xref: /openbmc/linux/tools/perf/builtin-lock.c (revision ab3a560a)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <inttypes.h>
4 #include "builtin.h"
5 #include "perf.h"
6 
7 #include "util/evlist.h" // for struct evsel_str_handler
8 #include "util/evsel.h"
9 #include "util/symbol.h"
10 #include "util/thread.h"
11 #include "util/header.h"
12 
13 #include <subcmd/pager.h>
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 #include "util/string2.h"
22 
23 #include <sys/types.h>
24 #include <sys/prctl.h>
25 #include <semaphore.h>
26 #include <pthread.h>
27 #include <math.h>
28 #include <limits.h>
29 
30 #include <linux/list.h>
31 #include <linux/hash.h>
32 #include <linux/kernel.h>
33 #include <linux/zalloc.h>
34 #include <linux/err.h>
35 
36 static struct perf_session *session;
37 
38 /* based on kernel/lockdep.c */
39 #define LOCKHASH_BITS		12
40 #define LOCKHASH_SIZE		(1UL << LOCKHASH_BITS)
41 
42 static struct hlist_head lockhash_table[LOCKHASH_SIZE];
43 
44 #define __lockhashfn(key)	hash_long((unsigned long)key, LOCKHASH_BITS)
45 #define lockhashentry(key)	(lockhash_table + __lockhashfn((key)))
46 
47 struct lock_stat {
48 	struct hlist_node	hash_entry;
49 	struct rb_node		rb;		/* used for sorting */
50 
51 	u64			addr;		/* address of lockdep_map, used as ID */
52 	char			*name;		/* for strcpy(), we cannot use const */
53 
54 	unsigned int		nr_acquire;
55 	unsigned int		nr_acquired;
56 	unsigned int		nr_contended;
57 	unsigned int		nr_release;
58 
59 	unsigned int		nr_readlock;
60 	unsigned int		nr_trylock;
61 
62 	/* these times are in nano sec. */
63 	u64                     avg_wait_time;
64 	u64			wait_time_total;
65 	u64			wait_time_min;
66 	u64			wait_time_max;
67 
68 	int			discard; /* flag of blacklist */
69 	int			combined;
70 };
71 
72 /*
73  * States of lock_seq_stat
74  *
75  * UNINITIALIZED is required for detecting first event of acquire.
76  * As the nature of lock events, there is no guarantee
77  * that the first event for the locks are acquire,
78  * it can be acquired, contended or release.
79  */
80 #define SEQ_STATE_UNINITIALIZED      0	       /* initial state */
81 #define SEQ_STATE_RELEASED	1
82 #define SEQ_STATE_ACQUIRING	2
83 #define SEQ_STATE_ACQUIRED	3
84 #define SEQ_STATE_READ_ACQUIRED	4
85 #define SEQ_STATE_CONTENDED	5
86 
87 /*
88  * MAX_LOCK_DEPTH
89  * Imported from include/linux/sched.h.
90  * Should this be synchronized?
91  */
92 #define MAX_LOCK_DEPTH 48
93 
94 /*
95  * struct lock_seq_stat:
96  * Place to put on state of one lock sequence
97  * 1) acquire -> acquired -> release
98  * 2) acquire -> contended -> acquired -> release
99  * 3) acquire (with read or try) -> release
100  * 4) Are there other patterns?
101  */
102 struct lock_seq_stat {
103 	struct list_head        list;
104 	int			state;
105 	u64			prev_event_time;
106 	u64                     addr;
107 
108 	int                     read_count;
109 };
110 
111 struct thread_stat {
112 	struct rb_node		rb;
113 
114 	u32                     tid;
115 	struct list_head        seq_list;
116 };
117 
118 static struct rb_root		thread_stats;
119 
120 static bool combine_locks;
121 
122 static struct thread_stat *thread_stat_find(u32 tid)
123 {
124 	struct rb_node *node;
125 	struct thread_stat *st;
126 
127 	node = thread_stats.rb_node;
128 	while (node) {
129 		st = container_of(node, struct thread_stat, rb);
130 		if (st->tid == tid)
131 			return st;
132 		else if (tid < st->tid)
133 			node = node->rb_left;
134 		else
135 			node = node->rb_right;
136 	}
137 
138 	return NULL;
139 }
140 
141 static void thread_stat_insert(struct thread_stat *new)
142 {
143 	struct rb_node **rb = &thread_stats.rb_node;
144 	struct rb_node *parent = NULL;
145 	struct thread_stat *p;
146 
147 	while (*rb) {
148 		p = container_of(*rb, struct thread_stat, rb);
149 		parent = *rb;
150 
151 		if (new->tid < p->tid)
152 			rb = &(*rb)->rb_left;
153 		else if (new->tid > p->tid)
154 			rb = &(*rb)->rb_right;
155 		else
156 			BUG_ON("inserting invalid thread_stat\n");
