builtin-top.c 31.5 KB
Newer Older
1
/*
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
 * builtin-top.c
 *
 * Builtin top command: Display a continuously updated profile of
 * any workload, CPU or specific PID.
 *
 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
 *
 * Improvements and fixes by:
 *
 *   Arjan van de Ven <arjan@linux.intel.com>
 *   Yanmin Zhang <yanmin.zhang@intel.com>
 *   Wu Fengguang <fengguang.wu@intel.com>
 *   Mike Galbraith <efault@gmx.de>
 *   Paul Mackerras <paulus@samba.org>
 *
 * Released under the GPL v2. (and only v2, not any later version)
18
 */
19
#include "builtin.h"
20

21
#include "perf.h"
22

23
#include "util/symbol.h"
24
#include "util/color.h"
25
#include "util/thread.h"
26
#include "util/util.h"
27
#include <linux/rbtree.h>
28 29
#include "util/parse-options.h"
#include "util/parse-events.h"
30

31 32
#include "util/debug.h"

33 34
#include <assert.h>
#include <fcntl.h>
35

36
#include <stdio.h>
37 38
#include <termios.h>
#include <unistd.h>
39

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55
#include <errno.h>
#include <time.h>
#include <sched.h>
#include <pthread.h>

#include <sys/syscall.h>
#include <sys/ioctl.h>
#include <sys/poll.h>
#include <sys/prctl.h>
#include <sys/wait.h>
#include <sys/uio.h>
#include <sys/mman.h>

#include <linux/unistd.h>
#include <linux/types.h>

56
static int			fd[MAX_NR_CPUS][MAX_COUNTERS];
57

58
static int			system_wide			=      0;
59

60
static int			default_interval		=      0;
61

62
static int			count_filter			=      5;
63
static int			print_entries;
64

65 66 67 68 69 70
static int			target_pid			=     -1;
static int			inherit				=      0;
static int			profile_cpu			=     -1;
static int			nr_cpus				=      0;
static unsigned int		realtime_prio			=      0;
static int			group				=      0;
71
static unsigned int		page_size;
72 73
static unsigned int		mmap_pages			=     16;
static int			freq				=   1000; /* 1 KHz */
74

75 76 77
static int			delay_secs			=      2;
static int			zero                            =      0;
static int			dump_symtab                     =      0;
78

79 80
static bool			hide_kernel_symbols		=  false;
static bool			hide_user_symbols		=  false;
81 82 83 84 85 86 87 88
static struct winsize		winsize;
static const char		*graph_line			=
	"_____________________________________________________________________"
	"_____________________________________________________________________";
static const char		*graph_dotted_line			=
	"---------------------------------------------------------------------"
	"---------------------------------------------------------------------"
	"---------------------------------------------------------------------";
89

90 91 92 93 94 95 96 97 98 99 100
/*
 * Source
 */

struct source_line {
	u64			eip;
	unsigned long		count[MAX_COUNTERS];
	char			*line;
	struct source_line	*next;
};

101 102 103 104 105
static char			*sym_filter			=   NULL;
struct sym_entry		*sym_filter_entry		=   NULL;
static int			sym_pcnt_filter			=      5;
static int			sym_counter			=      0;
static int			display_weighted		=     -1;
106

107 108 109 110
/*
 * Symbols
 */

111 112 113 114 115 116 117
struct sym_entry_source {
	struct source_line	*source;
	struct source_line	*lines;
	struct source_line	**lines_tail;
	pthread_mutex_t		lock;
};

118
struct sym_entry {
119 120
	struct rb_node		rb_node;
	struct list_head	node;
121 122
	unsigned long		snap_count;
	double			weight;
123
	int			skip;
124
	u16			name_len;
125
	u8			origin;
126
	struct map		*map;
127
	struct sym_entry_source	*src;
128
	unsigned long		count[0];
129 130
};

131 132 133 134
/*
 * Source functions
 */

135 136
static inline struct symbol *sym_entry__symbol(struct sym_entry *self)
{
137
       return ((void *)self) + symbol__priv_size;
138 139
}

140
static void get_term_dimensions(struct winsize *ws)
141
{
142 143 144 145 146 147 148 149 150 151
	char *s = getenv("LINES");

	if (s != NULL) {
		ws->ws_row = atoi(s);
		s = getenv("COLUMNS");
		if (s != NULL) {
			ws->ws_col = atoi(s);
			if (ws->ws_row && ws->ws_col)
				return;
		}
152
	}
153 154 155 156
#ifdef TIOCGWINSZ
	if (ioctl(1, TIOCGWINSZ, ws) == 0 &&
	    ws->ws_row && ws->ws_col)
		return;
157
#endif
158 159
	ws->ws_row = 25;
	ws->ws_col = 80;
160 161
}

162
static void update_print_entries(struct winsize *ws)
163
{
164 165
	print_entries = ws->ws_row;

166 167 168 169 170 171
	if (print_entries > 9)
		print_entries -= 9;
}

static void sig_winch_handler(int sig __used)
{
172 173
	get_term_dimensions(&winsize);
	update_print_entries(&winsize);
174 175
}

176 177 178
static void parse_source(struct sym_entry *syme)
{
	struct symbol *sym;
179
	struct sym_entry_source *source;
180
	struct map *map;
181
	FILE *file;
182
	char command[PATH_MAX*2];
183 184
	const char *path;
	u64 len;
185 186 187 188

	if (!syme)
		return;

