builtin-report.c 31.5 KB
Newer Older
1 2 3 4 5 6 7
/*
 * builtin-report.c
 *
 * Builtin report command: Analyze the perf.data input file,
 * look up and read DSOs and symbol information and display
 * a histogram of results, along various sorting keys.
 */
8
#include "builtin.h"
9

10 11
#include "util/util.h"

12
#include "util/color.h"
13
#include "util/list.h"
14
#include "util/cache.h"
15
#include "util/rbtree.h"
16
#include "util/symbol.h"
17
#include "util/string.h"
18

19
#include "perf.h"
20
#include "util/header.h"
21 22 23 24

#include "util/parse-options.h"
#include "util/parse-events.h"

25 26 27 28
#define SHOW_KERNEL	1
#define SHOW_USER	2
#define SHOW_HV		4

29
static char		const *input_name = "perf.data";
30
static char		*vmlinux = NULL;
31 32 33 34

static char		default_sort_order[] = "comm,dso";
static char		*sort_order = default_sort_order;

35 36 37
static int		input;
static int		show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;

38
static int		dump_trace = 0;
I
Ingo Molnar 已提交
39
#define dprintf(x...)	do { if (dump_trace) printf(x); } while (0)
40
#define cdprintf(x...)	do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
I
Ingo Molnar 已提交
41

42
static int		verbose;
43 44
#define eprintf(x...)	do { if (verbose) fprintf(stderr, x); } while (0)

45
static int		full_paths;
46

47 48 49
static unsigned long	page_size;
static unsigned long	mmap_window = 32;

50 51
static char		default_parent_pattern[] = "^sys_|^do_page_fault";
static char		*parent_pattern = default_parent_pattern;
52
static regex_t		parent_regex;
53

54 55
static int		exclude_other = 1;

56 57
static u64		sample_type;

58 59
struct ip_event {
	struct perf_event_header header;
60 61
	u64 ip;
	u32 pid, tid;
62
	unsigned char __more_data[];
63
};
64

65
struct ip_callchain {
66 67
	u64 nr;
	u64 ips[0];
68 69
};

70 71
struct mmap_event {
	struct perf_event_header header;
72 73 74 75
	u32 pid, tid;
	u64 start;
	u64 len;
	u64 pgoff;
76 77
	char filename[PATH_MAX];
};
78

79 80
struct comm_event {
	struct perf_event_header header;
81
	u32 pid, tid;
82 83 84
	char comm[16];
};

85 86
struct fork_event {
	struct perf_event_header header;
87
	u32 pid, ppid;
88 89
};

90
struct period_event {
91
	struct perf_event_header header;
92 93 94
	u64 time;
	u64 id;
	u64 sample_period;
95 96
};

97 98
struct lost_event {
	struct perf_event_header header;
99 100
	u64 id;
	u64 lost;
101 102
};

103 104 105 106 107 108 109
typedef union event_union {
	struct perf_event_header	header;
	struct ip_event			ip;
	struct mmap_event		mmap;
	struct comm_event		comm;
	struct fork_event		fork;
	struct period_event		period;
110
	struct lost_event		lost;
111 112 113 114
} event_t;

static LIST_HEAD(dsos);
static struct dso *kernel_dso;
P
Peter Zijlstra 已提交
115
static struct dso *vdso;
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134

static void dsos__add(struct dso *dso)
{
	list_add_tail(&dso->node, &dsos);
}

static struct dso *dsos__find(const char *name)
{
	struct dso *pos;

	list_for_each_entry(pos, &dsos, node)
		if (strcmp(pos->name, name) == 0)
			return pos;
	return NULL;
}

static struct dso *dsos__findnew(const char *name)
{
	struct dso *dso = dsos__find(name);
135
	int nr;
136

137 138 139 140 141 142
	if (dso)
		return dso;

	dso = dso__new(name, 0);
	if (!dso)
		goto out_delete_dso;
143

144
	nr = dso__load(dso, NULL, verbose);
145
	if (nr < 0) {
146
		eprintf("Failed to open: %s\n", name);
147
		goto out_delete_dso;
148
	}
149 150
	if (!nr)
		eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
151 152

	dsos__add(dso);
153 154 155 156 157 158 159 160

	return dso;

out_delete_dso:
	dso__delete(dso);
	return NULL;
}

161
static void dsos__fprintf(FILE *fp)
162 163 164 165 166 167 168
{
	struct dso *pos;

	list_for_each_entry(pos, &dsos, node)
		dso__fprintf(pos, fp);
}

169
static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
P
Peter Zijlstra 已提交
170 171 172 173
{
	return dso__find_symbol(kernel_dso, ip);
}

174 175
static int load_kernel(void)
{
176
	int err;
177

178
	kernel_dso = dso__new("[kernel]", 0);
179
	if (!kernel_dso)
180
		return -1;
181

182
	err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
183 184 185 186 187
	if (err) {
		dso__delete(kernel_dso);
		kernel_dso = NULL;
	} else
		dsos__add(kernel_dso);
188

P
Peter Zijlstra 已提交
189 190 191 192 193 194 195 196
	vdso = dso__new("[vdso]", 0);
	if (!vdso)
		return -1;

	vdso->find_symbol = vdso__find_symbol;

	dsos__add(vdso);

