xref: /linux/tools/perf/builtin-report.c (revision 6000fc4d6f3e55ad52cce8d76317187fe01af2aa)
1 /*
2  * builtin-report.c
3  *
4  * Builtin report command: Analyze the perf.data input file,
5  * look up and read DSOs and symbol information and display
6  * a histogram of results, along various sorting keys.
7  */
8 #include "builtin.h"
9 
10 #include "util/util.h"
11 
12 #include "util/color.h"
13 #include <linux/list.h>
14 #include "util/cache.h"
15 #include <linux/rbtree.h>
16 #include "util/symbol.h"
17 #include "util/string.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 
21 #include "perf.h"
22 #include "util/header.h"
23 
24 #include "util/parse-options.h"
25 #include "util/parse-events.h"
26 
27 #define SHOW_KERNEL	1
28 #define SHOW_USER	2
29 #define SHOW_HV		4
30 
31 static char		const *input_name = "perf.data";
32 static char		*vmlinux = NULL;
33 
34 static char		default_sort_order[] = "comm,dso,symbol";
35 static char		*sort_order = default_sort_order;
36 static char		*dso_list_str, *comm_list_str, *sym_list_str,
37 			*col_width_list_str;
38 static struct strlist	*dso_list, *comm_list, *sym_list;
39 static char		*field_sep;
40 
41 static int		input;
42 static int		show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
43 
44 static int		dump_trace = 0;
45 #define dprintf(x...)	do { if (dump_trace) printf(x); } while (0)
46 #define cdprintf(x...)	do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
47 
48 static int		verbose;
49 #define eprintf(x...)	do { if (verbose) fprintf(stderr, x); } while (0)
50 
51 static int		modules;
52 
53 static int		full_paths;
54 static int		show_nr_samples;
55 
56 static unsigned long	page_size;
57 static unsigned long	mmap_window = 32;
58 
59 static char		default_parent_pattern[] = "^sys_|^do_page_fault";
60 static char		*parent_pattern = default_parent_pattern;
61 static regex_t		parent_regex;
62 
63 static int		exclude_other = 1;
64 
65 static char		callchain_default_opt[] = "fractal,0.5";
66 
67 static int		callchain;
68 
69 static
70 struct callchain_param	callchain_param = {
71 	.mode	= CHAIN_GRAPH_REL,
72 	.min_percent = 0.5
73 };
74 
75 static u64		sample_type;
76 
77 struct ip_event {
78 	struct perf_event_header header;
79 	u64 ip;
80 	u32 pid, tid;
81 	unsigned char __more_data[];
82 };
83 
84 struct mmap_event {
85 	struct perf_event_header header;
86 	u32 pid, tid;
87 	u64 start;
88 	u64 len;
89 	u64 pgoff;
90 	char filename[PATH_MAX];
91 };
92 
93 struct comm_event {
94 	struct perf_event_header header;
95 	u32 pid, tid;
96 	char comm[16];
97 };
98 
99 struct fork_event {
100 	struct perf_event_header header;
101 	u32 pid, ppid;
102 	u32 tid, ptid;
103 };
104 
105 struct lost_event {
106 	struct perf_event_header header;
107 	u64 id;
108 	u64 lost;
109 };
110 
111 struct read_event {
112 	struct perf_event_header header;
113 	u32 pid,tid;
114 	u64 value;
115 	u64 time_enabled;
116 	u64 time_running;
117 	u64 id;
118 };
119 
120 typedef union event_union {
121 	struct perf_event_header	header;
122 	struct ip_event			ip;
123 	struct mmap_event		mmap;
124 	struct comm_event		comm;
125 	struct fork_event		fork;
126 	struct lost_event		lost;
127 	struct read_event		read;
128 } event_t;
129 
130 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
131 {
132 	int n;
133 	va_list ap;
134 
135 	va_start(ap, fmt);
136 	if (!field_sep)
137 		n = vfprintf(fp, fmt, ap);
138 	else {
139 		char *bf = NULL;
140 		n = vasprintf(&bf, fmt, ap);
141 		if (n > 0) {
142 			char *sep = bf;
143 			while (1) {
144 				sep = strchr(sep, *field_sep);
145 				if (sep == NULL)
146 					break;
147 				*sep = '.';
148 			}
149 		}
150 		fputs(bf, fp);
151 		free(bf);
152 	}
153 	va_end(ap);
154 	return n;
155 }
156 
157 static LIST_HEAD(dsos);
158 static struct dso *kernel_dso;
159 static struct dso *vdso;
160 static struct dso *hypervisor_dso;
161 
162 static void dsos__add(struct dso *dso)
163 {
164 	list_add_tail(&dso->node, &dsos);
165 }
166 
167 static struct dso *dsos__find(const char *name)
168 {
169 	struct dso *pos;
170 
171 	list_for_each_entry(pos, &dsos, node)
172 		if (strcmp(pos->name, name) == 0)
173 			return pos;
174 	return NULL;
175 }
176 
177 static struct dso *dsos__findnew(const char *name)
178 {
179 	struct dso *dso = dsos__find(name);
180 	int nr;
181 
182 	if (dso)
183 		return dso;
184 
185 	dso = dso__new(name, 0);
186 	if (!dso)
187 		goto out_delete_dso;
188 
189 	nr = dso__load(dso, NULL, verbose);
190 	if (nr < 0) {
191 		eprintf("Failed to open: %s\n", name);
192 		goto out_delete_dso;
193 	}
194 	if (!nr)
195 		eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
196 
197 	dsos__add(dso);
198 
199 	return dso;
200 
201 out_delete_dso:
202 	dso__delete(dso);
203 	return NULL;
204 }
205 
206 static void dsos__fprintf(FILE *fp)
207 {
208 	struct dso *pos;
209 
210 	list_for_each_entry(pos, &dsos, node)
211 		dso__fprintf(pos, fp);
212 }
213 
214 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
215 {
216 	return dso__find_symbol(dso, ip);
217 }
218 
219 static int load_kernel(void)
220 {
221 	int err;
222 
223 	kernel_dso = dso__new("[kernel]", 0);
224 	if (!kernel_dso)
225 		return -1;
226 
227 	err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose, modules);
228 	if (err <= 0) {
229 		dso__delete(kernel_dso);
230 		kernel_dso = NULL;
231 	} else
232 		dsos__add(kernel_dso);
233 
234 	vdso = dso__new("[vdso]", 0);
235 	if (!vdso)
236 		return -1;
237 
238 	vdso->find_symbol = vdso__find_symbol;
239 
240 	dsos__add(vdso);
241 
242 	hypervisor_dso = dso__new("[hypervisor]", 0);
243 	if (!hypervisor_dso)
244 		return -1;
245 	dsos__add(hypervisor_dso);
246 
247 	return err;
248 }
249 
250 static char __cwd[PATH_MAX];
251 static char *cwd = __cwd;
252 static int cwdlen;
253 
254 static int strcommon(const char *pathname)
255 {
256 	int n = 0;
257 
258 	while (n < cwdlen && pathname[n] == cwd[n])
259 		++n;
260 
261 	return n;
262 }
263 
264 struct map {
265 	struct list_head node;
266 	u64	 start;
267 	u64	 end;
268 	u64	 pgoff;
269 	u64	 (*map_ip)(struct map *, u64);
270 	struct dso	 *dso;
271 };
272 
273 static u64 map__map_ip(struct map *map, u64 ip)
274 {
275 	return ip - map->start + map->pgoff;
276 }
277 
278 static u64 vdso__map_ip(struct map *map __used, u64 ip)
279 {
280 	return ip;
281 }
282 
283 static inline int is_anon_memory(const char *filename)
284 {
285 	return strcmp(filename, "//anon") == 0;
286 }
287 
288 static struct map *map__new(struct mmap_event *event)
289 {
290 	struct map *self = malloc(sizeof(*self));
291 
292 	if (self != NULL) {
293 		const char *filename = event->filename;
294 		char newfilename[PATH_MAX];
295 		int anon;
296 
297 		if (cwd) {
298 			int n = strcommon(filename);
299 
300 			if (n == cwdlen) {
301 				snprintf(newfilename, sizeof(newfilename),
302 					 ".%s", filename + n);
303 				filename = newfilename;
304 			}
305 		}
306 
307 		anon = is_anon_memory(filename);
308 
309 		if (anon) {
310 			snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
311 			filename = newfilename;
312 		}
313 
314 		self->start = event->start;
315 		self->end   = event->start + event->len;
316 		self->pgoff = event->pgoff;
317 
318 		self->dso = dsos__findnew(filename);
319 		if (self->dso == NULL)
320 			goto out_delete;
321 
322 		if (self->dso == vdso || anon)
323 			self->map_ip = vdso__map_ip;
