builtin-report.c 29.6 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 20 21 22 23
#include "perf.h"

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

24 25 26 27
#define SHOW_KERNEL	1
#define SHOW_USER	2
#define SHOW_HV		4

28
static char		const *input_name = "perf.data";
29
static char		*vmlinux = NULL;
30 31 32 33

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

34 35 36
static int		input;
static int		show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;

37
static int		dump_trace = 0;
I
Ingo Molnar 已提交
38
#define dprintf(x...)	do { if (dump_trace) printf(x); } while (0)
39
#define cdprintf(x...)	do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
I
Ingo Molnar 已提交
40

41
static int		verbose;
42 43
#define eprintf(x...)	do { if (verbose) fprintf(stderr, x); } while (0)

44
static int		full_paths;
45

46 47 48
static unsigned long	page_size;
static unsigned long	mmap_window = 32;

49 50
static char		*parent_pattern = "^sys_|^do_page_fault";
static regex_t		parent_regex;
51

52 53 54 55
struct ip_event {
	struct perf_event_header header;
	__u64 ip;
	__u32 pid, tid;
56
	unsigned char __more_data[];
57
};
58

59 60 61 62 63 64 65 66
struct mmap_event {
	struct perf_event_header header;
	__u32 pid, tid;
	__u64 start;
	__u64 len;
	__u64 pgoff;
	char filename[PATH_MAX];
};
67

68 69
struct comm_event {
	struct perf_event_header header;
70
	__u32 pid, tid;
71 72 73
	char comm[16];
};

74 75 76 77 78
struct fork_event {
	struct perf_event_header header;
	__u32 pid, ppid;
};

79
struct period_event {
80
	struct perf_event_header header;
81 82 83 84 85 86 87 88 89 90 91 92
	__u64 time;
	__u64 id;
	__u64 sample_period;
};

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;
93 94 95 96
} event_t;

static LIST_HEAD(dsos);
static struct dso *kernel_dso;
P
Peter Zijlstra 已提交
97
static struct dso *vdso;
98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116

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);
117
	int nr;
118

119 120 121 122 123 124
	if (dso)
		return dso;

	dso = dso__new(name, 0);
	if (!dso)
		goto out_delete_dso;
125

126
	nr = dso__load(dso, NULL, verbose);
127
	if (nr < 0) {
128
		eprintf("Failed to open: %s\n", name);
129
		goto out_delete_dso;
130
	}
131 132
	if (!nr)
		eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
133 134

	dsos__add(dso);
135 136 137 138 139 140 141 142

	return dso;

out_delete_dso:
	dso__delete(dso);
	return NULL;
}

143
static void dsos__fprintf(FILE *fp)
144 145 146 147 148 149 150
{
	struct dso *pos;

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

151
static struct symbol *vdso__find_symbol(struct dso *dso, __u64 ip)
P
Peter Zijlstra 已提交
152 153 154 155
{
	return dso__find_symbol(kernel_dso, ip);
}

156 157
static int load_kernel(void)
{
158
	int err;
159

160
	kernel_dso = dso__new("[kernel]", 0);
161
	if (!kernel_dso)
162
		return -1;
163

164
	err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
165 166 167 168 169
	if (err) {
		dso__delete(kernel_dso);
		kernel_dso = NULL;
	} else
		dsos__add(kernel_dso);
170

P
Peter Zijlstra 已提交
171 172 173 174 175 176 177 178
	vdso = dso__new("[vdso]", 0);
	if (!vdso)
		return -1;

	vdso->find_symbol = vdso__find_symbol;

	dsos__add(vdso);

179
	return err;
180 181
}

182 183 184 185 186
static char __cwd[PATH_MAX];
static char *cwd = __cwd;
static int cwdlen;

static int strcommon(const char *pathname)
187 188 189 190 191 192 193 194 195
{
	int n = 0;

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

	return n;
}

196 197
struct map {
	struct list_head node;
198 199 200 201
	__u64	 start;
	__u64	 end;
	__u64	 pgoff;
	__u64	 (*map_ip)(struct map *, __u64);
202 203 204
	struct dso	 *dso;
};

205
static __u64 map__map_ip(struct map *map, __u64 ip)
P
Peter Zijlstra 已提交
206 207 208 209
{
	return ip - map->start + map->pgoff;
}

210
static __u64 vdso__map_ip(struct map *map, __u64 ip)
P
Peter Zijlstra 已提交
211 212 213 214
{
	return ip;
}