197
	return err;
198 199
}

200 201 202 203 204
static char __cwd[PATH_MAX];
static char *cwd = __cwd;
static int cwdlen;

static int strcommon(const char *pathname)
205 206 207 208 209 210 211 212 213
{
	int n = 0;

	while (pathname[n] == cwd[n] && n < cwdlen)
		++n;

	return n;
}

214 215
struct map {
	struct list_head node;
216 217 218 219
	u64	 start;
	u64	 end;
	u64	 pgoff;
	u64	 (*map_ip)(struct map *, u64);
220 221 222
	struct dso	 *dso;
};

223
static u64 map__map_ip(struct map *map, u64 ip)
P
Peter Zijlstra 已提交
224 225 226 227
{
	return ip - map->start + map->pgoff;
}

228
static u64 vdso__map_ip(struct map *map, u64 ip)
P
Peter Zijlstra 已提交
229 230 231 232
{
	return ip;
}

233 234 235 236 237
static inline int is_anon_memory(const char *filename)
{
     return strcmp(filename, "//anon") == 0;
}

238
static struct map *map__new(struct mmap_event *event)
239 240 241 242
{
	struct map *self = malloc(sizeof(*self));

	if (self != NULL) {
243 244
		const char *filename = event->filename;
		char newfilename[PATH_MAX];
245
		int anon;
246 247

		if (cwd) {
248 249
			int n = strcommon(filename);

250 251 252 253 254 255 256
			if (n == cwdlen) {
				snprintf(newfilename, sizeof(newfilename),
					 ".%s", filename + n);
				filename = newfilename;
			}
		}

257 258 259 260 261 262 263
		anon = is_anon_memory(filename);

		if (anon) {
			snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
			filename = newfilename;
		}

264 265 266 267
		self->start = event->start;
		self->end   = event->start + event->len;
		self->pgoff = event->pgoff;

268
		self->dso = dsos__findnew(filename);
269 270
		if (self->dso == NULL)
			goto out_delete;
P
Peter Zijlstra 已提交
271

272
		if (self->dso == vdso || anon)
P
Peter Zijlstra 已提交
273 274 275
			self->map_ip = vdso__map_ip;
		else
			self->map_ip = map__map_ip;
276 277 278 279 280 281 282
	}
	return self;
out_delete:
	free(self);
	return NULL;
}

283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307
static struct map *map__clone(struct map *self)
{
	struct map *map = malloc(sizeof(*self));

	if (!map)
		return NULL;

	memcpy(map, self, sizeof(*self));

	return map;
}

static int map__overlap(struct map *l, struct map *r)
{
	if (l->start > r->start) {
		struct map *t = l;
		l = r;
		r = t;
	}

	if (l->end > r->start)
		return 1;

	return 0;
}
308

309 310
static size_t map__fprintf(struct map *self, FILE *fp)
{
311
	return fprintf(fp, " %Lx-%Lx %Lx %s\n",
312 313 314 315
		       self->start, self->end, self->pgoff, self->dso->name);
}


316
struct thread {
317
	struct rb_node	 rb_node;
318 319 320 321 322 323 324 325 326 327 328
	struct list_head maps;
	pid_t		 pid;
	char		 *comm;
};

static struct thread *thread__new(pid_t pid)
{
	struct thread *self = malloc(sizeof(*self));

	if (self != NULL) {
		self->pid = pid;
P
Peter Zijlstra 已提交
329
		self->comm = malloc(32);
330
		if (self->comm)
P
Peter Zijlstra 已提交
331
			snprintf(self->comm, 32, ":%d", self->pid);
332 333 334 335 336 337 338 339
		INIT_LIST_HEAD(&self->maps);
	}

	return self;
}

static int thread__set_comm(struct thread *self, const char *comm)
{
P
Peter Zijlstra 已提交
340 341
	if (self->comm)
		free(self->comm);
342 343 344 345
	self->comm = strdup(comm);
	return self->comm ? 0 : -ENOMEM;
}

346 347 348 349 350 351 352 353 354 355 356 357
static size_t thread__fprintf(struct thread *self, FILE *fp)
{
	struct map *pos;
	size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);

	list_for_each_entry(pos, &self->maps, node)
		ret += map__fprintf(pos, fp);

	return ret;
}


358
static struct rb_root threads;
359
static struct thread *last_match;
360

361
static struct thread *threads__findnew(pid_t pid)
362
{
363 364 365
	struct rb_node **p = &threads.rb_node;
	struct rb_node *parent = NULL;
	struct thread *th;
366

367 368 369 370 371 372 373 374
	/*
	 * Font-end cache - PID lookups come in blocks,
	 * so most of the time we dont have to look up
	 * the full rbtree:
	 */
	if (last_match && last_match->pid == pid)
		return last_match;

375 376 377
	while (*p != NULL) {
		parent = *p;
		th = rb_entry(parent, struct thread, rb_node);
378

379 380
		if (th->pid == pid) {
			last_match = th;
381
			return th;
382
		}
383

384 385 386 387
		if (pid < th->pid)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
388 389
	}

390 391 392 393
	th = thread__new(pid);
	if (th != NULL) {
		rb_link_node(&th->rb_node, parent, p);
		rb_insert_color(&th->rb_node, &threads);
394
		last_match = th;
395
	}
396

397
	return th;
398 399 400 401
}

static void thread__insert_map(struct thread *self, struct map *map)
{
402 403 404 405
	struct map *pos, *tmp;

	list_for_each_entry_safe(pos, tmp, &self->maps, node) {
		if (map__overlap(pos, map)) {
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
			if (verbose >= 2) {
				printf("overlapping maps:\n");
				map__fprintf(map, stdout);
				map__fprintf(pos, stdout);
			}

			if (map->start <= pos->start && map->end > pos->start)
				pos->start = map->end;

			if (map->end >= pos->end && map->start < pos->end)
				pos->end = map->start;

			if (verbose >= 2) {
				printf("after collision:\n");
				map__fprintf(pos, stdout);
			}

			if (pos->start >= pos->end) {
				list_del_init(&pos->node);
				free(pos);
			}
427 428 429
		}
	}

