builtin-report.c 27.0 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
static int		collapse_syscalls;
44

45 46 47
static unsigned long	page_size;
static unsigned long	mmap_window = 32;

48 49 50 51 52 53 54 55
struct ip_chain_event {
	__u16 nr;
	__u16 hv;
	__u16 kernel;
	__u16 user;
	__u64 ips[];
};

56 57 58 59
struct ip_event {
	struct perf_event_header header;
	__u64 ip;
	__u32 pid, tid;
60
	unsigned char __more_data[];
61
};
62

63 64 65 66 67 68 69 70
struct mmap_event {
	struct perf_event_header header;
	__u32 pid, tid;
	__u64 start;
	__u64 len;
	__u64 pgoff;
	char filename[PATH_MAX];
};
71

72 73
struct comm_event {
	struct perf_event_header header;
74
	__u32 pid, tid;
75 76 77
	char comm[16];
};

78 79 80 81 82
struct fork_event {
	struct perf_event_header header;
	__u32 pid, ppid;
};

83
struct period_event {
84
	struct perf_event_header header;
85 86 87 88 89 90 91 92 93 94 95 96
	__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;
97 98 99 100
} event_t;

static LIST_HEAD(dsos);
static struct dso *kernel_dso;
P
Peter Zijlstra 已提交
101
static struct dso *vdso;
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120

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);
121
	int nr;
122

123 124 125 126 127 128
	if (dso)
		return dso;

	dso = dso__new(name, 0);
	if (!dso)
		goto out_delete_dso;
129

130
	nr = dso__load(dso, NULL, verbose);
131
	if (nr < 0) {
132 133
		if (verbose)
			fprintf(stderr, "Failed to open: %s\n", name);
134
		goto out_delete_dso;
135
	}
136 137 138 139 140 141 142
	if (!nr && verbose) {
		fprintf(stderr,
		"No symbols found in: %s, maybe install a debug package?\n",
				name);
	}

	dsos__add(dso);
143 144 145 146 147 148 149 150

	return dso;

out_delete_dso:
	dso__delete(dso);
	return NULL;
}

151
static void dsos__fprintf(FILE *fp)
152 153 154 155 156 157 158
{
	struct dso *pos;

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

159
static struct symbol *vdso__find_symbol(struct dso *dso, __u64 ip)
P
Peter Zijlstra 已提交
160 161 162 163
{
	return dso__find_symbol(kernel_dso, ip);
}

164 165
static int load_kernel(void)
{
166
	int err;
167

168
	kernel_dso = dso__new("[kernel]", 0);
169
	if (!kernel_dso)
170
		return -1;
171

172
	err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
173 174 175 176 177
	if (err) {
		dso__delete(kernel_dso);
		kernel_dso = NULL;
	} else
		dsos__add(kernel_dso);
178

P
Peter Zijlstra 已提交
179 180 181 182 183 184 185 186
	vdso = dso__new("[vdso]", 0);
	if (!vdso)
		return -1;

	vdso->find_symbol = vdso__find_symbol;

	dsos__add(vdso);

187
	return err;
188 189
}

190 191 192 193 194
static char __cwd[PATH_MAX];
static char *cwd = __cwd;
static int cwdlen;

static int strcommon(const char *pathname)
195 196 197 198 199 200 201 202 203
{
	int n = 0;

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

	return n;
}

204 205
struct map {
	struct list_head node;
206 207 208 209
	__u64	 start;
	__u64	 end;
	__u64	 pgoff;
	__u64	 (*map_ip)(struct map *, __u64);
210 211 212
	struct dso	 *dso;
};

213
static __u64 map__map_ip(struct map *map, __u64 ip)
P
Peter Zijlstra 已提交
214 215 216 217
{
	return ip - map->start + map->pgoff;
}

218
static __u64 vdso__map_ip(struct map *map, __u64 ip)
P
Peter Zijlstra 已提交
219 220 221 222
{
	return ip;
}

223 224 225 226 227
static inline int is_anon_memory(const char *filename)
{
     return strcmp(filename, "//anon") == 0;
}

228
static struct map *map__new(struct mmap_event *event)
229 230 231 232
{
	struct map *self = malloc(sizeof(*self));

	if (self != NULL) {
233 234
		const char *filename = event->filename;
		char newfilename[PATH_MAX];
235
		int anon;
236 237

		if (cwd) {
238 239
			int n = strcommon(filename);