215 216 217 218 219
static inline int is_anon_memory(const char *filename)
{
     return strcmp(filename, "//anon") == 0;
}

220
static struct map *map__new(struct mmap_event *event)
221 222 223 224
{
	struct map *self = malloc(sizeof(*self));

	if (self != NULL) {
225 226
		const char *filename = event->filename;
		char newfilename[PATH_MAX];
227
		int anon;
228 229

		if (cwd) {
230 231
			int n = strcommon(filename);

232 233 234 235 236 237 238
			if (n == cwdlen) {
				snprintf(newfilename, sizeof(newfilename),
					 ".%s", filename + n);
				filename = newfilename;
			}
		}

239 240 241 242 243 244 245
		anon = is_anon_memory(filename);

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

246 247 248 249
		self->start = event->start;
		self->end   = event->start + event->len;
		self->pgoff = event->pgoff;

250
		self->dso = dsos__findnew(filename);
251 252
		if (self->dso == NULL)
			goto out_delete;
P
Peter Zijlstra 已提交
253

254
		if (self->dso == vdso || anon)
P
Peter Zijlstra 已提交
255 256 257
			self->map_ip = vdso__map_ip;
		else
			self->map_ip = map__map_ip;
258 259 260 261 262 263 264
	}
	return self;
out_delete:
	free(self);
	return NULL;
}

265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
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;
}
290

291 292
static size_t map__fprintf(struct map *self, FILE *fp)
{
293
	return fprintf(fp, " %Lx-%Lx %Lx %s\n",
294 295 296 297
		       self->start, self->end, self->pgoff, self->dso->name);
}


298
struct thread {
299
	struct rb_node	 rb_node;
300 301 302 303 304 305 306 307 308 309 310
	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
311
		self->comm = malloc(32);
312
		if (self->comm)
P
Peter Zijlstra 已提交
313
			snprintf(self->comm, 32, ":%d", self->pid);
314 315 316 317 318 319 320 321
		INIT_LIST_HEAD(&self->maps);
	}

	return self;
}

static int thread__set_comm(struct thread *self, const char *comm)
{
P
Peter Zijlstra 已提交
322 323
	if (self->comm)
		free(self->comm);
324 325 326 327
	self->comm = strdup(comm);
	return self->comm ? 0 : -ENOMEM;
}

328 329 330 331 332 333 334 335 336 337 338 339
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;
}


340
static struct rb_root threads;
341
static struct thread *last_match;
342

343
static struct thread *threads__findnew(pid_t pid)
344
{
345 346 347
	struct rb_node **p = &threads.rb_node;
	struct rb_node *parent = NULL;
	struct thread *th;
348

349 350 351 352 353 354 355 356
	/*
	 * 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;

357 358 359
	while (*p != NULL) {
		parent = *p;
		th = rb_entry(parent, struct thread, rb_node);
360

361 362
		if (th->pid == pid) {
			last_match = th;
363
			return th;
364
		}
365

366 367 368 369
		if (pid < th->pid)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
370 371
	}

372 373 374 375
	th = thread__new(pid);
	if (th != NULL) {
		rb_link_node(&th->rb_node, parent, p);
		rb_insert_color(&th->rb_node, &threads);
376
		last_match = th;
377
	}
378

379
	return th;
380 381 382 383
}

static void thread__insert_map(struct thread *self, struct map *map)
{
384 385 386 387 388 389 390 391 392 393
	struct map *pos, *tmp;

	list_for_each_entry_safe(pos, tmp, &self->maps, node) {
		if (map__overlap(pos, map)) {
			list_del_init(&pos->node);
			/* XXX leaks dsos */
			free(pos);
		}
	}

394 395 396
	list_add_tail(&map->node, &self->maps);
}

397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416
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;
}

417
static struct map *thread__find_map(struct thread *self, __u64 ip)
418
{
419 420
	struct map *pos;

421 422 423 424 425 426 427 428 429 430
	if (self == NULL)
		return NULL;

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

	return NULL;
}

431 432 433 434 435 436 437 438 439 440 441 442 443 444
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;
}

445 446 447 448 449 450 451 452 453 454 455 456 457
/*
 * 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;
458
	struct symbol	 *parent;
459
	__u64		 ip;
460 461
	char		 level;

462
	__u64		 count;
463 464
};