430 431 432
	list_add_tail(&map->node, &self->maps);
}

433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452
static int thread__fork(struct thread *self, struct thread *parent)
{
	struct map *map;

	if (self->comm)
		free(self->comm);
	self->comm = strdup(parent->comm);
	if (!self->comm)
		return -ENOMEM;

	list_for_each_entry(map, &parent->maps, node) {
		struct map *new = map__clone(map);
		if (!new)
			return -ENOMEM;
		thread__insert_map(self, new);
	}

	return 0;
}

453
static struct map *thread__find_map(struct thread *self, u64 ip)
454
{
455 456
	struct map *pos;

457 458 459 460 461 462 463 464 465 466
	if (self == NULL)
		return NULL;

	list_for_each_entry(pos, &self->maps, node)
		if (ip >= pos->start && ip <= pos->end)
			return pos;

	return NULL;
}

467 468 469 470 471 472 473 474 475 476 477 478 479 480
static size_t threads__fprintf(FILE *fp)
{
	size_t ret = 0;
	struct rb_node *nd;

	for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
		struct thread *pos = rb_entry(nd, struct thread, rb_node);

		ret += thread__fprintf(pos, fp);
	}

	return ret;
}

481 482 483 484 485 486 487 488 489 490 491 492 493
/*
 * histogram, sorted on item, collects counts
 */

static struct rb_root hist;

struct hist_entry {
	struct rb_node	 rb_node;

	struct thread	 *thread;
	struct map	 *map;
	struct dso	 *dso;
	struct symbol	 *sym;
494
	struct symbol	 *parent;
495
	u64		 ip;
496 497
	char		 level;

498
	u64		 count;
499 500
};

501 502 503 504 505 506 507
/*
 * configurable sorting bits
 */

struct sort_entry {
	struct list_head list;

508 509
	char *header;

510
	int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
P
Peter Zijlstra 已提交
511
	int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
512 513 514
	size_t	(*print)(FILE *fp, struct hist_entry *);
};

515 516 517 518 519 520 521 522 523 524
static int64_t cmp_null(void *l, void *r)
{
	if (!l && !r)
		return 0;
	else if (!l)
		return -1;
	else
		return 1;
}

P
Peter Zijlstra 已提交
525 526
/* --sort pid */

527
static int64_t
528
sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
529
{
530 531 532 533 534 535
	return right->thread->pid - left->thread->pid;
}

static size_t
sort__thread_print(FILE *fp, struct hist_entry *self)
{
536
	return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
537
}
538

539
static struct sort_entry sort_thread = {
540
	.header = "         Command:  Pid",
541 542 543 544
	.cmp	= sort__thread_cmp,
	.print	= sort__thread_print,
};

P
Peter Zijlstra 已提交
545 546
/* --sort comm */

547 548
static int64_t
sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
P
Peter Zijlstra 已提交
549 550 551 552 553 554
{
	return right->thread->pid - left->thread->pid;
}

static int64_t
sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
555 556 557 558
{
	char *comm_l = left->thread->comm;
	char *comm_r = right->thread->comm;

559 560
	if (!comm_l || !comm_r)
		return cmp_null(comm_l, comm_r);
561 562 563 564 565 566 567

	return strcmp(comm_l, comm_r);
}

static size_t
sort__comm_print(FILE *fp, struct hist_entry *self)
{
568
	return fprintf(fp, "%16s", self->thread->comm);
569 570 571
}

static struct sort_entry sort_comm = {
572
	.header		= "         Command",
P
Peter Zijlstra 已提交
573 574 575
	.cmp		= sort__comm_cmp,
	.collapse	= sort__comm_collapse,
	.print		= sort__comm_print,
576 577
};

P
Peter Zijlstra 已提交
578 579
/* --sort dso */

580 581 582 583 584 585
static int64_t
sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
{
	struct dso *dso_l = left->dso;
	struct dso *dso_r = right->dso;

586 587
	if (!dso_l || !dso_r)
		return cmp_null(dso_l, dso_r);
588 589 590 591 592 593 594

	return strcmp(dso_l->name, dso_r->name);
}

static size_t
sort__dso_print(FILE *fp, struct hist_entry *self)
{
595
	if (self->dso)
596
		return fprintf(fp, "%-25s", self->dso->name);
597

598
	return fprintf(fp, "%016llx         ", (u64)self->ip);
599 600 601
}

static struct sort_entry sort_dso = {
602
	.header = "Shared Object            ",
603 604 605 606
	.cmp	= sort__dso_cmp,
	.print	= sort__dso_print,
};

P
Peter Zijlstra 已提交
607 608
/* --sort symbol */

609 610 611
static int64_t
sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
{
612
	u64 ip_l, ip_r;
613 614 615 616 617 618 619 620 621 622

	if (left->sym == right->sym)
		return 0;

	ip_l = left->sym ? left->sym->start : left->ip;
	ip_r = right->sym ? right->sym->start : right->ip;

	return (int64_t)(ip_r - ip_l);
}

623 624 625 626 627 628
static size_t
sort__sym_print(FILE *fp, struct hist_entry *self)
{
	size_t ret = 0;

	if (verbose)
629
		ret += fprintf(fp, "%#018llx  ", (u64)self->ip);
630

631 632 633 634
	if (self->sym) {
		ret += fprintf(fp, "[%c] %s",
			self->dso == kernel_dso ? 'k' : '.', self->sym->name);
	} else {
635
		ret += fprintf(fp, "%#016llx", (u64)self->ip);
636
	}
637 638 639 640 641

	return ret;
}

static struct sort_entry sort_sym = {
642
	.header = "Symbol",
643 644
	.cmp	= sort__sym_cmp,
	.print	= sort__sym_print,
645 646
};