189 190 191 192 193 194 195 196 197 198 199
	if (syme->src == NULL) {
		syme->src = calloc(1, sizeof(*source));
		if (syme->src == NULL)
			return;
		pthread_mutex_init(&syme->src->lock, NULL);
	}

	source = syme->src;

	if (source->lines) {
		pthread_mutex_lock(&source->lock);
200 201 202
		goto out_assign;
	}

203
	sym = sym_entry__symbol(syme);
204 205
	map = syme->map;
	path = map->dso->long_name;
206 207 208

	len = sym->end - sym->start;

209 210 211
	sprintf(command,
		"objdump --start-address=0x%016Lx "
			 "--stop-address=0x%016Lx -dS %s",
212 213
		map->unmap_ip(map, sym->start),
		map->unmap_ip(map, sym->end), path);
214 215 216 217 218

	file = popen(command, "r");
	if (!file)
		return;

219 220
	pthread_mutex_lock(&source->lock);
	source->lines_tail = &source->lines;
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239
	while (!feof(file)) {
		struct source_line *src;
		size_t dummy = 0;
		char *c;

		src = malloc(sizeof(struct source_line));
		assert(src != NULL);
		memset(src, 0, sizeof(struct source_line));

		if (getline(&src->line, &dummy, file) < 0)
			break;
		if (!src->line)
			break;

		c = strchr(src->line, '\n');
		if (c)
			*c = 0;

		src->next = NULL;
240 241
		*source->lines_tail = src;
		source->lines_tail = &src->next;
242 243 244

		if (strlen(src->line)>8 && src->line[8] == ':') {
			src->eip = strtoull(src->line, NULL, 16);
245
			src->eip = map->unmap_ip(map, src->eip);
246 247 248
		}
		if (strlen(src->line)>8 && src->line[16] == ':') {
			src->eip = strtoull(src->line, NULL, 16);
249
			src->eip = map->unmap_ip(map, src->eip);
250 251 252 253 254
		}
	}
	pclose(file);
out_assign:
	sym_filter_entry = syme;
255
	pthread_mutex_unlock(&source->lock);
256 257 258 259 260 261 262
}

static void __zero_source_counters(struct sym_entry *syme)
{
	int i;
	struct source_line *line;

263
	line = syme->src->lines;
264 265 266 267 268 269 270 271 272 273 274 275 276 277
	while (line) {
		for (i = 0; i < nr_counters; i++)
			line->count[i] = 0;
		line = line->next;
	}
}

static void record_precise_ip(struct sym_entry *syme, int counter, u64 ip)
{
	struct source_line *line;

	if (syme != sym_filter_entry)
		return;

278
	if (pthread_mutex_trylock(&syme->src->lock))
279 280
		return;

281
	if (syme->src == NULL || syme->src->source == NULL)
282 283
		goto out_unlock;

284
	for (line = syme->src->lines; line; line = line->next) {
285 286 287 288 289 290 291 292
		if (line->eip == ip) {
			line->count[counter]++;
			break;
		}
		if (line->eip > ip)
			break;
	}
out_unlock:
293
	pthread_mutex_unlock(&syme->src->lock);
294 295 296 297
}

static void lookup_sym_source(struct sym_entry *syme)
{
298
	struct symbol *symbol = sym_entry__symbol(syme);
299 300 301 302 303
	struct source_line *line;
	char pattern[PATH_MAX];

	sprintf(pattern, "<%s>:", symbol->name);

304 305
	pthread_mutex_lock(&syme->src->lock);
	for (line = syme->src->lines; line; line = line->next) {
306
		if (strstr(line->line, pattern)) {
307
			syme->src->source = line;
308 309 310
			break;
		}
	}
311
	pthread_mutex_unlock(&syme->src->lock);
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
}

static void show_lines(struct source_line *queue, int count, int total)
{
	int i;
	struct source_line *line;

	line = queue;
	for (i = 0; i < count; i++) {
		float pcnt = 100.0*(float)line->count[sym_counter]/(float)total;

		printf("%8li %4.1f%%\t%s\n", line->count[sym_counter], pcnt, line->line);
		line = line->next;
	}
}

#define TRACE_COUNT     3

static void show_details(struct sym_entry *syme)
{
	struct symbol *symbol;
	struct source_line *line;
	struct source_line *line_queue = NULL;
	int displayed = 0;
	int line_queue_count = 0, total = 0, more = 0;

	if (!syme)
		return;

341
	if (!syme->src->source)
342 343
		lookup_sym_source(syme);

344
	if (!syme->src->source)
345 346
		return;

347
	symbol = sym_entry__symbol(syme);
348 349 350
	printf("Showing %s for %s\n", event_name(sym_counter), symbol->name);
	printf("  Events  Pcnt (>=%d%%)\n", sym_pcnt_filter);

351 352
	pthread_mutex_lock(&syme->src->lock);
	line = syme->src->source;
353 354 355 356 357
	while (line) {
		total += line->count[sym_counter];
		line = line->next;
	}