465 466 467 468 469 470 471
/*
 * configurable sorting bits
 */

struct sort_entry {
	struct list_head list;

472 473
	char *header;

474
	int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
P
Peter Zijlstra 已提交
475
	int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
476 477 478
	size_t	(*print)(FILE *fp, struct hist_entry *);
};

479 480 481 482 483 484 485 486 487 488
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
489 490
/* --sort pid */

491
static int64_t
492
sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
493
{
494 495 496 497 498 499
	return right->thread->pid - left->thread->pid;
}

static size_t
sort__thread_print(FILE *fp, struct hist_entry *self)
{
500
	return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
501
}
502

503
static struct sort_entry sort_thread = {
504
	.header = "         Command:  Pid",
505 506 507 508
	.cmp	= sort__thread_cmp,
	.print	= sort__thread_print,
};

P
Peter Zijlstra 已提交
509 510
/* --sort comm */

511 512
static int64_t
sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
P
Peter Zijlstra 已提交
513 514 515 516 517 518
{
	return right->thread->pid - left->thread->pid;
}

static int64_t
sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
519 520 521 522
{
	char *comm_l = left->thread->comm;
	char *comm_r = right->thread->comm;

523 524
	if (!comm_l || !comm_r)
		return cmp_null(comm_l, comm_r);
525 526 527 528 529 530 531

	return strcmp(comm_l, comm_r);
}

static size_t
sort__comm_print(FILE *fp, struct hist_entry *self)
{
532
	return fprintf(fp, "%16s", self->thread->comm);
533 534 535
}

static struct sort_entry sort_comm = {
536
	.header		= "         Command",
P
Peter Zijlstra 已提交
537 538 539
	.cmp		= sort__comm_cmp,
	.collapse	= sort__comm_collapse,
	.print		= sort__comm_print,
540 541
};

P
Peter Zijlstra 已提交
542 543
/* --sort dso */

544 545 546 547 548 549
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;

550 551
	if (!dso_l || !dso_r)
		return cmp_null(dso_l, dso_r);
552 553 554 555 556 557 558

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

static size_t
sort__dso_print(FILE *fp, struct hist_entry *self)
{
559
	if (self->dso)
560
		return fprintf(fp, "%-25s", self->dso->name);
561

562
	return fprintf(fp, "%016llx         ", (__u64)self->ip);
563 564 565
}

static struct sort_entry sort_dso = {
566
	.header = "Shared Object            ",
567 568 569 570
	.cmp	= sort__dso_cmp,
	.print	= sort__dso_print,
};

P
Peter Zijlstra 已提交
571 572
/* --sort symbol */

573 574 575
static int64_t
sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
{
576
	__u64 ip_l, ip_r;
577 578 579 580 581 582 583 584 585 586

	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);
}

587 588 589 590 591 592
static size_t
sort__sym_print(FILE *fp, struct hist_entry *self)
{
	size_t ret = 0;

	if (verbose)
593
		ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
594

595 596 597 598
	if (self->sym) {
		ret += fprintf(fp, "[%c] %s",
			self->dso == kernel_dso ? 'k' : '.', self->sym->name);
	} else {
599
		ret += fprintf(fp, "%#016llx", (__u64)self->ip);
600
	}
601 602 603 604 605

	return ret;
}

static struct sort_entry sort_sym = {
606
	.header = "Symbol",
607 608
	.cmp	= sort__sym_cmp,
	.print	= sort__sym_print,
609 610
};

611
/* --sort parent */
612 613

static int64_t
614
sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
615
{
616 617
	struct symbol *sym_l = left->parent;
	struct symbol *sym_r = right->parent;
618 619 620 621 622 623 624 625

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

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

static size_t
626
sort__parent_print(FILE *fp, struct hist_entry *self)
627 628 629
{
	size_t ret = 0;

630
	ret += fprintf(fp, "%-20s", self->parent ? self->parent->name : "[other]");
631 632 633 634

	return ret;
}

635 636 637 638
static struct sort_entry sort_parent = {
	.header = "Parent symbol       ",
	.cmp	= sort__parent_cmp,
	.print	= sort__parent_print,
639 640
};

P
Peter Zijlstra 已提交
641
static int sort__need_collapse = 0;
642
static int sort__has_parent = 0;
P
Peter Zijlstra 已提交
643

