min;
226 	if (s1 == ULLONG_MAX)
227 		s1 = 0;
228 	if (s2 == ULLONG_MAX)
229 		s2 = 0;
230 	return s1 > s2;
231 }
232 
233 struct lock_key {
234 	/*
235 	 * name: the value for specify by user
236 	 * this should be simpler than raw name of member
237 	 * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
238 	 */
239 	const char		*name;
240 	int			(*key)(struct lock_stat*, struct lock_stat*);
241 };
242 
243 static const char		*sort_key = "acquired";
244 
245 static int			(*compare)(struct lock_stat *, struct lock_stat *);
246 
247 static struct rb_root		result;	/* place to store sorted data */
248 
249 #define DEF_KEY_LOCK(name, fn_suffix)	\
250 	{ #name, lock_stat_key_ ## fn_suffix }
251 struct lock_key keys[] = {
252 	DEF_KEY_LOCK(acquired, nr_acquired),
253 	DEF_KEY_LOCK(contended, nr_contended),
254 	DEF_KEY_LOCK(avg_wait, avg_wait_time),
255 	DEF_KEY_LOCK(wait_total, wait_time_total),
256 	DEF_KEY_LOCK(wait_min, wait_time_min),
257 	DEF_KEY_LOCK(wait_max, wait_time_max),
258 
259 	/* extra comparisons much complicated should be here */
260 
261 	{ NULL, NULL }
262 };
263 
264 static int select_key(void)
265 {
266 	int i;
267 
268 	for (i = 0; keys[i].name; i++) {
269 		if (!strcmp(keys[i].name, sort_key)) {
270 			compare = keys[i].key;
271 			return 0;
272 		}
273 	}
274 
275 	pr_err("Unknown compare key: %s\n", sort_key);
276 
277 	return -1;
278 }
279 
280 static void insert_to_result(struct lock_stat *st,
281 			     int (*bigger)(struct lock_stat *, struct lock_stat *))
282 {
283 	struct rb_node **rb = &result.rb_node;
284 	struct rb_node *parent = NULL;
285 	struct lock_stat *p;
286 
287 	while (*rb) {
288 		p = container_of(*rb, struct lock_stat, rb);
289 		parent = *rb;
290 
291 		if (bigger(st, p))
292 			rb = &(*rb)->rb_left;
293 		else
294 			rb = &(*rb)->rb_right;
295 	}
296 
297 	rb_link_node(&st->rb, parent, rb);
298 	rb_insert_color(&st->rb, &result);
299 }
300 
301 /* returns left most element of result, and erase it */
302 static struct lock_stat *pop_from_result(void)
303 {
304 	struct rb_node *node = result.rb_node;
305 
306 	if (!node)
307 		return NULL;
308 
309 	while (node->rb_left)
310 		node = node->rb_left;
311 
312 	rb_erase(node, &result);
313 	return container_of(node, struct lock_stat, rb);
314 }
315 
316 static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
317 {
318 	struct list_head *entry = lockhashentry(addr);
319 	struct lock_stat *ret, *new;
320 
321 	list_for_each_entry(ret, entry, hash_entry) {
322 		if (ret->addr == addr)
323 			return ret;
324 	}
325 
326 	new = zalloc(sizeof(struct lock_stat));
327 	if (!new)
328 		goto alloc_failed;
329 
330 	new->addr = addr;
331 	new->name = zalloc(sizeof(char) * strlen(name) + 1);
332 	if (!new->name) {
333 		free(new);
334 		goto alloc_failed;
335 	}
336 
337 	strcpy(new->name, name);
338 	new->wait_time_min = ULLONG_MAX;
339 
340 	list_add(&new->hash_entry, entry);
341 	return new;
342 
343 alloc_failed:
344 	pr_err("memory allocation failed\n");
345 	return NULL;
346 }
347 
348 struct trace_lock_handler {
349 	int (*acquire_event)(struct perf_evsel *evsel,
350 			     struct perf_sample *sample);
351 
352 	int (*acquired_event)(struct perf_evsel *evsel,
353 			      struct perf_sample *sample);
354 
355 	int (*contended_event)(struct perf_evsel *evsel,
356 			       struct perf_sample *sample);
357 
358 	int (*release_event)(struct perf_evsel *evsel,
359 			     struct perf_sample *sample);
360 };
361 
362 static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
363 {
364 	struct lock_seq_stat *seq;
365 
366 	list_for_each_entry(seq, &ts->seq_list, list) {