358
	line = syme->src->source;
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
	while (line) {
		float pcnt = 0.0;

		if (!line_queue_count)
			line_queue = line;
		line_queue_count++;

		if (line->count[sym_counter])
			pcnt = 100.0 * line->count[sym_counter] / (float)total;
		if (pcnt >= (float)sym_pcnt_filter) {
			if (displayed <= print_entries)
				show_lines(line_queue, line_queue_count, total);
			else more++;
			displayed += line_queue_count;
			line_queue_count = 0;
			line_queue = NULL;
		} else if (line_queue_count > TRACE_COUNT) {
			line_queue = line_queue->next;
			line_queue_count--;
		}

		line->count[sym_counter] = zero ? 0 : line->count[sym_counter] * 7 / 8;
		line = line->next;
	}
383
	pthread_mutex_unlock(&syme->src->lock);
384 385 386
	if (more)
		printf("%d lines not displayed, maybe increase display entries [e]\n", more);
}
387

388
/*
389
 * Symbols will be added here in event__process_sample and will get out
390 391 392
 * after decayed.
 */
static LIST_HEAD(active_symbols);
393
static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER;
394 395 396 397 398 399

/*
 * Ordering weight: count-1 * count-2 * ... / count-n
 */
static double sym_weight(const struct sym_entry *sym)
{
400
	double weight = sym->snap_count;
401 402
	int counter;

403 404 405
	if (!display_weighted)
		return weight;

406 407 408 409 410 411 412 413
	for (counter = 1; counter < nr_counters-1; counter++)
		weight *= sym->count[counter];

	weight /= (sym->count[counter] + 1);

	return weight;
}

414 415
static long			samples;
static long			userspace_samples;
416 417
static const char		CONSOLE_CLEAR[] = "";

418
static void __list_insert_active_sym(struct sym_entry *syme)
419 420 421 422
{
	list_add(&syme->node, &active_symbols);
}

423 424 425 426 427 428 429
static void list_remove_active_sym(struct sym_entry *syme)
{
	pthread_mutex_lock(&active_symbols_lock);
	list_del_init(&syme->node);
	pthread_mutex_unlock(&active_symbols_lock);
}

430 431 432 433 434 435 436 437 438 439
static void rb_insert_active_sym(struct rb_root *tree, struct sym_entry *se)
{
	struct rb_node **p = &tree->rb_node;
	struct rb_node *parent = NULL;
	struct sym_entry *iter;

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

440
		if (se->weight > iter->weight)
441 442 443 444 445 446 447 448
			p = &(*p)->rb_left;
		else
			p = &(*p)->rb_right;
	}

	rb_link_node(&se->rb_node, parent, p);
	rb_insert_color(&se->rb_node, tree);
}
449 450 451

static void print_sym_table(void)
{
452
	int printed = 0, j;
453
	int counter, snap = !display_weighted ? sym_counter : 0;
454 455 456
	float samples_per_sec = samples/delay_secs;
	float ksamples_per_sec = (samples-userspace_samples)/delay_secs;
	float sum_ksamples = 0.0;
457 458 459
	struct sym_entry *syme, *n;
	struct rb_root tmp = RB_ROOT;
	struct rb_node *nd;
460
	int sym_width = 0, dso_width = 0;
461
	const int win_width = winsize.ws_col - 1;
462
	struct dso *unique_dso = NULL, *first_dso = NULL;
463

464
	samples = userspace_samples = 0;
465

466
	/* Sort the active symbols */
467 468 469 470 471
	pthread_mutex_lock(&active_symbols_lock);
	syme = list_entry(active_symbols.next, struct sym_entry, node);
	pthread_mutex_unlock(&active_symbols_lock);

	list_for_each_entry_safe_from(syme, n, &active_symbols, node) {
472
		syme->snap_count = syme->count[snap];
473
		if (syme->snap_count != 0) {
474

475 476 477 478 479 480 481
			if ((hide_user_symbols &&
			     syme->origin == PERF_RECORD_MISC_USER) ||
			    (hide_kernel_symbols &&
			     syme->origin == PERF_RECORD_MISC_KERNEL)) {
				list_remove_active_sym(syme);
				continue;
			}
482
			syme->weight = sym_weight(syme);
483
			rb_insert_active_sym(&tmp, syme);
484
			sum_ksamples += syme->snap_count;
485 486

			for (j = 0; j < nr_counters; j++)
487 488
				syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8;
		} else
489
			list_remove_active_sym(syme);
490 491
	}

492
	puts(CONSOLE_CLEAR);
493

494
	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
495
	printf( "   PerfTop:%8.0f irqs/sec  kernel:%4.1f%% [",
496 497
		samples_per_sec,
		100.0 - (100.0*((samples_per_sec-ksamples_per_sec)/samples_per_sec)));
498

499
	if (nr_counters == 1 || !display_weighted) {
500
		printf("%Ld", (u64)attrs[0].sample_period);
I
Ingo Molnar 已提交
501 502 503 504 505
		if (freq)
			printf("Hz ");
		else
			printf(" ");
	}
506

507 508 509
	if (!display_weighted)
		printf("%s", event_name(sym_counter));
	else for (counter = 0; counter < nr_counters; counter++) {
510 511 512 513 514 515 516 517
		if (counter)
			printf("/");

		printf("%s", event_name(counter));
	}

	printf( "], ");

518 519
	if (target_pid != -1)
		printf(" (target_pid: %d", target_pid);
520 521 522 523 524 525
	else
		printf(" (all");

	if (profile_cpu != -1)
		printf(", cpu: %d)\n", profile_cpu);
	else {
526
		if (target_pid != -1)
527 528 529 530 531
			printf(")\n");
		else
			printf(", %d CPUs)\n", nr_cpus);
	}

532
	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
533

534 535 536 537 538
	if (sym_filter_entry) {
		show_details(sym_filter_entry);
		return;
	}