157 	}
158 
159 	rb_link_node(&new->rb, parent, rb);
160 	rb_insert_color(&new->rb, &thread_stats);
161 }
162 
163 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
164 {
165 	struct thread_stat *st;
166 
167 	st = thread_stat_find(tid);
168 	if (st)
169 		return st;
170 
171 	st = zalloc(sizeof(struct thread_stat));
172 	if (!st) {
173 		pr_err("memory allocation failed\n");
174 		return NULL;
175 	}
176 
177 	st->tid = tid;
178 	INIT_LIST_HEAD(&st->seq_list);
179 
180 	thread_stat_insert(st);
181 
182 	return st;
183 }
184 
185 static struct thread_stat *thread_stat_findnew_first(u32 tid);
186 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
187 	thread_stat_findnew_first;
188 
189 static struct thread_stat *thread_stat_findnew_first(u32 tid)
190 {
191 	struct thread_stat *st;
192 
193 	st = zalloc(sizeof(struct thread_stat));
194 	if (!st) {
195 		pr_err("memory allocation failed\n");
196 		return NULL;
197 	}
198 	st->tid = tid;
199 	INIT_LIST_HEAD(&st->seq_list);
200 
201 	rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
202 	rb_insert_color(&st->rb, &thread_stats);
203 
204 	thread_stat_findnew = thread_stat_findnew_after_first;
205 	return st;
206 }
207 
208 /* build simple key function one is bigger than two */
209 #define SINGLE_KEY(member)						\
210 	static int lock_stat_key_ ## member(struct lock_stat *one,	\
211 					 struct lock_stat *two)		\
212 	{								\
213 		return one->member > two->member;			\
214 	}
215 
216 SINGLE_KEY(nr_acquired)
217 SINGLE_KEY(nr_contended)
218 SINGLE_KEY(avg_wait_time)
219 SINGLE_KEY(wait_time_total)
220 SINGLE_KEY(wait_time_max)
221 
222 static int lock_stat_key_wait_time_min(struct lock_stat *one,
223 					struct lock_stat *two)
224 {
225 	u64 s1 = one->wait_time_min;
226 	u64 s2 = two->wait_time_min;
227 	if (s1 == ULLONG_MAX)
228 		s1 = 0;
229 	if (s2 == ULLONG_MAX)
230 		s2 = 0;
231 	return s1 > s2;
232 }
233 
234 struct lock_key {
235 	/*
236 	 * name: the value for specify by user
237 	 * this should be simpler than raw name of member
238 	 * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
239 	 */
240 	const char		*name;
241 	/* header: the string printed on the header line */
242 	const char		*header;
243 	/* len: the printing width of the field */
244 	int			len;
245 	/* key: a pointer to function to compare two lock stats for sorting */
246 	int			(*key)(struct lock_stat*, struct lock_stat*);
247 	/* print: a pointer to function to print a given lock stats */
248 	void			(*print)(struct lock_key*, struct lock_stat*);
249 	/* list: list entry to link this */
250 	struct list_head	list;
251 };
252 
253 #define PRINT_KEY(member)						\
254 static void lock_stat_key_print_ ## member(struct lock_key *key,	\
255 					   struct lock_stat *ls)	\
256 {									\
257 	pr_info("%*llu", key->len, (unsigned long long)ls->member);	\
258 }
259 
260 PRINT_KEY(nr_acquired)
261 PRINT_KEY(nr_contended)
262 PRINT_KEY(avg_wait_time)
263 PRINT_KEY(wait_time_total)
264 PRINT_KEY(wait_time_max)
265 
266 static void lock_stat_key_print_wait_time_min(struct lock_key *key,
267 					      struct lock_stat *ls)
268 {
269 	u64 wait_time = ls->wait_time_min;
270 
271 	if (wait_time == ULLONG_MAX)
272 		wait_time = 0;
273 
274 	pr_info("%*"PRIu64, key->len, wait_time);
275 }
276 
277 
278 static const char		*sort_key = "acquired";
279 
280 static int			(*compare)(struct lock_stat *, struct lock_stat *);
281 
282 static struct rb_root		sorted; /* place to store intermediate data */
283 static struct rb_root		result;	/* place to store sorted data */
284 
285 static LIST_HEAD(lock_keys);
286 static const char		*output_fields;
287 
288 #define DEF_KEY_LOCK(name, header, fn_suffix, len)			\
289 	{ #name, header, len, lock_stat_key_ ## fn_suffix, lock_stat_key_print_ ## fn_suffix, {} }
290 struct lock_key keys[] = {
291 	DEF_KEY_LOCK(acquired, "acquired", nr_acquired, 10),
292 	DEF_KEY_LOCK(contended, "contended", nr_contended, 10),
293 	DEF_KEY_LOCK(avg_wait, "avg wait (ns)", avg_wait_time, 15),
