builtin-report.c 28.9 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
static int		full_paths;
43

44 45 46
static unsigned long	page_size;
static unsigned long	mmap_window = 32;

47 48 49
static char		*call = "^sys_";
static regex_t		call_regex;

50 51 52 53 54 55 56 57
struct ip_chain_event {
	__u16 nr;
	__u16 hv;
	__u16 kernel;
	__u16 user;
	__u64 ips[];
};

58 59 60 61
struct ip_event {
	struct perf_event_header header;
	__u64 ip;
	__u32 pid, tid;
62
	unsigned char __more_data[];
63
};
64

65 66 67 68 69 70 71 72
struct mmap_event {
	struct perf_event_header header;
	__u32 pid, tid;
	__u64 start;
	__u64 len;
	__u64 pgoff;
	char filename[PATH_MAX];
};
73

74 75
struct comm_event {
	struct perf_event_header header;
76
	__u32 pid, tid;
77 78 79
	char comm[16];
};

80 81 82 83 84
struct fork_event {
	struct perf_event_header header;
	__u32 pid, ppid;
};

85
struct period_event {
86
	struct perf_event_header header;
87 88 89 90 91 92 93 94 95 96 97 98
	__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;
99 100 101 102
} event_t;

static LIST_HEAD(dsos);
static struct dso *kernel_dso;
P
Peter Zijlstra 已提交
103
static struct dso *vdso;
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122

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);
123
	int nr;
124

125 126 127 128 129 130
	if (dso)
		return dso;

	dso = dso__new(name, 0);
	if (!dso)
		goto out_delete_dso;
131

132
	nr = dso__load(dso, NULL, verbose);
133
	if (nr < 0) {
134 135
		if (verbose)
			fprintf(stderr, "Failed to open: %s\n", name);
136
		goto out_delete_dso;
137
	}
138 139 140 141 142 143 144
	if (!nr && verbose) {
		fprintf(stderr,
		"No symbols found in: %s, maybe install a debug package?\n",
				name);
	}

	dsos__add(dso);
145 146 147 148 149 150 151 152

	return dso;

out_delete_dso:
	dso__delete(dso);
	return NULL;
}

153
static void dsos__fprintf(FILE *fp)
154 155 156 157 158 159 160
{
	struct dso *pos;

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

161
static struct symbol *vdso__find_symbol(struct dso *dso, __u64 ip)
P
Peter Zijlstra 已提交
162 163 164 165
{
	return dso__find_symbol(kernel_dso, ip);
}

166 167
static int load_kernel(void)
{
168
	int err;
169

170
	kernel_dso = dso__new("[kernel]", 0);
171
	if (!kernel_dso)
172
		return -1;
173

174
	err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
175 176 177 178 179
	if (err) {
		dso__delete(kernel_dso);
		kernel_dso = NULL;
	} else
		dsos__add(kernel_dso);
180

P
Peter Zijlstra 已提交
181 182 183 184 185 186 187 188
	vdso = dso__new("[vdso]", 0);
	if (!vdso)
		return -1;

	vdso->find_symbol = vdso__find_symbol;

	dsos__add(vdso);

189
	return err;
190 191
}

192 193 194 195 196
static char __cwd[PATH_MAX];
static char *cwd = __cwd;
static int cwdlen;

static int strcommon(const char *pathname)
197 198 199 200 201 202 203 204 205
{
	int n = 0;

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

	return n;
}

206 207
struct map {
	struct list_head node;
208 209 210 211
	__u64	 start;
	__u64	 end;
	__u64	 pgoff;
	__u64	 (*map_ip)(struct map *, __u64);
212 213 214
	struct dso	 *dso;
};

215
static __u64 map__map_ip(struct map *map, __u64 ip)
P
Peter Zijlstra 已提交
216 217 218 219
{
	return ip - map->start + map->pgoff;
}

220
static __u64 vdso__map_ip(struct map *map, __u64 ip)
P
Peter Zijlstra 已提交
221 222 223 224
{
	return ip;
}