539 540 541 542 543 544 545 546 547
	/*
	 * Find the longest symbol name that will be displayed
	 */
	for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
		syme = rb_entry(nd, struct sym_entry, rb_node);
		if (++printed > print_entries ||
		    (int)syme->snap_count < count_filter)
			continue;

548 549 550 551 552 553 554 555
		if (first_dso == NULL)
			unique_dso = first_dso = syme->map->dso;
		else if (syme->map->dso != first_dso)
			unique_dso = NULL;

		if (syme->map->dso->long_name_len > dso_width)
			dso_width = syme->map->dso->long_name_len;

556 557 558 559 560 561
		if (syme->name_len > sym_width)
			sym_width = syme->name_len;
	}

	printed = 0;

562 563 564 565 566 567 568 569
	if (unique_dso)
		printf("DSO: %s\n", unique_dso->long_name);
	else {
		int max_dso_width = winsize.ws_col - sym_width - 29;
		if (dso_width > max_dso_width)
			dso_width = max_dso_width;
		putchar('\n');
	}
570
	if (nr_counters == 1)
571
		printf("             samples  pcnt");
572
	else
573
		printf("   weight    samples  pcnt");
574

575 576
	if (verbose)
		printf("         RIP       ");
577 578 579 580
	printf(" %-*.*s", sym_width, sym_width, "function");
	if (!unique_dso)
		printf(" DSO");
	putchar('\n');
581
	printf("   %s    _______ _____",
582 583
	       nr_counters == 1 ? "      " : "______");
	if (verbose)
584
		printf(" ________________");
585 586 587 588
	printf(" %-*.*s", sym_width, sym_width, graph_line);
	if (!unique_dso)
		printf(" %-*.*s", dso_width, dso_width, graph_line);
	puts("\n");
589

590
	for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
591
		struct symbol *sym;
592
		double pcnt;
593

594
		syme = rb_entry(nd, struct sym_entry, rb_node);
595
		sym = sym_entry__symbol(syme);
596

597
		if (++printed > print_entries || (int)syme->snap_count < count_filter)
598
			continue;
599

600 601
		pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) /
					 sum_ksamples));
602

603
		if (nr_counters == 1 || !display_weighted)
604
			printf("%20.2f ", syme->weight);
605
		else
606
			printf("%9.1f %10ld ", syme->weight, syme->snap_count);
607

608
		percent_color_fprintf(stdout, "%4.1f%%", pcnt);
609
		if (verbose)
610
			printf(" %016llx", sym->start);
611
		printf(" %-*.*s", sym_width, sym_width, sym->name);
612 613 614 615 616
		if (!unique_dso)
			printf(" %-*.*s", dso_width, dso_width,
			       dso_width >= syme->map->dso->long_name_len ?
						syme->map->dso->long_name :
						syme->map->dso->short_name);
617
		printf("\n");
618 619 620
	}
}

621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663
static void prompt_integer(int *target, const char *msg)
{
	char *buf = malloc(0), *p;
	size_t dummy = 0;
	int tmp;

	fprintf(stdout, "\n%s: ", msg);
	if (getline(&buf, &dummy, stdin) < 0)
		return;

	p = strchr(buf, '\n');
	if (p)
		*p = 0;

	p = buf;
	while(*p) {
		if (!isdigit(*p))
			goto out_free;
		p++;
	}
	tmp = strtoul(buf, NULL, 10);
	*target = tmp;
out_free:
	free(buf);
}

static void prompt_percent(int *target, const char *msg)
{
	int tmp = 0;

	prompt_integer(&tmp, msg);
	if (tmp >= 0 && tmp <= 100)
		*target = tmp;
}

static void prompt_symbol(struct sym_entry **target, const char *msg)
{
	char *buf = malloc(0), *p;
	struct sym_entry *syme = *target, *n, *found = NULL;
	size_t dummy = 0;

	/* zero counters of active symbol */
	if (syme) {
664
		pthread_mutex_lock(&syme->src->lock);
665 666
		__zero_source_counters(syme);
		*target = NULL;
667
		pthread_mutex_unlock(&syme->src->lock);
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682
	}

	fprintf(stdout, "\n%s: ", msg);
	if (getline(&buf, &dummy, stdin) < 0)
		goto out_free;

	p = strchr(buf, '\n');
	if (p)
		*p = 0;

	pthread_mutex_lock(&active_symbols_lock);
	syme = list_entry(active_symbols.next, struct sym_entry, node);
	pthread_mutex_unlock(&active_symbols_lock);

	list_for_each_entry_safe_from(syme, n, &active_symbols, node) {
683
		struct symbol *sym = sym_entry__symbol(syme);
684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701

		if (!strcmp(buf, sym->name)) {
			found = syme;
			break;
		}
	}

	if (!found) {
		fprintf(stderr, "Sorry, %s is not active.\n", sym_filter);
		sleep(1);
		return;
	} else
		parse_source(found);

out_free:
	free(buf);
}

702
static void print_mapped_keys(void)
703
{
704 705 706
	char *name = NULL;

	if (sym_filter_entry) {
707
		struct symbol *sym = sym_entry__symbol(sym_filter_entry);
708 709 710 711 712 713 714 715 716 717 718 719
		name = sym->name;
	}

	fprintf(stdout, "\nMapped keys:\n");
	fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", delay_secs);
	fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", print_entries);

	if (nr_counters > 1)
		fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", event_name(sym_counter));

	fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", count_filter);