644
struct sort_dimension {
645 646 647
	char			*name;
	struct sort_entry	*entry;
	int			taken;
648 649 650 651
};

static struct sort_dimension sort_dimensions[] = {
	{ .name = "pid",	.entry = &sort_thread,	},
652
	{ .name = "comm",	.entry = &sort_comm,	},
653
	{ .name = "dso",	.entry = &sort_dso,	},
654
	{ .name = "symbol",	.entry = &sort_sym,	},
655
	{ .name = "parent",	.entry = &sort_parent,	},
656 657
};

658 659
static LIST_HEAD(hist_entry__sort_list);

660 661 662 663 664 665 666 667 668 669
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;

670
		if (strncasecmp(tok, sd->name, strlen(tok)))
671 672
			continue;

P
Peter Zijlstra 已提交
673 674 675
		if (sd->entry->collapse)
			sort__need_collapse = 1;

676 677
		if (sd->entry == &sort_parent) {
			int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
678 679 680
			if (ret) {
				char err[BUFSIZ];

681 682 683
				regerror(ret, &parent_regex, err, sizeof(err));
				fprintf(stderr, "Invalid regex: %s\n%s",
					parent_pattern, err);
684 685
				exit(-1);
			}
686
			sort__has_parent = 1;
687 688
		}

689 690
		list_add_tail(&sd->entry->list, &hist_entry__sort_list);
		sd->taken = 1;
691

692 693 694 695 696 697
		return 0;
	}

	return -ESRCH;
}

698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
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
713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731
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;
}

732
static size_t
733
hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples)
734 735 736 737 738
{
	struct sort_entry *se;
	size_t ret;

	if (total_samples) {
739 740 741 742
		double percent = self->count * 100.0 / total_samples;
		char *color = PERF_COLOR_NORMAL;

		/*
743 744 745
		 * We color high-overhead entries in red, mid-overhead
		 * entries in green - and keep the low overhead places
		 * normal:
746
		 */
747
		if (percent >= 5.0) {
748
			color = PERF_COLOR_RED;
749 750 751 752
		} else {
			if (percent >= 0.5)
				color = PERF_COLOR_GREEN;
		}
753 754

		ret = color_fprintf(fp, color, "   %6.2f%%",
755 756
				(self->count * 100.0) / total_samples);
	} else
757
		ret = fprintf(fp, "%12Ld ", self->count);
758

759 760
	list_for_each_entry(se, &hist_entry__sort_list, list) {
		fprintf(fp, "  ");
761
		ret += se->print(fp, self);
762
	}
763 764 765 766 767 768

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

	return ret;
}

769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
/*
 *
 */

static struct symbol *
resolve_symbol(struct thread *thread, struct map **mapp,
	       struct dso **dsop, __u64 *ipp)
{
	struct dso *dso = dsop ? *dsop : NULL;
	struct map *map = mapp ? *mapp : NULL;
	uint64_t ip = *ipp;

	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);
		*ipp  = 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>");

	if (dsop)
		*dsop = dso;

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

static struct symbol *call__match(struct symbol *sym)
{
	if (!sym)
		return NULL;

825
	if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
826 827 828 829 830
		return sym;

	return NULL;
}

831 832 833 834
/*
 * collect histogram counts
 */

835 836
static int
hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
837
		struct symbol *sym, __u64 ip, struct perf_callchain_entry *chain,
838
		char level, __u64 count)
839
{
840 841 842 843 844 845 846 847 848 849
	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,
850
		.count	= count,
851 852 853
	};
	int cmp;

854
	if (sort__has_parent && chain) {
855 856 857 858 859 860
		int i, nr = chain->hv;
		struct symbol *sym;
		struct dso *dso;
		__u64 ip;

		for (i = 0; i < chain->kernel; i++) {
861
			ip = chain->ip[nr + i];
862 863
			dso = kernel_dso;
			sym = resolve_symbol(thread, NULL, &dso, &ip);
864 865 866
			entry.parent = call__match(sym);
			if (entry.parent)
				goto got_parent;
867 868 869 870
		}
		nr += i;

		for (i = 0; i < chain->user; i++) {
871
			ip = chain->ip[nr + i];
872
			sym = resolve_symbol(thread, NULL, NULL, &ip);
873 874 875
			entry.parent = call__match(sym);
			if (entry.parent)
				goto got_parent;
876 877 878
		}
		nr += i;
	}
879
got_parent:
880