240 241 242 243 244 245 246
			if (n == cwdlen) {
				snprintf(newfilename, sizeof(newfilename),
					 ".%s", filename + n);
				filename = newfilename;
			}
		}

247 248 249 250 251 252 253
		anon = is_anon_memory(filename);

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

254 255 256 257
		self->start = event->start;
		self->end   = event->start + event->len;
		self->pgoff = event->pgoff;

258
		self->dso = dsos__findnew(filename);
259 260
		if (self->dso == NULL)
			goto out_delete;
P
Peter Zijlstra 已提交
261

262
		if (self->dso == vdso || anon)
P
Peter Zijlstra 已提交
263 264 265
			self->map_ip = vdso__map_ip;
		else
			self->map_ip = map__map_ip;
266 267 268 269 270 271 272
	}
	return self;
out_delete:
	free(self);
	return NULL;
}

273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
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;
}
298

299 300
static size_t map__fprintf(struct map *self, FILE *fp)
{
301
	return fprintf(fp, " %Lx-%Lx %Lx %s\n",
302 303 304 305
		       self->start, self->end, self->pgoff, self->dso->name);
}


306
struct thread {
307
	struct rb_node	 rb_node;
308 309 310 311 312 313 314 315 316 317 318
	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
319
		self->comm = malloc(32);
320
		if (self->comm)
P
Peter Zijlstra 已提交
321
			snprintf(self->comm, 32, ":%d", self->pid);
322 323 324 325 326 327 328 329
		INIT_LIST_HEAD(&self->maps);
	}

	return self;
}

static int thread__set_comm(struct thread *self, const char *comm)
{
P
Peter Zijlstra 已提交
330 331
	if (self->comm)
		free(self->comm);
332 333 334 335
	self->comm = strdup(comm);
	return self->comm ? 0 : -ENOMEM;
}

336 337 338 339 340 341 342 343 344 345 346 347
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;
}


348
static struct rb_root threads;
349
static struct thread *last_match;
350

351
static struct thread *threads__findnew(pid_t pid)
352
{
353 354 355
	struct rb_node **p = &threads.rb_node;
	struct rb_node *parent = NULL;
	struct thread *th;
356

357 358 359 360 361 362 363 364
	/*
	 * 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;

365 366 367
	while (*p != NULL) {
		parent = *p;
		th = rb_entry(parent, struct thread, rb_node);
368

369 370
		if (th->pid == pid) {
			last_match = th;
371
			return th;
372
		}
373

374 375 376 377
		if (pid < th->pid)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
378 379
	}

380 381 382 383
	th = thread__new(pid);
	if (th != NULL) {
		rb_link_node(&th->rb_node, parent, p);
		rb_insert_color(&th->rb_node, &threads);
384
		last_match = th;
385
	}
386

387
	return th;
388 389 390 391
}

static void thread__insert_map(struct thread *self, struct map *map)
{
392 393 394 395 396 397 398 399 400 401
	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);
		}
	}

402 403 404
	list_add_tail(&map->node, &self->maps);
}

405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424
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;
}

425
static struct map *thread__find_map(struct thread *self, __u64 ip)
426
{
427 428
	struct map *pos;

429 430 431 432 433 434 435 436 437 438
	if (self == NULL)
		return NULL;

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

	return NULL;
}

439 440 441 442 443 444 445 446 447 448 449 450 451 452
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;
}

453 454 455 456 457 458 459 460 461 462 463 464 465
/*
 * 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;
466
	__u64		 ip;
467 468
	char		 level;

469
	__u64		 count;
470 471
};

472 473 474 475 476 477 478
/*
 * configurable sorting bits
 */

struct sort_entry {
	struct list_head list;

479 480
	char *header;

481
	int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
P
Peter Zijlstra 已提交
482
	int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
483 484 485
	size_t	(*print)(FILE *fp, struct hist_entry *);
};

P
Peter Zijlstra 已提交
486 487
/* --sort pid */

488
static int64_t
489
sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
490
{
491 492 493 494 495 496
	return right->thread->pid - left->thread->pid;
}

static size_t
sort__thread_print(FILE *fp, struct hist_entry *self)
{
497
	return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
498
}
499