367 		if (seq->addr == addr)
368 			return seq;
369 	}
370 
371 	seq = zalloc(sizeof(struct lock_seq_stat));
372 	if (!seq) {
373 		pr_err("memory allocation failed\n");
374 		return NULL;
375 	}
376 	seq->state = SEQ_STATE_UNINITIALIZED;
377 	seq->addr = addr;
378 
379 	list_add(&seq->list, &ts->seq_list);
380 	return seq;
381 }
382 
383 enum broken_state {
384 	BROKEN_ACQUIRE,
385 	BROKEN_ACQUIRED,
386 	BROKEN_CONTENDED,
387 	BROKEN_RELEASE,
388 	BROKEN_MAX,
389 };
390 
391 static int bad_hist[BROKEN_MAX];
392 
393 enum acquire_flags {
394 	TRY_LOCK = 1,
395 	READ_LOCK = 2,
396 };
397 
398 static int report_lock_acquire_event(struct perf_evsel *evsel,
399 				     struct perf_sample *sample)
400 {
401 	void *addr;
402 	struct lock_stat *ls;
403 	struct thread_stat *ts;
404 	struct lock_seq_stat *seq;
405 	const char *name = perf_evsel__strval(evsel, sample, "name");
406 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
407 	int flag = perf_evsel__intval(evsel, sample, "flag");
408 
409 	memcpy(&addr, &tmp, sizeof(void *));
410 
411 	ls = lock_stat_findnew(addr, name);
412 	if (!ls)
413 		return -ENOMEM;
414 	if (ls->discard)
415 		return 0;
416 
417 	ts = thread_stat_findnew(sample->tid);
418 	if (!ts)
419 		return -ENOMEM;
420 
421 	seq = get_seq(ts, addr);
422 	if (!seq)
423 		return -ENOMEM;
424 
425 	switch (seq->state) {
426 	case SEQ_STATE_UNINITIALIZED:
427 	case SEQ_STATE_RELEASED:
428 		if (!flag) {
429 			seq->state = SEQ_STATE_ACQUIRING;
430 		} else {
431 			if (flag & TRY_LOCK)
432 				ls->nr_trylock++;
433 			if (flag & READ_LOCK)
434 				ls->nr_readlock++;
435 			seq->state = SEQ_STATE_READ_ACQUIRED;
436 			seq->read_count = 1;
437 			ls->nr_acquired++;
438 		}
439 		break;
440 	case SEQ_STATE_READ_ACQUIRED:
441 		if (flag & READ_LOCK) {
442 			seq->read_count++;
443 			ls->nr_acquired++;
444 			goto end;
445 		} else {
446 			goto broken;
447 		}
448 		break;
449 	case SEQ_STATE_ACQUIRED:
450 	case SEQ_STATE_ACQUIRING:
451 	case SEQ_STATE_CONTENDED:
452 broken:
453 		/* broken lock sequence, discard it */
454 		ls->discard = 1;
455 		bad_hist[BROKEN_ACQUIRE]++;
456 		list_del(&seq->list);
457 		free(seq);
458 		goto end;
459 	default:
460 		BUG_ON("Unknown state of lock sequence found!\n");
461 		break;
462 	}
463 
464 	ls->nr_acquire++;
465 	seq->prev_event_time = sample->time;
466 end:
467 	return 0;
468 }
469 
470 static int report_lock_acquired_event(struct perf_evsel *evsel,
471 				      struct perf_sample *sample)
472 {
473 	void *addr;
474 	struct lock_stat *ls;
475 	struct thread_stat *ts;
476 	struct lock_seq_stat *seq;
477 	u64 contended_term;
478 	const char *name = perf_evsel__strval(evsel, sample, "name");
479 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
480 
481 	memcpy(&addr, &tmp, sizeof(void *));
482 
483 	ls = lock_stat_findnew(addr, name);
484 	if (!ls)
485 		return -ENOMEM;
486 	if (ls->discard)
487 		return 0;
488 
489 	ts = thread_stat_findnew(sample->tid);
490 	if (!ts)
491 		return -ENOMEM;
492 
493 	seq = get_seq(ts, addr);
494 	if (!seq)
495 		return -ENOMEM;
496 
497 	switch (seq->state) {
498 	case SEQ_STATE_UNINITIALIZED:
499 		/* orphan event, do nothing */
500 		return 0;
501 	case SEQ_STATE_ACQUIRING:
502 		break;
503 	case SEQ_STATE_CONTENDED:
504 		contended_term = sample->time - seq->prev_event_time;
505 		ls->wait_time_total += contended_term;
506 		if (contended_term < ls->wait_time_min)
507 			ls->wait_time_min = contended_term;
508 		if (ls->wait_time_max < contended_term)
509 			ls->wait_time_max = contended_term;
510 		break;