225 226 227 228 229
static inline int is_anon_memory(const char *filename)
{
     return strcmp(filename, "//anon") == 0;
}

230
static struct map *map__new(struct mmap_event *event)
231 232 233 234
{
	struct map *self = malloc(sizeof(*self));

	if (self != NULL) {
235 236
		const char *filename = event->filename;
		char newfilename[PATH_MAX];
237
		int anon;
238 239

		if (cwd) {
240 241
			int n = strcommon(filename);

242 243 244 245 246 247 248
			if (n == cwdlen) {
				snprintf(newfilename, sizeof(newfilename),
					 ".%s", filename + n);
				filename = newfilename;
			}
		}

249 250 251 252 253 254 255
		anon = is_anon_memory(filename);

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

256 257 258 259
		self->start = event->start;
		self->end   = event->start + event->len;
		self->pgoff = event->pgoff;

260
		self->dso = dsos__findnew(filename);
261 262
		if (self->dso == NULL)
			goto out_delete;
P
Peter Zijlstra 已提交
263

264
		if (self->dso == vdso || anon)
P
Peter Zijlstra 已提交
265 266 267
			self->map_ip = vdso__map_ip;
		else
			self->map_ip = map__map_ip;
268 269 270 271 272 273 274
	}
	return self;
out_delete:
	free(self);
	return NULL;
}

275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
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;
}
300

301 302
static size_t map__fprintf(struct map *self, FILE *fp)
{
303
	return fprintf(fp, " %Lx-%Lx %Lx %s\n",
304 305 306 307
		       self->start, self->end, self->pgoff, self->dso->name);
}


308
struct thread {
309
	struct rb_node	 rb_node;
310 311 312 313 314 315 316 317 318 319 320
	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
321
		self->comm = malloc(32);
322
		if (self->comm)
P
Peter Zijlstra 已提交
323
			snprintf(self->comm, 32, ":%d", self->pid);
324 325 326 327 328 329 330 331
		INIT_LIST_HEAD(&self->maps);
	}

	return self;
}

static int thread__set_comm(struct thread *self, const char *comm)
{
P
Peter Zijlstra 已提交
332 333
	if (self->comm)
		free(self->comm);
334 335 336 337
	self->comm = strdup(comm);
	return self->comm ? 0 : -ENOMEM;
}

338 339 340 341 342 343 344 345 346 347 348 349
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;
}


350
static struct rb_root threads;
351
static struct thread *last_match;
352

353
static struct thread *threads__findnew(pid_t pid)
354
{
355 356 357
	struct rb_node **p = &threads.rb_node;
	struct rb_node *parent = NULL;
	struct thread *th;
358

359 360 361 362 363 364 365 366
	/*
	 * 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;

367 368 369
	while (*p != NULL) {
		parent = *p;
		th = rb_entry(parent, struct thread, rb_node);
370

371 372
		if (th->pid == pid) {
			last_match = th;
373
			return th;
374
		}
375

376 377 378 379
		if (pid < th->pid)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
380 381
	}

382 383 384 385
	th = thread__new(pid);
	if (th != NULL) {
		rb_link_node(&th->rb_node, parent, p);
		rb_insert_color(&th->rb_node, &threads);
386
		last_match = th;
387
	}
388

389
	return th;
390 391 392 393
}

static void thread__insert_map(struct thread *self, struct map *map)
{
394 395 396 397 398 399 400 401 402 403
	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);
		}
	}

404 405 406
	list_add_tail(&map->node, &self->maps);
}

407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
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;
}

427
static struct map *thread__find_map(struct thread *self, __u64 ip)
428
{
429 430
	struct map *pos;

431 432 433 434 435 436 437 438 439 440
	if (self == NULL)
		return NULL;

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

	return NULL;
}

441 442 443 444 445 446 447 448 449 450 451 452 453 454
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;
}

455 456 457 458 459 460 461 462 463 464 465 466 467
/*
 * 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;
468
	struct symbol	 *call;
469
	__u64		 ip;
470 471
	char		 level;