500
static struct sort_entry sort_thread = {
501
	.header = "         Command:  Pid",
502 503 504 505
	.cmp	= sort__thread_cmp,
	.print	= sort__thread_print,
};

P
Peter Zijlstra 已提交
506 507
/* --sort comm */

508 509
static int64_t
sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
P
Peter Zijlstra 已提交
510 511 512 513 514 515
{
	return right->thread->pid - left->thread->pid;
}

static int64_t
sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
{
	char *comm_l = left->thread->comm;
	char *comm_r = right->thread->comm;

	if (!comm_l || !comm_r) {
		if (!comm_l && !comm_r)
			return 0;
		else if (!comm_l)
			return -1;
		else
			return 1;
	}

	return strcmp(comm_l, comm_r);
}

static size_t
sort__comm_print(FILE *fp, struct hist_entry *self)
{
535
	return fprintf(fp, "%16s", self->thread->comm);
536 537 538
}

static struct sort_entry sort_comm = {
539
	.header		= "         Command",
P
Peter Zijlstra 已提交
540 541 542
	.cmp		= sort__comm_cmp,
	.collapse	= sort__comm_collapse,
	.print		= sort__comm_print,
543 544
};

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

547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
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;

	if (!dso_l || !dso_r) {
		if (!dso_l && !dso_r)
			return 0;
		else if (!dso_l)
			return -1;
		else
			return 1;
	}

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

static size_t
sort__dso_print(FILE *fp, struct hist_entry *self)
{
568
	if (self->dso)
569
		return fprintf(fp, "%-25s", self->dso->name);
570

571
	return fprintf(fp, "%016llx         ", (__u64)self->ip);
572 573 574
}

static struct sort_entry sort_dso = {
575
	.header = "Shared Object            ",
576 577 578 579
	.cmp	= sort__dso_cmp,
	.print	= sort__dso_print,
};

P
Peter Zijlstra 已提交
580 581
/* --sort symbol */

582 583 584
static int64_t
sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
{
585
	__u64 ip_l, ip_r;
586 587 588 589 590 591 592 593 594 595

	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);
}

596 597 598 599 600 601
static size_t
sort__sym_print(FILE *fp, struct hist_entry *self)
{
	size_t ret = 0;

	if (verbose)
602
		ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
603

604 605 606 607
	if (self->sym) {
		ret += fprintf(fp, "[%c] %s",
			self->dso == kernel_dso ? 'k' : '.', self->sym->name);
	} else {
608
		ret += fprintf(fp, "%#016llx", (__u64)self->ip);
609
	}
610 611 612 613 614

	return ret;
}

static struct sort_entry sort_sym = {
615
	.header = "Symbol",
616 617
	.cmp	= sort__sym_cmp,
	.print	= sort__sym_print,
618 619
};

P
Peter Zijlstra 已提交
620 621
static int sort__need_collapse = 0;

622
struct sort_dimension {
623 624 625
	char			*name;
	struct sort_entry	*entry;
	int			taken;
626 627 628 629
};

static struct sort_dimension sort_dimensions[] = {
	{ .name = "pid",	.entry = &sort_thread,	},
630
	{ .name = "comm",	.entry = &sort_comm,	},
631
	{ .name = "dso",	.entry = &sort_dso,	},
632 633 634
	{ .name = "symbol",	.entry = &sort_sym,	},
};

635 636
static LIST_HEAD(hist_entry__sort_list);

637 638 639 640 641 642 643 644 645 646
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;

647
		if (strncasecmp(tok, sd->name, strlen(tok)))
648 649
			continue;

P
Peter Zijlstra 已提交
650 651 652
		if (sd->entry->collapse)
			sort__need_collapse = 1;

653 654
		list_add_tail(&sd->entry->list, &hist_entry__sort_list);
		sd->taken = 1;
655

656 657 658 659 660 661
		return 0;
	}

	return -ESRCH;
}

662 663 664 665 666 667 668 669 670 671 672 673 674 675 676
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
677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695
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;
}

696
static size_t
697
hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples)
698 699 700 701 702
{
	struct sort_entry *se;
	size_t ret;

	if (total_samples) {
703 704 705 706
		double percent = self->count * 100.0 / total_samples;
		char *color = PERF_COLOR_NORMAL;