324 		else
325 			self->map_ip = map__map_ip;
326 	}
327 	return self;
328 out_delete:
329 	free(self);
330 	return NULL;
331 }
332 
333 static struct map *map__clone(struct map *self)
334 {
335 	struct map *map = malloc(sizeof(*self));
336 
337 	if (!map)
338 		return NULL;
339 
340 	memcpy(map, self, sizeof(*self));
341 
342 	return map;
343 }
344 
345 static int map__overlap(struct map *l, struct map *r)
346 {
347 	if (l->start > r->start) {
348 		struct map *t = l;
349 		l = r;
350 		r = t;
351 	}
352 
353 	if (l->end > r->start)
354 		return 1;
355 
356 	return 0;
357 }
358 
359 static size_t map__fprintf(struct map *self, FILE *fp)
360 {
361 	return fprintf(fp, " %Lx-%Lx %Lx %s\n",
362 		       self->start, self->end, self->pgoff, self->dso->name);
363 }
364 
365 
366 struct thread {
367 	struct rb_node	 rb_node;
368 	struct list_head maps;
369 	pid_t		 pid;
370 	char		 *comm;
371 };
372 
373 static struct thread *thread__new(pid_t pid)
374 {
375 	struct thread *self = malloc(sizeof(*self));
376 
377 	if (self != NULL) {
378 		self->pid = pid;
379 		self->comm = malloc(32);
380 		if (self->comm)
381 			snprintf(self->comm, 32, ":%d", self->pid);
382 		INIT_LIST_HEAD(&self->maps);
383 	}
384 
385 	return self;
386 }
387 
388 static unsigned int dsos__col_width,
389 		    comms__col_width,
390 		    threads__col_width;
391 
392 static int thread__set_comm(struct thread *self, const char *comm)
393 {
394 	if (self->comm)
395 		free(self->comm);
396 	self->comm = strdup(comm);
397 	if (!self->comm)
398 		return -ENOMEM;
399 
400 	if (!col_width_list_str && !field_sep &&
401 	    (!comm_list || strlist__has_entry(comm_list, comm))) {
402 		unsigned int slen = strlen(comm);
403 		if (slen > comms__col_width) {
404 			comms__col_width = slen;
405 			threads__col_width = slen + 6;
406 		}
407 	}
408 
409 	return 0;
410 }
411 
412 static size_t thread__fprintf(struct thread *self, FILE *fp)
413 {
414 	struct map *pos;
415 	size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
416 
417 	list_for_each_entry(pos, &self->maps, node)
418 		ret += map__fprintf(pos, fp);
419 
420 	return ret;
421 }
422 
423 
424 static struct rb_root threads;
425 static struct thread *last_match;
426 
427 static struct thread *threads__findnew(pid_t pid)
428 {
429 	struct rb_node **p = &threads.rb_node;
430 	struct rb_node *parent = NULL;
431 	struct thread *th;
432 
433 	/*
434 	 * Font-end cache - PID lookups come in blocks,
435 	 * so most of the time we dont have to look up
436 	 * the full rbtree:
437 	 */
438 	if (last_match && last_match->pid == pid)
439 		return last_match;
440 
441 	while (*p != NULL) {
442 		parent = *p;
443 		th = rb_entry(parent, struct thread, rb_node);
444 
445 		if (th->pid == pid) {
446 			last_match = th;
447 			return th;
448 		}
449 
450 		if (pid < th->pid)
451 			p = &(*p)->rb_left;
452 		else
453 			p = &(*p)->rb_right;
454 	}
455 
456 	th = thread__new(pid);
457 	if (th != NULL) {
458 		rb_link_node(&th->rb_node, parent, p);
459 		rb_insert_color(&th->rb_node, &threads);
460 		last_match = th;
461 	}
462 
463 	return th;
464 }
465 
466 static void thread__insert_map(struct thread *self, struct map *map)
467 {
468 	struct map *pos, *tmp;
469 
470 	list_for_each_entry_safe(pos, tmp, &self->maps, node) {
471 		if (map__overlap(pos, map)) {
472 			if (verbose >= 2) {
473 				printf("overlapping maps:\n");
474 				map__fprintf(map, stdout);
475 				map__fprintf(pos, stdout);
476 			}
477 
478 			if (map->start <= pos->start && map->end > pos->start)
479 				pos->start = map->end;
480 
481 			if (map->end >= pos->end && map->start < pos->end)
482 				pos->end = map->start;
483 
484 			if (verbose >= 2) {
485 				printf("after collision:\n");
486 				map__fprintf(pos, stdout);
487 			}
488 
489 			if (pos->start >= pos->end) {
490 				list_del_init(&pos->node);
491 				free(pos);
492 			}
493 		}
494 	}
495 
496 	list_add_tail(&map->node, &self->maps);
497 }
498 
499 static int thread__fork(struct thread *self, struct thread *parent)
500 {
501 	struct map *map;
502 
503 	if (self->comm)
504 		free(self->comm);
505 	self->comm = strdup(parent->comm);
506 	if (!self->comm)
507 		return -ENOMEM;
508 
509 	list_for_each_entry(map, &parent->maps, node) {
510 		struct map *new = map__clone(map);
511 		if (!new)
512 			return -ENOMEM;
513 		thread__insert_map(self, new);
514 	}
515 
516 	return 0;
517 }
518 
519 static struct map *thread__find_map(struct thread *self, u64 ip)
520 {
521 	struct map *pos;
522 
523 	if (self == NULL)
524 		return NULL;
525 
526 	list_for_each_entry(pos, &self->maps, node)
527 		if (ip >= pos->start && ip <= pos->end)
528 			return pos;
529 
530 	return NULL;
531 }
532 
533 static size_t threads__fprintf(FILE *fp)
534 {
535 	size_t ret = 0;
536 	struct rb_node *nd;
537 
538 	for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
539 		struct thread *pos = rb_entry(nd, struct thread, rb_node);
540 
541 		ret += thread__fprintf(pos, fp);
542 	}
543 
544 	return ret;
545 }
546 
547 /*
548  * histogram, sorted on item, collects counts
549  */
550 
551 static struct rb_root hist;
552 
553 struct hist_entry {
554 	struct rb_node		rb_node;
555 
556 	struct thread		*thread;
557 	struct map		*map;
558 	struct dso		*dso;
559 	struct symbol		*sym;
560 	struct symbol		*parent;
561 	u64			ip;
562 	char			level;
563 	struct callchain_node	callchain;
564 	struct rb_root		sorted_chain;
565 
566 	u64			count;
567 };
568 
569 /*
570  * configurable sorting bits
571  */
572 
573 struct sort_entry {
574 	struct list_head list;
575 
576 	char *header;
577 
578 	int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
579 	int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
580 	size_t	(*print)(FILE *fp, struct hist_entry *, unsigned int width);
581 	unsigned int *width;
582 	bool	elide;
583 };
584 
585 static int64_t cmp_null(void *l, void *r)
586 {
587 	if (!l && !r)
588 		return 0;
589 	else if (!l)
590 		return -1;
591 	else
592 		return 1;
593 }
594 
595 /* --sort pid */
596 
597 static int64_t
598 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
599 {
600 	return right->thread->pid - left->thread->pid;
601 }
602 
603 static size_t
604 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
605 {
606 	return repsep_fprintf(fp, "%*s:%5d", width - 6,
607 			      self->thread->comm ?: "", self->thread->pid);
608 }
609 
610 static struct sort_entry sort_thread = {
611 	.header = "Command:  Pid",
612 	.cmp	= sort__thread_cmp,
613 	.print	= sort__thread_print,
614 	.width	= &threads__col_width,
615 };
616 
617 /* --sort comm */
618 
619 static int64_t
620 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
621 {
622 	return right->thread->pid - left->thread->pid;
623 }
624 
625 static int64_t
626 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
627 {
628 	char *comm_l = left->thread->comm;
629 	char *comm_r = right->thread->comm;
630 
631 	if (!comm_l || !comm_r)
632 		return cmp_null(comm_l, comm_r);
633 
634 	return strcmp(comm_l, comm_r);
635 }
636 
637 static size_t
638 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
639 {
640 	return repsep_fprintf(fp, "%*s", width, self->thread->comm);
641 }
642 
643 static struct sort_entry sort_comm = {
644 	.header		= "Command",
645 	.cmp		= sort__comm_cmp,
646 	.collapse	= sort__comm_collapse,