511 	case SEQ_STATE_RELEASED:
512 	case SEQ_STATE_ACQUIRED:
513 	case SEQ_STATE_READ_ACQUIRED:
514 		/* broken lock sequence, discard it */
515 		ls->discard = 1;
516 		bad_hist[BROKEN_ACQUIRED]++;
517 		list_del(&seq->list);
518 		free(seq);
519 		goto end;
520 	default:
521 		BUG_ON("Unknown state of lock sequence found!\n");
522 		break;
523 	}
524 
525 	seq->state = SEQ_STATE_ACQUIRED;
526 	ls->nr_acquired++;
527 	ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
528 	seq->prev_event_time = sample->time;
529 end:
530 	return 0;
531 }
532 
533 static int report_lock_contended_event(struct perf_evsel *evsel,
534 				       struct perf_sample *sample)
535 {
536 	void *addr;
537 	struct lock_stat *ls;
538 	struct thread_stat *ts;
539 	struct lock_seq_stat *seq;
540 	const char *name = perf_evsel__strval(evsel, sample, "name");
541 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
542 
543 	memcpy(&addr, &tmp, sizeof(void *));
544 
545 	ls = lock_stat_findnew(addr, name);
546 	if (!ls)
547 		return -ENOMEM;
548 	if (ls->discard)
549 		return 0;
550 
551 	ts = thread_stat_findnew(sample->tid);
552 	if (!ts)
553 		return -ENOMEM;
554 
555 	seq = get_seq(ts, addr);
556 	if (!seq)
557 		return -ENOMEM;
558 
559 	switch (seq->state) {
560 	case SEQ_STATE_UNINITIALIZED:
561 		/* orphan event, do nothing */
562 		return 0;
563 	case SEQ_STATE_ACQUIRING:
564 		break;
565 	case SEQ_STATE_RELEASED:
566 	case SEQ_STATE_ACQUIRED:
567 	case SEQ_STATE_READ_ACQUIRED:
568 	case SEQ_STATE_CONTENDED:
569 		/* broken lock sequence, discard it */
570 		ls->discard = 1;
571 		bad_hist[BROKEN_CONTENDED]++;
572 		list_del(&seq->list);
573 		free(seq);
574 		goto end;
575 	default:
576 		BUG_ON("Unknown state of lock sequence found!\n");
577 		break;
578 	}
579 
580 	seq->state = SEQ_STATE_CONTENDED;
581 	ls->nr_contended++;
582 	ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
583 	seq->prev_event_time = sample->time;
584 end:
585 	return 0;
586 }
587 
588 static int report_lock_release_event(struct perf_evsel *evsel,
589 				     struct perf_sample *sample)
590 {
591 	void *addr;
592 	struct lock_stat *ls;
593 	struct thread_stat *ts;
594 	struct lock_seq_stat *seq;
595 	const char *name = perf_evsel__strval(evsel, sample, "name");
596 	u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
597 
598 	memcpy(&addr, &tmp, sizeof(void *));
599 
600 	ls = lock_stat_findnew(addr, name);
601 	if (!ls)
602 		return -ENOMEM;
603 	if (ls->discard)
604 		return 0;
605 
606 	ts = thread_stat_findnew(sample->tid);
607 	if (!ts)
608 		return -ENOMEM;
609 
610 	seq = get_seq(ts, addr);
611 	if (!seq)
612 		return -ENOMEM;
613 
614 	switch (seq->state) {
615 	case SEQ_STATE_UNINITIALIZED:
616 		goto end;
617 	case SEQ_STATE_ACQUIRED:
618 		break;
619 	case SEQ_STATE_READ_ACQUIRED:
620 		seq->read_count--;
621 		BUG_ON(seq->read_count < 0);
622 		if (!seq->read_count) {
623 			ls->nr_release++;
624 			goto end;
625 		}
626 		break;
627 	case SEQ_STATE_ACQUIRING:
628 	case SEQ_STATE_CONTENDED:
629 	case SEQ_STATE_RELEASED:
630 		/* broken lock sequence, discard it */
631 		ls->discard = 1;
632 		bad_hist[BROKEN_RELEASE]++;
633 		goto free_seq;
634 	default:
635 		BUG_ON("Unknown state of lock sequence found!\n");
636 		break;
637 	}
638 
639 	ls->nr_release++;
640 free_seq:
641 	list_del(&seq->list);
642 	free(seq);
643 end:
644 	return 0;
645 }
646 
647 /* lock oriented handlers */
648 /* TODO: handlers for CPU oriented, thread oriented */
649 static struct trace_lock_handler report_lock_ops  = {
650 	.acquire_event		= report_lock_acquire_event,
651 	.acquired_event		= report_lock_acquired_event,