294 	DEF_KEY_LOCK(wait_total, "total wait (ns)", wait_time_total, 15),
295 	DEF_KEY_LOCK(wait_max, "max wait (ns)", wait_time_max, 15),
296 	DEF_KEY_LOCK(wait_min, "min wait (ns)", wait_time_min, 15),
297 
298 	/* extra comparisons much complicated should be here */
299 	{ }
300 };
301 
302 static int select_key(void)
303 {
304 	int i;
305 
306 	for (i = 0; keys[i].name; i++) {
307 		if (!strcmp(keys[i].name, sort_key)) {
308 			compare = keys[i].key;
309 
310 			/* selected key should be in the output fields */
311 			if (list_empty(&keys[i].list))
312 				list_add_tail(&keys[i].list, &lock_keys);
313 
314 			return 0;
315 		}
316 	}
317 
318 	pr_err("Unknown compare key: %s\n", sort_key);
319 	return -1;
320 }
321 
322 static int add_output_field(struct list_head *head, char *name)
323 {
324 	int i;
325 
326 	for (i = 0; keys[i].name; i++) {
327 		if (strcmp(keys[i].name, name))
328 			continue;
329 
330 		/* prevent double link */
331 		if (list_empty(&keys[i].list))
332 			list_add_tail(&keys[i].list, head);
333 
334 		return 0;
335 	}
336 
337 	pr_err("Unknown output field: %s\n", name);
338 	return -1;
339 }
340 
341 static int setup_output_field(const char *str)
342 {
343 	char *tok, *tmp, *orig;
344 	int i, ret = 0;
345 
346 	/* no output field given: use all of them */
347 	if (str == NULL) {
348 		for (i = 0; keys[i].name; i++)
349 			list_add_tail(&keys[i].list, &lock_keys);
350 		return 0;
351 	}
352 
353 	for (i = 0; keys[i].name; i++)
354 		INIT_LIST_HEAD(&keys[i].list);
355 
356 	orig = tmp = strdup(str);
357 	if (orig == NULL)
358 		return -ENOMEM;
359 
360 	while ((tok = strsep(&tmp, ",")) != NULL){
361 		ret = add_output_field(&lock_keys, tok);
362 		if (ret < 0)
363 			break;
364 	}
365 	free(orig);
366 
367 	return ret;
368 }
369 
370 static void combine_lock_stats(struct lock_stat *st)
371 {
372 	struct rb_node **rb = &sorted.rb_node;
373 	struct rb_node *parent = NULL;
374 	struct lock_stat *p;
375 	int ret;
376 
377 	while (*rb) {
378 		p = container_of(*rb, struct lock_stat, rb);
379 		parent = *rb;
380 
381 		if (st->name && p->name)
382 			ret = strcmp(st->name, p->name);
383 		else
384 			ret = !!st->name - !!p->name;
385 
386 		if (ret == 0) {
387 			if (st->discard)
388 				goto out;
389 
390 			p->nr_acquired += st->nr_acquired;
391 			p->nr_contended += st->nr_contended;
392 			p->wait_time_total += st->wait_time_total;
393 
394 			if (p->nr_contended)
395 				p->avg_wait_time = p->wait_time_total / p->nr_contended;
396 
397 			if (p->wait_time_min > st->wait_time_min)
398 				p->wait_time_min = st->wait_time_min;
399 			if (p->wait_time_max < st->wait_time_max)
400 				p->wait_time_max = st->wait_time_max;
401 
402 			/* now it got a new !discard record */
403 			p->discard = 0;
404 
405 out:
406 			st->combined = 1;
407 			return;
408 		}
409 
410 		if (ret < 0)
411 			rb = &(*rb)->rb_left;
412 		else
413 			rb = &(*rb)->rb_right;
414 	}
415 
416 	rb_link_node(&st->rb, parent, rb);
417 	rb_insert_color(&st->rb, &sorted);
418 
419 	if (st->discard) {
420 		st->nr_acquired = 0;
421 		st->nr_contended = 0;
422 		st->wait_time_total = 0;
423 		st->avg_wait_time = 0;
424 		st->wait_time_min = ULLONG_MAX;
425 		st->wait_time_max = 0;
426 	}
427 }
428 
429 static void insert_to_result(struct lock_stat *st,
430 			     int (*bigger)(struct lock_stat *, struct lock_stat *))
431 {
432 	struct rb_node **rb = &result.rb_node;
433 	struct rb_node *parent = NULL;
434 	struct lock_stat *p;
435 
436 	if (combine_locks && st->combined)
437 		return;
438 
439 	while (*rb) {
440 		p = container_of(*rb, struct lock_stat, rb);
441 		parent = *rb;
442 
443 		if (bigger(st, p))
444 			rb = &(*rb)->rb_left;
445 		else
446 			rb = &(*rb)->rb_right;
447 	}
448 
449 	rb_link_node(&st->rb, parent, rb);
450 	rb_insert_color(&st->rb, &result);
451 }
452 
453 /* returns left most element of result, and erase it */
454 static struct lock_stat *pop_from_result(void)
455 {
456 	struct rb_node *node = result.rb_node;
457 
458 	if (!node)