		/*
707 708 709
		 * We color high-overhead entries in red, mid-overhead
		 * entries in green - and keep the low overhead places
		 * normal:
710
		 */
711
		if (percent >= 5.0) {
712
			color = PERF_COLOR_RED;
713 714 715 716
		} else {
			if (percent >= 0.5)
				color = PERF_COLOR_GREEN;
		}
717 718

		ret = color_fprintf(fp, color, "   %6.2f%%",
719 720
				(self->count * 100.0) / total_samples);
	} else
721
		ret = fprintf(fp, "%12Ld ", self->count);
722

723 724
	list_for_each_entry(se, &hist_entry__sort_list, list) {
		fprintf(fp, "  ");
725
		ret += se->print(fp, self);
726
	}
727 728 729 730 731 732 733 734 735 736

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

	return ret;
}

/*
 * collect histogram counts
 */

737 738
static int
hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
739
		struct symbol *sym, __u64 ip, char level, __u64 count)
740
{
741 742 743 744 745 746 747 748 749 750
	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,
751
		.count	= count,
752 753 754 755 756 757 758 759 760 761
	};
	int cmp;

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

		cmp = hist_entry__cmp(&entry, he);

		if (!cmp) {
762
			he->count += count;
763 764 765 766 767 768 769
			return 0;
		}

		if (cmp < 0)
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
770
	}
771 772 773 774 775 776 777 778 779

	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;
780 781
}

P
Peter Zijlstra 已提交
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 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);
	}
}

840 841 842 843 844 845 846
/*
 * reverse the map, sort on count.
 */

static struct rb_root output_hists;

static void output__insert_entry(struct hist_entry *he)
847
{
848
	struct rb_node **p = &output_hists.rb_node;
849
	struct rb_node *parent = NULL;
850
	struct hist_entry *iter;
851 852 853

	while (*p != NULL) {
		parent = *p;
854
		iter = rb_entry(parent, struct hist_entry, rb_node);
855

856
		if (he->count > iter->count)
857 858 859 860 861
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

862 863
	rb_link_node(&he->rb_node, parent, p);
	rb_insert_color(&he->rb_node, &output_hists);
864 865
}

866
static void output__resort(void)
867
{
P
Peter Zijlstra 已提交
868
	struct rb_node *next;
869
	struct hist_entry *n;
870
	struct rb_root *tree = &hist;
871

P
Peter Zijlstra 已提交
872
	if (sort__need_collapse)
873 874 875
		tree = &collapse_hists;

	next = rb_first(tree);
P
Peter Zijlstra 已提交
876

877 878 879
	while (next) {
		n = rb_entry(next, struct hist_entry, rb_node);
		next = rb_next(&n->rb_node);
880

881
		rb_erase(&n->rb_node, tree);
882
		output__insert_entry(n);
883 884 885
	}
}

886
static size_t output__fprintf(FILE *fp, __u64 total_samples)
887
{
888
	struct hist_entry *pos;
889
	struct sort_entry *se;
890 891 892
	struct rb_node *nd;
	size_t ret = 0;

893
	fprintf(fp, "\n");
894
	fprintf(fp, "#\n");
895
	fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
896 897 898 899
	fprintf(fp, "#\n");

	fprintf(fp, "# Overhead");
	list_for_each_entry(se, &hist_entry__sort_list, list)
900
		fprintf(fp, "  %s", se->header);
901 902 903
	fprintf(fp, "\n");

	fprintf(fp, "# ........");
904
	list_for_each_entry(se, &hist_entry__sort_list, list) {
905 906
		int i;

907
		fprintf(fp, "  ");
908
		for (i = 0; i < strlen(se->header); i++)
909
			fprintf(fp, ".");
910
	}
911 912 913
	fprintf(fp, "\n");

	fprintf(fp, "#\n");
914

915 916 917
	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);
918 919
	}

920 921
	if (!strcmp(sort_order, default_sort_order)) {
		fprintf(fp, "#\n");
922
		fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
923 924
		fprintf(fp, "#\n");
	}
925
	fprintf(fp, "\n");
926

927 928 929
	return ret;
}

930 931 932 933 934 935 936 937 938 939 940
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);
	}
}

941 942 943 944 945
static unsigned long total = 0,
		     total_mmap = 0,
		     total_comm = 0,
		     total_fork = 0,
		     total_unknown = 0;
946