647 	.print		= sort__comm_print,
648 	.width		= &comms__col_width,
649 };
650 
651 /* --sort dso */
652 
653 static int64_t
654 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
655 {
656 	struct dso *dso_l = left->dso;
657 	struct dso *dso_r = right->dso;
658 
659 	if (!dso_l || !dso_r)
660 		return cmp_null(dso_l, dso_r);
661 
662 	return strcmp(dso_l->name, dso_r->name);
663 }
664 
665 static size_t
666 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
667 {
668 	if (self->dso)
669 		return repsep_fprintf(fp, "%-*s", width, self->dso->name);
670 
671 	return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
672 }
673 
674 static struct sort_entry sort_dso = {
675 	.header = "Shared Object",
676 	.cmp	= sort__dso_cmp,
677 	.print	= sort__dso_print,
678 	.width	= &dsos__col_width,
679 };
680 
681 /* --sort symbol */
682 
683 static int64_t
684 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
685 {
686 	u64 ip_l, ip_r;
687 
688 	if (left->sym == right->sym)
689 		return 0;
690 
691 	ip_l = left->sym ? left->sym->start : left->ip;
692 	ip_r = right->sym ? right->sym->start : right->ip;
693 
694 	return (int64_t)(ip_r - ip_l);
695 }
696 
697 static size_t
698 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
699 {
700 	size_t ret = 0;
701 
702 	if (verbose)
703 		ret += repsep_fprintf(fp, "%#018llx %c ", (u64)self->ip,
704 				      dso__symtab_origin(self->dso));
705 
706 	ret += repsep_fprintf(fp, "[%c] ", self->level);
707 	if (self->sym) {
708 		ret += repsep_fprintf(fp, "%s", self->sym->name);
709 
710 		if (self->sym->module)
711 			ret += repsep_fprintf(fp, "\t[%s]",
712 					     self->sym->module->name);
713 	} else {
714 		ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
715 	}
716 
717 	return ret;
718 }
719 
720 static struct sort_entry sort_sym = {
721 	.header = "Symbol",
722 	.cmp	= sort__sym_cmp,
723 	.print	= sort__sym_print,
724 };
725 
726 /* --sort parent */
727 
728 static int64_t
729 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
730 {
731 	struct symbol *sym_l = left->parent;
732 	struct symbol *sym_r = right->parent;
733 
734 	if (!sym_l || !sym_r)
735 		return cmp_null(sym_l, sym_r);
736 
737 	return strcmp(sym_l->name, sym_r->name);
738 }
739 
740 static size_t
741 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
742 {
743 	return repsep_fprintf(fp, "%-*s", width,
744 			      self->parent ? self->parent->name : "[other]");
745 }
746 
747 static unsigned int parent_symbol__col_width;
748 
749 static struct sort_entry sort_parent = {
750 	.header = "Parent symbol",
751 	.cmp	= sort__parent_cmp,
752 	.print	= sort__parent_print,
753 	.width	= &parent_symbol__col_width,
754 };
755 
756 static int sort__need_collapse = 0;
757 static int sort__has_parent = 0;
758 
759 struct sort_dimension {
760 	char			*name;
761 	struct sort_entry	*entry;
762 	int			taken;
763 };
764 
765 static struct sort_dimension sort_dimensions[] = {
766 	{ .name = "pid",	.entry = &sort_thread,	},
767 	{ .name = "comm",	.entry = &sort_comm,	},
768 	{ .name = "dso",	.entry = &sort_dso,	},
769 	{ .name = "symbol",	.entry = &sort_sym,	},
770 	{ .name = "parent",	.entry = &sort_parent,	},
771 };
772 
773 static LIST_HEAD(hist_entry__sort_list);
774 
775 static int sort_dimension__add(char *tok)
776 {
777 	unsigned int i;
778 
779 	for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
780 		struct sort_dimension *sd = &sort_dimensions[i];
781 
782 		if (sd->taken)
783 			continue;
784 
785 		if (strncasecmp(tok, sd->name, strlen(tok)))
786 			continue;
787 
788 		if (sd->entry->collapse)
789 			sort__need_collapse = 1;
790 
791 		if (sd->entry == &sort_parent) {
792 			int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
793 			if (ret) {
794 				char err[BUFSIZ];
795 
796 				regerror(ret, &parent_regex, err, sizeof(err));
797 				fprintf(stderr, "Invalid regex: %s\n%s",
798 					parent_pattern, err);
799 				exit(-1);
800 			}
801 			sort__has_parent = 1;
802 		}
803 
804 		list_add_tail(&sd->entry->list, &hist_entry__sort_list);
805 		sd->taken = 1;
806 
807 		return 0;
808 	}
809 
810 	return -ESRCH;
811 }
812 
813 static int64_t
814 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
815 {
816 	struct sort_entry *se;
817 	int64_t cmp = 0;
818 
819 	list_for_each_entry(se, &hist_entry__sort_list, list) {
820 		cmp = se->cmp(left, right);
821 		if (cmp)
822 			break;
823 	}
824 
825 	return cmp;
826 }
827 
828 static int64_t
829 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
830 {
831 	struct sort_entry *se;
832 	int64_t cmp = 0;
833 
834 	list_for_each_entry(se, &hist_entry__sort_list, list) {
835 		int64_t (*f)(struct hist_entry *, struct hist_entry *);
836 
837 		f = se->collapse ?: se->cmp;
838 
839 		cmp = f(left, right);
840 		if (cmp)
841 			break;
842 	}
843 
844 	return cmp;
845 }
846 
847 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
848 {
849 	int i;
850 	size_t ret = 0;
851 
852 	ret += fprintf(fp, "%s", "                ");
853 
854 	for (i = 0; i < depth; i++)
855 		if (depth_mask & (1 << i))
856 			ret += fprintf(fp, "|          ");
857 		else
858 			ret += fprintf(fp, "           ");
859 
860 	ret += fprintf(fp, "\n");
861 
862 	return ret;
863 }
864 static size_t
865 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
866 		       int depth_mask, int count, u64 total_samples,
867 		       int hits)
868 {
869 	int i;
870 	size_t ret = 0;
871 
872 	ret += fprintf(fp, "%s", "                ");
873 	for (i = 0; i < depth; i++) {
874 		if (depth_mask & (1 << i))
875 			ret += fprintf(fp, "|");
876 		else
877 			ret += fprintf(fp, " ");
878 		if (!count && i == depth - 1) {
879 			double percent;
880 
881 			percent = hits * 100.0 / total_samples;
882 			ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
883 		} else
884 			ret += fprintf(fp, "%s", "          ");
885 	}
886 	if (chain->sym)
887 		ret += fprintf(fp, "%s\n", chain->sym->name);
888 	else
889 		ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
890 
891 	return ret;
892 }
893 
894 static struct symbol *rem_sq_bracket;
895 static struct callchain_list rem_hits;
896 
897 static void init_rem_hits(void)
898 {
899 	rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
900 	if (!rem_sq_bracket) {
901 		fprintf(stderr, "Not enough memory to display remaining hits\n");
902 		return;
903 	}
904 
905 	strcpy(rem_sq_bracket->name, "[...]");
906 	rem_hits.sym = rem_sq_bracket;
907 }
908 
909 static size_t
910 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
911 			u64 total_samples, int depth, int depth_mask)
912 {
913 	struct rb_node *node, *next;
914 	struct callchain_node *child;
915 	struct callchain_list *chain;
916 	int new_depth_mask = depth_mask;
917 	u64 new_total;
918 	u64 remaining;
919 	size_t ret = 0;
920 	int i;
921 
922 	if (callchain_param.mode == CHAIN_GRAPH_REL)
923 		new_total = self->children_hit;
924 	else
925 		new_total = total_samples;
926 
927 	remaining = new_total;
928 
929 	node = rb_first(&self->rb_root);
930 	while (node) {
931 		u64 cumul;
932 
933 		child = rb_entry(node, struct callchain_node, rb_node);
934 		cumul = cumul_hits(child);
935 		remaining -= cumul;
936 
937 		/*
938 		 * The depth mask manages the output of pipes that show
939 		 * the depth. We don't want to keep the pipes of the current
940 		 * level for the last child of this depth.