647
/* --sort parent */
648 649

static int64_t
650
sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
651
{
652 653
	struct symbol *sym_l = left->parent;
	struct symbol *sym_r = right->parent;
654 655 656 657 658 659 660 661

	if (!sym_l || !sym_r)
		return cmp_null(sym_l, sym_r);

	return strcmp(sym_l->name, sym_r->name);
}

static size_t
662
sort__parent_print(FILE *fp, struct hist_entry *self)
663 664 665
{
	size_t ret = 0;

666
	ret += fprintf(fp, "%-20s", self->parent ? self->parent->name : "[other]");
667 668 669 670

	return ret;
}

671 672 673 674
static struct sort_entry sort_parent = {
	.header = "Parent symbol       ",
	.cmp	= sort__parent_cmp,
	.print	= sort__parent_print,
675 676
};

P
Peter Zijlstra 已提交
677
static int sort__need_collapse = 0;
678
static int sort__has_parent = 0;
P
Peter Zijlstra 已提交
679

680
struct sort_dimension {
681 682 683
	char			*name;
	struct sort_entry	*entry;
	int			taken;
684 685 686 687
};

static struct sort_dimension sort_dimensions[] = {
	{ .name = "pid",	.entry = &sort_thread,	},
688
	{ .name = "comm",	.entry = &sort_comm,	},
689
	{ .name = "dso",	.entry = &sort_dso,	},
690
	{ .name = "symbol",	.entry = &sort_sym,	},
691
	{ .name = "parent",	.entry = &sort_parent,	},
692 693
};

694 695
static LIST_HEAD(hist_entry__sort_list);

696 697 698 699 700 701 702 703 704 705
static int sort_dimension__add(char *tok)
{
	int i;

	for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
		struct sort_dimension *sd = &sort_dimensions[i];

		if (sd->taken)
			continue;

706
		if (strncasecmp(tok, sd->name, strlen(tok)))
707 708
			continue;

P
Peter Zijlstra 已提交
709 710 711
		if (sd->entry->collapse)
			sort__need_collapse = 1;

712 713
		if (sd->entry == &sort_parent) {
			int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
714 715 716
			if (ret) {
				char err[BUFSIZ];

717 718 719
				regerror(ret, &parent_regex, err, sizeof(err));
				fprintf(stderr, "Invalid regex: %s\n%s",
					parent_pattern, err);
720 721
				exit(-1);
			}
722
			sort__has_parent = 1;
723 724
		}

725 726
		list_add_tail(&sd->entry->list, &hist_entry__sort_list);
		sd->taken = 1;
727

728 729 730 731 732 733
		return 0;
	}

	return -ESRCH;
}

734 735 736 737 738 739 740 741 742 743 744 745 746 747 748
static int64_t
hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
{
	struct sort_entry *se;
	int64_t cmp = 0;

	list_for_each_entry(se, &hist_entry__sort_list, list) {
		cmp = se->cmp(left, right);
		if (cmp)
			break;
	}

	return cmp;
}

P
Peter Zijlstra 已提交
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767
static int64_t
hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
{
	struct sort_entry *se;
	int64_t cmp = 0;

	list_for_each_entry(se, &hist_entry__sort_list, list) {
		int64_t (*f)(struct hist_entry *, struct hist_entry *);

		f = se->collapse ?: se->cmp;

		cmp = f(left, right);
		if (cmp)
			break;
	}

	return cmp;
}

768
static size_t
769
hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
770 771 772 773
{
	struct sort_entry *se;
	size_t ret;

774 775 776
	if (exclude_other && !self->parent)
		return 0;

777
	if (total_samples) {
778 779 780 781
		double percent = self->count * 100.0 / total_samples;
		char *color = PERF_COLOR_NORMAL;

		/*
782 783 784
		 * We color high-overhead entries in red, mid-overhead
		 * entries in green - and keep the low overhead places
		 * normal:
785
		 */
786
		if (percent >= 5.0) {
787
			color = PERF_COLOR_RED;
788 789 790 791
		} else {
			if (percent >= 0.5)
				color = PERF_COLOR_GREEN;
		}
792 793

		ret = color_fprintf(fp, color, "   %6.2f%%",
794 795
				(self->count * 100.0) / total_samples);
	} else
796
		ret = fprintf(fp, "%12Ld ", self->count);
797

798
	list_for_each_entry(se, &hist_entry__sort_list, list) {
799 800 801
		if (exclude_other && (se == &sort_parent))
			continue;