472
	__u64		 count;
473 474
};

475 476 477 478 479 480 481
/*
 * configurable sorting bits
 */

struct sort_entry {
	struct list_head list;

482 483
	char *header;

484
	int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
P
Peter Zijlstra 已提交
485
	int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
486 487 488
	size_t	(*print)(FILE *fp, struct hist_entry *);
};

489 490 491 492 493 494 495 496 497 498
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
499 500
/* --sort pid */

501
static int64_t
502
sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
503
{
504 505 506 507 508 509
	return right->thread->pid - left->thread->pid;
}

static size_t
sort__thread_print(FILE *fp, struct hist_entry *self)
{
510
	return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
511
}
512

513
static struct sort_entry sort_thread = {
514
	.header = "         Command:  Pid",
515 516 517 518
	.cmp	= sort__thread_cmp,
	.print	= sort__thread_print,
};

P
Peter Zijlstra 已提交
519 520
/* --sort comm */

521 522
static int64_t
sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
P
Peter Zijlstra 已提交
523 524 525 526 527 528
{
	return right->thread->pid - left->thread->pid;
}

static int64_t
sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
529 530 531 532
{
	char *comm_l = left->thread->comm;
	char *comm_r = right->thread->comm;

533 534
	if (!comm_l || !comm_r)
		return cmp_null(comm_l, comm_r);
535 536 537 538 539 540 541

	return strcmp(comm_l, comm_r);
}

static size_t
sort__comm_print(FILE *fp, struct hist_entry *self)
{
542
	return fprintf(fp, "%16s", self->thread->comm);
543 544 545
}

static struct sort_entry sort_comm = {
546
	.header		= "         Command",
P
Peter Zijlstra 已提交
547 548 549
	.cmp		= sort__comm_cmp,
	.collapse	= sort__comm_collapse,
	.print		= sort__comm_print,
550 551
};

P
Peter Zijlstra 已提交
552 553
/* --sort dso */

554 555 556 557 558 559
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;

560 561
	if (!dso_l || !dso_r)
		return cmp_null(dso_l, dso_r);
562 563 564 565 566 567 568

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

static size_t
sort__dso_print(FILE *fp, struct hist_entry *self)
{
569
	if (self->dso)
570
		return fprintf(fp, "%-25s", self->dso->name);
571

572
	return fprintf(fp, "%016llx         ", (__u64)self->ip);
573 574 575
}

static struct sort_entry sort_dso = {
576
	.header = "Shared Object            ",
577 578 579 580
	.cmp	= sort__dso_cmp,
	.print	= sort__dso_print,
};

P
Peter Zijlstra 已提交
581 582
/* --sort symbol */

583 584 585
static int64_t
sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
{
586
	__u64 ip_l, ip_r;
587 588 589 590 591 592 593 594 595 596

	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);
}

597 598 599 600 601 602
static size_t
sort__sym_print(FILE *fp, struct hist_entry *self)
{
	size_t ret = 0;

	if (verbose)
603
		ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
604

605 606 607 608
	if (self->sym) {
		ret += fprintf(fp, "[%c] %s",
			self->dso == kernel_dso ? 'k' : '.', self->sym->name);
	} else {
609
		ret += fprintf(fp, "%#016llx", (__u64)self->ip);
610
	}
611 612 613 614 615

	return ret;
}

static struct sort_entry sort_sym = {
616
	.header = "Symbol",
617 618
	.cmp	= sort__sym_cmp,
	.print	= sort__sym_print,
619 620
};

621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650
/* --sort call */

static int64_t
sort__call_cmp(struct hist_entry *left, struct hist_entry *right)
{
	struct symbol *sym_l = left->call;
	struct symbol *sym_r = right->call;

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

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

static size_t
sort__call_print(FILE *fp, struct hist_entry *self)
{
	size_t ret = 0;

	ret += fprintf(fp, "%-20s", self->call ? self->call->name : "[unmatched]");

	return ret;
}

static struct sort_entry sort_call = {
	.header = "Callchain symbol    ",
	.cmp	= sort__call_cmp,
	.print	= sort__call_print,
};