941 		 * Except if we have remaining filtered hits. They will
942 		 * supersede the last child
943 		 */
944 		next = rb_next(node);
945 		if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
946 			new_depth_mask &= ~(1 << (depth - 1));
947 
948 		/*
949 		 * But we keep the older depth mask for the line seperator
950 		 * to keep the level link until we reach the last child
951 		 */
952 		ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
953 		i = 0;
954 		list_for_each_entry(chain, &child->val, list) {
955 			if (chain->ip >= PERF_CONTEXT_MAX)
956 				continue;
957 			ret += ipchain__fprintf_graph(fp, chain, depth,
958 						      new_depth_mask, i++,
959 						      new_total,
960 						      cumul);
961 		}
962 		ret += callchain__fprintf_graph(fp, child, new_total,
963 						depth + 1,
964 						new_depth_mask | (1 << depth));
965 		node = next;
966 	}
967 
968 	if (callchain_param.mode == CHAIN_GRAPH_REL &&
969 		remaining && remaining != new_total) {
970 
971 		if (!rem_sq_bracket)
972 			return ret;
973 
974 		new_depth_mask &= ~(1 << (depth - 1));
975 
976 		ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
977 					      new_depth_mask, 0, new_total,
978 					      remaining);
979 	}
980 
981 	return ret;
982 }
983 
984 static size_t
985 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
986 			u64 total_samples)
987 {
988 	struct callchain_list *chain;
989 	size_t ret = 0;
990 
991 	if (!self)
992 		return 0;
993 
994 	ret += callchain__fprintf_flat(fp, self->parent, total_samples);
995 
996 
997 	list_for_each_entry(chain, &self->val, list) {
998 		if (chain->ip >= PERF_CONTEXT_MAX)
999 			continue;
1000 		if (chain->sym)
1001 			ret += fprintf(fp, "                %s\n", chain->sym->name);
1002 		else
1003 			ret += fprintf(fp, "                %p\n",
1004 					(void *)(long)chain->ip);
1005 	}
1006 
1007 	return ret;
1008 }
1009 
1010 static size_t
1011 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
1012 			      u64 total_samples)
1013 {
1014 	struct rb_node *rb_node;
1015 	struct callchain_node *chain;
1016 	size_t ret = 0;
1017 
1018 	rb_node = rb_first(&self->sorted_chain);
1019 	while (rb_node) {
1020 		double percent;
1021 
1022 		chain = rb_entry(rb_node, struct callchain_node, rb_node);
1023 		percent = chain->hit * 100.0 / total_samples;
1024 		switch (callchain_param.mode) {
1025 		case CHAIN_FLAT:
1026 			ret += percent_color_fprintf(fp, "           %6.2f%%\n",
1027 						     percent);
1028 			ret += callchain__fprintf_flat(fp, chain, total_samples);
1029 			break;
1030 		case CHAIN_GRAPH_ABS: /* Falldown */
1031 		case CHAIN_GRAPH_REL:
1032 			ret += callchain__fprintf_graph(fp, chain,
1033 							total_samples, 1, 1);
1034 		default:
1035 			break;
1036 		}
1037 		ret += fprintf(fp, "\n");
1038 		rb_node = rb_next(rb_node);
1039 	}
1040 
1041 	return ret;
1042 }
1043 
1044 
1045 static size_t
1046 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
1047 {
1048 	struct sort_entry *se;
1049 	size_t ret;
1050 
1051 	if (exclude_other && !self->parent)
1052 		return 0;
1053 
1054 	if (total_samples)
1055 		ret = percent_color_fprintf(fp,
1056 					    field_sep ? "%.2f" : "   %6.2f%%",
1057 					(self->count * 100.0) / total_samples);
1058 	else
1059 		ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
1060 
1061 	if (show_nr_samples) {
1062 		if (field_sep)
1063 			fprintf(fp, "%c%lld", *field_sep, self->count);
1064 		else
1065 			fprintf(fp, "%11lld", self->count);
1066 	}
1067 
1068 	list_for_each_entry(se, &hist_entry__sort_list, list) {
1069 		if (se->elide)
1070 			continue;
1071 
1072 		fprintf(fp, "%s", field_sep ?: "  ");
1073 		ret += se->print(fp, self, se->width ? *se->width : 0);
1074 	}
1075 
1076 	ret += fprintf(fp, "\n");
1077 
1078 	if (callchain)
1079 		hist_entry_callchain__fprintf(fp, self, total_samples);
1080 
1081 	return ret;
1082 }
1083 
1084 /*
1085  *
1086  */
1087 
1088 static void dso__calc_col_width(struct dso *self)
1089 {
1090 	if (!col_width_list_str && !field_sep &&
1091 	    (!dso_list || strlist__has_entry(dso_list, self->name))) {
1092 		unsigned int slen = strlen(self->name);
1093 		if (slen > dsos__col_width)
1094 			dsos__col_width = slen;
1095 	}
1096 
1097 	self->slen_calculated = 1;
1098 }
1099 
1100 static struct symbol *
1101 resolve_symbol(struct thread *thread, struct map **mapp,
1102 	       struct dso **dsop, u64 *ipp)
1103 {
1104 	struct dso *dso = dsop ? *dsop : NULL;
1105 	struct map *map = mapp ? *mapp : NULL;
1106 	u64 ip = *ipp;
1107 
1108 	if (!thread)
1109 		return NULL;
1110 
1111 	if (dso)
1112 		goto got_dso;
1113 
1114 	if (map)
1115 		goto got_map;
1116 
1117 	map = thread__find_map(thread, ip);
1118 	if (map != NULL) {
1119 		/*
1120 		 * We have to do this here as we may have a dso
1121 		 * with no symbol hit that has a name longer than
1122 		 * the ones with symbols sampled.