802
		fprintf(fp, "  ");
803
		ret += se->print(fp, self);
804
	}
805 806 807 808 809 810

	ret += fprintf(fp, "\n");

	return ret;
}

811 812 813 814 815 816
/*
 *
 */

static struct symbol *
resolve_symbol(struct thread *thread, struct map **mapp,
817
	       struct dso **dsop, u64 *ipp)
818 819 820
{
	struct dso *dso = dsop ? *dsop : NULL;
	struct map *map = mapp ? *mapp : NULL;
821
	u64 ip = *ipp;
822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850

	if (!thread)
		return NULL;

	if (dso)
		goto got_dso;

	if (map)
		goto got_map;

	map = thread__find_map(thread, ip);
	if (map != NULL) {
		if (mapp)
			*mapp = map;
got_map:
		ip = map->map_ip(map, ip);

		dso = map->dso;
	} else {
		/*
		 * If this is outside of all known maps,
		 * and is a negative address, try to look it
		 * up in the kernel dso, as it might be a
		 * vsyscall (which executes in user-mode):
		 */
		if ((long long)ip < 0)
		dso = kernel_dso;
	}
	dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
851 852
	dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
	*ipp  = ip;
853 854 855 856 857 858 859 860 861 862

	if (dsop)
		*dsop = dso;

	if (!dso)
		return NULL;
got_dso:
	return dso->find_symbol(dso, ip);
}

863
static int call__match(struct symbol *sym)
864
{
865
	if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
866
		return 1;
867

868
	return 0;
869 870
}

871 872 873 874
/*
 * collect histogram counts
 */

875 876
static int
hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
877 878
		struct symbol *sym, u64 ip, struct ip_callchain *chain,
		char level, u64 count)
879
{
880 881 882 883 884 885 886 887 888 889
	struct rb_node **p = &hist.rb_node;
	struct rb_node *parent = NULL;
	struct hist_entry *he;
	struct hist_entry entry = {
		.thread	= thread,
		.map	= map,
		.dso	= dso,
		.sym	= sym,
		.ip	= ip,
		.level	= level,
890
		.count	= count,
891
		.parent = NULL,
892 893 894
	};
	int cmp;

895
	if (sort__has_parent && chain) {
896
		u64 context = PERF_CONTEXT_MAX;
897 898 899
		int i;

		for (i = 0; i < chain->nr; i++) {
900
			u64 ip = chain->ips[i];
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916
			struct dso *dso = NULL;
			struct symbol *sym;

			if (ip >= PERF_CONTEXT_MAX) {
				context = ip;
				continue;
			}

			switch (context) {
			case PERF_CONTEXT_KERNEL:
				dso = kernel_dso;
				break;
			default:
				break;
			}

917
			sym = resolve_symbol(thread, NULL, &dso, &ip);
918 919 920 921 922

			if (sym && call__match(sym)) {
				entry.parent = sym;
				break;
			}
923 924 925
		}
	}

926 927 928 929 930 931 932
	while (*p != NULL) {
		parent = *p;
		he = rb_entry(parent, struct hist_entry, rb_node);

		cmp = hist_entry__cmp(&entry, he);

		if (!cmp) {
933
			he->count += count;
934 935 936 937 938 939 940
			return 0;
		}

		if (cmp < 0)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
941
	}
942 943 944 945 946 947 948 949 950

	he = malloc(sizeof(*he));
	if (!he)
		return -ENOMEM;
	*he = entry;
	rb_link_node(&he->rb_node, parent, p);
	rb_insert_color(&he->rb_node, &hist);

	return 0;
951 952
}

P
Peter Zijlstra 已提交
953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
static void hist_entry__free(struct hist_entry *he)
{
	free(he);
}

/*
 * collapse the histogram
 */

static struct rb_root collapse_hists;

static void collapse__insert_entry(struct hist_entry *he)
{
	struct rb_node **p = &collapse_hists.rb_node;
	struct rb_node *parent = NULL;
	struct hist_entry *iter;
	int64_t cmp;

	while (*p != NULL) {
		parent = *p;
		iter = rb_entry(parent, struct hist_entry, rb_node);

		cmp = hist_entry__collapse(iter, he);

		if (!cmp) {
			iter->count += he->count;
			hist_entry__free(he);
			return;
		}

		if (cmp < 0)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

	rb_link_node(&he->rb_node, parent, p);
	rb_insert_color(&he->rb_node, &collapse_hists);
}

static void collapse__resort(void)
{
	struct rb_node *next;
	struct hist_entry *n;

	if (!sort__need_collapse)
		return;

	next = rb_first(&hist);
	while (next) {
		n = rb_entry(next, struct hist_entry, rb_node);
		next = rb_next(&n->rb_node);

		rb_erase(&n->rb_node, &hist);
		collapse__insert_entry(n);
	}
}

1011 1012 1013 1014 1015 1016 1017
/*
 * reverse the map, sort on count.
 */

static struct rb_root output_hists;

static void output__insert_entry(struct hist_entry *he)
1018
{
1019
	struct rb_node **p = &output_hists.rb_node;
1020
	struct rb_node *parent = NULL;
1021
	struct hist_entry *iter;
1022 1023 1024

	while (*p != NULL) {
		parent = *p;
1025
		iter = rb_entry(parent, struct hist_entry, rb_node);
1026

1027
		if (he->count > iter->count)
1028 1029 1030 1031 1032
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

1033 1034
	rb_link_node(&he->rb_node, parent, p);
	rb_insert_color(&he->rb_node, &output_hists);
1035 1036
}

1037
static void output__resort(void)
1038
{
P
Peter Zijlstra 已提交
1039
	struct rb_node *next;
1040
	struct hist_entry *n;
1041
	struct rb_root *tree = &hist;
1042

P
Peter Zijlstra 已提交
1043
	if (sort__need_collapse)
1044 1045 1046
		tree = &collapse_hists;

	next = rb_first(tree);
P
Peter Zijlstra 已提交
1047

1048 1049 1050
	while (next) {
		n = rb_entry(next, struct hist_entry, rb_node);
		next = rb_next(&n->rb_node);
1051