881 882 883 884 885 886 887
	while (*p != NULL) {
		parent = *p;
		he = rb_entry(parent, struct hist_entry, rb_node);

		cmp = hist_entry__cmp(&entry, he);

		if (!cmp) {
888
			he->count += count;
889 890 891 892 893 894 895
			return 0;
		}

		if (cmp < 0)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
896
	}
897 898 899 900 901 902 903 904 905

	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;
906 907
}

P
Peter Zijlstra 已提交
908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965
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);
	}
}

966 967 968 969 970 971 972
/*
 * reverse the map, sort on count.
 */

static struct rb_root output_hists;

static void output__insert_entry(struct hist_entry *he)
973
{
974
	struct rb_node **p = &output_hists.rb_node;
975
	struct rb_node *parent = NULL;
976
	struct hist_entry *iter;
977 978 979

	while (*p != NULL) {
		parent = *p;
980
		iter = rb_entry(parent, struct hist_entry, rb_node);
981

982
		if (he->count > iter->count)
983 984 985 986 987
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

988 989
	rb_link_node(&he->rb_node, parent, p);
	rb_insert_color(&he->rb_node, &output_hists);
990 991
}

992
static void output__resort(void)
993
{
P
Peter Zijlstra 已提交
994
	struct rb_node *next;
995
	struct hist_entry *n;
996
	struct rb_root *tree = &hist;
997

P
Peter Zijlstra 已提交
998
	if (sort__need_collapse)
999 1000 1001
		tree = &collapse_hists;

	next = rb_first(tree);
P
Peter Zijlstra 已提交
1002

1003 1004 1005
	while (next) {
		n = rb_entry(next, struct hist_entry, rb_node);
		next = rb_next(&n->rb_node);
1006

1007
		rb_erase(&n->rb_node, tree);
1008
		output__insert_entry(n);
1009 1010 1011
	}
}

1012
static size_t output__fprintf(FILE *fp, __u64 total_samples)
1013
{
1014
	struct hist_entry *pos;
1015
	struct sort_entry *se;
1016 1017 1018
	struct rb_node *nd;
	size_t ret = 0;

1019
	fprintf(fp, "\n");
1020
	fprintf(fp, "#\n");
1021
	fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
1022 1023 1024 1025
	fprintf(fp, "#\n");

	fprintf(fp, "# Overhead");
	list_for_each_entry(se, &hist_entry__sort_list, list)
1026
		fprintf(fp, "  %s", se->header);
1027 1028 1029
	fprintf(fp, "\n");

	fprintf(fp, "# ........");
1030
	list_for_each_entry(se, &hist_entry__sort_list, list) {
1031 1032
		int i;

1033
		fprintf(fp, "  ");
1034
		for (i = 0; i < strlen(se->header); i++)
1035
			fprintf(fp, ".");
1036
	}
1037 1038 1039
	fprintf(fp, "\n");

	fprintf(fp, "#\n");
1040

1041 1042 1043
	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);
1044 1045
	}

1046 1047
	if (!strcmp(sort_order, default_sort_order)) {
		fprintf(fp, "#\n");
1048
		fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1049 1050
		fprintf(fp, "#\n");
	}
1051
	fprintf(fp, "\n");
1052

1053 1054 1055
	return ret;
}

1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
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);
	}
}

1067 1068 1069 1070 1071
static unsigned long total = 0,
		     total_mmap = 0,
		     total_comm = 0,
		     total_fork = 0,
		     total_unknown = 0;
1072

1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096
static int validate_chain(struct perf_callchain_entry *chain, event_t *event)
{
	unsigned int chain_size;

	if (chain->nr > MAX_STACK_DEPTH)
		return -1;
	if (chain->hv > MAX_STACK_DEPTH)
		return -1;
	if (chain->kernel > MAX_STACK_DEPTH)
		return -1;
	if (chain->user > MAX_STACK_DEPTH)
		return -1;
	if (chain->hv + chain->kernel + chain->user != chain->nr)
		return -1;

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

	if (chain->nr*sizeof(__u64) > chain_size)
		return -1;

	return 0;
}

1097
static int
1098 1099 1100 1101 1102 1103
process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
{
	char level;
	int show = 0;
	struct dso *dso = NULL;
	struct thread *thread = threads__findnew(event->ip.pid);
1104 1105
	__u64 ip = event->ip.ip;
	__u64 period = 1;
1106
	struct map *map = NULL;
1107
	void *more_data = event->ip.__more_data;
1108
	struct perf_callchain_entry *chain = NULL;
1109