720
	if (vmlinux_name) {
721 722 723 724 725 726 727 728
		fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", sym_pcnt_filter);
		fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
		fprintf(stdout, "\t[S]     stop annotation.\n");
	}

	if (nr_counters > 1)
		fprintf(stdout, "\t[w]     toggle display weighted/count[E]r. \t(%d)\n", display_weighted ? 1 : 0);

729 730 731 732 733 734
	fprintf(stdout,
		"\t[K]     hide kernel_symbols symbols.             \t(%s)\n",
		hide_kernel_symbols ? "yes" : "no");
	fprintf(stdout,
		"\t[U]     hide user symbols.               \t(%s)\n",
		hide_user_symbols ? "yes" : "no");
735
	fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", zero ? 1 : 0);
736 737 738 739 740 741 742 743 744 745 746 747
	fprintf(stdout, "\t[qQ]    quit.\n");
}

static int key_mapped(int c)
{
	switch (c) {
		case 'd':
		case 'e':
		case 'f':
		case 'z':
		case 'q':
		case 'Q':
748 749
		case 'K':
		case 'U':
750 751 752 753 754 755 756
			return 1;
		case 'E':
		case 'w':
			return nr_counters > 1 ? 1 : 0;
		case 'F':
		case 's':
		case 'S':
757 758 759
			return vmlinux_name ? 1 : 0;
		default:
			break;
760 761 762
	}

	return 0;
763 764 765 766
}

static void handle_keypress(int c)
{
767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789
	if (!key_mapped(c)) {
		struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
		struct termios tc, save;

		print_mapped_keys();
		fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
		fflush(stdout);

		tcgetattr(0, &save);
		tc = save;
		tc.c_lflag &= ~(ICANON | ECHO);
		tc.c_cc[VMIN] = 0;
		tc.c_cc[VTIME] = 0;
		tcsetattr(0, TCSANOW, &tc);

		poll(&stdin_poll, 1, -1);
		c = getc(stdin);

		tcsetattr(0, TCSAFLUSH, &save);
		if (!key_mapped(c))
			return;
	}

790 791 792
	switch (c) {
		case 'd':
			prompt_integer(&delay_secs, "Enter display delay");
793 794
			if (delay_secs < 1)
				delay_secs = 1;
795 796 797
			break;
		case 'e':
			prompt_integer(&print_entries, "Enter display entries (lines)");
798
			if (print_entries == 0) {
799
				sig_winch_handler(SIGWINCH);
800 801 802
				signal(SIGWINCH, sig_winch_handler);
			} else
				signal(SIGWINCH, SIG_DFL);
803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826
			break;
		case 'E':
			if (nr_counters > 1) {
				int i;

				fprintf(stderr, "\nAvailable events:");
				for (i = 0; i < nr_counters; i++)
					fprintf(stderr, "\n\t%d %s", i, event_name(i));

				prompt_integer(&sym_counter, "Enter details event counter");

				if (sym_counter >= nr_counters) {
					fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(0));
					sym_counter = 0;
					sleep(1);
				}
			} else sym_counter = 0;
			break;
		case 'f':
			prompt_integer(&count_filter, "Enter display event count filter");
			break;
		case 'F':
			prompt_percent(&sym_pcnt_filter, "Enter details display event filter (percent)");
			break;
827 828 829
		case 'K':
			hide_kernel_symbols = !hide_kernel_symbols;
			break;
830 831 832 833 834 835 836 837 838 839 840 841 842
		case 'q':
		case 'Q':
			printf("exiting.\n");
			exit(0);
		case 's':
			prompt_symbol(&sym_filter_entry, "Enter details symbol");
			break;
		case 'S':
			if (!sym_filter_entry)
				break;
			else {
				struct sym_entry *syme = sym_filter_entry;

843
				pthread_mutex_lock(&syme->src->lock);
844 845
				sym_filter_entry = NULL;
				__zero_source_counters(syme);
846
				pthread_mutex_unlock(&syme->src->lock);
847 848
			}
			break;
849 850 851
		case 'U':
			hide_user_symbols = !hide_user_symbols;
			break;
852 853 854
		case 'w':
			display_weighted = ~display_weighted;
			break;
855 856 857
		case 'z':
			zero = ~zero;
			break;
858 859
		default:
			break;
860 861 862
	}
}

863
static void *display_thread(void *arg __used)
864
{
865
	struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
866 867 868 869 870 871 872 873
	struct termios tc, save;
	int delay_msecs, c;

	tcgetattr(0, &save);
	tc = save;
	tc.c_lflag &= ~(ICANON | ECHO);
	tc.c_cc[VMIN] = 0;
	tc.c_cc[VTIME] = 0;
874

875 876 877 878 879
repeat:
	delay_msecs = delay_secs * 1000;
	tcsetattr(0, TCSANOW, &tc);
	/* trash return*/
	getc(stdin);
880

881
	do {
882
		print_sym_table();
883 884
	} while (!poll(&stdin_poll, 1, delay_msecs) == 1);

885 886 887 888 889
	c = getc(stdin);
	tcsetattr(0, TCSAFLUSH, &save);

	handle_keypress(c);
	goto repeat;
890 891 892 893

	return NULL;
}

894
/* Tag samples to be skipped. */
895
static const char *skip_symbols[] = {
896 897 898 899 900
	"default_idle",
	"cpu_idle",
	"enter_idle",
	"exit_idle",
	"mwait_idle",
901
	"mwait_idle_with_hints",
902
	"poll_idle",
903 904
	"ppc64_runlatch_off",
	"pseries_dedicated_idle_sleep",
905 906 907
	NULL
};