1052
		rb_erase(&n->rb_node, tree);
1053
		output__insert_entry(n);
1054 1055 1056
	}
}

1057
static size_t output__fprintf(FILE *fp, u64 total_samples)
1058
{
1059
	struct hist_entry *pos;
1060
	struct sort_entry *se;
1061 1062 1063
	struct rb_node *nd;
	size_t ret = 0;

1064
	fprintf(fp, "\n");
1065
	fprintf(fp, "#\n");
1066
	fprintf(fp, "# (%Ld samples)\n", (u64)total_samples);
1067 1068 1069
	fprintf(fp, "#\n");

	fprintf(fp, "# Overhead");
1070 1071 1072
	list_for_each_entry(se, &hist_entry__sort_list, list) {
		if (exclude_other && (se == &sort_parent))
			continue;
1073
		fprintf(fp, "  %s", se->header);
1074
	}
1075 1076 1077
	fprintf(fp, "\n");

	fprintf(fp, "# ........");
1078
	list_for_each_entry(se, &hist_entry__sort_list, list) {
1079 1080
		int i;

1081 1082 1083
		if (exclude_other && (se == &sort_parent))
			continue;

1084
		fprintf(fp, "  ");
1085
		for (i = 0; i < strlen(se->header); i++)
1086
			fprintf(fp, ".");
1087
	}
1088 1089 1090
	fprintf(fp, "\n");

	fprintf(fp, "#\n");
1091

1092 1093 1094
	for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
		pos = rb_entry(nd, struct hist_entry, rb_node);
		ret += hist_entry__fprintf(fp, pos, total_samples);
1095 1096
	}

1097 1098
	if (sort_order == default_sort_order &&
			parent_pattern == default_parent_pattern) {
1099
		fprintf(fp, "#\n");
1100
		fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1101 1102
		fprintf(fp, "#\n");
	}
1103
	fprintf(fp, "\n");
1104

1105 1106 1107
	return ret;
}

1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118
static void register_idle_thread(void)
{
	struct thread *thread = threads__findnew(0);

	if (thread == NULL ||
			thread__set_comm(thread, "[idle]")) {
		fprintf(stderr, "problem inserting idle task.\n");
		exit(-1);
	}
}

1119 1120 1121 1122
static unsigned long total = 0,
		     total_mmap = 0,
		     total_comm = 0,
		     total_fork = 0,
1123 1124
		     total_unknown = 0,
		     total_lost = 0;
1125

1126
static int validate_chain(struct ip_callchain *chain, event_t *event)
1127 1128 1129 1130 1131 1132
{
	unsigned int chain_size;

	chain_size = event->header.size;
	chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;

1133
	if (chain->nr*sizeof(u64) > chain_size)
1134 1135 1136 1137 1138
		return -1;

	return 0;
}

1139
static int
1140
process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1141 1142 1143 1144 1145
{
	char level;
	int show = 0;
	struct dso *dso = NULL;
	struct thread *thread = threads__findnew(event->ip.pid);
1146 1147
	u64 ip = event->ip.ip;
	u64 period = 1;
1148
	struct map *map = NULL;
1149
	void *more_data = event->ip.__more_data;
1150
	struct ip_callchain *chain = NULL;
1151

1152
	if (sample_type & PERF_SAMPLE_PERIOD) {
1153 1154
		period = *(u64 *)more_data;
		more_data += sizeof(u64);
1155
	}
1156

1157
	dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1158 1159 1160 1161
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->header.misc,
		event->ip.pid,
1162
		(void *)(long)ip,
1163
		(long long)period);
1164

1165
	if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1166 1167 1168 1169
		int i;

		chain = (void *)more_data;

1170
		dprintf("... chain: nr:%Lu\n", chain->nr);
1171

1172 1173 1174 1175 1176 1177
		if (validate_chain(chain, event) < 0) {
			eprintf("call-chain problem with event, skipping it.\n");
			return 0;
		}

		if (dump_trace) {
1178
			for (i = 0; i < chain->nr; i++)
1179
				dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1180 1181 1182
		}
	}

1183 1184 1185
	dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);

	if (thread == NULL) {
1186
		eprintf("problem processing %d event, skipping it.\n",
1187 1188 1189
			event->header.type);
		return -1;
	}
1190

1191 1192 1193
	if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
		show = SHOW_KERNEL;
		level = 'k';
1194

1195
		dso = kernel_dso;
1196

1197
		dprintf(" ...... dso: %s\n", dso->name);
1198

1199
	} else if (event->header.misc & PERF_EVENT_MISC_USER) {
1200

1201 1202
		show = SHOW_USER;
		level = '.';
1203

1204 1205 1206 1207 1208
	} else {
		show = SHOW_HV;
		level = 'H';
		dprintf(" ...... dso: [hypervisor]\n");
	}
1209

1210
	if (show & show_mask) {
1211
		struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1212

1213
		if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1214
			eprintf("problem incrementing symbol count, skipping event\n");
1215
			return -1;
1216
		}
1217
	}
1218
	total += period;
1219

1220 1221
	return 0;
}
I
Ingo Molnar 已提交
1222

1223 1224 1225 1226 1227 1228
static int
process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
{
	struct thread *thread = threads__findnew(event->mmap.pid);
	struct map *map = map__new(&event->mmap);