P
Peter Zijlstra 已提交
651
static int sort__need_collapse = 0;
652
static int sort__has_call = 0;
P
Peter Zijlstra 已提交
653

654
struct sort_dimension {
655 656 657
	char			*name;
	struct sort_entry	*entry;
	int			taken;
658 659 660 661
};

static struct sort_dimension sort_dimensions[] = {
	{ .name = "pid",	.entry = &sort_thread,	},
662
	{ .name = "comm",	.entry = &sort_comm,	},
663
	{ .name = "dso",	.entry = &sort_dso,	},
664
	{ .name = "symbol",	.entry = &sort_sym,	},
665
	{ .name = "call",	.entry = &sort_call,	},
666 667
};

668 669
static LIST_HEAD(hist_entry__sort_list);

670 671 672 673 674 675 676 677 678 679
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;

680
		if (strncasecmp(tok, sd->name, strlen(tok)))
681 682
			continue;

P
Peter Zijlstra 已提交
683 684 685
		if (sd->entry->collapse)
			sort__need_collapse = 1;

686 687 688 689 690 691 692 693 694 695 696 697
		if (sd->entry == &sort_call) {
			int ret = regcomp(&call_regex, call, REG_EXTENDED);
			if (ret) {
				char err[BUFSIZ];

				regerror(ret, &call_regex, err, sizeof(err));
				fprintf(stderr, "Invalid regex: %s\n%s", call, err);
				exit(-1);
			}
			sort__has_call = 1;
		}

698 699
		list_add_tail(&sd->entry->list, &hist_entry__sort_list);
		sd->taken = 1;
700

701 702 703 704 705 706
		return 0;
	}

	return -ESRCH;
}

707 708 709 710 711 712 713 714 715 716 717 718 719 720 721
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
722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740
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;
}

741
static size_t
742
hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples)
743 744 745 746 747
{
	struct sort_entry *se;
	size_t ret;

	if (total_samples) {
748 749 750 751
		double percent = self->count * 100.0 / total_samples;
		char *color = PERF_COLOR_NORMAL;

		/*
752 753 754
		 * We color high-overhead entries in red, mid-overhead
		 * entries in green - and keep the low overhead places
		 * normal:
755
		 */
756
		if (percent >= 5.0) {
757
			color = PERF_COLOR_RED;
758 759 760 761
		} else {
			if (percent >= 0.5)
				color = PERF_COLOR_GREEN;
		}
762 763

		ret = color_fprintf(fp, color, "   %6.2f%%",
764 765
				(self->count * 100.0) / total_samples);
	} else
766
		ret = fprintf(fp, "%12Ld ", self->count);
767

768 769
	list_for_each_entry(se, &hist_entry__sort_list, list) {
		fprintf(fp, "  ");
770
		ret += se->print(fp, self);
771
	}
772 773 774 775 776 777

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

	return ret;
}

778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839
/*
 *
 */

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;

	if (sym->name && !regexec(&call_regex, sym->name, 0, NULL, 0))
		return sym;

	return NULL;
}

840 841 842 843
/*
 * collect histogram counts
 */

844 845
static int
hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
846 847
		struct symbol *sym, __u64 ip, struct ip_chain_event *chain,
	       	char level, __u64 count)
848
{
849 850 851 852 853 854 855 856 857 858
	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,
859
		.count	= count,
860 861 862
	};
	int cmp;

863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889
	if (sort__has_call && chain) {
		int i, nr = chain->hv;
		struct symbol *sym;
		struct dso *dso;
		__u64 ip;

		for (i = 0; i < chain->kernel; i++) {
			ip = chain->ips[nr + i];
			dso = kernel_dso;
			sym = resolve_symbol(thread, NULL, &dso, &ip);
			entry.call = call__match(sym);
			if (entry.call)
				goto got_call;
		}
		nr += i;

		for (i = 0; i < chain->user; i++) {
			ip = chain->ips[nr + i];
			sym = resolve_symbol(thread, NULL, NULL, &ip);
			entry.call = call__match(sym);
			if (entry.call)
				goto got_call;
		}
		nr += i;
	}
got_call:

890 891 892 893 894 895 896
	while (*p != NULL) {
		parent = *p;
		he = rb_entry(parent, struct hist_entry, rb_node);

		cmp = hist_entry__cmp(&entry, he);

		if (!cmp) {
897
			he->count += count;
898 899 900 901 902 903 904
			return 0;
		}

		if (cmp < 0)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
905
	}
906 907 908 909 910 911 912 913 914

	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;
915 916
}

P
Peter Zijlstra 已提交
917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974
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);
	}
}

975 976 977 978 979 980 981
/*
 * reverse the map, sort on count.
 */

static struct rb_root output_hists;

static void output__insert_entry(struct hist_entry *he)
982
{
983
	struct rb_node **p = &output_hists.rb_node;
984
	struct rb_node *parent = NULL;
985
	struct hist_entry *iter;
986 987 988

	while (*p != NULL) {
		parent = *p;
989
		iter = rb_entry(parent, struct hist_entry, rb_node);
990

991
		if (he->count > iter->count)
992 993 994 995 996
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

997 998
	rb_link_node(&he->rb_node, parent, p);
	rb_insert_color(&he->rb_node, &output_hists);
999 1000
}

1001
static void output__resort(void)
1002
{
P
Peter Zijlstra 已提交
1003
	struct rb_node *next;
1004
	struct hist_entry *n;
1005
	struct rb_root *tree = &hist;
1006

P
Peter Zijlstra 已提交
1007
	if (sort__need_collapse)
1008 1009 1010
		tree = &collapse_hists;

	next = rb_first(tree);
P
Peter Zijlstra 已提交
1011

1012 1013 1014
	while (next) {
		n = rb_entry(next, struct hist_entry, rb_node);
		next = rb_next(&n->rb_node);
1015

1016
		rb_erase(&n->rb_node, tree);
1017
		output__insert_entry(n);
1018 1019 1020
	}
}

1021
static size_t output__fprintf(FILE *fp, __u64 total_samples)
1022
{
1023
	struct hist_entry *pos;
1024
	struct sort_entry *se;
1025 1026 1027
	struct rb_node *nd;
	size_t ret = 0;

1028
	fprintf(fp, "\n");
1029
	fprintf(fp, "#\n");
1030
	fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
1031 1032 1033 1034
	fprintf(fp, "#\n");

	fprintf(fp, "# Overhead");
	list_for_each_entry(se, &hist_entry__sort_list, list)
1035
		fprintf(fp, "  %s", se->header);
1036 1037 1038
	fprintf(fp, "\n");

	fprintf(fp, "# ........");
1039
	list_for_each_entry(se, &hist_entry__sort_list, list) {
1040 1041
		int i;

1042
		fprintf(fp, "  ");
1043
		for (i = 0; i < strlen(se->header); i++)
1044
			fprintf(fp, ".");
1045
	}
1046 1047 1048
	fprintf(fp, "\n");

	fprintf(fp, "#\n");
1049

1050 1051 1052
	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);
1053 1054
	}

1055 1056
	if (!strcmp(sort_order, default_sort_order)) {
		fprintf(fp, "#\n");
1057
		fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1058 1059
		fprintf(fp, "#\n");
	}
1060
	fprintf(fp, "\n");
1061

1062 1063 1064
	return ret;
}

1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
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);
	}
}

1076 1077 1078 1079 1080
static unsigned long total = 0,
		     total_mmap = 0,
		     total_comm = 0,
		     total_fork = 0,
		     total_unknown = 0;
1081

1082
static int
1083 1084 1085 1086 1087 1088
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);
1089 1090
	__u64 ip = event->ip.ip;
	__u64 period = 1;
1091
	struct map *map = NULL;
1092
	void *more_data = event->ip.__more_data;
1093
	struct ip_chain_event *chain = NULL;
1094