1110 1111 1112 1113
	if (event->header.type & PERF_SAMPLE_PERIOD) {
		period = *(__u64 *)more_data;
		more_data += sizeof(__u64);
	}
1114

1115
	dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p period: %Ld\n",
1116 1117 1118 1119
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->header.misc,
		event->ip.pid,
1120
		(void *)(long)ip,
1121
		(long long)period);
1122

1123 1124 1125 1126 1127
	if (event->header.type & PERF_SAMPLE_CALLCHAIN) {
		int i;

		chain = (void *)more_data;

1128 1129 1130 1131
		dprintf("... chain: u:%d, k:%d, nr:%d\n",
			chain->user,
			chain->kernel,
			chain->nr);
1132

1133 1134 1135 1136 1137 1138
		if (validate_chain(chain, event) < 0) {
			eprintf("call-chain problem with event, skipping it.\n");
			return 0;
		}

		if (dump_trace) {
1139
			for (i = 0; i < chain->nr; i++)
1140
				dprintf("..... %2d: %016Lx\n", i, chain->ip[i]);
1141 1142 1143
		}
	}

1144 1145 1146
	dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);

	if (thread == NULL) {
1147
		eprintf("problem processing %d event, skipping it.\n",
1148 1149 1150
			event->header.type);
		return -1;
	}
1151

1152 1153 1154
	if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
		show = SHOW_KERNEL;
		level = 'k';
1155

1156
		dso = kernel_dso;
1157

1158
		dprintf(" ...... dso: %s\n", dso->name);
1159

1160
	} else if (event->header.misc & PERF_EVENT_MISC_USER) {
1161

1162 1163
		show = SHOW_USER;
		level = '.';
1164

1165 1166 1167 1168 1169
	} else {
		show = SHOW_HV;
		level = 'H';
		dprintf(" ...... dso: [hypervisor]\n");
	}
1170

1171
	if (show & show_mask) {
1172
		struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1173

1174
		if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1175
			eprintf("problem incrementing symbol count, skipping event\n");
1176
			return -1;
1177
		}
1178
	}
1179
	total += period;
1180

1181 1182
	return 0;
}
I
Ingo Molnar 已提交
1183

1184 1185 1186 1187 1188 1189
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);

1190
	dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1191 1192
		(void *)(offset + head),
		(void *)(long)(event->header.size),
1193
		event->mmap.pid,
1194 1195 1196 1197 1198 1199 1200
		(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");
1201
		return 0;
1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223
	}

	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;
1224
	}
1225 1226 1227 1228 1229
	total_comm++;

	return 0;
}

1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249
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;
}

1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
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;
}

1263 1264 1265
static void trace_event(event_t *event)
{
	unsigned char *raw_event = (void *)event;
1266
	char *color = PERF_COLOR_BLUE;
1267 1268 1269 1270 1271
	int i, j;

	if (!dump_trace)
		return;

1272 1273
	dprintf(".");
	cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1274 1275

	for (i = 0; i < event->header.size; i++) {
1276 1277 1278 1279
		if ((i & 15) == 0) {
			dprintf(".");
			cdprintf("  %04x: ", i);
		}
1280

1281
		cdprintf(" %02x", raw_event[i]);
1282 1283

		if (((i & 15) == 15) || i == event->header.size-1) {
1284
			cdprintf("  ");
1285
			for (j = 0; j < 15-(i & 15); j++)
1286
				cdprintf("   ");
1287
			for (j = 0; j < (i & 15); j++) {
1288
				if (isprint(raw_event[i-15+j]))
1289
					cdprintf("%c", raw_event[i-15+j]);
1290
				else
1291
					cdprintf(".");
1292
			}
1293
			cdprintf("\n");
1294 1295 1296 1297 1298
		}
	}
	dprintf(".\n");
}

1299 1300 1301
static int
process_event(event_t *event, unsigned long offset, unsigned long head)
{
1302 1303
	trace_event(event);

1304 1305 1306 1307 1308 1309 1310 1311 1312 1313
	if (event->header.misc & PERF_EVENT_MISC_OVERFLOW)
		return process_overflow_event(event, offset, head);

	switch (event->header.type) {
	case PERF_EVENT_MMAP:
		return process_mmap_event(event, offset, head);

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

1314 1315 1316
	case PERF_EVENT_FORK:
		return process_fork_event(event, offset, head);