652 	.contended_event	= report_lock_contended_event,
653 	.release_event		= report_lock_release_event,
654 };
655 
656 static struct trace_lock_handler *trace_handler;
657 
658 static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
659 					     struct perf_sample *sample)
660 {
661 	if (trace_handler->acquire_event)
662 		return trace_handler->acquire_event(evsel, sample);
663 	return 0;
664 }
665 
666 static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
667 					      struct perf_sample *sample)
668 {
669 	if (trace_handler->acquired_event)
670 		return trace_handler->acquired_event(evsel, sample);
671 	return 0;
672 }
673 
674 static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
675 					      struct perf_sample *sample)
676 {
677 	if (trace_handler->contended_event)
678 		return trace_handler->contended_event(evsel, sample);
679 	return 0;
680 }
681 
682 static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
683 					    struct perf_sample *sample)
684 {
685 	if (trace_handler->release_event)
686 		return trace_handler->release_event(evsel, sample);
687 	return 0;
688 }
689 
690 static void print_bad_events(int bad, int total)
691 {
692 	/* Output for debug, this have to be removed */
693 	int i;
694 	const char *name[4] =
695 		{ "acquire", "acquired", "contended", "release" };
696 
697 	pr_info("\n=== output for debug===\n\n");
698 	pr_info("bad: %d, total: %d\n", bad, total);
699 	pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
700 	pr_info("histogram of events caused bad sequence\n");
701 	for (i = 0; i < BROKEN_MAX; i++)
702 		pr_info(" %10s: %d\n", name[i], bad_hist[i]);
703 }
704 
705 /* TODO: various way to print, coloring, nano or milli sec */
706 static void print_result(void)
707 {
708 	struct lock_stat *st;
709 	char cut_name[20];
710 	int bad, total;
711 
712 	pr_info("%20s ", "Name");
713 	pr_info("%10s ", "acquired");
714 	pr_info("%10s ", "contended");
715 
716 	pr_info("%15s ", "avg wait (ns)");
717 	pr_info("%15s ", "total wait (ns)");
718 	pr_info("%15s ", "max wait (ns)");
719 	pr_info("%15s ", "min wait (ns)");
720 
721 	pr_info("\n\n");
722 
723 	bad = total = 0;
724 	while ((st = pop_from_result())) {
725 		total++;
726 		if (st->discard) {
727 			bad++;
728 			continue;
729 		}
730 		bzero(cut_name, 20);
731 
732 		if (strlen(st->name) < 16) {
733 			/* output raw name */
734 			pr_info("%20s ", st->name);
735 		} else {
736 			strncpy(cut_name, st->name, 16);
737 			cut_name[16] = '.';
738 			cut_name[17] = '.';
739 			cut_name[18] = '.';
740 			cut_name[19] = '\0';
741 			/* cut off name for saving output style */
742 			pr_info("%20s ", cut_name);
743 		}
744 
745 		pr_info("%10u ", st->nr_acquired);
746 		pr_info("%10u ", st->nr_contended);
747 
748 		pr_info("%15" PRIu64 " ", st->avg_wait_time);
749 		pr_info("%15" PRIu64 " ", st->wait_time_total);
750 		pr_info("%15" PRIu64 " ", st->wait_time_max);
751 		pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
752 		       0 : st->wait_time_min);
753 		pr_info("\n");
754 	}
755 
756 	print_bad_events(bad, total);
757 }
758 
759 static bool info_threads, info_map;
760 
761 static void dump_threads(void)
762 {
763 	struct thread_stat *st;
764 	struct rb_node *node;
765 	struct thread *t;
766 
767 	pr_info("%10s: comm\n", "Thread ID");
768 
769 	node = rb_first(&thread_stats);
770 	while (node) {
771 		st = container_of(node, struct thread_stat, rb);
772 		t = perf_session__findnew(session, st->tid);
773 		pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
774 		node = rb_next(node);
775 		thread__put(t);