908
static int symbol_filter(struct map *map, struct symbol *sym)
909
{
910 911
	struct sym_entry *syme;
	const char *name = sym->name;
912
	int i;
913

914 915 916 917 918 919 920
	/*
	 * ppc64 uses function descriptors and appends a '.' to the
	 * start of every instruction address. Remove it.
	 */
	if (name[0] == '.')
		name++;

921 922 923 924 925 926 927
	if (!strcmp(name, "_text") ||
	    !strcmp(name, "_etext") ||
	    !strcmp(name, "_sinittext") ||
	    !strncmp("init_module", name, 11) ||
	    !strncmp("cleanup_module", name, 14) ||
	    strstr(name, "_text_start") ||
	    strstr(name, "_text_end"))
928 929
		return 1;

930
	syme = symbol__priv(sym);
931
	syme->map = map;
932
	syme->src = NULL;
933 934 935
	if (!sym_filter_entry && sym_filter && !strcmp(name, sym_filter))
		sym_filter_entry = syme;

936 937 938 939 940 941
	for (i = 0; skip_symbols[i]; i++) {
		if (!strcmp(skip_symbols[i], name)) {
			syme->skip = 1;
			break;
		}
	}
942

943 944 945
	if (!syme->skip)
		syme->name_len = strlen(sym->name);

946 947 948
	return 0;
}

949
static int parse_symbols(void)
950
{
951 952 953 954 955
	struct dso *kernel = dsos__load_kernel();

	if (kernel == NULL)
		return -1;

956 957 958 959 960 961 962
	if (dsos__load_modules() < 0)
		pr_debug("Couldn't read the complete list of modules, "
			 "continuing...\n");

	if (dsos__load_modules_sym(symbol_filter) < 0)
		pr_warning("Failed to read module symbols, continuing...\n");

963
	if (dso__load_kernel_sym(kernel, symbol_filter, 1) <= 0)
964 965
		pr_debug("Couldn't read the complete list of kernel symbols, "
			 "continuing...\n");
966

967
	if (dump_symtab)
968
		dsos__fprintf(stderr);
969

970
	return 0;
971 972
}

973
static void event__process_sample(const event_t *self, int counter)
974
{
975
	u64 ip = self->ip.ip;
976
	struct map *map;
977 978
	struct sym_entry *syme;
	struct symbol *sym;
979
	u8 origin = self->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
980

981
	switch (origin) {
982
	case PERF_RECORD_MISC_USER: {
983
		struct thread *thread;
984

985 986 987 988
		if (hide_user_symbols)
			return;

		thread = threads__findnew(self->ip.pid);
989
		if (thread == NULL)
990
			return;
991 992 993 994

		map = thread__find_map(thread, ip);
		if (map != NULL) {
			ip = map->map_ip(map, ip);
995
			sym = map__find_symbol(map, ip, symbol_filter);
996 997 998 999
			if (sym == NULL)
				return;
			userspace_samples++;
			break;
1000 1001
		}
	}
1002 1003 1004 1005 1006 1007 1008 1009 1010 1011
		/*
		 * 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 or vdso (which executes in user-mode).
		 */
		if ((long long)ip >= 0)
			return;
		/* Fall thru */
	case PERF_RECORD_MISC_KERNEL:
1012 1013 1014
		if (hide_kernel_symbols)
			return;

1015 1016 1017 1018 1019 1020 1021 1022
		sym = kernel_maps__find_symbol(ip, &map);
		if (sym == NULL)
			return;
		break;
	default:
		return;
	}

1023
	syme = symbol__priv(sym);
1024

1025 1026
	if (!syme->skip) {
		syme->count[counter]++;
1027
		syme->origin = origin;
1028 1029 1030 1031 1032 1033 1034 1035
		record_precise_ip(syme, counter, ip);
		pthread_mutex_lock(&active_symbols_lock);
		if (list_empty(&syme->node) || !syme->node.next)
			__list_insert_active_sym(syme);
		pthread_mutex_unlock(&active_symbols_lock);
		++samples;
		return;
	}
1036 1037
}

1038
static void event__process_mmap(event_t *self)
1039
{
1040 1041 1042
	struct thread *thread = threads__findnew(self->mmap.pid);

	if (thread != NULL) {
1043
		struct map *map = map__new(&self->mmap, NULL, 0);
1044 1045 1046 1047
		if (map != NULL)
			thread__insert_map(thread, map);
	}
}
1048

1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
static void event__process_comm(event_t *self)
{
	struct thread *thread = threads__findnew(self->comm.pid);

	if (thread != NULL)
		thread__set_comm(thread, self->comm.comm);
}

static int event__process(event_t *event)
{
	switch (event->header.type) {
	case PERF_RECORD_COMM:
		event__process_comm(event);
		break;
	case PERF_RECORD_MMAP:
		event__process_mmap(event);
		break;
	default:
		break;
1068 1069
	}