459 		return NULL;
460 
461 	while (node->rb_left)
462 		node = node->rb_left;
463 
464 	rb_erase(node, &result);
465 	return container_of(node, struct lock_stat, rb);
466 }
467 
468 static struct lock_stat *lock_stat_findnew(u64 addr, const char *name)
469 {
470 	struct hlist_head *entry = lockhashentry(addr);
471 	struct lock_stat *ret, *new;
472 
473 	hlist_for_each_entry(ret, entry, hash_entry) {
474 		if (ret->addr == addr)
475 			return ret;
476 	}
477 
478 	new = zalloc(sizeof(struct lock_stat));
479 	if (!new)
480 		goto alloc_failed;
481 
482 	new->addr = addr;
483 	new->name = zalloc(sizeof(char) * strlen(name) + 1);
484 	if (!new->name) {
485 		free(new);
486 		goto alloc_failed;
487 	}
488 
489 	strcpy(new->name, name);
490 	new->wait_time_min = ULLONG_MAX;
491 
492 	hlist_add_head(&new->hash_entry, entry);
493 	return new;
494 
495 alloc_failed:
496 	pr_err("memory allocation failed\n");
497 	return NULL;
498 }
499 
500 struct trace_lock_handler {
501 	int (*acquire_event)(struct evsel *evsel,
502 			     struct perf_sample *sample);
503 
504 	int (*acquired_event)(struct evsel *evsel,
505 			      struct perf_sample *sample);
506 
507 	int (*contended_event)(struct evsel *evsel,
508 			       struct perf_sample *sample);
509 
510 	int (*release_event)(struct evsel *evsel,
511 			     struct perf_sample *sample);
512 };
513 
514 static struct lock_seq_stat *get_seq(struct thread_stat *ts, u64 addr)
515 {
516 	struct lock_seq_stat *seq;
517 
518 	list_for_each_entry(seq, &ts->seq_list, list) {
519 		if (seq->addr == addr)
520 			return seq;
521 	}
522 
523 	seq = zalloc(sizeof(struct lock_seq_stat));
524 	if (!seq) {
525 		pr_err("memory allocation failed\n");
526 		return NULL;
527 	}
528 	seq->state = SEQ_STATE_UNINITIALIZED;
529 	seq->addr = addr;
530 
531 	list_add(&seq->list, &ts->seq_list);
532 	return seq;
533 }
534 
535 enum broken_state {
536 	BROKEN_ACQUIRE,
537 	BROKEN_ACQUIRED,
538 	BROKEN_CONTENDED,
539 	BROKEN_RELEASE,
540 	BROKEN_MAX,
541 };
542 
543 static int bad_hist[BROKEN_MAX];
544 
545 enum acquire_flags {
546 	TRY_LOCK = 1,
547 	READ_LOCK = 2,
548 };
549 
550 static int report_lock_acquire_event(struct evsel *evsel,
551 				     struct perf_sample *sample)
552 {
553 	struct lock_stat *ls;
554 	struct thread_stat *ts;
555 	struct lock_seq_stat *seq;
556 	const char *name = evsel__strval(evsel, sample, "name");
557 	u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
558 	int flag = evsel__intval(evsel, sample, "flags");
559 
560 	ls = lock_stat_findnew(addr, name);
561 	if (!ls)
562 		return -ENOMEM;
563 	if (ls->discard)
564 		return 0;
565 
566 	ts = thread_stat_findnew(sample->tid);
567 	if (!ts)
568 		return -ENOMEM;
569 
570 	seq = get_seq(ts, addr);
571 	if (!seq)
572 		return -ENOMEM;
573 
574 	switch (seq->state) {
575 	case SEQ_STATE_UNINITIALIZED:
576 	case SEQ_STATE_RELEASED:
577 		if (!flag) {
578 			seq->state = SEQ_STATE_ACQUIRING;
579 		} else {
580 			if (flag & TRY_LOCK)
581 				ls->nr_trylock++;
582 			if (flag & READ_LOCK)
583 				ls->nr_readlock++;
584 			seq->state = SEQ_STATE_READ_ACQUIRED;
585 			seq->read_count = 1;
586 			ls->nr_acquired++;
587 		}
588 		break;
589 	case SEQ_STATE_READ_ACQUIRED:
590 		if (flag & READ_LOCK) {
591 			seq->read_count++;
592 			ls->nr_acquired++;
593 			goto end;
594 		} else {
595 			goto broken;
596 		}
597 		break;
598 	case SEQ_STATE_ACQUIRED:
599 	case SEQ_STATE_ACQUIRING:
600 	case SEQ_STATE_CONTENDED:
601 broken:
602 		/* broken lock sequence, discard it */
603 		ls->discard = 1;
604 		bad_hist[BROKEN_ACQUIRE]++;
605 		list_del_init(&seq->list);
606 		free(seq);
607 		goto end;
608 	default:
609 		BUG_ON("Unknown state of lock sequence found!\n");
610 		break;
611 	}
612 
613 	ls->nr_acquire++;
614 	seq->prev_event_time = sample->time;
615 end:
616 	return 0;
617 }
618 
619 static int report_lock_acquired_event(struct evsel *evsel,
620 				      struct perf_sample *sample)
621 {
622 	struct lock_stat *ls;
623 	struct thread_stat *ts;