776 	};
777 }
778 
779 static void dump_map(void)
780 {
781 	unsigned int i;
782 	struct lock_stat *st;
783 
784 	pr_info("Address of instance: name of class\n");
785 	for (i = 0; i < LOCKHASH_SIZE; i++) {
786 		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
787 			pr_info(" %p: %s\n", st->addr, st->name);
788 		}
789 	}
790 }
791 
792 static int dump_info(void)
793 {
794 	int rc = 0;
795 
796 	if (info_threads)
797 		dump_threads();
798 	else if (info_map)
799 		dump_map();
800 	else {
801 		rc = -1;
802 		pr_err("Unknown type of information\n");
803 	}
804 
805 	return rc;
806 }
807 
808 typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
809 				  struct perf_sample *sample);
810 
811 static int process_sample_event(struct perf_tool *tool __maybe_unused,
812 				union perf_event *event,
813 				struct perf_sample *sample,
814 				struct perf_evsel *evsel,
815 				struct machine *machine)
816 {
817 	int err = 0;
818 	struct thread *thread = machine__findnew_thread(machine, sample->pid,
819 							sample->tid);
820 
821 	if (thread == NULL) {
822 		pr_debug("problem processing %d event, skipping it.\n",
823 			event->header.type);
824 		return -1;
825 	}
826 
827 	if (evsel->handler != NULL) {
828 		tracepoint_handler f = evsel->handler;
829 		err = f(evsel, sample);
830 	}
831 
832 	thread__put(thread);
833 
834 	return err;
835 }
836 
837 static void sort_result(void)
838 {
839 	unsigned int i;
840 	struct lock_stat *st;
841 
842 	for (i = 0; i < LOCKHASH_SIZE; i++) {
843 		list_for_each_entry(st, &lockhash_table[i], hash_entry) {
844 			insert_to_result(st, compare);
845 		}
846 	}
847 }
848 
849 static const struct perf_evsel_str_handler lock_tracepoints[] = {
850 	{ "lock:lock_acquire",	 perf_evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
851 	{ "lock:lock_acquired",	 perf_evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
852 	{ "lock:lock_contended", perf_evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
853 	{ "lock:lock_release",	 perf_evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
854 };
855 
856 static bool force;
857 
858 static int __cmd_report(bool display_info)
859 {
860 	int err = -EINVAL;
861 	struct perf_tool eops = {
862 		.sample		 = process_sample_event,
863 		.comm		 = perf_event__process_comm,
864 		.namespaces	 = perf_event__process_namespaces,
865 		.ordered_events	 = true,
866 	};
867 	struct perf_data_file file = {
868 		.path = input_name,
869 		.mode = PERF_DATA_MODE_READ,
870 		.force = force,
871 	};
872 
873 	session = perf_session__new(&file, false, &eops);
874 	if (!session) {
875 		pr_err("Initializing perf session failed\n");
876 		return -1;
877 	}
878 
879 	symbol__init(&session->header.env);
880 
881 	if (!perf_session__has_traces(session, "lock record"))
882 		goto out_delete;
883 
884 	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
885 		pr_err("Initializing perf session tracepoint handlers failed\n");
886 		goto out_delete;
887 	}
888 
889 	if (select_key())
890 		goto out_delete;
891 
892 	err = perf_session__process_events(session);
893 	if (err)
894 		goto out_delete;
895 
896 	setup_pager();
897 	if (display_info) /* used for info subcommand */
898 		err = dump_info();
899 	else {
900 		sort_result();
901 		print_result();
902 	}
903 
904 out_delete:
905 	perf_session__delete(session);
906 	return err;
907 }
908 
909 static int __cmd_record(int argc, const char **argv)
910 {
911 	const char *record_args[] = {
912 		"record", "-R", "-m", "1024", "-c", "1",
913 	};
914 	unsigned int rec_argc, i, j, ret;
915 	const char **rec_argv;
916 