947
static int
948 949 950 951 952 953
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);
954 955
	__u64 ip = event->ip.ip;
	__u64 period = 1;
956
	struct map *map = NULL;
957 958
	void *more_data = event->ip.__more_data;
	struct ip_chain_event *chain;
959

960 961 962 963
	if (event->header.type & PERF_SAMPLE_PERIOD) {
		period = *(__u64 *)more_data;
		more_data += sizeof(__u64);
	}
964

965
	dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p period: %Ld\n",
966 967 968 969
		(void *)(offset + head),
		(void *)(long)(event->header.size),
		event->header.misc,
		event->ip.pid,
970
		(void *)(long)ip,
971
		(long long)period);
972

973 974 975 976 977 978 979 980 981 982 983 984
	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++)
985
				dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
986
		}
987 988 989 990 991 992 993 994 995
		if (collapse_syscalls) {
			/*
			 * Find the all-but-last kernel entry
			 * amongst the call-chains - to get
			 * to the level of system calls:
			 */
			if (chain->kernel >= 2)
				ip = chain->ips[chain->kernel-2];
		}
996 997
	}

998 999 1000 1001 1002 1003 1004
	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;
	}
1005

1006 1007 1008
	if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
		show = SHOW_KERNEL;
		level = 'k';
1009

1010
		dso = kernel_dso;
1011

1012
		dprintf(" ...... dso: %s\n", dso->name);
1013

1014
	} else if (event->header.misc & PERF_EVENT_MISC_USER) {
1015

1016 1017
		show = SHOW_USER;
		level = '.';
1018

1019 1020
		map = thread__find_map(thread, ip);
		if (map != NULL) {
P
Peter Zijlstra 已提交
1021
			ip = map->map_ip(map, ip);
1022
			dso = map->dso;
1023
		} else {
1024 1025 1026 1027 1028 1029 1030 1031
			/*
			 * 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;
1032
		}
1033 1034 1035 1036 1037 1038 1039
		dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");

	} else {
		show = SHOW_HV;
		level = 'H';
		dprintf(" ...... dso: [hypervisor]\n");
	}
1040

1041
	if (show & show_mask) {
P
Peter Zijlstra 已提交
1042 1043 1044 1045
		struct symbol *sym = NULL;

		if (dso)
			sym = dso->find_symbol(dso, ip);
1046

1047
		if (hist_entry__add(thread, map, dso, sym, ip, level, period)) {
1048
			fprintf(stderr,
1049
		"problem incrementing symbol count, skipping event\n");
1050
			return -1;
1051
		}
1052
	}
1053
	total += period;
1054

1055 1056
	return 0;
}
I
Ingo Molnar 已提交
1057

1058 1059 1060 1061 1062 1063
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);

1064
	dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1065 1066
		(void *)(offset + head),
		(void *)(long)(event->header.size),
1067
		event->mmap.pid,
1068 1069 1070 1071 1072 1073 1074
		(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");
1075
		return 0;
1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097
	}

	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;
1098
	}
1099 1100 1101 1102 1103
	total_comm++;

	return 0;
}

1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123
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;
}

1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
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;
}

1137 1138 1139
static void trace_event(event_t *event)
{
	unsigned char *raw_event = (void *)event;
1140
	char *color = PERF_COLOR_BLUE;
1141 1142 1143 1144 1145
	int i, j;

	if (!dump_trace)
		return;

1146 1147
	dprintf(".");
	cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1148 1149

	for (i = 0; i < event->header.size; i++) {
1150 1151 1152 1153
		if ((i & 15) == 0) {
			dprintf(".");
			cdprintf("  %04x: ", i);
		}
1154

1155
		cdprintf(" %02x", raw_event[i]);
1156 1157

		if (((i & 15) == 15) || i == event->header.size-1) {
1158
			cdprintf("  ");
1159
			for (j = 0; j < 15-(i & 15); j++)
1160
				cdprintf("   ");
1161 1162
			for (j = 0; j < (i & 15); j++) {
				if (isprint(raw_event[i-15+j]))
1163
					cdprintf("%c", raw_event[i-15+j]);
1164
				else
1165
					cdprintf(".");
1166
			}
1167
			cdprintf("\n");
1168 1169 1170 1171 1172
		}
	}
	dprintf(".\n");
}

1173 1174 1175
static int
process_event(event_t *event, unsigned long offset, unsigned long head)
{
1176 1177
	trace_event(event);