624 	struct lock_seq_stat *seq;
625 	u64 contended_term;
626 	const char *name = evsel__strval(evsel, sample, "name");
627 	u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
628 
629 	ls = lock_stat_findnew(addr, name);
630 	if (!ls)
631 		return -ENOMEM;
632 	if (ls->discard)
633 		return 0;
634 
635 	ts = thread_stat_findnew(sample->tid);
636 	if (!ts)
637 		return -ENOMEM;
638 
639 	seq = get_seq(ts, addr);
640 	if (!seq)
641 		return -ENOMEM;
642 
643 	switch (seq->state) {
644 	case SEQ_STATE_UNINITIALIZED:
645 		/* orphan event, do nothing */
646 		return 0;
647 	case SEQ_STATE_ACQUIRING:
648 		break;
649 	case SEQ_STATE_CONTENDED:
650 		contended_term = sample->time - seq->prev_event_time;
651 		ls->wait_time_total += contended_term;
652 		if (contended_term < ls->wait_time_min)
653 			ls->wait_time_min = contended_term;
654 		if (ls->wait_time_max < contended_term)
655 			ls->wait_time_max = contended_term;
656 		break;
657 	case SEQ_STATE_RELEASED:
658 	case SEQ_STATE_ACQUIRED:
659 	case SEQ_STATE_READ_ACQUIRED:
660 		/* broken lock sequence, discard it */
661 		ls->discard = 1;
662 		bad_hist[BROKEN_ACQUIRED]++;
663 		list_del_init(&seq->list);
664 		free(seq);
665 		goto end;
666 	default:
667 		BUG_ON("Unknown state of lock sequence found!\n");
668 		break;
669 	}
670 
671 	seq->state = SEQ_STATE_ACQUIRED;
672 	ls->nr_acquired++;
673 	ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
674 	seq->prev_event_time = sample->time;
675 end:
676 	return 0;
677 }
678 
679 static int report_lock_contended_event(struct evsel *evsel,
680 				       struct perf_sample *sample)
681 {
682 	struct lock_stat *ls;
683 	struct thread_stat *ts;
684 	struct lock_seq_stat *seq;
685 	const char *name = evsel__strval(evsel, sample, "name");
686 	u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
687 
688 	ls = lock_stat_findnew(addr, name);
689 	if (!ls)
690 		return -ENOMEM;
691 	if (ls->discard)
692 		return 0;
693 
694 	ts = thread_stat_findnew(sample->tid);
695 	if (!ts)
696 		return -ENOMEM;
697 
698 	seq = get_seq(ts, addr);
699 	if (!seq)
700 		return -ENOMEM;
701 
702 	switch (seq->state) {
703 	case SEQ_STATE_UNINITIALIZED:
704 		/* orphan event, do nothing */
705 		return 0;
706 	case SEQ_STATE_ACQUIRING:
707 		break;
708 	case SEQ_STATE_RELEASED:
709 	case SEQ_STATE_ACQUIRED:
710 	case SEQ_STATE_READ_ACQUIRED:
711 	case SEQ_STATE_CONTENDED:
712 		/* broken lock sequence, discard it */
713 		ls->discard = 1;
714 		bad_hist[BROKEN_CONTENDED]++;
715 		list_del_init(&seq->list);
716 		free(seq);
717 		goto end;
718 	default:
719 		BUG_ON("Unknown state of lock sequence found!\n");
720 		break;
721 	}
722 
723 	seq->state = SEQ_STATE_CONTENDED;
724 	ls->nr_contended++;
725 	ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
726 	seq->prev_event_time = sample->time;
727 end:
728 	return 0;
729 }
730 
731 static int report_lock_release_event(struct evsel *evsel,
732 				     struct perf_sample *sample)
733 {
734 	struct lock_stat *ls;
735 	struct thread_stat *ts;
736 	struct lock_seq_stat *seq;
737 	const char *name = evsel__strval(evsel, sample, "name");
738 	u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
739 
740 	ls = lock_stat_findnew(addr, name);
741 	if (!ls)
742 		return -ENOMEM;
743 	if (ls->discard)
744 		return 0;
745 
746 	ts = thread_stat_findnew(sample->tid);
747 	if (!ts)
748 		return -ENOMEM;
749 
750 	seq = get_seq(ts, addr);
751 	if (!seq)
752 		return -ENOMEM;
753 
754 	switch (seq->state) {
755 	case SEQ_STATE_UNINITIALIZED:
756 		goto end;
757 	case SEQ_STATE_ACQUIRED:
758 		break;
759 	case SEQ_STATE_READ_ACQUIRED:
760 		seq->read_count--;
761 		BUG_ON(seq->read_count < 0);
762 		if (seq->read_count) {
763 			ls->nr_release++;
764 			goto end;
765 		}
766 		break;
767 	case SEQ_STATE_ACQUIRING:
768 	case SEQ_STATE_CONTENDED:
769 	case SEQ_STATE_RELEASED:
770 		/* broken lock sequence, discard it */
771 		ls->discard = 1;
772 		bad_hist[BROKEN_RELEASE]++;
773 		goto free_seq;
774 	default:
775 		BUG_ON("Unknown state of lock sequence found!\n");
776 		break;
777 	}
778 
779 	ls->nr_release++;
780 free_seq:
781 	list_del_init(&seq->list);
782 	free(seq);
783 end:
784 	return 0;
785 }
786 
787 /* lock oriented handlers */
788 /* TODO: handlers for CPU oriented, thread oriented */
789 static struct trace_lock_handler report_lock_ops  = {
790 	.acquire_event		= report_lock_acquire_event,
791 	.acquired_event		= report_lock_acquired_event,
792 	.contended_event	= report_lock_contended_event,
793 	.release_event		= report_lock_release_event,
794 };
795 
796 static struct trace_lock_handler *trace_handler;
797 
798 static int evsel__process_lock_acquire(struct evsel *evsel, struct perf_sample *sample)
799 {
800 	if (trace_handler->acquire_event)
801 		return trace_handler->acquire_event(evsel, sample);
802 	return 0;
803 }
804 
805 static int evsel__process_lock_acquired(struct evsel *evsel, struct perf_sample *sample)
806 {
807 	if (trace_handler->acquired_event)
808 		return trace_handler->acquired_event(evsel, sample);
809 	return 0;
810 }
811 
812 static int evsel__process_lock_contended(struct evsel *evsel, struct perf_sample *sample)
813 {
814 	if (trace_handler->contended_event)
815 		return trace_handler->contended_event(evsel, sample);
816 	return 0;
817 }
818 
819 static int evsel__process_lock_release(struct evsel *evsel, struct perf_sample *sample)
820 {
821 	if (trace_handler->release_event)
822 		return trace_handler->release_event(evsel, sample);
823 	return 0;
824 }
825 
826 static void print_bad_events(int bad, int total)
827 {
828 	/* Output for debug, this have to be removed */
829 	int i;
830 	const char *name[4] =
831 		{ "acquire", "acquired", "contended", "release" };
832 
833 	pr_info("\n=== output for debug===\n\n");
834 	pr_info("bad: %d, total: %d\n", bad, total);
835 	pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
836 	pr_info("histogram of events caused bad sequence\n");
837 	for (i = 0; i < BROKEN_MAX; i++)
838 		pr_info(" %10s: %d\n", name[i], bad_hist[i]);
839 }
840 
841 /* TODO: various way to print, coloring, nano or milli sec */
842 static void print_result(void)
843 {
844 	struct lock_stat *st;
845 	struct lock_key *key;
846 	char cut_name[20];
847 	int bad, total;
848 
849 	pr_info("%20s ", "Name");
850 	list_for_each_entry(key, &lock_keys, list)
851 		pr_info("%*s ", key->len, key->header);
852 	pr_info("\n\n");
853 
854 	bad = total = 0;
855 	while ((st = pop_from_result())) {
856 		total++;
857 		if (st->discard) {
858 			bad++;
859 			continue;
860 		}
861 		bzero(cut_name, 20);
862 
863 		if (strlen(st->name) < 20) {
864 			/* output raw name */
865 			pr_info("%20s ", st->name);
866 		} else {
867 			strncpy(cut_name, st->name, 16);
868 			cut_name[16] = '.';
869 			cut_name[17] = '.';
870 			cut_name[18] = '.';
871 			cut_name[19] = '\0';
872 			/* cut off name for saving output style */
873 			pr_info("%20s ", cut_name);
874 		}
875 
876 		list_for_each_entry(key, &lock_keys, list) {
877 			key->print(key, st);
878 			pr_info(" ");
879 		}
880 		pr_info("\n");
881 	}
882 
883 	print_bad_events(bad, total);
884 }
885 
886 static bool info_threads, info_map;
887 
888 static void dump_threads(void)
889 {
890 	struct thread_stat *st;
891 	struct rb_node *node;
892 	struct thread *t;
893 
894 	pr_info("%10s: comm\n", "Thread ID");
895 
896 	node = rb_first(&thread_stats);
897 	while (node) {
898 		st = container_of(node, struct thread_stat, rb);
899 		t = perf_session__findnew(session, st->tid);
900 		pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
901 		node = rb_next(node);
902 		thread__put(t);
903 	}
904 }
905 
906 static int compare_maps(struct lock_stat *a, struct lock_stat *b)
907 {
908 	int ret;
909 
910 	if (a->name && b->name)
911 		ret = strcmp(a->name, b->name);
912 	else
913 		ret = !!a->name - !!b->name;
914 
915 	if (!ret)
916 		return a->addr < b->addr;
917 	else
918 		return ret < 0;
919 }
920 
921 static void dump_map(void)
922 {
923 	unsigned int i;
924 	struct lock_stat *st;
925 