1317 1318
	case PERF_EVENT_PERIOD:
		return process_period_event(event, offset, head);
1319 1320 1321
	/*
	 * We dont process them right now but they are fine:
	 */
1322

1323 1324 1325 1326
	case PERF_EVENT_THROTTLE:
	case PERF_EVENT_UNTHROTTLE:
		return 0;

1327 1328 1329 1330 1331 1332 1333 1334 1335
	default:
		return -1;
	}

	return 0;
}

static int __cmd_report(void)
{
1336
	int ret, rc = EXIT_FAILURE;
1337 1338 1339 1340 1341
	unsigned long offset = 0;
	unsigned long head = 0;
	struct stat stat;
	event_t *event;
	uint32_t size;
1342
	char *buf;
1343 1344 1345 1346 1347

	register_idle_thread();

	input = open(input_name, O_RDONLY);
	if (input < 0) {
1348 1349 1350 1351
		fprintf(stderr, " failed to open file: %s", input_name);
		if (!strcmp(input_name, "perf.data"))
			fprintf(stderr, "  (try 'perf record' first)");
		fprintf(stderr, "\n");
1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
		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);
	}

	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;
	}
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) {
		unsigned long shift = page_size * (head / page_size);
		int ret;

		ret = munmap(buf, page_size * mmap_window);
		assert(ret == 0);

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

	size = event->header.size;

1410
	dprintf("\n%p [%p]: event: %d\n",
1411 1412 1413 1414
			(void *)(offset + head),
			(void *)(long)event->header.size,
			event->header.type);

1415 1416
	if (!size || process_event(event, offset, head) < 0) {

I
Ingo Molnar 已提交
1417 1418 1419 1420
		dprintf("%p [%p]: skipping unknown header type: %d\n",
			(void *)(offset + head),
			(void *)(long)(event->header.size),
			event->header.type);
1421

1422
		total_unknown++;
1423 1424 1425 1426 1427 1428 1429 1430 1431 1432

		/*
		 * 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;
1433
	}
1434

1435
	head += size;
I
Ingo Molnar 已提交
1436

1437 1438 1439 1440 1441
	if (offset + head < stat.st_size)
		goto more;

	rc = EXIT_SUCCESS;
	close(input);
1442

I
Ingo Molnar 已提交
1443 1444 1445
	dprintf("      IP events: %10ld\n", total);
	dprintf("    mmap events: %10ld\n", total_mmap);
	dprintf("    comm events: %10ld\n", total_comm);
1446
	dprintf("    fork events: %10ld\n", total_fork);
I
Ingo Molnar 已提交
1447
	dprintf(" unknown events: %10ld\n", total_unknown);
1448

I
Ingo Molnar 已提交
1449
	if (dump_trace)
1450 1451
		return 0;

1452 1453 1454
	if (verbose >= 3)
		threads__fprintf(stdout);

1455
	if (verbose >= 2)
1456 1457
		dsos__fprintf(stdout);

P
Peter Zijlstra 已提交
1458
	collapse__resort();
1459 1460
	output__resort();
	output__fprintf(stdout, total);
1461 1462 1463 1464

	return rc;
}

1465 1466 1467 1468 1469 1470 1471 1472
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"),
1473 1474
	OPT_BOOLEAN('v', "verbose", &verbose,
		    "be more verbose (show symbol address, etc)"),
1475 1476
	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
		    "dump raw trace in ASCII"),
1477
	OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1478
	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1479
		   "sort by key(s): pid, comm, dso, symbol, parent"),
1480 1481
	OPT_BOOLEAN('P', "full-paths", &full_paths,
		    "Don't shorten the pathnames taking into account the cwd"),
1482 1483
	OPT_STRING('p', "parent", &parent_pattern, "regex",
		   "regex filter to identify parent, see: '--sort parent'"),
1484 1485 1486
	OPT_END()
};

1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501
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);
}

1502 1503
int cmd_report(int argc, const char **argv, const char *prefix)
{
1504
	symbol__init();
1505 1506 1507

	page_size = getpagesize();

1508
	argc = parse_options(argc, argv, options, report_usage, 0);
1509

1510 1511
	setup_sorting();

1512 1513 1514 1515 1516 1517
	/*
	 * Any (unrecognized) arguments left?
	 */
	if (argc)
		usage_with_options(report_usage, options);

1518 1519
	setup_pager();

1520 1521
	return __cmd_report();
}