1070
	return 0;
1071 1072 1073
}

struct mmap_data {
1074 1075
	int			counter;
	void			*base;
1076
	int			mask;
1077
	unsigned int		prev;
1078 1079 1080 1081
};

static unsigned int mmap_read_head(struct mmap_data *md)
{
1082
	struct perf_event_mmap_page *pc = md->base;
1083 1084 1085 1086 1087 1088 1089 1090
	int head;

	head = pc->data_head;
	rmb();

	return head;
}

1091
static void mmap_read_counter(struct mmap_data *md)
1092 1093 1094 1095 1096 1097 1098 1099
{
	unsigned int head = mmap_read_head(md);
	unsigned int old = md->prev;
	unsigned char *data = md->base + page_size;
	int diff;

	/*
	 * If we're further behind than half the buffer, there's a chance
1100
	 * the writer will bite our tail and mess up the samples under us.
1101 1102 1103 1104 1105 1106 1107
	 *
	 * If we somehow ended up ahead of the head, we got messed up.
	 *
	 * In either case, truncate and restart at head.
	 */
	diff = head - old;
	if (diff > md->mask / 2 || diff < 0) {
1108
		fprintf(stderr, "WARNING: failed to keep up with mmap data.\n");
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120

		/*
		 * head points to a known good entry, start there.
		 */
		old = head;
	}

	for (; old != head;) {
		event_t *event = (event_t *)&data[old & md->mask];

		event_t event_copy;

1121
		size_t size = event->header.size;
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142

		/*
		 * Event straddles the mmap boundary -- header should always
		 * be inside due to u64 alignment of output.
		 */
		if ((old & md->mask) + size != ((old + size) & md->mask)) {
			unsigned int offset = old;
			unsigned int len = min(sizeof(*event), size), cpy;
			void *dst = &event_copy;

			do {
				cpy = min(md->mask + 1 - (offset & md->mask), len);
				memcpy(dst, &data[offset & md->mask], cpy);
				offset += cpy;
				dst += cpy;
				len -= cpy;
			} while (len);

			event = &event_copy;
		}

1143 1144 1145 1146
		if (event->header.type == PERF_RECORD_SAMPLE)
			event__process_sample(event, md->counter);
		else
			event__process(event);
1147 1148 1149 1150 1151 1152
		old += size;
	}

	md->prev = old;
}

M
Mike Galbraith 已提交
1153 1154 1155
static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS];
static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS];

1156 1157 1158 1159 1160 1161 1162 1163 1164 1165
static void mmap_read(void)
{
	int i, counter;

	for (i = 0; i < nr_cpus; i++) {
		for (counter = 0; counter < nr_counters; counter++)
			mmap_read_counter(&mmap_array[i][counter]);
	}
}

1166 1167 1168 1169
int nr_poll;
int group_fd;

static void start_counter(int i, int counter)
1170
{
1171
	struct perf_event_attr *attr;
1172
	int cpu;
1173 1174 1175 1176 1177 1178 1179 1180

	cpu = profile_cpu;
	if (target_pid == -1 && profile_cpu == -1)
		cpu = i;

	attr = attrs + counter;

	attr->sample_type	= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
1181 1182 1183 1184 1185 1186 1187

	if (freq) {
		attr->sample_type	|= PERF_SAMPLE_PERIOD;
		attr->freq		= 1;
		attr->sample_freq	= freq;
	}

1188
	attr->inherit		= (cpu < 0) && inherit;
1189
	attr->mmap		= 1;
1190 1191

try_again:
1192
	fd[i][counter] = sys_perf_event_open(attr, target_pid, cpu, group_fd, 0);
1193 1194 1195 1196

	if (fd[i][counter] < 0) {
		int err = errno;

P
Pekka Enberg 已提交
1197
		if (err == EPERM || err == EACCES)
1198
			die("No permission - are you root?\n");
1199 1200 1201 1202 1203 1204
		/*
		 * If it's cycles then fall back to hrtimer
		 * based cpu-clock-tick sw counter, which
		 * is always available even if no PMU support:
		 */
		if (attr->type == PERF_TYPE_HARDWARE
1205
			&& attr->config == PERF_COUNT_HW_CPU_CYCLES) {
1206

1207 1208 1209
			if (verbose)
				warning(" ... trying to fall back to cpu-clock-ticks\n");

1210
			attr->type = PERF_TYPE_SOFTWARE;
1211
			attr->config = PERF_COUNT_SW_CPU_CLOCK;
1212 1213
			goto try_again;
		}
1214 1215 1216
		printf("\n");
		error("perfcounter syscall returned with %d (%s)\n",
			fd[i][counter], strerror(err));
1217
		die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
		exit(-1);
	}
	assert(fd[i][counter] >= 0);
	fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);

	/*
	 * First counter acts as the group leader:
	 */
	if (group && group_fd == -1)
		group_fd = fd[i][counter];

	event_array[nr_poll].fd = fd[i][counter];
	event_array[nr_poll].events = POLLIN;
	nr_poll++;

	mmap_array[i][counter].counter = counter;
	mmap_array[i][counter].prev = 0;
	mmap_array[i][counter].mask = mmap_pages*page_size - 1;
	mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
			PROT_READ, MAP_SHARED, fd[i][counter], 0);
	if (mmap_array[i][counter].base == MAP_FAILED)
		die("failed to mmap with %d (%s)\n", errno, strerror(errno));
}

static int __cmd_top(void)
{
	pthread_t thread;
	int i, counter;
1246 1247
	int ret;

1248 1249 1250 1251 1252
	if (target_pid != -1)
		event__synthesize_thread(target_pid, event__process);
	else
		event__synthesize_threads(event__process);

1253 1254
	for (i = 0; i < nr_cpus; i++) {
		group_fd = -1;
1255 1256
		for (counter = 0; counter < nr_counters; counter++)
			start_counter(i, counter);
1257 1258
	}