917 	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
918 		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
919 				pr_err("tracepoint %s is not enabled. "
920 				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
921 				       lock_tracepoints[i].name);
922 				return 1;
923 		}
924 	}
925 
926 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
927 	/* factor of 2 is for -e in front of each tracepoint */
928 	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
929 
930 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
931 	if (!rec_argv)
932 		return -ENOMEM;
933 
934 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
935 		rec_argv[i] = strdup(record_args[i]);
936 
937 	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
938 		rec_argv[i++] = "-e";
939 		rec_argv[i++] = strdup(lock_tracepoints[j].name);
940 	}
941 
942 	for (j = 1; j < (unsigned int)argc; j++, i++)
943 		rec_argv[i] = argv[j];
944 
945 	BUG_ON(i != rec_argc);
946 
947 	ret = cmd_record(i, rec_argv);
948 	free(rec_argv);
949 	return ret;
950 }
951 
952 int cmd_lock(int argc, const char **argv)
953 {
954 	const struct option lock_options[] = {
955 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
956 	OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
957 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
958 	OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
959 	OPT_END()
960 	};
961 
962 	const struct option info_options[] = {
963 	OPT_BOOLEAN('t', "threads", &info_threads,
964 		    "dump thread list in perf.data"),
965 	OPT_BOOLEAN('m', "map", &info_map,
966 		    "map of lock instances (address:name table)"),
967 	OPT_PARENT(lock_options)
968 	};
969 
970 	const struct option report_options[] = {
971 	OPT_STRING('k', "key", &sort_key, "acquired",
972 		    "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
973 	/* TODO: type */
974 	OPT_PARENT(lock_options)
975 	};
976 
977 	const char * const info_usage[] = {
978 		"perf lock info [<options>]",
979 		NULL
980 	};
981 	const char *const lock_subcommands[] = { "record", "report", "script",
982 						 "info", NULL };
983 	const char *lock_usage[] = {
984 		NULL,
985 		NULL
986 	};
987 	const char * const report_usage[] = {
988 		"perf lock report [<options>]",
989 		NULL
990 	};
991 	unsigned int i;
992 	int rc = 0;
993 
994 	for (i = 0; i < LOCKHASH_SIZE; i++)
995 		INIT_LIST_HEAD(lockhash_table + i);
996 
997 	argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
998 					lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
999 	if (!argc)
1000 		usage_with_options(lock_usage, lock_options);
1001 
1002 	if (!strncmp(argv[0], "rec", 3)) {
1003 		return __cmd_record(argc, argv);
1004 	} else if (!strncmp(argv[0], "report", 6)) {
1005 		trace_handler = &report_lock_ops;
1006 		if (argc) {
1007 			argc = parse_options(argc, argv,
1008 					     report_options, report_usage, 0);
1009 			if (argc)
1010 				usage_with_options(report_usage, report_options);
1011 		}
1012 		rc = __cmd_report(false);
1013 	} else if (!strcmp(argv[0], "script")) {
1014 		/* Aliased to 'perf script' */
1015 		return cmd_script(argc, argv);
1016 	} else if (!strcmp(argv[0], "info")) {
1017 		if (argc) {
1018 			argc = parse_options(argc, argv,
1019 					     info_options, info_usage, 0);
1020 			if (argc)
1021 				usage_with_options(info_usage, info_options);
1022 		}
1023 		/* recycling report_lock_ops */
1024 		trace_handler = &report_lock_ops;
1025 		rc = __cmd_report(true);
1026 	} else {
1027 		usage_with_options(lock_usage, lock_options);
1028 	}
1029 
1030 	return rc;
1031 }
1032