1123 		 */
1124 		if (!sort_dso.elide && !map->dso->slen_calculated)
1125 			dso__calc_col_width(map->dso);
1126 
1127 		if (mapp)
1128 			*mapp = map;
1129 got_map:
1130 		ip = map->map_ip(map, ip);
1131 
1132 		dso = map->dso;
1133 	} else {
1134 		/*
1135 		 * If this is outside of all known maps,
1136 		 * and is a negative address, try to look it
1137 		 * up in the kernel dso, as it might be a
1138 		 * vsyscall (which executes in user-mode):
1139 		 */
1140 		if ((long long)ip < 0)
1141 		dso = kernel_dso;
1142 	}
1143 	dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
1144 	dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
1145 	*ipp  = ip;
1146 
1147 	if (dsop)
1148 		*dsop = dso;
1149 
1150 	if (!dso)
1151 		return NULL;
1152 got_dso:
1153 	return dso->find_symbol(dso, ip);
1154 }
1155 
1156 static int call__match(struct symbol *sym)
1157 {
1158 	if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
1159 		return 1;
1160 
1161 	return 0;
1162 }
1163 
1164 static struct symbol **
1165 resolve_callchain(struct thread *thread, struct map *map __used,
1166 		    struct ip_callchain *chain, struct hist_entry *entry)
1167 {
1168 	u64 context = PERF_CONTEXT_MAX;
1169 	struct symbol **syms = NULL;
1170 	unsigned int i;
1171 
1172 	if (callchain) {
1173 		syms = calloc(chain->nr, sizeof(*syms));
1174 		if (!syms) {
1175 			fprintf(stderr, "Can't allocate memory for symbols\n");
1176 			exit(-1);
1177 		}
1178 	}
1179 
1180 	for (i = 0; i < chain->nr; i++) {
1181 		u64 ip = chain->ips[i];
1182 		struct dso *dso = NULL;
1183 		struct symbol *sym;
1184 
1185 		if (ip >= PERF_CONTEXT_MAX) {
1186 			context = ip;
1187 			continue;
1188 		}
1189 
1190 		switch (context) {
1191 		case PERF_CONTEXT_HV:
1192 			dso = hypervisor_dso;
1193 			break;
1194 		case PERF_CONTEXT_KERNEL:
1195 			dso = kernel_dso;
1196 			break;
1197 		default:
1198 			break;
1199 		}
1200 
1201 		sym = resolve_symbol(thread, NULL, &dso, &ip);
1202 
1203 		if (sym) {
1204 			if (sort__has_parent && call__match(sym) &&
1205 			    !entry->parent)
1206 				entry->parent = sym;
1207 			if (!callchain)
1208 				break;
1209 			syms[i] = sym;
1210 		}
1211 	}
1212 
1213 	return syms;
1214 }
1215 
1216 /*
1217  * collect histogram counts
1218  */
1219 
1220 static int
1221 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
1222 		struct symbol *sym, u64 ip, struct ip_callchain *chain,
1223 		char level, u64 count)
1224 {
1225 	struct rb_node **p = &hist.rb_node;
1226 	struct rb_node *parent = NULL;
1227 	struct hist_entry *he;
1228 	struct symbol **syms = NULL;
1229 	struct hist_entry entry = {
1230 		.thread	= thread,
1231 		.map	= map,
1232 		.dso	= dso,
1233 		.sym	= sym,
1234 		.ip	= ip,
1235 		.level	= level,
1236 		.count	= count,
1237 		.parent = NULL,
1238 		.sorted_chain = RB_ROOT
1239 	};
1240 	int cmp;
1241 
1242 	if ((sort__has_parent || callchain) && chain)
1243 		syms = resolve_callchain(thread, map, chain, &entry);
1244 
1245 	while (*p != NULL) {
1246 		parent = *p;
1247 		he = rb_entry(parent, struct hist_entry, rb_node);
1248 
1249 		cmp = hist_entry__cmp(&entry, he);
1250 
1251 		if (!cmp) {
1252 			he->count += count;
1253 			if (callchain) {
1254 				append_chain(&he->callchain, chain, syms);
1255 				free(syms);
1256 			}
1257 			return 0;
1258 		}
1259 
1260 		if (cmp < 0)
1261 			p = &(*p)->rb_left;
1262 		else
1263 			p = &(*p)->rb_right;
1264 	}
1265 
1266 	he = malloc(sizeof(*he));
1267 	if (!he)
1268 		return -ENOMEM;
1269 	*he = entry;
1270 	if (callchain) {
1271 		callchain_init(&he->callchain);
1272 		append_chain(&he->callchain, chain, syms);
1273 		free(syms);
1274 	}
1275 	rb_link_node(&he->rb_node, parent, p);
1276 	rb_insert_color(&he->rb_node, &hist);
1277 
1278 	return 0;
1279 }
1280 
1281 static void hist_entry__free(struct hist_entry *he)
1282 {
1283 	free(he);
1284 }
1285 
1286 /*
1287  * collapse the histogram
1288  */
1289 
1290 static struct rb_root collapse_hists;
1291 
1292 static void collapse__insert_entry(struct hist_entry *he)
1293 {
1294 	struct rb_node **p = &collapse_hists.rb_node;
1295 	struct rb_node *parent = NULL;
1296 	struct hist_entry *iter;
1297 	int64_t cmp;
1298 
1299 	while (*p != NULL) {
1300 		parent = *p;
1301 		iter = rb_entry(parent, struct hist_entry, rb_node);
1302 
1303 		cmp = hist_entry__collapse(iter, he);
1304 
1305 		if (!cmp) {
1306 			iter->count += he->count;
1307 			hist_entry__free(he);
1308 			return;
1309 		}
1310 
1311 		if (cmp < 0)
1312 			p = &(*p)->rb_left;
1313 		else
1314 			p = &(*p)->rb_right;
1315 	}
1316 
1317 	rb_link_node(&he->rb_node, parent, p);
1318 	rb_insert_color(&he->rb_node, &collapse_hists);
1319 }
1320 
1321 static void collapse__resort(void)
1322 {
1323 	struct rb_node *next;
1324 	struct hist_entry *n;
1325 
1326 	if (!sort__need_collapse)
1327 		return;
1328 
1329 	next = rb_first(&hist);
1330 	while (next) {
1331 		n = rb_entry(next, struct hist_entry, rb_node);
1332 		next = rb_next(&n->rb_node);
1333 
1334 		rb_erase(&n->rb_node, &hist);
1335 		collapse__insert_entry(n);
1336 	}
1337 }
1338 
1339 /*
1340  * reverse the map, sort on count.