1229
	dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1230 1231
		(void *)(offset + head),
		(void *)(long)(event->header.size),
1232
		event->mmap.pid,
1233 1234 1235 1236 1237 1238 1239
		(void *)(long)event->mmap.start,
		(void *)(long)event->mmap.len,
		(void *)(long)event->mmap.pgoff,
		event->mmap.filename);

	if (thread == NULL || map == NULL) {
		dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1240
		return 0;
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
	}

	thread__insert_map(thread, map);
	total_mmap++;

	return 0;
}

static int
process_comm_event(event_t *event, unsigned long offset, unsigned long head)
{
	struct thread *thread = threads__findnew(event->comm.pid);

	dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->comm.comm, event->comm.pid);

	if (thread == NULL ||
	    thread__set_comm(thread, event->comm.comm)) {
		dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
		return -1;
1263
	}
1264 1265 1266 1267 1268
	total_comm++;

	return 0;
}

1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288
static int
process_fork_event(event_t *event, unsigned long offset, unsigned long head)
{
	struct thread *thread = threads__findnew(event->fork.pid);
	struct thread *parent = threads__findnew(event->fork.ppid);

	dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->fork.pid, event->fork.ppid);

	if (!thread || !parent || thread__fork(thread, parent)) {
		dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
		return -1;
	}
	total_fork++;

	return 0;
}

1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301
static int
process_period_event(event_t *event, unsigned long offset, unsigned long head)
{
	dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->period.time,
		event->period.id,
		event->period.sample_period);

	return 0;
}

1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315
static int
process_lost_event(event_t *event, unsigned long offset, unsigned long head)
{
	dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->lost.id,
		event->lost.lost);

	total_lost += event->lost.lost;

	return 0;
}

1316 1317 1318
static void trace_event(event_t *event)
{
	unsigned char *raw_event = (void *)event;
1319
	char *color = PERF_COLOR_BLUE;
1320 1321 1322 1323 1324
	int i, j;

	if (!dump_trace)
		return;

1325 1326
	dprintf(".");
	cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1327 1328

	for (i = 0; i < event->header.size; i++) {
1329 1330 1331 1332
		if ((i & 15) == 0) {
			dprintf(".");
			cdprintf("  %04x: ", i);
		}
1333

1334
		cdprintf(" %02x", raw_event[i]);
1335 1336

		if (((i & 15) == 15) || i == event->header.size-1) {
1337
			cdprintf("  ");
1338
			for (j = 0; j < 15-(i & 15); j++)
1339
				cdprintf("   ");
1340
			for (j = 0; j < (i & 15); j++) {
1341
				if (isprint(raw_event[i-15+j]))
1342
					cdprintf("%c", raw_event[i-15+j]);
1343
				else
1344
					cdprintf(".");
1345
			}
1346
			cdprintf("\n");
1347 1348 1349 1350 1351
		}
	}
	dprintf(".\n");
}

1352 1353 1354
static int
process_event(event_t *event, unsigned long offset, unsigned long head)
{
1355 1356
	trace_event(event);

1357
	switch (event->header.type) {
1358 1359 1360
	case PERF_EVENT_SAMPLE:
		return process_sample_event(event, offset, head);

1361 1362 1363 1364 1365 1366
	case PERF_EVENT_MMAP:
		return process_mmap_event(event, offset, head);

	case PERF_EVENT_COMM:
		return process_comm_event(event, offset, head);

1367 1368 1369
	case PERF_EVENT_FORK:
		return process_fork_event(event, offset, head);

1370 1371
	case PERF_EVENT_PERIOD:
		return process_period_event(event, offset, head);
1372 1373 1374 1375

	case PERF_EVENT_LOST:
		return process_lost_event(event, offset, head);

1376 1377 1378
	/*
	 * We dont process them right now but they are fine:
	 */
1379

1380 1381 1382 1383
	case PERF_EVENT_THROTTLE:
	case PERF_EVENT_UNTHROTTLE:
		return 0;

1384 1385 1386 1387 1388 1389 1390
	default:
		return -1;
	}

	return 0;
}

1391 1392
static struct perf_header	*header;

1393
static u64 perf_header__sample_type(void)
1394
{
1395
	u64 sample_type = 0;
1396 1397 1398 1399 1400
	int i;

	for (i = 0; i < header->attrs; i++) {
		struct perf_header_attr *attr = header->attr[i];

1401 1402 1403 1404
		if (!sample_type)
			sample_type = attr->attr.sample_type;
		else if (sample_type != attr->attr.sample_type)
			die("non matching sample_type");
1405 1406
	}

1407
	return sample_type;
1408
}
1409

1410 1411
static int __cmd_report(void)
{
1412
	int ret, rc = EXIT_FAILURE;
1413
	unsigned long offset = 0;
1414
	unsigned long head, shift;
1415 1416 1417
	struct stat stat;
	event_t *event;
	uint32_t size;
1418
	char *buf;
1419 1420 1421 1422 1423

	register_idle_thread();

	input = open(input_name, O_RDONLY);
	if (input < 0) {
1424 1425 1426 1427
		fprintf(stderr, " failed to open file: %s", input_name);
		if (!strcmp(input_name, "perf.data"))
			fprintf(stderr, "  (try 'perf record' first)");
		fprintf(stderr, "\n");
1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441
		exit(-1);
	}

	ret = fstat(input, &stat);
	if (ret < 0) {
		perror("failed to stat file");
		exit(-1);
	}

	if (!stat.st_size) {
		fprintf(stderr, "zero-sized file, nothing to do!\n");
		exit(0);
	}

1442 1443
	header = perf_header__read(input);
	head = header->data_offset;
1444

1445 1446 1447
	sample_type = perf_header__sample_type();

	if (sort__has_parent && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1448 1449 1450 1451
		fprintf(stderr, "selected --sort parent, but no callchain data\n");
		exit(-1);
	}