1259 1260 1261 1262 1263
	/* Wait for a minimal set of events before starting the snapshot */
	poll(event_array, nr_poll, 100);

	mmap_read();

1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279
	if (pthread_create(&thread, NULL, display_thread, NULL)) {
		printf("Could not create display thread.\n");
		exit(-1);
	}

	if (realtime_prio) {
		struct sched_param param;

		param.sched_priority = realtime_prio;
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
			printf("Could not set realtime priority.\n");
			exit(-1);
		}
	}

	while (1) {
1280
		int hits = samples;
1281

1282
		mmap_read();
1283

1284
		if (hits == samples)
1285 1286 1287 1288 1289
			ret = poll(event_array, nr_poll, 100);
	}

	return 0;
}
1290 1291 1292 1293 1294 1295 1296 1297

static const char * const top_usage[] = {
	"perf top [<options>]",
	NULL
};

static const struct option options[] = {
	OPT_CALLBACK('e', "event", NULL, "event",
1298 1299
		     "event selector. use 'perf list' to list available events",
		     parse_events),
1300 1301 1302 1303 1304 1305 1306 1307
	OPT_INTEGER('c', "count", &default_interval,
		    "event period to sample"),
	OPT_INTEGER('p', "pid", &target_pid,
		    "profile events on existing pid"),
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
	OPT_INTEGER('C', "CPU", &profile_cpu,
		    "CPU to profile on"),
1308
	OPT_STRING('k', "vmlinux", &vmlinux_name, "file", "vmlinux pathname"),
1309 1310
	OPT_BOOLEAN('K', "hide_kernel_symbols", &hide_kernel_symbols,
		    "hide kernel symbols"),
1311 1312 1313 1314
	OPT_INTEGER('m', "mmap-pages", &mmap_pages,
		    "number of mmap data pages"),
	OPT_INTEGER('r', "realtime", &realtime_prio,
		    "collect data with this RT SCHED_FIFO priority"),
M
Mike Galbraith 已提交
1315
	OPT_INTEGER('d', "delay", &delay_secs,
1316 1317 1318
		    "number of seconds to delay between refreshes"),
	OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
			    "dump the symbol table used for profiling"),
1319
	OPT_INTEGER('f', "count-filter", &count_filter,
1320 1321 1322
		    "only display functions with more events than this"),
	OPT_BOOLEAN('g', "group", &group,
			    "put the counters into a counter group"),
1323 1324
	OPT_BOOLEAN('i', "inherit", &inherit,
		    "child tasks inherit counters"),
1325 1326
	OPT_STRING('s', "sym-annotate", &sym_filter, "symbol name",
		    "symbol to annotate - requires -k option"),
A
Anton Blanchard 已提交
1327
	OPT_BOOLEAN('z', "zero", &zero,
1328
		    "zero history across updates"),
1329
	OPT_INTEGER('F', "freq", &freq,
1330
		    "profile at this frequency"),
1331 1332
	OPT_INTEGER('E', "entries", &print_entries,
		    "display this many functions"),
1333 1334
	OPT_BOOLEAN('U', "hide_user_symbols", &hide_user_symbols,
		    "hide user symbols"),
1335 1336
	OPT_BOOLEAN('v', "verbose", &verbose,
		    "be more verbose (show counter open errors, etc)"),
1337 1338 1339
	OPT_END()
};

1340
int cmd_top(int argc, const char **argv, const char *prefix __used)
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356
{
	int counter;

	page_size = sysconf(_SC_PAGE_SIZE);

	argc = parse_options(argc, argv, options, top_usage, 0);
	if (argc)
		usage_with_options(top_usage, options);

	/* CPU and PID are mutually exclusive */
	if (target_pid != -1 && profile_cpu != -1) {
		printf("WARNING: PID switch overriding CPU\n");
		sleep(1);
		profile_cpu = -1;
	}

1357
	if (!nr_counters)
1358 1359
		nr_counters = 1;

1360 1361 1362
	symbol__init(sizeof(struct sym_entry) +
		     (nr_counters + 1) * sizeof(unsigned long));

1363 1364 1365
	if (delay_secs < 1)
		delay_secs = 1;

1366
	parse_symbols();
1367
	parse_source(sym_filter_entry);
1368

1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381

	/*
	 * User specified count overrides default frequency.
	 */
	if (default_interval)
		freq = 0;
	else if (freq) {
		default_interval = freq;
	} else {
		fprintf(stderr, "frequency and count are zero, aborting\n");
		exit(EXIT_FAILURE);
	}

1382 1383 1384
	/*
	 * Fill in the ones not specifically initialized via -c:
	 */
1385
	for (counter = 0; counter < nr_counters; counter++) {
1386
		if (attrs[counter].sample_period)
1387 1388
			continue;

1389
		attrs[counter].sample_period = default_interval;
1390 1391 1392 1393 1394 1395 1396 1397 1398
	}

	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	assert(nr_cpus <= MAX_NR_CPUS);
	assert(nr_cpus >= 0);

	if (target_pid != -1 || profile_cpu != -1)
		nr_cpus = 1;

1399
	get_term_dimensions(&winsize);
1400
	if (print_entries == 0) {
1401
		update_print_entries(&winsize);
1402 1403 1404
		signal(SIGWINCH, sig_winch_handler);
	}

1405 1406
	return __cmd_top();
}