1095 1096 1097 1098
	if (event->header.type & PERF_SAMPLE_PERIOD) {
		period = *(__u64 *)more_data;
		more_data += sizeof(__u64);
	}
1099

1100
	dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p period: %Ld\n",
1101 1102 1103 1104
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->header.misc,
		event->ip.pid,
1105
		(void *)(long)ip,
1106
		(long long)period);
1107

1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
	if (event->header.type & PERF_SAMPLE_CALLCHAIN) {
		int i;

		chain = (void *)more_data;

		if (dump_trace) {
			dprintf("... chain: u:%d, k:%d, nr:%d\n",
				chain->user,
				chain->kernel,
				chain->nr);

			for (i = 0; i < chain->nr; i++)
1120
				dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1121 1122 1123
		}
	}

1124 1125 1126 1127 1128 1129 1130
	dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);

	if (thread == NULL) {
		fprintf(stderr, "problem processing %d event, skipping it.\n",
			event->header.type);
		return -1;
	}
1131

1132 1133 1134
	if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
		show = SHOW_KERNEL;
		level = 'k';
1135

1136
		dso = kernel_dso;
1137

1138
		dprintf(" ...... dso: %s\n", dso->name);
1139

1140
	} else if (event->header.misc & PERF_EVENT_MISC_USER) {
1141

1142 1143
		show = SHOW_USER;
		level = '.';
1144

1145 1146 1147 1148 1149
	} else {
		show = SHOW_HV;
		level = 'H';
		dprintf(" ...... dso: [hypervisor]\n");
	}
1150

1151
	if (show & show_mask) {
1152
		struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1153

1154
		if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1155
			fprintf(stderr,
1156
		"problem incrementing symbol count, skipping event\n");
1157
			return -1;
1158
		}
1159
	}
1160
	total += period;
1161

1162 1163
	return 0;
}
I
Ingo Molnar 已提交
1164

1165 1166 1167 1168 1169 1170
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);

1171
	dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1172 1173
		(void *)(offset + head),
		(void *)(long)(event->header.size),
1174
		event->mmap.pid,
1175 1176 1177 1178 1179 1180 1181
		(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");
1182
		return 0;
1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
	}

	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;
1205
	}
1206 1207 1208 1209 1210
	total_comm++;

	return 0;
}

1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230
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;
}

1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243
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;
}

1244 1245 1246
static void trace_event(event_t *event)
{
	unsigned char *raw_event = (void *)event;
1247
	char *color = PERF_COLOR_BLUE;
1248 1249 1250 1251 1252
	int i, j;

	if (!dump_trace)
		return;

1253 1254
	dprintf(".");
	cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1255 1256

	for (i = 0; i < event->header.size; i++) {
1257 1258 1259 1260
		if ((i & 15) == 0) {
			dprintf(".");
			cdprintf("  %04x: ", i);
		}
1261

1262
		cdprintf(" %02x", raw_event[i]);
1263 1264

		if (((i & 15) == 15) || i == event->header.size-1) {
1265
			cdprintf("  ");
1266
			for (j = 0; j < 15-(i & 15); j++)
1267
				cdprintf("   ");
1268 1269
			for (j = 0; j < (i & 15); j++) {
				if (isprint(raw_event[i-15+j]))
1270
					cdprintf("%c", raw_event[i-15+j]);
1271
				else
1272
					cdprintf(".");
1273
			}
1274
			cdprintf("\n");
1275 1276 1277 1278 1279
		}
	}
	dprintf(".\n");
}

1280 1281 1282
static int
process_event(event_t *event, unsigned long offset, unsigned long head)
{
1283 1284
	trace_event(event);

1285 1286 1287 1288 1289 1290 1291 1292 1293 1294
	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);

1295 1296 1297
	case PERF_EVENT_FORK:
		return process_fork_event(event, offset, head);

1298 1299
	case PERF_EVENT_PERIOD:
		return process_period_event(event, offset, head);
1300 1301 1302
	/*
	 * We dont process them right now but they are fine:
	 */
1303