1341  */
1342 
1343 static struct rb_root output_hists;
1344 
1345 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
1346 {
1347 	struct rb_node **p = &output_hists.rb_node;
1348 	struct rb_node *parent = NULL;
1349 	struct hist_entry *iter;
1350 
1351 	if (callchain)
1352 		callchain_param.sort(&he->sorted_chain, &he->callchain,
1353 				      min_callchain_hits, &callchain_param);
1354 
1355 	while (*p != NULL) {
1356 		parent = *p;
1357 		iter = rb_entry(parent, struct hist_entry, rb_node);
1358 
1359 		if (he->count > iter->count)
1360 			p = &(*p)->rb_left;
1361 		else
1362 			p = &(*p)->rb_right;
1363 	}
1364 
1365 	rb_link_node(&he->rb_node, parent, p);
1366 	rb_insert_color(&he->rb_node, &output_hists);
1367 }
1368 
1369 static void output__resort(u64 total_samples)
1370 {
1371 	struct rb_node *next;
1372 	struct hist_entry *n;
1373 	struct rb_root *tree = &hist;
1374 	u64 min_callchain_hits;
1375 
1376 	min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
1377 
1378 	if (sort__need_collapse)
1379 		tree = &collapse_hists;
1380 
1381 	next = rb_first(tree);
1382 
1383 	while (next) {
1384 		n = rb_entry(next, struct hist_entry, rb_node);
1385 		next = rb_next(&n->rb_node);
1386 
1387 		rb_erase(&n->rb_node, tree);
1388 		output__insert_entry(n, min_callchain_hits);
1389 	}
1390 }
1391 
1392 static size_t output__fprintf(FILE *fp, u64 total_samples)
1393 {
1394 	struct hist_entry *pos;
1395 	struct sort_entry *se;
1396 	struct rb_node *nd;
1397 	size_t ret = 0;
1398 	unsigned int width;
1399 	char *col_width = col_width_list_str;
1400 
1401 	init_rem_hits();
1402 
1403 	fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
1404 	fprintf(fp, "#\n");
1405 
1406 	fprintf(fp, "# Overhead");
1407 	if (show_nr_samples) {
1408 		if (field_sep)
1409 			fprintf(fp, "%cSamples", *field_sep);
1410 		else
1411 			fputs("  Samples  ", fp);
1412 	}
1413 	list_for_each_entry(se, &hist_entry__sort_list, list) {
1414 		if (se->elide)
1415 			continue;
1416 		if (field_sep) {
1417 			fprintf(fp, "%c%s", *field_sep, se->header);
1418 			continue;
1419 		}
1420 		width = strlen(se->header);
1421 		if (se->width) {
1422 			if (col_width_list_str) {
1423 				if (col_width) {
1424 					*se->width = atoi(col_width);
1425 					col_width = strchr(col_width, ',');
1426 					if (col_width)
1427 						++col_width;
1428 				}
1429 			}
1430 			width = *se->width = max(*se->width, width);
1431 		}
1432 		fprintf(fp, "  %*s", width, se->header);
1433 	}
1434 	fprintf(fp, "\n");
1435 
1436 	if (field_sep)
1437 		goto print_entries;
1438 
1439 	fprintf(fp, "# ........");
1440 	if (show_nr_samples)
1441 		fprintf(fp, " ..........");
1442 	list_for_each_entry(se, &hist_entry__sort_list, list) {
1443 		unsigned int i;
1444 
1445 		if (se->elide)
1446 			continue;
1447 
1448 		fprintf(fp, "  ");
1449 		if (se->width)
1450 			width = *se->width;
1451 		else
1452 			width = strlen(se->header);
1453 		for (i = 0; i < width; i++)
1454 			fprintf(fp, ".");
1455 	}
1456 	fprintf(fp, "\n");
1457 
1458 	fprintf(fp, "#\n");
1459 
1460 print_entries:
1461 	for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1462 		pos = rb_entry(nd, struct hist_entry, rb_node);
1463 		ret += hist_entry__fprintf(fp, pos, total_samples);
1464 	}
1465 
1466 	if (sort_order == default_sort_order &&
1467 			parent_pattern == default_parent_pattern) {
1468 		fprintf(fp, "#\n");
1469 		fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1470 		fprintf(fp, "#\n");
1471 	}
1472 	fprintf(fp, "\n");
1473 
1474 	free(rem_sq_bracket);
1475 
1476 	return ret;
1477 }
1478 
1479 static void register_idle_thread(void)
1480 {
1481 	struct thread *thread = threads__findnew(0);
1482 
1483 	if (thread == NULL ||
1484 			thread__set_comm(thread, "[idle]")) {
1485 		fprintf(stderr, "problem inserting idle task.\n");
1486 		exit(-1);
1487 	}
1488 }
1489 
1490 static unsigned long total = 0,
1491 		     total_mmap = 0,
1492 		     total_comm = 0,
1493 		     total_fork = 0,
1494 		     total_unknown = 0,
1495 		     total_lost = 0;
1496 
1497 static int validate_chain(struct ip_callchain *chain, event_t *event)
1498 {
1499 	unsigned int chain_size;
1500 
1501 	chain_size = event->header.size;
1502 	chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1503 
1504 	if (chain->nr*sizeof(u64) > chain_size)
1505 		return -1;
1506 
1507 	return 0;
1508 }
1509 
1510 static int
1511 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1512 {
1513 	char level;
1514 	int show = 0;
1515 	struct dso *dso = NULL;
1516 	struct thread *thread = threads__findnew(event->ip.pid);
1517 	u64 ip = event->ip.ip;
1518 	u64 period = 1;
1519 	struct map *map = NULL;
1520 	void *more_data = event->ip.__more_data;
1521 	struct ip_callchain *chain = NULL;
1522 	int cpumode;
1523 
1524 	if (sample_type & PERF_SAMPLE_PERIOD) {
1525 		period = *(u64 *)more_data;
1526 		more_data += sizeof(u64);
1527 	}
1528 
1529 	dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1530 		(void *)(offset + head),
1531 		(void *)(long)(event->header.size),
1532 		event->header.misc,
1533 		event->ip.pid, event->ip.tid,
1534 		(void *)(long)ip,
1535 		(long long)period);
1536 
1537 	if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1538 		unsigned int i;
1539 
1540 		chain = (void *)more_data;
1541 
1542 		dprintf("... chain: nr:%Lu\n", chain->nr);
1543 
1544 		if (validate_chain(chain, event) < 0) {
1545 			eprintf("call-chain problem with event, skipping it.\n");
1546 			return 0;
1547 		}
1548 
1549 		if (dump_trace) {
1550 			for (i = 0; i < chain->nr; i++)
1551 				dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1552 		}
1553 	}
1554 
1555 	dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1556 
1557 	if (thread == NULL) {
1558 		eprintf("problem processing %d event, skipping it.\n",
1559 			event->header.type);
1560 		return -1;
1561 	}
1562 
1563 	if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1564 		return 0;
1565 
1566 	cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1567 
1568 	if (cpumode == PERF_EVENT_MISC_KERNEL) {
1569 		show = SHOW_KERNEL;
1570 		level = 'k';
1571 
1572 		dso = kernel_dso;
1573 
1574 		dprintf(" ...... dso: %s\n", dso->name);
1575 
1576 	} else if (cpumode == PERF_EVENT_MISC_USER) {
1577 
1578 		show = SHOW_USER;
1579 		level = '.';
1580 
1581 	} else {
1582 		show = SHOW_HV;
1583 		level = 'H';
1584 
1585 		dso = hypervisor_dso;
1586 
1587 		dprintf(" ...... dso: [hypervisor]\n");
1588 	}
1589 
1590 	if (show & show_mask) {
1591 		struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1592 
1593 		if (dso_list && (!dso || !dso->name ||
1594 				 !strlist__has_entry(dso_list, dso->name)))
1595 			return 0;
1596 
1597 		if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1598 			return 0;
1599 
1600 		if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1601 			eprintf("problem incrementing symbol count, skipping event\n");
1602 			return -1;
1603 		}
1604 	}
1605 	total += period;
1606 
1607 	return 0;
1608 }
1609 
1610 static int
1611 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1612 {
1613 	struct thread *thread = threads__findnew(event->mmap.pid);
1614 	struct map *map = map__new(&event->mmap);
1615 
1616 	dprintf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1617 		(void *)(offset + head),
1618 		(void *)(long)(event->header.size),
1619 		event->mmap.pid,
1620 		event->mmap.tid,
1621 		(void *)(long)event->mmap.start,
1622 		(void *)(long)event->mmap.len,
1623 		(void *)(long)event->mmap.pgoff,
1624 		event->mmap.filename);
1625 
1626 	if (thread == NULL || map == NULL) {
1627 		dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1628 		return 0;
1629 	}
1630 
1631 	thread__insert_map(thread, map);
1632 	total_mmap++;
1633 
1634 	return 0;
1635 }
1636 
1637 static int
1638 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1639 {
1640 	struct thread *thread = threads__findnew(event->comm.pid);
1641 
1642 	dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1643 		(void *)(offset + head),
1644 		(void *)(long)(event->header.size),
1645 		event->comm.comm, event->comm.pid);
1646 
1647 	if (thread == NULL ||
1648 	    thread__set_comm(thread, event->comm.comm)) {
1649 		dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1650 		return -1;
1651 	}
1652 	total_comm++;
1653 
1654 	return 0;
1655 }
1656 
1657 static int
1658 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1659 {
1660 	struct thread *thread = threads__findnew(event->fork.pid);
1661 	struct thread *parent = threads__findnew(event->fork.ppid);
1662 
1663 	dprintf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1664 		(void *)(offset + head),
1665 		(void *)(long)(event->header.size),
1666 		event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT",
1667 		event->fork.pid, event->fork.tid,
1668 		event->fork.ppid, event->fork.ptid);
1669 
1670 	/*
1671 	 * A thread clone will have the same PID for both
1672 	 * parent and child.