1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466
	if (load_kernel() < 0) {
		perror("failed to load kernel symbols");
		return EXIT_FAILURE;
	}

	if (!full_paths) {
		if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
			perror("failed to get the current directory");
			return EXIT_FAILURE;
		}
		cwdlen = strlen(cwd);
	} else {
		cwd = NULL;
		cwdlen = 0;
	}
1467 1468 1469 1470 1471

	shift = page_size * (head / page_size);
	offset += shift;
	head -= shift;

1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489
remap:
	buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
			   MAP_SHARED, input, offset);
	if (buf == MAP_FAILED) {
		perror("failed to mmap file");
		exit(-1);
	}

more:
	event = (event_t *)(buf + head);

	size = event->header.size;
	if (!size)
		size = 8;

	if (head + event->header.size >= page_size * mmap_window) {
		int ret;

1490 1491
		shift = page_size * (head / page_size);

1492 1493 1494 1495 1496 1497 1498 1499 1500 1501
		ret = munmap(buf, page_size * mmap_window);
		assert(ret == 0);

		offset += shift;
		head -= shift;
		goto remap;
	}

	size = event->header.size;

1502
	dprintf("\n%p [%p]: event: %d\n",
1503 1504 1505 1506
			(void *)(offset + head),
			(void *)(long)event->header.size,
			event->header.type);

1507 1508
	if (!size || process_event(event, offset, head) < 0) {

I
Ingo Molnar 已提交
1509 1510 1511 1512
		dprintf("%p [%p]: skipping unknown header type: %d\n",
			(void *)(offset + head),
			(void *)(long)(event->header.size),
			event->header.type);
1513

1514
		total_unknown++;
1515 1516 1517 1518 1519 1520 1521 1522 1523 1524

		/*
		 * assume we lost track of the stream, check alignment, and
		 * increment a single u64 in the hope to catch on again 'soon'.
		 */

		if (unlikely(head & 7))
			head &= ~7ULL;

		size = 8;
1525
	}
1526

1527
	head += size;
I
Ingo Molnar 已提交
1528

1529
	if (offset + head >= header->data_offset + header->data_size)
1530 1531
		goto done;

1532 1533 1534
	if (offset + head < stat.st_size)
		goto more;

1535
done:
1536 1537
	rc = EXIT_SUCCESS;
	close(input);
1538

I
Ingo Molnar 已提交
1539 1540 1541
	dprintf("      IP events: %10ld\n", total);
	dprintf("    mmap events: %10ld\n", total_mmap);
	dprintf("    comm events: %10ld\n", total_comm);
1542
	dprintf("    fork events: %10ld\n", total_fork);
1543
	dprintf("    lost events: %10ld\n", total_lost);
I
Ingo Molnar 已提交
1544
	dprintf(" unknown events: %10ld\n", total_unknown);
1545

I
Ingo Molnar 已提交
1546
	if (dump_trace)
1547 1548
		return 0;

1549 1550 1551
	if (verbose >= 3)
		threads__fprintf(stdout);

1552
	if (verbose >= 2)
1553 1554
		dsos__fprintf(stdout);

P
Peter Zijlstra 已提交
1555
	collapse__resort();
1556 1557
	output__resort();
	output__fprintf(stdout, total);
1558 1559 1560 1561

	return rc;
}

1562 1563 1564 1565 1566 1567 1568 1569
static const char * const report_usage[] = {
	"perf report [<options>] <command>",
	NULL
};

static const struct option options[] = {
	OPT_STRING('i', "input", &input_name, "file",
		    "input file name"),
1570 1571
	OPT_BOOLEAN('v', "verbose", &verbose,
		    "be more verbose (show symbol address, etc)"),
1572 1573
	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
		    "dump raw trace in ASCII"),
1574
	OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1575
	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1576
		   "sort by key(s): pid, comm, dso, symbol, parent"),
1577 1578
	OPT_BOOLEAN('P', "full-paths", &full_paths,
		    "Don't shorten the pathnames taking into account the cwd"),
1579 1580
	OPT_STRING('p', "parent", &parent_pattern, "regex",
		   "regex filter to identify parent, see: '--sort parent'"),
1581 1582
	OPT_BOOLEAN('x', "exclude-other", &exclude_other,
		    "Only display entries with parent-match"),
1583 1584 1585
	OPT_END()
};

1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600
static void setup_sorting(void)
{
	char *tmp, *tok, *str = strdup(sort_order);

	for (tok = strtok_r(str, ", ", &tmp);
			tok; tok = strtok_r(NULL, ", ", &tmp)) {
		if (sort_dimension__add(tok) < 0) {
			error("Unknown --sort key: `%s'", tok);
			usage_with_options(report_usage, options);
		}
	}

	free(str);
}

1601 1602
int cmd_report(int argc, const char **argv, const char *prefix)
{
1603
	symbol__init();
1604 1605 1606

	page_size = getpagesize();

1607
	argc = parse_options(argc, argv, options, report_usage, 0);
1608

1609 1610
	setup_sorting();

1611 1612 1613 1614 1615
	if (parent_pattern != default_parent_pattern)
		sort_dimension__add("parent");
	else
		exclude_other = 0;

1616 1617 1618 1619 1620 1621
	/*
	 * Any (unrecognized) arguments left?
	 */
	if (argc)
		usage_with_options(report_usage, options);

1622 1623
	setup_pager();

1624 1625
	return __cmd_report();
}