1304 1305 1306 1307
	case PERF_EVENT_THROTTLE:
	case PERF_EVENT_UNTHROTTLE:
		return 0;

1308 1309 1310 1311 1312 1313 1314 1315 1316
	default:
		return -1;
	}

	return 0;
}

static int __cmd_report(void)
{
1317
	int ret, rc = EXIT_FAILURE;
1318 1319 1320 1321 1322
	unsigned long offset = 0;
	unsigned long head = 0;
	struct stat stat;
	event_t *event;
	uint32_t size;
1323
	char *buf;
1324 1325 1326 1327 1328

	register_idle_thread();

	input = open(input_name, O_RDONLY);
	if (input < 0) {
1329 1330 1331 1332
		fprintf(stderr, " failed to open file: %s", input_name);
		if (!strcmp(input_name, "perf.data"))
			fprintf(stderr, "  (try 'perf record' first)");
		fprintf(stderr, "\n");
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 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
		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;

1391
	dprintf("\n%p [%p]: event: %d\n",
1392 1393 1394 1395
			(void *)(offset + head),
			(void *)(long)event->header.size,
			event->header.type);

1396 1397
	if (!size || process_event(event, offset, head) < 0) {

I
Ingo Molnar 已提交
1398 1399 1400 1401
		dprintf("%p [%p]: skipping unknown header type: %d\n",
			(void *)(offset + head),
			(void *)(long)(event->header.size),
			event->header.type);
1402

1403
		total_unknown++;
1404 1405 1406 1407 1408 1409 1410 1411 1412 1413

		/*
		 * 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;
1414
	}
1415

1416
	head += size;
I
Ingo Molnar 已提交
1417

1418 1419 1420 1421 1422
	if (offset + head < stat.st_size)
		goto more;

	rc = EXIT_SUCCESS;
	close(input);
1423

I
Ingo Molnar 已提交
1424 1425 1426
	dprintf("      IP events: %10ld\n", total);
	dprintf("    mmap events: %10ld\n", total_mmap);
	dprintf("    comm events: %10ld\n", total_comm);
1427
	dprintf("    fork events: %10ld\n", total_fork);
I
Ingo Molnar 已提交
1428
	dprintf(" unknown events: %10ld\n", total_unknown);
1429

I
Ingo Molnar 已提交
1430
	if (dump_trace)
1431 1432
		return 0;

1433 1434 1435
	if (verbose >= 3)
		threads__fprintf(stdout);

1436
	if (verbose >= 2)
1437 1438
		dsos__fprintf(stdout);

P
Peter Zijlstra 已提交
1439
	collapse__resort();
1440 1441
	output__resort();
	output__fprintf(stdout, total);
1442 1443 1444 1445

	return rc;
}

1446 1447 1448 1449 1450 1451 1452 1453
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"),
1454 1455
	OPT_BOOLEAN('v', "verbose", &verbose,
		    "be more verbose (show symbol address, etc)"),
1456 1457
	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
		    "dump raw trace in ASCII"),
1458
	OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1459 1460
	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
		   "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"),
1461 1462
	OPT_BOOLEAN('P', "full-paths", &full_paths,
		    "Don't shorten the pathnames taking into account the cwd"),
1463 1464
	OPT_STRING('c', "call", &call, "regex",
		   "regex to use for --sort call"),
1465 1466 1467
	OPT_END()
};

1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482
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);
}

1483 1484
int cmd_report(int argc, const char **argv, const char *prefix)
{
1485
	symbol__init();
1486 1487 1488

	page_size = getpagesize();

1489
	argc = parse_options(argc, argv, options, report_usage, 0);
1490

1491 1492
	setup_sorting();

1493 1494 1495 1496 1497 1498
	/*
	 * Any (unrecognized) arguments left?
	 */
	if (argc)
		usage_with_options(report_usage, options);

1499 1500
	setup_pager();

1501 1502
	return __cmd_report();
}