1673 	 */
1674 	if (thread == parent)
1675 		return 0;
1676 
1677 	if (event->header.type == PERF_EVENT_EXIT)
1678 		return 0;
1679 
1680 	if (!thread || !parent || thread__fork(thread, parent)) {
1681 		dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1682 		return -1;
1683 	}
1684 	total_fork++;
1685 
1686 	return 0;
1687 }
1688 
1689 static int
1690 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1691 {
1692 	dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1693 		(void *)(offset + head),
1694 		(void *)(long)(event->header.size),
1695 		event->lost.id,
1696 		event->lost.lost);
1697 
1698 	total_lost += event->lost.lost;
1699 
1700 	return 0;
1701 }
1702 
1703 static void trace_event(event_t *event)
1704 {
1705 	unsigned char *raw_event = (void *)event;
1706 	char *color = PERF_COLOR_BLUE;
1707 	int i, j;
1708 
1709 	if (!dump_trace)
1710 		return;
1711 
1712 	dprintf(".");
1713 	cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1714 
1715 	for (i = 0; i < event->header.size; i++) {
1716 		if ((i & 15) == 0) {
1717 			dprintf(".");
1718 			cdprintf("  %04x: ", i);
1719 		}
1720 
1721 		cdprintf(" %02x", raw_event[i]);
1722 
1723 		if (((i & 15) == 15) || i == event->header.size-1) {
1724 			cdprintf("  ");
1725 			for (j = 0; j < 15-(i & 15); j++)
1726 				cdprintf("   ");
1727 			for (j = 0; j < (i & 15); j++) {
1728 				if (isprint(raw_event[i-15+j]))
1729 					cdprintf("%c", raw_event[i-15+j]);
1730 				else
1731 					cdprintf(".");
1732 			}
1733 			cdprintf("\n");
1734 		}
1735 	}
1736 	dprintf(".\n");
1737 }
1738 
1739 static struct perf_header	*header;
1740 
1741 static struct perf_counter_attr *perf_header__find_attr(u64 id)
1742 {
1743 	int i;
1744 
1745 	for (i = 0; i < header->attrs; i++) {
1746 		struct perf_header_attr *attr = header->attr[i];
1747 		int j;
1748 
1749 		for (j = 0; j < attr->ids; j++) {
1750 			if (attr->id[j] == id)
1751 				return &attr->attr;
1752 		}
1753 	}
1754 
1755 	return NULL;
1756 }
1757 
1758 static int
1759 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1760 {
1761 	struct perf_counter_attr *attr = perf_header__find_attr(event->read.id);
1762 
1763 	dprintf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1764 			(void *)(offset + head),
1765 			(void *)(long)(event->header.size),
1766 			event->read.pid,
1767 			event->read.tid,
1768 			attr ? __event_name(attr->type, attr->config)
1769 			     : "FAIL",
1770 			event->read.value);
1771 
1772 	return 0;
1773 }
1774 
1775 static int
1776 process_event(event_t *event, unsigned long offset, unsigned long head)
1777 {
1778 	trace_event(event);
1779 
1780 	switch (event->header.type) {
1781 	case PERF_EVENT_SAMPLE:
1782 		return process_sample_event(event, offset, head);
1783 
1784 	case PERF_EVENT_MMAP:
1785 		return process_mmap_event(event, offset, head);
1786 
1787 	case PERF_EVENT_COMM:
1788 		return process_comm_event(event, offset, head);
1789 
1790 	case PERF_EVENT_FORK:
1791 	case PERF_EVENT_EXIT:
1792 		return process_task_event(event, offset, head);
1793 
1794 	case PERF_EVENT_LOST:
1795 		return process_lost_event(event, offset, head);
1796 
1797 	case PERF_EVENT_READ:
1798 		return process_read_event(event, offset, head);
1799 
1800 	/*
1801 	 * We dont process them right now but they are fine:
1802 	 */
1803 
1804 	case PERF_EVENT_THROTTLE:
1805 	case PERF_EVENT_UNTHROTTLE:
1806 		return 0;
1807 
1808 	default:
1809 		return -1;
1810 	}
1811 
1812 	return 0;
1813 }
1814 
1815 static u64 perf_header__sample_type(void)
1816 {
1817 	u64 sample_type = 0;
1818 	int i;
1819 
1820 	for (i = 0; i < header->attrs; i++) {
1821 		struct perf_header_attr *attr = header->attr[i];
1822 
1823 		if (!sample_type)
1824 			sample_type = attr->attr.sample_type;
1825 		else if (sample_type != attr->attr.sample_type)
1826 			die("non matching sample_type");
1827 	}
1828 
1829 	return sample_type;
1830 }
1831 
1832 static int __cmd_report(void)
1833 {
1834 	int ret, rc = EXIT_FAILURE;
1835 	unsigned long offset = 0;
1836 	unsigned long head, shift;
1837 	struct stat stat;
1838 	event_t *event;
1839 	uint32_t size;
1840 	char *buf;
1841 
1842 	register_idle_thread();
1843 
1844 	input = open(input_name, O_RDONLY);
1845 	if (input < 0) {
1846 		fprintf(stderr, " failed to open file: %s", input_name);
1847 		if (!strcmp(input_name, "perf.data"))
1848 			fprintf(stderr, "  (try 'perf record' first)");
1849 		fprintf(stderr, "\n");
1850 		exit(-1);
1851 	}
1852 
1853 	ret = fstat(input, &stat);
1854 	if (ret < 0) {
1855 		perror("failed to stat file");
1856 		exit(-1);
1857 	}
1858 
1859 	if (!stat.st_size) {
1860 		fprintf(stderr, "zero-sized file, nothing to do!\n");
1861 		exit(0);
1862 	}
1863 
1864 	header = perf_header__read(input);
1865 	head = header->data_offset;
1866 
1867 	sample_type = perf_header__sample_type();
1868 
1869 	if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1870 		if (sort__has_parent) {
1871 			fprintf(stderr, "selected --sort parent, but no"
1872 					" callchain data. Did you call"
1873 					" perf record without -g?\n");
1874 			exit(-1);
1875 		}
1876 		if (callchain) {
1877 			fprintf(stderr, "selected -c but no callchain data."
1878 					" Did you call perf record without"
1879 					" -g?\n");
1880 			exit(-1);
1881 		}
1882 	} else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1883 			callchain = 1;
1884 			if (register_callchain_param(&callchain_param) < 0) {
1885 				fprintf(stderr, "Can't register callchain"
1886 						" params\n");
1887 				exit(-1);
1888 			}
1889 	}
1890 
1891 	if (load_kernel() < 0) {
1892 		perror("failed to load kernel symbols");
1893 		return EXIT_FAILURE;
1894 	}
1895 
1896 	if (!full_paths) {
1897 		if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1898 			perror("failed to get the current directory");
1899 			return EXIT_FAILURE;
1900 		}
1901 		cwdlen = strlen(cwd);
1902 	} else {
1903 		cwd = NULL;
1904 		cwdlen = 0;
1905 	}
1906 
1907 	shift = page_size * (head / page_size);
1908 	offset += shift;
1909 	head -= shift;
1910 
1911 remap:
1912 	buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1913 			   MAP_SHARED, input, offset);
1914 	if (buf == MAP_FAILED) {
1915 		perror("failed to mmap file");
1916 		exit(-1);
1917 	}
1918 
1919 more:
1920 	event = (event_t *)(buf + head);
1921 
1922 	size = event->header.size;
1923 	if (!size)
1924 		size = 8;
1925 
1926 	if (head + event->header.size >= page_size * mmap_window) {
1927 		int ret;
1928 
1929 		shift = page_size * (head / page_size);
1930 
1931 		ret = munmap(buf, page_size * mmap_window);
1932 		assert(ret == 0);
1933 
1934 		offset += shift;
1935 		head -= shift;
1936 		goto remap;
1937 	}
1938 
1939 	size = event->header.size;
1940 
1941 	dprintf("\n%p [%p]: event: %d\n",
1942 			(void *)(offset + head),
1943 			(void *)(long)event->header.size,
1944 			event->header.type);
1945 
1946 	if (!size || process_event(event, offset, head) < 0) {
1947 
1948 		dprintf("%p [%p]: skipping unknown header type: %d\n",
1949 			(void *)(offset + head),
1950 			(void *)(long)(event->header.size),
1951 			event->header.type);
1952 
1953 		total_unknown++;
1954 
1955 		/*
1956 		 * assume we lost track of the stream, check alignment, and
1957 		 * increment a single u64 in the hope to catch on again 'soon'.