926 	pr_info("Address of instance: name of class\n");
927 	for (i = 0; i < LOCKHASH_SIZE; i++) {
928 		hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
929 			insert_to_result(st, compare_maps);
930 		}
931 	}
932 
933 	while ((st = pop_from_result()))
934 		pr_info(" %#llx: %s\n", (unsigned long long)st->addr, st->name);
935 }
936 
937 static int dump_info(void)
938 {
939 	int rc = 0;
940 
941 	if (info_threads)
942 		dump_threads();
943 	else if (info_map)
944 		dump_map();
945 	else {
946 		rc = -1;
947 		pr_err("Unknown type of information\n");
948 	}
949 
950 	return rc;
951 }
952 
953 typedef int (*tracepoint_handler)(struct evsel *evsel,
954 				  struct perf_sample *sample);
955 
956 static int process_sample_event(struct perf_tool *tool __maybe_unused,
957 				union perf_event *event,
958 				struct perf_sample *sample,
959 				struct evsel *evsel,
960 				struct machine *machine)
961 {
962 	int err = 0;
963 	struct thread *thread = machine__findnew_thread(machine, sample->pid,
964 							sample->tid);
965 
966 	if (thread == NULL) {
967 		pr_debug("problem processing %d event, skipping it.\n",
968 			event->header.type);
969 		return -1;
970 	}
971 
972 	if (evsel->handler != NULL) {
973 		tracepoint_handler f = evsel->handler;
974 		err = f(evsel, sample);
975 	}
976 
977 	thread__put(thread);
978 
979 	return err;
980 }
981 
982 static void combine_result(void)
983 {
984 	unsigned int i;
985 	struct lock_stat *st;
986 
987 	if (!combine_locks)
988 		return;
989 
990 	for (i = 0; i < LOCKHASH_SIZE; i++) {
991 		hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
992 			combine_lock_stats(st);
993 		}
994 	}
995 }
996 
997 static void sort_result(void)
998 {
999 	unsigned int i;
1000 	struct lock_stat *st;
1001 
1002 	for (i = 0; i < LOCKHASH_SIZE; i++) {
1003 		hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1004 			insert_to_result(st, compare);
1005 		}
1006 	}
1007 }
1008 
1009 static const struct evsel_str_handler lock_tracepoints[] = {
1010 	{ "lock:lock_acquire",	 evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
1011 	{ "lock:lock_acquired",	 evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1012 	{ "lock:lock_contended", evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1013 	{ "lock:lock_release",	 evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
1014 };
1015 
1016 static bool force;
1017 
1018 static int __cmd_report(bool display_info)
1019 {
1020 	int err = -EINVAL;
1021 	struct perf_tool eops = {
1022 		.sample		 = process_sample_event,
1023 		.comm		 = perf_event__process_comm,
1024 		.namespaces	 = perf_event__process_namespaces,
1025 		.ordered_events	 = true,
1026 	};
1027 	struct perf_data data = {
1028 		.path  = input_name,
1029 		.mode  = PERF_DATA_MODE_READ,
1030 		.force = force,
1031 	};
1032 
1033 	session = perf_session__new(&data, &eops);
1034 	if (IS_ERR(session)) {
1035 		pr_err("Initializing perf session failed\n");
1036 		return PTR_ERR(session);
1037 	}
1038 
1039 	symbol__init(&session->header.env);
1040 
1041 	if (!perf_session__has_traces(session, "lock record"))
1042 		goto out_delete;
1043 
1044 	if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
1045 		pr_err("Initializing perf session tracepoint handlers failed\n");
1046 		goto out_delete;
1047 	}
1048 
1049 	if (setup_output_field(output_fields))
1050 		goto out_delete;
1051 
1052 	if (select_key())
1053 		goto out_delete;
1054 
1055 	err = perf_session__process_events(session);
1056 	if (err)
1057 		goto out_delete;
1058 
1059 	setup_pager();
1060 	if (display_info) /* used for info subcommand */
1061 		err = dump_info();
1062 	else {
1063 		combine_result();
1064 		sort_result();
1065 		print_result();
1066 	}
1067 
1068 out_delete:
1069 	perf_session__delete(session);
1070 	return err;
1071 }
1072 
1073 static int __cmd_record(int argc, const char **argv)
1074 {
1075 	const char *record_args[] = {
1076 		"record", "-R", "-m", "1024", "-c", "1", "--synth", "no",
1077 	};
1078 	unsigned int rec_argc, i, j, ret;