1178 1179 1180 1181 1182 1183 1184 1185 1186 1187
	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);

1188 1189 1190
	case PERF_EVENT_FORK:
		return process_fork_event(event, offset, head);

1191 1192
	case PERF_EVENT_PERIOD:
		return process_period_event(event, offset, head);
1193 1194 1195
	/*
	 * We dont process them right now but they are fine:
	 */
1196

1197 1198 1199 1200
	case PERF_EVENT_THROTTLE:
	case PERF_EVENT_UNTHROTTLE:
		return 0;

1201 1202 1203 1204 1205 1206 1207 1208 1209
	default:
		return -1;
	}

	return 0;
}

static int __cmd_report(void)
{
1210
	int ret, rc = EXIT_FAILURE;
1211 1212 1213 1214 1215
	unsigned long offset = 0;
	unsigned long head = 0;
	struct stat stat;
	event_t *event;
	uint32_t size;
1216
	char *buf;
1217 1218 1219 1220 1221

	register_idle_thread();

	input = open(input_name, O_RDONLY);
	if (input < 0) {
1222 1223 1224 1225
		fprintf(stderr, " failed to open file: %s", input_name);
		if (!strcmp(input_name, "perf.data"))
			fprintf(stderr, "  (try 'perf record' first)");
		fprintf(stderr, "\n");
1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283
		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;

1284
	dprintf("\n%p [%p]: event: %d\n",
1285 1286 1287 1288
			(void *)(offset + head),
			(void *)(long)event->header.size,
			event->header.type);

1289 1290
	if (!size || process_event(event, offset, head) < 0) {

I
Ingo Molnar 已提交
1291 1292 1293 1294
		dprintf("%p [%p]: skipping unknown header type: %d\n",
			(void *)(offset + head),
			(void *)(long)(event->header.size),
			event->header.type);
1295

1296
		total_unknown++;
1297 1298 1299 1300 1301 1302 1303 1304 1305 1306

		/*
		 * 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;
1307
	}
1308

1309
	head += size;
I
Ingo Molnar 已提交
1310

1311 1312 1313 1314 1315
	if (offset + head < stat.st_size)
		goto more;

	rc = EXIT_SUCCESS;
	close(input);
1316

I
Ingo Molnar 已提交
1317 1318 1319
	dprintf("      IP events: %10ld\n", total);
	dprintf("    mmap events: %10ld\n", total_mmap);
	dprintf("    comm events: %10ld\n", total_comm);
1320
	dprintf("    fork events: %10ld\n", total_fork);
I
Ingo Molnar 已提交
1321
	dprintf(" unknown events: %10ld\n", total_unknown);
1322

I
Ingo Molnar 已提交
1323
	if (dump_trace)
1324 1325
		return 0;

1326 1327 1328
	if (verbose >= 3)
		threads__fprintf(stdout);

1329
	if (verbose >= 2)
1330 1331
		dsos__fprintf(stdout);

P
Peter Zijlstra 已提交
1332
	collapse__resort();
1333 1334
	output__resort();
	output__fprintf(stdout, total);
1335 1336 1337 1338

	return rc;
}

1339 1340 1341 1342 1343 1344 1345 1346
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"),
1347 1348
	OPT_BOOLEAN('v', "verbose", &verbose,
		    "be more verbose (show symbol address, etc)"),
1349 1350
	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
		    "dump raw trace in ASCII"),
1351
	OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1352 1353
	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
		   "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"),
1354 1355
	OPT_BOOLEAN('P', "full-paths", &full_paths,
		    "Don't shorten the pathnames taking into account the cwd"),
1356 1357
	OPT_BOOLEAN('S', "syscalls", &collapse_syscalls,
		    "show per syscall summary overhead, using call graph"),
1358 1359 1360
	OPT_END()
};

1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375
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);
}

1376 1377
int cmd_report(int argc, const char **argv, const char *prefix)
{
1378
	symbol__init();
1379 1380 1381

	page_size = getpagesize();

1382
	argc = parse_options(argc, argv, options, report_usage, 0);
1383

1384 1385
	setup_sorting();

1386 1387 1388 1389 1390 1391
	/*
	 * Any (unrecognized) arguments left?
	 */
	if (argc)
		usage_with_options(report_usage, options);

1392 1393
	setup_pager();

1394 1395
	return __cmd_report();
}