1958 		 */
1959 
1960 		if (unlikely(head & 7))
1961 			head &= ~7ULL;
1962 
1963 		size = 8;
1964 	}
1965 
1966 	head += size;
1967 
1968 	if (offset + head >= header->data_offset + header->data_size)
1969 		goto done;
1970 
1971 	if (offset + head < (unsigned long)stat.st_size)
1972 		goto more;
1973 
1974 done:
1975 	rc = EXIT_SUCCESS;
1976 	close(input);
1977 
1978 	dprintf("      IP events: %10ld\n", total);
1979 	dprintf("    mmap events: %10ld\n", total_mmap);
1980 	dprintf("    comm events: %10ld\n", total_comm);
1981 	dprintf("    fork events: %10ld\n", total_fork);
1982 	dprintf("    lost events: %10ld\n", total_lost);
1983 	dprintf(" unknown events: %10ld\n", total_unknown);
1984 
1985 	if (dump_trace)
1986 		return 0;
1987 
1988 	if (verbose >= 3)
1989 		threads__fprintf(stdout);
1990 
1991 	if (verbose >= 2)
1992 		dsos__fprintf(stdout);
1993 
1994 	collapse__resort();
1995 	output__resort(total);
1996 	output__fprintf(stdout, total);
1997 
1998 	return rc;
1999 }
2000 
2001 static int
2002 parse_callchain_opt(const struct option *opt __used, const char *arg,
2003 		    int unset __used)
2004 {
2005 	char *tok;
2006 	char *endptr;
2007 
2008 	callchain = 1;
2009 
2010 	if (!arg)
2011 		return 0;
2012 
2013 	tok = strtok((char *)arg, ",");
2014 	if (!tok)
2015 		return -1;
2016 
2017 	/* get the output mode */
2018 	if (!strncmp(tok, "graph", strlen(arg)))
2019 		callchain_param.mode = CHAIN_GRAPH_ABS;
2020 
2021 	else if (!strncmp(tok, "flat", strlen(arg)))
2022 		callchain_param.mode = CHAIN_FLAT;
2023 
2024 	else if (!strncmp(tok, "fractal", strlen(arg)))
2025 		callchain_param.mode = CHAIN_GRAPH_REL;
2026 
2027 	else if (!strncmp(tok, "none", strlen(arg))) {
2028 		callchain_param.mode = CHAIN_NONE;
2029 		callchain = 0;
2030 
2031 		return 0;
2032 	}
2033 
2034 	else
2035 		return -1;
2036 
2037 	/* get the min percentage */
2038 	tok = strtok(NULL, ",");
2039 	if (!tok)
2040 		goto setup;
2041 
2042 	callchain_param.min_percent = strtod(tok, &endptr);
2043 	if (tok == endptr)
2044 		return -1;
2045 
2046 setup:
2047 	if (register_callchain_param(&callchain_param) < 0) {
2048 		fprintf(stderr, "Can't register callchain params\n");
2049 		return -1;
2050 	}
2051 	return 0;
2052 }
2053 
2054 static const char * const report_usage[] = {
2055 	"perf report [<options>] <command>",
2056 	NULL
2057 };
2058 
2059 static const struct option options[] = {
2060 	OPT_STRING('i', "input", &input_name, "file",
2061 		    "input file name"),
2062 	OPT_BOOLEAN('v', "verbose", &verbose,
2063 		    "be more verbose (show symbol address, etc)"),
2064 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
2065 		    "dump raw trace in ASCII"),
2066 	OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
2067 	OPT_BOOLEAN('m', "modules", &modules,
2068 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
2069 	OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
2070 		    "Show a column with the number of samples"),
2071 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
2072 		   "sort by key(s): pid, comm, dso, symbol, parent"),
2073 	OPT_BOOLEAN('P', "full-paths", &full_paths,
2074 		    "Don't shorten the pathnames taking into account the cwd"),
2075 	OPT_STRING('p', "parent", &parent_pattern, "regex",
2076 		   "regex filter to identify parent, see: '--sort parent'"),
2077 	OPT_BOOLEAN('x', "exclude-other", &exclude_other,
2078 		    "Only display entries with parent-match"),
2079 	OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
2080 		     "Display callchains using output_type and min percent threshold. "
2081 		     "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
2082 	OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
2083 		   "only consider symbols in these dsos"),
2084 	OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
2085 		   "only consider symbols in these comms"),
2086 	OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
2087 		   "only consider these symbols"),
2088 	OPT_STRING('w', "column-widths", &col_width_list_str,
2089 		   "width[,width...]",
2090 		   "don't try to adjust column width, use these fixed values"),
2091 	OPT_STRING('t', "field-separator", &field_sep, "separator",
2092 		   "separator for columns, no spaces will be added between "
2093 		   "columns '.' is reserved."),
2094 	OPT_END()
2095 };
2096 
2097 static void setup_sorting(void)
2098 {
2099 	char *tmp, *tok, *str = strdup(sort_order);
2100 
2101 	for (tok = strtok_r(str, ", ", &tmp);
2102 			tok; tok = strtok_r(NULL, ", ", &tmp)) {
2103 		if (sort_dimension__add(tok) < 0) {
2104 			error("Unknown --sort key: `%s'", tok);
2105 			usage_with_options(report_usage, options);
2106 		}
2107 	}
2108 
2109 	free(str);
2110 }
2111 
2112 static void setup_list(struct strlist **list, const char *list_str,
2113 		       struct sort_entry *se, const char *list_name,
2114 		       FILE *fp)
2115 {
2116 	if (list_str) {
2117 		*list = strlist__new(true, list_str);
2118 		if (!*list) {
2119 			fprintf(stderr, "problems parsing %s list\n",
2120 				list_name);
2121 			exit(129);
2122 		}
2123 		if (strlist__nr_entries(*list) == 1) {
2124 			fprintf(fp, "# %s: %s\n", list_name,
2125 				strlist__entry(*list, 0)->s);
2126 			se->elide = true;
2127 		}
2128 	}
2129 }
2130 
2131 int cmd_report(int argc, const char **argv, const char *prefix __used)
2132 {
2133 	symbol__init();
2134 
2135 	page_size = getpagesize();
2136 
2137 	argc = parse_options(argc, argv, options, report_usage, 0);
2138 
2139 	setup_sorting();
2140 
2141 	if (parent_pattern != default_parent_pattern) {
2142 		sort_dimension__add("parent");
2143 		sort_parent.elide = 1;
2144 	} else
2145 		exclude_other = 0;
2146 
2147 	/*
2148 	 * Any (unrecognized) arguments left?
2149 	 */
2150 	if (argc)
2151 		usage_with_options(report_usage, options);
2152 
2153 	setup_pager();
2154 
2155 	setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
2156 	setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
2157 	setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
2158 
2159 	if (field_sep && *field_sep == '.') {
2160 		fputs("'.' is the only non valid --field-separator argument\n",
2161 		      stderr);
2162 		exit(129);
2163 	}
2164 
2165 	return __cmd_report();
2166 }
2167