1079 	const char **rec_argv;
1080 
1081 	for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
1082 		if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
1083 				pr_err("tracepoint %s is not enabled. "
1084 				       "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
1085 				       lock_tracepoints[i].name);
1086 				return 1;
1087 		}
1088 	}
1089 
1090 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1091 	/* factor of 2 is for -e in front of each tracepoint */
1092 	rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
1093 
1094 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1095 	if (!rec_argv)
1096 		return -ENOMEM;
1097 
1098 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
1099 		rec_argv[i] = strdup(record_args[i]);
1100 
1101 	for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
1102 		rec_argv[i++] = "-e";
1103 		rec_argv[i++] = strdup(lock_tracepoints[j].name);
1104 	}
1105 
1106 	for (j = 1; j < (unsigned int)argc; j++, i++)
1107 		rec_argv[i] = argv[j];
1108 
1109 	BUG_ON(i != rec_argc);
1110 
1111 	ret = cmd_record(i, rec_argv);
1112 	free(rec_argv);
1113 	return ret;
1114 }
1115 
1116 int cmd_lock(int argc, const char **argv)
1117 {
1118 	const struct option lock_options[] = {
1119 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1120 	OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
1121 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
1122 	OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
1123 	OPT_END()
1124 	};
1125 
1126 	const struct option info_options[] = {
1127 	OPT_BOOLEAN('t', "threads", &info_threads,
1128 		    "dump thread list in perf.data"),
1129 	OPT_BOOLEAN('m', "map", &info_map,
1130 		    "map of lock instances (address:name table)"),
1131 	OPT_PARENT(lock_options)
1132 	};
1133 
1134 	const struct option report_options[] = {
1135 	OPT_STRING('k', "key", &sort_key, "acquired",
1136 		    "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
1137 	OPT_STRING('F', "field", &output_fields, NULL,
1138 		    "output fields (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
1139 	/* TODO: type */
1140 	OPT_BOOLEAN('c', "combine-locks", &combine_locks,
1141 		    "combine locks in the same class"),
1142 	OPT_PARENT(lock_options)
1143 	};
1144 
1145 	const char * const info_usage[] = {
1146 		"perf lock info [<options>]",
1147 		NULL
1148 	};
1149 	const char *const lock_subcommands[] = { "record", "report", "script",
1150 						 "info", NULL };
1151 	const char *lock_usage[] = {
1152 		NULL,
1153 		NULL
1154 	};
1155 	const char * const report_usage[] = {
1156 		"perf lock report [<options>]",
1157 		NULL
1158 	};
1159 	unsigned int i;
1160 	int rc = 0;
1161 
1162 	for (i = 0; i < LOCKHASH_SIZE; i++)
1163 		INIT_HLIST_HEAD(lockhash_table + i);
1164 
1165 	argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
1166 					lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1167 	if (!argc)
1168 		usage_with_options(lock_usage, lock_options);
1169 
1170 	if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
1171 		return __cmd_record(argc, argv);
1172 	} else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
1173 		trace_handler = &report_lock_ops;
1174 		if (argc) {
1175 			argc = parse_options(argc, argv,
1176 					     report_options, report_usage, 0);
1177 			if (argc)
1178 				usage_with_options(report_usage, report_options);
1179 		}
1180 		rc = __cmd_report(false);
1181 	} else if (!strcmp(argv[0], "script")) {
1182 		/* Aliased to 'perf script' */
1183 		return cmd_script(argc, argv);
1184 	} else if (!strcmp(argv[0], "info")) {
1185 		if (argc) {
1186 			argc = parse_options(argc, argv,
1187 					     info_options, info_usage, 0);
1188 			if (argc)
1189 				usage_with_options(info_usage, info_options);
1190 		}
1191 		/* recycling report_lock_ops */
1192 		trace_handler = &report_lock_ops;
1193 		rc = __cmd_report(true);
1194 	} else {
1195 		usage_with_options(lock_usage, lock_options);
1196 	}
1197 
1198 	return rc;
1199 }
1200